]>
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 | |
2d24d165 CL |
359 | /* Identifier used to distinguish vhost devices from each other. */ |
360 | char vhost_id[PATH_MAX]; | |
7d1ced01 | 361 | |
8a9562d2 | 362 | /* In dpdk_list. */ |
ca6ba700 | 363 | struct ovs_list list_node OVS_GUARDED_BY(dpdk_mutex); |
0bf765f7 IS |
364 | |
365 | /* QoS configuration and lock for the device */ | |
366 | struct qos_conf *qos_conf; | |
367 | rte_spinlock_t qos_lock; | |
368 | ||
050c60bf DDP |
369 | /* The following properties cannot be changed when a device is running, |
370 | * so we remember the request and update them next time | |
371 | * netdev_dpdk*_reconfigure() is called */ | |
0072e931 | 372 | int requested_mtu; |
050c60bf DDP |
373 | int requested_n_txq; |
374 | int requested_n_rxq; | |
9509913a | 375 | |
db8f13b0 CL |
376 | /* Socket ID detected when vHost device is brought up */ |
377 | int requested_socket_id; | |
378 | ||
c1ff66ac CL |
379 | /* Denotes whether vHost port is client/server mode */ |
380 | uint64_t vhost_driver_flags; | |
381 | ||
9509913a IS |
382 | /* Ingress Policer */ |
383 | OVSRCU_TYPE(struct ingress_policer *) ingress_policer; | |
384 | uint32_t policer_rate; | |
385 | uint32_t policer_burst; | |
9fd39370 SC |
386 | |
387 | /* DPDK-ETH Flow control */ | |
388 | struct rte_eth_fc_conf fc_conf; | |
8a9562d2 PS |
389 | }; |
390 | ||
391 | struct netdev_rxq_dpdk { | |
392 | struct netdev_rxq up; | |
393 | int port_id; | |
394 | }; | |
395 | ||
5f17de68 | 396 | static bool dpdk_thread_is_pmd(void); |
db73f716 | 397 | |
8a9562d2 PS |
398 | static int netdev_dpdk_construct(struct netdev *); |
399 | ||
0a0f39df | 400 | int netdev_dpdk_get_vid(const struct netdev_dpdk *dev); |
58397e6c | 401 | |
9509913a IS |
402 | struct ingress_policer * |
403 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev); | |
404 | ||
8a9562d2 PS |
405 | static bool |
406 | is_dpdk_class(const struct netdev_class *class) | |
407 | { | |
408 | return class->construct == netdev_dpdk_construct; | |
409 | } | |
410 | ||
4be4d22c MK |
411 | /* DPDK NIC drivers allocate RX buffers at a particular granularity, typically |
412 | * aligned at 1k or less. If a declared mbuf size is not a multiple of this | |
413 | * value, insufficient buffers are allocated to accomodate the packet in its | |
414 | * entirety. Furthermore, certain drivers need to ensure that there is also | |
415 | * sufficient space in the Rx buffer to accommodate two VLAN tags (for QinQ | |
416 | * frames). If the RX buffer is too small, then the driver enables scatter RX | |
417 | * behaviour, which reduces performance. To prevent this, use a buffer size that | |
418 | * is closest to 'mtu', but which satisfies the aforementioned criteria. | |
419 | */ | |
420 | static uint32_t | |
421 | dpdk_buf_size(int mtu) | |
422 | { | |
423 | return ROUND_UP((MTU_TO_MAX_FRAME_LEN(mtu) + RTE_PKTMBUF_HEADROOM), | |
424 | NETDEV_DPDK_MBUF_ALIGN); | |
425 | } | |
426 | ||
58397e6c KT |
427 | /* XXX: use dpdk malloc for entire OVS. in fact huge page should be used |
428 | * for all other segments data, bss and text. */ | |
8a9562d2 PS |
429 | |
430 | static void * | |
431 | dpdk_rte_mzalloc(size_t sz) | |
432 | { | |
433 | void *ptr; | |
434 | ||
435 | ptr = rte_zmalloc(OVS_VPORT_DPDK, sz, OVS_CACHE_LINE_SIZE); | |
436 | if (ptr == NULL) { | |
437 | out_of_memory(); | |
438 | } | |
439 | return ptr; | |
440 | } | |
441 | ||
db73f716 DDP |
442 | /* XXX this function should be called only by pmd threads (or by non pmd |
443 | * threads holding the nonpmd_mempool_mutex) */ | |
8a9562d2 | 444 | void |
e14deea0 | 445 | free_dpdk_buf(struct dp_packet *p) |
8a9562d2 | 446 | { |
db73f716 | 447 | struct rte_mbuf *pkt = (struct rte_mbuf *) p; |
8a9562d2 | 448 | |
b00b4a81 | 449 | rte_pktmbuf_free(pkt); |
8a9562d2 PS |
450 | } |
451 | ||
b3cd9f9d PS |
452 | static void |
453 | ovs_rte_pktmbuf_init(struct rte_mempool *mp, | |
454 | void *opaque_arg OVS_UNUSED, | |
455 | void *_m, | |
456 | unsigned i OVS_UNUSED) | |
457 | { | |
458 | struct rte_mbuf *m = _m; | |
459 | ||
4be4d22c | 460 | rte_pktmbuf_init(mp, opaque_arg, _m, i); |
b3cd9f9d | 461 | |
cf62fa4c | 462 | dp_packet_init_dpdk((struct dp_packet *) m, m->buf_len); |
b3cd9f9d PS |
463 | } |
464 | ||
8a9562d2 PS |
465 | static struct dpdk_mp * |
466 | dpdk_mp_get(int socket_id, int mtu) OVS_REQUIRES(dpdk_mutex) | |
467 | { | |
468 | struct dpdk_mp *dmp = NULL; | |
469 | char mp_name[RTE_MEMPOOL_NAMESIZE]; | |
da79ce2b | 470 | unsigned mp_size; |
4be4d22c | 471 | struct rte_pktmbuf_pool_private mbp_priv; |
8a9562d2 PS |
472 | |
473 | LIST_FOR_EACH (dmp, list_node, &dpdk_mp_list) { | |
474 | if (dmp->socket_id == socket_id && dmp->mtu == mtu) { | |
475 | dmp->refcount++; | |
476 | return dmp; | |
477 | } | |
478 | } | |
479 | ||
480 | dmp = dpdk_rte_mzalloc(sizeof *dmp); | |
481 | dmp->socket_id = socket_id; | |
482 | dmp->mtu = mtu; | |
483 | dmp->refcount = 1; | |
4be4d22c | 484 | mbp_priv.mbuf_data_room_size = MBUF_SIZE(mtu) - sizeof(struct dp_packet); |
0072e931 MK |
485 | mbp_priv.mbuf_priv_size = sizeof (struct dp_packet) |
486 | - sizeof (struct rte_mbuf); | |
487 | /* XXX: this is a really rough method of provisioning memory. | |
488 | * It's impossible to determine what the exact memory requirements are when | |
489 | * the number of ports and rxqs that utilize a particular mempool can change | |
490 | * dynamically at runtime. For the moment, use this rough heurisitic. | |
491 | */ | |
492 | if (mtu >= ETHER_MTU) { | |
493 | mp_size = MAX_NB_MBUF; | |
494 | } else { | |
495 | mp_size = MIN_NB_MBUF; | |
496 | } | |
8a9562d2 | 497 | |
da79ce2b DDP |
498 | do { |
499 | if (snprintf(mp_name, RTE_MEMPOOL_NAMESIZE, "ovs_mp_%d_%d_%u", | |
500 | dmp->mtu, dmp->socket_id, mp_size) < 0) { | |
5f88de0d | 501 | goto fail; |
da79ce2b | 502 | } |
95fb793a | 503 | |
da79ce2b DDP |
504 | dmp->mp = rte_mempool_create(mp_name, mp_size, MBUF_SIZE(mtu), |
505 | MP_CACHE_SZ, | |
506 | sizeof(struct rte_pktmbuf_pool_private), | |
4be4d22c | 507 | rte_pktmbuf_pool_init, &mbp_priv, |
da79ce2b DDP |
508 | ovs_rte_pktmbuf_init, NULL, |
509 | socket_id, 0); | |
510 | } while (!dmp->mp && rte_errno == ENOMEM && (mp_size /= 2) >= MIN_NB_MBUF); | |
8a9562d2 PS |
511 | |
512 | if (dmp->mp == NULL) { | |
5f88de0d | 513 | goto fail; |
da79ce2b DDP |
514 | } else { |
515 | VLOG_DBG("Allocated \"%s\" mempool with %u mbufs", mp_name, mp_size ); | |
8a9562d2 PS |
516 | } |
517 | ||
417e7e66 | 518 | ovs_list_push_back(&dpdk_mp_list, &dmp->list_node); |
8a9562d2 | 519 | return dmp; |
5f88de0d IM |
520 | |
521 | fail: | |
522 | rte_free(dmp); | |
523 | return NULL; | |
8a9562d2 PS |
524 | } |
525 | ||
526 | static void | |
8d38823b | 527 | dpdk_mp_put(struct dpdk_mp *dmp) OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 528 | { |
8a9562d2 PS |
529 | if (!dmp) { |
530 | return; | |
531 | } | |
532 | ||
8d38823b | 533 | ovs_assert(dmp->refcount); |
8a9562d2 | 534 | |
8d38823b MK |
535 | if (!--dmp->refcount) { |
536 | ovs_list_remove(&dmp->list_node); | |
537 | rte_mempool_free(dmp->mp); | |
5f88de0d | 538 | rte_free(dmp); |
8a9562d2 | 539 | } |
8a9562d2 PS |
540 | } |
541 | ||
0072e931 MK |
542 | /* Tries to allocate new mempool on requested_socket_id with |
543 | * mbuf size corresponding to requested_mtu. | |
544 | * On success new configuration will be applied. | |
545 | * On error, device will be left unchanged. */ | |
546 | static int | |
547 | netdev_dpdk_mempool_configure(struct netdev_dpdk *dev) | |
548 | OVS_REQUIRES(dpdk_mutex) | |
549 | OVS_REQUIRES(dev->mutex) | |
550 | { | |
551 | uint32_t buf_size = dpdk_buf_size(dev->requested_mtu); | |
552 | struct dpdk_mp *mp; | |
553 | ||
554 | mp = dpdk_mp_get(dev->requested_socket_id, FRAME_LEN_TO_MTU(buf_size)); | |
555 | if (!mp) { | |
556 | VLOG_ERR("Insufficient memory to create memory pool for netdev " | |
557 | "%s, with MTU %d on socket %d\n", | |
558 | dev->up.name, dev->requested_mtu, dev->requested_socket_id); | |
559 | return ENOMEM; | |
560 | } else { | |
561 | dpdk_mp_put(dev->dpdk_mp); | |
562 | dev->dpdk_mp = mp; | |
563 | dev->mtu = dev->requested_mtu; | |
564 | dev->socket_id = dev->requested_socket_id; | |
565 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); | |
566 | } | |
567 | ||
568 | return 0; | |
569 | } | |
570 | ||
8a9562d2 PS |
571 | static void |
572 | check_link_status(struct netdev_dpdk *dev) | |
573 | { | |
574 | struct rte_eth_link link; | |
575 | ||
576 | rte_eth_link_get_nowait(dev->port_id, &link); | |
577 | ||
578 | if (dev->link.link_status != link.link_status) { | |
3e912ffc | 579 | netdev_change_seq_changed(&dev->up); |
8a9562d2 PS |
580 | |
581 | dev->link_reset_cnt++; | |
582 | dev->link = link; | |
583 | if (dev->link.link_status) { | |
584 | VLOG_DBG_RL(&rl, "Port %d Link Up - speed %u Mbps - %s", | |
585 | dev->port_id, (unsigned)dev->link.link_speed, | |
586 | (dev->link.link_duplex == ETH_LINK_FULL_DUPLEX) ? | |
587 | ("full-duplex") : ("half-duplex")); | |
588 | } else { | |
589 | VLOG_DBG_RL(&rl, "Port %d Link Down", dev->port_id); | |
590 | } | |
591 | } | |
592 | } | |
593 | ||
594 | static void * | |
595 | dpdk_watchdog(void *dummy OVS_UNUSED) | |
596 | { | |
597 | struct netdev_dpdk *dev; | |
598 | ||
599 | pthread_detach(pthread_self()); | |
600 | ||
601 | for (;;) { | |
602 | ovs_mutex_lock(&dpdk_mutex); | |
603 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { | |
604 | ovs_mutex_lock(&dev->mutex); | |
1f5b157e IM |
605 | if (dev->type == DPDK_DEV_ETH) { |
606 | check_link_status(dev); | |
607 | } | |
8a9562d2 PS |
608 | ovs_mutex_unlock(&dev->mutex); |
609 | } | |
610 | ovs_mutex_unlock(&dpdk_mutex); | |
611 | xsleep(DPDK_PORT_WATCHDOG_INTERVAL); | |
612 | } | |
613 | ||
614 | return NULL; | |
615 | } | |
616 | ||
b98d7669 DDP |
617 | static int |
618 | dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) | |
619 | { | |
620 | int diag = 0; | |
621 | int i; | |
0072e931 | 622 | struct rte_eth_conf conf = port_conf; |
b98d7669 | 623 | |
0072e931 MK |
624 | if (dev->mtu > ETHER_MTU) { |
625 | conf.rxmode.jumbo_frame = 1; | |
626 | conf.rxmode.max_rx_pkt_len = dev->max_packet_len; | |
627 | } else { | |
628 | conf.rxmode.jumbo_frame = 0; | |
629 | conf.rxmode.max_rx_pkt_len = 0; | |
630 | } | |
b98d7669 DDP |
631 | /* A device may report more queues than it makes available (this has |
632 | * been observed for Intel xl710, which reserves some of them for | |
633 | * SRIOV): rte_eth_*_queue_setup will fail if a queue is not | |
634 | * available. When this happens we can retry the configuration | |
635 | * and request less queues */ | |
636 | while (n_rxq && n_txq) { | |
637 | if (diag) { | |
638 | VLOG_INFO("Retrying setup with (rxq:%d txq:%d)", n_rxq, n_txq); | |
639 | } | |
640 | ||
0072e931 | 641 | diag = rte_eth_dev_configure(dev->port_id, n_rxq, n_txq, &conf); |
b98d7669 | 642 | if (diag) { |
0072e931 MK |
643 | VLOG_WARN("Interface %s eth_dev setup error %s\n", |
644 | dev->up.name, rte_strerror(-diag)); | |
b98d7669 DDP |
645 | break; |
646 | } | |
647 | ||
648 | for (i = 0; i < n_txq; i++) { | |
649 | diag = rte_eth_tx_queue_setup(dev->port_id, i, NIC_PORT_TX_Q_SIZE, | |
650 | dev->socket_id, NULL); | |
651 | if (diag) { | |
652 | VLOG_INFO("Interface %s txq(%d) setup error: %s", | |
653 | dev->up.name, i, rte_strerror(-diag)); | |
654 | break; | |
655 | } | |
656 | } | |
657 | ||
658 | if (i != n_txq) { | |
659 | /* Retry with less tx queues */ | |
660 | n_txq = i; | |
661 | continue; | |
662 | } | |
663 | ||
664 | for (i = 0; i < n_rxq; i++) { | |
665 | diag = rte_eth_rx_queue_setup(dev->port_id, i, NIC_PORT_RX_Q_SIZE, | |
666 | dev->socket_id, NULL, | |
667 | dev->dpdk_mp->mp); | |
668 | if (diag) { | |
669 | VLOG_INFO("Interface %s rxq(%d) setup error: %s", | |
670 | dev->up.name, i, rte_strerror(-diag)); | |
671 | break; | |
672 | } | |
673 | } | |
674 | ||
675 | if (i != n_rxq) { | |
676 | /* Retry with less rx queues */ | |
677 | n_rxq = i; | |
678 | continue; | |
679 | } | |
680 | ||
681 | dev->up.n_rxq = n_rxq; | |
81acebda | 682 | dev->up.n_txq = n_txq; |
b98d7669 DDP |
683 | |
684 | return 0; | |
685 | } | |
686 | ||
687 | return diag; | |
688 | } | |
689 | ||
9fd39370 SC |
690 | static void |
691 | dpdk_eth_flow_ctrl_setup(struct netdev_dpdk *dev) OVS_REQUIRES(dev->mutex) | |
692 | { | |
693 | if (rte_eth_dev_flow_ctrl_set(dev->port_id, &dev->fc_conf)) { | |
694 | VLOG_WARN("Failed to enable flow control on device %d", dev->port_id); | |
695 | } | |
696 | } | |
b98d7669 | 697 | |
8a9562d2 PS |
698 | static int |
699 | dpdk_eth_dev_init(struct netdev_dpdk *dev) OVS_REQUIRES(dpdk_mutex) | |
700 | { | |
701 | struct rte_pktmbuf_pool_private *mbp_priv; | |
a0cb2d66 | 702 | struct rte_eth_dev_info info; |
8a9562d2 PS |
703 | struct ether_addr eth_addr; |
704 | int diag; | |
b98d7669 | 705 | int n_rxq, n_txq; |
8a9562d2 | 706 | |
7cd1261d | 707 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
95fb793a | 708 | return ENODEV; |
8a9562d2 PS |
709 | } |
710 | ||
a0cb2d66 | 711 | rte_eth_dev_info_get(dev->port_id, &info); |
a0cb2d66 | 712 | |
b98d7669 DDP |
713 | n_rxq = MIN(info.max_rx_queues, dev->up.n_rxq); |
714 | n_txq = MIN(info.max_tx_queues, dev->up.n_txq); | |
715 | ||
716 | diag = dpdk_eth_dev_queue_setup(dev, n_rxq, n_txq); | |
8a9562d2 | 717 | if (diag) { |
b98d7669 DDP |
718 | VLOG_ERR("Interface %s(rxq:%d txq:%d) configure error: %s", |
719 | dev->up.name, n_rxq, n_txq, rte_strerror(-diag)); | |
95fb793a | 720 | return -diag; |
8a9562d2 PS |
721 | } |
722 | ||
8a9562d2 PS |
723 | diag = rte_eth_dev_start(dev->port_id); |
724 | if (diag) { | |
b98d7669 DDP |
725 | VLOG_ERR("Interface %s start error: %s", dev->up.name, |
726 | rte_strerror(-diag)); | |
95fb793a | 727 | return -diag; |
8a9562d2 PS |
728 | } |
729 | ||
730 | rte_eth_promiscuous_enable(dev->port_id); | |
731 | rte_eth_allmulticast_enable(dev->port_id); | |
732 | ||
733 | memset(ð_addr, 0x0, sizeof(eth_addr)); | |
734 | rte_eth_macaddr_get(dev->port_id, ð_addr); | |
735 | VLOG_INFO_RL(&rl, "Port %d: "ETH_ADDR_FMT"", | |
ca92d173 | 736 | dev->port_id, ETH_ADDR_BYTES_ARGS(eth_addr.addr_bytes)); |
8a9562d2 | 737 | |
ca92d173 | 738 | memcpy(dev->hwaddr.ea, eth_addr.addr_bytes, ETH_ADDR_LEN); |
8a9562d2 PS |
739 | rte_eth_link_get_nowait(dev->port_id, &dev->link); |
740 | ||
741 | mbp_priv = rte_mempool_get_priv(dev->dpdk_mp->mp); | |
742 | dev->buf_size = mbp_priv->mbuf_data_room_size - RTE_PKTMBUF_HEADROOM; | |
743 | ||
744 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
9fd39370 SC |
745 | |
746 | /* Get the Flow control configuration for DPDK-ETH */ | |
747 | diag = rte_eth_dev_flow_ctrl_get(dev->port_id, &dev->fc_conf); | |
748 | if (diag) { | |
749 | VLOG_DBG("cannot get flow control parameters on port=%d, err=%d", | |
750 | dev->port_id, diag); | |
751 | } | |
752 | ||
8a9562d2 PS |
753 | return 0; |
754 | } | |
755 | ||
756 | static struct netdev_dpdk * | |
757 | netdev_dpdk_cast(const struct netdev *netdev) | |
758 | { | |
759 | return CONTAINER_OF(netdev, struct netdev_dpdk, up); | |
760 | } | |
761 | ||
762 | static struct netdev * | |
763 | netdev_dpdk_alloc(void) | |
764 | { | |
bab69409 AC |
765 | struct netdev_dpdk *dev; |
766 | ||
767 | if (!rte_eal_init_ret) { /* Only after successful initialization */ | |
768 | dev = dpdk_rte_mzalloc(sizeof *dev); | |
769 | if (dev) { | |
770 | return &dev->up; | |
771 | } | |
772 | } | |
773 | return NULL; | |
8a9562d2 PS |
774 | } |
775 | ||
5a034064 | 776 | static void |
d46285a2 | 777 | netdev_dpdk_alloc_txq(struct netdev_dpdk *dev, unsigned int n_txqs) |
5a034064 | 778 | { |
bd5131ba | 779 | unsigned i; |
5a034064 | 780 | |
d46285a2 | 781 | dev->tx_q = dpdk_rte_mzalloc(n_txqs * sizeof *dev->tx_q); |
5a034064 | 782 | for (i = 0; i < n_txqs; i++) { |
585a5bea | 783 | /* Initialize map for vhost devices. */ |
d46285a2 DDP |
784 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
785 | rte_spinlock_init(&dev->tx_q[i].tx_lock); | |
5a034064 AW |
786 | } |
787 | } | |
788 | ||
8a9562d2 | 789 | static int |
d46285a2 | 790 | netdev_dpdk_init(struct netdev *netdev, unsigned int port_no, |
58397e6c | 791 | enum dpdk_dev_type type) |
5a034064 | 792 | OVS_REQUIRES(dpdk_mutex) |
8a9562d2 | 793 | { |
d46285a2 | 794 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b7a04e0 | 795 | int sid; |
95fb793a | 796 | int err = 0; |
8a9562d2 | 797 | |
d46285a2 DDP |
798 | ovs_mutex_init(&dev->mutex); |
799 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 800 | |
d46285a2 | 801 | rte_spinlock_init(&dev->stats_lock); |
45d947c4 | 802 | |
1b7a04e0 AW |
803 | /* If the 'sid' is negative, it means that the kernel fails |
804 | * to obtain the pci numa info. In that situation, always | |
805 | * use 'SOCKET0'. */ | |
58397e6c KT |
806 | if (type == DPDK_DEV_ETH) { |
807 | sid = rte_eth_dev_socket_id(port_no); | |
808 | } else { | |
809 | sid = rte_lcore_to_socket_id(rte_get_master_lcore()); | |
810 | } | |
811 | ||
d46285a2 | 812 | dev->socket_id = sid < 0 ? SOCKET0 : sid; |
db8f13b0 | 813 | dev->requested_socket_id = dev->socket_id; |
d46285a2 DDP |
814 | dev->port_id = port_no; |
815 | dev->type = type; | |
816 | dev->flags = 0; | |
0072e931 | 817 | dev->requested_mtu = dev->mtu = ETHER_MTU; |
d46285a2 | 818 | dev->max_packet_len = MTU_TO_FRAME_LEN(dev->mtu); |
0a0f39df CL |
819 | ovsrcu_index_init(&dev->vid, -1); |
820 | dev->vhost_reconfigured = false; | |
8a9562d2 | 821 | |
0072e931 MK |
822 | err = netdev_dpdk_mempool_configure(dev); |
823 | if (err) { | |
95fb793a | 824 | goto unlock; |
8a9562d2 PS |
825 | } |
826 | ||
0bf765f7 | 827 | /* Initialise QoS configuration to NULL and qos lock to unlocked */ |
d46285a2 DDP |
828 | dev->qos_conf = NULL; |
829 | rte_spinlock_init(&dev->qos_lock); | |
0bf765f7 | 830 | |
9509913a IS |
831 | /* Initialise rcu pointer for ingress policer to NULL */ |
832 | ovsrcu_init(&dev->ingress_policer, NULL); | |
833 | dev->policer_rate = 0; | |
834 | dev->policer_burst = 0; | |
835 | ||
d46285a2 | 836 | netdev->n_rxq = NR_QUEUE; |
81acebda IM |
837 | netdev->n_txq = NR_QUEUE; |
838 | dev->requested_n_rxq = netdev->n_rxq; | |
839 | dev->requested_n_txq = netdev->n_txq; | |
58397e6c | 840 | |
9fd39370 SC |
841 | /* Initialize the flow control to NULL */ |
842 | memset(&dev->fc_conf, 0, sizeof dev->fc_conf); | |
58397e6c | 843 | if (type == DPDK_DEV_ETH) { |
d46285a2 | 844 | err = dpdk_eth_dev_init(dev); |
1b99bb05 MG |
845 | if (err) { |
846 | goto unlock; | |
847 | } | |
81acebda | 848 | netdev_dpdk_alloc_txq(dev, netdev->n_txq); |
585a5bea | 849 | } else { |
d46285a2 | 850 | netdev_dpdk_alloc_txq(dev, OVS_VHOST_MAX_QUEUE_NUM); |
e543851d ZB |
851 | /* Enable DPDK_DEV_VHOST device and set promiscuous mode flag. */ |
852 | dev->flags = NETDEV_UP | NETDEV_PROMISC; | |
8a9562d2 | 853 | } |
8a9562d2 | 854 | |
d46285a2 | 855 | ovs_list_push_back(&dpdk_list, &dev->list_node); |
8a9562d2 | 856 | |
95fb793a | 857 | unlock: |
d46285a2 | 858 | ovs_mutex_unlock(&dev->mutex); |
95fb793a | 859 | return err; |
860 | } | |
861 | ||
b83a2df1 MV |
862 | /* dev_name must be the prefix followed by a positive decimal number. |
863 | * (no leading + or - signs are allowed) */ | |
95fb793a | 864 | static int |
865 | dpdk_dev_parse_name(const char dev_name[], const char prefix[], | |
866 | unsigned int *port_no) | |
867 | { | |
868 | const char *cport; | |
869 | ||
870 | if (strncmp(dev_name, prefix, strlen(prefix))) { | |
871 | return ENODEV; | |
872 | } | |
873 | ||
874 | cport = dev_name + strlen(prefix); | |
b83a2df1 MV |
875 | |
876 | if (str_to_uint(cport, 10, port_no)) { | |
877 | return 0; | |
878 | } else { | |
879 | return ENODEV; | |
880 | } | |
95fb793a | 881 | } |
882 | ||
7d1ced01 | 883 | static int |
53f50d24 | 884 | netdev_dpdk_vhost_construct(struct netdev *netdev) |
7d1ced01 | 885 | { |
d46285a2 DDP |
886 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
887 | const char *name = netdev->name; | |
7d1ced01 | 888 | int err; |
a0cb2d66 | 889 | |
1af27e8a DDP |
890 | /* 'name' is appended to 'vhost_sock_dir' and used to create a socket in |
891 | * the file system. '/' or '\' would traverse directories, so they're not | |
892 | * acceptable in 'name'. */ | |
893 | if (strchr(name, '/') || strchr(name, '\\')) { | |
894 | VLOG_ERR("\"%s\" is not a valid name for a vhost-user port. " | |
895 | "A valid name must not include '/' or '\\'", | |
896 | name); | |
897 | return EINVAL; | |
898 | } | |
899 | ||
bab69409 AC |
900 | if (rte_eal_init_ret) { |
901 | return rte_eal_init_ret; | |
902 | } | |
903 | ||
7d1ced01 CL |
904 | ovs_mutex_lock(&dpdk_mutex); |
905 | /* Take the name of the vhost-user port and append it to the location where | |
2d24d165 | 906 | * the socket is to be created, then register the socket. |
7d1ced01 | 907 | */ |
2d24d165 | 908 | snprintf(dev->vhost_id, sizeof dev->vhost_id, "%s/%s", |
1af27e8a DDP |
909 | vhost_sock_dir, name); |
910 | ||
2d24d165 CL |
911 | dev->vhost_driver_flags &= ~RTE_VHOST_USER_CLIENT; |
912 | err = rte_vhost_driver_register(dev->vhost_id, dev->vhost_driver_flags); | |
7d1ced01 CL |
913 | if (err) { |
914 | VLOG_ERR("vhost-user socket device setup failure for socket %s\n", | |
2d24d165 | 915 | dev->vhost_id); |
e5c0f5a4 | 916 | } else { |
2d24d165 CL |
917 | fatal_signal_add_file_to_unlink(dev->vhost_id); |
918 | VLOG_INFO("Socket %s created for vhost-user port %s\n", | |
919 | dev->vhost_id, name); | |
920 | } | |
921 | err = netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); | |
922 | ||
923 | ovs_mutex_unlock(&dpdk_mutex); | |
924 | return err; | |
925 | } | |
926 | ||
927 | static int | |
928 | netdev_dpdk_vhost_client_construct(struct netdev *netdev) | |
929 | { | |
930 | int err; | |
931 | ||
932 | if (rte_eal_init_ret) { | |
933 | return rte_eal_init_ret; | |
7d1ced01 | 934 | } |
e5c0f5a4 | 935 | |
2d24d165 CL |
936 | ovs_mutex_lock(&dpdk_mutex); |
937 | err = netdev_dpdk_init(netdev, -1, DPDK_DEV_VHOST); | |
7d1ced01 | 938 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c KT |
939 | return err; |
940 | } | |
941 | ||
95fb793a | 942 | static int |
943 | netdev_dpdk_construct(struct netdev *netdev) | |
944 | { | |
945 | unsigned int port_no; | |
946 | int err; | |
947 | ||
948 | if (rte_eal_init_ret) { | |
949 | return rte_eal_init_ret; | |
950 | } | |
951 | ||
952 | /* Names always start with "dpdk" */ | |
953 | err = dpdk_dev_parse_name(netdev->name, "dpdk", &port_no); | |
954 | if (err) { | |
955 | return err; | |
956 | } | |
957 | ||
958 | ovs_mutex_lock(&dpdk_mutex); | |
58397e6c | 959 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
8a9562d2 PS |
960 | ovs_mutex_unlock(&dpdk_mutex); |
961 | return err; | |
962 | } | |
963 | ||
964 | static void | |
d46285a2 | 965 | netdev_dpdk_destruct(struct netdev *netdev) |
8a9562d2 | 966 | { |
d46285a2 | 967 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 968 | |
8d38823b | 969 | ovs_mutex_lock(&dpdk_mutex); |
8a9562d2 | 970 | ovs_mutex_lock(&dev->mutex); |
8d38823b | 971 | |
8a9562d2 | 972 | rte_eth_dev_stop(dev->port_id); |
9509913a IS |
973 | free(ovsrcu_get_protected(struct ingress_policer *, |
974 | &dev->ingress_policer)); | |
8a9562d2 | 975 | |
5a034064 | 976 | rte_free(dev->tx_q); |
417e7e66 | 977 | ovs_list_remove(&dev->list_node); |
8a9562d2 | 978 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b MK |
979 | |
980 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 | 981 | ovs_mutex_unlock(&dpdk_mutex); |
58397e6c | 982 | } |
8a9562d2 | 983 | |
3f891bbe DDP |
984 | /* rte_vhost_driver_unregister() can call back destroy_device(), which will |
985 | * try to acquire 'dpdk_mutex' and possibly 'dev->mutex'. To avoid a | |
986 | * deadlock, none of the mutexes must be held while calling this function. */ | |
987 | static int | |
c1ff66ac CL |
988 | dpdk_vhost_driver_unregister(struct netdev_dpdk *dev OVS_UNUSED, |
989 | char *vhost_id) | |
3f891bbe DDP |
990 | OVS_EXCLUDED(dpdk_mutex) |
991 | OVS_EXCLUDED(dev->mutex) | |
992 | { | |
c1ff66ac | 993 | return rte_vhost_driver_unregister(vhost_id); |
3f891bbe DDP |
994 | } |
995 | ||
58397e6c | 996 | static void |
d46285a2 | 997 | netdev_dpdk_vhost_destruct(struct netdev *netdev) |
58397e6c | 998 | { |
d46285a2 | 999 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
c1ff66ac | 1000 | char *vhost_id; |
58397e6c | 1001 | |
8d38823b MK |
1002 | ovs_mutex_lock(&dpdk_mutex); |
1003 | ovs_mutex_lock(&dev->mutex); | |
1004 | ||
c62da695 | 1005 | /* Guest becomes an orphan if still attached. */ |
c1ff66ac CL |
1006 | if (netdev_dpdk_get_vid(dev) >= 0 |
1007 | && !(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { | |
c62da695 | 1008 | VLOG_ERR("Removing port '%s' while vhost device still attached.", |
d46285a2 | 1009 | netdev->name); |
c62da695 | 1010 | VLOG_ERR("To restore connectivity after re-adding of port, VM on socket" |
2d24d165 | 1011 | " '%s' must be restarted.", dev->vhost_id); |
58397e6c KT |
1012 | } |
1013 | ||
9509913a IS |
1014 | free(ovsrcu_get_protected(struct ingress_policer *, |
1015 | &dev->ingress_policer)); | |
9509913a | 1016 | |
6019cb63 | 1017 | rte_free(dev->tx_q); |
417e7e66 | 1018 | ovs_list_remove(&dev->list_node); |
58397e6c | 1019 | dpdk_mp_put(dev->dpdk_mp); |
8d38823b | 1020 | |
2d24d165 | 1021 | vhost_id = xstrdup(dev->vhost_id); |
c1ff66ac | 1022 | |
8d38823b | 1023 | ovs_mutex_unlock(&dev->mutex); |
58397e6c | 1024 | ovs_mutex_unlock(&dpdk_mutex); |
3f891bbe | 1025 | |
c1ff66ac | 1026 | if (dpdk_vhost_driver_unregister(dev, vhost_id)) { |
41964543 IM |
1027 | VLOG_ERR("%s: Unable to unregister vhost driver for socket '%s'.\n", |
1028 | netdev->name, vhost_id); | |
c1ff66ac CL |
1029 | } else if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1030 | /* OVS server mode - remove this socket from list for deletion */ | |
1031 | fatal_signal_remove_file_to_unlink(vhost_id); | |
3f891bbe | 1032 | } |
c1ff66ac | 1033 | free(vhost_id); |
8a9562d2 PS |
1034 | } |
1035 | ||
1036 | static void | |
d46285a2 | 1037 | netdev_dpdk_dealloc(struct netdev *netdev) |
8a9562d2 | 1038 | { |
d46285a2 | 1039 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 | 1040 | |
d46285a2 | 1041 | rte_free(dev); |
8a9562d2 PS |
1042 | } |
1043 | ||
1044 | static int | |
a14b8947 | 1045 | netdev_dpdk_get_config(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 1046 | { |
a14b8947 | 1047 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1048 | |
1049 | ovs_mutex_lock(&dev->mutex); | |
1050 | ||
050c60bf | 1051 | smap_add_format(args, "requested_rx_queues", "%d", dev->requested_n_rxq); |
a14b8947 | 1052 | smap_add_format(args, "configured_rx_queues", "%d", netdev->n_rxq); |
81acebda IM |
1053 | smap_add_format(args, "requested_tx_queues", "%d", dev->requested_n_txq); |
1054 | smap_add_format(args, "configured_tx_queues", "%d", netdev->n_txq); | |
0072e931 | 1055 | smap_add_format(args, "mtu", "%d", dev->mtu); |
8a9562d2 PS |
1056 | ovs_mutex_unlock(&dev->mutex); |
1057 | ||
1058 | return 0; | |
1059 | } | |
1060 | ||
c3d062a7 CL |
1061 | static void |
1062 | dpdk_set_rxq_config(struct netdev_dpdk *dev, const struct smap *args) | |
a14b8947 | 1063 | { |
050c60bf | 1064 | int new_n_rxq; |
a14b8947 | 1065 | |
050c60bf DDP |
1066 | new_n_rxq = MAX(smap_get_int(args, "n_rxq", dev->requested_n_rxq), 1); |
1067 | if (new_n_rxq != dev->requested_n_rxq) { | |
1068 | dev->requested_n_rxq = new_n_rxq; | |
c3d062a7 | 1069 | netdev_request_reconfigure(&dev->up); |
050c60bf | 1070 | } |
c3d062a7 CL |
1071 | } |
1072 | ||
1073 | static int | |
1074 | netdev_dpdk_set_config(struct netdev *netdev, const struct smap *args) | |
1075 | { | |
1076 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1077 | ||
1078 | ovs_mutex_lock(&dev->mutex); | |
1079 | ||
1080 | dpdk_set_rxq_config(dev, args); | |
1081 | ||
1082 | /* Flow control support is only available for DPDK Ethernet ports. */ | |
1083 | bool rx_fc_en = false; | |
1084 | bool tx_fc_en = false; | |
1085 | enum rte_eth_fc_mode fc_mode_set[2][2] = | |
1086 | {{RTE_FC_NONE, RTE_FC_TX_PAUSE}, | |
1087 | {RTE_FC_RX_PAUSE, RTE_FC_FULL} | |
1088 | }; | |
1089 | rx_fc_en = smap_get_bool(args, "rx-flow-ctrl", false); | |
1090 | tx_fc_en = smap_get_bool(args, "tx-flow-ctrl", false); | |
1091 | dev->fc_conf.autoneg = smap_get_bool(args, "flow-ctrl-autoneg", false); | |
1092 | dev->fc_conf.mode = fc_mode_set[tx_fc_en][rx_fc_en]; | |
1093 | ||
1094 | dpdk_eth_flow_ctrl_setup(dev); | |
9fd39370 | 1095 | |
c3d062a7 CL |
1096 | ovs_mutex_unlock(&dev->mutex); |
1097 | ||
1098 | return 0; | |
1099 | } | |
1100 | ||
1101 | static int | |
1102 | netdev_dpdk_ring_set_config(struct netdev *netdev, const struct smap *args) | |
1103 | { | |
1104 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1105 | ||
1106 | ovs_mutex_lock(&dev->mutex); | |
1107 | dpdk_set_rxq_config(dev, args); | |
a14b8947 IM |
1108 | ovs_mutex_unlock(&dev->mutex); |
1109 | ||
1110 | return 0; | |
1111 | } | |
1112 | ||
c1ff66ac | 1113 | static int |
2d24d165 CL |
1114 | netdev_dpdk_vhost_client_set_config(struct netdev *netdev, |
1115 | const struct smap *args) | |
c1ff66ac CL |
1116 | { |
1117 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1118 | const char *path; | |
1119 | ||
6881885a | 1120 | ovs_mutex_lock(&dev->mutex); |
c1ff66ac CL |
1121 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT)) { |
1122 | path = smap_get(args, "vhost-server-path"); | |
2d24d165 CL |
1123 | if (path && strcmp(path, dev->vhost_id)) { |
1124 | strcpy(dev->vhost_id, path); | |
c1ff66ac CL |
1125 | netdev_request_reconfigure(netdev); |
1126 | } | |
1127 | } | |
6881885a | 1128 | ovs_mutex_unlock(&dev->mutex); |
c1ff66ac CL |
1129 | |
1130 | return 0; | |
1131 | } | |
1132 | ||
7dec44fe | 1133 | static int |
d46285a2 | 1134 | netdev_dpdk_get_numa_id(const struct netdev *netdev) |
7dec44fe | 1135 | { |
d46285a2 | 1136 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
7dec44fe | 1137 | |
d46285a2 | 1138 | return dev->socket_id; |
7dec44fe AW |
1139 | } |
1140 | ||
050c60bf | 1141 | /* Sets the number of tx queues for the dpdk interface. */ |
5496878c | 1142 | static int |
050c60bf | 1143 | netdev_dpdk_set_tx_multiq(struct netdev *netdev, unsigned int n_txq) |
5496878c | 1144 | { |
d46285a2 | 1145 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
5496878c | 1146 | |
d46285a2 | 1147 | ovs_mutex_lock(&dev->mutex); |
91968eb0 | 1148 | |
050c60bf DDP |
1149 | if (dev->requested_n_txq == n_txq) { |
1150 | goto out; | |
4573fbd3 FL |
1151 | } |
1152 | ||
050c60bf DDP |
1153 | dev->requested_n_txq = n_txq; |
1154 | netdev_request_reconfigure(netdev); | |
58397e6c | 1155 | |
050c60bf | 1156 | out: |
d46285a2 | 1157 | ovs_mutex_unlock(&dev->mutex); |
050c60bf | 1158 | return 0; |
58397e6c KT |
1159 | } |
1160 | ||
8a9562d2 PS |
1161 | static struct netdev_rxq * |
1162 | netdev_dpdk_rxq_alloc(void) | |
1163 | { | |
1164 | struct netdev_rxq_dpdk *rx = dpdk_rte_mzalloc(sizeof *rx); | |
1165 | ||
1166 | return &rx->up; | |
1167 | } | |
1168 | ||
1169 | static struct netdev_rxq_dpdk * | |
d46285a2 | 1170 | netdev_rxq_dpdk_cast(const struct netdev_rxq *rxq) |
8a9562d2 | 1171 | { |
d46285a2 | 1172 | return CONTAINER_OF(rxq, struct netdev_rxq_dpdk, up); |
8a9562d2 PS |
1173 | } |
1174 | ||
1175 | static int | |
d46285a2 | 1176 | netdev_dpdk_rxq_construct(struct netdev_rxq *rxq) |
8a9562d2 | 1177 | { |
d46285a2 DDP |
1178 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1179 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
8a9562d2 | 1180 | |
d46285a2 DDP |
1181 | ovs_mutex_lock(&dev->mutex); |
1182 | rx->port_id = dev->port_id; | |
1183 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
1184 | |
1185 | return 0; | |
1186 | } | |
1187 | ||
1188 | static void | |
d46285a2 | 1189 | netdev_dpdk_rxq_destruct(struct netdev_rxq *rxq OVS_UNUSED) |
8a9562d2 PS |
1190 | { |
1191 | } | |
1192 | ||
1193 | static void | |
d46285a2 | 1194 | netdev_dpdk_rxq_dealloc(struct netdev_rxq *rxq) |
8a9562d2 | 1195 | { |
d46285a2 | 1196 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
8a9562d2 PS |
1197 | |
1198 | rte_free(rx); | |
1199 | } | |
1200 | ||
b170db2a | 1201 | static inline void |
b59cc14e IM |
1202 | netdev_dpdk_eth_tx_burst(struct netdev_dpdk *dev, int qid, |
1203 | struct rte_mbuf **pkts, int cnt) | |
8a9562d2 | 1204 | { |
1304f1f8 DDP |
1205 | uint32_t nb_tx = 0; |
1206 | ||
b59cc14e | 1207 | while (nb_tx != cnt) { |
1304f1f8 DDP |
1208 | uint32_t ret; |
1209 | ||
b59cc14e | 1210 | ret = rte_eth_tx_burst(dev->port_id, qid, pkts + nb_tx, cnt - nb_tx); |
1304f1f8 DDP |
1211 | if (!ret) { |
1212 | break; | |
1213 | } | |
1214 | ||
1215 | nb_tx += ret; | |
1216 | } | |
8a9562d2 | 1217 | |
b59cc14e | 1218 | if (OVS_UNLIKELY(nb_tx != cnt)) { |
db73f716 DDP |
1219 | /* free buffers, which we couldn't transmit, one at a time (each |
1220 | * packet could come from a different mempool) */ | |
1221 | int i; | |
1222 | ||
b59cc14e IM |
1223 | for (i = nb_tx; i < cnt; i++) { |
1224 | rte_pktmbuf_free(pkts[i]); | |
db73f716 | 1225 | } |
45d947c4 | 1226 | rte_spinlock_lock(&dev->stats_lock); |
b59cc14e | 1227 | dev->stats.tx_dropped += cnt - nb_tx; |
45d947c4 | 1228 | rte_spinlock_unlock(&dev->stats_lock); |
8a9562d2 | 1229 | } |
8a9562d2 PS |
1230 | } |
1231 | ||
f3926f29 IS |
1232 | static inline bool |
1233 | netdev_dpdk_policer_pkt_handle(struct rte_meter_srtcm *meter, | |
1234 | struct rte_mbuf *pkt, uint64_t time) | |
1235 | { | |
1236 | uint32_t pkt_len = rte_pktmbuf_pkt_len(pkt) - sizeof(struct ether_hdr); | |
1237 | ||
1238 | return rte_meter_srtcm_color_blind_check(meter, time, pkt_len) == | |
1239 | e_RTE_METER_GREEN; | |
1240 | } | |
1241 | ||
1242 | static int | |
1243 | netdev_dpdk_policer_run(struct rte_meter_srtcm *meter, | |
1244 | struct rte_mbuf **pkts, int pkt_cnt) | |
1245 | { | |
1246 | int i = 0; | |
1247 | int cnt = 0; | |
1248 | struct rte_mbuf *pkt = NULL; | |
1249 | uint64_t current_time = rte_rdtsc(); | |
1250 | ||
1251 | for (i = 0; i < pkt_cnt; i++) { | |
1252 | pkt = pkts[i]; | |
1253 | /* Handle current packet */ | |
1254 | if (netdev_dpdk_policer_pkt_handle(meter, pkt, current_time)) { | |
1255 | if (cnt != i) { | |
1256 | pkts[cnt] = pkt; | |
1257 | } | |
1258 | cnt++; | |
1259 | } else { | |
1260 | rte_pktmbuf_free(pkt); | |
1261 | } | |
1262 | } | |
1263 | ||
1264 | return cnt; | |
1265 | } | |
1266 | ||
9509913a IS |
1267 | static int |
1268 | ingress_policer_run(struct ingress_policer *policer, struct rte_mbuf **pkts, | |
1269 | int pkt_cnt) | |
1270 | { | |
1271 | int cnt = 0; | |
1272 | ||
1273 | rte_spinlock_lock(&policer->policer_lock); | |
1274 | cnt = netdev_dpdk_policer_run(&policer->in_policer, pkts, pkt_cnt); | |
1275 | rte_spinlock_unlock(&policer->policer_lock); | |
1276 | ||
1277 | return cnt; | |
1278 | } | |
1279 | ||
58397e6c | 1280 | static bool |
0a0f39df | 1281 | is_vhost_running(struct netdev_dpdk *dev) |
58397e6c | 1282 | { |
0a0f39df | 1283 | return (netdev_dpdk_get_vid(dev) >= 0 && dev->vhost_reconfigured); |
58397e6c KT |
1284 | } |
1285 | ||
d6e3feb5 | 1286 | static inline void |
1287 | netdev_dpdk_vhost_update_rx_size_counters(struct netdev_stats *stats, | |
1288 | unsigned int packet_size) | |
1289 | { | |
1290 | /* Hard-coded search for the size bucket. */ | |
1291 | if (packet_size < 256) { | |
1292 | if (packet_size >= 128) { | |
1293 | stats->rx_128_to_255_packets++; | |
1294 | } else if (packet_size <= 64) { | |
1295 | stats->rx_1_to_64_packets++; | |
1296 | } else { | |
1297 | stats->rx_65_to_127_packets++; | |
1298 | } | |
1299 | } else { | |
1300 | if (packet_size >= 1523) { | |
1301 | stats->rx_1523_to_max_packets++; | |
1302 | } else if (packet_size >= 1024) { | |
1303 | stats->rx_1024_to_1522_packets++; | |
1304 | } else if (packet_size < 512) { | |
1305 | stats->rx_256_to_511_packets++; | |
1306 | } else { | |
1307 | stats->rx_512_to_1023_packets++; | |
1308 | } | |
1309 | } | |
1310 | } | |
1311 | ||
9e3ddd45 TP |
1312 | static inline void |
1313 | netdev_dpdk_vhost_update_rx_counters(struct netdev_stats *stats, | |
9509913a IS |
1314 | struct dp_packet **packets, int count, |
1315 | int dropped) | |
9e3ddd45 TP |
1316 | { |
1317 | int i; | |
d6e3feb5 | 1318 | unsigned int packet_size; |
9e3ddd45 TP |
1319 | struct dp_packet *packet; |
1320 | ||
1321 | stats->rx_packets += count; | |
9509913a | 1322 | stats->rx_dropped += dropped; |
9e3ddd45 TP |
1323 | for (i = 0; i < count; i++) { |
1324 | packet = packets[i]; | |
d6e3feb5 | 1325 | packet_size = dp_packet_size(packet); |
9e3ddd45 | 1326 | |
d6e3feb5 | 1327 | if (OVS_UNLIKELY(packet_size < ETH_HEADER_LEN)) { |
9e3ddd45 TP |
1328 | /* This only protects the following multicast counting from |
1329 | * too short packets, but it does not stop the packet from | |
1330 | * further processing. */ | |
1331 | stats->rx_errors++; | |
1332 | stats->rx_length_errors++; | |
1333 | continue; | |
1334 | } | |
1335 | ||
d6e3feb5 | 1336 | netdev_dpdk_vhost_update_rx_size_counters(stats, packet_size); |
1337 | ||
9e3ddd45 TP |
1338 | struct eth_header *eh = (struct eth_header *) dp_packet_data(packet); |
1339 | if (OVS_UNLIKELY(eth_addr_is_multicast(eh->eth_dst))) { | |
1340 | stats->multicast++; | |
1341 | } | |
1342 | ||
d6e3feb5 | 1343 | stats->rx_bytes += packet_size; |
9e3ddd45 TP |
1344 | } |
1345 | } | |
1346 | ||
58397e6c KT |
1347 | /* |
1348 | * The receive path for the vhost port is the TX path out from guest. | |
1349 | */ | |
1350 | static int | |
d46285a2 | 1351 | netdev_dpdk_vhost_rxq_recv(struct netdev_rxq *rxq, |
64839cf4 | 1352 | struct dp_packet_batch *batch) |
58397e6c | 1353 | { |
d46285a2 | 1354 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); |
d46285a2 | 1355 | int qid = rxq->queue_id; |
9509913a | 1356 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
58397e6c | 1357 | uint16_t nb_rx = 0; |
9509913a | 1358 | uint16_t dropped = 0; |
58397e6c | 1359 | |
0a0f39df | 1360 | if (OVS_UNLIKELY(!is_vhost_running(dev) |
e543851d | 1361 | || !(dev->flags & NETDEV_UP))) { |
58397e6c KT |
1362 | return EAGAIN; |
1363 | } | |
1364 | ||
0a0f39df CL |
1365 | nb_rx = rte_vhost_dequeue_burst(netdev_dpdk_get_vid(dev), |
1366 | qid * VIRTIO_QNUM + VIRTIO_TXQ, | |
d46285a2 | 1367 | dev->dpdk_mp->mp, |
64839cf4 | 1368 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1369 | NETDEV_MAX_BURST); |
58397e6c KT |
1370 | if (!nb_rx) { |
1371 | return EAGAIN; | |
1372 | } | |
1373 | ||
9509913a IS |
1374 | if (policer) { |
1375 | dropped = nb_rx; | |
64839cf4 WT |
1376 | nb_rx = ingress_policer_run(policer, |
1377 | (struct rte_mbuf **) batch->packets, | |
1378 | nb_rx); | |
9509913a IS |
1379 | dropped -= nb_rx; |
1380 | } | |
1381 | ||
d46285a2 | 1382 | rte_spinlock_lock(&dev->stats_lock); |
64839cf4 WT |
1383 | netdev_dpdk_vhost_update_rx_counters(&dev->stats, batch->packets, |
1384 | nb_rx, dropped); | |
d46285a2 | 1385 | rte_spinlock_unlock(&dev->stats_lock); |
45d947c4 | 1386 | |
64839cf4 | 1387 | batch->count = (int) nb_rx; |
58397e6c KT |
1388 | return 0; |
1389 | } | |
1390 | ||
8a9562d2 | 1391 | static int |
64839cf4 | 1392 | netdev_dpdk_rxq_recv(struct netdev_rxq *rxq, struct dp_packet_batch *batch) |
8a9562d2 | 1393 | { |
d46285a2 DDP |
1394 | struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq); |
1395 | struct netdev_dpdk *dev = netdev_dpdk_cast(rxq->netdev); | |
9509913a | 1396 | struct ingress_policer *policer = netdev_dpdk_get_ingress_policer(dev); |
8a9562d2 | 1397 | int nb_rx; |
9509913a | 1398 | int dropped = 0; |
8a9562d2 | 1399 | |
d46285a2 | 1400 | nb_rx = rte_eth_rx_burst(rx->port_id, rxq->queue_id, |
64839cf4 | 1401 | (struct rte_mbuf **) batch->packets, |
cd159f1a | 1402 | NETDEV_MAX_BURST); |
8a9562d2 PS |
1403 | if (!nb_rx) { |
1404 | return EAGAIN; | |
1405 | } | |
1406 | ||
9509913a IS |
1407 | if (policer) { |
1408 | dropped = nb_rx; | |
64839cf4 WT |
1409 | nb_rx = ingress_policer_run(policer, |
1410 | (struct rte_mbuf **)batch->packets, | |
1411 | nb_rx); | |
9509913a IS |
1412 | dropped -= nb_rx; |
1413 | } | |
1414 | ||
1415 | /* Update stats to reflect dropped packets */ | |
1416 | if (OVS_UNLIKELY(dropped)) { | |
1417 | rte_spinlock_lock(&dev->stats_lock); | |
1418 | dev->stats.rx_dropped += dropped; | |
1419 | rte_spinlock_unlock(&dev->stats_lock); | |
1420 | } | |
1421 | ||
64839cf4 | 1422 | batch->count = nb_rx; |
8a9562d2 PS |
1423 | |
1424 | return 0; | |
1425 | } | |
1426 | ||
0bf765f7 IS |
1427 | static inline int |
1428 | netdev_dpdk_qos_run__(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1429 | int cnt) | |
1430 | { | |
1431 | struct netdev *netdev = &dev->up; | |
1432 | ||
1433 | if (dev->qos_conf != NULL) { | |
1434 | rte_spinlock_lock(&dev->qos_lock); | |
1435 | if (dev->qos_conf != NULL) { | |
1436 | cnt = dev->qos_conf->ops->qos_run(netdev, pkts, cnt); | |
1437 | } | |
1438 | rte_spinlock_unlock(&dev->qos_lock); | |
1439 | } | |
1440 | ||
1441 | return cnt; | |
1442 | } | |
1443 | ||
c6ec9d17 IM |
1444 | static int |
1445 | netdev_dpdk_filter_packet_len(struct netdev_dpdk *dev, struct rte_mbuf **pkts, | |
1446 | int pkt_cnt) | |
1447 | { | |
1448 | int i = 0; | |
1449 | int cnt = 0; | |
1450 | struct rte_mbuf *pkt; | |
1451 | ||
1452 | for (i = 0; i < pkt_cnt; i++) { | |
1453 | pkt = pkts[i]; | |
1454 | if (OVS_UNLIKELY(pkt->pkt_len > dev->max_packet_len)) { | |
1455 | VLOG_WARN_RL(&rl, "%s: Too big size %" PRIu32 " max_packet_len %d", | |
1456 | dev->up.name, pkt->pkt_len, dev->max_packet_len); | |
1457 | rte_pktmbuf_free(pkt); | |
1458 | continue; | |
1459 | } | |
1460 | ||
1461 | if (OVS_UNLIKELY(i != cnt)) { | |
1462 | pkts[cnt] = pkt; | |
1463 | } | |
1464 | cnt++; | |
1465 | } | |
1466 | ||
1467 | return cnt; | |
1468 | } | |
1469 | ||
9e3ddd45 TP |
1470 | static inline void |
1471 | netdev_dpdk_vhost_update_tx_counters(struct netdev_stats *stats, | |
1472 | struct dp_packet **packets, | |
1473 | int attempted, | |
1474 | int dropped) | |
1475 | { | |
1476 | int i; | |
1477 | int sent = attempted - dropped; | |
1478 | ||
1479 | stats->tx_packets += sent; | |
1480 | stats->tx_dropped += dropped; | |
1481 | ||
1482 | for (i = 0; i < sent; i++) { | |
1483 | stats->tx_bytes += dp_packet_size(packets[i]); | |
1484 | } | |
1485 | } | |
1486 | ||
58397e6c | 1487 | static void |
4573fbd3 | 1488 | __netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
dd52de45 | 1489 | struct dp_packet **pkts, int cnt) |
58397e6c | 1490 | { |
d46285a2 | 1491 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
95e9881f KT |
1492 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) pkts; |
1493 | unsigned int total_pkts = cnt; | |
c6ec9d17 | 1494 | unsigned int dropped = 0; |
dd52de45 | 1495 | int i, retries = 0; |
58397e6c | 1496 | |
81acebda | 1497 | qid = dev->tx_q[qid % netdev->n_txq].map; |
585a5bea | 1498 | |
0a0f39df | 1499 | if (OVS_UNLIKELY(!is_vhost_running(dev) || qid < 0 |
e543851d | 1500 | || !(dev->flags & NETDEV_UP))) { |
d46285a2 DDP |
1501 | rte_spinlock_lock(&dev->stats_lock); |
1502 | dev->stats.tx_dropped+= cnt; | |
1503 | rte_spinlock_unlock(&dev->stats_lock); | |
1b99bb05 | 1504 | goto out; |
58397e6c KT |
1505 | } |
1506 | ||
d46285a2 | 1507 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
58397e6c | 1508 | |
c6ec9d17 | 1509 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
0bf765f7 | 1510 | /* Check has QoS has been configured for the netdev */ |
d46285a2 | 1511 | cnt = netdev_dpdk_qos_run__(dev, cur_pkts, cnt); |
c6ec9d17 | 1512 | dropped = total_pkts - cnt; |
0bf765f7 | 1513 | |
95e9881f | 1514 | do { |
4573fbd3 | 1515 | int vhost_qid = qid * VIRTIO_QNUM + VIRTIO_RXQ; |
95e9881f KT |
1516 | unsigned int tx_pkts; |
1517 | ||
0a0f39df | 1518 | tx_pkts = rte_vhost_enqueue_burst(netdev_dpdk_get_vid(dev), |
c6ec9d17 | 1519 | vhost_qid, cur_pkts, cnt); |
95e9881f KT |
1520 | if (OVS_LIKELY(tx_pkts)) { |
1521 | /* Packets have been sent.*/ | |
1522 | cnt -= tx_pkts; | |
31871ee3 | 1523 | /* Prepare for possible retry.*/ |
95e9881f KT |
1524 | cur_pkts = &cur_pkts[tx_pkts]; |
1525 | } else { | |
31871ee3 KT |
1526 | /* No packets sent - do not retry.*/ |
1527 | break; | |
95e9881f | 1528 | } |
c6ec9d17 | 1529 | } while (cnt && (retries++ <= VHOST_ENQ_RETRY_NUM)); |
4573fbd3 | 1530 | |
d46285a2 | 1531 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
95e9881f | 1532 | |
d46285a2 | 1533 | rte_spinlock_lock(&dev->stats_lock); |
0072e931 | 1534 | netdev_dpdk_vhost_update_tx_counters(&dev->stats, pkts, total_pkts, |
c6ec9d17 | 1535 | cnt + dropped); |
d46285a2 | 1536 | rte_spinlock_unlock(&dev->stats_lock); |
58397e6c KT |
1537 | |
1538 | out: | |
c6ec9d17 | 1539 | for (i = 0; i < total_pkts - dropped; i++) { |
dd52de45 | 1540 | dp_packet_delete(pkts[i]); |
58397e6c KT |
1541 | } |
1542 | } | |
1543 | ||
8a9562d2 PS |
1544 | /* Tx function. Transmit packets indefinitely */ |
1545 | static void | |
64839cf4 | 1546 | dpdk_do_tx_copy(struct netdev *netdev, int qid, struct dp_packet_batch *batch) |
db73f716 | 1547 | OVS_NO_THREAD_SAFETY_ANALYSIS |
8a9562d2 | 1548 | { |
bce01e3a | 1549 | #if !defined(__CHECKER__) && !defined(_WIN32) |
64839cf4 | 1550 | const size_t PKT_ARRAY_SIZE = batch->count; |
bce01e3a EJ |
1551 | #else |
1552 | /* Sparse or MSVC doesn't like variable length array. */ | |
cd159f1a | 1553 | enum { PKT_ARRAY_SIZE = NETDEV_MAX_BURST }; |
bce01e3a | 1554 | #endif |
8a9562d2 | 1555 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
bce01e3a | 1556 | struct rte_mbuf *mbufs[PKT_ARRAY_SIZE]; |
175cf4de RW |
1557 | int dropped = 0; |
1558 | int newcnt = 0; | |
1559 | int i; | |
8a9562d2 | 1560 | |
db73f716 DDP |
1561 | /* If we are on a non pmd thread we have to use the mempool mutex, because |
1562 | * every non pmd thread shares the same mempool cache */ | |
1563 | ||
5f17de68 | 1564 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1565 | ovs_mutex_lock(&nonpmd_mempool_mutex); |
1566 | } | |
1567 | ||
7d6d1a40 WT |
1568 | dp_packet_batch_apply_cutlen(batch); |
1569 | ||
64839cf4 WT |
1570 | for (i = 0; i < batch->count; i++) { |
1571 | int size = dp_packet_size(batch->packets[i]); | |
95fb793a | 1572 | |
f98d7864 | 1573 | if (OVS_UNLIKELY(size > dev->max_packet_len)) { |
f4fd623c DDP |
1574 | VLOG_WARN_RL(&rl, "Too big size %d max_packet_len %d", |
1575 | (int)size , dev->max_packet_len); | |
1576 | ||
175cf4de | 1577 | dropped++; |
f4fd623c DDP |
1578 | continue; |
1579 | } | |
8a9562d2 | 1580 | |
f4fd623c | 1581 | mbufs[newcnt] = rte_pktmbuf_alloc(dev->dpdk_mp->mp); |
8a9562d2 | 1582 | |
f4fd623c | 1583 | if (!mbufs[newcnt]) { |
64839cf4 | 1584 | dropped += batch->count - i; |
175cf4de | 1585 | break; |
f4fd623c DDP |
1586 | } |
1587 | ||
1588 | /* We have to do a copy for now */ | |
64839cf4 WT |
1589 | memcpy(rte_pktmbuf_mtod(mbufs[newcnt], void *), |
1590 | dp_packet_data(batch->packets[i]), size); | |
f4fd623c DDP |
1591 | |
1592 | rte_pktmbuf_data_len(mbufs[newcnt]) = size; | |
1593 | rte_pktmbuf_pkt_len(mbufs[newcnt]) = size; | |
1594 | ||
1595 | newcnt++; | |
1596 | } | |
8a9562d2 | 1597 | |
58397e6c | 1598 | if (dev->type == DPDK_DEV_VHOST) { |
64839cf4 | 1599 | __netdev_dpdk_vhost_send(netdev, qid, (struct dp_packet **) mbufs, |
dd52de45 | 1600 | newcnt); |
58397e6c | 1601 | } else { |
0bf765f7 IS |
1602 | unsigned int qos_pkts = newcnt; |
1603 | ||
1604 | /* Check if QoS has been configured for this netdev. */ | |
1605 | newcnt = netdev_dpdk_qos_run__(dev, mbufs, newcnt); | |
1606 | ||
1607 | dropped += qos_pkts - newcnt; | |
b59cc14e | 1608 | netdev_dpdk_eth_tx_burst(dev, qid, mbufs, newcnt); |
58397e6c | 1609 | } |
db73f716 | 1610 | |
0bf765f7 IS |
1611 | if (OVS_UNLIKELY(dropped)) { |
1612 | rte_spinlock_lock(&dev->stats_lock); | |
1613 | dev->stats.tx_dropped += dropped; | |
1614 | rte_spinlock_unlock(&dev->stats_lock); | |
1615 | } | |
1616 | ||
5f17de68 | 1617 | if (!dpdk_thread_is_pmd()) { |
db73f716 DDP |
1618 | ovs_mutex_unlock(&nonpmd_mempool_mutex); |
1619 | } | |
8a9562d2 PS |
1620 | } |
1621 | ||
58397e6c | 1622 | static int |
64839cf4 WT |
1623 | netdev_dpdk_vhost_send(struct netdev *netdev, int qid, |
1624 | struct dp_packet_batch *batch, | |
324c8374 | 1625 | bool may_steal, bool concurrent_txq OVS_UNUSED) |
58397e6c | 1626 | { |
58397e6c | 1627 | |
dd52de45 | 1628 | if (OVS_UNLIKELY(!may_steal || batch->packets[0]->source != DPBUF_DPDK)) { |
64839cf4 WT |
1629 | dpdk_do_tx_copy(netdev, qid, batch); |
1630 | dp_packet_delete_batch(batch, may_steal); | |
58397e6c | 1631 | } else { |
64839cf4 | 1632 | dp_packet_batch_apply_cutlen(batch); |
dd52de45 | 1633 | __netdev_dpdk_vhost_send(netdev, qid, batch->packets, batch->count); |
58397e6c KT |
1634 | } |
1635 | return 0; | |
1636 | } | |
1637 | ||
7251515e DV |
1638 | static inline void |
1639 | netdev_dpdk_send__(struct netdev_dpdk *dev, int qid, | |
324c8374 IM |
1640 | struct dp_packet_batch *batch, bool may_steal, |
1641 | bool concurrent_txq) | |
8a9562d2 | 1642 | { |
324c8374 | 1643 | if (OVS_UNLIKELY(concurrent_txq)) { |
81acebda | 1644 | qid = qid % dev->up.n_txq; |
a0cb2d66 DDP |
1645 | rte_spinlock_lock(&dev->tx_q[qid].tx_lock); |
1646 | } | |
1647 | ||
7251515e | 1648 | if (OVS_UNLIKELY(!may_steal || |
64839cf4 | 1649 | batch->packets[0]->source != DPBUF_DPDK)) { |
7251515e DV |
1650 | struct netdev *netdev = &dev->up; |
1651 | ||
64839cf4 WT |
1652 | dpdk_do_tx_copy(netdev, qid, batch); |
1653 | dp_packet_delete_batch(batch, may_steal); | |
8a9562d2 | 1654 | } else { |
6b094bf4 | 1655 | int dropped; |
64839cf4 | 1656 | int cnt = batch->count; |
6b094bf4 | 1657 | struct rte_mbuf **cur_pkts = (struct rte_mbuf **) batch->packets; |
8a9562d2 | 1658 | |
7d6d1a40 WT |
1659 | dp_packet_batch_apply_cutlen(batch); |
1660 | ||
6b094bf4 IM |
1661 | cnt = netdev_dpdk_filter_packet_len(dev, cur_pkts, cnt); |
1662 | cnt = netdev_dpdk_qos_run__(dev, cur_pkts, cnt); | |
1663 | dropped = batch->count - cnt; | |
1b99bb05 | 1664 | |
6b094bf4 | 1665 | netdev_dpdk_eth_tx_burst(dev, qid, cur_pkts, cnt); |
8a9562d2 | 1666 | |
f4fd623c | 1667 | if (OVS_UNLIKELY(dropped)) { |
45d947c4 | 1668 | rte_spinlock_lock(&dev->stats_lock); |
f4fd623c | 1669 | dev->stats.tx_dropped += dropped; |
45d947c4 | 1670 | rte_spinlock_unlock(&dev->stats_lock); |
f4fd623c | 1671 | } |
8a9562d2 | 1672 | } |
a0cb2d66 | 1673 | |
324c8374 | 1674 | if (OVS_UNLIKELY(concurrent_txq)) { |
a0cb2d66 DDP |
1675 | rte_spinlock_unlock(&dev->tx_q[qid].tx_lock); |
1676 | } | |
7251515e DV |
1677 | } |
1678 | ||
1679 | static int | |
1680 | netdev_dpdk_eth_send(struct netdev *netdev, int qid, | |
324c8374 IM |
1681 | struct dp_packet_batch *batch, bool may_steal, |
1682 | bool concurrent_txq) | |
7251515e DV |
1683 | { |
1684 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
8a9562d2 | 1685 | |
324c8374 | 1686 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e | 1687 | return 0; |
8a9562d2 PS |
1688 | } |
1689 | ||
1690 | static int | |
74ff3298 | 1691 | netdev_dpdk_set_etheraddr(struct netdev *netdev, const struct eth_addr mac) |
8a9562d2 PS |
1692 | { |
1693 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1694 | ||
1695 | ovs_mutex_lock(&dev->mutex); | |
1696 | if (!eth_addr_equals(dev->hwaddr, mac)) { | |
74ff3298 | 1697 | dev->hwaddr = mac; |
045c0d1a | 1698 | netdev_change_seq_changed(netdev); |
8a9562d2 PS |
1699 | } |
1700 | ovs_mutex_unlock(&dev->mutex); | |
1701 | ||
1702 | return 0; | |
1703 | } | |
1704 | ||
1705 | static int | |
74ff3298 | 1706 | netdev_dpdk_get_etheraddr(const struct netdev *netdev, struct eth_addr *mac) |
8a9562d2 PS |
1707 | { |
1708 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1709 | ||
1710 | ovs_mutex_lock(&dev->mutex); | |
74ff3298 | 1711 | *mac = dev->hwaddr; |
8a9562d2 PS |
1712 | ovs_mutex_unlock(&dev->mutex); |
1713 | ||
1714 | return 0; | |
1715 | } | |
1716 | ||
1717 | static int | |
1718 | netdev_dpdk_get_mtu(const struct netdev *netdev, int *mtup) | |
1719 | { | |
1720 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1721 | ||
1722 | ovs_mutex_lock(&dev->mutex); | |
1723 | *mtup = dev->mtu; | |
1724 | ovs_mutex_unlock(&dev->mutex); | |
1725 | ||
1726 | return 0; | |
1727 | } | |
1728 | ||
0072e931 MK |
1729 | static int |
1730 | netdev_dpdk_set_mtu(struct netdev *netdev, int mtu) | |
1731 | { | |
1732 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1733 | ||
1734 | if (MTU_TO_FRAME_LEN(mtu) > NETDEV_DPDK_MAX_PKT_LEN | |
1735 | || mtu < ETHER_MIN_MTU) { | |
1736 | VLOG_WARN("%s: unsupported MTU %d\n", dev->up.name, mtu); | |
1737 | return EINVAL; | |
1738 | } | |
1739 | ||
1740 | ovs_mutex_lock(&dev->mutex); | |
1741 | if (dev->requested_mtu != mtu) { | |
1742 | dev->requested_mtu = mtu; | |
1743 | netdev_request_reconfigure(netdev); | |
1744 | } | |
1745 | ovs_mutex_unlock(&dev->mutex); | |
1746 | ||
1747 | return 0; | |
1748 | } | |
1749 | ||
8a9562d2 | 1750 | static int |
d46285a2 | 1751 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier); |
8a9562d2 | 1752 | |
58397e6c KT |
1753 | static int |
1754 | netdev_dpdk_vhost_get_stats(const struct netdev *netdev, | |
1755 | struct netdev_stats *stats) | |
1756 | { | |
1757 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1758 | ||
1759 | ovs_mutex_lock(&dev->mutex); | |
58397e6c | 1760 | |
45d947c4 | 1761 | rte_spinlock_lock(&dev->stats_lock); |
58397e6c KT |
1762 | /* Supported Stats */ |
1763 | stats->rx_packets += dev->stats.rx_packets; | |
1764 | stats->tx_packets += dev->stats.tx_packets; | |
9509913a | 1765 | stats->rx_dropped = dev->stats.rx_dropped; |
58397e6c | 1766 | stats->tx_dropped += dev->stats.tx_dropped; |
9e3ddd45 TP |
1767 | stats->multicast = dev->stats.multicast; |
1768 | stats->rx_bytes = dev->stats.rx_bytes; | |
1769 | stats->tx_bytes = dev->stats.tx_bytes; | |
1770 | stats->rx_errors = dev->stats.rx_errors; | |
1771 | stats->rx_length_errors = dev->stats.rx_length_errors; | |
d6e3feb5 | 1772 | |
1773 | stats->rx_1_to_64_packets = dev->stats.rx_1_to_64_packets; | |
1774 | stats->rx_65_to_127_packets = dev->stats.rx_65_to_127_packets; | |
1775 | stats->rx_128_to_255_packets = dev->stats.rx_128_to_255_packets; | |
1776 | stats->rx_256_to_511_packets = dev->stats.rx_256_to_511_packets; | |
1777 | stats->rx_512_to_1023_packets = dev->stats.rx_512_to_1023_packets; | |
1778 | stats->rx_1024_to_1522_packets = dev->stats.rx_1024_to_1522_packets; | |
1779 | stats->rx_1523_to_max_packets = dev->stats.rx_1523_to_max_packets; | |
1780 | ||
45d947c4 | 1781 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 | 1782 | |
58397e6c KT |
1783 | ovs_mutex_unlock(&dev->mutex); |
1784 | ||
1785 | return 0; | |
1786 | } | |
1787 | ||
d6e3feb5 | 1788 | static void |
1789 | netdev_dpdk_convert_xstats(struct netdev_stats *stats, | |
0a0f39df CL |
1790 | const struct rte_eth_xstat *xstats, |
1791 | const struct rte_eth_xstat_name *names, | |
d6e3feb5 | 1792 | const unsigned int size) |
1793 | { | |
d6e3feb5 | 1794 | for (unsigned int i = 0; i < size; i++) { |
0a0f39df | 1795 | if (strcmp(XSTAT_RX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1796 | stats->rx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1797 | } else if (strcmp(XSTAT_RX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1798 | stats->rx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1799 | } else if (strcmp(XSTAT_RX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1800 | stats->rx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1801 | } else if (strcmp(XSTAT_RX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1802 | stats->rx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1803 | } else if (strcmp(XSTAT_RX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1804 | stats->rx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1805 | } else if (strcmp(XSTAT_RX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1806 | stats->rx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1807 | } else if (strcmp(XSTAT_RX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1808 | stats->rx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1809 | } else if (strcmp(XSTAT_TX_64_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1810 | stats->tx_1_to_64_packets = xstats[i].value; |
0a0f39df | 1811 | } else if (strcmp(XSTAT_TX_65_TO_127_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1812 | stats->tx_65_to_127_packets = xstats[i].value; |
0a0f39df | 1813 | } else if (strcmp(XSTAT_TX_128_TO_255_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1814 | stats->tx_128_to_255_packets = xstats[i].value; |
0a0f39df | 1815 | } else if (strcmp(XSTAT_TX_256_TO_511_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1816 | stats->tx_256_to_511_packets = xstats[i].value; |
0a0f39df | 1817 | } else if (strcmp(XSTAT_TX_512_TO_1023_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1818 | stats->tx_512_to_1023_packets = xstats[i].value; |
0a0f39df | 1819 | } else if (strcmp(XSTAT_TX_1024_TO_1522_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1820 | stats->tx_1024_to_1522_packets = xstats[i].value; |
0a0f39df | 1821 | } else if (strcmp(XSTAT_TX_1523_TO_MAX_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1822 | stats->tx_1523_to_max_packets = xstats[i].value; |
0a0f39df | 1823 | } else if (strcmp(XSTAT_TX_MULTICAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1824 | stats->tx_multicast_packets = xstats[i].value; |
0a0f39df | 1825 | } else if (strcmp(XSTAT_RX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1826 | stats->rx_broadcast_packets = xstats[i].value; |
0a0f39df | 1827 | } else if (strcmp(XSTAT_TX_BROADCAST_PACKETS, names[i].name) == 0) { |
d6e3feb5 | 1828 | stats->tx_broadcast_packets = xstats[i].value; |
0a0f39df | 1829 | } else if (strcmp(XSTAT_RX_UNDERSIZED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1830 | stats->rx_undersized_errors = xstats[i].value; |
0a0f39df | 1831 | } else if (strcmp(XSTAT_RX_FRAGMENTED_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1832 | stats->rx_fragmented_errors = xstats[i].value; |
0a0f39df | 1833 | } else if (strcmp(XSTAT_RX_JABBER_ERRORS, names[i].name) == 0) { |
d6e3feb5 | 1834 | stats->rx_jabber_errors = xstats[i].value; |
1835 | } | |
1836 | } | |
1837 | } | |
1838 | ||
8a9562d2 PS |
1839 | static int |
1840 | netdev_dpdk_get_stats(const struct netdev *netdev, struct netdev_stats *stats) | |
1841 | { | |
1842 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1843 | struct rte_eth_stats rte_stats; | |
1844 | bool gg; | |
1845 | ||
1846 | netdev_dpdk_get_carrier(netdev, &gg); | |
1847 | ovs_mutex_lock(&dev->mutex); | |
8a9562d2 | 1848 | |
0a0f39df CL |
1849 | struct rte_eth_xstat *rte_xstats = NULL; |
1850 | struct rte_eth_xstat_name *rte_xstats_names = NULL; | |
1851 | int rte_xstats_len, rte_xstats_new_len, rte_xstats_ret; | |
d6e3feb5 | 1852 | |
1853 | if (rte_eth_stats_get(dev->port_id, &rte_stats)) { | |
1854 | VLOG_ERR("Can't get ETH statistics for port: %i.", dev->port_id); | |
f9256822 | 1855 | ovs_mutex_unlock(&dev->mutex); |
d6e3feb5 | 1856 | return EPROTO; |
1857 | } | |
1858 | ||
0a0f39df CL |
1859 | /* Get length of statistics */ |
1860 | rte_xstats_len = rte_eth_xstats_get_names(dev->port_id, NULL, 0); | |
1861 | if (rte_xstats_len < 0) { | |
1862 | VLOG_WARN("Cannot get XSTATS values for port: %i", dev->port_id); | |
1863 | goto out; | |
1864 | } | |
1865 | /* Reserve memory for xstats names and values */ | |
1866 | rte_xstats_names = xcalloc(rte_xstats_len, sizeof *rte_xstats_names); | |
1867 | rte_xstats = xcalloc(rte_xstats_len, sizeof *rte_xstats); | |
1868 | ||
1869 | /* Retreive xstats names */ | |
1870 | rte_xstats_new_len = rte_eth_xstats_get_names(dev->port_id, | |
1871 | rte_xstats_names, | |
1872 | rte_xstats_len); | |
1873 | if (rte_xstats_new_len != rte_xstats_len) { | |
1874 | VLOG_WARN("Cannot get XSTATS names for port: %i.", dev->port_id); | |
1875 | goto out; | |
1876 | } | |
1877 | /* Retreive xstats values */ | |
1878 | memset(rte_xstats, 0xff, sizeof *rte_xstats * rte_xstats_len); | |
1879 | rte_xstats_ret = rte_eth_xstats_get(dev->port_id, rte_xstats, | |
1880 | rte_xstats_len); | |
1881 | if (rte_xstats_ret > 0 && rte_xstats_ret <= rte_xstats_len) { | |
1882 | netdev_dpdk_convert_xstats(stats, rte_xstats, rte_xstats_names, | |
1883 | rte_xstats_len); | |
d6e3feb5 | 1884 | } else { |
0a0f39df | 1885 | VLOG_WARN("Cannot get XSTATS values for port: %i.", dev->port_id); |
d6e3feb5 | 1886 | } |
8a9562d2 | 1887 | |
0a0f39df CL |
1888 | out: |
1889 | free(rte_xstats); | |
1890 | free(rte_xstats_names); | |
1891 | ||
2f9dd77f PS |
1892 | stats->rx_packets = rte_stats.ipackets; |
1893 | stats->tx_packets = rte_stats.opackets; | |
1894 | stats->rx_bytes = rte_stats.ibytes; | |
1895 | stats->tx_bytes = rte_stats.obytes; | |
9e3ddd45 TP |
1896 | /* DPDK counts imissed as errors, but count them here as dropped instead */ |
1897 | stats->rx_errors = rte_stats.ierrors - rte_stats.imissed; | |
2f9dd77f | 1898 | stats->tx_errors = rte_stats.oerrors; |
8a9562d2 | 1899 | |
45d947c4 | 1900 | rte_spinlock_lock(&dev->stats_lock); |
2f9dd77f | 1901 | stats->tx_dropped = dev->stats.tx_dropped; |
9509913a | 1902 | stats->rx_dropped = dev->stats.rx_dropped; |
45d947c4 | 1903 | rte_spinlock_unlock(&dev->stats_lock); |
9e3ddd45 TP |
1904 | |
1905 | /* These are the available DPDK counters for packets not received due to | |
1906 | * local resource constraints in DPDK and NIC respectively. */ | |
9509913a | 1907 | stats->rx_dropped += rte_stats.rx_nombuf + rte_stats.imissed; |
9e3ddd45 TP |
1908 | stats->rx_missed_errors = rte_stats.imissed; |
1909 | ||
8a9562d2 PS |
1910 | ovs_mutex_unlock(&dev->mutex); |
1911 | ||
1912 | return 0; | |
1913 | } | |
1914 | ||
1915 | static int | |
d46285a2 | 1916 | netdev_dpdk_get_features(const struct netdev *netdev, |
8a9562d2 PS |
1917 | enum netdev_features *current, |
1918 | enum netdev_features *advertised OVS_UNUSED, | |
1919 | enum netdev_features *supported OVS_UNUSED, | |
1920 | enum netdev_features *peer OVS_UNUSED) | |
1921 | { | |
d46285a2 | 1922 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
1923 | struct rte_eth_link link; |
1924 | ||
1925 | ovs_mutex_lock(&dev->mutex); | |
1926 | link = dev->link; | |
1927 | ovs_mutex_unlock(&dev->mutex); | |
1928 | ||
362ca396 | 1929 | if (link.link_duplex == ETH_LINK_HALF_DUPLEX) { |
1930 | if (link.link_speed == ETH_SPEED_NUM_10M) { | |
8a9562d2 PS |
1931 | *current = NETDEV_F_10MB_HD; |
1932 | } | |
362ca396 | 1933 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1934 | *current = NETDEV_F_100MB_HD; |
1935 | } | |
362ca396 | 1936 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1937 | *current = NETDEV_F_1GB_HD; |
1938 | } | |
1939 | } else if (link.link_duplex == ETH_LINK_FULL_DUPLEX) { | |
362ca396 | 1940 | if (link.link_speed == ETH_SPEED_NUM_10M) { |
8a9562d2 PS |
1941 | *current = NETDEV_F_10MB_FD; |
1942 | } | |
362ca396 | 1943 | if (link.link_speed == ETH_SPEED_NUM_100M) { |
8a9562d2 PS |
1944 | *current = NETDEV_F_100MB_FD; |
1945 | } | |
362ca396 | 1946 | if (link.link_speed == ETH_SPEED_NUM_1G) { |
8a9562d2 PS |
1947 | *current = NETDEV_F_1GB_FD; |
1948 | } | |
362ca396 | 1949 | if (link.link_speed == ETH_SPEED_NUM_10G) { |
8a9562d2 PS |
1950 | *current = NETDEV_F_10GB_FD; |
1951 | } | |
1952 | } | |
1953 | ||
362ca396 | 1954 | if (link.link_autoneg) { |
1955 | *current |= NETDEV_F_AUTONEG; | |
1956 | } | |
1957 | ||
8a9562d2 PS |
1958 | return 0; |
1959 | } | |
1960 | ||
9509913a IS |
1961 | static struct ingress_policer * |
1962 | netdev_dpdk_policer_construct(uint32_t rate, uint32_t burst) | |
1963 | { | |
1964 | struct ingress_policer *policer = NULL; | |
1965 | uint64_t rate_bytes; | |
1966 | uint64_t burst_bytes; | |
1967 | int err = 0; | |
1968 | ||
1969 | policer = xmalloc(sizeof *policer); | |
1970 | rte_spinlock_init(&policer->policer_lock); | |
1971 | ||
1972 | /* rte_meter requires bytes so convert kbits rate and burst to bytes. */ | |
1973 | rate_bytes = rate * 1000/8; | |
1974 | burst_bytes = burst * 1000/8; | |
1975 | ||
1976 | policer->app_srtcm_params.cir = rate_bytes; | |
1977 | policer->app_srtcm_params.cbs = burst_bytes; | |
1978 | policer->app_srtcm_params.ebs = 0; | |
1979 | err = rte_meter_srtcm_config(&policer->in_policer, | |
1980 | &policer->app_srtcm_params); | |
1981 | if(err) { | |
1982 | VLOG_ERR("Could not create rte meter for ingress policer"); | |
1983 | return NULL; | |
1984 | } | |
1985 | ||
1986 | return policer; | |
1987 | } | |
1988 | ||
1989 | static int | |
1990 | netdev_dpdk_set_policing(struct netdev* netdev, uint32_t policer_rate, | |
1991 | uint32_t policer_burst) | |
1992 | { | |
1993 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
1994 | struct ingress_policer *policer; | |
1995 | ||
1996 | /* Force to 0 if no rate specified, | |
1997 | * default to 8000 kbits if burst is 0, | |
1998 | * else stick with user-specified value. | |
1999 | */ | |
2000 | policer_burst = (!policer_rate ? 0 | |
2001 | : !policer_burst ? 8000 | |
2002 | : policer_burst); | |
2003 | ||
2004 | ovs_mutex_lock(&dev->mutex); | |
2005 | ||
2006 | policer = ovsrcu_get_protected(struct ingress_policer *, | |
2007 | &dev->ingress_policer); | |
2008 | ||
2009 | if (dev->policer_rate == policer_rate && | |
2010 | dev->policer_burst == policer_burst) { | |
2011 | /* Assume that settings haven't changed since we last set them. */ | |
2012 | ovs_mutex_unlock(&dev->mutex); | |
2013 | return 0; | |
2014 | } | |
2015 | ||
2016 | /* Destroy any existing ingress policer for the device if one exists */ | |
2017 | if (policer) { | |
2018 | ovsrcu_postpone(free, policer); | |
2019 | } | |
2020 | ||
2021 | if (policer_rate != 0) { | |
2022 | policer = netdev_dpdk_policer_construct(policer_rate, policer_burst); | |
2023 | } else { | |
2024 | policer = NULL; | |
2025 | } | |
2026 | ovsrcu_set(&dev->ingress_policer, policer); | |
2027 | dev->policer_rate = policer_rate; | |
2028 | dev->policer_burst = policer_burst; | |
2029 | ovs_mutex_unlock(&dev->mutex); | |
2030 | ||
2031 | return 0; | |
2032 | } | |
2033 | ||
8a9562d2 PS |
2034 | static int |
2035 | netdev_dpdk_get_ifindex(const struct netdev *netdev) | |
2036 | { | |
2037 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2038 | int ifindex; | |
2039 | ||
2040 | ovs_mutex_lock(&dev->mutex); | |
2041 | ifindex = dev->port_id; | |
2042 | ovs_mutex_unlock(&dev->mutex); | |
2043 | ||
2044 | return ifindex; | |
2045 | } | |
2046 | ||
2047 | static int | |
d46285a2 | 2048 | netdev_dpdk_get_carrier(const struct netdev *netdev, bool *carrier) |
8a9562d2 | 2049 | { |
d46285a2 | 2050 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2051 | |
2052 | ovs_mutex_lock(&dev->mutex); | |
2053 | check_link_status(dev); | |
2054 | *carrier = dev->link.link_status; | |
58397e6c KT |
2055 | |
2056 | ovs_mutex_unlock(&dev->mutex); | |
2057 | ||
2058 | return 0; | |
2059 | } | |
2060 | ||
2061 | static int | |
d46285a2 | 2062 | netdev_dpdk_vhost_get_carrier(const struct netdev *netdev, bool *carrier) |
58397e6c | 2063 | { |
d46285a2 | 2064 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
58397e6c KT |
2065 | |
2066 | ovs_mutex_lock(&dev->mutex); | |
2067 | ||
0a0f39df | 2068 | if (is_vhost_running(dev)) { |
58397e6c KT |
2069 | *carrier = 1; |
2070 | } else { | |
2071 | *carrier = 0; | |
2072 | } | |
2073 | ||
8a9562d2 PS |
2074 | ovs_mutex_unlock(&dev->mutex); |
2075 | ||
2076 | return 0; | |
2077 | } | |
2078 | ||
2079 | static long long int | |
d46285a2 | 2080 | netdev_dpdk_get_carrier_resets(const struct netdev *netdev) |
8a9562d2 | 2081 | { |
d46285a2 | 2082 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2083 | long long int carrier_resets; |
2084 | ||
2085 | ovs_mutex_lock(&dev->mutex); | |
2086 | carrier_resets = dev->link_reset_cnt; | |
2087 | ovs_mutex_unlock(&dev->mutex); | |
2088 | ||
2089 | return carrier_resets; | |
2090 | } | |
2091 | ||
2092 | static int | |
d46285a2 | 2093 | netdev_dpdk_set_miimon(struct netdev *netdev OVS_UNUSED, |
8a9562d2 PS |
2094 | long long int interval OVS_UNUSED) |
2095 | { | |
ee32150e | 2096 | return EOPNOTSUPP; |
8a9562d2 PS |
2097 | } |
2098 | ||
2099 | static int | |
2100 | netdev_dpdk_update_flags__(struct netdev_dpdk *dev, | |
2101 | enum netdev_flags off, enum netdev_flags on, | |
64839cf4 WT |
2102 | enum netdev_flags *old_flagsp) |
2103 | OVS_REQUIRES(dev->mutex) | |
8a9562d2 PS |
2104 | { |
2105 | int err; | |
2106 | ||
2107 | if ((off | on) & ~(NETDEV_UP | NETDEV_PROMISC)) { | |
2108 | return EINVAL; | |
2109 | } | |
2110 | ||
2111 | *old_flagsp = dev->flags; | |
2112 | dev->flags |= on; | |
2113 | dev->flags &= ~off; | |
2114 | ||
2115 | if (dev->flags == *old_flagsp) { | |
2116 | return 0; | |
2117 | } | |
2118 | ||
58397e6c KT |
2119 | if (dev->type == DPDK_DEV_ETH) { |
2120 | if (dev->flags & NETDEV_UP) { | |
2121 | err = rte_eth_dev_start(dev->port_id); | |
2122 | if (err) | |
2123 | return -err; | |
2124 | } | |
8a9562d2 | 2125 | |
58397e6c KT |
2126 | if (dev->flags & NETDEV_PROMISC) { |
2127 | rte_eth_promiscuous_enable(dev->port_id); | |
2128 | } | |
8a9562d2 | 2129 | |
58397e6c KT |
2130 | if (!(dev->flags & NETDEV_UP)) { |
2131 | rte_eth_dev_stop(dev->port_id); | |
2132 | } | |
e543851d ZB |
2133 | } else { |
2134 | /* If DPDK_DEV_VHOST device's NETDEV_UP flag was changed and vhost is | |
2135 | * running then change netdev's change_seq to trigger link state | |
2136 | * update. */ | |
e543851d ZB |
2137 | |
2138 | if ((NETDEV_UP & ((*old_flagsp ^ on) | (*old_flagsp ^ off))) | |
0a0f39df | 2139 | && is_vhost_running(dev)) { |
e543851d ZB |
2140 | netdev_change_seq_changed(&dev->up); |
2141 | ||
2142 | /* Clear statistics if device is getting up. */ | |
2143 | if (NETDEV_UP & on) { | |
2144 | rte_spinlock_lock(&dev->stats_lock); | |
2145 | memset(&dev->stats, 0, sizeof(dev->stats)); | |
2146 | rte_spinlock_unlock(&dev->stats_lock); | |
2147 | } | |
2148 | } | |
8a9562d2 PS |
2149 | } |
2150 | ||
2151 | return 0; | |
2152 | } | |
2153 | ||
2154 | static int | |
d46285a2 | 2155 | netdev_dpdk_update_flags(struct netdev *netdev, |
8a9562d2 PS |
2156 | enum netdev_flags off, enum netdev_flags on, |
2157 | enum netdev_flags *old_flagsp) | |
2158 | { | |
d46285a2 | 2159 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2160 | int error; |
2161 | ||
d46285a2 DDP |
2162 | ovs_mutex_lock(&dev->mutex); |
2163 | error = netdev_dpdk_update_flags__(dev, off, on, old_flagsp); | |
2164 | ovs_mutex_unlock(&dev->mutex); | |
8a9562d2 PS |
2165 | |
2166 | return error; | |
2167 | } | |
2168 | ||
2169 | static int | |
d46285a2 | 2170 | netdev_dpdk_get_status(const struct netdev *netdev, struct smap *args) |
8a9562d2 | 2171 | { |
d46285a2 | 2172 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
8a9562d2 PS |
2173 | struct rte_eth_dev_info dev_info; |
2174 | ||
7cd1261d | 2175 | if (!rte_eth_dev_is_valid_port(dev->port_id)) { |
8a9562d2 | 2176 | return ENODEV; |
7cd1261d | 2177 | } |
8a9562d2 PS |
2178 | |
2179 | ovs_mutex_lock(&dev->mutex); | |
2180 | rte_eth_dev_info_get(dev->port_id, &dev_info); | |
2181 | ovs_mutex_unlock(&dev->mutex); | |
2182 | ||
95fb793a | 2183 | smap_add_format(args, "port_no", "%d", dev->port_id); |
8a9562d2 PS |
2184 | smap_add_format(args, "numa_id", "%d", rte_eth_dev_socket_id(dev->port_id)); |
2185 | smap_add_format(args, "driver_name", "%s", dev_info.driver_name); | |
2186 | smap_add_format(args, "min_rx_bufsize", "%u", dev_info.min_rx_bufsize); | |
4be4d22c | 2187 | smap_add_format(args, "max_rx_pktlen", "%u", dev->max_packet_len); |
8a9562d2 PS |
2188 | smap_add_format(args, "max_rx_queues", "%u", dev_info.max_rx_queues); |
2189 | smap_add_format(args, "max_tx_queues", "%u", dev_info.max_tx_queues); | |
2190 | smap_add_format(args, "max_mac_addrs", "%u", dev_info.max_mac_addrs); | |
2191 | smap_add_format(args, "max_hash_mac_addrs", "%u", dev_info.max_hash_mac_addrs); | |
2192 | smap_add_format(args, "max_vfs", "%u", dev_info.max_vfs); | |
2193 | smap_add_format(args, "max_vmdq_pools", "%u", dev_info.max_vmdq_pools); | |
2194 | ||
39c2baa9 | 2195 | if (dev_info.pci_dev) { |
2196 | smap_add_format(args, "pci-vendor_id", "0x%u", | |
2197 | dev_info.pci_dev->id.vendor_id); | |
2198 | smap_add_format(args, "pci-device_id", "0x%x", | |
2199 | dev_info.pci_dev->id.device_id); | |
2200 | } | |
8a9562d2 PS |
2201 | |
2202 | return 0; | |
2203 | } | |
2204 | ||
2205 | static void | |
2206 | netdev_dpdk_set_admin_state__(struct netdev_dpdk *dev, bool admin_state) | |
2207 | OVS_REQUIRES(dev->mutex) | |
2208 | { | |
2209 | enum netdev_flags old_flags; | |
2210 | ||
2211 | if (admin_state) { | |
2212 | netdev_dpdk_update_flags__(dev, 0, NETDEV_UP, &old_flags); | |
2213 | } else { | |
2214 | netdev_dpdk_update_flags__(dev, NETDEV_UP, 0, &old_flags); | |
2215 | } | |
2216 | } | |
2217 | ||
2218 | static void | |
2219 | netdev_dpdk_set_admin_state(struct unixctl_conn *conn, int argc, | |
2220 | const char *argv[], void *aux OVS_UNUSED) | |
2221 | { | |
2222 | bool up; | |
2223 | ||
2224 | if (!strcasecmp(argv[argc - 1], "up")) { | |
2225 | up = true; | |
2226 | } else if ( !strcasecmp(argv[argc - 1], "down")) { | |
2227 | up = false; | |
2228 | } else { | |
2229 | unixctl_command_reply_error(conn, "Invalid Admin State"); | |
2230 | return; | |
2231 | } | |
2232 | ||
2233 | if (argc > 2) { | |
2234 | struct netdev *netdev = netdev_from_name(argv[1]); | |
2235 | if (netdev && is_dpdk_class(netdev->netdev_class)) { | |
2236 | struct netdev_dpdk *dpdk_dev = netdev_dpdk_cast(netdev); | |
2237 | ||
2238 | ovs_mutex_lock(&dpdk_dev->mutex); | |
2239 | netdev_dpdk_set_admin_state__(dpdk_dev, up); | |
2240 | ovs_mutex_unlock(&dpdk_dev->mutex); | |
2241 | ||
2242 | netdev_close(netdev); | |
2243 | } else { | |
2244 | unixctl_command_reply_error(conn, "Not a DPDK Interface"); | |
2245 | netdev_close(netdev); | |
2246 | return; | |
2247 | } | |
2248 | } else { | |
2249 | struct netdev_dpdk *netdev; | |
2250 | ||
2251 | ovs_mutex_lock(&dpdk_mutex); | |
2252 | LIST_FOR_EACH (netdev, list_node, &dpdk_list) { | |
2253 | ovs_mutex_lock(&netdev->mutex); | |
2254 | netdev_dpdk_set_admin_state__(netdev, up); | |
2255 | ovs_mutex_unlock(&netdev->mutex); | |
2256 | } | |
2257 | ovs_mutex_unlock(&dpdk_mutex); | |
2258 | } | |
2259 | unixctl_command_reply(conn, "OK"); | |
2260 | } | |
2261 | ||
58397e6c KT |
2262 | /* |
2263 | * Set virtqueue flags so that we do not receive interrupts. | |
2264 | */ | |
2265 | static void | |
0a0f39df | 2266 | set_irq_status(int vid) |
58397e6c | 2267 | { |
4573fbd3 FL |
2268 | uint32_t i; |
2269 | uint64_t idx; | |
2270 | ||
0a0f39df | 2271 | for (i = 0; i < rte_vhost_get_queue_num(vid); i++) { |
4573fbd3 | 2272 | idx = i * VIRTIO_QNUM; |
0a0f39df CL |
2273 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_RXQ, 0); |
2274 | rte_vhost_enable_guest_notification(vid, idx + VIRTIO_TXQ, 0); | |
4573fbd3 FL |
2275 | } |
2276 | } | |
2277 | ||
585a5bea IM |
2278 | /* |
2279 | * Fixes mapping for vhost-user tx queues. Must be called after each | |
81acebda | 2280 | * enabling/disabling of queues and n_txq modifications. |
585a5bea IM |
2281 | */ |
2282 | static void | |
d46285a2 DDP |
2283 | netdev_dpdk_remap_txqs(struct netdev_dpdk *dev) |
2284 | OVS_REQUIRES(dev->mutex) | |
585a5bea IM |
2285 | { |
2286 | int *enabled_queues, n_enabled = 0; | |
81acebda | 2287 | int i, k, total_txqs = dev->up.n_txq; |
585a5bea IM |
2288 | |
2289 | enabled_queues = dpdk_rte_mzalloc(total_txqs * sizeof *enabled_queues); | |
2290 | ||
2291 | for (i = 0; i < total_txqs; i++) { | |
2292 | /* Enabled queues always mapped to themselves. */ | |
d46285a2 | 2293 | if (dev->tx_q[i].map == i) { |
585a5bea IM |
2294 | enabled_queues[n_enabled++] = i; |
2295 | } | |
2296 | } | |
2297 | ||
2298 | if (n_enabled == 0 && total_txqs != 0) { | |
f3ea2ad2 | 2299 | enabled_queues[0] = OVS_VHOST_QUEUE_DISABLED; |
585a5bea IM |
2300 | n_enabled = 1; |
2301 | } | |
2302 | ||
2303 | k = 0; | |
2304 | for (i = 0; i < total_txqs; i++) { | |
d46285a2 DDP |
2305 | if (dev->tx_q[i].map != i) { |
2306 | dev->tx_q[i].map = enabled_queues[k]; | |
585a5bea IM |
2307 | k = (k + 1) % n_enabled; |
2308 | } | |
2309 | } | |
2310 | ||
2d24d165 | 2311 | VLOG_DBG("TX queue mapping for %s\n", dev->vhost_id); |
585a5bea | 2312 | for (i = 0; i < total_txqs; i++) { |
d46285a2 | 2313 | VLOG_DBG("%2d --> %2d", i, dev->tx_q[i].map); |
585a5bea IM |
2314 | } |
2315 | ||
2316 | rte_free(enabled_queues); | |
2317 | } | |
4573fbd3 | 2318 | |
58397e6c KT |
2319 | /* |
2320 | * A new virtio-net device is added to a vhost port. | |
2321 | */ | |
2322 | static int | |
0a0f39df | 2323 | new_device(int vid) |
58397e6c | 2324 | { |
d46285a2 | 2325 | struct netdev_dpdk *dev; |
58397e6c | 2326 | bool exists = false; |
db8f13b0 | 2327 | int newnode = 0; |
0a0f39df CL |
2328 | char ifname[IF_NAME_SZ]; |
2329 | ||
2330 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
58397e6c KT |
2331 | |
2332 | ovs_mutex_lock(&dpdk_mutex); | |
2333 | /* Add device to the vhost port with the same name as that passed down. */ | |
d46285a2 | 2334 | LIST_FOR_EACH(dev, list_node, &dpdk_list) { |
c1ff66ac | 2335 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2336 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
0a0f39df | 2337 | uint32_t qp_num = rte_vhost_get_queue_num(vid); |
db8f13b0 CL |
2338 | |
2339 | /* Get NUMA information */ | |
0a0f39df CL |
2340 | newnode = rte_vhost_get_numa_node(vid); |
2341 | if (newnode == -1) { | |
5b9bf9e0 | 2342 | #ifdef VHOST_NUMA |
db8f13b0 | 2343 | VLOG_INFO("Error getting NUMA info for vHost Device '%s'", |
0a0f39df | 2344 | ifname); |
5b9bf9e0 | 2345 | #endif |
db8f13b0 | 2346 | newnode = dev->socket_id; |
db8f13b0 CL |
2347 | } |
2348 | ||
7f5f2bd0 IM |
2349 | if (dev->requested_n_txq != qp_num |
2350 | || dev->requested_n_rxq != qp_num | |
2351 | || dev->requested_socket_id != newnode) { | |
2352 | dev->requested_socket_id = newnode; | |
2353 | dev->requested_n_rxq = qp_num; | |
2354 | dev->requested_n_txq = qp_num; | |
2355 | netdev_request_reconfigure(&dev->up); | |
2356 | } else { | |
2357 | /* Reconfiguration not required. */ | |
2358 | dev->vhost_reconfigured = true; | |
2359 | } | |
81acebda | 2360 | |
0a0f39df | 2361 | ovsrcu_index_set(&dev->vid, vid); |
81acebda IM |
2362 | exists = true; |
2363 | ||
58397e6c | 2364 | /* Disable notifications. */ |
0a0f39df | 2365 | set_irq_status(vid); |
e543851d | 2366 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2367 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2368 | break; |
2369 | } | |
c1ff66ac | 2370 | ovs_mutex_unlock(&dev->mutex); |
58397e6c KT |
2371 | } |
2372 | ovs_mutex_unlock(&dpdk_mutex); | |
2373 | ||
2374 | if (!exists) { | |
0a0f39df | 2375 | VLOG_INFO("vHost Device '%s' can't be added - name not found", ifname); |
58397e6c KT |
2376 | |
2377 | return -1; | |
2378 | } | |
2379 | ||
0a0f39df CL |
2380 | VLOG_INFO("vHost Device '%s' has been added on numa node %i", |
2381 | ifname, newnode); | |
2382 | ||
58397e6c KT |
2383 | return 0; |
2384 | } | |
2385 | ||
f3ea2ad2 IM |
2386 | /* Clears mapping for all available queues of vhost interface. */ |
2387 | static void | |
2388 | netdev_dpdk_txq_map_clear(struct netdev_dpdk *dev) | |
2389 | OVS_REQUIRES(dev->mutex) | |
2390 | { | |
2391 | int i; | |
2392 | ||
81acebda | 2393 | for (i = 0; i < dev->up.n_txq; i++) { |
f3ea2ad2 IM |
2394 | dev->tx_q[i].map = OVS_VHOST_QUEUE_MAP_UNKNOWN; |
2395 | } | |
2396 | } | |
2397 | ||
58397e6c KT |
2398 | /* |
2399 | * Remove a virtio-net device from the specific vhost port. Use dev->remove | |
2400 | * flag to stop any more packets from being sent or received to/from a VM and | |
2401 | * ensure all currently queued packets have been sent/received before removing | |
2402 | * the device. | |
2403 | */ | |
2404 | static void | |
0a0f39df | 2405 | destroy_device(int vid) |
58397e6c | 2406 | { |
d46285a2 | 2407 | struct netdev_dpdk *dev; |
afee281f | 2408 | bool exists = false; |
0a0f39df CL |
2409 | char ifname[IF_NAME_SZ]; |
2410 | ||
2411 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
58397e6c KT |
2412 | |
2413 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2414 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
0a0f39df | 2415 | if (netdev_dpdk_get_vid(dev) == vid) { |
58397e6c | 2416 | |
d46285a2 | 2417 | ovs_mutex_lock(&dev->mutex); |
0a0f39df CL |
2418 | dev->vhost_reconfigured = false; |
2419 | ovsrcu_index_set(&dev->vid, -1); | |
d46285a2 | 2420 | netdev_dpdk_txq_map_clear(dev); |
81acebda | 2421 | |
e543851d | 2422 | netdev_change_seq_changed(&dev->up); |
d46285a2 | 2423 | ovs_mutex_unlock(&dev->mutex); |
81acebda | 2424 | exists = true; |
afee281f | 2425 | break; |
58397e6c KT |
2426 | } |
2427 | } | |
afee281f | 2428 | |
58397e6c KT |
2429 | ovs_mutex_unlock(&dpdk_mutex); |
2430 | ||
0a0f39df | 2431 | if (exists) { |
afee281f KT |
2432 | /* |
2433 | * Wait for other threads to quiesce after setting the 'virtio_dev' | |
2434 | * to NULL, before returning. | |
2435 | */ | |
2436 | ovsrcu_synchronize(); | |
2437 | /* | |
2438 | * As call to ovsrcu_synchronize() will end the quiescent state, | |
2439 | * put thread back into quiescent state before returning. | |
2440 | */ | |
2441 | ovsrcu_quiesce_start(); | |
0a0f39df | 2442 | VLOG_INFO("vHost Device '%s' has been removed", ifname); |
afee281f | 2443 | } else { |
0a0f39df | 2444 | VLOG_INFO("vHost Device '%s' not found", ifname); |
afee281f | 2445 | } |
58397e6c KT |
2446 | } |
2447 | ||
585a5bea | 2448 | static int |
0a0f39df | 2449 | vring_state_changed(int vid, uint16_t queue_id, int enable) |
585a5bea | 2450 | { |
d46285a2 | 2451 | struct netdev_dpdk *dev; |
585a5bea IM |
2452 | bool exists = false; |
2453 | int qid = queue_id / VIRTIO_QNUM; | |
0a0f39df CL |
2454 | char ifname[IF_NAME_SZ]; |
2455 | ||
2456 | rte_vhost_get_ifname(vid, ifname, sizeof(ifname)); | |
585a5bea IM |
2457 | |
2458 | if (queue_id % VIRTIO_QNUM == VIRTIO_TXQ) { | |
2459 | return 0; | |
2460 | } | |
2461 | ||
2462 | ovs_mutex_lock(&dpdk_mutex); | |
d46285a2 | 2463 | LIST_FOR_EACH (dev, list_node, &dpdk_list) { |
c1ff66ac | 2464 | ovs_mutex_lock(&dev->mutex); |
2d24d165 | 2465 | if (strncmp(ifname, dev->vhost_id, IF_NAME_SZ) == 0) { |
585a5bea | 2466 | if (enable) { |
d46285a2 | 2467 | dev->tx_q[qid].map = qid; |
585a5bea | 2468 | } else { |
d46285a2 | 2469 | dev->tx_q[qid].map = OVS_VHOST_QUEUE_DISABLED; |
585a5bea | 2470 | } |
d46285a2 | 2471 | netdev_dpdk_remap_txqs(dev); |
585a5bea | 2472 | exists = true; |
d46285a2 | 2473 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2474 | break; |
2475 | } | |
c1ff66ac | 2476 | ovs_mutex_unlock(&dev->mutex); |
585a5bea IM |
2477 | } |
2478 | ovs_mutex_unlock(&dpdk_mutex); | |
2479 | ||
2480 | if (exists) { | |
0a0f39df CL |
2481 | VLOG_INFO("State of queue %d ( tx_qid %d ) of vhost device '%s'" |
2482 | "changed to \'%s\'", queue_id, qid, ifname, | |
d46285a2 | 2483 | (enable == 1) ? "enabled" : "disabled"); |
585a5bea | 2484 | } else { |
0a0f39df | 2485 | VLOG_INFO("vHost Device '%s' not found", ifname); |
585a5bea IM |
2486 | return -1; |
2487 | } | |
2488 | ||
2489 | return 0; | |
2490 | } | |
2491 | ||
0a0f39df CL |
2492 | int |
2493 | netdev_dpdk_get_vid(const struct netdev_dpdk *dev) | |
58397e6c | 2494 | { |
0a0f39df | 2495 | return ovsrcu_index_get(&dev->vid); |
58397e6c KT |
2496 | } |
2497 | ||
9509913a IS |
2498 | struct ingress_policer * |
2499 | netdev_dpdk_get_ingress_policer(const struct netdev_dpdk *dev) | |
2500 | { | |
2501 | return ovsrcu_get(struct ingress_policer *, &dev->ingress_policer); | |
2502 | } | |
2503 | ||
58397e6c KT |
2504 | /* |
2505 | * These callbacks allow virtio-net devices to be added to vhost ports when | |
2506 | * configuration has been fully complete. | |
2507 | */ | |
bce01e3a | 2508 | static const struct virtio_net_device_ops virtio_net_device_ops = |
58397e6c KT |
2509 | { |
2510 | .new_device = new_device, | |
2511 | .destroy_device = destroy_device, | |
585a5bea | 2512 | .vring_state_changed = vring_state_changed |
58397e6c KT |
2513 | }; |
2514 | ||
2515 | static void * | |
7d1ced01 | 2516 | start_vhost_loop(void *dummy OVS_UNUSED) |
58397e6c KT |
2517 | { |
2518 | pthread_detach(pthread_self()); | |
41987644 | 2519 | /* Put the vhost thread into quiescent state. */ |
618f44f7 | 2520 | ovsrcu_quiesce_start(); |
58397e6c KT |
2521 | rte_vhost_driver_session_start(); |
2522 | return NULL; | |
2523 | } | |
2524 | ||
2525 | static int | |
2526 | dpdk_vhost_class_init(void) | |
7d1ced01 CL |
2527 | { |
2528 | rte_vhost_driver_callback_register(&virtio_net_device_ops); | |
362ca396 | 2529 | rte_vhost_feature_disable(1ULL << VIRTIO_NET_F_HOST_TSO4 |
2530 | | 1ULL << VIRTIO_NET_F_HOST_TSO6 | |
2531 | | 1ULL << VIRTIO_NET_F_CSUM); | |
2532 | ||
7d1ced01 CL |
2533 | ovs_thread_create("vhost_thread", start_vhost_loop, NULL); |
2534 | return 0; | |
2535 | } | |
2536 | ||
033e9df2 DDP |
2537 | static void |
2538 | dpdk_common_init(void) | |
2539 | { | |
2540 | unixctl_command_register("netdev-dpdk/set-admin-state", | |
2541 | "[netdev] up|down", 1, 2, | |
2542 | netdev_dpdk_set_admin_state, NULL); | |
2543 | ||
033e9df2 DDP |
2544 | } |
2545 | ||
95fb793a | 2546 | /* Client Rings */ |
2547 | ||
95fb793a | 2548 | static int |
2549 | dpdk_ring_create(const char dev_name[], unsigned int port_no, | |
2550 | unsigned int *eth_port_id) | |
2551 | { | |
2552 | struct dpdk_ring *ivshmem; | |
ca7e7bee | 2553 | char ring_name[RTE_RING_NAMESIZE]; |
95fb793a | 2554 | int err; |
2555 | ||
2556 | ivshmem = dpdk_rte_mzalloc(sizeof *ivshmem); | |
2557 | if (ivshmem == NULL) { | |
2558 | return ENOMEM; | |
2559 | } | |
2560 | ||
7251515e | 2561 | /* XXX: Add support for multiquque ring. */ |
ca7e7bee | 2562 | err = snprintf(ring_name, sizeof(ring_name), "%s_tx", dev_name); |
95fb793a | 2563 | if (err < 0) { |
2564 | return -err; | |
2565 | } | |
2566 | ||
8f0a76c9 | 2567 | /* Create single producer tx ring, netdev does explicit locking. */ |
7251515e | 2568 | ivshmem->cring_tx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2569 | RING_F_SP_ENQ); |
95fb793a | 2570 | if (ivshmem->cring_tx == NULL) { |
2571 | rte_free(ivshmem); | |
2572 | return ENOMEM; | |
2573 | } | |
2574 | ||
ca7e7bee | 2575 | err = snprintf(ring_name, sizeof(ring_name), "%s_rx", dev_name); |
95fb793a | 2576 | if (err < 0) { |
2577 | return -err; | |
2578 | } | |
2579 | ||
8f0a76c9 | 2580 | /* Create single consumer rx ring, netdev does explicit locking. */ |
7251515e | 2581 | ivshmem->cring_rx = rte_ring_create(ring_name, DPDK_RING_SIZE, SOCKET0, |
8f0a76c9 | 2582 | RING_F_SC_DEQ); |
95fb793a | 2583 | if (ivshmem->cring_rx == NULL) { |
2584 | rte_free(ivshmem); | |
2585 | return ENOMEM; | |
2586 | } | |
2587 | ||
d7310583 DDP |
2588 | err = rte_eth_from_rings(dev_name, &ivshmem->cring_rx, 1, |
2589 | &ivshmem->cring_tx, 1, SOCKET0); | |
2590 | ||
95fb793a | 2591 | if (err < 0) { |
2592 | rte_free(ivshmem); | |
2593 | return ENODEV; | |
2594 | } | |
2595 | ||
2596 | ivshmem->user_port_id = port_no; | |
2597 | ivshmem->eth_port_id = rte_eth_dev_count() - 1; | |
417e7e66 | 2598 | ovs_list_push_back(&dpdk_ring_list, &ivshmem->list_node); |
95fb793a | 2599 | |
2600 | *eth_port_id = ivshmem->eth_port_id; | |
2601 | return 0; | |
2602 | } | |
2603 | ||
2604 | static int | |
64839cf4 WT |
2605 | dpdk_ring_open(const char dev_name[], unsigned int *eth_port_id) |
2606 | OVS_REQUIRES(dpdk_mutex) | |
95fb793a | 2607 | { |
2608 | struct dpdk_ring *ivshmem; | |
2609 | unsigned int port_no; | |
2610 | int err = 0; | |
2611 | ||
2612 | /* Names always start with "dpdkr" */ | |
2613 | err = dpdk_dev_parse_name(dev_name, "dpdkr", &port_no); | |
2614 | if (err) { | |
2615 | return err; | |
2616 | } | |
2617 | ||
2618 | /* look through our list to find the device */ | |
2619 | LIST_FOR_EACH (ivshmem, list_node, &dpdk_ring_list) { | |
2620 | if (ivshmem->user_port_id == port_no) { | |
58397e6c | 2621 | VLOG_INFO("Found dpdk ring device %s:", dev_name); |
95fb793a | 2622 | *eth_port_id = ivshmem->eth_port_id; /* really all that is needed */ |
2623 | return 0; | |
2624 | } | |
2625 | } | |
2626 | /* Need to create the device rings */ | |
2627 | return dpdk_ring_create(dev_name, port_no, eth_port_id); | |
2628 | } | |
2629 | ||
7251515e | 2630 | static int |
d46285a2 | 2631 | netdev_dpdk_ring_send(struct netdev *netdev, int qid, |
324c8374 IM |
2632 | struct dp_packet_batch *batch, bool may_steal, |
2633 | bool concurrent_txq) | |
7251515e | 2634 | { |
d46285a2 | 2635 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
1b99bb05 MG |
2636 | unsigned i; |
2637 | ||
2638 | /* When using 'dpdkr' and sending to a DPDK ring, we want to ensure that the | |
2639 | * rss hash field is clear. This is because the same mbuf may be modified by | |
2640 | * the consumer of the ring and return into the datapath without recalculating | |
2641 | * the RSS hash. */ | |
64839cf4 WT |
2642 | for (i = 0; i < batch->count; i++) { |
2643 | dp_packet_rss_invalidate(batch->packets[i]); | |
1b99bb05 | 2644 | } |
7251515e | 2645 | |
324c8374 | 2646 | netdev_dpdk_send__(dev, qid, batch, may_steal, concurrent_txq); |
7251515e DV |
2647 | return 0; |
2648 | } | |
2649 | ||
95fb793a | 2650 | static int |
2651 | netdev_dpdk_ring_construct(struct netdev *netdev) | |
2652 | { | |
2653 | unsigned int port_no = 0; | |
2654 | int err = 0; | |
2655 | ||
2656 | if (rte_eal_init_ret) { | |
2657 | return rte_eal_init_ret; | |
2658 | } | |
2659 | ||
2660 | ovs_mutex_lock(&dpdk_mutex); | |
2661 | ||
2662 | err = dpdk_ring_open(netdev->name, &port_no); | |
2663 | if (err) { | |
2664 | goto unlock_dpdk; | |
2665 | } | |
2666 | ||
58397e6c | 2667 | err = netdev_dpdk_init(netdev, port_no, DPDK_DEV_ETH); |
95fb793a | 2668 | |
2669 | unlock_dpdk: | |
2670 | ovs_mutex_unlock(&dpdk_mutex); | |
2671 | return err; | |
2672 | } | |
2673 | ||
0bf765f7 IS |
2674 | /* QoS Functions */ |
2675 | ||
2676 | /* | |
2677 | * Initialize QoS configuration operations. | |
2678 | */ | |
2679 | static void | |
2680 | qos_conf_init(struct qos_conf *conf, const struct dpdk_qos_ops *ops) | |
2681 | { | |
2682 | conf->ops = ops; | |
2683 | } | |
2684 | ||
2685 | /* | |
2686 | * Search existing QoS operations in qos_ops and compare each set of | |
2687 | * operations qos_name to name. Return a dpdk_qos_ops pointer to a match, | |
2688 | * else return NULL | |
2689 | */ | |
2690 | static const struct dpdk_qos_ops * | |
2691 | qos_lookup_name(const char *name) | |
2692 | { | |
2693 | const struct dpdk_qos_ops *const *opsp; | |
2694 | ||
2695 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2696 | const struct dpdk_qos_ops *ops = *opsp; | |
2697 | if (!strcmp(name, ops->qos_name)) { | |
2698 | return ops; | |
2699 | } | |
2700 | } | |
2701 | return NULL; | |
2702 | } | |
2703 | ||
2704 | /* | |
2705 | * Call qos_destruct to clean up items associated with the netdevs | |
2706 | * qos_conf. Set netdevs qos_conf to NULL. | |
2707 | */ | |
2708 | static void | |
d46285a2 | 2709 | qos_delete_conf(struct netdev *netdev) |
0bf765f7 | 2710 | { |
d46285a2 | 2711 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2712 | |
d46285a2 DDP |
2713 | rte_spinlock_lock(&dev->qos_lock); |
2714 | if (dev->qos_conf) { | |
2715 | if (dev->qos_conf->ops->qos_destruct) { | |
2716 | dev->qos_conf->ops->qos_destruct(netdev, dev->qos_conf); | |
0bf765f7 | 2717 | } |
d46285a2 | 2718 | dev->qos_conf = NULL; |
0bf765f7 | 2719 | } |
d46285a2 | 2720 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2721 | } |
2722 | ||
2723 | static int | |
2724 | netdev_dpdk_get_qos_types(const struct netdev *netdev OVS_UNUSED, | |
2725 | struct sset *types) | |
2726 | { | |
2727 | const struct dpdk_qos_ops *const *opsp; | |
2728 | ||
2729 | for (opsp = qos_confs; *opsp != NULL; opsp++) { | |
2730 | const struct dpdk_qos_ops *ops = *opsp; | |
2731 | if (ops->qos_construct && ops->qos_name[0] != '\0') { | |
2732 | sset_add(types, ops->qos_name); | |
2733 | } | |
2734 | } | |
2735 | return 0; | |
2736 | } | |
2737 | ||
2738 | static int | |
d46285a2 | 2739 | netdev_dpdk_get_qos(const struct netdev *netdev, |
0bf765f7 IS |
2740 | const char **typep, struct smap *details) |
2741 | { | |
d46285a2 | 2742 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2743 | int error = 0; |
2744 | ||
d46285a2 DDP |
2745 | ovs_mutex_lock(&dev->mutex); |
2746 | if(dev->qos_conf) { | |
2747 | *typep = dev->qos_conf->ops->qos_name; | |
2748 | error = (dev->qos_conf->ops->qos_get | |
2749 | ? dev->qos_conf->ops->qos_get(netdev, details): 0); | |
d03603c4 MC |
2750 | } else { |
2751 | /* No QoS configuration set, return an empty string */ | |
2752 | *typep = ""; | |
0bf765f7 | 2753 | } |
d46285a2 | 2754 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2755 | |
2756 | return error; | |
2757 | } | |
2758 | ||
2759 | static int | |
d46285a2 | 2760 | netdev_dpdk_set_qos(struct netdev *netdev, |
0bf765f7 IS |
2761 | const char *type, const struct smap *details) |
2762 | { | |
d46285a2 | 2763 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2764 | const struct dpdk_qos_ops *new_ops = NULL; |
2765 | int error = 0; | |
2766 | ||
2767 | /* If type is empty or unsupported then the current QoS configuration | |
2768 | * for the dpdk-netdev can be destroyed */ | |
2769 | new_ops = qos_lookup_name(type); | |
2770 | ||
2771 | if (type[0] == '\0' || !new_ops || !new_ops->qos_construct) { | |
d46285a2 | 2772 | qos_delete_conf(netdev); |
0bf765f7 IS |
2773 | return EOPNOTSUPP; |
2774 | } | |
2775 | ||
d46285a2 | 2776 | ovs_mutex_lock(&dev->mutex); |
0bf765f7 | 2777 | |
d46285a2 DDP |
2778 | if (dev->qos_conf) { |
2779 | if (new_ops == dev->qos_conf->ops) { | |
2780 | error = new_ops->qos_set ? new_ops->qos_set(netdev, details) : 0; | |
0bf765f7 IS |
2781 | } else { |
2782 | /* Delete existing QoS configuration. */ | |
d46285a2 DDP |
2783 | qos_delete_conf(netdev); |
2784 | ovs_assert(dev->qos_conf == NULL); | |
0bf765f7 IS |
2785 | |
2786 | /* Install new QoS configuration. */ | |
d46285a2 | 2787 | error = new_ops->qos_construct(netdev, details); |
0bf765f7 IS |
2788 | } |
2789 | } else { | |
d46285a2 | 2790 | error = new_ops->qos_construct(netdev, details); |
7ea266e9 IS |
2791 | } |
2792 | ||
2793 | ovs_assert((error == 0) == (dev->qos_conf != NULL)); | |
2794 | if (error) { | |
2795 | VLOG_ERR("Failed to set QoS type %s on port %s, returned error: %s", | |
2796 | type, netdev->name, rte_strerror(-error)); | |
0bf765f7 IS |
2797 | } |
2798 | ||
d46285a2 | 2799 | ovs_mutex_unlock(&dev->mutex); |
0bf765f7 IS |
2800 | return error; |
2801 | } | |
2802 | ||
2803 | /* egress-policer details */ | |
2804 | ||
2805 | struct egress_policer { | |
2806 | struct qos_conf qos_conf; | |
2807 | struct rte_meter_srtcm_params app_srtcm_params; | |
2808 | struct rte_meter_srtcm egress_meter; | |
2809 | }; | |
2810 | ||
2811 | static struct egress_policer * | |
d46285a2 | 2812 | egress_policer_get__(const struct netdev *netdev) |
0bf765f7 | 2813 | { |
d46285a2 DDP |
2814 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
2815 | return CONTAINER_OF(dev->qos_conf, struct egress_policer, qos_conf); | |
0bf765f7 IS |
2816 | } |
2817 | ||
2818 | static int | |
d46285a2 DDP |
2819 | egress_policer_qos_construct(struct netdev *netdev, |
2820 | const struct smap *details) | |
0bf765f7 | 2821 | { |
d46285a2 | 2822 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 | 2823 | struct egress_policer *policer; |
0bf765f7 IS |
2824 | int err = 0; |
2825 | ||
d46285a2 | 2826 | rte_spinlock_lock(&dev->qos_lock); |
0bf765f7 IS |
2827 | policer = xmalloc(sizeof *policer); |
2828 | qos_conf_init(&policer->qos_conf, &egress_policer_ops); | |
d46285a2 | 2829 | dev->qos_conf = &policer->qos_conf; |
13c1637f BP |
2830 | policer->app_srtcm_params.cir = smap_get_ullong(details, "cir", 0); |
2831 | policer->app_srtcm_params.cbs = smap_get_ullong(details, "cbs", 0); | |
0bf765f7 IS |
2832 | policer->app_srtcm_params.ebs = 0; |
2833 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2834 | &policer->app_srtcm_params); | |
7ea266e9 IS |
2835 | |
2836 | if (err < 0) { | |
2837 | /* Error occurred during rte_meter creation, destroy the policer | |
2838 | * and set the qos configuration for the netdev dpdk to NULL | |
2839 | */ | |
2840 | free(policer); | |
2841 | dev->qos_conf = NULL; | |
2842 | err = -err; | |
2843 | } | |
d46285a2 | 2844 | rte_spinlock_unlock(&dev->qos_lock); |
0bf765f7 IS |
2845 | |
2846 | return err; | |
2847 | } | |
2848 | ||
2849 | static void | |
d46285a2 | 2850 | egress_policer_qos_destruct(struct netdev *netdev OVS_UNUSED, |
0bf765f7 IS |
2851 | struct qos_conf *conf) |
2852 | { | |
2853 | struct egress_policer *policer = CONTAINER_OF(conf, struct egress_policer, | |
2854 | qos_conf); | |
2855 | free(policer); | |
2856 | } | |
2857 | ||
2858 | static int | |
2859 | egress_policer_qos_get(const struct netdev *netdev, struct smap *details) | |
2860 | { | |
2861 | struct egress_policer *policer = egress_policer_get__(netdev); | |
2862 | smap_add_format(details, "cir", "%llu", | |
2863 | 1ULL * policer->app_srtcm_params.cir); | |
2864 | smap_add_format(details, "cbs", "%llu", | |
2865 | 1ULL * policer->app_srtcm_params.cbs); | |
050c60bf | 2866 | |
0bf765f7 IS |
2867 | return 0; |
2868 | } | |
2869 | ||
2870 | static int | |
d46285a2 | 2871 | egress_policer_qos_set(struct netdev *netdev, const struct smap *details) |
0bf765f7 IS |
2872 | { |
2873 | struct egress_policer *policer; | |
7ea266e9 | 2874 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); |
0bf765f7 IS |
2875 | int err = 0; |
2876 | ||
d46285a2 | 2877 | policer = egress_policer_get__(netdev); |
7ea266e9 | 2878 | rte_spinlock_lock(&dev->qos_lock); |
13c1637f BP |
2879 | policer->app_srtcm_params.cir = smap_get_ullong(details, "cir", 0); |
2880 | policer->app_srtcm_params.cbs = smap_get_ullong(details, "cbs", 0); | |
0bf765f7 IS |
2881 | policer->app_srtcm_params.ebs = 0; |
2882 | err = rte_meter_srtcm_config(&policer->egress_meter, | |
2883 | &policer->app_srtcm_params); | |
2884 | ||
7ea266e9 IS |
2885 | if (err < 0) { |
2886 | /* Error occurred during rte_meter creation, destroy the policer | |
2887 | * and set the qos configuration for the netdev dpdk to NULL | |
2888 | */ | |
2889 | free(policer); | |
2890 | dev->qos_conf = NULL; | |
2891 | err = -err; | |
2892 | } | |
2893 | rte_spinlock_unlock(&dev->qos_lock); | |
2894 | ||
0bf765f7 IS |
2895 | return err; |
2896 | } | |
2897 | ||
0bf765f7 | 2898 | static int |
f3926f29 | 2899 | egress_policer_run(struct netdev *netdev, struct rte_mbuf **pkts, int pkt_cnt) |
0bf765f7 | 2900 | { |
0bf765f7 | 2901 | int cnt = 0; |
d46285a2 | 2902 | struct egress_policer *policer = egress_policer_get__(netdev); |
0bf765f7 | 2903 | |
f3926f29 | 2904 | cnt = netdev_dpdk_policer_run(&policer->egress_meter, pkts, pkt_cnt); |
0bf765f7 IS |
2905 | |
2906 | return cnt; | |
2907 | } | |
2908 | ||
2909 | static const struct dpdk_qos_ops egress_policer_ops = { | |
2910 | "egress-policer", /* qos_name */ | |
2911 | egress_policer_qos_construct, | |
2912 | egress_policer_qos_destruct, | |
2913 | egress_policer_qos_get, | |
2914 | egress_policer_qos_set, | |
2915 | egress_policer_run | |
2916 | }; | |
2917 | ||
050c60bf DDP |
2918 | static int |
2919 | netdev_dpdk_reconfigure(struct netdev *netdev) | |
2920 | { | |
2921 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2922 | int err = 0; | |
2923 | ||
2924 | ovs_mutex_lock(&dpdk_mutex); | |
2925 | ovs_mutex_lock(&dev->mutex); | |
2926 | ||
2927 | if (netdev->n_txq == dev->requested_n_txq | |
0072e931 MK |
2928 | && netdev->n_rxq == dev->requested_n_rxq |
2929 | && dev->mtu == dev->requested_mtu) { | |
050c60bf DDP |
2930 | /* Reconfiguration is unnecessary */ |
2931 | ||
2932 | goto out; | |
2933 | } | |
2934 | ||
2935 | rte_eth_dev_stop(dev->port_id); | |
2936 | ||
0072e931 MK |
2937 | if (dev->mtu != dev->requested_mtu) { |
2938 | netdev_dpdk_mempool_configure(dev); | |
2939 | } | |
2940 | ||
050c60bf DDP |
2941 | netdev->n_txq = dev->requested_n_txq; |
2942 | netdev->n_rxq = dev->requested_n_rxq; | |
2943 | ||
2944 | rte_free(dev->tx_q); | |
2945 | err = dpdk_eth_dev_init(dev); | |
81acebda | 2946 | netdev_dpdk_alloc_txq(dev, netdev->n_txq); |
050c60bf | 2947 | |
0072e931 MK |
2948 | netdev_change_seq_changed(netdev); |
2949 | ||
050c60bf DDP |
2950 | out: |
2951 | ||
2952 | ovs_mutex_unlock(&dev->mutex); | |
2953 | ovs_mutex_unlock(&dpdk_mutex); | |
2954 | ||
2955 | return err; | |
2956 | } | |
2957 | ||
2d24d165 CL |
2958 | static void |
2959 | dpdk_vhost_reconfigure_helper(struct netdev_dpdk *dev) | |
2960 | OVS_REQUIRES(dpdk_mutex) | |
2961 | OVS_REQUIRES(dev->mutex) | |
050c60bf | 2962 | { |
2d24d165 CL |
2963 | dev->up.n_txq = dev->requested_n_txq; |
2964 | dev->up.n_rxq = dev->requested_n_rxq; | |
050c60bf | 2965 | |
81acebda IM |
2966 | /* Enable TX queue 0 by default if it wasn't disabled. */ |
2967 | if (dev->tx_q[0].map == OVS_VHOST_QUEUE_MAP_UNKNOWN) { | |
2968 | dev->tx_q[0].map = 0; | |
2969 | } | |
2970 | ||
2971 | netdev_dpdk_remap_txqs(dev); | |
2972 | ||
0072e931 MK |
2973 | if (dev->requested_socket_id != dev->socket_id |
2974 | || dev->requested_mtu != dev->mtu) { | |
2975 | if (!netdev_dpdk_mempool_configure(dev)) { | |
2d24d165 | 2976 | netdev_change_seq_changed(&dev->up); |
db8f13b0 CL |
2977 | } |
2978 | } | |
2979 | ||
0a0f39df CL |
2980 | if (netdev_dpdk_get_vid(dev) >= 0) { |
2981 | dev->vhost_reconfigured = true; | |
81acebda | 2982 | } |
2d24d165 CL |
2983 | } |
2984 | ||
2985 | static int | |
2986 | netdev_dpdk_vhost_reconfigure(struct netdev *netdev) | |
2987 | { | |
2988 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
2989 | ||
2990 | ovs_mutex_lock(&dpdk_mutex); | |
2991 | ovs_mutex_lock(&dev->mutex); | |
2992 | ||
2993 | dpdk_vhost_reconfigure_helper(dev); | |
2994 | ||
2995 | ovs_mutex_unlock(&dev->mutex); | |
2996 | ovs_mutex_unlock(&dpdk_mutex); | |
2997 | ||
2998 | return 0; | |
2999 | } | |
3000 | ||
3001 | static int | |
3002 | netdev_dpdk_vhost_client_reconfigure(struct netdev *netdev) | |
3003 | { | |
3004 | struct netdev_dpdk *dev = netdev_dpdk_cast(netdev); | |
3005 | int err = 0; | |
3006 | ||
3007 | ovs_mutex_lock(&dpdk_mutex); | |
3008 | ovs_mutex_lock(&dev->mutex); | |
3009 | ||
3010 | dpdk_vhost_reconfigure_helper(dev); | |
81acebda | 3011 | |
c1ff66ac CL |
3012 | /* Configure vHost client mode if requested and if the following criteria |
3013 | * are met: | |
2d24d165 CL |
3014 | * 1. Device hasn't been registered yet. |
3015 | * 2. A path has been specified. | |
c1ff66ac CL |
3016 | */ |
3017 | if (!(dev->vhost_driver_flags & RTE_VHOST_USER_CLIENT) | |
2d24d165 CL |
3018 | && strlen(dev->vhost_id)) { |
3019 | /* Register client-mode device */ | |
3020 | err = rte_vhost_driver_register(dev->vhost_id, | |
3021 | RTE_VHOST_USER_CLIENT); | |
c1ff66ac | 3022 | if (err) { |
2d24d165 CL |
3023 | VLOG_ERR("vhost-user device setup failure for device %s\n", |
3024 | dev->vhost_id); | |
c1ff66ac | 3025 | } else { |
2d24d165 CL |
3026 | /* Configuration successful */ |
3027 | dev->vhost_driver_flags |= RTE_VHOST_USER_CLIENT; | |
3028 | VLOG_INFO("vHost User device '%s' created in 'client' mode, " | |
3029 | "using client socket '%s'", | |
3030 | dev->up.name, dev->vhost_id); | |
c1ff66ac CL |
3031 | } |
3032 | } | |
3033 | ||
050c60bf DDP |
3034 | ovs_mutex_unlock(&dev->mutex); |
3035 | ovs_mutex_unlock(&dpdk_mutex); | |
3036 | ||
0072e931 | 3037 | return 0; |
050c60bf DDP |
3038 | } |
3039 | ||
53f50d24 | 3040 | #define NETDEV_DPDK_CLASS(NAME, CONSTRUCT, DESTRUCT, \ |
81acebda IM |
3041 | SET_CONFIG, SET_TX_MULTIQ, SEND, \ |
3042 | GET_CARRIER, GET_STATS, \ | |
3043 | GET_FEATURES, GET_STATUS, \ | |
3044 | RECONFIGURE, RXQ_RECV) \ | |
95fb793a | 3045 | { \ |
3046 | NAME, \ | |
118c77b1 | 3047 | true, /* is_pmd */ \ |
53f50d24 | 3048 | NULL, /* init */ \ |
95fb793a | 3049 | NULL, /* netdev_dpdk_run */ \ |
3050 | NULL, /* netdev_dpdk_wait */ \ | |
3051 | \ | |
3052 | netdev_dpdk_alloc, \ | |
3053 | CONSTRUCT, \ | |
58397e6c | 3054 | DESTRUCT, \ |
95fb793a | 3055 | netdev_dpdk_dealloc, \ |
3056 | netdev_dpdk_get_config, \ | |
81acebda | 3057 | SET_CONFIG, \ |
95fb793a | 3058 | NULL, /* get_tunnel_config */ \ |
58397e6c KT |
3059 | NULL, /* build header */ \ |
3060 | NULL, /* push header */ \ | |
3061 | NULL, /* pop header */ \ | |
7dec44fe | 3062 | netdev_dpdk_get_numa_id, /* get_numa_id */ \ |
81acebda | 3063 | SET_TX_MULTIQ, \ |
95fb793a | 3064 | \ |
7251515e | 3065 | SEND, /* send */ \ |
95fb793a | 3066 | NULL, /* send_wait */ \ |
3067 | \ | |
3068 | netdev_dpdk_set_etheraddr, \ | |
3069 | netdev_dpdk_get_etheraddr, \ | |
3070 | netdev_dpdk_get_mtu, \ | |
0072e931 | 3071 | netdev_dpdk_set_mtu, \ |
95fb793a | 3072 | netdev_dpdk_get_ifindex, \ |
58397e6c | 3073 | GET_CARRIER, \ |
95fb793a | 3074 | netdev_dpdk_get_carrier_resets, \ |
3075 | netdev_dpdk_set_miimon, \ | |
58397e6c KT |
3076 | GET_STATS, \ |
3077 | GET_FEATURES, \ | |
95fb793a | 3078 | NULL, /* set_advertisements */ \ |
3079 | \ | |
9509913a | 3080 | netdev_dpdk_set_policing, \ |
0bf765f7 | 3081 | netdev_dpdk_get_qos_types, \ |
95fb793a | 3082 | NULL, /* get_qos_capabilities */ \ |
0bf765f7 IS |
3083 | netdev_dpdk_get_qos, \ |
3084 | netdev_dpdk_set_qos, \ | |
95fb793a | 3085 | NULL, /* get_queue */ \ |
3086 | NULL, /* set_queue */ \ | |
3087 | NULL, /* delete_queue */ \ | |
3088 | NULL, /* get_queue_stats */ \ | |
3089 | NULL, /* queue_dump_start */ \ | |
3090 | NULL, /* queue_dump_next */ \ | |
3091 | NULL, /* queue_dump_done */ \ | |
3092 | NULL, /* dump_queue_stats */ \ | |
3093 | \ | |
95fb793a | 3094 | NULL, /* set_in4 */ \ |
a8704b50 | 3095 | NULL, /* get_addr_list */ \ |
95fb793a | 3096 | NULL, /* add_router */ \ |
3097 | NULL, /* get_next_hop */ \ | |
58397e6c | 3098 | GET_STATUS, \ |
95fb793a | 3099 | NULL, /* arp_lookup */ \ |
3100 | \ | |
3101 | netdev_dpdk_update_flags, \ | |
050c60bf | 3102 | RECONFIGURE, \ |
95fb793a | 3103 | \ |
3104 | netdev_dpdk_rxq_alloc, \ | |
3105 | netdev_dpdk_rxq_construct, \ | |
3106 | netdev_dpdk_rxq_destruct, \ | |
3107 | netdev_dpdk_rxq_dealloc, \ | |
58397e6c | 3108 | RXQ_RECV, \ |
95fb793a | 3109 | NULL, /* rx_wait */ \ |
3110 | NULL, /* rxq_drain */ \ | |
3111 | } | |
8a9562d2 | 3112 | |
7d1ced01 CL |
3113 | static int |
3114 | process_vhost_flags(char *flag, char *default_val, int size, | |
bab69409 AC |
3115 | const struct smap *ovs_other_config, |
3116 | char **new_val) | |
7d1ced01 | 3117 | { |
bab69409 | 3118 | const char *val; |
7d1ced01 CL |
3119 | int changed = 0; |
3120 | ||
bab69409 AC |
3121 | val = smap_get(ovs_other_config, flag); |
3122 | ||
41987644 CL |
3123 | /* Process the vhost-sock-dir flag if it is provided, otherwise resort to |
3124 | * default value. | |
7d1ced01 | 3125 | */ |
bab69409 | 3126 | if (val && (strlen(val) <= size)) { |
7d1ced01 | 3127 | changed = 1; |
bab69409 | 3128 | *new_val = xstrdup(val); |
7d1ced01 CL |
3129 | VLOG_INFO("User-provided %s in use: %s", flag, *new_val); |
3130 | } else { | |
3131 | VLOG_INFO("No %s provided - defaulting to %s", flag, default_val); | |
3132 | *new_val = default_val; | |
3133 | } | |
3134 | ||
3135 | return changed; | |
3136 | } | |
3137 | ||
bab69409 AC |
3138 | static char ** |
3139 | grow_argv(char ***argv, size_t cur_siz, size_t grow_by) | |
8a9562d2 | 3140 | { |
bab69409 AC |
3141 | return xrealloc(*argv, sizeof(char *) * (cur_siz + grow_by)); |
3142 | } | |
8a9562d2 | 3143 | |
bab69409 AC |
3144 | static void |
3145 | dpdk_option_extend(char ***argv, int argc, const char *option, | |
3146 | const char *value) | |
3147 | { | |
3148 | char **newargv = grow_argv(argv, argc, 2); | |
3149 | *argv = newargv; | |
3150 | newargv[argc] = xstrdup(option); | |
3151 | newargv[argc+1] = xstrdup(value); | |
3152 | } | |
8a9562d2 | 3153 | |
52a57e36 AC |
3154 | static char ** |
3155 | move_argv(char ***argv, size_t cur_size, char **src_argv, size_t src_argc) | |
3156 | { | |
3157 | char **newargv = grow_argv(argv, cur_size, src_argc); | |
3158 | while (src_argc--) { | |
3159 | newargv[cur_size+src_argc] = src_argv[src_argc]; | |
3160 | src_argv[src_argc] = NULL; | |
3161 | } | |
3162 | return newargv; | |
3163 | } | |
3164 | ||
eac84432 AC |
3165 | static int |
3166 | extra_dpdk_args(const char *ovs_extra_config, char ***argv, int argc) | |
3167 | { | |
3168 | int ret = argc; | |
3169 | char *release_tok = xstrdup(ovs_extra_config); | |
27373420 | 3170 | char *tok, *endptr = NULL; |
eac84432 AC |
3171 | |
3172 | for (tok = strtok_r(release_tok, " ", &endptr); tok != NULL; | |
3173 | tok = strtok_r(NULL, " ", &endptr)) { | |
3174 | char **newarg = grow_argv(argv, ret, 1); | |
3175 | *argv = newarg; | |
3176 | newarg[ret++] = xstrdup(tok); | |
3177 | } | |
3178 | free(release_tok); | |
3179 | return ret; | |
3180 | } | |
3181 | ||
52a57e36 AC |
3182 | static bool |
3183 | argv_contains(char **argv_haystack, const size_t argc_haystack, | |
3184 | const char *needle) | |
3185 | { | |
3186 | for (size_t i = 0; i < argc_haystack; ++i) { | |
3187 | if (!strcmp(argv_haystack[i], needle)) | |
3188 | return true; | |
3189 | } | |
3190 | return false; | |
3191 | } | |
3192 | ||
bab69409 AC |
3193 | static int |
3194 | construct_dpdk_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
3195 | char ***argv, const int initial_size, |
3196 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
3197 | { |
3198 | struct dpdk_options_map { | |
3199 | const char *ovs_configuration; | |
3200 | const char *dpdk_option; | |
3201 | bool default_enabled; | |
3202 | const char *default_value; | |
3203 | } opts[] = { | |
3204 | {"dpdk-lcore-mask", "-c", false, NULL}, | |
3205 | {"dpdk-hugepage-dir", "--huge-dir", false, NULL}, | |
3206 | }; | |
3207 | ||
3208 | int i, ret = initial_size; | |
3209 | ||
3210 | /*First, construct from the flat-options (non-mutex)*/ | |
3211 | for (i = 0; i < ARRAY_SIZE(opts); ++i) { | |
3212 | const char *lookup = smap_get(ovs_other_config, | |
3213 | opts[i].ovs_configuration); | |
3214 | if (!lookup && opts[i].default_enabled) { | |
3215 | lookup = opts[i].default_value; | |
3216 | } | |
8a9562d2 | 3217 | |
bab69409 | 3218 | if (lookup) { |
52a57e36 AC |
3219 | if (!argv_contains(extra_args, extra_argc, opts[i].dpdk_option)) { |
3220 | dpdk_option_extend(argv, ret, opts[i].dpdk_option, lookup); | |
3221 | ret += 2; | |
3222 | } else { | |
3223 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3224 | "dpdk_extras config", opts[i].dpdk_option); | |
3225 | } | |
bab69409 AC |
3226 | } |
3227 | } | |
3228 | ||
3229 | return ret; | |
3230 | } | |
3231 | ||
3232 | #define MAX_DPDK_EXCL_OPTS 10 | |
3233 | ||
3234 | static int | |
3235 | construct_dpdk_mutex_options(const struct smap *ovs_other_config, | |
52a57e36 AC |
3236 | char ***argv, const int initial_size, |
3237 | char **extra_args, const size_t extra_argc) | |
bab69409 AC |
3238 | { |
3239 | struct dpdk_exclusive_options_map { | |
3240 | const char *category; | |
3241 | const char *ovs_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
3242 | const char *eal_dpdk_options[MAX_DPDK_EXCL_OPTS]; | |
3243 | const char *default_value; | |
3244 | int default_option; | |
3245 | } excl_opts[] = { | |
3246 | {"memory type", | |
3247 | {"dpdk-alloc-mem", "dpdk-socket-mem", NULL,}, | |
3248 | {"-m", "--socket-mem", NULL,}, | |
3249 | "1024,0", 1 | |
3250 | }, | |
3251 | }; | |
3252 | ||
3253 | int i, ret = initial_size; | |
3254 | for (i = 0; i < ARRAY_SIZE(excl_opts); ++i) { | |
3255 | int found_opts = 0, scan, found_pos = -1; | |
3256 | const char *found_value; | |
3257 | struct dpdk_exclusive_options_map *popt = &excl_opts[i]; | |
3258 | ||
3259 | for (scan = 0; scan < MAX_DPDK_EXCL_OPTS | |
3260 | && popt->ovs_dpdk_options[scan]; ++scan) { | |
3261 | const char *lookup = smap_get(ovs_other_config, | |
3262 | popt->ovs_dpdk_options[scan]); | |
3263 | if (lookup && strlen(lookup)) { | |
3264 | found_opts++; | |
3265 | found_pos = scan; | |
3266 | found_value = lookup; | |
3267 | } | |
3268 | } | |
3269 | ||
3270 | if (!found_opts) { | |
3271 | if (popt->default_option) { | |
3272 | found_pos = popt->default_option; | |
3273 | found_value = popt->default_value; | |
3274 | } else { | |
3275 | continue; | |
3276 | } | |
dfc89a55 | 3277 | } |
bab69409 AC |
3278 | |
3279 | if (found_opts > 1) { | |
3280 | VLOG_ERR("Multiple defined options for %s. Please check your" | |
3281 | " database settings and reconfigure if necessary.", | |
3282 | popt->category); | |
3283 | } | |
3284 | ||
52a57e36 AC |
3285 | if (!argv_contains(extra_args, extra_argc, |
3286 | popt->eal_dpdk_options[found_pos])) { | |
3287 | dpdk_option_extend(argv, ret, popt->eal_dpdk_options[found_pos], | |
3288 | found_value); | |
3289 | ret += 2; | |
3290 | } else { | |
3291 | VLOG_WARN("Ignoring database defined option '%s' due to " | |
3292 | "dpdk_extras config", popt->eal_dpdk_options[found_pos]); | |
3293 | } | |
bab69409 AC |
3294 | } |
3295 | ||
3296 | return ret; | |
3297 | } | |
3298 | ||
3299 | static int | |
88964e64 AC |
3300 | get_dpdk_args(const struct smap *ovs_other_config, char ***argv, |
3301 | int argc) | |
bab69409 | 3302 | { |
eac84432 | 3303 | const char *extra_configuration; |
52a57e36 AC |
3304 | char **extra_args = NULL; |
3305 | int i; | |
3306 | size_t extra_argc = 0; | |
eac84432 AC |
3307 | |
3308 | extra_configuration = smap_get(ovs_other_config, "dpdk-extra"); | |
3309 | if (extra_configuration) { | |
52a57e36 | 3310 | extra_argc = extra_dpdk_args(extra_configuration, &extra_args, 0); |
eac84432 | 3311 | } |
52a57e36 AC |
3312 | |
3313 | i = construct_dpdk_options(ovs_other_config, argv, argc, extra_args, | |
3314 | extra_argc); | |
3315 | i = construct_dpdk_mutex_options(ovs_other_config, argv, i, extra_args, | |
3316 | extra_argc); | |
3317 | ||
3318 | if (extra_configuration) { | |
3319 | *argv = move_argv(argv, i, extra_args, extra_argc); | |
3320 | } | |
3321 | ||
3322 | return i + extra_argc; | |
bab69409 AC |
3323 | } |
3324 | ||
3325 | static char **dpdk_argv; | |
3326 | static int dpdk_argc; | |
3327 | ||
3328 | static void | |
3329 | deferred_argv_release(void) | |
3330 | { | |
3331 | int result; | |
3332 | for (result = 0; result < dpdk_argc; ++result) { | |
3333 | free(dpdk_argv[result]); | |
3334 | } | |
3335 | ||
3336 | free(dpdk_argv); | |
3337 | } | |
3338 | ||
3339 | static void | |
3340 | dpdk_init__(const struct smap *ovs_other_config) | |
3341 | { | |
3342 | char **argv = NULL; | |
3343 | int result; | |
88964e64 AC |
3344 | int argc, argc_tmp; |
3345 | bool auto_determine = true; | |
d8e2f4cc | 3346 | int err = 0; |
bab69409 | 3347 | cpu_set_t cpuset; |
d8a8f353 | 3348 | char *sock_dir_subcomponent; |
bab69409 AC |
3349 | |
3350 | if (!smap_get_bool(ovs_other_config, "dpdk-init", false)) { | |
3351 | VLOG_INFO("DPDK Disabled - to change this requires a restart.\n"); | |
3352 | return; | |
dfc89a55 AZ |
3353 | } |
3354 | ||
bab69409 | 3355 | VLOG_INFO("DPDK Enabled, initializing"); |
6eb51f9a | 3356 | if (process_vhost_flags("vhost-sock-dir", xstrdup(ovs_rundir()), |
d8a8f353 AC |
3357 | NAME_MAX, ovs_other_config, |
3358 | &sock_dir_subcomponent)) { | |
7d1ced01 | 3359 | struct stat s; |
d8a8f353 AC |
3360 | if (!strstr(sock_dir_subcomponent, "..")) { |
3361 | vhost_sock_dir = xasprintf("%s/%s", ovs_rundir(), | |
3362 | sock_dir_subcomponent); | |
3363 | ||
3364 | err = stat(vhost_sock_dir, &s); | |
3365 | if (err) { | |
3366 | VLOG_ERR("vhost-user sock directory '%s' does not exist.", | |
3367 | vhost_sock_dir); | |
3368 | } | |
3369 | } else { | |
3370 | vhost_sock_dir = xstrdup(ovs_rundir()); | |
3371 | VLOG_ERR("vhost-user sock directory request '%s/%s' has invalid" | |
3372 | "characters '..' - using %s instead.", | |
3373 | ovs_rundir(), sock_dir_subcomponent, ovs_rundir()); | |
7d1ced01 | 3374 | } |
d8a8f353 AC |
3375 | free(sock_dir_subcomponent); |
3376 | } else { | |
6eb51f9a | 3377 | vhost_sock_dir = sock_dir_subcomponent; |
58397e6c KT |
3378 | } |
3379 | ||
bab69409 | 3380 | argv = grow_argv(&argv, 0, 1); |
88964e64 | 3381 | argc = 1; |
bab69409 | 3382 | argv[0] = xstrdup(ovs_get_program_name()); |
88964e64 AC |
3383 | argc_tmp = get_dpdk_args(ovs_other_config, &argv, argc); |
3384 | ||
3385 | while (argc_tmp != argc) { | |
3386 | if (!strcmp("-c", argv[argc]) || !strcmp("-l", argv[argc])) { | |
3387 | auto_determine = false; | |
3388 | break; | |
3389 | } | |
3390 | argc++; | |
3391 | } | |
3392 | argc = argc_tmp; | |
3393 | ||
3394 | /** | |
3395 | * NOTE: This is an unsophisticated mechanism for determining the DPDK | |
3396 | * lcore for the DPDK Master. | |
3397 | */ | |
3398 | if (auto_determine) { | |
3399 | int i; | |
d8e2f4cc KT |
3400 | /* Get the main thread affinity */ |
3401 | CPU_ZERO(&cpuset); | |
3402 | err = pthread_getaffinity_np(pthread_self(), sizeof(cpu_set_t), | |
3403 | &cpuset); | |
3404 | if (!err) { | |
3405 | for (i = 0; i < CPU_SETSIZE; i++) { | |
3406 | if (CPU_ISSET(i, &cpuset)) { | |
3407 | argv = grow_argv(&argv, argc, 2); | |
3408 | argv[argc++] = xstrdup("-c"); | |
3409 | argv[argc++] = xasprintf("0x%08llX", (1ULL<<i)); | |
3410 | i = CPU_SETSIZE; | |
3411 | } | |
88964e64 | 3412 | } |
d8e2f4cc KT |
3413 | } else { |
3414 | VLOG_ERR("Thread getaffinity error %d. Using core 0x1", err); | |
3415 | /* User did not set dpdk-lcore-mask and unable to get current | |
3416 | * thread affintity - default to core 0x1 */ | |
3417 | argv = grow_argv(&argv, argc, 2); | |
3418 | argv[argc++] = xstrdup("-c"); | |
3419 | argv[argc++] = xasprintf("0x%X", 1); | |
88964e64 AC |
3420 | } |
3421 | } | |
bab69409 AC |
3422 | |
3423 | argv = grow_argv(&argv, argc, 1); | |
3424 | argv[argc] = NULL; | |
3425 | ||
3426 | optind = 1; | |
58397e6c | 3427 | |
eac84432 AC |
3428 | if (VLOG_IS_INFO_ENABLED()) { |
3429 | struct ds eal_args; | |
3430 | int opt; | |
3431 | ds_init(&eal_args); | |
3432 | ds_put_cstr(&eal_args, "EAL ARGS:"); | |
3433 | for (opt = 0; opt < argc; ++opt) { | |
3434 | ds_put_cstr(&eal_args, " "); | |
3435 | ds_put_cstr(&eal_args, argv[opt]); | |
3436 | } | |
3437 | VLOG_INFO("%s", ds_cstr_ro(&eal_args)); | |
3438 | ds_destroy(&eal_args); | |
3439 | } | |
3440 | ||
8a9562d2 PS |
3441 | /* Make sure things are initialized ... */ |
3442 | result = rte_eal_init(argc, argv); | |
451450fa | 3443 | if (result < 0) { |
58397e6c | 3444 | ovs_abort(result, "Cannot init EAL"); |
451450fa | 3445 | } |
8a9562d2 | 3446 | |
563c98d8 | 3447 | /* Set the main thread affinity back to pre rte_eal_init() value */ |
d8e2f4cc | 3448 | if (auto_determine && !err) { |
bab69409 AC |
3449 | err = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), |
3450 | &cpuset); | |
3451 | if (err) { | |
3452 | VLOG_ERR("Thread setaffinity error %d", err); | |
3453 | } | |
563c98d8 AC |
3454 | } |
3455 | ||
bab69409 AC |
3456 | dpdk_argv = argv; |
3457 | dpdk_argc = argc; | |
3458 | ||
3459 | atexit(deferred_argv_release); | |
3460 | ||
d7310583 | 3461 | rte_memzone_dump(stdout); |
8a9562d2 PS |
3462 | rte_eal_init_ret = 0; |
3463 | ||
db73f716 | 3464 | /* We are called from the main thread here */ |
d5c199ea | 3465 | RTE_PER_LCORE(_lcore_id) = NON_PMD_CORE_ID; |
db73f716 | 3466 | |
bab69409 AC |
3467 | ovs_thread_create("dpdk_watchdog", dpdk_watchdog, NULL); |
3468 | ||
bab69409 AC |
3469 | dpdk_vhost_class_init(); |
3470 | ||
4b88d678 CL |
3471 | #ifdef DPDK_PDUMP |
3472 | VLOG_INFO("DPDK pdump packet capture enabled"); | |
3473 | err = rte_pdump_init(ovs_rundir()); | |
3474 | if (err) { | |
3475 | VLOG_INFO("Error initialising DPDK pdump"); | |
3476 | rte_pdump_uninit(); | |
3477 | } else { | |
3478 | char *server_socket_path; | |
3479 | ||
3480 | server_socket_path = xasprintf("%s/%s", ovs_rundir(), | |
3481 | "pdump_server_socket"); | |
3482 | fatal_signal_add_file_to_unlink(server_socket_path); | |
3483 | free(server_socket_path); | |
3484 | } | |
3485 | #endif | |
3486 | ||
bab69409 AC |
3487 | /* Finally, register the dpdk classes */ |
3488 | netdev_dpdk_register(); | |
3489 | } | |
3490 | ||
3491 | void | |
3492 | dpdk_init(const struct smap *ovs_other_config) | |
3493 | { | |
3494 | static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; | |
3495 | ||
3496 | if (ovs_other_config && ovsthread_once_start(&once)) { | |
3497 | dpdk_init__(ovs_other_config); | |
3498 | ovsthread_once_done(&once); | |
3499 | } | |
8a9562d2 PS |
3500 | } |
3501 | ||
bce01e3a | 3502 | static const struct netdev_class dpdk_class = |
95fb793a | 3503 | NETDEV_DPDK_CLASS( |
3504 | "dpdk", | |
5496878c | 3505 | netdev_dpdk_construct, |
58397e6c | 3506 | netdev_dpdk_destruct, |
81acebda IM |
3507 | netdev_dpdk_set_config, |
3508 | netdev_dpdk_set_tx_multiq, | |
58397e6c KT |
3509 | netdev_dpdk_eth_send, |
3510 | netdev_dpdk_get_carrier, | |
3511 | netdev_dpdk_get_stats, | |
3512 | netdev_dpdk_get_features, | |
3513 | netdev_dpdk_get_status, | |
050c60bf | 3514 | netdev_dpdk_reconfigure, |
58397e6c | 3515 | netdev_dpdk_rxq_recv); |
95fb793a | 3516 | |
bce01e3a | 3517 | static const struct netdev_class dpdk_ring_class = |
95fb793a | 3518 | NETDEV_DPDK_CLASS( |
3519 | "dpdkr", | |
5496878c | 3520 | netdev_dpdk_ring_construct, |
58397e6c | 3521 | netdev_dpdk_destruct, |
c3d062a7 | 3522 | netdev_dpdk_ring_set_config, |
81acebda | 3523 | netdev_dpdk_set_tx_multiq, |
58397e6c KT |
3524 | netdev_dpdk_ring_send, |
3525 | netdev_dpdk_get_carrier, | |
3526 | netdev_dpdk_get_stats, | |
3527 | netdev_dpdk_get_features, | |
3528 | netdev_dpdk_get_status, | |
050c60bf | 3529 | netdev_dpdk_reconfigure, |
58397e6c KT |
3530 | netdev_dpdk_rxq_recv); |
3531 | ||
53f50d24 | 3532 | static const struct netdev_class dpdk_vhost_class = |
7d1ced01 CL |
3533 | NETDEV_DPDK_CLASS( |
3534 | "dpdkvhostuser", | |
53f50d24 | 3535 | netdev_dpdk_vhost_construct, |
58397e6c | 3536 | netdev_dpdk_vhost_destruct, |
2d24d165 | 3537 | NULL, |
81acebda | 3538 | NULL, |
58397e6c KT |
3539 | netdev_dpdk_vhost_send, |
3540 | netdev_dpdk_vhost_get_carrier, | |
3541 | netdev_dpdk_vhost_get_stats, | |
3542 | NULL, | |
7251515e | 3543 | NULL, |
53f50d24 | 3544 | netdev_dpdk_vhost_reconfigure, |
58397e6c | 3545 | netdev_dpdk_vhost_rxq_recv); |
2d24d165 CL |
3546 | static const struct netdev_class dpdk_vhost_client_class = |
3547 | NETDEV_DPDK_CLASS( | |
3548 | "dpdkvhostuserclient", | |
3549 | netdev_dpdk_vhost_client_construct, | |
3550 | netdev_dpdk_vhost_destruct, | |
3551 | netdev_dpdk_vhost_client_set_config, | |
3552 | NULL, | |
3553 | netdev_dpdk_vhost_send, | |
3554 | netdev_dpdk_vhost_get_carrier, | |
3555 | netdev_dpdk_vhost_get_stats, | |
3556 | NULL, | |
3557 | NULL, | |
3558 | netdev_dpdk_vhost_client_reconfigure, | |
3559 | netdev_dpdk_vhost_rxq_recv); | |
95fb793a | 3560 | |
8a9562d2 PS |
3561 | void |
3562 | netdev_dpdk_register(void) | |
3563 | { | |
bab69409 AC |
3564 | dpdk_common_init(); |
3565 | netdev_register_provider(&dpdk_class); | |
3566 | netdev_register_provider(&dpdk_ring_class); | |
53f50d24 | 3567 | netdev_register_provider(&dpdk_vhost_class); |
2d24d165 | 3568 | netdev_register_provider(&dpdk_vhost_client_class); |
8a9562d2 | 3569 | } |
8617afff | 3570 | |
6930c7e0 DDP |
3571 | void |
3572 | dpdk_set_lcore_id(unsigned cpu) | |
8617afff | 3573 | { |
abb5943d AW |
3574 | /* NON_PMD_CORE_ID is reserved for use by non pmd threads. */ |
3575 | ovs_assert(cpu != NON_PMD_CORE_ID); | |
65f13b50 | 3576 | RTE_PER_LCORE(_lcore_id) = cpu; |
8617afff | 3577 | } |
db73f716 | 3578 | |
db73f716 | 3579 | static bool |
5f17de68 | 3580 | dpdk_thread_is_pmd(void) |
db73f716 | 3581 | { |
abb5943d | 3582 | return rte_lcore_id() != NON_PMD_CORE_ID; |
db73f716 | 3583 | } |