1 /*- 2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org> 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice unmodified, this list of conditions, and the following 10 * disclaimer. 11 * 2. Redistributions in binary form must reproduce the above copyright 12 * notice, this list of conditions and the following disclaimer in the 13 * documentation and/or other materials provided with the distribution. 14 * 15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 /* Driver for VirtIO network devices. */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include <sys/param.h> 33 #include <sys/eventhandler.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/sockio.h> 37 #include <sys/mbuf.h> 38 #include <sys/malloc.h> 39 #include <sys/module.h> 40 #include <sys/socket.h> 41 #include <sys/sysctl.h> 42 #include <sys/random.h> 43 #include <sys/sglist.h> 44 #include <sys/lock.h> 45 #include <sys/mutex.h> 46 #include <sys/taskqueue.h> 47 #include <sys/smp.h> 48 #include <machine/smp.h> 49 50 #include <vm/uma.h> 51 52 #include <net/ethernet.h> 53 #include <net/if.h> 54 #include <net/if_var.h> 55 #include <net/if_arp.h> 56 #include <net/if_dl.h> 57 #include <net/if_types.h> 58 #include <net/if_media.h> 59 #include <net/if_vlan_var.h> 60 61 #include <net/bpf.h> 62 63 #include <netinet/in_systm.h> 64 #include <netinet/in.h> 65 #include <netinet/ip.h> 66 #include <netinet/ip6.h> 67 #include <netinet6/ip6_var.h> 68 #include <netinet/udp.h> 69 #include <netinet/tcp.h> 70 #include <netinet/sctp.h> 71 72 #include <machine/bus.h> 73 #include <machine/resource.h> 74 #include <sys/bus.h> 75 #include <sys/rman.h> 76 77 #include <dev/virtio/virtio.h> 78 #include <dev/virtio/virtqueue.h> 79 #include <dev/virtio/network/virtio_net.h> 80 #include <dev/virtio/network/if_vtnetvar.h> 81 82 #include "virtio_if.h" 83 84 #include "opt_inet.h" 85 #include "opt_inet6.h" 86 87 static int vtnet_modevent(module_t, int, void *); 88 89 static int vtnet_probe(device_t); 90 static int vtnet_attach(device_t); 91 static int vtnet_detach(device_t); 92 static int vtnet_suspend(device_t); 93 static int vtnet_resume(device_t); 94 static int vtnet_shutdown(device_t); 95 static int vtnet_attach_completed(device_t); 96 static int vtnet_config_change(device_t); 97 98 static void vtnet_negotiate_features(struct vtnet_softc *); 99 static void vtnet_setup_features(struct vtnet_softc *); 100 static int vtnet_init_rxq(struct vtnet_softc *, int); 101 static int vtnet_init_txq(struct vtnet_softc *, int); 102 static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); 103 static void vtnet_free_rxtx_queues(struct vtnet_softc *); 104 static int vtnet_alloc_rx_filters(struct vtnet_softc *); 105 static void vtnet_free_rx_filters(struct vtnet_softc *); 106 static int vtnet_alloc_virtqueues(struct vtnet_softc *); 107 static int vtnet_setup_interface(struct vtnet_softc *); 108 static int vtnet_change_mtu(struct vtnet_softc *, int); 109 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t); 110 111 static int vtnet_rxq_populate(struct vtnet_rxq *); 112 static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); 113 static struct mbuf * 114 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); 115 static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *, 116 struct mbuf *, int); 117 static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); 118 static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); 119 static int vtnet_rxq_new_buf(struct vtnet_rxq *); 120 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, 121 struct virtio_net_hdr *); 122 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); 123 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); 124 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); 125 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, 126 struct virtio_net_hdr *); 127 static int vtnet_rxq_eof(struct vtnet_rxq *); 128 static void vtnet_rx_vq_intr(void *); 129 static void vtnet_rxq_tq_intr(void *, int); 130 131 static void vtnet_txq_free_mbufs(struct vtnet_txq *); 132 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, 133 int *, int *, int *); 134 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, 135 int, struct virtio_net_hdr *); 136 static struct mbuf * 137 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, 138 struct virtio_net_hdr *); 139 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, 140 struct vtnet_tx_header *); 141 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **); 142 #ifdef VTNET_LEGACY_TX 143 static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *); 144 static void vtnet_start(struct ifnet *); 145 #else 146 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); 147 static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *); 148 static void vtnet_txq_tq_deferred(void *, int); 149 #endif 150 static void vtnet_txq_start(struct vtnet_txq *); 151 static void vtnet_txq_tq_intr(void *, int); 152 static void vtnet_txq_eof(struct vtnet_txq *); 153 static void vtnet_tx_vq_intr(void *); 154 static void vtnet_tx_start_all(struct vtnet_softc *); 155 156 #ifndef VTNET_LEGACY_TX 157 static void vtnet_qflush(struct ifnet *); 158 #endif 159 160 static int vtnet_watchdog(struct vtnet_txq *); 161 static void vtnet_rxq_accum_stats(struct vtnet_rxq *, 162 struct vtnet_rxq_stats *); 163 static void vtnet_txq_accum_stats(struct vtnet_txq *, 164 struct vtnet_txq_stats *); 165 static void vtnet_accumulate_stats(struct vtnet_softc *); 166 static void vtnet_tick(void *); 167 168 static void vtnet_start_taskqueues(struct vtnet_softc *); 169 static void vtnet_free_taskqueues(struct vtnet_softc *); 170 static void vtnet_drain_taskqueues(struct vtnet_softc *); 171 172 static void vtnet_drain_rxtx_queues(struct vtnet_softc *); 173 static void vtnet_stop_rendezvous(struct vtnet_softc *); 174 static void vtnet_stop(struct vtnet_softc *); 175 static int vtnet_virtio_reinit(struct vtnet_softc *); 176 static void vtnet_init_rx_filters(struct vtnet_softc *); 177 static int vtnet_init_rx_queues(struct vtnet_softc *); 178 static int vtnet_init_tx_queues(struct vtnet_softc *); 179 static int vtnet_init_rxtx_queues(struct vtnet_softc *); 180 static void vtnet_set_active_vq_pairs(struct vtnet_softc *); 181 static int vtnet_reinit(struct vtnet_softc *); 182 static void vtnet_init_locked(struct vtnet_softc *); 183 static void vtnet_init(void *); 184 185 static void vtnet_free_ctrl_vq(struct vtnet_softc *); 186 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 187 struct sglist *, int, int); 188 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 189 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); 190 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int); 191 static int vtnet_set_promisc(struct vtnet_softc *, int); 192 static int vtnet_set_allmulti(struct vtnet_softc *, int); 193 static void vtnet_attach_disable_promisc(struct vtnet_softc *); 194 static void vtnet_rx_filter(struct vtnet_softc *); 195 static void vtnet_rx_filter_mac(struct vtnet_softc *); 196 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 197 static void vtnet_rx_filter_vlan(struct vtnet_softc *); 198 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 199 static void vtnet_register_vlan(void *, struct ifnet *, uint16_t); 200 static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t); 201 202 static int vtnet_is_link_up(struct vtnet_softc *); 203 static void vtnet_update_link_status(struct vtnet_softc *); 204 static int vtnet_ifmedia_upd(struct ifnet *); 205 static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *); 206 static void vtnet_get_hwaddr(struct vtnet_softc *); 207 static void vtnet_set_hwaddr(struct vtnet_softc *); 208 static void vtnet_vlan_tag_remove(struct mbuf *); 209 210 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, 211 struct sysctl_oid_list *, struct vtnet_rxq *); 212 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, 213 struct sysctl_oid_list *, struct vtnet_txq *); 214 static void vtnet_setup_queue_sysctl(struct vtnet_softc *); 215 static void vtnet_setup_sysctl(struct vtnet_softc *); 216 217 static int vtnet_rxq_enable_intr(struct vtnet_rxq *); 218 static void vtnet_rxq_disable_intr(struct vtnet_rxq *); 219 static int vtnet_txq_enable_intr(struct vtnet_txq *); 220 static void vtnet_txq_disable_intr(struct vtnet_txq *); 221 static void vtnet_enable_rx_interrupts(struct vtnet_softc *); 222 static void vtnet_enable_tx_interrupts(struct vtnet_softc *); 223 static void vtnet_enable_interrupts(struct vtnet_softc *); 224 static void vtnet_disable_rx_interrupts(struct vtnet_softc *); 225 static void vtnet_disable_tx_interrupts(struct vtnet_softc *); 226 static void vtnet_disable_interrupts(struct vtnet_softc *); 227 228 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); 229 230 /* Tunables. */ 231 static int vtnet_csum_disable = 0; 232 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable); 233 static int vtnet_tso_disable = 0; 234 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable); 235 static int vtnet_lro_disable = 0; 236 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable); 237 static int vtnet_mq_disable = 0; 238 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable); 239 static int vtnet_mq_max_pairs = 0; 240 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs); 241 static int vtnet_rx_process_limit = 512; 242 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit); 243 244 /* 245 * Reducing the number of transmit completed interrupts can improve 246 * performance. To do so, the define below keeps the Tx vq interrupt 247 * disabled and adds calls to vtnet_txeof() in the start and watchdog 248 * paths. The price to pay for this is the m_free'ing of transmitted 249 * mbufs may be delayed until the watchdog fires. 250 * 251 * BMV: Reintroduce this later as a run-time option, if it makes 252 * sense after the EVENT_IDX feature is supported. 253 * 254 * #define VTNET_TX_INTR_MODERATION 255 */ 256 257 static uma_zone_t vtnet_tx_header_zone; 258 259 static struct virtio_feature_desc vtnet_feature_desc[] = { 260 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 261 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 262 { VIRTIO_NET_F_MAC, "MacAddress" }, 263 { VIRTIO_NET_F_GSO, "TxAllGSO" }, 264 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" }, 265 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" }, 266 { VIRTIO_NET_F_GUEST_ECN, "RxECN" }, 267 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 268 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 269 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 270 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 271 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 272 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 273 { VIRTIO_NET_F_STATUS, "Status" }, 274 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" }, 275 { VIRTIO_NET_F_CTRL_RX, "RxMode" }, 276 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" }, 277 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" }, 278 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 279 { VIRTIO_NET_F_MQ, "Multiqueue" }, 280 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" }, 281 282 { 0, NULL } 283 }; 284 285 static device_method_t vtnet_methods[] = { 286 /* Device methods. */ 287 DEVMETHOD(device_probe, vtnet_probe), 288 DEVMETHOD(device_attach, vtnet_attach), 289 DEVMETHOD(device_detach, vtnet_detach), 290 DEVMETHOD(device_suspend, vtnet_suspend), 291 DEVMETHOD(device_resume, vtnet_resume), 292 DEVMETHOD(device_shutdown, vtnet_shutdown), 293 294 /* VirtIO methods. */ 295 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), 296 DEVMETHOD(virtio_config_change, vtnet_config_change), 297 298 DEVMETHOD_END 299 }; 300 301 static driver_t vtnet_driver = { 302 "vtnet", 303 vtnet_methods, 304 sizeof(struct vtnet_softc) 305 }; 306 static devclass_t vtnet_devclass; 307 308 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass, 309 vtnet_modevent, 0); 310 MODULE_VERSION(vtnet, 1); 311 MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 312 313 static int 314 vtnet_modevent(module_t mod, int type, void *unused) 315 { 316 int error; 317 318 error = 0; 319 320 switch (type) { 321 case MOD_LOAD: 322 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", 323 sizeof(struct vtnet_tx_header), 324 NULL, NULL, NULL, NULL, 0, 0); 325 break; 326 case MOD_QUIESCE: 327 case MOD_UNLOAD: 328 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) 329 error = EBUSY; 330 else if (type == MOD_UNLOAD) { 331 uma_zdestroy(vtnet_tx_header_zone); 332 vtnet_tx_header_zone = NULL; 333 } 334 break; 335 case MOD_SHUTDOWN: 336 break; 337 default: 338 error = EOPNOTSUPP; 339 break; 340 } 341 342 return (error); 343 } 344 345 static int 346 vtnet_probe(device_t dev) 347 { 348 349 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK) 350 return (ENXIO); 351 352 device_set_desc(dev, "VirtIO Networking Adapter"); 353 354 return (BUS_PROBE_DEFAULT); 355 } 356 357 static int 358 vtnet_attach(device_t dev) 359 { 360 struct vtnet_softc *sc; 361 int error; 362 363 sc = device_get_softc(dev); 364 sc->vtnet_dev = dev; 365 366 /* Register our feature descriptions. */ 367 virtio_set_feature_desc(dev, vtnet_feature_desc); 368 369 VTNET_CORE_LOCK_INIT(sc); 370 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); 371 372 vtnet_setup_sysctl(sc); 373 vtnet_setup_features(sc); 374 375 error = vtnet_alloc_rx_filters(sc); 376 if (error) { 377 device_printf(dev, "cannot allocate Rx filters\n"); 378 goto fail; 379 } 380 381 error = vtnet_alloc_rxtx_queues(sc); 382 if (error) { 383 device_printf(dev, "cannot allocate queues\n"); 384 goto fail; 385 } 386 387 error = vtnet_alloc_virtqueues(sc); 388 if (error) { 389 device_printf(dev, "cannot allocate virtqueues\n"); 390 goto fail; 391 } 392 393 error = vtnet_setup_interface(sc); 394 if (error) { 395 device_printf(dev, "cannot setup interface\n"); 396 goto fail; 397 } 398 399 error = virtio_setup_intr(dev, INTR_TYPE_NET); 400 if (error) { 401 device_printf(dev, "cannot setup virtqueue interrupts\n"); 402 /* BMV: This will crash if during boot! */ 403 ether_ifdetach(sc->vtnet_ifp); 404 goto fail; 405 } 406 407 vtnet_start_taskqueues(sc); 408 409 fail: 410 if (error) 411 vtnet_detach(dev); 412 413 return (error); 414 } 415 416 static int 417 vtnet_detach(device_t dev) 418 { 419 struct vtnet_softc *sc; 420 struct ifnet *ifp; 421 422 sc = device_get_softc(dev); 423 ifp = sc->vtnet_ifp; 424 425 if (device_is_attached(dev)) { 426 VTNET_CORE_LOCK(sc); 427 vtnet_stop(sc); 428 VTNET_CORE_UNLOCK(sc); 429 430 callout_drain(&sc->vtnet_tick_ch); 431 vtnet_drain_taskqueues(sc); 432 433 ether_ifdetach(ifp); 434 } 435 436 vtnet_free_taskqueues(sc); 437 438 if (sc->vtnet_vlan_attach != NULL) { 439 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 440 sc->vtnet_vlan_attach = NULL; 441 } 442 if (sc->vtnet_vlan_detach != NULL) { 443 EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach); 444 sc->vtnet_vlan_detach = NULL; 445 } 446 447 ifmedia_removeall(&sc->vtnet_media); 448 449 if (ifp != NULL) { 450 if_free(ifp); 451 sc->vtnet_ifp = NULL; 452 } 453 454 vtnet_free_rxtx_queues(sc); 455 vtnet_free_rx_filters(sc); 456 457 if (sc->vtnet_ctrl_vq != NULL) 458 vtnet_free_ctrl_vq(sc); 459 460 VTNET_CORE_LOCK_DESTROY(sc); 461 462 return (0); 463 } 464 465 static int 466 vtnet_suspend(device_t dev) 467 { 468 struct vtnet_softc *sc; 469 470 sc = device_get_softc(dev); 471 472 VTNET_CORE_LOCK(sc); 473 vtnet_stop(sc); 474 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 475 VTNET_CORE_UNLOCK(sc); 476 477 return (0); 478 } 479 480 static int 481 vtnet_resume(device_t dev) 482 { 483 struct vtnet_softc *sc; 484 struct ifnet *ifp; 485 486 sc = device_get_softc(dev); 487 ifp = sc->vtnet_ifp; 488 489 VTNET_CORE_LOCK(sc); 490 if (ifp->if_flags & IFF_UP) 491 vtnet_init_locked(sc); 492 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 493 VTNET_CORE_UNLOCK(sc); 494 495 return (0); 496 } 497 498 static int 499 vtnet_shutdown(device_t dev) 500 { 501 502 /* 503 * Suspend already does all of what we need to 504 * do here; we just never expect to be resumed. 505 */ 506 return (vtnet_suspend(dev)); 507 } 508 509 static int 510 vtnet_attach_completed(device_t dev) 511 { 512 513 vtnet_attach_disable_promisc(device_get_softc(dev)); 514 515 return (0); 516 } 517 518 static int 519 vtnet_config_change(device_t dev) 520 { 521 struct vtnet_softc *sc; 522 523 sc = device_get_softc(dev); 524 525 VTNET_CORE_LOCK(sc); 526 vtnet_update_link_status(sc); 527 if (sc->vtnet_link_active != 0) 528 vtnet_tx_start_all(sc); 529 VTNET_CORE_UNLOCK(sc); 530 531 return (0); 532 } 533 534 static void 535 vtnet_negotiate_features(struct vtnet_softc *sc) 536 { 537 device_t dev; 538 uint64_t mask, features; 539 540 dev = sc->vtnet_dev; 541 mask = 0; 542 543 /* 544 * TSO and LRO are only available when their corresponding checksum 545 * offload feature is also negotiated. 546 */ 547 if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) { 548 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM; 549 mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES; 550 } 551 if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) 552 mask |= VTNET_TSO_FEATURES; 553 if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) 554 mask |= VTNET_LRO_FEATURES; 555 #ifndef VTNET_LEGACY_TX 556 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 557 mask |= VIRTIO_NET_F_MQ; 558 #else 559 mask |= VIRTIO_NET_F_MQ; 560 #endif 561 562 features = VTNET_FEATURES & ~mask; 563 sc->vtnet_features = virtio_negotiate_features(dev, features); 564 565 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 566 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 567 /* 568 * LRO without mergeable buffers requires special care. This 569 * is not ideal because every receive buffer must be large 570 * enough to hold the maximum TCP packet, the Ethernet header, 571 * and the header. This requires up to 34 descriptors with 572 * MCLBYTES clusters. If we do not have indirect descriptors, 573 * LRO is disabled since the virtqueue will not contain very 574 * many receive buffers. 575 */ 576 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 577 device_printf(dev, 578 "LRO disabled due to both mergeable buffers and " 579 "indirect descriptors not negotiated\n"); 580 581 features &= ~VTNET_LRO_FEATURES; 582 sc->vtnet_features = 583 virtio_negotiate_features(dev, features); 584 } else 585 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 586 } 587 } 588 589 static void 590 vtnet_setup_features(struct vtnet_softc *sc) 591 { 592 device_t dev; 593 int max_pairs, max; 594 595 dev = sc->vtnet_dev; 596 597 vtnet_negotiate_features(sc); 598 599 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 600 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 601 602 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 603 /* This feature should always be negotiated. */ 604 sc->vtnet_flags |= VTNET_FLAG_MAC; 605 } 606 607 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 608 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 609 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 610 } else 611 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 612 613 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 614 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS; 615 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 616 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS; 617 else 618 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS; 619 620 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 621 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 622 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 623 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS; 624 else 625 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS; 626 627 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 628 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 629 630 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 631 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 632 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 633 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 634 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 635 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 636 } 637 638 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) && 639 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 640 max_pairs = virtio_read_dev_config_2(dev, 641 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 642 if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 643 max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) 644 max_pairs = 1; 645 } else 646 max_pairs = 1; 647 648 if (max_pairs > 1) { 649 /* 650 * Limit the maximum number of queue pairs to the number of 651 * CPUs or the configured maximum. The actual number of 652 * queues that get used may be less. 653 */ 654 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 655 if (max > 0 && max_pairs > max) 656 max_pairs = max; 657 if (max_pairs > mp_ncpus) 658 max_pairs = mp_ncpus; 659 if (max_pairs > VTNET_MAX_QUEUE_PAIRS) 660 max_pairs = VTNET_MAX_QUEUE_PAIRS; 661 if (max_pairs > 1) 662 sc->vtnet_flags |= VTNET_FLAG_MULTIQ; 663 } 664 665 sc->vtnet_max_vq_pairs = max_pairs; 666 } 667 668 static int 669 vtnet_init_rxq(struct vtnet_softc *sc, int id) 670 { 671 struct vtnet_rxq *rxq; 672 673 rxq = &sc->vtnet_rxqs[id]; 674 675 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 676 device_get_nameunit(sc->vtnet_dev), id); 677 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 678 679 rxq->vtnrx_sc = sc; 680 rxq->vtnrx_id = id; 681 682 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 683 if (rxq->vtnrx_sg == NULL) 684 return (ENOMEM); 685 686 TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 687 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 688 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 689 690 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 691 } 692 693 static int 694 vtnet_init_txq(struct vtnet_softc *sc, int id) 695 { 696 struct vtnet_txq *txq; 697 698 txq = &sc->vtnet_txqs[id]; 699 700 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 701 device_get_nameunit(sc->vtnet_dev), id); 702 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 703 704 txq->vtntx_sc = sc; 705 txq->vtntx_id = id; 706 707 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 708 if (txq->vtntx_sg == NULL) 709 return (ENOMEM); 710 711 #ifndef VTNET_LEGACY_TX 712 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 713 M_NOWAIT, &txq->vtntx_mtx); 714 if (txq->vtntx_br == NULL) 715 return (ENOMEM); 716 717 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 718 #endif 719 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 720 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 721 taskqueue_thread_enqueue, &txq->vtntx_tq); 722 if (txq->vtntx_tq == NULL) 723 return (ENOMEM); 724 725 return (0); 726 } 727 728 static int 729 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 730 { 731 int i, npairs, error; 732 733 npairs = sc->vtnet_max_vq_pairs; 734 735 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 736 M_NOWAIT | M_ZERO); 737 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 738 M_NOWAIT | M_ZERO); 739 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 740 return (ENOMEM); 741 742 for (i = 0; i < npairs; i++) { 743 error = vtnet_init_rxq(sc, i); 744 if (error) 745 return (error); 746 error = vtnet_init_txq(sc, i); 747 if (error) 748 return (error); 749 } 750 751 vtnet_setup_queue_sysctl(sc); 752 753 return (0); 754 } 755 756 static void 757 vtnet_destroy_rxq(struct vtnet_rxq *rxq) 758 { 759 760 rxq->vtnrx_sc = NULL; 761 rxq->vtnrx_id = -1; 762 763 if (rxq->vtnrx_sg != NULL) { 764 sglist_free(rxq->vtnrx_sg); 765 rxq->vtnrx_sg = NULL; 766 } 767 768 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 769 mtx_destroy(&rxq->vtnrx_mtx); 770 } 771 772 static void 773 vtnet_destroy_txq(struct vtnet_txq *txq) 774 { 775 776 txq->vtntx_sc = NULL; 777 txq->vtntx_id = -1; 778 779 if (txq->vtntx_sg != NULL) { 780 sglist_free(txq->vtntx_sg); 781 txq->vtntx_sg = NULL; 782 } 783 784 #ifndef VTNET_LEGACY_TX 785 if (txq->vtntx_br != NULL) { 786 buf_ring_free(txq->vtntx_br, M_DEVBUF); 787 txq->vtntx_br = NULL; 788 } 789 #endif 790 791 if (mtx_initialized(&txq->vtntx_mtx) != 0) 792 mtx_destroy(&txq->vtntx_mtx); 793 } 794 795 static void 796 vtnet_free_rxtx_queues(struct vtnet_softc *sc) 797 { 798 int i; 799 800 if (sc->vtnet_rxqs != NULL) { 801 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 802 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 803 free(sc->vtnet_rxqs, M_DEVBUF); 804 sc->vtnet_rxqs = NULL; 805 } 806 807 if (sc->vtnet_txqs != NULL) { 808 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 809 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 810 free(sc->vtnet_txqs, M_DEVBUF); 811 sc->vtnet_txqs = NULL; 812 } 813 } 814 815 static int 816 vtnet_alloc_rx_filters(struct vtnet_softc *sc) 817 { 818 819 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 820 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 821 M_DEVBUF, M_NOWAIT | M_ZERO); 822 if (sc->vtnet_mac_filter == NULL) 823 return (ENOMEM); 824 } 825 826 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 827 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 828 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 829 if (sc->vtnet_vlan_filter == NULL) 830 return (ENOMEM); 831 } 832 833 return (0); 834 } 835 836 static void 837 vtnet_free_rx_filters(struct vtnet_softc *sc) 838 { 839 840 if (sc->vtnet_mac_filter != NULL) { 841 free(sc->vtnet_mac_filter, M_DEVBUF); 842 sc->vtnet_mac_filter = NULL; 843 } 844 845 if (sc->vtnet_vlan_filter != NULL) { 846 free(sc->vtnet_vlan_filter, M_DEVBUF); 847 sc->vtnet_vlan_filter = NULL; 848 } 849 } 850 851 static int 852 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 853 { 854 device_t dev; 855 struct vq_alloc_info *info; 856 struct vtnet_rxq *rxq; 857 struct vtnet_txq *txq; 858 int i, idx, flags, nvqs, error; 859 860 dev = sc->vtnet_dev; 861 flags = 0; 862 863 nvqs = sc->vtnet_max_vq_pairs * 2; 864 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 865 nvqs++; 866 867 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); 868 if (info == NULL) 869 return (ENOMEM); 870 871 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) { 872 rxq = &sc->vtnet_rxqs[i]; 873 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, 874 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, 875 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id); 876 877 txq = &sc->vtnet_txqs[i]; 878 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs, 879 vtnet_tx_vq_intr, txq, &txq->vtntx_vq, 880 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id); 881 } 882 883 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 884 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, 885 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); 886 } 887 888 /* 889 * Enable interrupt binding if this is multiqueue. This only matters 890 * when per-vq MSIX is available. 891 */ 892 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ) 893 flags |= 0; 894 895 error = virtio_alloc_virtqueues(dev, flags, nvqs, info); 896 free(info, M_TEMP); 897 898 return (error); 899 } 900 901 static int 902 vtnet_setup_interface(struct vtnet_softc *sc) 903 { 904 device_t dev; 905 struct ifnet *ifp; 906 int limit; 907 908 dev = sc->vtnet_dev; 909 910 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER); 911 if (ifp == NULL) { 912 device_printf(dev, "cannot allocate ifnet structure\n"); 913 return (ENOSPC); 914 } 915 916 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 917 ifp->if_baudrate = IF_Gbps(10); /* Approx. */ 918 ifp->if_softc = sc; 919 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 920 ifp->if_init = vtnet_init; 921 ifp->if_ioctl = vtnet_ioctl; 922 923 #ifndef VTNET_LEGACY_TX 924 ifp->if_transmit = vtnet_txq_mq_start; 925 ifp->if_qflush = vtnet_qflush; 926 #else 927 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 928 ifp->if_start = vtnet_start; 929 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1); 930 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1; 931 IFQ_SET_READY(&ifp->if_snd); 932 #endif 933 934 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd, 935 vtnet_ifmedia_sts); 936 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL); 937 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE); 938 939 /* Read (or generate) the MAC address for the adapter. */ 940 vtnet_get_hwaddr(sc); 941 942 ether_ifattach(ifp, sc->vtnet_hwaddr); 943 944 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 945 ifp->if_capabilities |= IFCAP_LINKSTATE; 946 947 /* Tell the upper layer(s) we support long frames. */ 948 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); 949 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU; 950 951 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 952 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6; 953 954 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) { 955 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6; 956 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 957 } else { 958 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 959 ifp->if_capabilities |= IFCAP_TSO4; 960 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 961 ifp->if_capabilities |= IFCAP_TSO6; 962 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 963 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 964 } 965 966 if (ifp->if_capabilities & IFCAP_TSO) 967 ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 968 } 969 970 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) 971 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6; 972 973 if (ifp->if_capabilities & IFCAP_HWCSUM) { 974 /* 975 * VirtIO does not support VLAN tagging, but we can fake 976 * it by inserting and removing the 802.1Q header during 977 * transmit and receive. We are then able to do checksum 978 * offloading of VLAN frames. 979 */ 980 ifp->if_capabilities |= 981 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; 982 } 983 984 ifp->if_capenable = ifp->if_capabilities; 985 986 /* 987 * Capabilities after here are not enabled by default. 988 */ 989 990 if (ifp->if_capabilities & IFCAP_RXCSUM) { 991 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) || 992 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6)) 993 ifp->if_capabilities |= IFCAP_LRO; 994 } 995 996 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 997 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 998 999 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1000 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1001 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1002 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1003 } 1004 1005 limit = vtnet_tunable_int(sc, "rx_process_limit", 1006 vtnet_rx_process_limit); 1007 if (limit < 0) 1008 limit = INT_MAX; 1009 sc->vtnet_rx_process_limit = limit; 1010 1011 return (0); 1012 } 1013 1014 static int 1015 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu) 1016 { 1017 struct ifnet *ifp; 1018 int frame_size, clsize; 1019 1020 ifp = sc->vtnet_ifp; 1021 1022 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU) 1023 return (EINVAL); 1024 1025 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) + 1026 new_mtu; 1027 1028 /* 1029 * Based on the new MTU (and hence frame size) determine which 1030 * cluster size is most appropriate for the receive queues. 1031 */ 1032 if (frame_size <= MCLBYTES) { 1033 clsize = MCLBYTES; 1034 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1035 /* Avoid going past 9K jumbos. */ 1036 if (frame_size > MJUM9BYTES) 1037 return (EINVAL); 1038 clsize = MJUM9BYTES; 1039 } else 1040 clsize = MJUMPAGESIZE; 1041 1042 ifp->if_mtu = new_mtu; 1043 sc->vtnet_rx_new_clsize = clsize; 1044 1045 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1046 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1047 vtnet_init_locked(sc); 1048 } 1049 1050 return (0); 1051 } 1052 1053 static int 1054 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1055 { 1056 struct vtnet_softc *sc; 1057 struct ifreq *ifr; 1058 int reinit, mask, error; 1059 1060 sc = ifp->if_softc; 1061 ifr = (struct ifreq *) data; 1062 error = 0; 1063 1064 switch (cmd) { 1065 case SIOCSIFMTU: 1066 if (ifp->if_mtu != ifr->ifr_mtu) { 1067 VTNET_CORE_LOCK(sc); 1068 error = vtnet_change_mtu(sc, ifr->ifr_mtu); 1069 VTNET_CORE_UNLOCK(sc); 1070 } 1071 break; 1072 1073 case SIOCSIFFLAGS: 1074 VTNET_CORE_LOCK(sc); 1075 if ((ifp->if_flags & IFF_UP) == 0) { 1076 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1077 vtnet_stop(sc); 1078 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1079 if ((ifp->if_flags ^ sc->vtnet_if_flags) & 1080 (IFF_PROMISC | IFF_ALLMULTI)) { 1081 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1082 vtnet_rx_filter(sc); 1083 else 1084 error = ENOTSUP; 1085 } 1086 } else 1087 vtnet_init_locked(sc); 1088 1089 if (error == 0) 1090 sc->vtnet_if_flags = ifp->if_flags; 1091 VTNET_CORE_UNLOCK(sc); 1092 break; 1093 1094 case SIOCADDMULTI: 1095 case SIOCDELMULTI: 1096 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) 1097 break; 1098 VTNET_CORE_LOCK(sc); 1099 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1100 vtnet_rx_filter_mac(sc); 1101 VTNET_CORE_UNLOCK(sc); 1102 break; 1103 1104 case SIOCSIFMEDIA: 1105 case SIOCGIFMEDIA: 1106 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1107 break; 1108 1109 case SIOCSIFCAP: 1110 VTNET_CORE_LOCK(sc); 1111 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1112 1113 if (mask & IFCAP_TXCSUM) 1114 ifp->if_capenable ^= IFCAP_TXCSUM; 1115 if (mask & IFCAP_TXCSUM_IPV6) 1116 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; 1117 if (mask & IFCAP_TSO4) 1118 ifp->if_capenable ^= IFCAP_TSO4; 1119 if (mask & IFCAP_TSO6) 1120 ifp->if_capenable ^= IFCAP_TSO6; 1121 1122 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO | 1123 IFCAP_VLAN_HWFILTER)) { 1124 /* These Rx features require us to renegotiate. */ 1125 reinit = 1; 1126 1127 if (mask & IFCAP_RXCSUM) 1128 ifp->if_capenable ^= IFCAP_RXCSUM; 1129 if (mask & IFCAP_RXCSUM_IPV6) 1130 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; 1131 if (mask & IFCAP_LRO) 1132 ifp->if_capenable ^= IFCAP_LRO; 1133 if (mask & IFCAP_VLAN_HWFILTER) 1134 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 1135 } else 1136 reinit = 0; 1137 1138 if (mask & IFCAP_VLAN_HWTSO) 1139 ifp->if_capenable ^= IFCAP_VLAN_HWTSO; 1140 if (mask & IFCAP_VLAN_HWTAGGING) 1141 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 1142 1143 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1144 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1145 vtnet_init_locked(sc); 1146 } 1147 1148 VTNET_CORE_UNLOCK(sc); 1149 VLAN_CAPABILITIES(ifp); 1150 1151 break; 1152 1153 default: 1154 error = ether_ioctl(ifp, cmd, data); 1155 break; 1156 } 1157 1158 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1159 1160 return (error); 1161 } 1162 1163 static int 1164 vtnet_rxq_populate(struct vtnet_rxq *rxq) 1165 { 1166 struct virtqueue *vq; 1167 int nbufs, error; 1168 1169 vq = rxq->vtnrx_vq; 1170 error = ENOSPC; 1171 1172 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1173 error = vtnet_rxq_new_buf(rxq); 1174 if (error) 1175 break; 1176 } 1177 1178 if (nbufs > 0) { 1179 virtqueue_notify(vq); 1180 /* 1181 * EMSGSIZE signifies the virtqueue did not have enough 1182 * entries available to hold the last mbuf. This is not 1183 * an error. 1184 */ 1185 if (error == EMSGSIZE) 1186 error = 0; 1187 } 1188 1189 return (error); 1190 } 1191 1192 static void 1193 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1194 { 1195 struct virtqueue *vq; 1196 struct mbuf *m; 1197 int last; 1198 1199 vq = rxq->vtnrx_vq; 1200 last = 0; 1201 1202 while ((m = virtqueue_drain(vq, &last)) != NULL) 1203 m_freem(m); 1204 1205 KASSERT(virtqueue_empty(vq), 1206 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1207 } 1208 1209 static struct mbuf * 1210 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1211 { 1212 struct mbuf *m_head, *m_tail, *m; 1213 int i, clsize; 1214 1215 clsize = sc->vtnet_rx_clsize; 1216 1217 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1218 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs)); 1219 1220 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize); 1221 if (m_head == NULL) 1222 goto fail; 1223 1224 m_head->m_len = clsize; 1225 m_tail = m_head; 1226 1227 /* Allocate the rest of the chain. */ 1228 for (i = 1; i < nbufs; i++) { 1229 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize); 1230 if (m == NULL) 1231 goto fail; 1232 1233 m->m_len = clsize; 1234 m_tail->m_next = m; 1235 m_tail = m; 1236 } 1237 1238 if (m_tailp != NULL) 1239 *m_tailp = m_tail; 1240 1241 return (m_head); 1242 1243 fail: 1244 sc->vtnet_stats.mbuf_alloc_failed++; 1245 m_freem(m_head); 1246 1247 return (NULL); 1248 } 1249 1250 /* 1251 * Slow path for when LRO without mergeable buffers is negotiated. 1252 */ 1253 static int 1254 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1255 int len0) 1256 { 1257 struct vtnet_softc *sc; 1258 struct mbuf *m, *m_prev; 1259 struct mbuf *m_new, *m_tail; 1260 int len, clsize, nreplace, error; 1261 1262 sc = rxq->vtnrx_sc; 1263 clsize = sc->vtnet_rx_clsize; 1264 1265 m_prev = NULL; 1266 m_tail = NULL; 1267 nreplace = 0; 1268 1269 m = m0; 1270 len = len0; 1271 1272 /* 1273 * Since these mbuf chains are so large, we avoid allocating an 1274 * entire replacement chain if possible. When the received frame 1275 * did not consume the entire chain, the unused mbufs are moved 1276 * to the replacement chain. 1277 */ 1278 while (len > 0) { 1279 /* 1280 * Something is seriously wrong if we received a frame 1281 * larger than the chain. Drop it. 1282 */ 1283 if (m == NULL) { 1284 sc->vtnet_stats.rx_frame_too_large++; 1285 return (EMSGSIZE); 1286 } 1287 1288 /* We always allocate the same cluster size. */ 1289 KASSERT(m->m_len == clsize, 1290 ("%s: mbuf size %d is not the cluster size %d", 1291 __func__, m->m_len, clsize)); 1292 1293 m->m_len = MIN(m->m_len, len); 1294 len -= m->m_len; 1295 1296 m_prev = m; 1297 m = m->m_next; 1298 nreplace++; 1299 } 1300 1301 KASSERT(nreplace <= sc->vtnet_rx_nmbufs, 1302 ("%s: too many replacement mbufs %d max %d", __func__, nreplace, 1303 sc->vtnet_rx_nmbufs)); 1304 1305 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1306 if (m_new == NULL) { 1307 m_prev->m_len = clsize; 1308 return (ENOBUFS); 1309 } 1310 1311 /* 1312 * Move any unused mbufs from the received chain onto the end 1313 * of the new chain. 1314 */ 1315 if (m_prev->m_next != NULL) { 1316 m_tail->m_next = m_prev->m_next; 1317 m_prev->m_next = NULL; 1318 } 1319 1320 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1321 if (error) { 1322 /* 1323 * BAD! We could not enqueue the replacement mbuf chain. We 1324 * must restore the m0 chain to the original state if it was 1325 * modified so we can subsequently discard it. 1326 * 1327 * NOTE: The replacement is suppose to be an identical copy 1328 * to the one just dequeued so this is an unexpected error. 1329 */ 1330 sc->vtnet_stats.rx_enq_replacement_failed++; 1331 1332 if (m_tail->m_next != NULL) { 1333 m_prev->m_next = m_tail->m_next; 1334 m_tail->m_next = NULL; 1335 } 1336 1337 m_prev->m_len = clsize; 1338 m_freem(m_new); 1339 } 1340 1341 return (error); 1342 } 1343 1344 static int 1345 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1346 { 1347 struct vtnet_softc *sc; 1348 struct mbuf *m_new; 1349 int error; 1350 1351 sc = rxq->vtnrx_sc; 1352 1353 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1354 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1355 1356 if (m->m_next == NULL) { 1357 /* Fast-path for the common case of just one mbuf. */ 1358 if (m->m_len < len) 1359 return (EINVAL); 1360 1361 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1362 if (m_new == NULL) 1363 return (ENOBUFS); 1364 1365 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1366 if (error) { 1367 /* 1368 * The new mbuf is suppose to be an identical 1369 * copy of the one just dequeued so this is an 1370 * unexpected error. 1371 */ 1372 m_freem(m_new); 1373 sc->vtnet_stats.rx_enq_replacement_failed++; 1374 } else 1375 m->m_len = len; 1376 } else 1377 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len); 1378 1379 return (error); 1380 } 1381 1382 static int 1383 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1384 { 1385 struct vtnet_softc *sc; 1386 struct sglist *sg; 1387 struct vtnet_rx_header *rxhdr; 1388 uint8_t *mdata; 1389 int offset, error; 1390 1391 sc = rxq->vtnrx_sc; 1392 sg = rxq->vtnrx_sg; 1393 mdata = mtod(m, uint8_t *); 1394 1395 VTNET_RXQ_LOCK_ASSERT(rxq); 1396 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1397 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1398 KASSERT(m->m_len == sc->vtnet_rx_clsize, 1399 ("%s: unexpected cluster size %d/%d", __func__, m->m_len, 1400 sc->vtnet_rx_clsize)); 1401 1402 sglist_reset(sg); 1403 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1404 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1405 rxhdr = (struct vtnet_rx_header *) mdata; 1406 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1407 offset = sizeof(struct vtnet_rx_header); 1408 } else 1409 offset = 0; 1410 1411 sglist_append(sg, mdata + offset, m->m_len - offset); 1412 if (m->m_next != NULL) { 1413 error = sglist_append_mbuf(sg, m->m_next); 1414 MPASS(error == 0); 1415 } 1416 1417 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg); 1418 1419 return (error); 1420 } 1421 1422 static int 1423 vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1424 { 1425 struct vtnet_softc *sc; 1426 struct mbuf *m; 1427 int error; 1428 1429 sc = rxq->vtnrx_sc; 1430 1431 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1432 if (m == NULL) 1433 return (ENOBUFS); 1434 1435 error = vtnet_rxq_enqueue_buf(rxq, m); 1436 if (error) 1437 m_freem(m); 1438 1439 return (error); 1440 } 1441 1442 /* 1443 * Use the checksum offset in the VirtIO header to set the 1444 * correct CSUM_* flags. 1445 */ 1446 static int 1447 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m, 1448 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1449 { 1450 struct vtnet_softc *sc; 1451 #if defined(INET) || defined(INET6) 1452 int offset = hdr->csum_start + hdr->csum_offset; 1453 #endif 1454 1455 sc = rxq->vtnrx_sc; 1456 1457 /* Only do a basic sanity check on the offset. */ 1458 switch (eth_type) { 1459 #if defined(INET) 1460 case ETHERTYPE_IP: 1461 if (__predict_false(offset < ip_start + sizeof(struct ip))) 1462 return (1); 1463 break; 1464 #endif 1465 #if defined(INET6) 1466 case ETHERTYPE_IPV6: 1467 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr))) 1468 return (1); 1469 break; 1470 #endif 1471 default: 1472 sc->vtnet_stats.rx_csum_bad_ethtype++; 1473 return (1); 1474 } 1475 1476 /* 1477 * Use the offset to determine the appropriate CSUM_* flags. This is 1478 * a bit dirty, but we can get by with it since the checksum offsets 1479 * happen to be different. We assume the host host does not do IPv4 1480 * header checksum offloading. 1481 */ 1482 switch (hdr->csum_offset) { 1483 case offsetof(struct udphdr, uh_sum): 1484 case offsetof(struct tcphdr, th_sum): 1485 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1486 m->m_pkthdr.csum_data = 0xFFFF; 1487 break; 1488 case offsetof(struct sctphdr, checksum): 1489 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 1490 break; 1491 default: 1492 sc->vtnet_stats.rx_csum_bad_offset++; 1493 return (1); 1494 } 1495 1496 return (0); 1497 } 1498 1499 static int 1500 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m, 1501 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1502 { 1503 struct vtnet_softc *sc; 1504 int offset, proto; 1505 1506 sc = rxq->vtnrx_sc; 1507 1508 switch (eth_type) { 1509 #if defined(INET) 1510 case ETHERTYPE_IP: { 1511 struct ip *ip; 1512 if (__predict_false(m->m_len < ip_start + sizeof(struct ip))) 1513 return (1); 1514 ip = (struct ip *)(m->m_data + ip_start); 1515 proto = ip->ip_p; 1516 offset = ip_start + (ip->ip_hl << 2); 1517 break; 1518 } 1519 #endif 1520 #if defined(INET6) 1521 case ETHERTYPE_IPV6: 1522 if (__predict_false(m->m_len < ip_start + 1523 sizeof(struct ip6_hdr))) 1524 return (1); 1525 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto); 1526 if (__predict_false(offset < 0)) 1527 return (1); 1528 break; 1529 #endif 1530 default: 1531 sc->vtnet_stats.rx_csum_bad_ethtype++; 1532 return (1); 1533 } 1534 1535 switch (proto) { 1536 case IPPROTO_TCP: 1537 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) 1538 return (1); 1539 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1540 m->m_pkthdr.csum_data = 0xFFFF; 1541 break; 1542 case IPPROTO_UDP: 1543 if (__predict_false(m->m_len < offset + sizeof(struct udphdr))) 1544 return (1); 1545 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1546 m->m_pkthdr.csum_data = 0xFFFF; 1547 break; 1548 case IPPROTO_SCTP: 1549 if (__predict_false(m->m_len < offset + sizeof(struct sctphdr))) 1550 return (1); 1551 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 1552 break; 1553 default: 1554 /* 1555 * For the remaining protocols, FreeBSD does not support 1556 * checksum offloading, so the checksum will be recomputed. 1557 */ 1558 #if 0 1559 if_printf(sc->vtnet_ifp, "cksum offload of unsupported " 1560 "protocol eth_type=%#x proto=%d csum_start=%d " 1561 "csum_offset=%d\n", __func__, eth_type, proto, 1562 hdr->csum_start, hdr->csum_offset); 1563 #endif 1564 break; 1565 } 1566 1567 return (0); 1568 } 1569 1570 /* 1571 * Set the appropriate CSUM_* flags. Unfortunately, the information 1572 * provided is not directly useful to us. The VirtIO header gives the 1573 * offset of the checksum, which is all Linux needs, but this is not 1574 * how FreeBSD does things. We are forced to peek inside the packet 1575 * a bit. 1576 * 1577 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD 1578 * could accept the offsets and let the stack figure it out. 1579 */ 1580 static int 1581 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1582 struct virtio_net_hdr *hdr) 1583 { 1584 struct ether_header *eh; 1585 struct ether_vlan_header *evh; 1586 uint16_t eth_type; 1587 int offset, error; 1588 1589 eh = mtod(m, struct ether_header *); 1590 eth_type = ntohs(eh->ether_type); 1591 if (eth_type == ETHERTYPE_VLAN) { 1592 /* BMV: We should handle nested VLAN tags too. */ 1593 evh = mtod(m, struct ether_vlan_header *); 1594 eth_type = ntohs(evh->evl_proto); 1595 offset = sizeof(struct ether_vlan_header); 1596 } else 1597 offset = sizeof(struct ether_header); 1598 1599 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1600 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr); 1601 else 1602 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr); 1603 1604 return (error); 1605 } 1606 1607 static void 1608 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1609 { 1610 struct mbuf *m; 1611 1612 while (--nbufs > 0) { 1613 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1614 if (m == NULL) 1615 break; 1616 vtnet_rxq_discard_buf(rxq, m); 1617 } 1618 } 1619 1620 static void 1621 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1622 { 1623 int error; 1624 1625 /* 1626 * Requeue the discarded mbuf. This should always be successful 1627 * since it was just dequeued. 1628 */ 1629 error = vtnet_rxq_enqueue_buf(rxq, m); 1630 KASSERT(error == 0, 1631 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1632 } 1633 1634 static int 1635 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1636 { 1637 struct vtnet_softc *sc; 1638 struct ifnet *ifp; 1639 struct virtqueue *vq; 1640 struct mbuf *m, *m_tail; 1641 int len; 1642 1643 sc = rxq->vtnrx_sc; 1644 vq = rxq->vtnrx_vq; 1645 ifp = sc->vtnet_ifp; 1646 m_tail = m_head; 1647 1648 while (--nbufs > 0) { 1649 m = virtqueue_dequeue(vq, &len); 1650 if (m == NULL) { 1651 rxq->vtnrx_stats.vrxs_ierrors++; 1652 goto fail; 1653 } 1654 1655 if (vtnet_rxq_new_buf(rxq) != 0) { 1656 rxq->vtnrx_stats.vrxs_iqdrops++; 1657 vtnet_rxq_discard_buf(rxq, m); 1658 if (nbufs > 1) 1659 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1660 goto fail; 1661 } 1662 1663 if (m->m_len < len) 1664 len = m->m_len; 1665 1666 m->m_len = len; 1667 m->m_flags &= ~M_PKTHDR; 1668 1669 m_head->m_pkthdr.len += len; 1670 m_tail->m_next = m; 1671 m_tail = m; 1672 } 1673 1674 return (0); 1675 1676 fail: 1677 sc->vtnet_stats.rx_mergeable_failed++; 1678 m_freem(m_head); 1679 1680 return (1); 1681 } 1682 1683 static void 1684 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 1685 struct virtio_net_hdr *hdr) 1686 { 1687 struct vtnet_softc *sc; 1688 struct ifnet *ifp; 1689 struct ether_header *eh; 1690 1691 sc = rxq->vtnrx_sc; 1692 ifp = sc->vtnet_ifp; 1693 1694 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { 1695 eh = mtod(m, struct ether_header *); 1696 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1697 vtnet_vlan_tag_remove(m); 1698 /* 1699 * With the 802.1Q header removed, update the 1700 * checksum starting location accordingly. 1701 */ 1702 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1703 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 1704 } 1705 } 1706 1707 m->m_pkthdr.flowid = rxq->vtnrx_id; 1708 m->m_flags |= M_FLOWID; 1709 1710 /* 1711 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum 1712 * distinction that Linux does. Need to reevaluate if performing 1713 * offloading for the NEEDS_CSUM case is really appropriate. 1714 */ 1715 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM | 1716 VIRTIO_NET_HDR_F_DATA_VALID)) { 1717 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 1718 rxq->vtnrx_stats.vrxs_csum++; 1719 else 1720 rxq->vtnrx_stats.vrxs_csum_failed++; 1721 } 1722 1723 rxq->vtnrx_stats.vrxs_ipackets++; 1724 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 1725 1726 VTNET_RXQ_UNLOCK(rxq); 1727 (*ifp->if_input)(ifp, m); 1728 VTNET_RXQ_LOCK(rxq); 1729 } 1730 1731 static int 1732 vtnet_rxq_eof(struct vtnet_rxq *rxq) 1733 { 1734 struct virtio_net_hdr lhdr, *hdr; 1735 struct vtnet_softc *sc; 1736 struct ifnet *ifp; 1737 struct virtqueue *vq; 1738 struct mbuf *m; 1739 struct virtio_net_hdr_mrg_rxbuf *mhdr; 1740 int len, deq, nbufs, adjsz, count; 1741 1742 sc = rxq->vtnrx_sc; 1743 vq = rxq->vtnrx_vq; 1744 ifp = sc->vtnet_ifp; 1745 hdr = &lhdr; 1746 deq = 0; 1747 count = sc->vtnet_rx_process_limit; 1748 1749 VTNET_RXQ_LOCK_ASSERT(rxq); 1750 1751 while (count-- > 0) { 1752 m = virtqueue_dequeue(vq, &len); 1753 if (m == NULL) 1754 break; 1755 deq++; 1756 1757 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 1758 rxq->vtnrx_stats.vrxs_ierrors++; 1759 vtnet_rxq_discard_buf(rxq, m); 1760 continue; 1761 } 1762 1763 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1764 nbufs = 1; 1765 adjsz = sizeof(struct vtnet_rx_header); 1766 /* 1767 * Account for our pad inserted between the header 1768 * and the actual start of the frame. 1769 */ 1770 len += VTNET_RX_HEADER_PAD; 1771 } else { 1772 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 1773 nbufs = mhdr->num_buffers; 1774 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1775 } 1776 1777 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 1778 rxq->vtnrx_stats.vrxs_iqdrops++; 1779 vtnet_rxq_discard_buf(rxq, m); 1780 if (nbufs > 1) 1781 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1782 continue; 1783 } 1784 1785 m->m_pkthdr.len = len; 1786 m->m_pkthdr.rcvif = ifp; 1787 m->m_pkthdr.csum_flags = 0; 1788 1789 if (nbufs > 1) { 1790 /* Dequeue the rest of chain. */ 1791 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 1792 continue; 1793 } 1794 1795 /* 1796 * Save copy of header before we strip it. For both mergeable 1797 * and non-mergeable, the header is at the beginning of the 1798 * mbuf data. We no longer need num_buffers, so always use a 1799 * regular header. 1800 * 1801 * BMV: Is this memcpy() expensive? We know the mbuf data is 1802 * still valid even after the m_adj(). 1803 */ 1804 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr)); 1805 m_adj(m, adjsz); 1806 1807 vtnet_rxq_input(rxq, m, hdr); 1808 1809 /* Must recheck after dropping the Rx lock. */ 1810 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1811 break; 1812 } 1813 1814 if (deq > 0) 1815 virtqueue_notify(vq); 1816 1817 return (count > 0 ? 0 : EAGAIN); 1818 } 1819 1820 static void 1821 vtnet_rx_vq_intr(void *xrxq) 1822 { 1823 struct vtnet_softc *sc; 1824 struct vtnet_rxq *rxq; 1825 struct ifnet *ifp; 1826 int tries, more; 1827 1828 rxq = xrxq; 1829 sc = rxq->vtnrx_sc; 1830 ifp = sc->vtnet_ifp; 1831 tries = 0; 1832 1833 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 1834 /* 1835 * Ignore this interrupt. Either this is a spurious interrupt 1836 * or multiqueue without per-VQ MSIX so every queue needs to 1837 * be polled (a brain dead configuration we could try harder 1838 * to avoid). 1839 */ 1840 vtnet_rxq_disable_intr(rxq); 1841 return; 1842 } 1843 1844 VTNET_RXQ_LOCK(rxq); 1845 1846 again: 1847 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1848 VTNET_RXQ_UNLOCK(rxq); 1849 return; 1850 } 1851 1852 more = vtnet_rxq_eof(rxq); 1853 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1854 if (!more) 1855 vtnet_rxq_disable_intr(rxq); 1856 /* 1857 * This is an occasional condition or race (when !more), 1858 * so retry a few times before scheduling the taskqueue. 1859 */ 1860 if (tries++ < VTNET_INTR_DISABLE_RETRIES) 1861 goto again; 1862 1863 VTNET_RXQ_UNLOCK(rxq); 1864 rxq->vtnrx_stats.vrxs_rescheduled++; 1865 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1866 } else 1867 VTNET_RXQ_UNLOCK(rxq); 1868 } 1869 1870 static void 1871 vtnet_rxq_tq_intr(void *xrxq, int pending) 1872 { 1873 struct vtnet_softc *sc; 1874 struct vtnet_rxq *rxq; 1875 struct ifnet *ifp; 1876 int more; 1877 1878 rxq = xrxq; 1879 sc = rxq->vtnrx_sc; 1880 ifp = sc->vtnet_ifp; 1881 1882 VTNET_RXQ_LOCK(rxq); 1883 1884 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1885 VTNET_RXQ_UNLOCK(rxq); 1886 return; 1887 } 1888 1889 more = vtnet_rxq_eof(rxq); 1890 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1891 if (!more) 1892 vtnet_rxq_disable_intr(rxq); 1893 rxq->vtnrx_stats.vrxs_rescheduled++; 1894 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1895 } 1896 1897 VTNET_RXQ_UNLOCK(rxq); 1898 } 1899 1900 static void 1901 vtnet_txq_free_mbufs(struct vtnet_txq *txq) 1902 { 1903 struct virtqueue *vq; 1904 struct vtnet_tx_header *txhdr; 1905 int last; 1906 1907 vq = txq->vtntx_vq; 1908 last = 0; 1909 1910 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 1911 m_freem(txhdr->vth_mbuf); 1912 uma_zfree(vtnet_tx_header_zone, txhdr); 1913 } 1914 1915 KASSERT(virtqueue_empty(vq), 1916 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 1917 } 1918 1919 /* 1920 * BMV: Much of this can go away once we finally have offsets in 1921 * the mbuf packet header. Bug andre@. 1922 */ 1923 static int 1924 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, 1925 int *etype, int *proto, int *start) 1926 { 1927 struct vtnet_softc *sc; 1928 struct ether_vlan_header *evh; 1929 int offset; 1930 1931 sc = txq->vtntx_sc; 1932 1933 evh = mtod(m, struct ether_vlan_header *); 1934 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 1935 /* BMV: We should handle nested VLAN tags too. */ 1936 *etype = ntohs(evh->evl_proto); 1937 offset = sizeof(struct ether_vlan_header); 1938 } else { 1939 *etype = ntohs(evh->evl_encap_proto); 1940 offset = sizeof(struct ether_header); 1941 } 1942 1943 switch (*etype) { 1944 #if defined(INET) 1945 case ETHERTYPE_IP: { 1946 struct ip *ip, iphdr; 1947 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 1948 m_copydata(m, offset, sizeof(struct ip), 1949 (caddr_t) &iphdr); 1950 ip = &iphdr; 1951 } else 1952 ip = (struct ip *)(m->m_data + offset); 1953 *proto = ip->ip_p; 1954 *start = offset + (ip->ip_hl << 2); 1955 break; 1956 } 1957 #endif 1958 #if defined(INET6) 1959 case ETHERTYPE_IPV6: 1960 *proto = -1; 1961 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 1962 /* Assert the network stack sent us a valid packet. */ 1963 KASSERT(*start > offset, 1964 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 1965 *start, offset, *proto)); 1966 break; 1967 #endif 1968 default: 1969 sc->vtnet_stats.tx_csum_bad_ethtype++; 1970 return (EINVAL); 1971 } 1972 1973 return (0); 1974 } 1975 1976 static int 1977 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 1978 int offset, struct virtio_net_hdr *hdr) 1979 { 1980 static struct timeval lastecn; 1981 static int curecn; 1982 struct vtnet_softc *sc; 1983 struct tcphdr *tcp, tcphdr; 1984 1985 sc = txq->vtntx_sc; 1986 1987 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 1988 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 1989 tcp = &tcphdr; 1990 } else 1991 tcp = (struct tcphdr *)(m->m_data + offset); 1992 1993 hdr->hdr_len = offset + (tcp->th_off << 2); 1994 hdr->gso_size = m->m_pkthdr.tso_segsz; 1995 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 1996 VIRTIO_NET_HDR_GSO_TCPV6; 1997 1998 if (tcp->th_flags & TH_CWR) { 1999 /* 2000 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD, 2001 * ECN support is not on a per-interface basis, but globally via 2002 * the net.inet.tcp.ecn.enable sysctl knob. The default is off. 2003 */ 2004 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2005 if (ppsratecheck(&lastecn, &curecn, 1)) 2006 if_printf(sc->vtnet_ifp, 2007 "TSO with ECN not negotiated with host\n"); 2008 return (ENOTSUP); 2009 } 2010 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2011 } 2012 2013 txq->vtntx_stats.vtxs_tso++; 2014 2015 return (0); 2016 } 2017 2018 static struct mbuf * 2019 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2020 struct virtio_net_hdr *hdr) 2021 { 2022 struct vtnet_softc *sc; 2023 int flags, etype, csum_start, proto, error; 2024 2025 sc = txq->vtntx_sc; 2026 flags = m->m_pkthdr.csum_flags; 2027 2028 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2029 if (error) 2030 goto drop; 2031 2032 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) || 2033 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) { 2034 /* 2035 * We could compare the IP protocol vs the CSUM_ flag too, 2036 * but that really should not be necessary. 2037 */ 2038 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2039 hdr->csum_start = csum_start; 2040 hdr->csum_offset = m->m_pkthdr.csum_data; 2041 txq->vtntx_stats.vtxs_csum++; 2042 } 2043 2044 if (flags & CSUM_TSO) { 2045 if (__predict_false(proto != IPPROTO_TCP)) { 2046 /* Likely failed to correctly parse the mbuf. */ 2047 sc->vtnet_stats.tx_tso_not_tcp++; 2048 goto drop; 2049 } 2050 2051 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM, 2052 ("%s: mbuf %p TSO without checksum offload %#x", 2053 __func__, m, flags)); 2054 2055 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2056 if (error) 2057 goto drop; 2058 } 2059 2060 return (m); 2061 2062 drop: 2063 m_freem(m); 2064 return (NULL); 2065 } 2066 2067 static int 2068 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2069 struct vtnet_tx_header *txhdr) 2070 { 2071 struct vtnet_softc *sc; 2072 struct virtqueue *vq; 2073 struct sglist *sg; 2074 struct mbuf *m; 2075 int error; 2076 2077 sc = txq->vtntx_sc; 2078 vq = txq->vtntx_vq; 2079 sg = txq->vtntx_sg; 2080 m = *m_head; 2081 2082 sglist_reset(sg); 2083 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2084 KASSERT(error == 0 && sg->sg_nseg == 1, 2085 ("%s: error %d adding header to sglist", __func__, error)); 2086 2087 error = sglist_append_mbuf(sg, m); 2088 if (error) { 2089 m = m_defrag(m, M_NOWAIT); 2090 if (m == NULL) 2091 goto fail; 2092 2093 *m_head = m; 2094 sc->vtnet_stats.tx_defragged++; 2095 2096 error = sglist_append_mbuf(sg, m); 2097 if (error) 2098 goto fail; 2099 } 2100 2101 txhdr->vth_mbuf = m; 2102 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2103 2104 return (error); 2105 2106 fail: 2107 sc->vtnet_stats.tx_defrag_failed++; 2108 m_freem(*m_head); 2109 *m_head = NULL; 2110 2111 return (ENOBUFS); 2112 } 2113 2114 static int 2115 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head) 2116 { 2117 struct vtnet_tx_header *txhdr; 2118 struct virtio_net_hdr *hdr; 2119 struct mbuf *m; 2120 int error; 2121 2122 m = *m_head; 2123 M_ASSERTPKTHDR(m); 2124 2125 txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO); 2126 if (txhdr == NULL) { 2127 m_freem(m); 2128 *m_head = NULL; 2129 return (ENOMEM); 2130 } 2131 2132 /* 2133 * Always use the non-mergeable header, regardless if the feature 2134 * was negotiated. For transmit, num_buffers is always zero. The 2135 * vtnet_hdr_size is used to enqueue the correct header size. 2136 */ 2137 hdr = &txhdr->vth_uhdr.hdr; 2138 2139 if (m->m_flags & M_VLANTAG) { 2140 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2141 if ((*m_head = m) == NULL) { 2142 error = ENOBUFS; 2143 goto fail; 2144 } 2145 m->m_flags &= ~M_VLANTAG; 2146 } 2147 2148 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2149 m = vtnet_txq_offload(txq, m, hdr); 2150 if ((*m_head = m) == NULL) { 2151 error = ENOBUFS; 2152 goto fail; 2153 } 2154 } 2155 2156 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2157 if (error == 0) 2158 return (0); 2159 2160 fail: 2161 uma_zfree(vtnet_tx_header_zone, txhdr); 2162 2163 return (error); 2164 } 2165 2166 #ifdef VTNET_LEGACY_TX 2167 2168 static void 2169 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp) 2170 { 2171 struct vtnet_softc *sc; 2172 struct virtqueue *vq; 2173 struct mbuf *m0; 2174 int enq; 2175 2176 sc = txq->vtntx_sc; 2177 vq = txq->vtntx_vq; 2178 enq = 0; 2179 2180 VTNET_TXQ_LOCK_ASSERT(txq); 2181 2182 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2183 sc->vtnet_link_active == 0) 2184 return; 2185 2186 vtnet_txq_eof(txq); 2187 2188 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { 2189 if (virtqueue_full(vq)) 2190 break; 2191 2192 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0); 2193 if (m0 == NULL) 2194 break; 2195 2196 if (vtnet_txq_encap(txq, &m0) != 0) { 2197 if (m0 != NULL) 2198 IFQ_DRV_PREPEND(&ifp->if_snd, m0); 2199 break; 2200 } 2201 2202 enq++; 2203 ETHER_BPF_MTAP(ifp, m0); 2204 } 2205 2206 if (enq > 0) { 2207 virtqueue_notify(vq); 2208 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2209 } 2210 } 2211 2212 static void 2213 vtnet_start(struct ifnet *ifp) 2214 { 2215 struct vtnet_softc *sc; 2216 struct vtnet_txq *txq; 2217 2218 sc = ifp->if_softc; 2219 txq = &sc->vtnet_txqs[0]; 2220 2221 VTNET_TXQ_LOCK(txq); 2222 vtnet_start_locked(txq, ifp); 2223 VTNET_TXQ_UNLOCK(txq); 2224 } 2225 2226 #else /* !VTNET_LEGACY_TX */ 2227 2228 static int 2229 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2230 { 2231 struct vtnet_softc *sc; 2232 struct virtqueue *vq; 2233 struct buf_ring *br; 2234 struct ifnet *ifp; 2235 int enq, error; 2236 2237 sc = txq->vtntx_sc; 2238 vq = txq->vtntx_vq; 2239 br = txq->vtntx_br; 2240 ifp = sc->vtnet_ifp; 2241 enq = 0; 2242 error = 0; 2243 2244 VTNET_TXQ_LOCK_ASSERT(txq); 2245 2246 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2247 sc->vtnet_link_active == 0) { 2248 if (m != NULL) 2249 error = drbr_enqueue(ifp, br, m); 2250 return (error); 2251 } 2252 2253 if (m != NULL) { 2254 error = drbr_enqueue(ifp, br, m); 2255 if (error) 2256 return (error); 2257 } 2258 2259 vtnet_txq_eof(txq); 2260 2261 while ((m = drbr_peek(ifp, br)) != NULL) { 2262 if (virtqueue_full(vq)) { 2263 drbr_putback(ifp, br, m); 2264 break; 2265 } 2266 2267 error = vtnet_txq_encap(txq, &m); 2268 if (error) { 2269 if (m != NULL) 2270 drbr_putback(ifp, br, m); 2271 else 2272 drbr_advance(ifp, br); 2273 break; 2274 } 2275 drbr_advance(ifp, br); 2276 2277 enq++; 2278 ETHER_BPF_MTAP(ifp, m); 2279 } 2280 2281 if (enq > 0) { 2282 virtqueue_notify(vq); 2283 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2284 } 2285 2286 return (0); 2287 } 2288 2289 static int 2290 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m) 2291 { 2292 struct vtnet_softc *sc; 2293 struct vtnet_txq *txq; 2294 int i, npairs, error; 2295 2296 sc = ifp->if_softc; 2297 npairs = sc->vtnet_act_vq_pairs; 2298 2299 if (m->m_flags & M_FLOWID) 2300 i = m->m_pkthdr.flowid % npairs; 2301 else 2302 i = curcpu % npairs; 2303 2304 txq = &sc->vtnet_txqs[i]; 2305 2306 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2307 error = vtnet_txq_mq_start_locked(txq, m); 2308 VTNET_TXQ_UNLOCK(txq); 2309 } else { 2310 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2311 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2312 } 2313 2314 return (error); 2315 } 2316 2317 static void 2318 vtnet_txq_tq_deferred(void *xtxq, int pending) 2319 { 2320 struct vtnet_softc *sc; 2321 struct vtnet_txq *txq; 2322 2323 txq = xtxq; 2324 sc = txq->vtntx_sc; 2325 2326 VTNET_TXQ_LOCK(txq); 2327 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2328 vtnet_txq_mq_start_locked(txq, NULL); 2329 VTNET_TXQ_UNLOCK(txq); 2330 } 2331 2332 #endif /* VTNET_LEGACY_TX */ 2333 2334 static void 2335 vtnet_txq_start(struct vtnet_txq *txq) 2336 { 2337 struct vtnet_softc *sc; 2338 struct ifnet *ifp; 2339 2340 sc = txq->vtntx_sc; 2341 ifp = sc->vtnet_ifp; 2342 2343 #ifdef VTNET_LEGACY_TX 2344 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 2345 vtnet_start_locked(txq, ifp); 2346 #else 2347 if (!drbr_empty(ifp, txq->vtntx_br)) 2348 vtnet_txq_mq_start_locked(txq, NULL); 2349 #endif 2350 } 2351 2352 static void 2353 vtnet_txq_tq_intr(void *xtxq, int pending) 2354 { 2355 struct vtnet_softc *sc; 2356 struct vtnet_txq *txq; 2357 struct ifnet *ifp; 2358 2359 txq = xtxq; 2360 sc = txq->vtntx_sc; 2361 ifp = sc->vtnet_ifp; 2362 2363 VTNET_TXQ_LOCK(txq); 2364 2365 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2366 VTNET_TXQ_UNLOCK(txq); 2367 return; 2368 } 2369 2370 vtnet_txq_eof(txq); 2371 2372 vtnet_txq_start(txq); 2373 2374 if (vtnet_txq_enable_intr(txq) != 0) { 2375 vtnet_txq_disable_intr(txq); 2376 txq->vtntx_stats.vtxs_rescheduled++; 2377 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2378 } 2379 2380 VTNET_TXQ_UNLOCK(txq); 2381 } 2382 2383 static void 2384 vtnet_txq_eof(struct vtnet_txq *txq) 2385 { 2386 struct virtqueue *vq; 2387 struct vtnet_tx_header *txhdr; 2388 struct mbuf *m; 2389 2390 vq = txq->vtntx_vq; 2391 VTNET_TXQ_LOCK_ASSERT(txq); 2392 2393 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2394 m = txhdr->vth_mbuf; 2395 2396 txq->vtntx_stats.vtxs_opackets++; 2397 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2398 if (m->m_flags & M_MCAST) 2399 txq->vtntx_stats.vtxs_omcasts++; 2400 2401 m_freem(m); 2402 uma_zfree(vtnet_tx_header_zone, txhdr); 2403 } 2404 2405 if (virtqueue_empty(vq)) 2406 txq->vtntx_watchdog = 0; 2407 } 2408 2409 static void 2410 vtnet_tx_vq_intr(void *xtxq) 2411 { 2412 struct vtnet_softc *sc; 2413 struct vtnet_txq *txq; 2414 struct ifnet *ifp; 2415 int tries; 2416 2417 txq = xtxq; 2418 sc = txq->vtntx_sc; 2419 ifp = sc->vtnet_ifp; 2420 tries = 0; 2421 2422 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2423 /* 2424 * Ignore this interrupt. Either this is a spurious interrupt 2425 * or multiqueue without per-VQ MSIX so every queue needs to 2426 * be polled (a brain dead configuration we could try harder 2427 * to avoid). 2428 */ 2429 vtnet_txq_disable_intr(txq); 2430 return; 2431 } 2432 2433 VTNET_TXQ_LOCK(txq); 2434 2435 again: 2436 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2437 VTNET_TXQ_UNLOCK(txq); 2438 return; 2439 } 2440 2441 vtnet_txq_eof(txq); 2442 2443 vtnet_txq_start(txq); 2444 2445 if (vtnet_txq_enable_intr(txq) != 0) { 2446 vtnet_txq_disable_intr(txq); 2447 /* 2448 * This is an occasional race, so retry a few times 2449 * before scheduling the taskqueue. 2450 */ 2451 if (tries++ < VTNET_INTR_DISABLE_RETRIES) 2452 goto again; 2453 2454 VTNET_TXQ_UNLOCK(txq); 2455 txq->vtntx_stats.vtxs_rescheduled++; 2456 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2457 } else 2458 VTNET_TXQ_UNLOCK(txq); 2459 } 2460 2461 static void 2462 vtnet_tx_start_all(struct vtnet_softc *sc) 2463 { 2464 struct vtnet_txq *txq; 2465 int i; 2466 2467 VTNET_CORE_LOCK_ASSERT(sc); 2468 2469 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2470 txq = &sc->vtnet_txqs[i]; 2471 2472 VTNET_TXQ_LOCK(txq); 2473 vtnet_txq_start(txq); 2474 VTNET_TXQ_UNLOCK(txq); 2475 } 2476 } 2477 2478 #ifndef VTNET_LEGACY_TX 2479 static void 2480 vtnet_qflush(struct ifnet *ifp) 2481 { 2482 struct vtnet_softc *sc; 2483 struct vtnet_txq *txq; 2484 struct mbuf *m; 2485 int i; 2486 2487 sc = ifp->if_softc; 2488 2489 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2490 txq = &sc->vtnet_txqs[i]; 2491 2492 VTNET_TXQ_LOCK(txq); 2493 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2494 m_freem(m); 2495 VTNET_TXQ_UNLOCK(txq); 2496 } 2497 2498 if_qflush(ifp); 2499 } 2500 #endif 2501 2502 static int 2503 vtnet_watchdog(struct vtnet_txq *txq) 2504 { 2505 struct vtnet_softc *sc; 2506 2507 sc = txq->vtntx_sc; 2508 2509 VTNET_TXQ_LOCK(txq); 2510 if (sc->vtnet_flags & VTNET_FLAG_EVENT_IDX) 2511 vtnet_txq_eof(txq); 2512 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 2513 VTNET_TXQ_UNLOCK(txq); 2514 return (0); 2515 } 2516 VTNET_TXQ_UNLOCK(txq); 2517 2518 if_printf(sc->vtnet_ifp, "watchdog timeout on queue %d\n", 2519 txq->vtntx_id); 2520 return (1); 2521 } 2522 2523 static void 2524 vtnet_rxq_accum_stats(struct vtnet_rxq *rxq, struct vtnet_rxq_stats *accum) 2525 { 2526 struct vtnet_rxq_stats *st; 2527 2528 st = &rxq->vtnrx_stats; 2529 2530 accum->vrxs_ipackets += st->vrxs_ipackets; 2531 accum->vrxs_ibytes += st->vrxs_ibytes; 2532 accum->vrxs_iqdrops += st->vrxs_iqdrops; 2533 accum->vrxs_csum += st->vrxs_csum; 2534 accum->vrxs_csum_failed += st->vrxs_csum_failed; 2535 accum->vrxs_rescheduled += st->vrxs_rescheduled; 2536 } 2537 2538 static void 2539 vtnet_txq_accum_stats(struct vtnet_txq *txq, struct vtnet_txq_stats *accum) 2540 { 2541 struct vtnet_txq_stats *st; 2542 2543 st = &txq->vtntx_stats; 2544 2545 accum->vtxs_opackets += st->vtxs_opackets; 2546 accum->vtxs_obytes += st->vtxs_obytes; 2547 accum->vtxs_csum += st->vtxs_csum; 2548 accum->vtxs_tso += st->vtxs_tso; 2549 accum->vtxs_rescheduled += st->vtxs_rescheduled; 2550 } 2551 2552 static void 2553 vtnet_accumulate_stats(struct vtnet_softc *sc) 2554 { 2555 struct ifnet *ifp; 2556 struct vtnet_statistics *st; 2557 struct vtnet_rxq_stats rxaccum; 2558 struct vtnet_txq_stats txaccum; 2559 int i; 2560 2561 ifp = sc->vtnet_ifp; 2562 st = &sc->vtnet_stats; 2563 bzero(&rxaccum, sizeof(struct vtnet_rxq_stats)); 2564 bzero(&txaccum, sizeof(struct vtnet_txq_stats)); 2565 2566 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2567 vtnet_rxq_accum_stats(&sc->vtnet_rxqs[i], &rxaccum); 2568 vtnet_txq_accum_stats(&sc->vtnet_txqs[i], &txaccum); 2569 } 2570 2571 st->rx_csum_offloaded = rxaccum.vrxs_csum; 2572 st->rx_csum_failed = rxaccum.vrxs_csum_failed; 2573 st->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 2574 st->tx_csum_offloaded = txaccum.vtxs_csum; 2575 st->tx_tso_offloaded = txaccum.vtxs_tso; 2576 st->tx_task_rescheduled = txaccum.vtxs_rescheduled; 2577 2578 /* 2579 * With the exception of if_ierrors, these ifnet statistics are 2580 * only updated in the driver, so just set them to our accumulated 2581 * values. if_ierrors is updated in ether_input() for malformed 2582 * frames that we should have already discarded. 2583 */ 2584 ifp->if_ipackets = rxaccum.vrxs_ipackets; 2585 ifp->if_iqdrops = rxaccum.vrxs_iqdrops; 2586 ifp->if_ierrors = rxaccum.vrxs_ierrors; 2587 ifp->if_opackets = txaccum.vtxs_opackets; 2588 #ifndef VTNET_LEGACY_TX 2589 ifp->if_obytes = txaccum.vtxs_obytes; 2590 ifp->if_omcasts = txaccum.vtxs_omcasts; 2591 #endif 2592 } 2593 2594 static void 2595 vtnet_tick(void *xsc) 2596 { 2597 struct vtnet_softc *sc; 2598 struct ifnet *ifp; 2599 int i, timedout; 2600 2601 sc = xsc; 2602 ifp = sc->vtnet_ifp; 2603 timedout = 0; 2604 2605 VTNET_CORE_LOCK_ASSERT(sc); 2606 vtnet_accumulate_stats(sc); 2607 2608 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 2609 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 2610 2611 if (timedout != 0) { 2612 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2613 vtnet_init_locked(sc); 2614 } else 2615 callout_schedule(&sc->vtnet_tick_ch, hz); 2616 } 2617 2618 static void 2619 vtnet_start_taskqueues(struct vtnet_softc *sc) 2620 { 2621 device_t dev; 2622 struct vtnet_rxq *rxq; 2623 struct vtnet_txq *txq; 2624 int i, error; 2625 2626 dev = sc->vtnet_dev; 2627 2628 /* 2629 * Errors here are very difficult to recover from - we cannot 2630 * easily fail because, if this is during boot, we will hang 2631 * when freeing any successfully started taskqueues because 2632 * the scheduler isn't up yet. 2633 * 2634 * Most drivers just ignore the return value - it only fails 2635 * with ENOMEM so an error is not likely. 2636 */ 2637 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2638 rxq = &sc->vtnet_rxqs[i]; 2639 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 2640 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 2641 if (error) { 2642 device_printf(dev, "failed to start rx taskq %d\n", 2643 rxq->vtnrx_id); 2644 } 2645 2646 txq = &sc->vtnet_txqs[i]; 2647 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 2648 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 2649 if (error) { 2650 device_printf(dev, "failed to start tx taskq %d\n", 2651 txq->vtntx_id); 2652 } 2653 } 2654 } 2655 2656 static void 2657 vtnet_free_taskqueues(struct vtnet_softc *sc) 2658 { 2659 struct vtnet_rxq *rxq; 2660 struct vtnet_txq *txq; 2661 int i; 2662 2663 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2664 rxq = &sc->vtnet_rxqs[i]; 2665 if (rxq->vtnrx_tq != NULL) { 2666 taskqueue_free(rxq->vtnrx_tq); 2667 rxq->vtnrx_vq = NULL; 2668 } 2669 2670 txq = &sc->vtnet_txqs[i]; 2671 if (txq->vtntx_tq != NULL) { 2672 taskqueue_free(txq->vtntx_tq); 2673 txq->vtntx_tq = NULL; 2674 } 2675 } 2676 } 2677 2678 static void 2679 vtnet_drain_taskqueues(struct vtnet_softc *sc) 2680 { 2681 struct vtnet_rxq *rxq; 2682 struct vtnet_txq *txq; 2683 int i; 2684 2685 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2686 rxq = &sc->vtnet_rxqs[i]; 2687 if (rxq->vtnrx_tq != NULL) 2688 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2689 2690 txq = &sc->vtnet_txqs[i]; 2691 if (txq->vtntx_tq != NULL) { 2692 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 2693 #ifndef VTNET_LEGACY_TX 2694 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 2695 #endif 2696 } 2697 } 2698 } 2699 2700 static void 2701 vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 2702 { 2703 struct vtnet_rxq *rxq; 2704 struct vtnet_txq *txq; 2705 int i; 2706 2707 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2708 rxq = &sc->vtnet_rxqs[i]; 2709 vtnet_rxq_free_mbufs(rxq); 2710 2711 txq = &sc->vtnet_txqs[i]; 2712 vtnet_txq_free_mbufs(txq); 2713 } 2714 } 2715 2716 static void 2717 vtnet_stop_rendezvous(struct vtnet_softc *sc) 2718 { 2719 struct vtnet_rxq *rxq; 2720 struct vtnet_txq *txq; 2721 int i; 2722 2723 /* 2724 * Lock and unlock the per-queue mutex so we known the stop 2725 * state is visible. Doing only the active queues should be 2726 * sufficient, but it does not cost much extra to do all the 2727 * queues. Note we hold the core mutex here too. 2728 */ 2729 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2730 rxq = &sc->vtnet_rxqs[i]; 2731 VTNET_RXQ_LOCK(rxq); 2732 VTNET_RXQ_UNLOCK(rxq); 2733 2734 txq = &sc->vtnet_txqs[i]; 2735 VTNET_TXQ_LOCK(txq); 2736 VTNET_TXQ_UNLOCK(txq); 2737 } 2738 } 2739 2740 static void 2741 vtnet_stop(struct vtnet_softc *sc) 2742 { 2743 device_t dev; 2744 struct ifnet *ifp; 2745 2746 dev = sc->vtnet_dev; 2747 ifp = sc->vtnet_ifp; 2748 2749 VTNET_CORE_LOCK_ASSERT(sc); 2750 2751 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2752 sc->vtnet_link_active = 0; 2753 callout_stop(&sc->vtnet_tick_ch); 2754 2755 /* Only advisory. */ 2756 vtnet_disable_interrupts(sc); 2757 2758 /* 2759 * Stop the host adapter. This resets it to the pre-initialized 2760 * state. It will not generate any interrupts until after it is 2761 * reinitialized. 2762 */ 2763 virtio_stop(dev); 2764 vtnet_stop_rendezvous(sc); 2765 2766 /* Free any mbufs left in the virtqueues. */ 2767 vtnet_drain_rxtx_queues(sc); 2768 } 2769 2770 static int 2771 vtnet_virtio_reinit(struct vtnet_softc *sc) 2772 { 2773 device_t dev; 2774 struct ifnet *ifp; 2775 uint64_t features; 2776 int mask, error; 2777 2778 dev = sc->vtnet_dev; 2779 ifp = sc->vtnet_ifp; 2780 features = sc->vtnet_features; 2781 2782 mask = 0; 2783 #if defined(INET) 2784 mask |= IFCAP_RXCSUM; 2785 #endif 2786 #if defined (INET6) 2787 mask |= IFCAP_RXCSUM_IPV6; 2788 #endif 2789 2790 /* 2791 * Re-negotiate with the host, removing any disabled receive 2792 * features. Transmit features are disabled only on our side 2793 * via if_capenable and if_hwassist. 2794 */ 2795 2796 if (ifp->if_capabilities & mask) { 2797 /* 2798 * We require both IPv4 and IPv6 offloading to be enabled 2799 * in order to negotiated it: VirtIO does not distinguish 2800 * between the two. 2801 */ 2802 if ((ifp->if_capenable & mask) != mask) 2803 features &= ~VIRTIO_NET_F_GUEST_CSUM; 2804 } 2805 2806 if (ifp->if_capabilities & IFCAP_LRO) { 2807 if ((ifp->if_capenable & IFCAP_LRO) == 0) 2808 features &= ~VTNET_LRO_FEATURES; 2809 } 2810 2811 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) { 2812 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) 2813 features &= ~VIRTIO_NET_F_CTRL_VLAN; 2814 } 2815 2816 error = virtio_reinit(dev, features); 2817 if (error) 2818 device_printf(dev, "virtio reinit error %d\n", error); 2819 2820 return (error); 2821 } 2822 2823 static void 2824 vtnet_init_rx_filters(struct vtnet_softc *sc) 2825 { 2826 struct ifnet *ifp; 2827 2828 ifp = sc->vtnet_ifp; 2829 2830 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 2831 /* Restore promiscuous and all-multicast modes. */ 2832 vtnet_rx_filter(sc); 2833 /* Restore filtered MAC addresses. */ 2834 vtnet_rx_filter_mac(sc); 2835 } 2836 2837 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) 2838 vtnet_rx_filter_vlan(sc); 2839 } 2840 2841 static int 2842 vtnet_init_rx_queues(struct vtnet_softc *sc) 2843 { 2844 device_t dev; 2845 struct vtnet_rxq *rxq; 2846 int i, clsize, error; 2847 2848 dev = sc->vtnet_dev; 2849 2850 /* 2851 * Use the new cluster size if one has been set (via a MTU 2852 * change). Otherwise, use the standard 2K clusters. 2853 * 2854 * BMV: It might make sense to use page sized clusters as 2855 * the default (depending on the features negotiated). 2856 */ 2857 if (sc->vtnet_rx_new_clsize != 0) { 2858 clsize = sc->vtnet_rx_new_clsize; 2859 sc->vtnet_rx_new_clsize = 0; 2860 } else 2861 clsize = MCLBYTES; 2862 2863 sc->vtnet_rx_clsize = clsize; 2864 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize); 2865 2866 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS || 2867 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 2868 ("%s: too many rx mbufs %d for %d segments", __func__, 2869 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 2870 2871 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2872 rxq = &sc->vtnet_rxqs[i]; 2873 2874 /* Hold the lock to satisfy asserts. */ 2875 VTNET_RXQ_LOCK(rxq); 2876 error = vtnet_rxq_populate(rxq); 2877 VTNET_RXQ_UNLOCK(rxq); 2878 2879 if (error) { 2880 device_printf(dev, 2881 "cannot allocate mbufs for Rx queue %d\n", i); 2882 return (error); 2883 } 2884 } 2885 2886 return (0); 2887 } 2888 2889 static int 2890 vtnet_init_tx_queues(struct vtnet_softc *sc) 2891 { 2892 struct vtnet_txq *txq; 2893 int i; 2894 2895 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2896 txq = &sc->vtnet_txqs[i]; 2897 txq->vtntx_watchdog = 0; 2898 } 2899 2900 return (0); 2901 } 2902 2903 static int 2904 vtnet_init_rxtx_queues(struct vtnet_softc *sc) 2905 { 2906 int error; 2907 2908 error = vtnet_init_rx_queues(sc); 2909 if (error) 2910 return (error); 2911 2912 error = vtnet_init_tx_queues(sc); 2913 if (error) 2914 return (error); 2915 2916 return (0); 2917 } 2918 2919 static void 2920 vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 2921 { 2922 device_t dev; 2923 int npairs; 2924 2925 dev = sc->vtnet_dev; 2926 2927 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) { 2928 MPASS(sc->vtnet_max_vq_pairs == 1); 2929 sc->vtnet_act_vq_pairs = 1; 2930 return; 2931 } 2932 2933 /* BMV: Just use the maximum configured for now. */ 2934 npairs = sc->vtnet_max_vq_pairs; 2935 2936 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 2937 device_printf(dev, 2938 "cannot set active queue pairs to %d\n", npairs); 2939 npairs = 1; 2940 } 2941 2942 sc->vtnet_act_vq_pairs = npairs; 2943 } 2944 2945 static int 2946 vtnet_reinit(struct vtnet_softc *sc) 2947 { 2948 struct ifnet *ifp; 2949 int error; 2950 2951 ifp = sc->vtnet_ifp; 2952 2953 /* Use the current MAC address. */ 2954 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 2955 vtnet_set_hwaddr(sc); 2956 2957 vtnet_set_active_vq_pairs(sc); 2958 2959 ifp->if_hwassist = 0; 2960 if (ifp->if_capenable & IFCAP_TXCSUM) 2961 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 2962 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) 2963 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6; 2964 if (ifp->if_capenable & IFCAP_TSO4) 2965 ifp->if_hwassist |= CSUM_TSO; 2966 if (ifp->if_capenable & IFCAP_TSO6) 2967 ifp->if_hwassist |= CSUM_TSO; /* No CSUM_TSO_IPV6. */ 2968 2969 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 2970 vtnet_init_rx_filters(sc); 2971 2972 error = vtnet_init_rxtx_queues(sc); 2973 if (error) 2974 return (error); 2975 2976 vtnet_enable_interrupts(sc); 2977 ifp->if_drv_flags |= IFF_DRV_RUNNING; 2978 2979 return (0); 2980 } 2981 2982 static void 2983 vtnet_init_locked(struct vtnet_softc *sc) 2984 { 2985 device_t dev; 2986 struct ifnet *ifp; 2987 2988 dev = sc->vtnet_dev; 2989 ifp = sc->vtnet_ifp; 2990 2991 VTNET_CORE_LOCK_ASSERT(sc); 2992 2993 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 2994 return; 2995 2996 vtnet_stop(sc); 2997 2998 /* Reinitialize with the host. */ 2999 if (vtnet_virtio_reinit(sc) != 0) 3000 goto fail; 3001 3002 if (vtnet_reinit(sc) != 0) 3003 goto fail; 3004 3005 virtio_reinit_complete(dev); 3006 3007 vtnet_update_link_status(sc); 3008 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3009 3010 return; 3011 3012 fail: 3013 vtnet_stop(sc); 3014 } 3015 3016 static void 3017 vtnet_init(void *xsc) 3018 { 3019 struct vtnet_softc *sc; 3020 3021 sc = xsc; 3022 3023 VTNET_CORE_LOCK(sc); 3024 vtnet_init_locked(sc); 3025 VTNET_CORE_UNLOCK(sc); 3026 } 3027 3028 static void 3029 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3030 { 3031 struct virtqueue *vq; 3032 3033 vq = sc->vtnet_ctrl_vq; 3034 3035 /* 3036 * The control virtqueue is only polled and therefore it should 3037 * already be empty. 3038 */ 3039 KASSERT(virtqueue_empty(vq), 3040 ("%s: ctrl vq %p not empty", __func__, vq)); 3041 } 3042 3043 static void 3044 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3045 struct sglist *sg, int readable, int writable) 3046 { 3047 struct virtqueue *vq; 3048 3049 vq = sc->vtnet_ctrl_vq; 3050 3051 VTNET_CORE_LOCK_ASSERT(sc); 3052 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ, 3053 ("%s: CTRL_VQ feature not negotiated", __func__)); 3054 3055 if (!virtqueue_empty(vq)) 3056 return; 3057 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0) 3058 return; 3059 3060 /* 3061 * Poll for the response, but the command is likely already 3062 * done when we return from the notify. 3063 */ 3064 virtqueue_notify(vq); 3065 virtqueue_poll(vq, NULL); 3066 } 3067 3068 static int 3069 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3070 { 3071 struct virtio_net_ctrl_hdr hdr __aligned(2); 3072 struct sglist_seg segs[3]; 3073 struct sglist sg; 3074 uint8_t ack; 3075 int error; 3076 3077 hdr.class = VIRTIO_NET_CTRL_MAC; 3078 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3079 ack = VIRTIO_NET_ERR; 3080 3081 sglist_init(&sg, 3, segs); 3082 error = 0; 3083 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3084 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN); 3085 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3086 KASSERT(error == 0 && sg.sg_nseg == 3, 3087 ("%s: error %d adding set MAC msg to sglist", __func__, error)); 3088 3089 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3090 3091 return (ack == VIRTIO_NET_OK ? 0 : EIO); 3092 } 3093 3094 static int 3095 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3096 { 3097 struct sglist_seg segs[3]; 3098 struct sglist sg; 3099 struct { 3100 struct virtio_net_ctrl_hdr hdr; 3101 uint8_t pad1; 3102 struct virtio_net_ctrl_mq mq; 3103 uint8_t pad2; 3104 uint8_t ack; 3105 } s __aligned(2); 3106 int error; 3107 3108 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3109 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3110 s.mq.virtqueue_pairs = npairs; 3111 s.ack = VIRTIO_NET_ERR; 3112 3113 sglist_init(&sg, 3, segs); 3114 error = 0; 3115 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3116 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3117 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3118 KASSERT(error == 0 && sg.sg_nseg == 3, 3119 ("%s: error %d adding MQ message to sglist", __func__, error)); 3120 3121 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3122 3123 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3124 } 3125 3126 static int 3127 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on) 3128 { 3129 struct sglist_seg segs[3]; 3130 struct sglist sg; 3131 struct { 3132 struct virtio_net_ctrl_hdr hdr; 3133 uint8_t pad1; 3134 uint8_t onoff; 3135 uint8_t pad2; 3136 uint8_t ack; 3137 } s __aligned(2); 3138 int error; 3139 3140 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3141 ("%s: CTRL_RX feature not negotiated", __func__)); 3142 3143 s.hdr.class = VIRTIO_NET_CTRL_RX; 3144 s.hdr.cmd = cmd; 3145 s.onoff = !!on; 3146 s.ack = VIRTIO_NET_ERR; 3147 3148 sglist_init(&sg, 3, segs); 3149 error = 0; 3150 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3151 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3152 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3153 KASSERT(error == 0 && sg.sg_nseg == 3, 3154 ("%s: error %d adding Rx message to sglist", __func__, error)); 3155 3156 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3157 3158 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3159 } 3160 3161 static int 3162 vtnet_set_promisc(struct vtnet_softc *sc, int on) 3163 { 3164 3165 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3166 } 3167 3168 static int 3169 vtnet_set_allmulti(struct vtnet_softc *sc, int on) 3170 { 3171 3172 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3173 } 3174 3175 /* 3176 * The device defaults to promiscuous mode for backwards compatibility. 3177 * Turn it off at attach time if possible. 3178 */ 3179 static void 3180 vtnet_attach_disable_promisc(struct vtnet_softc *sc) 3181 { 3182 struct ifnet *ifp; 3183 3184 ifp = sc->vtnet_ifp; 3185 3186 VTNET_CORE_LOCK(sc); 3187 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) { 3188 ifp->if_flags |= IFF_PROMISC; 3189 } else if (vtnet_set_promisc(sc, 0) != 0) { 3190 ifp->if_flags |= IFF_PROMISC; 3191 device_printf(sc->vtnet_dev, 3192 "cannot disable default promiscuous mode\n"); 3193 } 3194 VTNET_CORE_UNLOCK(sc); 3195 } 3196 3197 static void 3198 vtnet_rx_filter(struct vtnet_softc *sc) 3199 { 3200 device_t dev; 3201 struct ifnet *ifp; 3202 3203 dev = sc->vtnet_dev; 3204 ifp = sc->vtnet_ifp; 3205 3206 VTNET_CORE_LOCK_ASSERT(sc); 3207 3208 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) 3209 device_printf(dev, "cannot %s promiscuous mode\n", 3210 ifp->if_flags & IFF_PROMISC ? "enable" : "disable"); 3211 3212 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) 3213 device_printf(dev, "cannot %s all-multicast mode\n", 3214 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable"); 3215 } 3216 3217 static void 3218 vtnet_rx_filter_mac(struct vtnet_softc *sc) 3219 { 3220 struct virtio_net_ctrl_hdr hdr __aligned(2); 3221 struct vtnet_mac_filter *filter; 3222 struct sglist_seg segs[4]; 3223 struct sglist sg; 3224 struct ifnet *ifp; 3225 struct ifaddr *ifa; 3226 struct ifmultiaddr *ifma; 3227 int ucnt, mcnt, promisc, allmulti, error; 3228 uint8_t ack; 3229 3230 ifp = sc->vtnet_ifp; 3231 filter = sc->vtnet_mac_filter; 3232 ucnt = 0; 3233 mcnt = 0; 3234 promisc = 0; 3235 allmulti = 0; 3236 3237 VTNET_CORE_LOCK_ASSERT(sc); 3238 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3239 ("%s: CTRL_RX feature not negotiated", __func__)); 3240 3241 /* Unicast MAC addresses: */ 3242 if_addr_rlock(ifp); 3243 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 3244 if (ifa->ifa_addr->sa_family != AF_LINK) 3245 continue; 3246 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 3247 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3248 continue; 3249 else if (ucnt == VTNET_MAX_MAC_ENTRIES) { 3250 promisc = 1; 3251 break; 3252 } 3253 3254 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 3255 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN); 3256 ucnt++; 3257 } 3258 if_addr_runlock(ifp); 3259 3260 if (promisc != 0) { 3261 filter->vmf_unicast.nentries = 0; 3262 if_printf(ifp, "more than %d MAC addresses assigned, " 3263 "falling back to promiscuous mode\n", 3264 VTNET_MAX_MAC_ENTRIES); 3265 } else 3266 filter->vmf_unicast.nentries = ucnt; 3267 3268 /* Multicast MAC addresses: */ 3269 if_maddr_rlock(ifp); 3270 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3271 if (ifma->ifma_addr->sa_family != AF_LINK) 3272 continue; 3273 else if (mcnt == VTNET_MAX_MAC_ENTRIES) { 3274 allmulti = 1; 3275 break; 3276 } 3277 3278 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 3279 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN); 3280 mcnt++; 3281 } 3282 if_maddr_runlock(ifp); 3283 3284 if (allmulti != 0) { 3285 filter->vmf_multicast.nentries = 0; 3286 if_printf(ifp, "more than %d multicast MAC addresses " 3287 "assigned, falling back to all-multicast mode\n", 3288 VTNET_MAX_MAC_ENTRIES); 3289 } else 3290 filter->vmf_multicast.nentries = mcnt; 3291 3292 if (promisc != 0 && allmulti != 0) 3293 goto out; 3294 3295 hdr.class = VIRTIO_NET_CTRL_MAC; 3296 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3297 ack = VIRTIO_NET_ERR; 3298 3299 sglist_init(&sg, 4, segs); 3300 error = 0; 3301 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3302 error |= sglist_append(&sg, &filter->vmf_unicast, 3303 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN); 3304 error |= sglist_append(&sg, &filter->vmf_multicast, 3305 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN); 3306 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3307 KASSERT(error == 0 && sg.sg_nseg == 4, 3308 ("%s: error %d adding MAC filter msg to sglist", __func__, error)); 3309 3310 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3311 3312 if (ack != VIRTIO_NET_OK) 3313 if_printf(ifp, "error setting host MAC filter table\n"); 3314 3315 out: 3316 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0) 3317 if_printf(ifp, "cannot enable promiscuous mode\n"); 3318 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0) 3319 if_printf(ifp, "cannot enable all-multicast mode\n"); 3320 } 3321 3322 static int 3323 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3324 { 3325 struct sglist_seg segs[3]; 3326 struct sglist sg; 3327 struct { 3328 struct virtio_net_ctrl_hdr hdr; 3329 uint8_t pad1; 3330 uint16_t tag; 3331 uint8_t pad2; 3332 uint8_t ack; 3333 } s __aligned(2); 3334 int error; 3335 3336 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3337 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3338 s.tag = tag; 3339 s.ack = VIRTIO_NET_ERR; 3340 3341 sglist_init(&sg, 3, segs); 3342 error = 0; 3343 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3344 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3345 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3346 KASSERT(error == 0 && sg.sg_nseg == 3, 3347 ("%s: error %d adding VLAN message to sglist", __func__, error)); 3348 3349 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3350 3351 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3352 } 3353 3354 static void 3355 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3356 { 3357 uint32_t w; 3358 uint16_t tag; 3359 int i, bit; 3360 3361 VTNET_CORE_LOCK_ASSERT(sc); 3362 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER, 3363 ("%s: VLAN_FILTER feature not negotiated", __func__)); 3364 3365 /* Enable the filter for each configured VLAN. */ 3366 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3367 w = sc->vtnet_vlan_filter[i]; 3368 3369 while ((bit = ffs(w) - 1) != -1) { 3370 w &= ~(1 << bit); 3371 tag = sizeof(w) * CHAR_BIT * i + bit; 3372 3373 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3374 device_printf(sc->vtnet_dev, 3375 "cannot enable VLAN %d filter\n", tag); 3376 } 3377 } 3378 } 3379 } 3380 3381 static void 3382 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3383 { 3384 struct ifnet *ifp; 3385 int idx, bit; 3386 3387 ifp = sc->vtnet_ifp; 3388 idx = (tag >> 5) & 0x7F; 3389 bit = tag & 0x1F; 3390 3391 if (tag == 0 || tag > 4095) 3392 return; 3393 3394 VTNET_CORE_LOCK(sc); 3395 3396 if (add) 3397 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3398 else 3399 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3400 3401 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && 3402 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3403 device_printf(sc->vtnet_dev, 3404 "cannot %s VLAN %d %s the host filter table\n", 3405 add ? "add" : "remove", tag, add ? "to" : "from"); 3406 } 3407 3408 VTNET_CORE_UNLOCK(sc); 3409 } 3410 3411 static void 3412 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3413 { 3414 3415 if (ifp->if_softc != arg) 3416 return; 3417 3418 vtnet_update_vlan_filter(arg, 1, tag); 3419 } 3420 3421 static void 3422 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3423 { 3424 3425 if (ifp->if_softc != arg) 3426 return; 3427 3428 vtnet_update_vlan_filter(arg, 0, tag); 3429 } 3430 3431 static int 3432 vtnet_is_link_up(struct vtnet_softc *sc) 3433 { 3434 device_t dev; 3435 struct ifnet *ifp; 3436 uint16_t status; 3437 3438 dev = sc->vtnet_dev; 3439 ifp = sc->vtnet_ifp; 3440 3441 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0) 3442 status = VIRTIO_NET_S_LINK_UP; 3443 else 3444 status = virtio_read_dev_config_2(dev, 3445 offsetof(struct virtio_net_config, status)); 3446 3447 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3448 } 3449 3450 static void 3451 vtnet_update_link_status(struct vtnet_softc *sc) 3452 { 3453 struct ifnet *ifp; 3454 int link; 3455 3456 ifp = sc->vtnet_ifp; 3457 3458 VTNET_CORE_LOCK_ASSERT(sc); 3459 link = vtnet_is_link_up(sc); 3460 3461 /* Notify if the link status has changed. */ 3462 if (link != 0 && sc->vtnet_link_active == 0) { 3463 sc->vtnet_link_active = 1; 3464 if_link_state_change(ifp, LINK_STATE_UP); 3465 } else if (link == 0 && sc->vtnet_link_active != 0) { 3466 sc->vtnet_link_active = 0; 3467 if_link_state_change(ifp, LINK_STATE_DOWN); 3468 } 3469 } 3470 3471 static int 3472 vtnet_ifmedia_upd(struct ifnet *ifp) 3473 { 3474 struct vtnet_softc *sc; 3475 struct ifmedia *ifm; 3476 3477 sc = ifp->if_softc; 3478 ifm = &sc->vtnet_media; 3479 3480 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) 3481 return (EINVAL); 3482 3483 return (0); 3484 } 3485 3486 static void 3487 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 3488 { 3489 struct vtnet_softc *sc; 3490 3491 sc = ifp->if_softc; 3492 3493 ifmr->ifm_status = IFM_AVALID; 3494 ifmr->ifm_active = IFM_ETHER; 3495 3496 VTNET_CORE_LOCK(sc); 3497 if (vtnet_is_link_up(sc) != 0) { 3498 ifmr->ifm_status |= IFM_ACTIVE; 3499 ifmr->ifm_active |= VTNET_MEDIATYPE; 3500 } else 3501 ifmr->ifm_active |= IFM_NONE; 3502 VTNET_CORE_UNLOCK(sc); 3503 } 3504 3505 static void 3506 vtnet_set_hwaddr(struct vtnet_softc *sc) 3507 { 3508 device_t dev; 3509 int i; 3510 3511 dev = sc->vtnet_dev; 3512 3513 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 3514 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0) 3515 device_printf(dev, "unable to set MAC address\n"); 3516 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) { 3517 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3518 virtio_write_dev_config_1(dev, 3519 offsetof(struct virtio_net_config, mac) + i, 3520 sc->vtnet_hwaddr[i]); 3521 } 3522 } 3523 } 3524 3525 static void 3526 vtnet_get_hwaddr(struct vtnet_softc *sc) 3527 { 3528 device_t dev; 3529 int i; 3530 3531 dev = sc->vtnet_dev; 3532 3533 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { 3534 /* 3535 * Generate a random locally administered unicast address. 3536 * 3537 * It would be nice to generate the same MAC address across 3538 * reboots, but it seems all the hosts currently available 3539 * support the MAC feature, so this isn't too important. 3540 */ 3541 sc->vtnet_hwaddr[0] = 0xB2; 3542 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 3543 vtnet_set_hwaddr(sc); 3544 return; 3545 } 3546 3547 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3548 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev, 3549 offsetof(struct virtio_net_config, mac) + i); 3550 } 3551 } 3552 3553 static void 3554 vtnet_vlan_tag_remove(struct mbuf *m) 3555 { 3556 struct ether_vlan_header *evh; 3557 3558 evh = mtod(m, struct ether_vlan_header *); 3559 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 3560 m->m_flags |= M_VLANTAG; 3561 3562 /* Strip the 802.1Q header. */ 3563 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 3564 ETHER_HDR_LEN - ETHER_TYPE_LEN); 3565 m_adj(m, ETHER_VLAN_ENCAP_LEN); 3566 } 3567 3568 static void 3569 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 3570 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 3571 { 3572 struct sysctl_oid *node; 3573 struct sysctl_oid_list *list; 3574 struct vtnet_rxq_stats *stats; 3575 char namebuf[16]; 3576 3577 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 3578 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3579 CTLFLAG_RD, NULL, "Receive Queue"); 3580 list = SYSCTL_CHILDREN(node); 3581 3582 stats = &rxq->vtnrx_stats; 3583 3584 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, 3585 &stats->vrxs_ipackets, "Receive packets"); 3586 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, 3587 &stats->vrxs_ibytes, "Receive bytes"); 3588 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, 3589 &stats->vrxs_iqdrops, "Receive drops"); 3590 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, 3591 &stats->vrxs_ierrors, "Receive errors"); 3592 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3593 &stats->vrxs_csum, "Receive checksum offloaded"); 3594 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, 3595 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 3596 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3597 &stats->vrxs_rescheduled, 3598 "Receive interrupt handler rescheduled"); 3599 } 3600 3601 static void 3602 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 3603 struct sysctl_oid_list *child, struct vtnet_txq *txq) 3604 { 3605 struct sysctl_oid *node; 3606 struct sysctl_oid_list *list; 3607 struct vtnet_txq_stats *stats; 3608 char namebuf[16]; 3609 3610 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 3611 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3612 CTLFLAG_RD, NULL, "Transmit Queue"); 3613 list = SYSCTL_CHILDREN(node); 3614 3615 stats = &txq->vtntx_stats; 3616 3617 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, 3618 &stats->vtxs_opackets, "Transmit packets"); 3619 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, 3620 &stats->vtxs_obytes, "Transmit bytes"); 3621 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, 3622 &stats->vtxs_omcasts, "Transmit multicasts"); 3623 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3624 &stats->vtxs_csum, "Transmit checksum offloaded"); 3625 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, 3626 &stats->vtxs_tso, "Transmit segmentation offloaded"); 3627 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3628 &stats->vtxs_rescheduled, 3629 "Transmit interrupt handler rescheduled"); 3630 } 3631 3632 static void 3633 vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 3634 { 3635 device_t dev; 3636 struct sysctl_ctx_list *ctx; 3637 struct sysctl_oid *tree; 3638 struct sysctl_oid_list *child; 3639 int i; 3640 3641 dev = sc->vtnet_dev; 3642 ctx = device_get_sysctl_ctx(dev); 3643 tree = device_get_sysctl_tree(dev); 3644 child = SYSCTL_CHILDREN(tree); 3645 3646 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3647 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 3648 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 3649 } 3650 } 3651 3652 static void 3653 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 3654 struct sysctl_oid_list *child, struct vtnet_softc *sc) 3655 { 3656 struct vtnet_statistics *stats; 3657 3658 stats = &sc->vtnet_stats; 3659 3660 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 3661 CTLFLAG_RD, &stats->mbuf_alloc_failed, 3662 "Mbuf cluster allocation failures"); 3663 3664 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 3665 CTLFLAG_RD, &stats->rx_frame_too_large, 3666 "Received frame larger than the mbuf chain"); 3667 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 3668 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 3669 "Enqueuing the replacement receive mbuf failed"); 3670 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 3671 CTLFLAG_RD, &stats->rx_mergeable_failed, 3672 "Mergeable buffers receive failures"); 3673 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 3674 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 3675 "Received checksum offloaded buffer with unsupported " 3676 "Ethernet type"); 3677 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 3678 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 3679 "Received checksum offloaded buffer with incorrect IP protocol"); 3680 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 3681 CTLFLAG_RD, &stats->rx_csum_bad_offset, 3682 "Received checksum offloaded buffer with incorrect offset"); 3683 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", 3684 CTLFLAG_RD, &stats->rx_csum_bad_proto, 3685 "Received checksum offloaded buffer with incorrect protocol"); 3686 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 3687 CTLFLAG_RD, &stats->rx_csum_failed, 3688 "Received buffer checksum offload failed"); 3689 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 3690 CTLFLAG_RD, &stats->rx_csum_offloaded, 3691 "Received buffer checksum offload succeeded"); 3692 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 3693 CTLFLAG_RD, &stats->rx_task_rescheduled, 3694 "Times the receive interrupt task rescheduled itself"); 3695 3696 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype", 3697 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 3698 "Aborted transmit of checksum offloaded buffer with unknown " 3699 "Ethernet type"); 3700 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype", 3701 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 3702 "Aborted transmit of TSO buffer with unknown Ethernet type"); 3703 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 3704 CTLFLAG_RD, &stats->tx_tso_not_tcp, 3705 "Aborted transmit of TSO buffer with non TCP protocol"); 3706 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 3707 CTLFLAG_RD, &stats->tx_defragged, 3708 "Transmit mbufs defragged"); 3709 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 3710 CTLFLAG_RD, &stats->tx_defrag_failed, 3711 "Aborted transmit of buffer because defrag failed"); 3712 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 3713 CTLFLAG_RD, &stats->tx_csum_offloaded, 3714 "Offloaded checksum of transmitted buffer"); 3715 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 3716 CTLFLAG_RD, &stats->tx_tso_offloaded, 3717 "Segmentation offload of transmitted buffer"); 3718 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 3719 CTLFLAG_RD, &stats->tx_task_rescheduled, 3720 "Times the transmit interrupt task rescheduled itself"); 3721 } 3722 3723 static void 3724 vtnet_setup_sysctl(struct vtnet_softc *sc) 3725 { 3726 device_t dev; 3727 struct sysctl_ctx_list *ctx; 3728 struct sysctl_oid *tree; 3729 struct sysctl_oid_list *child; 3730 3731 dev = sc->vtnet_dev; 3732 ctx = device_get_sysctl_ctx(dev); 3733 tree = device_get_sysctl_tree(dev); 3734 child = SYSCTL_CHILDREN(tree); 3735 3736 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 3737 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 3738 "Maximum number of supported virtqueue pairs"); 3739 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 3740 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 3741 "Number of active virtqueue pairs"); 3742 3743 vtnet_setup_stat_sysctl(ctx, child, sc); 3744 } 3745 3746 static int 3747 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 3748 { 3749 3750 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 3751 } 3752 3753 static void 3754 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 3755 { 3756 3757 virtqueue_disable_intr(rxq->vtnrx_vq); 3758 } 3759 3760 static int 3761 vtnet_txq_enable_intr(struct vtnet_txq *txq) 3762 { 3763 3764 return (virtqueue_postpone_intr(txq->vtntx_vq, VQ_POSTPONE_LONG)); 3765 } 3766 3767 static void 3768 vtnet_txq_disable_intr(struct vtnet_txq *txq) 3769 { 3770 3771 virtqueue_disable_intr(txq->vtntx_vq); 3772 } 3773 3774 static void 3775 vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 3776 { 3777 int i; 3778 3779 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3780 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]); 3781 } 3782 3783 static void 3784 vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 3785 { 3786 int i; 3787 3788 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3789 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 3790 } 3791 3792 static void 3793 vtnet_enable_interrupts(struct vtnet_softc *sc) 3794 { 3795 3796 vtnet_enable_rx_interrupts(sc); 3797 vtnet_enable_tx_interrupts(sc); 3798 } 3799 3800 static void 3801 vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 3802 { 3803 int i; 3804 3805 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3806 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 3807 } 3808 3809 static void 3810 vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 3811 { 3812 int i; 3813 3814 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3815 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 3816 } 3817 3818 static void 3819 vtnet_disable_interrupts(struct vtnet_softc *sc) 3820 { 3821 3822 vtnet_disable_rx_interrupts(sc); 3823 vtnet_disable_tx_interrupts(sc); 3824 } 3825 3826 static int 3827 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 3828 { 3829 char path[64]; 3830 3831 snprintf(path, sizeof(path), 3832 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 3833 TUNABLE_INT_FETCH(path, &def); 3834 3835 return (def); 3836 } 3837