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 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 556 mask |= VIRTIO_NET_F_MQ; 557 #ifdef VTNET_LEGACY_TX 558 mask |= VIRTIO_NET_F_MQ; 559 #endif 560 561 features = VTNET_FEATURES & ~mask; 562 sc->vtnet_features = virtio_negotiate_features(dev, features); 563 564 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) == 0) 565 return; 566 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) 567 return; 568 569 /* 570 * LRO without mergeable buffers requires special care. This is not 571 * ideal because every receive buffer must be large enough to hold 572 * the maximum TCP packet, the Ethernet header, and the header. This 573 * requires up to 34 descriptors with MCLBYTES clusters. If we do 574 * not have indirect descriptors, LRO is disabled since the virtqueue 575 * will not contain very many receive buffers. 576 */ 577 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC) == 0) { 578 device_printf(dev, 579 "LRO disabled due to both mergeable buffers and indirect " 580 "descriptors not negotiated\n"); 581 582 features &= ~VTNET_LRO_FEATURES; 583 sc->vtnet_features = virtio_negotiate_features(dev, features); 584 } else 585 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 586 } 587 588 static void 589 vtnet_setup_features(struct vtnet_softc *sc) 590 { 591 device_t dev; 592 int max_pairs, max; 593 594 dev = sc->vtnet_dev; 595 596 vtnet_negotiate_features(sc); 597 598 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 599 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 600 601 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 602 /* This feature should always be negotiated. */ 603 sc->vtnet_flags |= VTNET_FLAG_MAC; 604 } 605 606 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 607 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 608 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 609 } else 610 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 611 612 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 613 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS; 614 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 615 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS; 616 else 617 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS; 618 619 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 620 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 621 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 622 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS; 623 else 624 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS; 625 626 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 627 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 628 629 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 630 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 631 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 632 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 633 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 634 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 635 } 636 637 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) && 638 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 639 max_pairs = virtio_read_dev_config_2(dev, 640 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 641 if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 642 max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) 643 max_pairs = 1; 644 } else 645 max_pairs = 1; 646 647 if (max_pairs > 1) { 648 /* 649 * Limit the maximum number of queue pairs to the number of 650 * CPUs or the configured maximum. The actual number of 651 * queues that get used may be less. 652 */ 653 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 654 if (max > 0 && max_pairs > max) 655 max_pairs = max; 656 if (max_pairs > mp_ncpus) 657 max_pairs = mp_ncpus; 658 if (max_pairs > VTNET_MAX_QUEUE_PAIRS) 659 max_pairs = VTNET_MAX_QUEUE_PAIRS; 660 if (max_pairs > 1) 661 sc->vtnet_flags |= VTNET_FLAG_MULTIQ; 662 } 663 664 sc->vtnet_max_vq_pairs = max_pairs; 665 } 666 667 static int 668 vtnet_init_rxq(struct vtnet_softc *sc, int id) 669 { 670 struct vtnet_rxq *rxq; 671 672 rxq = &sc->vtnet_rxqs[id]; 673 674 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 675 device_get_nameunit(sc->vtnet_dev), id); 676 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 677 678 rxq->vtnrx_sc = sc; 679 rxq->vtnrx_id = id; 680 681 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 682 if (rxq->vtnrx_sg == NULL) 683 return (ENOMEM); 684 685 TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 686 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 687 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 688 689 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 690 } 691 692 static int 693 vtnet_init_txq(struct vtnet_softc *sc, int id) 694 { 695 struct vtnet_txq *txq; 696 697 txq = &sc->vtnet_txqs[id]; 698 699 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 700 device_get_nameunit(sc->vtnet_dev), id); 701 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 702 703 txq->vtntx_sc = sc; 704 txq->vtntx_id = id; 705 706 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 707 if (txq->vtntx_sg == NULL) 708 return (ENOMEM); 709 710 #ifndef VTNET_LEGACY_TX 711 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 712 M_NOWAIT, &txq->vtntx_mtx); 713 if (txq->vtntx_br == NULL) 714 return (ENOMEM); 715 716 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 717 #endif 718 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 719 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 720 taskqueue_thread_enqueue, &txq->vtntx_tq); 721 if (txq->vtntx_tq == NULL) 722 return (ENOMEM); 723 724 return (0); 725 } 726 727 static int 728 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 729 { 730 int i, npairs, error; 731 732 npairs = sc->vtnet_max_vq_pairs; 733 734 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 735 M_NOWAIT | M_ZERO); 736 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 737 M_NOWAIT | M_ZERO); 738 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 739 return (ENOMEM); 740 741 for (i = 0; i < npairs; i++) { 742 error = vtnet_init_rxq(sc, i); 743 if (error) 744 return (error); 745 error = vtnet_init_txq(sc, i); 746 if (error) 747 return (error); 748 } 749 750 vtnet_setup_queue_sysctl(sc); 751 752 return (0); 753 } 754 755 static void 756 vtnet_destroy_rxq(struct vtnet_rxq *rxq) 757 { 758 759 rxq->vtnrx_sc = NULL; 760 rxq->vtnrx_id = -1; 761 762 if (rxq->vtnrx_sg != NULL) { 763 sglist_free(rxq->vtnrx_sg); 764 rxq->vtnrx_sg = NULL; 765 } 766 767 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 768 mtx_destroy(&rxq->vtnrx_mtx); 769 } 770 771 static void 772 vtnet_destroy_txq(struct vtnet_txq *txq) 773 { 774 775 txq->vtntx_sc = NULL; 776 txq->vtntx_id = -1; 777 778 if (txq->vtntx_sg != NULL) { 779 sglist_free(txq->vtntx_sg); 780 txq->vtntx_sg = NULL; 781 } 782 783 #ifndef VTNET_LEGACY_TX 784 if (txq->vtntx_br != NULL) { 785 buf_ring_free(txq->vtntx_br, M_DEVBUF); 786 txq->vtntx_br = NULL; 787 } 788 #endif 789 790 if (mtx_initialized(&txq->vtntx_mtx) != 0) 791 mtx_destroy(&txq->vtntx_mtx); 792 } 793 794 static void 795 vtnet_free_rxtx_queues(struct vtnet_softc *sc) 796 { 797 int i; 798 799 if (sc->vtnet_rxqs != NULL) { 800 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 801 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 802 free(sc->vtnet_rxqs, M_DEVBUF); 803 sc->vtnet_rxqs = NULL; 804 } 805 806 if (sc->vtnet_txqs != NULL) { 807 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 808 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 809 free(sc->vtnet_txqs, M_DEVBUF); 810 sc->vtnet_txqs = NULL; 811 } 812 } 813 814 static int 815 vtnet_alloc_rx_filters(struct vtnet_softc *sc) 816 { 817 818 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 819 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 820 M_DEVBUF, M_NOWAIT | M_ZERO); 821 if (sc->vtnet_mac_filter == NULL) 822 return (ENOMEM); 823 } 824 825 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 826 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 827 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 828 if (sc->vtnet_vlan_filter == NULL) 829 return (ENOMEM); 830 } 831 832 return (0); 833 } 834 835 static void 836 vtnet_free_rx_filters(struct vtnet_softc *sc) 837 { 838 839 if (sc->vtnet_mac_filter != NULL) { 840 free(sc->vtnet_mac_filter, M_DEVBUF); 841 sc->vtnet_mac_filter = NULL; 842 } 843 844 if (sc->vtnet_vlan_filter != NULL) { 845 free(sc->vtnet_vlan_filter, M_DEVBUF); 846 sc->vtnet_vlan_filter = NULL; 847 } 848 } 849 850 static int 851 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 852 { 853 device_t dev; 854 struct vq_alloc_info *info; 855 struct vtnet_rxq *rxq; 856 struct vtnet_txq *txq; 857 int i, idx, flags, nvqs, error; 858 859 dev = sc->vtnet_dev; 860 flags = 0; 861 862 nvqs = sc->vtnet_max_vq_pairs * 2; 863 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 864 nvqs++; 865 866 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); 867 if (info == NULL) 868 return (ENOMEM); 869 870 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) { 871 rxq = &sc->vtnet_rxqs[i]; 872 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, 873 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, 874 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id); 875 876 txq = &sc->vtnet_txqs[i]; 877 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs, 878 vtnet_tx_vq_intr, txq, &txq->vtntx_vq, 879 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id); 880 } 881 882 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 883 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, 884 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); 885 } 886 887 /* 888 * Enable interrupt binding if this is multiqueue. This only matters 889 * when per-vq MSIX is available. 890 */ 891 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ) 892 flags |= 0; 893 894 error = virtio_alloc_virtqueues(dev, flags, nvqs, info); 895 free(info, M_TEMP); 896 897 return (error); 898 } 899 900 static int 901 vtnet_setup_interface(struct vtnet_softc *sc) 902 { 903 device_t dev; 904 struct ifnet *ifp; 905 int limit; 906 907 dev = sc->vtnet_dev; 908 909 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER); 910 if (ifp == NULL) { 911 device_printf(dev, "cannot allocate ifnet structure\n"); 912 return (ENOSPC); 913 } 914 915 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 916 if_initbaudrate(ifp, IF_Gbps(10)); /* Approx. */ 917 ifp->if_softc = sc; 918 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 919 ifp->if_init = vtnet_init; 920 ifp->if_ioctl = vtnet_ioctl; 921 922 #ifndef VTNET_LEGACY_TX 923 ifp->if_transmit = vtnet_txq_mq_start; 924 ifp->if_qflush = vtnet_qflush; 925 #else 926 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 927 ifp->if_start = vtnet_start; 928 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1); 929 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1; 930 IFQ_SET_READY(&ifp->if_snd); 931 #endif 932 933 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd, 934 vtnet_ifmedia_sts); 935 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL); 936 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE); 937 938 /* Read (or generate) the MAC address for the adapter. */ 939 vtnet_get_hwaddr(sc); 940 941 ether_ifattach(ifp, sc->vtnet_hwaddr); 942 943 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 944 ifp->if_capabilities |= IFCAP_LINKSTATE; 945 946 /* Tell the upper layer(s) we support long frames. */ 947 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); 948 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU; 949 950 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 951 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6; 952 953 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) { 954 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6; 955 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 956 } else { 957 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 958 ifp->if_capabilities |= IFCAP_TSO4; 959 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 960 ifp->if_capabilities |= IFCAP_TSO6; 961 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 962 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 963 } 964 965 if (ifp->if_capabilities & IFCAP_TSO) 966 ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 967 } 968 969 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) 970 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6; 971 972 if (ifp->if_capabilities & IFCAP_HWCSUM) { 973 /* 974 * VirtIO does not support VLAN tagging, but we can fake 975 * it by inserting and removing the 802.1Q header during 976 * transmit and receive. We are then able to do checksum 977 * offloading of VLAN frames. 978 */ 979 ifp->if_capabilities |= 980 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; 981 } 982 983 ifp->if_capenable = ifp->if_capabilities; 984 985 /* 986 * Capabilities after here are not enabled by default. 987 */ 988 989 if (ifp->if_capabilities & IFCAP_RXCSUM) { 990 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) || 991 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6)) 992 ifp->if_capabilities |= IFCAP_LRO; 993 } 994 995 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 996 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 997 998 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 999 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1000 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1001 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1002 } 1003 1004 limit = vtnet_tunable_int(sc, "rx_process_limit", 1005 vtnet_rx_process_limit); 1006 if (limit < 0) 1007 limit = INT_MAX; 1008 sc->vtnet_rx_process_limit = limit; 1009 1010 return (0); 1011 } 1012 1013 static int 1014 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu) 1015 { 1016 struct ifnet *ifp; 1017 int frame_size, clsize; 1018 1019 ifp = sc->vtnet_ifp; 1020 1021 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU) 1022 return (EINVAL); 1023 1024 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) + 1025 new_mtu; 1026 1027 /* 1028 * Based on the new MTU (and hence frame size) determine which 1029 * cluster size is most appropriate for the receive queues. 1030 */ 1031 if (frame_size <= MCLBYTES) { 1032 clsize = MCLBYTES; 1033 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1034 /* Avoid going past 9K jumbos. */ 1035 if (frame_size > MJUM9BYTES) 1036 return (EINVAL); 1037 clsize = MJUM9BYTES; 1038 } else 1039 clsize = MJUMPAGESIZE; 1040 1041 ifp->if_mtu = new_mtu; 1042 sc->vtnet_rx_new_clsize = clsize; 1043 1044 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1045 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1046 vtnet_init_locked(sc); 1047 } 1048 1049 return (0); 1050 } 1051 1052 static int 1053 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1054 { 1055 struct vtnet_softc *sc; 1056 struct ifreq *ifr; 1057 int reinit, mask, error; 1058 1059 sc = ifp->if_softc; 1060 ifr = (struct ifreq *) data; 1061 error = 0; 1062 1063 switch (cmd) { 1064 case SIOCSIFMTU: 1065 if (ifp->if_mtu != ifr->ifr_mtu) { 1066 VTNET_CORE_LOCK(sc); 1067 error = vtnet_change_mtu(sc, ifr->ifr_mtu); 1068 VTNET_CORE_UNLOCK(sc); 1069 } 1070 break; 1071 1072 case SIOCSIFFLAGS: 1073 VTNET_CORE_LOCK(sc); 1074 if ((ifp->if_flags & IFF_UP) == 0) { 1075 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1076 vtnet_stop(sc); 1077 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1078 if ((ifp->if_flags ^ sc->vtnet_if_flags) & 1079 (IFF_PROMISC | IFF_ALLMULTI)) { 1080 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1081 vtnet_rx_filter(sc); 1082 else 1083 error = ENOTSUP; 1084 } 1085 } else 1086 vtnet_init_locked(sc); 1087 1088 if (error == 0) 1089 sc->vtnet_if_flags = ifp->if_flags; 1090 VTNET_CORE_UNLOCK(sc); 1091 break; 1092 1093 case SIOCADDMULTI: 1094 case SIOCDELMULTI: 1095 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) 1096 break; 1097 VTNET_CORE_LOCK(sc); 1098 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1099 vtnet_rx_filter_mac(sc); 1100 VTNET_CORE_UNLOCK(sc); 1101 break; 1102 1103 case SIOCSIFMEDIA: 1104 case SIOCGIFMEDIA: 1105 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1106 break; 1107 1108 case SIOCSIFCAP: 1109 VTNET_CORE_LOCK(sc); 1110 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1111 1112 if (mask & IFCAP_TXCSUM) 1113 ifp->if_capenable ^= IFCAP_TXCSUM; 1114 if (mask & IFCAP_TXCSUM_IPV6) 1115 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; 1116 if (mask & IFCAP_TSO4) 1117 ifp->if_capenable ^= IFCAP_TSO4; 1118 if (mask & IFCAP_TSO6) 1119 ifp->if_capenable ^= IFCAP_TSO6; 1120 1121 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO | 1122 IFCAP_VLAN_HWFILTER)) { 1123 /* These Rx features require us to renegotiate. */ 1124 reinit = 1; 1125 1126 if (mask & IFCAP_RXCSUM) 1127 ifp->if_capenable ^= IFCAP_RXCSUM; 1128 if (mask & IFCAP_RXCSUM_IPV6) 1129 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; 1130 if (mask & IFCAP_LRO) 1131 ifp->if_capenable ^= IFCAP_LRO; 1132 if (mask & IFCAP_VLAN_HWFILTER) 1133 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 1134 } else 1135 reinit = 0; 1136 1137 if (mask & IFCAP_VLAN_HWTSO) 1138 ifp->if_capenable ^= IFCAP_VLAN_HWTSO; 1139 if (mask & IFCAP_VLAN_HWTAGGING) 1140 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 1141 1142 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1143 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1144 vtnet_init_locked(sc); 1145 } 1146 1147 VTNET_CORE_UNLOCK(sc); 1148 VLAN_CAPABILITIES(ifp); 1149 1150 break; 1151 1152 default: 1153 error = ether_ioctl(ifp, cmd, data); 1154 break; 1155 } 1156 1157 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1158 1159 return (error); 1160 } 1161 1162 static int 1163 vtnet_rxq_populate(struct vtnet_rxq *rxq) 1164 { 1165 struct virtqueue *vq; 1166 int nbufs, error; 1167 1168 vq = rxq->vtnrx_vq; 1169 error = ENOSPC; 1170 1171 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1172 error = vtnet_rxq_new_buf(rxq); 1173 if (error) 1174 break; 1175 } 1176 1177 if (nbufs > 0) { 1178 virtqueue_notify(vq); 1179 /* 1180 * EMSGSIZE signifies the virtqueue did not have enough 1181 * entries available to hold the last mbuf. This is not 1182 * an error. 1183 */ 1184 if (error == EMSGSIZE) 1185 error = 0; 1186 } 1187 1188 return (error); 1189 } 1190 1191 static void 1192 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1193 { 1194 struct virtqueue *vq; 1195 struct mbuf *m; 1196 int last; 1197 1198 vq = rxq->vtnrx_vq; 1199 last = 0; 1200 1201 while ((m = virtqueue_drain(vq, &last)) != NULL) 1202 m_freem(m); 1203 1204 KASSERT(virtqueue_empty(vq), 1205 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1206 } 1207 1208 static struct mbuf * 1209 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1210 { 1211 struct mbuf *m_head, *m_tail, *m; 1212 int i, clsize; 1213 1214 clsize = sc->vtnet_rx_clsize; 1215 1216 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1217 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs)); 1218 1219 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize); 1220 if (m_head == NULL) 1221 goto fail; 1222 1223 m_head->m_len = clsize; 1224 m_tail = m_head; 1225 1226 /* Allocate the rest of the chain. */ 1227 for (i = 1; i < nbufs; i++) { 1228 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize); 1229 if (m == NULL) 1230 goto fail; 1231 1232 m->m_len = clsize; 1233 m_tail->m_next = m; 1234 m_tail = m; 1235 } 1236 1237 if (m_tailp != NULL) 1238 *m_tailp = m_tail; 1239 1240 return (m_head); 1241 1242 fail: 1243 sc->vtnet_stats.mbuf_alloc_failed++; 1244 m_freem(m_head); 1245 1246 return (NULL); 1247 } 1248 1249 /* 1250 * Slow path for when LRO without mergeable buffers is negotiated. 1251 */ 1252 static int 1253 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1254 int len0) 1255 { 1256 struct vtnet_softc *sc; 1257 struct mbuf *m, *m_prev; 1258 struct mbuf *m_new, *m_tail; 1259 int len, clsize, nreplace, error; 1260 1261 sc = rxq->vtnrx_sc; 1262 clsize = sc->vtnet_rx_clsize; 1263 1264 m_prev = NULL; 1265 m_tail = NULL; 1266 nreplace = 0; 1267 1268 m = m0; 1269 len = len0; 1270 1271 /* 1272 * Since these mbuf chains are so large, we avoid allocating an 1273 * entire replacement chain if possible. When the received frame 1274 * did not consume the entire chain, the unused mbufs are moved 1275 * to the replacement chain. 1276 */ 1277 while (len > 0) { 1278 /* 1279 * Something is seriously wrong if we received a frame 1280 * larger than the chain. Drop it. 1281 */ 1282 if (m == NULL) { 1283 sc->vtnet_stats.rx_frame_too_large++; 1284 return (EMSGSIZE); 1285 } 1286 1287 /* We always allocate the same cluster size. */ 1288 KASSERT(m->m_len == clsize, 1289 ("%s: mbuf size %d is not the cluster size %d", 1290 __func__, m->m_len, clsize)); 1291 1292 m->m_len = MIN(m->m_len, len); 1293 len -= m->m_len; 1294 1295 m_prev = m; 1296 m = m->m_next; 1297 nreplace++; 1298 } 1299 1300 KASSERT(nreplace <= sc->vtnet_rx_nmbufs, 1301 ("%s: too many replacement mbufs %d max %d", __func__, nreplace, 1302 sc->vtnet_rx_nmbufs)); 1303 1304 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1305 if (m_new == NULL) { 1306 m_prev->m_len = clsize; 1307 return (ENOBUFS); 1308 } 1309 1310 /* 1311 * Move any unused mbufs from the received chain onto the end 1312 * of the new chain. 1313 */ 1314 if (m_prev->m_next != NULL) { 1315 m_tail->m_next = m_prev->m_next; 1316 m_prev->m_next = NULL; 1317 } 1318 1319 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1320 if (error) { 1321 /* 1322 * BAD! We could not enqueue the replacement mbuf chain. We 1323 * must restore the m0 chain to the original state if it was 1324 * modified so we can subsequently discard it. 1325 * 1326 * NOTE: The replacement is suppose to be an identical copy 1327 * to the one just dequeued so this is an unexpected error. 1328 */ 1329 sc->vtnet_stats.rx_enq_replacement_failed++; 1330 1331 if (m_tail->m_next != NULL) { 1332 m_prev->m_next = m_tail->m_next; 1333 m_tail->m_next = NULL; 1334 } 1335 1336 m_prev->m_len = clsize; 1337 m_freem(m_new); 1338 } 1339 1340 return (error); 1341 } 1342 1343 static int 1344 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1345 { 1346 struct vtnet_softc *sc; 1347 struct mbuf *m_new; 1348 int error; 1349 1350 sc = rxq->vtnrx_sc; 1351 1352 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1353 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1354 1355 if (m->m_next == NULL) { 1356 /* Fast-path for the common case of just one mbuf. */ 1357 if (m->m_len < len) 1358 return (EINVAL); 1359 1360 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1361 if (m_new == NULL) 1362 return (ENOBUFS); 1363 1364 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1365 if (error) { 1366 /* 1367 * The new mbuf is suppose to be an identical 1368 * copy of the one just dequeued so this is an 1369 * unexpected error. 1370 */ 1371 m_freem(m_new); 1372 sc->vtnet_stats.rx_enq_replacement_failed++; 1373 } else 1374 m->m_len = len; 1375 } else 1376 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len); 1377 1378 return (error); 1379 } 1380 1381 static int 1382 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1383 { 1384 struct vtnet_softc *sc; 1385 struct sglist *sg; 1386 struct vtnet_rx_header *rxhdr; 1387 uint8_t *mdata; 1388 int offset, error; 1389 1390 sc = rxq->vtnrx_sc; 1391 sg = rxq->vtnrx_sg; 1392 mdata = mtod(m, uint8_t *); 1393 1394 VTNET_RXQ_LOCK_ASSERT(rxq); 1395 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1396 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1397 KASSERT(m->m_len == sc->vtnet_rx_clsize, 1398 ("%s: unexpected cluster size %d/%d", __func__, m->m_len, 1399 sc->vtnet_rx_clsize)); 1400 1401 sglist_reset(sg); 1402 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1403 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1404 rxhdr = (struct vtnet_rx_header *) mdata; 1405 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1406 offset = sizeof(struct vtnet_rx_header); 1407 } else 1408 offset = 0; 1409 1410 sglist_append(sg, mdata + offset, m->m_len - offset); 1411 if (m->m_next != NULL) { 1412 error = sglist_append_mbuf(sg, m->m_next); 1413 MPASS(error == 0); 1414 } 1415 1416 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg); 1417 1418 return (error); 1419 } 1420 1421 static int 1422 vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1423 { 1424 struct vtnet_softc *sc; 1425 struct mbuf *m; 1426 int error; 1427 1428 sc = rxq->vtnrx_sc; 1429 1430 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1431 if (m == NULL) 1432 return (ENOBUFS); 1433 1434 error = vtnet_rxq_enqueue_buf(rxq, m); 1435 if (error) 1436 m_freem(m); 1437 1438 return (error); 1439 } 1440 1441 /* 1442 * Use the checksum offset in the VirtIO header to set the 1443 * correct CSUM_* flags. 1444 */ 1445 static int 1446 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m, 1447 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1448 { 1449 struct vtnet_softc *sc; 1450 #if defined(INET) || defined(INET6) 1451 int offset = hdr->csum_start + hdr->csum_offset; 1452 #endif 1453 1454 sc = rxq->vtnrx_sc; 1455 1456 /* Only do a basic sanity check on the offset. */ 1457 switch (eth_type) { 1458 #if defined(INET) 1459 case ETHERTYPE_IP: 1460 if (__predict_false(offset < ip_start + sizeof(struct ip))) 1461 return (1); 1462 break; 1463 #endif 1464 #if defined(INET6) 1465 case ETHERTYPE_IPV6: 1466 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr))) 1467 return (1); 1468 break; 1469 #endif 1470 default: 1471 sc->vtnet_stats.rx_csum_bad_ethtype++; 1472 return (1); 1473 } 1474 1475 /* 1476 * Use the offset to determine the appropriate CSUM_* flags. This is 1477 * a bit dirty, but we can get by with it since the checksum offsets 1478 * happen to be different. We assume the host host does not do IPv4 1479 * header checksum offloading. 1480 */ 1481 switch (hdr->csum_offset) { 1482 case offsetof(struct udphdr, uh_sum): 1483 case offsetof(struct tcphdr, th_sum): 1484 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1485 m->m_pkthdr.csum_data = 0xFFFF; 1486 break; 1487 case offsetof(struct sctphdr, checksum): 1488 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 1489 break; 1490 default: 1491 sc->vtnet_stats.rx_csum_bad_offset++; 1492 return (1); 1493 } 1494 1495 return (0); 1496 } 1497 1498 static int 1499 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m, 1500 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1501 { 1502 struct vtnet_softc *sc; 1503 int offset, proto; 1504 1505 sc = rxq->vtnrx_sc; 1506 1507 switch (eth_type) { 1508 #if defined(INET) 1509 case ETHERTYPE_IP: { 1510 struct ip *ip; 1511 if (__predict_false(m->m_len < ip_start + sizeof(struct ip))) 1512 return (1); 1513 ip = (struct ip *)(m->m_data + ip_start); 1514 proto = ip->ip_p; 1515 offset = ip_start + (ip->ip_hl << 2); 1516 break; 1517 } 1518 #endif 1519 #if defined(INET6) 1520 case ETHERTYPE_IPV6: 1521 if (__predict_false(m->m_len < ip_start + 1522 sizeof(struct ip6_hdr))) 1523 return (1); 1524 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto); 1525 if (__predict_false(offset < 0)) 1526 return (1); 1527 break; 1528 #endif 1529 default: 1530 sc->vtnet_stats.rx_csum_bad_ethtype++; 1531 return (1); 1532 } 1533 1534 switch (proto) { 1535 case IPPROTO_TCP: 1536 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) 1537 return (1); 1538 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1539 m->m_pkthdr.csum_data = 0xFFFF; 1540 break; 1541 case IPPROTO_UDP: 1542 if (__predict_false(m->m_len < offset + sizeof(struct udphdr))) 1543 return (1); 1544 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1545 m->m_pkthdr.csum_data = 0xFFFF; 1546 break; 1547 case IPPROTO_SCTP: 1548 if (__predict_false(m->m_len < offset + sizeof(struct sctphdr))) 1549 return (1); 1550 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 1551 break; 1552 default: 1553 /* 1554 * For the remaining protocols, FreeBSD does not support 1555 * checksum offloading, so the checksum will be recomputed. 1556 */ 1557 #if 0 1558 if_printf(sc->vtnet_ifp, "cksum offload of unsupported " 1559 "protocol eth_type=%#x proto=%d csum_start=%d " 1560 "csum_offset=%d\n", __func__, eth_type, proto, 1561 hdr->csum_start, hdr->csum_offset); 1562 #endif 1563 break; 1564 } 1565 1566 return (0); 1567 } 1568 1569 /* 1570 * Set the appropriate CSUM_* flags. Unfortunately, the information 1571 * provided is not directly useful to us. The VirtIO header gives the 1572 * offset of the checksum, which is all Linux needs, but this is not 1573 * how FreeBSD does things. We are forced to peek inside the packet 1574 * a bit. 1575 * 1576 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD 1577 * could accept the offsets and let the stack figure it out. 1578 */ 1579 static int 1580 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1581 struct virtio_net_hdr *hdr) 1582 { 1583 struct ether_header *eh; 1584 struct ether_vlan_header *evh; 1585 uint16_t eth_type; 1586 int offset, error; 1587 1588 eh = mtod(m, struct ether_header *); 1589 eth_type = ntohs(eh->ether_type); 1590 if (eth_type == ETHERTYPE_VLAN) { 1591 /* BMV: We should handle nested VLAN tags too. */ 1592 evh = mtod(m, struct ether_vlan_header *); 1593 eth_type = ntohs(evh->evl_proto); 1594 offset = sizeof(struct ether_vlan_header); 1595 } else 1596 offset = sizeof(struct ether_header); 1597 1598 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1599 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr); 1600 else 1601 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr); 1602 1603 return (error); 1604 } 1605 1606 static void 1607 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1608 { 1609 struct mbuf *m; 1610 1611 while (--nbufs > 0) { 1612 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1613 if (m == NULL) 1614 break; 1615 vtnet_rxq_discard_buf(rxq, m); 1616 } 1617 } 1618 1619 static void 1620 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1621 { 1622 int error; 1623 1624 /* 1625 * Requeue the discarded mbuf. This should always be successful 1626 * since it was just dequeued. 1627 */ 1628 error = vtnet_rxq_enqueue_buf(rxq, m); 1629 KASSERT(error == 0, 1630 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1631 } 1632 1633 static int 1634 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1635 { 1636 struct vtnet_softc *sc; 1637 struct ifnet *ifp; 1638 struct virtqueue *vq; 1639 struct mbuf *m, *m_tail; 1640 int len; 1641 1642 sc = rxq->vtnrx_sc; 1643 vq = rxq->vtnrx_vq; 1644 ifp = sc->vtnet_ifp; 1645 m_tail = m_head; 1646 1647 while (--nbufs > 0) { 1648 m = virtqueue_dequeue(vq, &len); 1649 if (m == NULL) { 1650 rxq->vtnrx_stats.vrxs_ierrors++; 1651 goto fail; 1652 } 1653 1654 if (vtnet_rxq_new_buf(rxq) != 0) { 1655 rxq->vtnrx_stats.vrxs_iqdrops++; 1656 vtnet_rxq_discard_buf(rxq, m); 1657 if (nbufs > 1) 1658 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1659 goto fail; 1660 } 1661 1662 if (m->m_len < len) 1663 len = m->m_len; 1664 1665 m->m_len = len; 1666 m->m_flags &= ~M_PKTHDR; 1667 1668 m_head->m_pkthdr.len += len; 1669 m_tail->m_next = m; 1670 m_tail = m; 1671 } 1672 1673 return (0); 1674 1675 fail: 1676 sc->vtnet_stats.rx_mergeable_failed++; 1677 m_freem(m_head); 1678 1679 return (1); 1680 } 1681 1682 static void 1683 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 1684 struct virtio_net_hdr *hdr) 1685 { 1686 struct vtnet_softc *sc; 1687 struct ifnet *ifp; 1688 struct ether_header *eh; 1689 1690 sc = rxq->vtnrx_sc; 1691 ifp = sc->vtnet_ifp; 1692 1693 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { 1694 eh = mtod(m, struct ether_header *); 1695 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1696 vtnet_vlan_tag_remove(m); 1697 /* 1698 * With the 802.1Q header removed, update the 1699 * checksum starting location accordingly. 1700 */ 1701 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1702 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 1703 } 1704 } 1705 1706 m->m_pkthdr.flowid = rxq->vtnrx_id; 1707 m->m_flags |= M_FLOWID; 1708 1709 /* 1710 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum 1711 * distinction that Linux does. Need to reevaluate if performing 1712 * offloading for the NEEDS_CSUM case is really appropriate. 1713 */ 1714 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM | 1715 VIRTIO_NET_HDR_F_DATA_VALID)) { 1716 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 1717 rxq->vtnrx_stats.vrxs_csum++; 1718 else 1719 rxq->vtnrx_stats.vrxs_csum_failed++; 1720 } 1721 1722 rxq->vtnrx_stats.vrxs_ipackets++; 1723 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 1724 1725 VTNET_RXQ_UNLOCK(rxq); 1726 (*ifp->if_input)(ifp, m); 1727 VTNET_RXQ_LOCK(rxq); 1728 } 1729 1730 static int 1731 vtnet_rxq_eof(struct vtnet_rxq *rxq) 1732 { 1733 struct virtio_net_hdr lhdr, *hdr; 1734 struct vtnet_softc *sc; 1735 struct ifnet *ifp; 1736 struct virtqueue *vq; 1737 struct mbuf *m; 1738 struct virtio_net_hdr_mrg_rxbuf *mhdr; 1739 int len, deq, nbufs, adjsz, count; 1740 1741 sc = rxq->vtnrx_sc; 1742 vq = rxq->vtnrx_vq; 1743 ifp = sc->vtnet_ifp; 1744 hdr = &lhdr; 1745 deq = 0; 1746 count = sc->vtnet_rx_process_limit; 1747 1748 VTNET_RXQ_LOCK_ASSERT(rxq); 1749 1750 while (count-- > 0) { 1751 m = virtqueue_dequeue(vq, &len); 1752 if (m == NULL) 1753 break; 1754 deq++; 1755 1756 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 1757 rxq->vtnrx_stats.vrxs_ierrors++; 1758 vtnet_rxq_discard_buf(rxq, m); 1759 continue; 1760 } 1761 1762 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1763 nbufs = 1; 1764 adjsz = sizeof(struct vtnet_rx_header); 1765 /* 1766 * Account for our pad inserted between the header 1767 * and the actual start of the frame. 1768 */ 1769 len += VTNET_RX_HEADER_PAD; 1770 } else { 1771 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 1772 nbufs = mhdr->num_buffers; 1773 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1774 } 1775 1776 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 1777 rxq->vtnrx_stats.vrxs_iqdrops++; 1778 vtnet_rxq_discard_buf(rxq, m); 1779 if (nbufs > 1) 1780 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1781 continue; 1782 } 1783 1784 m->m_pkthdr.len = len; 1785 m->m_pkthdr.rcvif = ifp; 1786 m->m_pkthdr.csum_flags = 0; 1787 1788 if (nbufs > 1) { 1789 /* Dequeue the rest of chain. */ 1790 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 1791 continue; 1792 } 1793 1794 /* 1795 * Save copy of header before we strip it. For both mergeable 1796 * and non-mergeable, the header is at the beginning of the 1797 * mbuf data. We no longer need num_buffers, so always use a 1798 * regular header. 1799 * 1800 * BMV: Is this memcpy() expensive? We know the mbuf data is 1801 * still valid even after the m_adj(). 1802 */ 1803 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr)); 1804 m_adj(m, adjsz); 1805 1806 vtnet_rxq_input(rxq, m, hdr); 1807 1808 /* Must recheck after dropping the Rx lock. */ 1809 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1810 break; 1811 } 1812 1813 if (deq > 0) 1814 virtqueue_notify(vq); 1815 1816 return (count > 0 ? 0 : EAGAIN); 1817 } 1818 1819 static void 1820 vtnet_rx_vq_intr(void *xrxq) 1821 { 1822 struct vtnet_softc *sc; 1823 struct vtnet_rxq *rxq; 1824 struct ifnet *ifp; 1825 int tries, more; 1826 1827 rxq = xrxq; 1828 sc = rxq->vtnrx_sc; 1829 ifp = sc->vtnet_ifp; 1830 tries = 0; 1831 1832 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 1833 /* 1834 * Ignore this interrupt. Either this is a spurious interrupt 1835 * or multiqueue without per-VQ MSIX so every queue needs to 1836 * be polled (a brain dead configuration we could try harder 1837 * to avoid). 1838 */ 1839 vtnet_rxq_disable_intr(rxq); 1840 return; 1841 } 1842 1843 VTNET_RXQ_LOCK(rxq); 1844 1845 again: 1846 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1847 VTNET_RXQ_UNLOCK(rxq); 1848 return; 1849 } 1850 1851 more = vtnet_rxq_eof(rxq); 1852 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1853 if (!more) 1854 vtnet_rxq_disable_intr(rxq); 1855 /* 1856 * This is an occasional condition or race (when !more), 1857 * so retry a few times before scheduling the taskqueue. 1858 */ 1859 if (tries++ < VTNET_INTR_DISABLE_RETRIES) 1860 goto again; 1861 1862 VTNET_RXQ_UNLOCK(rxq); 1863 rxq->vtnrx_stats.vrxs_rescheduled++; 1864 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1865 } else 1866 VTNET_RXQ_UNLOCK(rxq); 1867 } 1868 1869 static void 1870 vtnet_rxq_tq_intr(void *xrxq, int pending) 1871 { 1872 struct vtnet_softc *sc; 1873 struct vtnet_rxq *rxq; 1874 struct ifnet *ifp; 1875 int more; 1876 1877 rxq = xrxq; 1878 sc = rxq->vtnrx_sc; 1879 ifp = sc->vtnet_ifp; 1880 1881 VTNET_RXQ_LOCK(rxq); 1882 1883 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1884 VTNET_RXQ_UNLOCK(rxq); 1885 return; 1886 } 1887 1888 more = vtnet_rxq_eof(rxq); 1889 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1890 if (!more) 1891 vtnet_rxq_disable_intr(rxq); 1892 rxq->vtnrx_stats.vrxs_rescheduled++; 1893 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1894 } 1895 1896 VTNET_RXQ_UNLOCK(rxq); 1897 } 1898 1899 static void 1900 vtnet_txq_free_mbufs(struct vtnet_txq *txq) 1901 { 1902 struct virtqueue *vq; 1903 struct vtnet_tx_header *txhdr; 1904 int last; 1905 1906 vq = txq->vtntx_vq; 1907 last = 0; 1908 1909 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 1910 m_freem(txhdr->vth_mbuf); 1911 uma_zfree(vtnet_tx_header_zone, txhdr); 1912 } 1913 1914 KASSERT(virtqueue_empty(vq), 1915 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 1916 } 1917 1918 /* 1919 * BMV: Much of this can go away once we finally have offsets in 1920 * the mbuf packet header. Bug andre@. 1921 */ 1922 static int 1923 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, 1924 int *etype, int *proto, int *start) 1925 { 1926 struct vtnet_softc *sc; 1927 struct ether_vlan_header *evh; 1928 int offset; 1929 1930 sc = txq->vtntx_sc; 1931 1932 evh = mtod(m, struct ether_vlan_header *); 1933 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 1934 /* BMV: We should handle nested VLAN tags too. */ 1935 *etype = ntohs(evh->evl_proto); 1936 offset = sizeof(struct ether_vlan_header); 1937 } else { 1938 *etype = ntohs(evh->evl_encap_proto); 1939 offset = sizeof(struct ether_header); 1940 } 1941 1942 switch (*etype) { 1943 #if defined(INET) 1944 case ETHERTYPE_IP: { 1945 struct ip *ip, iphdr; 1946 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 1947 m_copydata(m, offset, sizeof(struct ip), 1948 (caddr_t) &iphdr); 1949 ip = &iphdr; 1950 } else 1951 ip = (struct ip *)(m->m_data + offset); 1952 *proto = ip->ip_p; 1953 *start = offset + (ip->ip_hl << 2); 1954 break; 1955 } 1956 #endif 1957 #if defined(INET6) 1958 case ETHERTYPE_IPV6: 1959 *proto = -1; 1960 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 1961 /* Assert the network stack sent us a valid packet. */ 1962 KASSERT(*start > offset, 1963 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 1964 *start, offset, *proto)); 1965 break; 1966 #endif 1967 default: 1968 sc->vtnet_stats.tx_csum_bad_ethtype++; 1969 return (EINVAL); 1970 } 1971 1972 return (0); 1973 } 1974 1975 static int 1976 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 1977 int offset, struct virtio_net_hdr *hdr) 1978 { 1979 static struct timeval lastecn; 1980 static int curecn; 1981 struct vtnet_softc *sc; 1982 struct tcphdr *tcp, tcphdr; 1983 1984 sc = txq->vtntx_sc; 1985 1986 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 1987 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 1988 tcp = &tcphdr; 1989 } else 1990 tcp = (struct tcphdr *)(m->m_data + offset); 1991 1992 hdr->hdr_len = offset + (tcp->th_off << 2); 1993 hdr->gso_size = m->m_pkthdr.tso_segsz; 1994 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 1995 VIRTIO_NET_HDR_GSO_TCPV6; 1996 1997 if (tcp->th_flags & TH_CWR) { 1998 /* 1999 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD, 2000 * ECN support is not on a per-interface basis, but globally via 2001 * the net.inet.tcp.ecn.enable sysctl knob. The default is off. 2002 */ 2003 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2004 if (ppsratecheck(&lastecn, &curecn, 1)) 2005 if_printf(sc->vtnet_ifp, 2006 "TSO with ECN not negotiated with host\n"); 2007 return (ENOTSUP); 2008 } 2009 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2010 } 2011 2012 txq->vtntx_stats.vtxs_tso++; 2013 2014 return (0); 2015 } 2016 2017 static struct mbuf * 2018 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2019 struct virtio_net_hdr *hdr) 2020 { 2021 struct vtnet_softc *sc; 2022 int flags, etype, csum_start, proto, error; 2023 2024 sc = txq->vtntx_sc; 2025 flags = m->m_pkthdr.csum_flags; 2026 2027 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2028 if (error) 2029 goto drop; 2030 2031 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) || 2032 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) { 2033 /* 2034 * We could compare the IP protocol vs the CSUM_ flag too, 2035 * but that really should not be necessary. 2036 */ 2037 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2038 hdr->csum_start = csum_start; 2039 hdr->csum_offset = m->m_pkthdr.csum_data; 2040 txq->vtntx_stats.vtxs_csum++; 2041 } 2042 2043 if (flags & CSUM_TSO) { 2044 if (__predict_false(proto != IPPROTO_TCP)) { 2045 /* Likely failed to correctly parse the mbuf. */ 2046 sc->vtnet_stats.tx_tso_not_tcp++; 2047 goto drop; 2048 } 2049 2050 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM, 2051 ("%s: mbuf %p TSO without checksum offload %#x", 2052 __func__, m, flags)); 2053 2054 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2055 if (error) 2056 goto drop; 2057 } 2058 2059 return (m); 2060 2061 drop: 2062 m_freem(m); 2063 return (NULL); 2064 } 2065 2066 static int 2067 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2068 struct vtnet_tx_header *txhdr) 2069 { 2070 struct vtnet_softc *sc; 2071 struct virtqueue *vq; 2072 struct sglist *sg; 2073 struct mbuf *m; 2074 int error; 2075 2076 sc = txq->vtntx_sc; 2077 vq = txq->vtntx_vq; 2078 sg = txq->vtntx_sg; 2079 m = *m_head; 2080 2081 sglist_reset(sg); 2082 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2083 KASSERT(error == 0 && sg->sg_nseg == 1, 2084 ("%s: error %d adding header to sglist", __func__, error)); 2085 2086 error = sglist_append_mbuf(sg, m); 2087 if (error) { 2088 m = m_defrag(m, M_NOWAIT); 2089 if (m == NULL) 2090 goto fail; 2091 2092 *m_head = m; 2093 sc->vtnet_stats.tx_defragged++; 2094 2095 error = sglist_append_mbuf(sg, m); 2096 if (error) 2097 goto fail; 2098 } 2099 2100 txhdr->vth_mbuf = m; 2101 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2102 2103 return (error); 2104 2105 fail: 2106 sc->vtnet_stats.tx_defrag_failed++; 2107 m_freem(*m_head); 2108 *m_head = NULL; 2109 2110 return (ENOBUFS); 2111 } 2112 2113 static int 2114 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head) 2115 { 2116 struct vtnet_softc *sc; 2117 struct vtnet_tx_header *txhdr; 2118 struct virtio_net_hdr *hdr; 2119 struct mbuf *m; 2120 int error; 2121 2122 sc = txq->vtntx_sc; 2123 m = *m_head; 2124 M_ASSERTPKTHDR(m); 2125 2126 txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO); 2127 if (txhdr == NULL) { 2128 m_freem(m); 2129 *m_head = NULL; 2130 return (ENOMEM); 2131 } 2132 2133 /* 2134 * Always use the non-mergeable header, regardless if the feature 2135 * was negotiated. For transmit, num_buffers is always zero. The 2136 * vtnet_hdr_size is used to enqueue the correct header size. 2137 */ 2138 hdr = &txhdr->vth_uhdr.hdr; 2139 2140 if (m->m_flags & M_VLANTAG) { 2141 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2142 if ((*m_head = m) == NULL) { 2143 error = ENOBUFS; 2144 goto fail; 2145 } 2146 m->m_flags &= ~M_VLANTAG; 2147 } 2148 2149 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2150 m = vtnet_txq_offload(txq, m, hdr); 2151 if ((*m_head = m) == NULL) { 2152 error = ENOBUFS; 2153 goto fail; 2154 } 2155 } 2156 2157 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2158 if (error == 0) 2159 return (0); 2160 2161 fail: 2162 uma_zfree(vtnet_tx_header_zone, txhdr); 2163 2164 return (error); 2165 } 2166 2167 #ifdef VTNET_LEGACY_TX 2168 2169 static void 2170 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp) 2171 { 2172 struct vtnet_softc *sc; 2173 struct virtqueue *vq; 2174 struct mbuf *m0; 2175 int enq; 2176 2177 sc = txq->vtntx_sc; 2178 vq = txq->vtntx_vq; 2179 enq = 0; 2180 2181 VTNET_TXQ_LOCK_ASSERT(txq); 2182 2183 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2184 sc->vtnet_link_active == 0) 2185 return; 2186 2187 vtnet_txq_eof(txq); 2188 2189 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { 2190 if (virtqueue_full(vq)) 2191 break; 2192 2193 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0); 2194 if (m0 == NULL) 2195 break; 2196 2197 if (vtnet_txq_encap(txq, &m0) != 0) { 2198 if (m0 != NULL) 2199 IFQ_DRV_PREPEND(&ifp->if_snd, m0); 2200 break; 2201 } 2202 2203 enq++; 2204 ETHER_BPF_MTAP(ifp, m0); 2205 } 2206 2207 if (enq > 0) { 2208 virtqueue_notify(vq); 2209 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2210 } 2211 } 2212 2213 static void 2214 vtnet_start(struct ifnet *ifp) 2215 { 2216 struct vtnet_softc *sc; 2217 struct vtnet_txq *txq; 2218 2219 sc = ifp->if_softc; 2220 txq = &sc->vtnet_txqs[0]; 2221 2222 VTNET_TXQ_LOCK(txq); 2223 vtnet_start_locked(txq, ifp); 2224 VTNET_TXQ_UNLOCK(txq); 2225 } 2226 2227 #else /* !VTNET_LEGACY_TX */ 2228 2229 static int 2230 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2231 { 2232 struct vtnet_softc *sc; 2233 struct virtqueue *vq; 2234 struct buf_ring *br; 2235 struct ifnet *ifp; 2236 int enq, error; 2237 2238 sc = txq->vtntx_sc; 2239 vq = txq->vtntx_vq; 2240 br = txq->vtntx_br; 2241 ifp = sc->vtnet_ifp; 2242 enq = 0; 2243 error = 0; 2244 2245 VTNET_TXQ_LOCK_ASSERT(txq); 2246 2247 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2248 sc->vtnet_link_active == 0) { 2249 if (m != NULL) 2250 error = drbr_enqueue(ifp, br, m); 2251 return (error); 2252 } 2253 2254 if (m != NULL) { 2255 error = drbr_enqueue(ifp, br, m); 2256 if (error) 2257 return (error); 2258 } 2259 2260 vtnet_txq_eof(txq); 2261 2262 while ((m = drbr_peek(ifp, br)) != NULL) { 2263 if (virtqueue_full(vq)) { 2264 drbr_putback(ifp, br, m); 2265 error = ENOBUFS; 2266 break; 2267 } 2268 2269 error = vtnet_txq_encap(txq, &m); 2270 if (error) { 2271 if (m != NULL) 2272 drbr_putback(ifp, br, m); 2273 else 2274 drbr_advance(ifp, br); 2275 break; 2276 } 2277 drbr_advance(ifp, br); 2278 2279 enq++; 2280 ETHER_BPF_MTAP(ifp, m); 2281 } 2282 2283 if (enq > 0) { 2284 virtqueue_notify(vq); 2285 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2286 } 2287 2288 return (error); 2289 } 2290 2291 static int 2292 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m) 2293 { 2294 struct vtnet_softc *sc; 2295 struct vtnet_txq *txq; 2296 int i, npairs, error; 2297 2298 sc = ifp->if_softc; 2299 npairs = sc->vtnet_act_vq_pairs; 2300 2301 if (m->m_flags & M_FLOWID) 2302 i = m->m_pkthdr.flowid % npairs; 2303 else 2304 i = curcpu % npairs; 2305 2306 txq = &sc->vtnet_txqs[i]; 2307 2308 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2309 error = vtnet_txq_mq_start_locked(txq, m); 2310 VTNET_TXQ_UNLOCK(txq); 2311 } else { 2312 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2313 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2314 } 2315 2316 return (error); 2317 } 2318 2319 static void 2320 vtnet_txq_tq_deferred(void *xtxq, int pending) 2321 { 2322 struct vtnet_softc *sc; 2323 struct vtnet_txq *txq; 2324 2325 txq = xtxq; 2326 sc = txq->vtntx_sc; 2327 2328 VTNET_TXQ_LOCK(txq); 2329 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2330 vtnet_txq_mq_start_locked(txq, NULL); 2331 VTNET_TXQ_UNLOCK(txq); 2332 } 2333 2334 #endif /* VTNET_LEGACY_TX */ 2335 2336 static void 2337 vtnet_txq_start(struct vtnet_txq *txq) 2338 { 2339 struct vtnet_softc *sc; 2340 struct ifnet *ifp; 2341 2342 sc = txq->vtntx_sc; 2343 ifp = sc->vtnet_ifp; 2344 2345 #ifdef VTNET_LEGACY_TX 2346 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 2347 vtnet_start_locked(txq, ifp); 2348 #else 2349 if (!drbr_empty(ifp, txq->vtntx_br)) 2350 vtnet_txq_mq_start_locked(txq, NULL); 2351 #endif 2352 } 2353 2354 static void 2355 vtnet_txq_tq_intr(void *xtxq, int pending) 2356 { 2357 struct vtnet_softc *sc; 2358 struct vtnet_txq *txq; 2359 struct ifnet *ifp; 2360 2361 txq = xtxq; 2362 sc = txq->vtntx_sc; 2363 ifp = sc->vtnet_ifp; 2364 2365 VTNET_TXQ_LOCK(txq); 2366 2367 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2368 VTNET_TXQ_UNLOCK(txq); 2369 return; 2370 } 2371 2372 vtnet_txq_eof(txq); 2373 2374 vtnet_txq_start(txq); 2375 2376 if (vtnet_txq_enable_intr(txq) != 0) { 2377 vtnet_txq_disable_intr(txq); 2378 txq->vtntx_stats.vtxs_rescheduled++; 2379 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2380 } 2381 2382 VTNET_TXQ_UNLOCK(txq); 2383 } 2384 2385 static void 2386 vtnet_txq_eof(struct vtnet_txq *txq) 2387 { 2388 struct virtqueue *vq; 2389 struct vtnet_tx_header *txhdr; 2390 struct mbuf *m; 2391 2392 vq = txq->vtntx_vq; 2393 VTNET_TXQ_LOCK_ASSERT(txq); 2394 2395 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2396 m = txhdr->vth_mbuf; 2397 2398 txq->vtntx_stats.vtxs_opackets++; 2399 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2400 if (m->m_flags & M_MCAST) 2401 txq->vtntx_stats.vtxs_omcasts++; 2402 2403 m_freem(m); 2404 uma_zfree(vtnet_tx_header_zone, txhdr); 2405 } 2406 2407 if (virtqueue_empty(vq)) 2408 txq->vtntx_watchdog = 0; 2409 } 2410 2411 static void 2412 vtnet_tx_vq_intr(void *xtxq) 2413 { 2414 struct vtnet_softc *sc; 2415 struct vtnet_txq *txq; 2416 struct ifnet *ifp; 2417 int tries; 2418 2419 txq = xtxq; 2420 sc = txq->vtntx_sc; 2421 ifp = sc->vtnet_ifp; 2422 tries = 0; 2423 2424 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2425 /* 2426 * Ignore this interrupt. Either this is a spurious interrupt 2427 * or multiqueue without per-VQ MSIX so every queue needs to 2428 * be polled (a brain dead configuration we could try harder 2429 * to avoid). 2430 */ 2431 vtnet_txq_disable_intr(txq); 2432 return; 2433 } 2434 2435 VTNET_TXQ_LOCK(txq); 2436 2437 again: 2438 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2439 VTNET_TXQ_UNLOCK(txq); 2440 return; 2441 } 2442 2443 vtnet_txq_eof(txq); 2444 2445 vtnet_txq_start(txq); 2446 2447 if (vtnet_txq_enable_intr(txq) != 0) { 2448 vtnet_txq_disable_intr(txq); 2449 /* 2450 * This is an occasional race, so retry a few times 2451 * before scheduling the taskqueue. 2452 */ 2453 if (tries++ < VTNET_INTR_DISABLE_RETRIES) 2454 goto again; 2455 2456 VTNET_TXQ_UNLOCK(txq); 2457 txq->vtntx_stats.vtxs_rescheduled++; 2458 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2459 } else 2460 VTNET_TXQ_UNLOCK(txq); 2461 } 2462 2463 static void 2464 vtnet_tx_start_all(struct vtnet_softc *sc) 2465 { 2466 struct vtnet_txq *txq; 2467 int i; 2468 2469 VTNET_CORE_LOCK_ASSERT(sc); 2470 2471 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2472 txq = &sc->vtnet_txqs[i]; 2473 2474 VTNET_TXQ_LOCK(txq); 2475 vtnet_txq_start(txq); 2476 VTNET_TXQ_UNLOCK(txq); 2477 } 2478 } 2479 2480 #ifndef VTNET_LEGACY_TX 2481 static void 2482 vtnet_qflush(struct ifnet *ifp) 2483 { 2484 struct vtnet_softc *sc; 2485 struct vtnet_txq *txq; 2486 struct mbuf *m; 2487 int i; 2488 2489 sc = ifp->if_softc; 2490 2491 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2492 txq = &sc->vtnet_txqs[i]; 2493 2494 VTNET_TXQ_LOCK(txq); 2495 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2496 m_freem(m); 2497 VTNET_TXQ_UNLOCK(txq); 2498 } 2499 2500 if_qflush(ifp); 2501 } 2502 #endif 2503 2504 static int 2505 vtnet_watchdog(struct vtnet_txq *txq) 2506 { 2507 struct vtnet_softc *sc; 2508 2509 sc = txq->vtntx_sc; 2510 2511 VTNET_TXQ_LOCK(txq); 2512 if (sc->vtnet_flags & VTNET_FLAG_EVENT_IDX) 2513 vtnet_txq_eof(txq); 2514 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 2515 VTNET_TXQ_UNLOCK(txq); 2516 return (0); 2517 } 2518 VTNET_TXQ_UNLOCK(txq); 2519 2520 if_printf(sc->vtnet_ifp, "watchdog timeout on queue %d\n", 2521 txq->vtntx_id); 2522 return (1); 2523 } 2524 2525 static void 2526 vtnet_rxq_accum_stats(struct vtnet_rxq *rxq, struct vtnet_rxq_stats *accum) 2527 { 2528 struct vtnet_rxq_stats *st; 2529 2530 st = &rxq->vtnrx_stats; 2531 2532 accum->vrxs_ipackets += st->vrxs_ipackets; 2533 accum->vrxs_ibytes += st->vrxs_ibytes; 2534 accum->vrxs_iqdrops += st->vrxs_iqdrops; 2535 accum->vrxs_csum += st->vrxs_csum; 2536 accum->vrxs_csum_failed += st->vrxs_csum_failed; 2537 accum->vrxs_rescheduled += st->vrxs_rescheduled; 2538 } 2539 2540 static void 2541 vtnet_txq_accum_stats(struct vtnet_txq *txq, struct vtnet_txq_stats *accum) 2542 { 2543 struct vtnet_txq_stats *st; 2544 2545 st = &txq->vtntx_stats; 2546 2547 accum->vtxs_opackets += st->vtxs_opackets; 2548 accum->vtxs_obytes += st->vtxs_obytes; 2549 accum->vtxs_csum += st->vtxs_csum; 2550 accum->vtxs_tso += st->vtxs_tso; 2551 accum->vtxs_rescheduled += st->vtxs_rescheduled; 2552 } 2553 2554 static void 2555 vtnet_accumulate_stats(struct vtnet_softc *sc) 2556 { 2557 struct ifnet *ifp; 2558 struct vtnet_statistics *st; 2559 struct vtnet_rxq_stats rxaccum; 2560 struct vtnet_txq_stats txaccum; 2561 int i; 2562 2563 ifp = sc->vtnet_ifp; 2564 st = &sc->vtnet_stats; 2565 bzero(&rxaccum, sizeof(struct vtnet_rxq_stats)); 2566 bzero(&txaccum, sizeof(struct vtnet_txq_stats)); 2567 2568 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2569 vtnet_rxq_accum_stats(&sc->vtnet_rxqs[i], &rxaccum); 2570 vtnet_txq_accum_stats(&sc->vtnet_txqs[i], &txaccum); 2571 } 2572 2573 st->rx_csum_offloaded = rxaccum.vrxs_csum; 2574 st->rx_csum_failed = rxaccum.vrxs_csum_failed; 2575 st->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 2576 st->tx_csum_offloaded = txaccum.vtxs_csum; 2577 st->tx_tso_offloaded = txaccum.vtxs_tso; 2578 st->tx_task_rescheduled = txaccum.vtxs_rescheduled; 2579 2580 /* 2581 * With the exception of if_ierrors, these ifnet statistics are 2582 * only updated in the driver, so just set them to our accumulated 2583 * values. if_ierrors is updated in ether_input() for malformed 2584 * frames that we should have already discarded. 2585 */ 2586 ifp->if_ipackets = rxaccum.vrxs_ipackets; 2587 ifp->if_iqdrops = rxaccum.vrxs_iqdrops; 2588 ifp->if_ierrors = rxaccum.vrxs_ierrors; 2589 ifp->if_opackets = txaccum.vtxs_opackets; 2590 #ifndef VTNET_LEGACY_TX 2591 ifp->if_obytes = txaccum.vtxs_obytes; 2592 ifp->if_omcasts = txaccum.vtxs_omcasts; 2593 #endif 2594 } 2595 2596 static void 2597 vtnet_tick(void *xsc) 2598 { 2599 struct vtnet_softc *sc; 2600 struct ifnet *ifp; 2601 int i, timedout; 2602 2603 sc = xsc; 2604 ifp = sc->vtnet_ifp; 2605 timedout = 0; 2606 2607 VTNET_CORE_LOCK_ASSERT(sc); 2608 vtnet_accumulate_stats(sc); 2609 2610 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 2611 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 2612 2613 if (timedout != 0) { 2614 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2615 vtnet_init_locked(sc); 2616 } else 2617 callout_schedule(&sc->vtnet_tick_ch, hz); 2618 } 2619 2620 static void 2621 vtnet_start_taskqueues(struct vtnet_softc *sc) 2622 { 2623 device_t dev; 2624 struct vtnet_rxq *rxq; 2625 struct vtnet_txq *txq; 2626 int i, error; 2627 2628 dev = sc->vtnet_dev; 2629 2630 /* 2631 * Errors here are very difficult to recover from - we cannot 2632 * easily fail because, if this is during boot, we will hang 2633 * when freeing any successfully started taskqueues because 2634 * the scheduler isn't up yet. 2635 * 2636 * Most drivers just ignore the return value - it only fails 2637 * with ENOMEM so an error is not likely. 2638 */ 2639 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2640 rxq = &sc->vtnet_rxqs[i]; 2641 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 2642 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 2643 if (error) { 2644 device_printf(dev, "failed to start rx taskq %d\n", 2645 rxq->vtnrx_id); 2646 } 2647 2648 txq = &sc->vtnet_txqs[i]; 2649 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 2650 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 2651 if (error) { 2652 device_printf(dev, "failed to start tx taskq %d\n", 2653 txq->vtntx_id); 2654 } 2655 } 2656 } 2657 2658 static void 2659 vtnet_free_taskqueues(struct vtnet_softc *sc) 2660 { 2661 struct vtnet_rxq *rxq; 2662 struct vtnet_txq *txq; 2663 int i; 2664 2665 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2666 rxq = &sc->vtnet_rxqs[i]; 2667 if (rxq->vtnrx_tq != NULL) { 2668 taskqueue_free(rxq->vtnrx_tq); 2669 rxq->vtnrx_vq = NULL; 2670 } 2671 2672 txq = &sc->vtnet_txqs[i]; 2673 if (txq->vtntx_tq != NULL) { 2674 taskqueue_free(txq->vtntx_tq); 2675 txq->vtntx_tq = NULL; 2676 } 2677 } 2678 } 2679 2680 static void 2681 vtnet_drain_taskqueues(struct vtnet_softc *sc) 2682 { 2683 struct vtnet_rxq *rxq; 2684 struct vtnet_txq *txq; 2685 int i; 2686 2687 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2688 rxq = &sc->vtnet_rxqs[i]; 2689 if (rxq->vtnrx_tq != NULL) 2690 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2691 2692 txq = &sc->vtnet_txqs[i]; 2693 if (txq->vtntx_tq != NULL) { 2694 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 2695 #ifndef VTNET_LEGACY_TX 2696 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 2697 #endif 2698 } 2699 } 2700 } 2701 2702 static void 2703 vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 2704 { 2705 struct vtnet_rxq *rxq; 2706 struct vtnet_txq *txq; 2707 int i; 2708 2709 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2710 rxq = &sc->vtnet_rxqs[i]; 2711 vtnet_rxq_free_mbufs(rxq); 2712 2713 txq = &sc->vtnet_txqs[i]; 2714 vtnet_txq_free_mbufs(txq); 2715 } 2716 } 2717 2718 static void 2719 vtnet_stop_rendezvous(struct vtnet_softc *sc) 2720 { 2721 struct vtnet_rxq *rxq; 2722 struct vtnet_txq *txq; 2723 int i; 2724 2725 /* 2726 * Lock and unlock the per-queue mutex so we known the stop 2727 * state is visible. Doing only the active queues should be 2728 * sufficient, but it does not cost much extra to do all the 2729 * queues. Note we hold the core mutex here too. 2730 */ 2731 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2732 rxq = &sc->vtnet_rxqs[i]; 2733 VTNET_RXQ_LOCK(rxq); 2734 VTNET_RXQ_UNLOCK(rxq); 2735 2736 txq = &sc->vtnet_txqs[i]; 2737 VTNET_TXQ_LOCK(txq); 2738 VTNET_TXQ_UNLOCK(txq); 2739 } 2740 } 2741 2742 static void 2743 vtnet_stop(struct vtnet_softc *sc) 2744 { 2745 device_t dev; 2746 struct ifnet *ifp; 2747 2748 dev = sc->vtnet_dev; 2749 ifp = sc->vtnet_ifp; 2750 2751 VTNET_CORE_LOCK_ASSERT(sc); 2752 2753 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2754 sc->vtnet_link_active = 0; 2755 callout_stop(&sc->vtnet_tick_ch); 2756 2757 /* Only advisory. */ 2758 vtnet_disable_interrupts(sc); 2759 2760 /* 2761 * Stop the host adapter. This resets it to the pre-initialized 2762 * state. It will not generate any interrupts until after it is 2763 * reinitialized. 2764 */ 2765 virtio_stop(dev); 2766 vtnet_stop_rendezvous(sc); 2767 2768 /* Free any mbufs left in the virtqueues. */ 2769 vtnet_drain_rxtx_queues(sc); 2770 } 2771 2772 static int 2773 vtnet_virtio_reinit(struct vtnet_softc *sc) 2774 { 2775 device_t dev; 2776 struct ifnet *ifp; 2777 uint64_t features; 2778 int mask, error; 2779 2780 dev = sc->vtnet_dev; 2781 ifp = sc->vtnet_ifp; 2782 features = sc->vtnet_features; 2783 2784 mask = 0; 2785 #if defined(INET) 2786 mask |= IFCAP_RXCSUM; 2787 #endif 2788 #if defined (INET6) 2789 mask |= IFCAP_RXCSUM_IPV6; 2790 #endif 2791 2792 /* 2793 * Re-negotiate with the host, removing any disabled receive 2794 * features. Transmit features are disabled only on our side 2795 * via if_capenable and if_hwassist. 2796 */ 2797 2798 if (ifp->if_capabilities & mask) { 2799 /* 2800 * We require both IPv4 and IPv6 offloading to be enabled 2801 * in order to negotiated it: VirtIO does not distinguish 2802 * between the two. 2803 */ 2804 if ((ifp->if_capenable & mask) != mask) 2805 features &= ~VIRTIO_NET_F_GUEST_CSUM; 2806 } 2807 2808 if (ifp->if_capabilities & IFCAP_LRO) { 2809 if ((ifp->if_capenable & IFCAP_LRO) == 0) 2810 features &= ~VTNET_LRO_FEATURES; 2811 } 2812 2813 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) { 2814 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) 2815 features &= ~VIRTIO_NET_F_CTRL_VLAN; 2816 } 2817 2818 error = virtio_reinit(dev, features); 2819 if (error) 2820 device_printf(dev, "virtio reinit error %d\n", error); 2821 2822 return (error); 2823 } 2824 2825 static void 2826 vtnet_init_rx_filters(struct vtnet_softc *sc) 2827 { 2828 struct ifnet *ifp; 2829 2830 ifp = sc->vtnet_ifp; 2831 2832 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 2833 /* Restore promiscuous and all-multicast modes. */ 2834 vtnet_rx_filter(sc); 2835 /* Restore filtered MAC addresses. */ 2836 vtnet_rx_filter_mac(sc); 2837 } 2838 2839 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) 2840 vtnet_rx_filter_vlan(sc); 2841 } 2842 2843 static int 2844 vtnet_init_rx_queues(struct vtnet_softc *sc) 2845 { 2846 device_t dev; 2847 struct vtnet_rxq *rxq; 2848 int i, clsize, error; 2849 2850 dev = sc->vtnet_dev; 2851 2852 /* 2853 * Use the new cluster size if one has been set (via a MTU 2854 * change). Otherwise, use the standard 2K clusters. 2855 * 2856 * BMV: It might make sense to use page sized clusters as 2857 * the default (depending on the features negotiated). 2858 */ 2859 if (sc->vtnet_rx_new_clsize != 0) { 2860 clsize = sc->vtnet_rx_new_clsize; 2861 sc->vtnet_rx_new_clsize = 0; 2862 } else 2863 clsize = MCLBYTES; 2864 2865 sc->vtnet_rx_clsize = clsize; 2866 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize); 2867 2868 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS || 2869 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 2870 ("%s: too many rx mbufs %d for %d segments", __func__, 2871 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 2872 2873 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2874 rxq = &sc->vtnet_rxqs[i]; 2875 2876 /* Hold the lock to satisfy asserts. */ 2877 VTNET_RXQ_LOCK(rxq); 2878 error = vtnet_rxq_populate(rxq); 2879 VTNET_RXQ_UNLOCK(rxq); 2880 2881 if (error) { 2882 device_printf(dev, 2883 "cannot allocate mbufs for Rx queue %d\n", i); 2884 return (error); 2885 } 2886 } 2887 2888 return (0); 2889 } 2890 2891 static int 2892 vtnet_init_tx_queues(struct vtnet_softc *sc) 2893 { 2894 struct vtnet_txq *txq; 2895 int i; 2896 2897 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2898 txq = &sc->vtnet_txqs[i]; 2899 txq->vtntx_watchdog = 0; 2900 } 2901 2902 return (0); 2903 } 2904 2905 static int 2906 vtnet_init_rxtx_queues(struct vtnet_softc *sc) 2907 { 2908 int error; 2909 2910 error = vtnet_init_rx_queues(sc); 2911 if (error) 2912 return (error); 2913 2914 error = vtnet_init_tx_queues(sc); 2915 if (error) 2916 return (error); 2917 2918 return (0); 2919 } 2920 2921 static void 2922 vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 2923 { 2924 device_t dev; 2925 int npairs; 2926 2927 dev = sc->vtnet_dev; 2928 2929 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) { 2930 MPASS(sc->vtnet_max_vq_pairs == 1); 2931 sc->vtnet_act_vq_pairs = 1; 2932 return; 2933 } 2934 2935 /* BMV: Just use the maximum configured for now. */ 2936 npairs = sc->vtnet_max_vq_pairs; 2937 2938 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 2939 device_printf(dev, 2940 "cannot set active queue pairs to %d\n", npairs); 2941 npairs = 1; 2942 } 2943 2944 sc->vtnet_act_vq_pairs = npairs; 2945 } 2946 2947 static int 2948 vtnet_reinit(struct vtnet_softc *sc) 2949 { 2950 device_t dev; 2951 struct ifnet *ifp; 2952 int error; 2953 2954 dev = sc->vtnet_dev; 2955 ifp = sc->vtnet_ifp; 2956 2957 /* Use the current MAC address. */ 2958 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 2959 vtnet_set_hwaddr(sc); 2960 2961 vtnet_set_active_vq_pairs(sc); 2962 2963 ifp->if_hwassist = 0; 2964 if (ifp->if_capenable & IFCAP_TXCSUM) 2965 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 2966 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) 2967 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6; 2968 if (ifp->if_capenable & IFCAP_TSO4) 2969 ifp->if_hwassist |= CSUM_TSO; 2970 if (ifp->if_capenable & IFCAP_TSO6) 2971 ifp->if_hwassist |= CSUM_TSO; /* No CSUM_TSO_IPV6. */ 2972 2973 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 2974 vtnet_init_rx_filters(sc); 2975 2976 error = vtnet_init_rxtx_queues(sc); 2977 if (error) 2978 return (error); 2979 2980 vtnet_enable_interrupts(sc); 2981 ifp->if_drv_flags |= IFF_DRV_RUNNING; 2982 2983 return (0); 2984 } 2985 2986 static void 2987 vtnet_init_locked(struct vtnet_softc *sc) 2988 { 2989 device_t dev; 2990 struct ifnet *ifp; 2991 2992 dev = sc->vtnet_dev; 2993 ifp = sc->vtnet_ifp; 2994 2995 VTNET_CORE_LOCK_ASSERT(sc); 2996 2997 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 2998 return; 2999 3000 vtnet_stop(sc); 3001 3002 /* Reinitialize with the host. */ 3003 if (vtnet_virtio_reinit(sc) != 0) 3004 goto fail; 3005 3006 if (vtnet_reinit(sc) != 0) 3007 goto fail; 3008 3009 virtio_reinit_complete(dev); 3010 3011 vtnet_update_link_status(sc); 3012 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3013 3014 return; 3015 3016 fail: 3017 vtnet_stop(sc); 3018 } 3019 3020 static void 3021 vtnet_init(void *xsc) 3022 { 3023 struct vtnet_softc *sc; 3024 3025 sc = xsc; 3026 3027 VTNET_CORE_LOCK(sc); 3028 vtnet_init_locked(sc); 3029 VTNET_CORE_UNLOCK(sc); 3030 } 3031 3032 static void 3033 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3034 { 3035 struct virtqueue *vq; 3036 3037 vq = sc->vtnet_ctrl_vq; 3038 3039 /* 3040 * The control virtqueue is only polled and therefore it should 3041 * already be empty. 3042 */ 3043 KASSERT(virtqueue_empty(vq), 3044 ("%s: ctrl vq %p not empty", __func__, vq)); 3045 } 3046 3047 static void 3048 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3049 struct sglist *sg, int readable, int writable) 3050 { 3051 struct virtqueue *vq; 3052 3053 vq = sc->vtnet_ctrl_vq; 3054 3055 VTNET_CORE_LOCK_ASSERT(sc); 3056 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ, 3057 ("%s: CTRL_VQ feature not negotiated", __func__)); 3058 3059 if (!virtqueue_empty(vq)) 3060 return; 3061 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0) 3062 return; 3063 3064 /* 3065 * Poll for the response, but the command is likely already 3066 * done when we return from the notify. 3067 */ 3068 virtqueue_notify(vq); 3069 virtqueue_poll(vq, NULL); 3070 } 3071 3072 static int 3073 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3074 { 3075 struct virtio_net_ctrl_hdr hdr; 3076 struct sglist_seg segs[3]; 3077 struct sglist sg; 3078 uint8_t ack; 3079 int error; 3080 3081 hdr.class = VIRTIO_NET_CTRL_MAC; 3082 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3083 ack = VIRTIO_NET_ERR; 3084 3085 sglist_init(&sg, 3, segs); 3086 error = 0; 3087 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3088 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN); 3089 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3090 KASSERT(error == 0 && sg.sg_nseg == 3, 3091 ("%s: error %d adding set MAC msg to sglist", __func__, error)); 3092 3093 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3094 3095 return (ack == VIRTIO_NET_OK ? 0 : EIO); 3096 } 3097 3098 static int 3099 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3100 { 3101 struct sglist_seg segs[3]; 3102 struct sglist sg; 3103 struct { 3104 struct virtio_net_ctrl_hdr hdr; 3105 uint8_t pad1; 3106 struct virtio_net_ctrl_mq mq; 3107 uint8_t pad2; 3108 uint8_t ack; 3109 } s; 3110 int error; 3111 3112 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3113 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3114 s.mq.virtqueue_pairs = npairs; 3115 s.ack = VIRTIO_NET_ERR; 3116 3117 sglist_init(&sg, 3, segs); 3118 error = 0; 3119 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3120 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3121 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3122 KASSERT(error == 0 && sg.sg_nseg == 3, 3123 ("%s: error %d adding MQ message to sglist", __func__, error)); 3124 3125 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3126 3127 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3128 } 3129 3130 static int 3131 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on) 3132 { 3133 struct sglist_seg segs[3]; 3134 struct sglist sg; 3135 struct { 3136 struct virtio_net_ctrl_hdr hdr; 3137 uint8_t pad1; 3138 uint8_t onoff; 3139 uint8_t pad2; 3140 uint8_t ack; 3141 } s; 3142 int error; 3143 3144 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3145 ("%s: CTRL_RX feature not negotiated", __func__)); 3146 3147 s.hdr.class = VIRTIO_NET_CTRL_RX; 3148 s.hdr.cmd = cmd; 3149 s.onoff = !!on; 3150 s.ack = VIRTIO_NET_ERR; 3151 3152 sglist_init(&sg, 3, segs); 3153 error = 0; 3154 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3155 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3156 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3157 KASSERT(error == 0 && sg.sg_nseg == 3, 3158 ("%s: error %d adding Rx message to sglist", __func__, error)); 3159 3160 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3161 3162 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3163 } 3164 3165 static int 3166 vtnet_set_promisc(struct vtnet_softc *sc, int on) 3167 { 3168 3169 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3170 } 3171 3172 static int 3173 vtnet_set_allmulti(struct vtnet_softc *sc, int on) 3174 { 3175 3176 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3177 } 3178 3179 /* 3180 * The device defaults to promiscuous mode for backwards compatibility. 3181 * Turn it off at attach time if possible. 3182 */ 3183 static void 3184 vtnet_attach_disable_promisc(struct vtnet_softc *sc) 3185 { 3186 struct ifnet *ifp; 3187 3188 ifp = sc->vtnet_ifp; 3189 3190 VTNET_CORE_LOCK(sc); 3191 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) { 3192 ifp->if_flags |= IFF_PROMISC; 3193 } else if (vtnet_set_promisc(sc, 0) != 0) { 3194 ifp->if_flags |= IFF_PROMISC; 3195 device_printf(sc->vtnet_dev, 3196 "cannot disable default promiscuous mode\n"); 3197 } 3198 VTNET_CORE_UNLOCK(sc); 3199 } 3200 3201 static void 3202 vtnet_rx_filter(struct vtnet_softc *sc) 3203 { 3204 device_t dev; 3205 struct ifnet *ifp; 3206 3207 dev = sc->vtnet_dev; 3208 ifp = sc->vtnet_ifp; 3209 3210 VTNET_CORE_LOCK_ASSERT(sc); 3211 3212 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) 3213 device_printf(dev, "cannot %s promiscuous mode\n", 3214 ifp->if_flags & IFF_PROMISC ? "enable" : "disable"); 3215 3216 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) 3217 device_printf(dev, "cannot %s all-multicast mode\n", 3218 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable"); 3219 } 3220 3221 static void 3222 vtnet_rx_filter_mac(struct vtnet_softc *sc) 3223 { 3224 struct virtio_net_ctrl_hdr hdr; 3225 struct vtnet_mac_filter *filter; 3226 struct sglist_seg segs[4]; 3227 struct sglist sg; 3228 struct ifnet *ifp; 3229 struct ifaddr *ifa; 3230 struct ifmultiaddr *ifma; 3231 int ucnt, mcnt, promisc, allmulti, error; 3232 uint8_t ack; 3233 3234 ifp = sc->vtnet_ifp; 3235 filter = sc->vtnet_mac_filter; 3236 ucnt = 0; 3237 mcnt = 0; 3238 promisc = 0; 3239 allmulti = 0; 3240 3241 VTNET_CORE_LOCK_ASSERT(sc); 3242 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3243 ("%s: CTRL_RX feature not negotiated", __func__)); 3244 3245 /* Unicast MAC addresses: */ 3246 if_addr_rlock(ifp); 3247 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 3248 if (ifa->ifa_addr->sa_family != AF_LINK) 3249 continue; 3250 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 3251 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3252 continue; 3253 else if (ucnt == VTNET_MAX_MAC_ENTRIES) { 3254 promisc = 1; 3255 break; 3256 } 3257 3258 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 3259 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN); 3260 ucnt++; 3261 } 3262 if_addr_runlock(ifp); 3263 3264 if (promisc != 0) { 3265 filter->vmf_unicast.nentries = 0; 3266 if_printf(ifp, "more than %d MAC addresses assigned, " 3267 "falling back to promiscuous mode\n", 3268 VTNET_MAX_MAC_ENTRIES); 3269 } else 3270 filter->vmf_unicast.nentries = ucnt; 3271 3272 /* Multicast MAC addresses: */ 3273 if_maddr_rlock(ifp); 3274 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3275 if (ifma->ifma_addr->sa_family != AF_LINK) 3276 continue; 3277 else if (mcnt == VTNET_MAX_MAC_ENTRIES) { 3278 allmulti = 1; 3279 break; 3280 } 3281 3282 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 3283 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN); 3284 mcnt++; 3285 } 3286 if_maddr_runlock(ifp); 3287 3288 if (allmulti != 0) { 3289 filter->vmf_multicast.nentries = 0; 3290 if_printf(ifp, "more than %d multicast MAC addresses " 3291 "assigned, falling back to all-multicast mode\n", 3292 VTNET_MAX_MAC_ENTRIES); 3293 } else 3294 filter->vmf_multicast.nentries = mcnt; 3295 3296 if (promisc != 0 && allmulti != 0) 3297 goto out; 3298 3299 hdr.class = VIRTIO_NET_CTRL_MAC; 3300 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3301 ack = VIRTIO_NET_ERR; 3302 3303 sglist_init(&sg, 4, segs); 3304 error = 0; 3305 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3306 error |= sglist_append(&sg, &filter->vmf_unicast, 3307 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN); 3308 error |= sglist_append(&sg, &filter->vmf_multicast, 3309 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN); 3310 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3311 KASSERT(error == 0 && sg.sg_nseg == 4, 3312 ("%s: error %d adding MAC filter msg to sglist", __func__, error)); 3313 3314 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3315 3316 if (ack != VIRTIO_NET_OK) 3317 if_printf(ifp, "error setting host MAC filter table\n"); 3318 3319 out: 3320 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0) 3321 if_printf(ifp, "cannot enable promiscuous mode\n"); 3322 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0) 3323 if_printf(ifp, "cannot enable all-multicast mode\n"); 3324 } 3325 3326 static int 3327 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3328 { 3329 struct sglist_seg segs[3]; 3330 struct sglist sg; 3331 struct { 3332 struct virtio_net_ctrl_hdr hdr; 3333 uint8_t pad1; 3334 uint16_t tag; 3335 uint8_t pad2; 3336 uint8_t ack; 3337 } s; 3338 int error; 3339 3340 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3341 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3342 s.tag = tag; 3343 s.ack = VIRTIO_NET_ERR; 3344 3345 sglist_init(&sg, 3, segs); 3346 error = 0; 3347 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3348 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3349 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3350 KASSERT(error == 0 && sg.sg_nseg == 3, 3351 ("%s: error %d adding VLAN message to sglist", __func__, error)); 3352 3353 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3354 3355 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3356 } 3357 3358 static void 3359 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3360 { 3361 uint32_t w; 3362 uint16_t tag; 3363 int i, bit; 3364 3365 VTNET_CORE_LOCK_ASSERT(sc); 3366 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER, 3367 ("%s: VLAN_FILTER feature not negotiated", __func__)); 3368 3369 /* Enable the filter for each configured VLAN. */ 3370 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3371 w = sc->vtnet_vlan_filter[i]; 3372 3373 while ((bit = ffs(w) - 1) != -1) { 3374 w &= ~(1 << bit); 3375 tag = sizeof(w) * CHAR_BIT * i + bit; 3376 3377 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3378 device_printf(sc->vtnet_dev, 3379 "cannot enable VLAN %d filter\n", tag); 3380 } 3381 } 3382 } 3383 } 3384 3385 static void 3386 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3387 { 3388 struct ifnet *ifp; 3389 int idx, bit; 3390 3391 ifp = sc->vtnet_ifp; 3392 idx = (tag >> 5) & 0x7F; 3393 bit = tag & 0x1F; 3394 3395 if (tag == 0 || tag > 4095) 3396 return; 3397 3398 VTNET_CORE_LOCK(sc); 3399 3400 if (add) 3401 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3402 else 3403 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3404 3405 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && 3406 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3407 device_printf(sc->vtnet_dev, 3408 "cannot %s VLAN %d %s the host filter table\n", 3409 add ? "add" : "remove", tag, add ? "to" : "from"); 3410 } 3411 3412 VTNET_CORE_UNLOCK(sc); 3413 } 3414 3415 static void 3416 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3417 { 3418 3419 if (ifp->if_softc != arg) 3420 return; 3421 3422 vtnet_update_vlan_filter(arg, 1, tag); 3423 } 3424 3425 static void 3426 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3427 { 3428 3429 if (ifp->if_softc != arg) 3430 return; 3431 3432 vtnet_update_vlan_filter(arg, 0, tag); 3433 } 3434 3435 static int 3436 vtnet_is_link_up(struct vtnet_softc *sc) 3437 { 3438 device_t dev; 3439 struct ifnet *ifp; 3440 uint16_t status; 3441 3442 dev = sc->vtnet_dev; 3443 ifp = sc->vtnet_ifp; 3444 3445 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0) 3446 status = VIRTIO_NET_S_LINK_UP; 3447 else 3448 status = virtio_read_dev_config_2(dev, 3449 offsetof(struct virtio_net_config, status)); 3450 3451 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3452 } 3453 3454 static void 3455 vtnet_update_link_status(struct vtnet_softc *sc) 3456 { 3457 struct ifnet *ifp; 3458 int link; 3459 3460 ifp = sc->vtnet_ifp; 3461 3462 VTNET_CORE_LOCK_ASSERT(sc); 3463 link = vtnet_is_link_up(sc); 3464 3465 /* Notify if the link status has changed. */ 3466 if (link != 0 && sc->vtnet_link_active == 0) { 3467 sc->vtnet_link_active = 1; 3468 if_link_state_change(ifp, LINK_STATE_UP); 3469 } else if (link == 0 && sc->vtnet_link_active != 0) { 3470 sc->vtnet_link_active = 0; 3471 if_link_state_change(ifp, LINK_STATE_DOWN); 3472 } 3473 } 3474 3475 static int 3476 vtnet_ifmedia_upd(struct ifnet *ifp) 3477 { 3478 struct vtnet_softc *sc; 3479 struct ifmedia *ifm; 3480 3481 sc = ifp->if_softc; 3482 ifm = &sc->vtnet_media; 3483 3484 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) 3485 return (EINVAL); 3486 3487 return (0); 3488 } 3489 3490 static void 3491 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 3492 { 3493 struct vtnet_softc *sc; 3494 3495 sc = ifp->if_softc; 3496 3497 ifmr->ifm_status = IFM_AVALID; 3498 ifmr->ifm_active = IFM_ETHER; 3499 3500 VTNET_CORE_LOCK(sc); 3501 if (vtnet_is_link_up(sc) != 0) { 3502 ifmr->ifm_status |= IFM_ACTIVE; 3503 ifmr->ifm_active |= VTNET_MEDIATYPE; 3504 } else 3505 ifmr->ifm_active |= IFM_NONE; 3506 VTNET_CORE_UNLOCK(sc); 3507 } 3508 3509 static void 3510 vtnet_set_hwaddr(struct vtnet_softc *sc) 3511 { 3512 device_t dev; 3513 int i; 3514 3515 dev = sc->vtnet_dev; 3516 3517 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 3518 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0) 3519 device_printf(dev, "unable to set MAC address\n"); 3520 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) { 3521 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3522 virtio_write_dev_config_1(dev, 3523 offsetof(struct virtio_net_config, mac) + i, 3524 sc->vtnet_hwaddr[i]); 3525 } 3526 } 3527 } 3528 3529 static void 3530 vtnet_get_hwaddr(struct vtnet_softc *sc) 3531 { 3532 device_t dev; 3533 int i; 3534 3535 dev = sc->vtnet_dev; 3536 3537 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { 3538 /* 3539 * Generate a random locally administered unicast address. 3540 * 3541 * It would be nice to generate the same MAC address across 3542 * reboots, but it seems all the hosts currently available 3543 * support the MAC feature, so this isn't too important. 3544 */ 3545 sc->vtnet_hwaddr[0] = 0xB2; 3546 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 3547 vtnet_set_hwaddr(sc); 3548 return; 3549 } 3550 3551 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3552 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev, 3553 offsetof(struct virtio_net_config, mac) + i); 3554 } 3555 } 3556 3557 static void 3558 vtnet_vlan_tag_remove(struct mbuf *m) 3559 { 3560 struct ether_vlan_header *evh; 3561 3562 evh = mtod(m, struct ether_vlan_header *); 3563 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 3564 m->m_flags |= M_VLANTAG; 3565 3566 /* Strip the 802.1Q header. */ 3567 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 3568 ETHER_HDR_LEN - ETHER_TYPE_LEN); 3569 m_adj(m, ETHER_VLAN_ENCAP_LEN); 3570 } 3571 3572 static void 3573 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 3574 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 3575 { 3576 struct sysctl_oid *node; 3577 struct sysctl_oid_list *list; 3578 struct vtnet_rxq_stats *stats; 3579 char namebuf[16]; 3580 3581 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 3582 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3583 CTLFLAG_RD, NULL, "Receive Queue"); 3584 list = SYSCTL_CHILDREN(node); 3585 3586 stats = &rxq->vtnrx_stats; 3587 3588 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, 3589 &stats->vrxs_ipackets, "Receive packets"); 3590 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, 3591 &stats->vrxs_ibytes, "Receive bytes"); 3592 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, 3593 &stats->vrxs_iqdrops, "Receive drops"); 3594 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, 3595 &stats->vrxs_ierrors, "Receive errors"); 3596 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3597 &stats->vrxs_csum, "Receive checksum offloaded"); 3598 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, 3599 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 3600 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3601 &stats->vrxs_rescheduled, 3602 "Receive interrupt handler rescheduled"); 3603 } 3604 3605 static void 3606 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 3607 struct sysctl_oid_list *child, struct vtnet_txq *txq) 3608 { 3609 struct sysctl_oid *node; 3610 struct sysctl_oid_list *list; 3611 struct vtnet_txq_stats *stats; 3612 char namebuf[16]; 3613 3614 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 3615 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3616 CTLFLAG_RD, NULL, "Transmit Queue"); 3617 list = SYSCTL_CHILDREN(node); 3618 3619 stats = &txq->vtntx_stats; 3620 3621 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, 3622 &stats->vtxs_opackets, "Transmit packets"); 3623 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, 3624 &stats->vtxs_obytes, "Transmit bytes"); 3625 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, 3626 &stats->vtxs_omcasts, "Transmit multicasts"); 3627 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3628 &stats->vtxs_csum, "Transmit checksum offloaded"); 3629 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, 3630 &stats->vtxs_tso, "Transmit segmentation offloaded"); 3631 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3632 &stats->vtxs_rescheduled, 3633 "Transmit interrupt handler rescheduled"); 3634 } 3635 3636 static void 3637 vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 3638 { 3639 device_t dev; 3640 struct sysctl_ctx_list *ctx; 3641 struct sysctl_oid *tree; 3642 struct sysctl_oid_list *child; 3643 int i; 3644 3645 dev = sc->vtnet_dev; 3646 ctx = device_get_sysctl_ctx(dev); 3647 tree = device_get_sysctl_tree(dev); 3648 child = SYSCTL_CHILDREN(tree); 3649 3650 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3651 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 3652 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 3653 } 3654 } 3655 3656 static void 3657 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 3658 struct sysctl_oid_list *child, struct vtnet_softc *sc) 3659 { 3660 struct vtnet_statistics *stats; 3661 3662 stats = &sc->vtnet_stats; 3663 3664 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 3665 CTLFLAG_RD, &stats->mbuf_alloc_failed, 3666 "Mbuf cluster allocation failures"); 3667 3668 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 3669 CTLFLAG_RD, &stats->rx_frame_too_large, 3670 "Received frame larger than the mbuf chain"); 3671 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 3672 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 3673 "Enqueuing the replacement receive mbuf failed"); 3674 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 3675 CTLFLAG_RD, &stats->rx_mergeable_failed, 3676 "Mergeable buffers receive failures"); 3677 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 3678 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 3679 "Received checksum offloaded buffer with unsupported " 3680 "Ethernet type"); 3681 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 3682 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 3683 "Received checksum offloaded buffer with incorrect IP protocol"); 3684 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 3685 CTLFLAG_RD, &stats->rx_csum_bad_offset, 3686 "Received checksum offloaded buffer with incorrect offset"); 3687 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", 3688 CTLFLAG_RD, &stats->rx_csum_bad_proto, 3689 "Received checksum offloaded buffer with incorrect protocol"); 3690 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 3691 CTLFLAG_RD, &stats->rx_csum_failed, 3692 "Received buffer checksum offload failed"); 3693 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 3694 CTLFLAG_RD, &stats->rx_csum_offloaded, 3695 "Received buffer checksum offload succeeded"); 3696 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 3697 CTLFLAG_RD, &stats->rx_task_rescheduled, 3698 "Times the receive interrupt task rescheduled itself"); 3699 3700 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype", 3701 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 3702 "Aborted transmit of checksum offloaded buffer with unknown " 3703 "Ethernet type"); 3704 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype", 3705 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 3706 "Aborted transmit of TSO buffer with unknown Ethernet type"); 3707 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 3708 CTLFLAG_RD, &stats->tx_tso_not_tcp, 3709 "Aborted transmit of TSO buffer with non TCP protocol"); 3710 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 3711 CTLFLAG_RD, &stats->tx_defragged, 3712 "Transmit mbufs defragged"); 3713 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 3714 CTLFLAG_RD, &stats->tx_defrag_failed, 3715 "Aborted transmit of buffer because defrag failed"); 3716 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 3717 CTLFLAG_RD, &stats->tx_csum_offloaded, 3718 "Offloaded checksum of transmitted buffer"); 3719 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 3720 CTLFLAG_RD, &stats->tx_tso_offloaded, 3721 "Segmentation offload of transmitted buffer"); 3722 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 3723 CTLFLAG_RD, &stats->tx_task_rescheduled, 3724 "Times the transmit interrupt task rescheduled itself"); 3725 } 3726 3727 static void 3728 vtnet_setup_sysctl(struct vtnet_softc *sc) 3729 { 3730 device_t dev; 3731 struct sysctl_ctx_list *ctx; 3732 struct sysctl_oid *tree; 3733 struct sysctl_oid_list *child; 3734 3735 dev = sc->vtnet_dev; 3736 ctx = device_get_sysctl_ctx(dev); 3737 tree = device_get_sysctl_tree(dev); 3738 child = SYSCTL_CHILDREN(tree); 3739 3740 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 3741 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 3742 "Maximum number of supported virtqueue pairs"); 3743 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 3744 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 3745 "Number of active virtqueue pairs"); 3746 3747 vtnet_setup_stat_sysctl(ctx, child, sc); 3748 } 3749 3750 static int 3751 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 3752 { 3753 3754 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 3755 } 3756 3757 static void 3758 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 3759 { 3760 3761 virtqueue_disable_intr(rxq->vtnrx_vq); 3762 } 3763 3764 static int 3765 vtnet_txq_enable_intr(struct vtnet_txq *txq) 3766 { 3767 3768 return (virtqueue_postpone_intr(txq->vtntx_vq, VQ_POSTPONE_LONG)); 3769 } 3770 3771 static void 3772 vtnet_txq_disable_intr(struct vtnet_txq *txq) 3773 { 3774 3775 virtqueue_disable_intr(txq->vtntx_vq); 3776 } 3777 3778 static void 3779 vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 3780 { 3781 int i; 3782 3783 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3784 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]); 3785 } 3786 3787 static void 3788 vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 3789 { 3790 int i; 3791 3792 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3793 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 3794 } 3795 3796 static void 3797 vtnet_enable_interrupts(struct vtnet_softc *sc) 3798 { 3799 3800 vtnet_enable_rx_interrupts(sc); 3801 vtnet_enable_tx_interrupts(sc); 3802 } 3803 3804 static void 3805 vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 3806 { 3807 int i; 3808 3809 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3810 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 3811 } 3812 3813 static void 3814 vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 3815 { 3816 int i; 3817 3818 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3819 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 3820 } 3821 3822 static void 3823 vtnet_disable_interrupts(struct vtnet_softc *sc) 3824 { 3825 3826 vtnet_disable_rx_interrupts(sc); 3827 vtnet_disable_tx_interrupts(sc); 3828 } 3829 3830 static int 3831 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 3832 { 3833 char path[64]; 3834 3835 snprintf(path, sizeof(path), 3836 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 3837 TUNABLE_INT_FETCH(path, &def); 3838 3839 return (def); 3840 } 3841