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/systm.h> 34#include <sys/kernel.h> 35#include <sys/sockio.h> 36#include <sys/mbuf.h> 37#include <sys/malloc.h> 38#include <sys/module.h> 39#include <sys/socket.h> 40#include <sys/sysctl.h> 41#include <sys/random.h> 42#include <sys/sglist.h> 43#include <sys/lock.h> 44#include <sys/mutex.h> 45#include <sys/taskqueue.h> 46#include <sys/smp.h> 47#include <machine/smp.h> 48 49#include <vm/uma.h> 50 51#include <net/ethernet.h> 52#include <net/if.h> 53#include <net/if_arp.h> 54#include <net/if_dl.h> 55#include <net/if_types.h> 56#include <net/if_media.h> 57#include <net/if_vlan_var.h> 58 59#include <net/bpf.h> 60 61#include <netinet/in_systm.h> 62#include <netinet/in.h> 63#include <netinet/ip.h> 64#include <netinet/ip6.h> 65#include <netinet6/ip6_var.h> 66#include <netinet/udp.h> 67#include <netinet/tcp.h> 68#include <netinet/sctp.h> 69 70#include <machine/bus.h> 71#include <machine/resource.h> 72#include <sys/bus.h> 73#include <sys/rman.h> 74 75#include <dev/virtio/virtio.h> 76#include <dev/virtio/virtqueue.h> 77#include <dev/virtio/network/virtio_net.h> 78#include <dev/virtio/network/if_vtnetvar.h> 79 80#include "virtio_if.h" 81 82#include "opt_inet.h" 83#include "opt_inet6.h" 84 85static int vtnet_modevent(module_t, int, void *); 86 87static int vtnet_probe(device_t); 88static int vtnet_attach(device_t); 89static int vtnet_detach(device_t); 90static int vtnet_suspend(device_t); 91static int vtnet_resume(device_t); 92static int vtnet_shutdown(device_t); 93static int vtnet_attach_completed(device_t); 94static int vtnet_config_change(device_t); 95 96static void vtnet_negotiate_features(struct vtnet_softc *); 97static void vtnet_setup_features(struct vtnet_softc *); 98static int vtnet_init_rxq(struct vtnet_softc *, int); 99static int vtnet_init_txq(struct vtnet_softc *, int); 100static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); 101static void vtnet_free_rxtx_queues(struct vtnet_softc *); 102static int vtnet_alloc_rx_filters(struct vtnet_softc *); 103static void vtnet_free_rx_filters(struct vtnet_softc *); 104static int vtnet_alloc_virtqueues(struct vtnet_softc *); 105static int vtnet_setup_interface(struct vtnet_softc *); 106static int vtnet_change_mtu(struct vtnet_softc *, int); 107static int vtnet_ioctl(struct ifnet *, u_long, caddr_t); 108 109static int vtnet_rxq_populate(struct vtnet_rxq *); 110static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); 111static struct mbuf * 112 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); 113static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *, 114 struct mbuf *, int); 115static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); 116static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); 117static int vtnet_rxq_new_buf(struct vtnet_rxq *); 118static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, 119 struct virtio_net_hdr *); 120static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); 121static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); 122static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); 123static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, 124 struct virtio_net_hdr *); 125static int vtnet_rxq_eof(struct vtnet_rxq *); 126static void vtnet_rx_vq_intr(void *); 127static void vtnet_rxq_tq_intr(void *, int); 128 129static int vtnet_txq_below_threshold(struct vtnet_txq *); 130static int vtnet_txq_notify(struct vtnet_txq *); 131static void vtnet_txq_free_mbufs(struct vtnet_txq *); 132static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, 133 int *, int *, int *); 134static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, 135 int, struct virtio_net_hdr *); 136static struct mbuf * 137 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, 138 struct virtio_net_hdr *); 139static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, 140 struct vtnet_tx_header *); 141static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **); 142#ifdef VTNET_LEGACY_TX 143static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *); 144static void vtnet_start(struct ifnet *); 145#else 146static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); 147static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *); 148static void vtnet_txq_tq_deferred(void *, int); 149#endif 150static void vtnet_txq_start(struct vtnet_txq *); 151static void vtnet_txq_tq_intr(void *, int); 152static int vtnet_txq_eof(struct vtnet_txq *); 153static void vtnet_tx_vq_intr(void *); 154static void vtnet_tx_start_all(struct vtnet_softc *); 155 156#ifndef VTNET_LEGACY_TX 157static void vtnet_qflush(struct ifnet *); 158#endif 159 160static int vtnet_watchdog(struct vtnet_txq *); 161static void vtnet_rxq_accum_stats(struct vtnet_rxq *, 162 struct vtnet_rxq_stats *); 163static void vtnet_txq_accum_stats(struct vtnet_txq *, 164 struct vtnet_txq_stats *); 165static void vtnet_accumulate_stats(struct vtnet_softc *); 166static void vtnet_tick(void *); 167 168static void vtnet_start_taskqueues(struct vtnet_softc *); 169static void vtnet_free_taskqueues(struct vtnet_softc *); 170static void vtnet_drain_taskqueues(struct vtnet_softc *); 171 172static void vtnet_drain_rxtx_queues(struct vtnet_softc *); 173static void vtnet_stop_rendezvous(struct vtnet_softc *); 174static void vtnet_stop(struct vtnet_softc *); 175static int vtnet_virtio_reinit(struct vtnet_softc *); 176static void vtnet_init_rx_filters(struct vtnet_softc *); 177static int vtnet_init_rx_queues(struct vtnet_softc *); 178static int vtnet_init_tx_queues(struct vtnet_softc *); 179static int vtnet_init_rxtx_queues(struct vtnet_softc *); 180static void vtnet_set_active_vq_pairs(struct vtnet_softc *); 181static int vtnet_reinit(struct vtnet_softc *); 182static void vtnet_init_locked(struct vtnet_softc *); 183static void vtnet_init(void *); 184 185static void vtnet_free_ctrl_vq(struct vtnet_softc *); 186static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 187 struct sglist *, int, int); 188static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 189static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); 190static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int); 191static int vtnet_set_promisc(struct vtnet_softc *, int); 192static int vtnet_set_allmulti(struct vtnet_softc *, int); 193static void vtnet_attach_disable_promisc(struct vtnet_softc *); 194static void vtnet_rx_filter(struct vtnet_softc *); 195static void vtnet_rx_filter_mac(struct vtnet_softc *); 196static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 197static void vtnet_rx_filter_vlan(struct vtnet_softc *); 198static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 199static void vtnet_register_vlan(void *, struct ifnet *, uint16_t); 200static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t); 201 202static int vtnet_is_link_up(struct vtnet_softc *); 203static void vtnet_update_link_status(struct vtnet_softc *); 204static int vtnet_ifmedia_upd(struct ifnet *); 205static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *); 206static void vtnet_get_hwaddr(struct vtnet_softc *); 207static void vtnet_set_hwaddr(struct vtnet_softc *); 208static void vtnet_vlan_tag_remove(struct mbuf *); 209static void vtnet_set_rx_process_limit(struct vtnet_softc *); 210static void vtnet_set_tx_intr_threshold(struct vtnet_softc *); 211 212static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, 213 struct sysctl_oid_list *, struct vtnet_rxq *); 214static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, 215 struct sysctl_oid_list *, struct vtnet_txq *); 216static void vtnet_setup_queue_sysctl(struct vtnet_softc *); 217static void vtnet_setup_sysctl(struct vtnet_softc *); 218 219static int vtnet_rxq_enable_intr(struct vtnet_rxq *); 220static void vtnet_rxq_disable_intr(struct vtnet_rxq *); 221static int vtnet_txq_enable_intr(struct vtnet_txq *); 222static void vtnet_txq_disable_intr(struct vtnet_txq *); 223static void vtnet_enable_rx_interrupts(struct vtnet_softc *); 224static void vtnet_enable_tx_interrupts(struct vtnet_softc *); 225static void vtnet_enable_interrupts(struct vtnet_softc *); 226static void vtnet_disable_rx_interrupts(struct vtnet_softc *); 227static void vtnet_disable_tx_interrupts(struct vtnet_softc *); 228static void vtnet_disable_interrupts(struct vtnet_softc *); 229 230static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); 231 232/* Tunables. */ 233static int vtnet_csum_disable = 0; 234TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable); 235static int vtnet_tso_disable = 0; 236TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable); 237static int vtnet_lro_disable = 0; 238TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable); 239static int vtnet_mq_disable = 0; 240TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable); 241static int vtnet_mq_max_pairs = 0; 242TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs); 243static int vtnet_rx_process_limit = 512; 244TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit); 245 246static uma_zone_t vtnet_tx_header_zone; 247 248static struct virtio_feature_desc vtnet_feature_desc[] = { 249 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 250 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 251 { VIRTIO_NET_F_MAC, "MacAddress" }, 252 { VIRTIO_NET_F_GSO, "TxAllGSO" }, 253 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" }, 254 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" }, 255 { VIRTIO_NET_F_GUEST_ECN, "RxECN" }, 256 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 257 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 258 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 259 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 260 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 261 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 262 { VIRTIO_NET_F_STATUS, "Status" }, 263 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" }, 264 { VIRTIO_NET_F_CTRL_RX, "RxMode" }, 265 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" }, 266 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" }, 267 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 268 { VIRTIO_NET_F_MQ, "Multiqueue" }, 269 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" }, 270 271 { 0, NULL } 272}; 273 274static device_method_t vtnet_methods[] = { 275 /* Device methods. */ 276 DEVMETHOD(device_probe, vtnet_probe), 277 DEVMETHOD(device_attach, vtnet_attach), 278 DEVMETHOD(device_detach, vtnet_detach), 279 DEVMETHOD(device_suspend, vtnet_suspend), 280 DEVMETHOD(device_resume, vtnet_resume), 281 DEVMETHOD(device_shutdown, vtnet_shutdown), 282 283 /* VirtIO methods. */ 284 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), 285 DEVMETHOD(virtio_config_change, vtnet_config_change), 286 287 DEVMETHOD_END 288}; 289 290#ifdef DEV_NETMAP 291#include <dev/netmap/if_vtnet_netmap.h> 292#endif /* DEV_NETMAP */ 293 294static driver_t vtnet_driver = { 295 "vtnet", 296 vtnet_methods, 297 sizeof(struct vtnet_softc) 298}; 299static devclass_t vtnet_devclass; 300 301DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass, 302 vtnet_modevent, 0); 303MODULE_VERSION(vtnet, 1); 304MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 305 306static int 307vtnet_modevent(module_t mod, int type, void *unused) 308{ 309 int error; 310 311 error = 0; 312 313 switch (type) { 314 case MOD_LOAD: 315 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", 316 sizeof(struct vtnet_tx_header), 317 NULL, NULL, NULL, NULL, 0, 0); 318 break; 319 case MOD_QUIESCE: 320 case MOD_UNLOAD: 321 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) 322 error = EBUSY; 323 else if (type == MOD_UNLOAD) { 324 uma_zdestroy(vtnet_tx_header_zone); 325 vtnet_tx_header_zone = NULL; 326 } 327 break; 328 case MOD_SHUTDOWN: 329 break; 330 default: 331 error = EOPNOTSUPP; 332 break; 333 } 334 335 return (error); 336} 337 338static int 339vtnet_probe(device_t dev) 340{ 341 342 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK) 343 return (ENXIO); 344 345 device_set_desc(dev, "VirtIO Networking Adapter"); 346 347 return (BUS_PROBE_DEFAULT); 348} 349 350static int 351vtnet_attach(device_t dev) 352{ 353 struct vtnet_softc *sc; 354 int error; 355 356 sc = device_get_softc(dev); 357 sc->vtnet_dev = dev; 358 359 /* Register our feature descriptions. */ 360 virtio_set_feature_desc(dev, vtnet_feature_desc); 361 362 VTNET_CORE_LOCK_INIT(sc); 363 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); 364 365 vtnet_setup_sysctl(sc); 366 vtnet_setup_features(sc); 367 368 error = vtnet_alloc_rx_filters(sc); 369 if (error) { 370 device_printf(dev, "cannot allocate Rx filters\n"); 371 goto fail; 372 } 373 374 error = vtnet_alloc_rxtx_queues(sc); 375 if (error) { 376 device_printf(dev, "cannot allocate queues\n"); 377 goto fail; 378 } 379 380 error = vtnet_alloc_virtqueues(sc); 381 if (error) { 382 device_printf(dev, "cannot allocate virtqueues\n"); 383 goto fail; 384 } 385 386 error = vtnet_setup_interface(sc); 387 if (error) { 388 device_printf(dev, "cannot setup interface\n"); 389 goto fail; 390 } 391 392 error = virtio_setup_intr(dev, INTR_TYPE_NET); 393 if (error) { 394 device_printf(dev, "cannot setup virtqueue interrupts\n"); 395 /* BMV: This will crash if during boot! */ 396 ether_ifdetach(sc->vtnet_ifp); 397 goto fail; 398 } 399 400#ifdef DEV_NETMAP 401 vtnet_netmap_attach(sc); 402#endif /* DEV_NETMAP */ 403 404 vtnet_start_taskqueues(sc); 405 406fail: 407 if (error) 408 vtnet_detach(dev); 409 410 return (error); 411} 412 413static int 414vtnet_detach(device_t dev) 415{ 416 struct vtnet_softc *sc; 417 struct ifnet *ifp; 418 419 sc = device_get_softc(dev); 420 ifp = sc->vtnet_ifp; 421 422 if (device_is_attached(dev)) { 423 VTNET_CORE_LOCK(sc); 424 vtnet_stop(sc); 425 VTNET_CORE_UNLOCK(sc); 426 427 callout_drain(&sc->vtnet_tick_ch); 428 vtnet_drain_taskqueues(sc); 429 430 ether_ifdetach(ifp); 431 } 432 433#ifdef DEV_NETMAP 434 netmap_detach(ifp); 435#endif /* DEV_NETMAP */ 436 437 vtnet_free_taskqueues(sc); 438 439 if (sc->vtnet_vlan_attach != NULL) { 440 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 441 sc->vtnet_vlan_attach = NULL; 442 } 443 if (sc->vtnet_vlan_detach != NULL) { 444 EVENTHANDLER_DEREGISTER(vlan_unconfg, sc->vtnet_vlan_detach); 445 sc->vtnet_vlan_detach = NULL; 446 } 447 448 ifmedia_removeall(&sc->vtnet_media); 449 450 if (ifp != NULL) { 451 if_free(ifp); 452 sc->vtnet_ifp = NULL; 453 } 454 455 vtnet_free_rxtx_queues(sc); 456 vtnet_free_rx_filters(sc); 457 458 if (sc->vtnet_ctrl_vq != NULL) 459 vtnet_free_ctrl_vq(sc); 460 461 VTNET_CORE_LOCK_DESTROY(sc); 462 463 return (0); 464} 465 466static int 467vtnet_suspend(device_t dev) 468{ 469 struct vtnet_softc *sc; 470 471 sc = device_get_softc(dev); 472 473 VTNET_CORE_LOCK(sc); 474 vtnet_stop(sc); 475 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 476 VTNET_CORE_UNLOCK(sc); 477 478 return (0); 479} 480 481static int 482vtnet_resume(device_t dev) 483{ 484 struct vtnet_softc *sc; 485 struct ifnet *ifp; 486 487 sc = device_get_softc(dev); 488 ifp = sc->vtnet_ifp; 489 490 VTNET_CORE_LOCK(sc); 491 if (ifp->if_flags & IFF_UP) 492 vtnet_init_locked(sc); 493 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 494 VTNET_CORE_UNLOCK(sc); 495 496 return (0); 497} 498 499static int 500vtnet_shutdown(device_t dev) 501{ 502 503 /* 504 * Suspend already does all of what we need to 505 * do here; we just never expect to be resumed. 506 */ 507 return (vtnet_suspend(dev)); 508} 509 510static int 511vtnet_attach_completed(device_t dev) 512{ 513 514 vtnet_attach_disable_promisc(device_get_softc(dev)); 515 516 return (0); 517} 518 519static int 520vtnet_config_change(device_t dev) 521{ 522 struct vtnet_softc *sc; 523 524 sc = device_get_softc(dev); 525 526 VTNET_CORE_LOCK(sc); 527 vtnet_update_link_status(sc); 528 if (sc->vtnet_link_active != 0) 529 vtnet_tx_start_all(sc); 530 VTNET_CORE_UNLOCK(sc); 531 532 return (0); 533} 534 535static void 536vtnet_negotiate_features(struct vtnet_softc *sc) 537{ 538 device_t dev; 539 uint64_t mask, features; 540 541 dev = sc->vtnet_dev; 542 mask = 0; 543 544 /* 545 * TSO and LRO are only available when their corresponding checksum 546 * offload feature is also negotiated. 547 */ 548 if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) { 549 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM; 550 mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES; 551 } 552 if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) 553 mask |= VTNET_TSO_FEATURES; 554 if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) 555 mask |= VTNET_LRO_FEATURES; 556#ifndef VTNET_LEGACY_TX 557 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 558 mask |= VIRTIO_NET_F_MQ; 559#else 560 mask |= VIRTIO_NET_F_MQ; 561#endif 562 563 features = VTNET_FEATURES & ~mask; 564 sc->vtnet_features = virtio_negotiate_features(dev, features); 565 566 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 567 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 568 /* 569 * LRO without mergeable buffers requires special care. This 570 * is not ideal because every receive buffer must be large 571 * enough to hold the maximum TCP packet, the Ethernet header, 572 * and the header. This requires up to 34 descriptors with 573 * MCLBYTES clusters. If we do not have indirect descriptors, 574 * LRO is disabled since the virtqueue will not contain very 575 * many receive buffers. 576 */ 577 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 578 device_printf(dev, 579 "LRO disabled due to both mergeable buffers and " 580 "indirect descriptors not negotiated\n"); 581 582 features &= ~VTNET_LRO_FEATURES; 583 sc->vtnet_features = 584 virtio_negotiate_features(dev, features); 585 } else 586 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 587 } 588} 589 590static void 591vtnet_setup_features(struct vtnet_softc *sc) 592{ 593 device_t dev; 594 int max_pairs, max; 595 596 dev = sc->vtnet_dev; 597 598 vtnet_negotiate_features(sc); 599 600 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 601 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 602 603 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 604 /* This feature should always be negotiated. */ 605 sc->vtnet_flags |= VTNET_FLAG_MAC; 606 } 607 608 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 609 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 610 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 611 } else 612 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 613 614 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 615 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS; 616 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 617 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS; 618 else 619 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS; 620 621 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 622 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 623 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 624 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS; 625 else 626 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS; 627 628 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 629 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 630 631 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 632 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 633 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 634 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 635 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 636 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 637 } 638 639 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) && 640 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 641 max_pairs = virtio_read_dev_config_2(dev, 642 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 643 if (max_pairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 644 max_pairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) 645 max_pairs = 1; 646 } else 647 max_pairs = 1; 648 649 if (max_pairs > 1) { 650 /* 651 * Limit the maximum number of queue pairs to the number of 652 * CPUs or the configured maximum. The actual number of 653 * queues that get used may be less. 654 */ 655 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 656 if (max > 0 && max_pairs > max) 657 max_pairs = max; 658 if (max_pairs > mp_ncpus) 659 max_pairs = mp_ncpus; 660 if (max_pairs > VTNET_MAX_QUEUE_PAIRS) 661 max_pairs = VTNET_MAX_QUEUE_PAIRS; 662 if (max_pairs > 1) 663 sc->vtnet_flags |= VTNET_FLAG_MULTIQ; 664 } 665 666 sc->vtnet_max_vq_pairs = max_pairs; 667} 668 669static int 670vtnet_init_rxq(struct vtnet_softc *sc, int id) 671{ 672 struct vtnet_rxq *rxq; 673 674 rxq = &sc->vtnet_rxqs[id]; 675 676 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 677 device_get_nameunit(sc->vtnet_dev), id); 678 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 679 680 rxq->vtnrx_sc = sc; 681 rxq->vtnrx_id = id; 682 683 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 684 if (rxq->vtnrx_sg == NULL) 685 return (ENOMEM); 686 687 TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 688 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 689 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 690 691 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 692} 693 694static int 695vtnet_init_txq(struct vtnet_softc *sc, int id) 696{ 697 struct vtnet_txq *txq; 698 699 txq = &sc->vtnet_txqs[id]; 700 701 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 702 device_get_nameunit(sc->vtnet_dev), id); 703 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 704 705 txq->vtntx_sc = sc; 706 txq->vtntx_id = id; 707 708 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 709 if (txq->vtntx_sg == NULL) 710 return (ENOMEM); 711 712#ifndef VTNET_LEGACY_TX 713 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 714 M_NOWAIT, &txq->vtntx_mtx); 715 if (txq->vtntx_br == NULL) 716 return (ENOMEM); 717 718 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 719#endif 720 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 721 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 722 taskqueue_thread_enqueue, &txq->vtntx_tq); 723 if (txq->vtntx_tq == NULL) 724 return (ENOMEM); 725 726 return (0); 727} 728 729static int 730vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 731{ 732 int i, npairs, error; 733 734 npairs = sc->vtnet_max_vq_pairs; 735 736 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 737 M_NOWAIT | M_ZERO); 738 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 739 M_NOWAIT | M_ZERO); 740 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 741 return (ENOMEM); 742 743 for (i = 0; i < npairs; i++) { 744 error = vtnet_init_rxq(sc, i); 745 if (error) 746 return (error); 747 error = vtnet_init_txq(sc, i); 748 if (error) 749 return (error); 750 } 751 752 vtnet_setup_queue_sysctl(sc); 753 754 return (0); 755} 756 757static void 758vtnet_destroy_rxq(struct vtnet_rxq *rxq) 759{ 760 761 rxq->vtnrx_sc = NULL; 762 rxq->vtnrx_id = -1; 763 764 if (rxq->vtnrx_sg != NULL) { 765 sglist_free(rxq->vtnrx_sg); 766 rxq->vtnrx_sg = NULL; 767 } 768 769 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 770 mtx_destroy(&rxq->vtnrx_mtx); 771} 772 773static void 774vtnet_destroy_txq(struct vtnet_txq *txq) 775{ 776 777 txq->vtntx_sc = NULL; 778 txq->vtntx_id = -1; 779 780 if (txq->vtntx_sg != NULL) { 781 sglist_free(txq->vtntx_sg); 782 txq->vtntx_sg = NULL; 783 } 784 785#ifndef VTNET_LEGACY_TX 786 if (txq->vtntx_br != NULL) { 787 buf_ring_free(txq->vtntx_br, M_DEVBUF); 788 txq->vtntx_br = NULL; 789 } 790#endif 791 792 if (mtx_initialized(&txq->vtntx_mtx) != 0) 793 mtx_destroy(&txq->vtntx_mtx); 794} 795 796static void 797vtnet_free_rxtx_queues(struct vtnet_softc *sc) 798{ 799 int i; 800 801 if (sc->vtnet_rxqs != NULL) { 802 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 803 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 804 free(sc->vtnet_rxqs, M_DEVBUF); 805 sc->vtnet_rxqs = NULL; 806 } 807 808 if (sc->vtnet_txqs != NULL) { 809 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 810 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 811 free(sc->vtnet_txqs, M_DEVBUF); 812 sc->vtnet_txqs = NULL; 813 } 814} 815 816static int 817vtnet_alloc_rx_filters(struct vtnet_softc *sc) 818{ 819 820 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 821 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 822 M_DEVBUF, M_NOWAIT | M_ZERO); 823 if (sc->vtnet_mac_filter == NULL) 824 return (ENOMEM); 825 } 826 827 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 828 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 829 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 830 if (sc->vtnet_vlan_filter == NULL) 831 return (ENOMEM); 832 } 833 834 return (0); 835} 836 837static void 838vtnet_free_rx_filters(struct vtnet_softc *sc) 839{ 840 841 if (sc->vtnet_mac_filter != NULL) { 842 free(sc->vtnet_mac_filter, M_DEVBUF); 843 sc->vtnet_mac_filter = NULL; 844 } 845 846 if (sc->vtnet_vlan_filter != NULL) { 847 free(sc->vtnet_vlan_filter, M_DEVBUF); 848 sc->vtnet_vlan_filter = NULL; 849 } 850} 851 852static int 853vtnet_alloc_virtqueues(struct vtnet_softc *sc) 854{ 855 device_t dev; 856 struct vq_alloc_info *info; 857 struct vtnet_rxq *rxq; 858 struct vtnet_txq *txq; 859 int i, idx, flags, nvqs, error; 860 861 dev = sc->vtnet_dev; 862 flags = 0; 863 864 nvqs = sc->vtnet_max_vq_pairs * 2; 865 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 866 nvqs++; 867 868 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); 869 if (info == NULL) 870 return (ENOMEM); 871 872 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) { 873 rxq = &sc->vtnet_rxqs[i]; 874 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, 875 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, 876 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id); 877 878 txq = &sc->vtnet_txqs[i]; 879 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs, 880 vtnet_tx_vq_intr, txq, &txq->vtntx_vq, 881 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id); 882 } 883 884 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 885 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, 886 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); 887 } 888 889 /* 890 * Enable interrupt binding if this is multiqueue. This only matters 891 * when per-vq MSIX is available. 892 */ 893 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ) 894 flags |= 0; 895 896 error = virtio_alloc_virtqueues(dev, flags, nvqs, info); 897 free(info, M_TEMP); 898 899 return (error); 900} 901 902static int 903vtnet_setup_interface(struct vtnet_softc *sc) 904{ 905 device_t dev; 906 struct ifnet *ifp; 907 908 dev = sc->vtnet_dev; 909 910 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER); 911 if (ifp == NULL) { 912 device_printf(dev, "cannot allocate ifnet structure\n"); 913 return (ENOSPC); 914 } 915 916 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 917 if_initbaudrate(ifp, IF_Gbps(10)); /* Approx. */ 918 ifp->if_softc = sc; 919 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 920 ifp->if_init = vtnet_init; 921 ifp->if_ioctl = vtnet_ioctl; 922 923#ifndef VTNET_LEGACY_TX 924 ifp->if_transmit = vtnet_txq_mq_start; 925 ifp->if_qflush = vtnet_qflush; 926#else 927 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 928 ifp->if_start = vtnet_start; 929 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1); 930 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1; 931 IFQ_SET_READY(&ifp->if_snd); 932#endif 933 934 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd, 935 vtnet_ifmedia_sts); 936 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL); 937 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE); 938 939 /* Read (or generate) the MAC address for the adapter. */ 940 vtnet_get_hwaddr(sc); 941 942 ether_ifattach(ifp, sc->vtnet_hwaddr); 943 944 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 945 ifp->if_capabilities |= IFCAP_LINKSTATE; 946 947 /* Tell the upper layer(s) we support long frames. */ 948 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header); 949 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU; 950 951 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 952 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6; 953 954 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) { 955 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6; 956 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 957 } else { 958 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 959 ifp->if_capabilities |= IFCAP_TSO4; 960 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 961 ifp->if_capabilities |= IFCAP_TSO6; 962 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 963 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 964 } 965 966 if (ifp->if_capabilities & IFCAP_TSO) 967 ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 968 } 969 970 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) 971 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6; 972 973 if (ifp->if_capabilities & IFCAP_HWCSUM) { 974 /* 975 * VirtIO does not support VLAN tagging, but we can fake 976 * it by inserting and removing the 802.1Q header during 977 * transmit and receive. We are then able to do checksum 978 * offloading of VLAN frames. 979 */ 980 ifp->if_capabilities |= 981 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; 982 } 983 984 ifp->if_capenable = ifp->if_capabilities; 985 986 /* 987 * Capabilities after here are not enabled by default. 988 */ 989 990 if (ifp->if_capabilities & IFCAP_RXCSUM) { 991 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) || 992 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6)) 993 ifp->if_capabilities |= IFCAP_LRO; 994 } 995 996 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 997 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 998 999 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1000 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1001 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1002 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1003 } 1004 1005 vtnet_set_rx_process_limit(sc); 1006 vtnet_set_tx_intr_threshold(sc); 1007 1008 return (0); 1009} 1010 1011static int 1012vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu) 1013{ 1014 struct ifnet *ifp; 1015 int frame_size, clsize; 1016 1017 ifp = sc->vtnet_ifp; 1018 1019 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU) 1020 return (EINVAL); 1021 1022 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) + 1023 new_mtu; 1024 1025 /* 1026 * Based on the new MTU (and hence frame size) determine which 1027 * cluster size is most appropriate for the receive queues. 1028 */ 1029 if (frame_size <= MCLBYTES) { 1030 clsize = MCLBYTES; 1031 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1032 /* Avoid going past 9K jumbos. */ 1033 if (frame_size > MJUM9BYTES) 1034 return (EINVAL); 1035 clsize = MJUM9BYTES; 1036 } else 1037 clsize = MJUMPAGESIZE; 1038 1039 ifp->if_mtu = new_mtu; 1040 sc->vtnet_rx_new_clsize = clsize; 1041 1042 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1043 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1044 vtnet_init_locked(sc); 1045 } 1046 1047 return (0); 1048} 1049 1050static int 1051vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1052{ 1053 struct vtnet_softc *sc; 1054 struct ifreq *ifr; 1055 int reinit, mask, error; 1056 1057 sc = ifp->if_softc; 1058 ifr = (struct ifreq *) data; 1059 error = 0; 1060 1061 switch (cmd) { 1062 case SIOCSIFMTU: 1063 if (ifp->if_mtu != ifr->ifr_mtu) { 1064 VTNET_CORE_LOCK(sc); 1065 error = vtnet_change_mtu(sc, ifr->ifr_mtu); 1066 VTNET_CORE_UNLOCK(sc); 1067 } 1068 break; 1069 1070 case SIOCSIFFLAGS: 1071 VTNET_CORE_LOCK(sc); 1072 if ((ifp->if_flags & IFF_UP) == 0) { 1073 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1074 vtnet_stop(sc); 1075 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1076 if ((ifp->if_flags ^ sc->vtnet_if_flags) & 1077 (IFF_PROMISC | IFF_ALLMULTI)) { 1078 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1079 vtnet_rx_filter(sc); 1080 else 1081 error = ENOTSUP; 1082 } 1083 } else 1084 vtnet_init_locked(sc); 1085 1086 if (error == 0) 1087 sc->vtnet_if_flags = ifp->if_flags; 1088 VTNET_CORE_UNLOCK(sc); 1089 break; 1090 1091 case SIOCADDMULTI: 1092 case SIOCDELMULTI: 1093 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) 1094 break; 1095 VTNET_CORE_LOCK(sc); 1096 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1097 vtnet_rx_filter_mac(sc); 1098 VTNET_CORE_UNLOCK(sc); 1099 break; 1100 1101 case SIOCSIFMEDIA: 1102 case SIOCGIFMEDIA: 1103 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1104 break; 1105 1106 case SIOCSIFCAP: 1107 VTNET_CORE_LOCK(sc); 1108 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1109 1110 if (mask & IFCAP_TXCSUM) 1111 ifp->if_capenable ^= IFCAP_TXCSUM; 1112 if (mask & IFCAP_TXCSUM_IPV6) 1113 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; 1114 if (mask & IFCAP_TSO4) 1115 ifp->if_capenable ^= IFCAP_TSO4; 1116 if (mask & IFCAP_TSO6) 1117 ifp->if_capenable ^= IFCAP_TSO6; 1118 1119 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO | 1120 IFCAP_VLAN_HWFILTER)) { 1121 /* These Rx features require us to renegotiate. */ 1122 reinit = 1; 1123 1124 if (mask & IFCAP_RXCSUM) 1125 ifp->if_capenable ^= IFCAP_RXCSUM; 1126 if (mask & IFCAP_RXCSUM_IPV6) 1127 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; 1128 if (mask & IFCAP_LRO) 1129 ifp->if_capenable ^= IFCAP_LRO; 1130 if (mask & IFCAP_VLAN_HWFILTER) 1131 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 1132 } else 1133 reinit = 0; 1134 1135 if (mask & IFCAP_VLAN_HWTSO) 1136 ifp->if_capenable ^= IFCAP_VLAN_HWTSO; 1137 if (mask & IFCAP_VLAN_HWTAGGING) 1138 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 1139 1140 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1141 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1142 vtnet_init_locked(sc); 1143 } 1144 1145 VTNET_CORE_UNLOCK(sc); 1146 VLAN_CAPABILITIES(ifp); 1147 1148 break; 1149 1150 default: 1151 error = ether_ioctl(ifp, cmd, data); 1152 break; 1153 } 1154 1155 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1156 1157 return (error); 1158} 1159 1160static int 1161vtnet_rxq_populate(struct vtnet_rxq *rxq) 1162{ 1163 struct virtqueue *vq; 1164 int nbufs, error; 1165 1166 vq = rxq->vtnrx_vq; 1167 error = ENOSPC; 1168 1169 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1170 error = vtnet_rxq_new_buf(rxq); 1171 if (error) 1172 break; 1173 } 1174 1175 if (nbufs > 0) { 1176 virtqueue_notify(vq); 1177 /* 1178 * EMSGSIZE signifies the virtqueue did not have enough 1179 * entries available to hold the last mbuf. This is not 1180 * an error. 1181 */ 1182 if (error == EMSGSIZE) 1183 error = 0; 1184 } 1185 1186 return (error); 1187} 1188 1189static void 1190vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1191{ 1192 struct virtqueue *vq; 1193 struct mbuf *m; 1194 int last; 1195 1196 vq = rxq->vtnrx_vq; 1197 last = 0; 1198 1199 while ((m = virtqueue_drain(vq, &last)) != NULL) 1200 m_freem(m); 1201 1202 KASSERT(virtqueue_empty(vq), 1203 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1204} 1205 1206static struct mbuf * 1207vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1208{ 1209 struct mbuf *m_head, *m_tail, *m; 1210 int i, clsize; 1211 1212 clsize = sc->vtnet_rx_clsize; 1213 1214 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1215 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs)); 1216 1217 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize); 1218 if (m_head == NULL) 1219 goto fail; 1220 1221 m_head->m_len = clsize; 1222 m_tail = m_head; 1223 1224 /* Allocate the rest of the chain. */ 1225 for (i = 1; i < nbufs; i++) { 1226 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize); 1227 if (m == NULL) 1228 goto fail; 1229 1230 m->m_len = clsize; 1231 m_tail->m_next = m; 1232 m_tail = m; 1233 } 1234 1235 if (m_tailp != NULL) 1236 *m_tailp = m_tail; 1237 1238 return (m_head); 1239 1240fail: 1241 sc->vtnet_stats.mbuf_alloc_failed++; 1242 m_freem(m_head); 1243 1244 return (NULL); 1245} 1246 1247/* 1248 * Slow path for when LRO without mergeable buffers is negotiated. 1249 */ 1250static int 1251vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1252 int len0) 1253{ 1254 struct vtnet_softc *sc; 1255 struct mbuf *m, *m_prev; 1256 struct mbuf *m_new, *m_tail; 1257 int len, clsize, nreplace, error; 1258 1259 sc = rxq->vtnrx_sc; 1260 clsize = sc->vtnet_rx_clsize; 1261 1262 m_prev = NULL; 1263 m_tail = NULL; 1264 nreplace = 0; 1265 1266 m = m0; 1267 len = len0; 1268 1269 /* 1270 * Since these mbuf chains are so large, we avoid allocating an 1271 * entire replacement chain if possible. When the received frame 1272 * did not consume the entire chain, the unused mbufs are moved 1273 * to the replacement chain. 1274 */ 1275 while (len > 0) { 1276 /* 1277 * Something is seriously wrong if we received a frame 1278 * larger than the chain. Drop it. 1279 */ 1280 if (m == NULL) { 1281 sc->vtnet_stats.rx_frame_too_large++; 1282 return (EMSGSIZE); 1283 } 1284 1285 /* We always allocate the same cluster size. */ 1286 KASSERT(m->m_len == clsize, 1287 ("%s: mbuf size %d is not the cluster size %d", 1288 __func__, m->m_len, clsize)); 1289 1290 m->m_len = MIN(m->m_len, len); 1291 len -= m->m_len; 1292 1293 m_prev = m; 1294 m = m->m_next; 1295 nreplace++; 1296 } 1297 1298 KASSERT(nreplace <= sc->vtnet_rx_nmbufs, 1299 ("%s: too many replacement mbufs %d max %d", __func__, nreplace, 1300 sc->vtnet_rx_nmbufs)); 1301 1302 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1303 if (m_new == NULL) { 1304 m_prev->m_len = clsize; 1305 return (ENOBUFS); 1306 } 1307 1308 /* 1309 * Move any unused mbufs from the received chain onto the end 1310 * of the new chain. 1311 */ 1312 if (m_prev->m_next != NULL) { 1313 m_tail->m_next = m_prev->m_next; 1314 m_prev->m_next = NULL; 1315 } 1316 1317 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1318 if (error) { 1319 /* 1320 * BAD! We could not enqueue the replacement mbuf chain. We 1321 * must restore the m0 chain to the original state if it was 1322 * modified so we can subsequently discard it. 1323 * 1324 * NOTE: The replacement is suppose to be an identical copy 1325 * to the one just dequeued so this is an unexpected error. 1326 */ 1327 sc->vtnet_stats.rx_enq_replacement_failed++; 1328 1329 if (m_tail->m_next != NULL) { 1330 m_prev->m_next = m_tail->m_next; 1331 m_tail->m_next = NULL; 1332 } 1333 1334 m_prev->m_len = clsize; 1335 m_freem(m_new); 1336 } 1337 1338 return (error); 1339} 1340 1341static int 1342vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1343{ 1344 struct vtnet_softc *sc; 1345 struct mbuf *m_new; 1346 int error; 1347 1348 sc = rxq->vtnrx_sc; 1349 1350 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1351 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1352 1353 if (m->m_next == NULL) { 1354 /* Fast-path for the common case of just one mbuf. */ 1355 if (m->m_len < len) 1356 return (EINVAL); 1357 1358 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1359 if (m_new == NULL) 1360 return (ENOBUFS); 1361 1362 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1363 if (error) { 1364 /* 1365 * The new mbuf is suppose to be an identical 1366 * copy of the one just dequeued so this is an 1367 * unexpected error. 1368 */ 1369 m_freem(m_new); 1370 sc->vtnet_stats.rx_enq_replacement_failed++; 1371 } else 1372 m->m_len = len; 1373 } else 1374 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len); 1375 1376 return (error); 1377} 1378 1379static int 1380vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1381{ 1382 struct vtnet_softc *sc; 1383 struct sglist *sg; 1384 struct vtnet_rx_header *rxhdr; 1385 uint8_t *mdata; 1386 int offset, error; 1387 1388 sc = rxq->vtnrx_sc; 1389 sg = rxq->vtnrx_sg; 1390 mdata = mtod(m, uint8_t *); 1391 1392 VTNET_RXQ_LOCK_ASSERT(rxq); 1393 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1394 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1395 KASSERT(m->m_len == sc->vtnet_rx_clsize, 1396 ("%s: unexpected cluster size %d/%d", __func__, m->m_len, 1397 sc->vtnet_rx_clsize)); 1398 1399 sglist_reset(sg); 1400 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1401 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1402 rxhdr = (struct vtnet_rx_header *) mdata; 1403 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1404 offset = sizeof(struct vtnet_rx_header); 1405 } else 1406 offset = 0; 1407 1408 sglist_append(sg, mdata + offset, m->m_len - offset); 1409 if (m->m_next != NULL) { 1410 error = sglist_append_mbuf(sg, m->m_next); 1411 MPASS(error == 0); 1412 } 1413 1414 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg); 1415 1416 return (error); 1417} 1418 1419static int 1420vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1421{ 1422 struct vtnet_softc *sc; 1423 struct mbuf *m; 1424 int error; 1425 1426 sc = rxq->vtnrx_sc; 1427 1428 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1429 if (m == NULL) 1430 return (ENOBUFS); 1431 1432 error = vtnet_rxq_enqueue_buf(rxq, m); 1433 if (error) 1434 m_freem(m); 1435 1436 return (error); 1437} 1438 1439/* 1440 * Use the checksum offset in the VirtIO header to set the 1441 * correct CSUM_* flags. 1442 */ 1443static int 1444vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m, 1445 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1446{ 1447 struct vtnet_softc *sc; 1448#if defined(INET) || defined(INET6) 1449 int offset = hdr->csum_start + hdr->csum_offset; 1450#endif 1451 1452 sc = rxq->vtnrx_sc; 1453 1454 /* Only do a basic sanity check on the offset. */ 1455 switch (eth_type) { 1456#if defined(INET) 1457 case ETHERTYPE_IP: 1458 if (__predict_false(offset < ip_start + sizeof(struct ip))) 1459 return (1); 1460 break; 1461#endif 1462#if defined(INET6) 1463 case ETHERTYPE_IPV6: 1464 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr))) 1465 return (1); 1466 break; 1467#endif 1468 default: 1469 sc->vtnet_stats.rx_csum_bad_ethtype++; 1470 return (1); 1471 } 1472 1473 /* 1474 * Use the offset to determine the appropriate CSUM_* flags. This is 1475 * a bit dirty, but we can get by with it since the checksum offsets 1476 * happen to be different. We assume the host host does not do IPv4 1477 * header checksum offloading. 1478 */ 1479 switch (hdr->csum_offset) { 1480 case offsetof(struct udphdr, uh_sum): 1481 case offsetof(struct tcphdr, th_sum): 1482 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1483 m->m_pkthdr.csum_data = 0xFFFF; 1484 break; 1485 case offsetof(struct sctphdr, checksum): 1486 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 1487 break; 1488 default: 1489 sc->vtnet_stats.rx_csum_bad_offset++; 1490 return (1); 1491 } 1492 1493 return (0); 1494} 1495 1496static int 1497vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m, 1498 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1499{ 1500 struct vtnet_softc *sc; 1501 int offset, proto; 1502 1503 sc = rxq->vtnrx_sc; 1504 1505 switch (eth_type) { 1506#if defined(INET) 1507 case ETHERTYPE_IP: { 1508 struct ip *ip; 1509 if (__predict_false(m->m_len < ip_start + sizeof(struct ip))) 1510 return (1); 1511 ip = (struct ip *)(m->m_data + ip_start); 1512 proto = ip->ip_p; 1513 offset = ip_start + (ip->ip_hl << 2); 1514 break; 1515 } 1516#endif 1517#if defined(INET6) 1518 case ETHERTYPE_IPV6: 1519 if (__predict_false(m->m_len < ip_start + 1520 sizeof(struct ip6_hdr))) 1521 return (1); 1522 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto); 1523 if (__predict_false(offset < 0)) 1524 return (1); 1525 break; 1526#endif 1527 default: 1528 sc->vtnet_stats.rx_csum_bad_ethtype++; 1529 return (1); 1530 } 1531 1532 switch (proto) { 1533 case IPPROTO_TCP: 1534 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) 1535 return (1); 1536 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1537 m->m_pkthdr.csum_data = 0xFFFF; 1538 break; 1539 case IPPROTO_UDP: 1540 if (__predict_false(m->m_len < offset + sizeof(struct udphdr))) 1541 return (1); 1542 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1543 m->m_pkthdr.csum_data = 0xFFFF; 1544 break; 1545 case IPPROTO_SCTP: 1546 if (__predict_false(m->m_len < offset + sizeof(struct sctphdr))) 1547 return (1); 1548 m->m_pkthdr.csum_flags |= CSUM_SCTP_VALID; 1549 break; 1550 default: 1551 /* 1552 * For the remaining protocols, FreeBSD does not support 1553 * checksum offloading, so the checksum will be recomputed. 1554 */ 1555#if 0 1556 if_printf(sc->vtnet_ifp, "cksum offload of unsupported " 1557 "protocol eth_type=%#x proto=%d csum_start=%d " 1558 "csum_offset=%d\n", __func__, eth_type, proto, 1559 hdr->csum_start, hdr->csum_offset); 1560#endif 1561 break; 1562 } 1563 1564 return (0); 1565} 1566 1567/* 1568 * Set the appropriate CSUM_* flags. Unfortunately, the information 1569 * provided is not directly useful to us. The VirtIO header gives the 1570 * offset of the checksum, which is all Linux needs, but this is not 1571 * how FreeBSD does things. We are forced to peek inside the packet 1572 * a bit. 1573 * 1574 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD 1575 * could accept the offsets and let the stack figure it out. 1576 */ 1577static int 1578vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1579 struct virtio_net_hdr *hdr) 1580{ 1581 struct ether_header *eh; 1582 struct ether_vlan_header *evh; 1583 uint16_t eth_type; 1584 int offset, error; 1585 1586 eh = mtod(m, struct ether_header *); 1587 eth_type = ntohs(eh->ether_type); 1588 if (eth_type == ETHERTYPE_VLAN) { 1589 /* BMV: We should handle nested VLAN tags too. */ 1590 evh = mtod(m, struct ether_vlan_header *); 1591 eth_type = ntohs(evh->evl_proto); 1592 offset = sizeof(struct ether_vlan_header); 1593 } else 1594 offset = sizeof(struct ether_header); 1595 1596 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1597 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr); 1598 else 1599 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr); 1600 1601 return (error); 1602} 1603 1604static void 1605vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1606{ 1607 struct mbuf *m; 1608 1609 while (--nbufs > 0) { 1610 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1611 if (m == NULL) 1612 break; 1613 vtnet_rxq_discard_buf(rxq, m); 1614 } 1615} 1616 1617static void 1618vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1619{ 1620 int error; 1621 1622 /* 1623 * Requeue the discarded mbuf. This should always be successful 1624 * since it was just dequeued. 1625 */ 1626 error = vtnet_rxq_enqueue_buf(rxq, m); 1627 KASSERT(error == 0, 1628 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1629} 1630 1631static int 1632vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1633{ 1634 struct vtnet_softc *sc; 1635 struct ifnet *ifp; 1636 struct virtqueue *vq; 1637 struct mbuf *m, *m_tail; 1638 int len; 1639 1640 sc = rxq->vtnrx_sc; 1641 vq = rxq->vtnrx_vq; 1642 ifp = sc->vtnet_ifp; 1643 m_tail = m_head; 1644 1645 while (--nbufs > 0) { 1646 m = virtqueue_dequeue(vq, &len); 1647 if (m == NULL) { 1648 rxq->vtnrx_stats.vrxs_ierrors++; 1649 goto fail; 1650 } 1651 1652 if (vtnet_rxq_new_buf(rxq) != 0) { 1653 rxq->vtnrx_stats.vrxs_iqdrops++; 1654 vtnet_rxq_discard_buf(rxq, m); 1655 if (nbufs > 1) 1656 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1657 goto fail; 1658 } 1659 1660 if (m->m_len < len) 1661 len = m->m_len; 1662 1663 m->m_len = len; 1664 m->m_flags &= ~M_PKTHDR; 1665 1666 m_head->m_pkthdr.len += len; 1667 m_tail->m_next = m; 1668 m_tail = m; 1669 } 1670 1671 return (0); 1672 1673fail: 1674 sc->vtnet_stats.rx_mergeable_failed++; 1675 m_freem(m_head); 1676 1677 return (1); 1678} 1679 1680static void 1681vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 1682 struct virtio_net_hdr *hdr) 1683{ 1684 struct vtnet_softc *sc; 1685 struct ifnet *ifp; 1686 struct ether_header *eh; 1687 1688 sc = rxq->vtnrx_sc; 1689 ifp = sc->vtnet_ifp; 1690 1691 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { 1692 eh = mtod(m, struct ether_header *); 1693 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1694 vtnet_vlan_tag_remove(m); 1695 /* 1696 * With the 802.1Q header removed, update the 1697 * checksum starting location accordingly. 1698 */ 1699 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1700 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 1701 } 1702 } 1703 1704 m->m_pkthdr.flowid = rxq->vtnrx_id; 1705 m->m_flags |= M_FLOWID; 1706 1707 /* 1708 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum 1709 * distinction that Linux does. Need to reevaluate if performing 1710 * offloading for the NEEDS_CSUM case is really appropriate. 1711 */ 1712 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM | 1713 VIRTIO_NET_HDR_F_DATA_VALID)) { 1714 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 1715 rxq->vtnrx_stats.vrxs_csum++; 1716 else 1717 rxq->vtnrx_stats.vrxs_csum_failed++; 1718 } 1719 1720 rxq->vtnrx_stats.vrxs_ipackets++; 1721 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 1722 1723 VTNET_RXQ_UNLOCK(rxq); 1724 (*ifp->if_input)(ifp, m); 1725 VTNET_RXQ_LOCK(rxq); 1726} 1727 1728static int 1729vtnet_rxq_eof(struct vtnet_rxq *rxq) 1730{ 1731 struct virtio_net_hdr lhdr, *hdr; 1732 struct vtnet_softc *sc; 1733 struct ifnet *ifp; 1734 struct virtqueue *vq; 1735 struct mbuf *m; 1736 struct virtio_net_hdr_mrg_rxbuf *mhdr; 1737 int len, deq, nbufs, adjsz, count; 1738 1739 sc = rxq->vtnrx_sc; 1740 vq = rxq->vtnrx_vq; 1741 ifp = sc->vtnet_ifp; 1742 hdr = &lhdr; 1743 deq = 0; 1744 count = sc->vtnet_rx_process_limit; 1745 1746 VTNET_RXQ_LOCK_ASSERT(rxq); 1747 1748#ifdef DEV_NETMAP 1749 if (netmap_rx_irq(ifp, 0, &deq)) { 1750 return (FALSE); 1751 } 1752#endif /* DEV_NETMAP */ 1753 1754 while (count-- > 0) { 1755 m = virtqueue_dequeue(vq, &len); 1756 if (m == NULL) 1757 break; 1758 deq++; 1759 1760 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 1761 rxq->vtnrx_stats.vrxs_ierrors++; 1762 vtnet_rxq_discard_buf(rxq, m); 1763 continue; 1764 } 1765 1766 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1767 nbufs = 1; 1768 adjsz = sizeof(struct vtnet_rx_header); 1769 /* 1770 * Account for our pad inserted between the header 1771 * and the actual start of the frame. 1772 */ 1773 len += VTNET_RX_HEADER_PAD; 1774 } else { 1775 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 1776 nbufs = mhdr->num_buffers; 1777 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1778 } 1779 1780 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 1781 rxq->vtnrx_stats.vrxs_iqdrops++; 1782 vtnet_rxq_discard_buf(rxq, m); 1783 if (nbufs > 1) 1784 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1785 continue; 1786 } 1787 1788 m->m_pkthdr.len = len; 1789 m->m_pkthdr.rcvif = ifp; 1790 m->m_pkthdr.csum_flags = 0; 1791 1792 if (nbufs > 1) { 1793 /* Dequeue the rest of chain. */ 1794 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 1795 continue; 1796 } 1797 1798 /* 1799 * Save copy of header before we strip it. For both mergeable 1800 * and non-mergeable, the header is at the beginning of the 1801 * mbuf data. We no longer need num_buffers, so always use a 1802 * regular header. 1803 * 1804 * BMV: Is this memcpy() expensive? We know the mbuf data is 1805 * still valid even after the m_adj(). 1806 */ 1807 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr)); 1808 m_adj(m, adjsz); 1809 1810 vtnet_rxq_input(rxq, m, hdr); 1811 1812 /* Must recheck after dropping the Rx lock. */ 1813 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1814 break; 1815 } 1816 1817 if (deq > 0) 1818 virtqueue_notify(vq); 1819 1820 return (count > 0 ? 0 : EAGAIN); 1821} 1822 1823static void 1824vtnet_rx_vq_intr(void *xrxq) 1825{ 1826 struct vtnet_softc *sc; 1827 struct vtnet_rxq *rxq; 1828 struct ifnet *ifp; 1829 int tries, more; 1830 1831 rxq = xrxq; 1832 sc = rxq->vtnrx_sc; 1833 ifp = sc->vtnet_ifp; 1834 tries = 0; 1835 1836 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 1837 /* 1838 * Ignore this interrupt. Either this is a spurious interrupt 1839 * or multiqueue without per-VQ MSIX so every queue needs to 1840 * be polled (a brain dead configuration we could try harder 1841 * to avoid). 1842 */ 1843 vtnet_rxq_disable_intr(rxq); 1844 return; 1845 } 1846 1847 VTNET_RXQ_LOCK(rxq); 1848 1849again: 1850 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1851 VTNET_RXQ_UNLOCK(rxq); 1852 return; 1853 } 1854 1855 more = vtnet_rxq_eof(rxq); 1856 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1857 if (!more) 1858 vtnet_rxq_disable_intr(rxq); 1859 /* 1860 * This is an occasional condition or race (when !more), 1861 * so retry a few times before scheduling the taskqueue. 1862 */ 1863 if (tries++ < VTNET_INTR_DISABLE_RETRIES) 1864 goto again; 1865 1866 VTNET_RXQ_UNLOCK(rxq); 1867 rxq->vtnrx_stats.vrxs_rescheduled++; 1868 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1869 } else 1870 VTNET_RXQ_UNLOCK(rxq); 1871} 1872 1873static void 1874vtnet_rxq_tq_intr(void *xrxq, int pending) 1875{ 1876 struct vtnet_softc *sc; 1877 struct vtnet_rxq *rxq; 1878 struct ifnet *ifp; 1879 int more; 1880 1881 rxq = xrxq; 1882 sc = rxq->vtnrx_sc; 1883 ifp = sc->vtnet_ifp; 1884 1885 VTNET_RXQ_LOCK(rxq); 1886 1887 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1888 VTNET_RXQ_UNLOCK(rxq); 1889 return; 1890 } 1891 1892 more = vtnet_rxq_eof(rxq); 1893 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1894 if (!more) 1895 vtnet_rxq_disable_intr(rxq); 1896 rxq->vtnrx_stats.vrxs_rescheduled++; 1897 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1898 } 1899 1900 VTNET_RXQ_UNLOCK(rxq); 1901} 1902 1903static int 1904vtnet_txq_below_threshold(struct vtnet_txq *txq) 1905{ 1906 struct vtnet_softc *sc; 1907 struct virtqueue *vq; 1908 1909 sc = txq->vtntx_sc; 1910 vq = txq->vtntx_vq; 1911 1912 return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh); 1913} 1914 1915static int 1916vtnet_txq_notify(struct vtnet_txq *txq) 1917{ 1918 struct virtqueue *vq; 1919 1920 vq = txq->vtntx_vq; 1921 1922 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 1923 virtqueue_notify(vq); 1924 1925 if (vtnet_txq_enable_intr(txq) == 0) 1926 return (0); 1927 1928 /* 1929 * Drain frames that were completed since last checked. If this 1930 * causes the queue to go above the threshold, the caller should 1931 * continue transmitting. 1932 */ 1933 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) { 1934 virtqueue_disable_intr(vq); 1935 return (1); 1936 } 1937 1938 return (0); 1939} 1940 1941static void 1942vtnet_txq_free_mbufs(struct vtnet_txq *txq) 1943{ 1944 struct virtqueue *vq; 1945 struct vtnet_tx_header *txhdr; 1946 int last; 1947 1948 vq = txq->vtntx_vq; 1949 last = 0; 1950 1951 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 1952 m_freem(txhdr->vth_mbuf); 1953 uma_zfree(vtnet_tx_header_zone, txhdr); 1954 } 1955 1956 KASSERT(virtqueue_empty(vq), 1957 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 1958} 1959 1960/* 1961 * BMV: Much of this can go away once we finally have offsets in 1962 * the mbuf packet header. Bug andre@. 1963 */ 1964static int 1965vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, 1966 int *etype, int *proto, int *start) 1967{ 1968 struct vtnet_softc *sc; 1969 struct ether_vlan_header *evh; 1970 int offset; 1971 1972 sc = txq->vtntx_sc; 1973 1974 evh = mtod(m, struct ether_vlan_header *); 1975 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 1976 /* BMV: We should handle nested VLAN tags too. */ 1977 *etype = ntohs(evh->evl_proto); 1978 offset = sizeof(struct ether_vlan_header); 1979 } else { 1980 *etype = ntohs(evh->evl_encap_proto); 1981 offset = sizeof(struct ether_header); 1982 } 1983 1984 switch (*etype) { 1985#if defined(INET) 1986 case ETHERTYPE_IP: { 1987 struct ip *ip, iphdr; 1988 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 1989 m_copydata(m, offset, sizeof(struct ip), 1990 (caddr_t) &iphdr); 1991 ip = &iphdr; 1992 } else 1993 ip = (struct ip *)(m->m_data + offset); 1994 *proto = ip->ip_p; 1995 *start = offset + (ip->ip_hl << 2); 1996 break; 1997 } 1998#endif 1999#if defined(INET6) 2000 case ETHERTYPE_IPV6: 2001 *proto = -1; 2002 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 2003 /* Assert the network stack sent us a valid packet. */ 2004 KASSERT(*start > offset, 2005 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 2006 *start, offset, *proto)); 2007 break; 2008#endif 2009 default: 2010 sc->vtnet_stats.tx_csum_bad_ethtype++; 2011 return (EINVAL); 2012 } 2013 2014 return (0); 2015} 2016 2017static int 2018vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 2019 int offset, struct virtio_net_hdr *hdr) 2020{ 2021 static struct timeval lastecn; 2022 static int curecn; 2023 struct vtnet_softc *sc; 2024 struct tcphdr *tcp, tcphdr; 2025 2026 sc = txq->vtntx_sc; 2027 2028 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 2029 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 2030 tcp = &tcphdr; 2031 } else 2032 tcp = (struct tcphdr *)(m->m_data + offset); 2033 2034 hdr->hdr_len = offset + (tcp->th_off << 2); 2035 hdr->gso_size = m->m_pkthdr.tso_segsz; 2036 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 2037 VIRTIO_NET_HDR_GSO_TCPV6; 2038 2039 if (tcp->th_flags & TH_CWR) { 2040 /* 2041 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD, 2042 * ECN support is not on a per-interface basis, but globally via 2043 * the net.inet.tcp.ecn.enable sysctl knob. The default is off. 2044 */ 2045 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2046 if (ppsratecheck(&lastecn, &curecn, 1)) 2047 if_printf(sc->vtnet_ifp, 2048 "TSO with ECN not negotiated with host\n"); 2049 return (ENOTSUP); 2050 } 2051 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2052 } 2053 2054 txq->vtntx_stats.vtxs_tso++; 2055 2056 return (0); 2057} 2058 2059static struct mbuf * 2060vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2061 struct virtio_net_hdr *hdr) 2062{ 2063 struct vtnet_softc *sc; 2064 int flags, etype, csum_start, proto, error; 2065 2066 sc = txq->vtntx_sc; 2067 flags = m->m_pkthdr.csum_flags; 2068 2069 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2070 if (error) 2071 goto drop; 2072 2073 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) || 2074 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) { 2075 /* 2076 * We could compare the IP protocol vs the CSUM_ flag too, 2077 * but that really should not be necessary. 2078 */ 2079 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2080 hdr->csum_start = csum_start; 2081 hdr->csum_offset = m->m_pkthdr.csum_data; 2082 txq->vtntx_stats.vtxs_csum++; 2083 } 2084 2085 if (flags & CSUM_TSO) { 2086 if (__predict_false(proto != IPPROTO_TCP)) { 2087 /* Likely failed to correctly parse the mbuf. */ 2088 sc->vtnet_stats.tx_tso_not_tcp++; 2089 goto drop; 2090 } 2091 2092 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM, 2093 ("%s: mbuf %p TSO without checksum offload %#x", 2094 __func__, m, flags)); 2095 2096 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2097 if (error) 2098 goto drop; 2099 } 2100 2101 return (m); 2102 2103drop: 2104 m_freem(m); 2105 return (NULL); 2106} 2107 2108static int 2109vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2110 struct vtnet_tx_header *txhdr) 2111{ 2112 struct vtnet_softc *sc; 2113 struct virtqueue *vq; 2114 struct sglist *sg; 2115 struct mbuf *m; 2116 int error; 2117 2118 sc = txq->vtntx_sc; 2119 vq = txq->vtntx_vq; 2120 sg = txq->vtntx_sg; 2121 m = *m_head; 2122 2123 sglist_reset(sg); 2124 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2125 KASSERT(error == 0 && sg->sg_nseg == 1, 2126 ("%s: error %d adding header to sglist", __func__, error)); 2127 2128 error = sglist_append_mbuf(sg, m); 2129 if (error) { 2130 m = m_defrag(m, M_NOWAIT); 2131 if (m == NULL) 2132 goto fail; 2133 2134 *m_head = m; 2135 sc->vtnet_stats.tx_defragged++; 2136 2137 error = sglist_append_mbuf(sg, m); 2138 if (error) 2139 goto fail; 2140 } 2141 2142 txhdr->vth_mbuf = m; 2143 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2144 2145 return (error); 2146 2147fail: 2148 sc->vtnet_stats.tx_defrag_failed++; 2149 m_freem(*m_head); 2150 *m_head = NULL; 2151 2152 return (ENOBUFS); 2153} 2154 2155static int 2156vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head) 2157{ 2158 struct vtnet_tx_header *txhdr; 2159 struct virtio_net_hdr *hdr; 2160 struct mbuf *m; 2161 int error; 2162 2163 m = *m_head; 2164 M_ASSERTPKTHDR(m); 2165 2166 txhdr = uma_zalloc(vtnet_tx_header_zone, M_NOWAIT | M_ZERO); 2167 if (txhdr == NULL) { 2168 m_freem(m); 2169 *m_head = NULL; 2170 return (ENOMEM); 2171 } 2172 2173 /* 2174 * Always use the non-mergeable header, regardless if the feature 2175 * was negotiated. For transmit, num_buffers is always zero. The 2176 * vtnet_hdr_size is used to enqueue the correct header size. 2177 */ 2178 hdr = &txhdr->vth_uhdr.hdr; 2179 2180 if (m->m_flags & M_VLANTAG) { 2181 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2182 if ((*m_head = m) == NULL) { 2183 error = ENOBUFS; 2184 goto fail; 2185 } 2186 m->m_flags &= ~M_VLANTAG; 2187 } 2188 2189 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2190 m = vtnet_txq_offload(txq, m, hdr); 2191 if ((*m_head = m) == NULL) { 2192 error = ENOBUFS; 2193 goto fail; 2194 } 2195 } 2196 2197 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2198 if (error == 0) 2199 return (0); 2200 2201fail: 2202 uma_zfree(vtnet_tx_header_zone, txhdr); 2203 2204 return (error); 2205} 2206 2207#ifdef VTNET_LEGACY_TX 2208 2209static void 2210vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp) 2211{ 2212 struct vtnet_softc *sc; 2213 struct virtqueue *vq; 2214 struct mbuf *m0; 2215 int tries, enq; 2216 2217 sc = txq->vtntx_sc; 2218 vq = txq->vtntx_vq; 2219 tries = 0; 2220 2221 VTNET_TXQ_LOCK_ASSERT(txq); 2222 2223 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2224 sc->vtnet_link_active == 0) 2225 return; 2226 2227 vtnet_txq_eof(txq); 2228 2229again: 2230 enq = 0; 2231 2232 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { 2233 if (virtqueue_full(vq)) 2234 break; 2235 2236 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0); 2237 if (m0 == NULL) 2238 break; 2239 2240 if (vtnet_txq_encap(txq, &m0) != 0) { 2241 if (m0 != NULL) 2242 IFQ_DRV_PREPEND(&ifp->if_snd, m0); 2243 break; 2244 } 2245 2246 enq++; 2247 ETHER_BPF_MTAP(ifp, m0); 2248 } 2249 2250 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2251 if (tries++ < VTNET_NOTIFY_RETRIES) 2252 goto again; 2253 2254 txq->vtntx_stats.vtxs_rescheduled++; 2255 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2256 } 2257} 2258 2259static void 2260vtnet_start(struct ifnet *ifp) 2261{ 2262 struct vtnet_softc *sc; 2263 struct vtnet_txq *txq; 2264 2265 sc = ifp->if_softc; 2266 txq = &sc->vtnet_txqs[0]; 2267 2268 VTNET_TXQ_LOCK(txq); 2269 vtnet_start_locked(txq, ifp); 2270 VTNET_TXQ_UNLOCK(txq); 2271} 2272 2273#else /* !VTNET_LEGACY_TX */ 2274 2275static int 2276vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2277{ 2278 struct vtnet_softc *sc; 2279 struct virtqueue *vq; 2280 struct buf_ring *br; 2281 struct ifnet *ifp; 2282 int enq, tries, error; 2283 2284 sc = txq->vtntx_sc; 2285 vq = txq->vtntx_vq; 2286 br = txq->vtntx_br; 2287 ifp = sc->vtnet_ifp; 2288 tries = 0; 2289 error = 0; 2290 2291 VTNET_TXQ_LOCK_ASSERT(txq); 2292 2293 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2294 sc->vtnet_link_active == 0) { 2295 if (m != NULL) 2296 error = drbr_enqueue(ifp, br, m); 2297 return (error); 2298 } 2299 2300 if (m != NULL) { 2301 error = drbr_enqueue(ifp, br, m); 2302 if (error) 2303 return (error); 2304 } 2305 2306 vtnet_txq_eof(txq); 2307 2308again: 2309 enq = 0; 2310 2311 while ((m = drbr_peek(ifp, br)) != NULL) { 2312 if (virtqueue_full(vq)) { 2313 drbr_putback(ifp, br, m); 2314 break; 2315 } 2316 2317 if (vtnet_txq_encap(txq, &m) != 0) { 2318 if (m != NULL) 2319 drbr_putback(ifp, br, m); 2320 else 2321 drbr_advance(ifp, br); 2322 break; 2323 } 2324 drbr_advance(ifp, br); 2325 2326 enq++; 2327 ETHER_BPF_MTAP(ifp, m); 2328 } 2329 2330 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2331 if (tries++ < VTNET_NOTIFY_RETRIES) 2332 goto again; 2333 2334 txq->vtntx_stats.vtxs_rescheduled++; 2335 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2336 } 2337 2338 return (0); 2339} 2340 2341static int 2342vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m) 2343{ 2344 struct vtnet_softc *sc; 2345 struct vtnet_txq *txq; 2346 int i, npairs, error; 2347 2348 sc = ifp->if_softc; 2349 npairs = sc->vtnet_act_vq_pairs; 2350 2351 if (m->m_flags & M_FLOWID) 2352 i = m->m_pkthdr.flowid % npairs; 2353 else 2354 i = curcpu % npairs; 2355 2356 txq = &sc->vtnet_txqs[i]; 2357 2358 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2359 error = vtnet_txq_mq_start_locked(txq, m); 2360 VTNET_TXQ_UNLOCK(txq); 2361 } else { 2362 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2363 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2364 } 2365 2366 return (error); 2367} 2368 2369static void 2370vtnet_txq_tq_deferred(void *xtxq, int pending) 2371{ 2372 struct vtnet_softc *sc; 2373 struct vtnet_txq *txq; 2374 2375 txq = xtxq; 2376 sc = txq->vtntx_sc; 2377 2378 VTNET_TXQ_LOCK(txq); 2379 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2380 vtnet_txq_mq_start_locked(txq, NULL); 2381 VTNET_TXQ_UNLOCK(txq); 2382} 2383 2384#endif /* VTNET_LEGACY_TX */ 2385 2386static void 2387vtnet_txq_start(struct vtnet_txq *txq) 2388{ 2389 struct vtnet_softc *sc; 2390 struct ifnet *ifp; 2391 2392 sc = txq->vtntx_sc; 2393 ifp = sc->vtnet_ifp; 2394 2395#ifdef VTNET_LEGACY_TX 2396 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 2397 vtnet_start_locked(txq, ifp); 2398#else 2399 if (!drbr_empty(ifp, txq->vtntx_br)) 2400 vtnet_txq_mq_start_locked(txq, NULL); 2401#endif 2402} 2403 2404static void 2405vtnet_txq_tq_intr(void *xtxq, int pending) 2406{ 2407 struct vtnet_softc *sc; 2408 struct vtnet_txq *txq; 2409 struct ifnet *ifp; 2410 2411 txq = xtxq; 2412 sc = txq->vtntx_sc; 2413 ifp = sc->vtnet_ifp; 2414 2415 VTNET_TXQ_LOCK(txq); 2416 2417 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2418 VTNET_TXQ_UNLOCK(txq); 2419 return; 2420 } 2421 2422 vtnet_txq_eof(txq); 2423 vtnet_txq_start(txq); 2424 2425 VTNET_TXQ_UNLOCK(txq); 2426} 2427 2428static int 2429vtnet_txq_eof(struct vtnet_txq *txq) 2430{ 2431 struct virtqueue *vq; 2432 struct vtnet_tx_header *txhdr; 2433 struct mbuf *m; 2434 int deq; 2435 2436 vq = txq->vtntx_vq; 2437 deq = 0; 2438 VTNET_TXQ_LOCK_ASSERT(txq); 2439 2440#ifdef DEV_NETMAP 2441 if (netmap_tx_irq(txq->vtntx_sc->vtnet_ifp, txq->vtntx_id)) { 2442 virtqueue_disable_intr(vq); // XXX luigi 2443 return 0; // XXX or 1 ? 2444 } 2445#endif /* DEV_NETMAP */ 2446 2447 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2448 m = txhdr->vth_mbuf; 2449 deq++; 2450 2451 txq->vtntx_stats.vtxs_opackets++; 2452 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2453 if (m->m_flags & M_MCAST) 2454 txq->vtntx_stats.vtxs_omcasts++; 2455 2456 m_freem(m); 2457 uma_zfree(vtnet_tx_header_zone, txhdr); 2458 } 2459 2460 if (virtqueue_empty(vq)) 2461 txq->vtntx_watchdog = 0; 2462 2463 return (deq); 2464} 2465 2466static void 2467vtnet_tx_vq_intr(void *xtxq) 2468{ 2469 struct vtnet_softc *sc; 2470 struct vtnet_txq *txq; 2471 struct ifnet *ifp; 2472 2473 txq = xtxq; 2474 sc = txq->vtntx_sc; 2475 ifp = sc->vtnet_ifp; 2476 2477 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2478 /* 2479 * Ignore this interrupt. Either this is a spurious interrupt 2480 * or multiqueue without per-VQ MSIX so every queue needs to 2481 * be polled (a brain dead configuration we could try harder 2482 * to avoid). 2483 */ 2484 vtnet_txq_disable_intr(txq); 2485 return; 2486 } 2487 2488 VTNET_TXQ_LOCK(txq); 2489 2490 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2491 VTNET_TXQ_UNLOCK(txq); 2492 return; 2493 } 2494 2495 vtnet_txq_eof(txq); 2496 vtnet_txq_start(txq); 2497 2498 VTNET_TXQ_UNLOCK(txq); 2499} 2500 2501static void 2502vtnet_tx_start_all(struct vtnet_softc *sc) 2503{ 2504 struct vtnet_txq *txq; 2505 int i; 2506 2507 VTNET_CORE_LOCK_ASSERT(sc); 2508 2509 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2510 txq = &sc->vtnet_txqs[i]; 2511 2512 VTNET_TXQ_LOCK(txq); 2513 vtnet_txq_start(txq); 2514 VTNET_TXQ_UNLOCK(txq); 2515 } 2516} 2517 2518#ifndef VTNET_LEGACY_TX 2519static void 2520vtnet_qflush(struct ifnet *ifp) 2521{ 2522 struct vtnet_softc *sc; 2523 struct vtnet_txq *txq; 2524 struct mbuf *m; 2525 int i; 2526 2527 sc = ifp->if_softc; 2528 2529 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2530 txq = &sc->vtnet_txqs[i]; 2531 2532 VTNET_TXQ_LOCK(txq); 2533 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2534 m_freem(m); 2535 VTNET_TXQ_UNLOCK(txq); 2536 } 2537 2538 if_qflush(ifp); 2539} 2540#endif 2541 2542static int 2543vtnet_watchdog(struct vtnet_txq *txq) 2544{ 2545 struct ifnet *ifp; 2546 2547 ifp = txq->vtntx_sc->vtnet_ifp; 2548 2549 VTNET_TXQ_LOCK(txq); 2550 if (txq->vtntx_watchdog == 1) { 2551 /* 2552 * Only drain completed frames if the watchdog is about to 2553 * expire. If any frames were drained, there may be enough 2554 * free descriptors now available to transmit queued frames. 2555 * In that case, the timer will immediately be decremented 2556 * below, but the timeout is generous enough that should not 2557 * be a problem. 2558 */ 2559 if (vtnet_txq_eof(txq) != 0) 2560 vtnet_txq_start(txq); 2561 } 2562 2563 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 2564 VTNET_TXQ_UNLOCK(txq); 2565 return (0); 2566 } 2567 VTNET_TXQ_UNLOCK(txq); 2568 2569 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id); 2570 return (1); 2571} 2572 2573static void 2574vtnet_rxq_accum_stats(struct vtnet_rxq *rxq, struct vtnet_rxq_stats *accum) 2575{ 2576 struct vtnet_rxq_stats *st; 2577 2578 st = &rxq->vtnrx_stats; 2579 2580 accum->vrxs_ipackets += st->vrxs_ipackets; 2581 accum->vrxs_ibytes += st->vrxs_ibytes; 2582 accum->vrxs_iqdrops += st->vrxs_iqdrops; 2583 accum->vrxs_csum += st->vrxs_csum; 2584 accum->vrxs_csum_failed += st->vrxs_csum_failed; 2585 accum->vrxs_rescheduled += st->vrxs_rescheduled; 2586} 2587 2588static void 2589vtnet_txq_accum_stats(struct vtnet_txq *txq, struct vtnet_txq_stats *accum) 2590{ 2591 struct vtnet_txq_stats *st; 2592 2593 st = &txq->vtntx_stats; 2594 2595 accum->vtxs_opackets += st->vtxs_opackets; 2596 accum->vtxs_obytes += st->vtxs_obytes; 2597 accum->vtxs_csum += st->vtxs_csum; 2598 accum->vtxs_tso += st->vtxs_tso; 2599 accum->vtxs_rescheduled += st->vtxs_rescheduled; 2600} 2601 2602static void 2603vtnet_accumulate_stats(struct vtnet_softc *sc) 2604{ 2605 struct ifnet *ifp; 2606 struct vtnet_statistics *st; 2607 struct vtnet_rxq_stats rxaccum; 2608 struct vtnet_txq_stats txaccum; 2609 int i; 2610 2611 ifp = sc->vtnet_ifp; 2612 st = &sc->vtnet_stats; 2613 bzero(&rxaccum, sizeof(struct vtnet_rxq_stats)); 2614 bzero(&txaccum, sizeof(struct vtnet_txq_stats)); 2615 2616 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2617 vtnet_rxq_accum_stats(&sc->vtnet_rxqs[i], &rxaccum); 2618 vtnet_txq_accum_stats(&sc->vtnet_txqs[i], &txaccum); 2619 } 2620 2621 st->rx_csum_offloaded = rxaccum.vrxs_csum; 2622 st->rx_csum_failed = rxaccum.vrxs_csum_failed; 2623 st->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 2624 st->tx_csum_offloaded = txaccum.vtxs_csum; 2625 st->tx_tso_offloaded = txaccum.vtxs_tso; 2626 st->tx_task_rescheduled = txaccum.vtxs_rescheduled; 2627 2628 /* 2629 * With the exception of if_ierrors, these ifnet statistics are 2630 * only updated in the driver, so just set them to our accumulated 2631 * values. if_ierrors is updated in ether_input() for malformed 2632 * frames that we should have already discarded. 2633 */ 2634 ifp->if_ipackets = rxaccum.vrxs_ipackets; 2635 ifp->if_iqdrops = rxaccum.vrxs_iqdrops; 2636 ifp->if_ierrors = rxaccum.vrxs_ierrors; 2637 ifp->if_opackets = txaccum.vtxs_opackets; 2638#ifndef VTNET_LEGACY_TX 2639 ifp->if_obytes = txaccum.vtxs_obytes; 2640 ifp->if_omcasts = txaccum.vtxs_omcasts; 2641#endif 2642} 2643 2644static void 2645vtnet_tick(void *xsc) 2646{ 2647 struct vtnet_softc *sc; 2648 struct ifnet *ifp; 2649 int i, timedout; 2650 2651 sc = xsc; 2652 ifp = sc->vtnet_ifp; 2653 timedout = 0; 2654 2655 VTNET_CORE_LOCK_ASSERT(sc); 2656 vtnet_accumulate_stats(sc); 2657 2658 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 2659 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 2660 2661 if (timedout != 0) { 2662 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2663 vtnet_init_locked(sc); 2664 } else 2665 callout_schedule(&sc->vtnet_tick_ch, hz); 2666} 2667 2668static void 2669vtnet_start_taskqueues(struct vtnet_softc *sc) 2670{ 2671 device_t dev; 2672 struct vtnet_rxq *rxq; 2673 struct vtnet_txq *txq; 2674 int i, error; 2675 2676 dev = sc->vtnet_dev; 2677 2678 /* 2679 * Errors here are very difficult to recover from - we cannot 2680 * easily fail because, if this is during boot, we will hang 2681 * when freeing any successfully started taskqueues because 2682 * the scheduler isn't up yet. 2683 * 2684 * Most drivers just ignore the return value - it only fails 2685 * with ENOMEM so an error is not likely. 2686 */ 2687 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2688 rxq = &sc->vtnet_rxqs[i]; 2689 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 2690 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 2691 if (error) { 2692 device_printf(dev, "failed to start rx taskq %d\n", 2693 rxq->vtnrx_id); 2694 } 2695 2696 txq = &sc->vtnet_txqs[i]; 2697 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 2698 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 2699 if (error) { 2700 device_printf(dev, "failed to start tx taskq %d\n", 2701 txq->vtntx_id); 2702 } 2703 } 2704} 2705 2706static void 2707vtnet_free_taskqueues(struct vtnet_softc *sc) 2708{ 2709 struct vtnet_rxq *rxq; 2710 struct vtnet_txq *txq; 2711 int i; 2712 2713 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2714 rxq = &sc->vtnet_rxqs[i]; 2715 if (rxq->vtnrx_tq != NULL) { 2716 taskqueue_free(rxq->vtnrx_tq); 2717 rxq->vtnrx_vq = NULL; 2718 } 2719 2720 txq = &sc->vtnet_txqs[i]; 2721 if (txq->vtntx_tq != NULL) { 2722 taskqueue_free(txq->vtntx_tq); 2723 txq->vtntx_tq = NULL; 2724 } 2725 } 2726} 2727 2728static void 2729vtnet_drain_taskqueues(struct vtnet_softc *sc) 2730{ 2731 struct vtnet_rxq *rxq; 2732 struct vtnet_txq *txq; 2733 int i; 2734 2735 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2736 rxq = &sc->vtnet_rxqs[i]; 2737 if (rxq->vtnrx_tq != NULL) 2738 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2739 2740 txq = &sc->vtnet_txqs[i]; 2741 if (txq->vtntx_tq != NULL) { 2742 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 2743#ifndef VTNET_LEGACY_TX 2744 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 2745#endif 2746 } 2747 } 2748} 2749 2750static void 2751vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 2752{ 2753 struct vtnet_rxq *rxq; 2754 struct vtnet_txq *txq; 2755 int i; 2756 2757 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2758 rxq = &sc->vtnet_rxqs[i]; 2759 vtnet_rxq_free_mbufs(rxq); 2760 2761 txq = &sc->vtnet_txqs[i]; 2762 vtnet_txq_free_mbufs(txq); 2763 } 2764} 2765 2766static void 2767vtnet_stop_rendezvous(struct vtnet_softc *sc) 2768{ 2769 struct vtnet_rxq *rxq; 2770 struct vtnet_txq *txq; 2771 int i; 2772 2773 /* 2774 * Lock and unlock the per-queue mutex so we known the stop 2775 * state is visible. Doing only the active queues should be 2776 * sufficient, but it does not cost much extra to do all the 2777 * queues. Note we hold the core mutex here too. 2778 */ 2779 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2780 rxq = &sc->vtnet_rxqs[i]; 2781 VTNET_RXQ_LOCK(rxq); 2782 VTNET_RXQ_UNLOCK(rxq); 2783 2784 txq = &sc->vtnet_txqs[i]; 2785 VTNET_TXQ_LOCK(txq); 2786 VTNET_TXQ_UNLOCK(txq); 2787 } 2788} 2789 2790static void 2791vtnet_stop(struct vtnet_softc *sc) 2792{ 2793 device_t dev; 2794 struct ifnet *ifp; 2795 2796 dev = sc->vtnet_dev; 2797 ifp = sc->vtnet_ifp; 2798 2799 VTNET_CORE_LOCK_ASSERT(sc); 2800 2801 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2802 sc->vtnet_link_active = 0; 2803 callout_stop(&sc->vtnet_tick_ch); 2804 2805 /* Only advisory. */ 2806 vtnet_disable_interrupts(sc); 2807 2808 /* 2809 * Stop the host adapter. This resets it to the pre-initialized 2810 * state. It will not generate any interrupts until after it is 2811 * reinitialized. 2812 */ 2813 virtio_stop(dev); 2814 vtnet_stop_rendezvous(sc); 2815 2816 /* Free any mbufs left in the virtqueues. */ 2817 vtnet_drain_rxtx_queues(sc); 2818} 2819 2820static int 2821vtnet_virtio_reinit(struct vtnet_softc *sc) 2822{ 2823 device_t dev; 2824 struct ifnet *ifp; 2825 uint64_t features; 2826 int mask, error; 2827 2828 dev = sc->vtnet_dev; 2829 ifp = sc->vtnet_ifp; 2830 features = sc->vtnet_features; 2831 2832 mask = 0; 2833#if defined(INET) 2834 mask |= IFCAP_RXCSUM; 2835#endif 2836#if defined (INET6) 2837 mask |= IFCAP_RXCSUM_IPV6; 2838#endif 2839 2840 /* 2841 * Re-negotiate with the host, removing any disabled receive 2842 * features. Transmit features are disabled only on our side 2843 * via if_capenable and if_hwassist. 2844 */ 2845 2846 if (ifp->if_capabilities & mask) { 2847 /* 2848 * We require both IPv4 and IPv6 offloading to be enabled 2849 * in order to negotiated it: VirtIO does not distinguish 2850 * between the two. 2851 */ 2852 if ((ifp->if_capenable & mask) != mask) 2853 features &= ~VIRTIO_NET_F_GUEST_CSUM; 2854 } 2855 2856 if (ifp->if_capabilities & IFCAP_LRO) { 2857 if ((ifp->if_capenable & IFCAP_LRO) == 0) 2858 features &= ~VTNET_LRO_FEATURES; 2859 } 2860 2861 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) { 2862 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) 2863 features &= ~VIRTIO_NET_F_CTRL_VLAN; 2864 } 2865 2866 error = virtio_reinit(dev, features); 2867 if (error) 2868 device_printf(dev, "virtio reinit error %d\n", error); 2869 2870 return (error); 2871} 2872 2873static void 2874vtnet_init_rx_filters(struct vtnet_softc *sc) 2875{ 2876 struct ifnet *ifp; 2877 2878 ifp = sc->vtnet_ifp; 2879 2880 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 2881 /* Restore promiscuous and all-multicast modes. */ 2882 vtnet_rx_filter(sc); 2883 /* Restore filtered MAC addresses. */ 2884 vtnet_rx_filter_mac(sc); 2885 } 2886 2887 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) 2888 vtnet_rx_filter_vlan(sc); 2889} 2890 2891static int 2892vtnet_init_rx_queues(struct vtnet_softc *sc) 2893{ 2894 device_t dev; 2895 struct vtnet_rxq *rxq; 2896 int i, clsize, error; 2897 2898 dev = sc->vtnet_dev; 2899 2900 /* 2901 * Use the new cluster size if one has been set (via a MTU 2902 * change). Otherwise, use the standard 2K clusters. 2903 * 2904 * BMV: It might make sense to use page sized clusters as 2905 * the default (depending on the features negotiated). 2906 */ 2907 if (sc->vtnet_rx_new_clsize != 0) { 2908 clsize = sc->vtnet_rx_new_clsize; 2909 sc->vtnet_rx_new_clsize = 0; 2910 } else 2911 clsize = MCLBYTES; 2912 2913 sc->vtnet_rx_clsize = clsize; 2914 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize); 2915 2916 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS || 2917 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 2918 ("%s: too many rx mbufs %d for %d segments", __func__, 2919 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 2920 2921#ifdef DEV_NETMAP 2922 if (vtnet_netmap_init_rx_buffers(sc)) 2923 return 0; 2924#endif /* DEV_NETMAP */ 2925 2926 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2927 rxq = &sc->vtnet_rxqs[i]; 2928 2929 /* Hold the lock to satisfy asserts. */ 2930 VTNET_RXQ_LOCK(rxq); 2931 error = vtnet_rxq_populate(rxq); 2932 VTNET_RXQ_UNLOCK(rxq); 2933 2934 if (error) { 2935 device_printf(dev, 2936 "cannot allocate mbufs for Rx queue %d\n", i); 2937 return (error); 2938 } 2939 } 2940 2941 return (0); 2942} 2943 2944static int 2945vtnet_init_tx_queues(struct vtnet_softc *sc) 2946{ 2947 struct vtnet_txq *txq; 2948 int i; 2949 2950 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2951 txq = &sc->vtnet_txqs[i]; 2952 txq->vtntx_watchdog = 0; 2953 } 2954 2955 return (0); 2956} 2957 2958static int 2959vtnet_init_rxtx_queues(struct vtnet_softc *sc) 2960{ 2961 int error; 2962 2963 error = vtnet_init_rx_queues(sc); 2964 if (error) 2965 return (error); 2966 2967 error = vtnet_init_tx_queues(sc); 2968 if (error) 2969 return (error); 2970 2971 return (0); 2972} 2973 2974static void 2975vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 2976{ 2977 device_t dev; 2978 int npairs; 2979 2980 dev = sc->vtnet_dev; 2981 2982 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) { 2983 MPASS(sc->vtnet_max_vq_pairs == 1); 2984 sc->vtnet_act_vq_pairs = 1; 2985 return; 2986 } 2987 2988 /* BMV: Just use the maximum configured for now. */ 2989 npairs = sc->vtnet_max_vq_pairs; 2990 2991 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 2992 device_printf(dev, 2993 "cannot set active queue pairs to %d\n", npairs); 2994 npairs = 1; 2995 } 2996 2997 sc->vtnet_act_vq_pairs = npairs; 2998} 2999 3000static int 3001vtnet_reinit(struct vtnet_softc *sc) 3002{ 3003 struct ifnet *ifp; 3004 int error; 3005 3006 ifp = sc->vtnet_ifp; 3007 3008 /* Use the current MAC address. */ 3009 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 3010 vtnet_set_hwaddr(sc); 3011 3012 vtnet_set_active_vq_pairs(sc); 3013 3014 ifp->if_hwassist = 0; 3015 if (ifp->if_capenable & IFCAP_TXCSUM) 3016 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 3017 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) 3018 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6; 3019 if (ifp->if_capenable & IFCAP_TSO4) 3020 ifp->if_hwassist |= CSUM_TSO; 3021 if (ifp->if_capenable & IFCAP_TSO6) 3022 ifp->if_hwassist |= CSUM_TSO; /* No CSUM_TSO_IPV6. */ 3023 3024 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 3025 vtnet_init_rx_filters(sc); 3026 3027 error = vtnet_init_rxtx_queues(sc); 3028 if (error) 3029 return (error); 3030 3031 vtnet_enable_interrupts(sc); 3032 ifp->if_drv_flags |= IFF_DRV_RUNNING; 3033 3034 return (0); 3035} 3036 3037static void 3038vtnet_init_locked(struct vtnet_softc *sc) 3039{ 3040 device_t dev; 3041 struct ifnet *ifp; 3042 3043 dev = sc->vtnet_dev; 3044 ifp = sc->vtnet_ifp; 3045 3046 VTNET_CORE_LOCK_ASSERT(sc); 3047 3048 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 3049 return; 3050 3051 vtnet_stop(sc); 3052 3053 /* Reinitialize with the host. */ 3054 if (vtnet_virtio_reinit(sc) != 0) 3055 goto fail; 3056 3057 if (vtnet_reinit(sc) != 0) 3058 goto fail; 3059 3060 virtio_reinit_complete(dev); 3061 3062 vtnet_update_link_status(sc); 3063 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3064 3065 return; 3066 3067fail: 3068 vtnet_stop(sc); 3069} 3070 3071static void 3072vtnet_init(void *xsc) 3073{ 3074 struct vtnet_softc *sc; 3075 3076 sc = xsc; 3077 3078#ifdef DEV_NETMAP 3079 if (!NA(sc->vtnet_ifp)) { 3080 D("try to attach again"); 3081 vtnet_netmap_attach(sc); 3082 } 3083#endif /* DEV_NETMAP */ 3084 3085 VTNET_CORE_LOCK(sc); 3086 vtnet_init_locked(sc); 3087 VTNET_CORE_UNLOCK(sc); 3088} 3089 3090static void 3091vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3092{ 3093 struct virtqueue *vq; 3094 3095 vq = sc->vtnet_ctrl_vq; 3096 3097 /* 3098 * The control virtqueue is only polled and therefore it should 3099 * already be empty. 3100 */ 3101 KASSERT(virtqueue_empty(vq), 3102 ("%s: ctrl vq %p not empty", __func__, vq)); 3103} 3104 3105static void 3106vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3107 struct sglist *sg, int readable, int writable) 3108{ 3109 struct virtqueue *vq; 3110 3111 vq = sc->vtnet_ctrl_vq; 3112 3113 VTNET_CORE_LOCK_ASSERT(sc); 3114 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ, 3115 ("%s: CTRL_VQ feature not negotiated", __func__)); 3116 3117 if (!virtqueue_empty(vq)) 3118 return; 3119 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0) 3120 return; 3121 3122 /* 3123 * Poll for the response, but the command is likely already 3124 * done when we return from the notify. 3125 */ 3126 virtqueue_notify(vq); 3127 virtqueue_poll(vq, NULL); 3128} 3129 3130static int 3131vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3132{ 3133 struct virtio_net_ctrl_hdr hdr __aligned(2); 3134 struct sglist_seg segs[3]; 3135 struct sglist sg; 3136 uint8_t ack; 3137 int error; 3138 3139 hdr.class = VIRTIO_NET_CTRL_MAC; 3140 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3141 ack = VIRTIO_NET_ERR; 3142 3143 sglist_init(&sg, 3, segs); 3144 error = 0; 3145 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3146 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN); 3147 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3148 KASSERT(error == 0 && sg.sg_nseg == 3, 3149 ("%s: error %d adding set MAC msg to sglist", __func__, error)); 3150 3151 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3152 3153 return (ack == VIRTIO_NET_OK ? 0 : EIO); 3154} 3155 3156static int 3157vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3158{ 3159 struct sglist_seg segs[3]; 3160 struct sglist sg; 3161 struct { 3162 struct virtio_net_ctrl_hdr hdr; 3163 uint8_t pad1; 3164 struct virtio_net_ctrl_mq mq; 3165 uint8_t pad2; 3166 uint8_t ack; 3167 } s __aligned(2); 3168 int error; 3169 3170 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3171 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3172 s.mq.virtqueue_pairs = npairs; 3173 s.ack = VIRTIO_NET_ERR; 3174 3175 sglist_init(&sg, 3, segs); 3176 error = 0; 3177 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3178 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3179 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3180 KASSERT(error == 0 && sg.sg_nseg == 3, 3181 ("%s: error %d adding MQ message to sglist", __func__, error)); 3182 3183 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3184 3185 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3186} 3187 3188static int 3189vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on) 3190{ 3191 struct sglist_seg segs[3]; 3192 struct sglist sg; 3193 struct { 3194 struct virtio_net_ctrl_hdr hdr; 3195 uint8_t pad1; 3196 uint8_t onoff; 3197 uint8_t pad2; 3198 uint8_t ack; 3199 } s __aligned(2); 3200 int error; 3201 3202 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3203 ("%s: CTRL_RX feature not negotiated", __func__)); 3204 3205 s.hdr.class = VIRTIO_NET_CTRL_RX; 3206 s.hdr.cmd = cmd; 3207 s.onoff = !!on; 3208 s.ack = VIRTIO_NET_ERR; 3209 3210 sglist_init(&sg, 3, segs); 3211 error = 0; 3212 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3213 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3214 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3215 KASSERT(error == 0 && sg.sg_nseg == 3, 3216 ("%s: error %d adding Rx message to sglist", __func__, error)); 3217 3218 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3219 3220 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3221} 3222 3223static int 3224vtnet_set_promisc(struct vtnet_softc *sc, int on) 3225{ 3226 3227 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3228} 3229 3230static int 3231vtnet_set_allmulti(struct vtnet_softc *sc, int on) 3232{ 3233 3234 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3235} 3236 3237/* 3238 * The device defaults to promiscuous mode for backwards compatibility. 3239 * Turn it off at attach time if possible. 3240 */ 3241static void 3242vtnet_attach_disable_promisc(struct vtnet_softc *sc) 3243{ 3244 struct ifnet *ifp; 3245 3246 ifp = sc->vtnet_ifp; 3247 3248 VTNET_CORE_LOCK(sc); 3249 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) { 3250 ifp->if_flags |= IFF_PROMISC; 3251 } else if (vtnet_set_promisc(sc, 0) != 0) { 3252 ifp->if_flags |= IFF_PROMISC; 3253 device_printf(sc->vtnet_dev, 3254 "cannot disable default promiscuous mode\n"); 3255 } 3256 VTNET_CORE_UNLOCK(sc); 3257} 3258 3259static void 3260vtnet_rx_filter(struct vtnet_softc *sc) 3261{ 3262 device_t dev; 3263 struct ifnet *ifp; 3264 3265 dev = sc->vtnet_dev; 3266 ifp = sc->vtnet_ifp; 3267 3268 VTNET_CORE_LOCK_ASSERT(sc); 3269 3270 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) 3271 device_printf(dev, "cannot %s promiscuous mode\n", 3272 ifp->if_flags & IFF_PROMISC ? "enable" : "disable"); 3273 3274 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) 3275 device_printf(dev, "cannot %s all-multicast mode\n", 3276 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable"); 3277} 3278 3279static void 3280vtnet_rx_filter_mac(struct vtnet_softc *sc) 3281{ 3282 struct virtio_net_ctrl_hdr hdr __aligned(2); 3283 struct vtnet_mac_filter *filter; 3284 struct sglist_seg segs[4]; 3285 struct sglist sg; 3286 struct ifnet *ifp; 3287 struct ifaddr *ifa; 3288 struct ifmultiaddr *ifma; 3289 int ucnt, mcnt, promisc, allmulti, error; 3290 uint8_t ack; 3291 3292 ifp = sc->vtnet_ifp; 3293 filter = sc->vtnet_mac_filter; 3294 ucnt = 0; 3295 mcnt = 0; 3296 promisc = 0; 3297 allmulti = 0; 3298 3299 VTNET_CORE_LOCK_ASSERT(sc); 3300 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3301 ("%s: CTRL_RX feature not negotiated", __func__)); 3302 3303 /* Unicast MAC addresses: */ 3304 if_addr_rlock(ifp); 3305 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 3306 if (ifa->ifa_addr->sa_family != AF_LINK) 3307 continue; 3308 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 3309 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3310 continue; 3311 else if (ucnt == VTNET_MAX_MAC_ENTRIES) { 3312 promisc = 1; 3313 break; 3314 } 3315 3316 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr), 3317 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN); 3318 ucnt++; 3319 } 3320 if_addr_runlock(ifp); 3321 3322 if (promisc != 0) { 3323 filter->vmf_unicast.nentries = 0; 3324 if_printf(ifp, "more than %d MAC addresses assigned, " 3325 "falling back to promiscuous mode\n", 3326 VTNET_MAX_MAC_ENTRIES); 3327 } else 3328 filter->vmf_unicast.nentries = ucnt; 3329 3330 /* Multicast MAC addresses: */ 3331 if_maddr_rlock(ifp); 3332 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { 3333 if (ifma->ifma_addr->sa_family != AF_LINK) 3334 continue; 3335 else if (mcnt == VTNET_MAX_MAC_ENTRIES) { 3336 allmulti = 1; 3337 break; 3338 } 3339 3340 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr), 3341 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN); 3342 mcnt++; 3343 } 3344 if_maddr_runlock(ifp); 3345 3346 if (allmulti != 0) { 3347 filter->vmf_multicast.nentries = 0; 3348 if_printf(ifp, "more than %d multicast MAC addresses " 3349 "assigned, falling back to all-multicast mode\n", 3350 VTNET_MAX_MAC_ENTRIES); 3351 } else 3352 filter->vmf_multicast.nentries = mcnt; 3353 3354 if (promisc != 0 && allmulti != 0) 3355 goto out; 3356 3357 hdr.class = VIRTIO_NET_CTRL_MAC; 3358 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3359 ack = VIRTIO_NET_ERR; 3360 3361 sglist_init(&sg, 4, segs); 3362 error = 0; 3363 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3364 error |= sglist_append(&sg, &filter->vmf_unicast, 3365 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN); 3366 error |= sglist_append(&sg, &filter->vmf_multicast, 3367 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN); 3368 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3369 KASSERT(error == 0 && sg.sg_nseg == 4, 3370 ("%s: error %d adding MAC filter msg to sglist", __func__, error)); 3371 3372 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3373 3374 if (ack != VIRTIO_NET_OK) 3375 if_printf(ifp, "error setting host MAC filter table\n"); 3376 3377out: 3378 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0) 3379 if_printf(ifp, "cannot enable promiscuous mode\n"); 3380 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0) 3381 if_printf(ifp, "cannot enable all-multicast mode\n"); 3382} 3383 3384static int 3385vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3386{ 3387 struct sglist_seg segs[3]; 3388 struct sglist sg; 3389 struct { 3390 struct virtio_net_ctrl_hdr hdr; 3391 uint8_t pad1; 3392 uint16_t tag; 3393 uint8_t pad2; 3394 uint8_t ack; 3395 } s __aligned(2); 3396 int error; 3397 3398 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3399 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3400 s.tag = tag; 3401 s.ack = VIRTIO_NET_ERR; 3402 3403 sglist_init(&sg, 3, segs); 3404 error = 0; 3405 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3406 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3407 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3408 KASSERT(error == 0 && sg.sg_nseg == 3, 3409 ("%s: error %d adding VLAN message to sglist", __func__, error)); 3410 3411 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3412 3413 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3414} 3415 3416static void 3417vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3418{ 3419 uint32_t w; 3420 uint16_t tag; 3421 int i, bit; 3422 3423 VTNET_CORE_LOCK_ASSERT(sc); 3424 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER, 3425 ("%s: VLAN_FILTER feature not negotiated", __func__)); 3426 3427 /* Enable the filter for each configured VLAN. */ 3428 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3429 w = sc->vtnet_vlan_filter[i]; 3430 3431 while ((bit = ffs(w) - 1) != -1) { 3432 w &= ~(1 << bit); 3433 tag = sizeof(w) * CHAR_BIT * i + bit; 3434 3435 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3436 device_printf(sc->vtnet_dev, 3437 "cannot enable VLAN %d filter\n", tag); 3438 } 3439 } 3440 } 3441} 3442 3443static void 3444vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3445{ 3446 struct ifnet *ifp; 3447 int idx, bit; 3448 3449 ifp = sc->vtnet_ifp; 3450 idx = (tag >> 5) & 0x7F; 3451 bit = tag & 0x1F; 3452 3453 if (tag == 0 || tag > 4095) 3454 return; 3455 3456 VTNET_CORE_LOCK(sc); 3457 3458 if (add) 3459 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3460 else 3461 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3462 3463 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && 3464 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3465 device_printf(sc->vtnet_dev, 3466 "cannot %s VLAN %d %s the host filter table\n", 3467 add ? "add" : "remove", tag, add ? "to" : "from"); 3468 } 3469 3470 VTNET_CORE_UNLOCK(sc); 3471} 3472 3473static void 3474vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3475{ 3476 3477 if (ifp->if_softc != arg) 3478 return; 3479 3480 vtnet_update_vlan_filter(arg, 1, tag); 3481} 3482 3483static void 3484vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3485{ 3486 3487 if (ifp->if_softc != arg) 3488 return; 3489 3490 vtnet_update_vlan_filter(arg, 0, tag); 3491} 3492 3493static int 3494vtnet_is_link_up(struct vtnet_softc *sc) 3495{ 3496 device_t dev; 3497 struct ifnet *ifp; 3498 uint16_t status; 3499 3500 dev = sc->vtnet_dev; 3501 ifp = sc->vtnet_ifp; 3502 3503 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0) 3504 status = VIRTIO_NET_S_LINK_UP; 3505 else 3506 status = virtio_read_dev_config_2(dev, 3507 offsetof(struct virtio_net_config, status)); 3508 3509 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3510} 3511 3512static void 3513vtnet_update_link_status(struct vtnet_softc *sc) 3514{ 3515 struct ifnet *ifp; 3516 int link; 3517 3518 ifp = sc->vtnet_ifp; 3519 3520 VTNET_CORE_LOCK_ASSERT(sc); 3521 link = vtnet_is_link_up(sc); 3522 3523 /* Notify if the link status has changed. */ 3524 if (link != 0 && sc->vtnet_link_active == 0) { 3525 sc->vtnet_link_active = 1; 3526 if_link_state_change(ifp, LINK_STATE_UP); 3527 } else if (link == 0 && sc->vtnet_link_active != 0) { 3528 sc->vtnet_link_active = 0; 3529 if_link_state_change(ifp, LINK_STATE_DOWN); 3530 } 3531} 3532 3533static int 3534vtnet_ifmedia_upd(struct ifnet *ifp) 3535{ 3536 struct vtnet_softc *sc; 3537 struct ifmedia *ifm; 3538 3539 sc = ifp->if_softc; 3540 ifm = &sc->vtnet_media; 3541 3542 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) 3543 return (EINVAL); 3544 3545 return (0); 3546} 3547 3548static void 3549vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 3550{ 3551 struct vtnet_softc *sc; 3552 3553 sc = ifp->if_softc; 3554 3555 ifmr->ifm_status = IFM_AVALID; 3556 ifmr->ifm_active = IFM_ETHER; 3557 3558 VTNET_CORE_LOCK(sc); 3559 if (vtnet_is_link_up(sc) != 0) { 3560 ifmr->ifm_status |= IFM_ACTIVE; 3561 ifmr->ifm_active |= VTNET_MEDIATYPE; 3562 } else 3563 ifmr->ifm_active |= IFM_NONE; 3564 VTNET_CORE_UNLOCK(sc); 3565} 3566 3567static void 3568vtnet_set_hwaddr(struct vtnet_softc *sc) 3569{ 3570 device_t dev; 3571 int i; 3572 3573 dev = sc->vtnet_dev; 3574 3575 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 3576 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0) 3577 device_printf(dev, "unable to set MAC address\n"); 3578 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) { 3579 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3580 virtio_write_dev_config_1(dev, 3581 offsetof(struct virtio_net_config, mac) + i, 3582 sc->vtnet_hwaddr[i]); 3583 } 3584 } 3585} 3586 3587static void 3588vtnet_get_hwaddr(struct vtnet_softc *sc) 3589{ 3590 device_t dev; 3591 int i; 3592 3593 dev = sc->vtnet_dev; 3594 3595 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { 3596 /* 3597 * Generate a random locally administered unicast address. 3598 * 3599 * It would be nice to generate the same MAC address across 3600 * reboots, but it seems all the hosts currently available 3601 * support the MAC feature, so this isn't too important. 3602 */ 3603 sc->vtnet_hwaddr[0] = 0xB2; 3604 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 3605 vtnet_set_hwaddr(sc); 3606 return; 3607 } 3608 3609 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3610 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev, 3611 offsetof(struct virtio_net_config, mac) + i); 3612 } 3613} 3614 3615static void 3616vtnet_vlan_tag_remove(struct mbuf *m) 3617{ 3618 struct ether_vlan_header *evh; 3619 3620 evh = mtod(m, struct ether_vlan_header *); 3621 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 3622 m->m_flags |= M_VLANTAG; 3623 3624 /* Strip the 802.1Q header. */ 3625 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 3626 ETHER_HDR_LEN - ETHER_TYPE_LEN); 3627 m_adj(m, ETHER_VLAN_ENCAP_LEN); 3628} 3629 3630static void 3631vtnet_set_rx_process_limit(struct vtnet_softc *sc) 3632{ 3633 int limit; 3634 3635 limit = vtnet_tunable_int(sc, "rx_process_limit", 3636 vtnet_rx_process_limit); 3637 if (limit < 0) 3638 limit = INT_MAX; 3639 sc->vtnet_rx_process_limit = limit; 3640} 3641 3642static void 3643vtnet_set_tx_intr_threshold(struct vtnet_softc *sc) 3644{ 3645 device_t dev; 3646 int size, thresh; 3647 3648 dev = sc->vtnet_dev; 3649 size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq); 3650 3651 /* 3652 * The Tx interrupt is disabled until the queue free count falls 3653 * below our threshold. Completed frames are drained from the Tx 3654 * virtqueue before transmitting new frames and in the watchdog 3655 * callout, so the frequency of Tx interrupts is greatly reduced, 3656 * at the cost of not freeing mbufs as quickly as they otherwise 3657 * would be. 3658 * 3659 * N.B. We assume all the Tx queues are the same size. 3660 */ 3661 thresh = size / 4; 3662 3663 /* 3664 * Without indirect descriptors, leave enough room for the most 3665 * segments we handle. 3666 */ 3667 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC) == 0 && 3668 thresh < sc->vtnet_tx_nsegs) 3669 thresh = sc->vtnet_tx_nsegs; 3670 3671 sc->vtnet_tx_intr_thresh = thresh; 3672} 3673 3674static void 3675vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 3676 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 3677{ 3678 struct sysctl_oid *node; 3679 struct sysctl_oid_list *list; 3680 struct vtnet_rxq_stats *stats; 3681 char namebuf[16]; 3682 3683 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 3684 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3685 CTLFLAG_RD, NULL, "Receive Queue"); 3686 list = SYSCTL_CHILDREN(node); 3687 3688 stats = &rxq->vtnrx_stats; 3689 3690 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, 3691 &stats->vrxs_ipackets, "Receive packets"); 3692 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, 3693 &stats->vrxs_ibytes, "Receive bytes"); 3694 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, 3695 &stats->vrxs_iqdrops, "Receive drops"); 3696 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, 3697 &stats->vrxs_ierrors, "Receive errors"); 3698 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3699 &stats->vrxs_csum, "Receive checksum offloaded"); 3700 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, 3701 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 3702 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3703 &stats->vrxs_rescheduled, 3704 "Receive interrupt handler rescheduled"); 3705} 3706 3707static void 3708vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 3709 struct sysctl_oid_list *child, struct vtnet_txq *txq) 3710{ 3711 struct sysctl_oid *node; 3712 struct sysctl_oid_list *list; 3713 struct vtnet_txq_stats *stats; 3714 char namebuf[16]; 3715 3716 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 3717 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3718 CTLFLAG_RD, NULL, "Transmit Queue"); 3719 list = SYSCTL_CHILDREN(node); 3720 3721 stats = &txq->vtntx_stats; 3722 3723 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, 3724 &stats->vtxs_opackets, "Transmit packets"); 3725 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, 3726 &stats->vtxs_obytes, "Transmit bytes"); 3727 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, 3728 &stats->vtxs_omcasts, "Transmit multicasts"); 3729 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3730 &stats->vtxs_csum, "Transmit checksum offloaded"); 3731 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, 3732 &stats->vtxs_tso, "Transmit segmentation offloaded"); 3733 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3734 &stats->vtxs_rescheduled, 3735 "Transmit interrupt handler rescheduled"); 3736} 3737 3738static void 3739vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 3740{ 3741 device_t dev; 3742 struct sysctl_ctx_list *ctx; 3743 struct sysctl_oid *tree; 3744 struct sysctl_oid_list *child; 3745 int i; 3746 3747 dev = sc->vtnet_dev; 3748 ctx = device_get_sysctl_ctx(dev); 3749 tree = device_get_sysctl_tree(dev); 3750 child = SYSCTL_CHILDREN(tree); 3751 3752 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3753 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 3754 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 3755 } 3756} 3757 3758static void 3759vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 3760 struct sysctl_oid_list *child, struct vtnet_softc *sc) 3761{ 3762 struct vtnet_statistics *stats; 3763 3764 stats = &sc->vtnet_stats; 3765 3766 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 3767 CTLFLAG_RD, &stats->mbuf_alloc_failed, 3768 "Mbuf cluster allocation failures"); 3769 3770 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 3771 CTLFLAG_RD, &stats->rx_frame_too_large, 3772 "Received frame larger than the mbuf chain"); 3773 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 3774 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 3775 "Enqueuing the replacement receive mbuf failed"); 3776 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 3777 CTLFLAG_RD, &stats->rx_mergeable_failed, 3778 "Mergeable buffers receive failures"); 3779 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 3780 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 3781 "Received checksum offloaded buffer with unsupported " 3782 "Ethernet type"); 3783 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 3784 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 3785 "Received checksum offloaded buffer with incorrect IP protocol"); 3786 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 3787 CTLFLAG_RD, &stats->rx_csum_bad_offset, 3788 "Received checksum offloaded buffer with incorrect offset"); 3789 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", 3790 CTLFLAG_RD, &stats->rx_csum_bad_proto, 3791 "Received checksum offloaded buffer with incorrect protocol"); 3792 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 3793 CTLFLAG_RD, &stats->rx_csum_failed, 3794 "Received buffer checksum offload failed"); 3795 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 3796 CTLFLAG_RD, &stats->rx_csum_offloaded, 3797 "Received buffer checksum offload succeeded"); 3798 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 3799 CTLFLAG_RD, &stats->rx_task_rescheduled, 3800 "Times the receive interrupt task rescheduled itself"); 3801 3802 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype", 3803 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 3804 "Aborted transmit of checksum offloaded buffer with unknown " 3805 "Ethernet type"); 3806 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype", 3807 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 3808 "Aborted transmit of TSO buffer with unknown Ethernet type"); 3809 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 3810 CTLFLAG_RD, &stats->tx_tso_not_tcp, 3811 "Aborted transmit of TSO buffer with non TCP protocol"); 3812 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 3813 CTLFLAG_RD, &stats->tx_defragged, 3814 "Transmit mbufs defragged"); 3815 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 3816 CTLFLAG_RD, &stats->tx_defrag_failed, 3817 "Aborted transmit of buffer because defrag failed"); 3818 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 3819 CTLFLAG_RD, &stats->tx_csum_offloaded, 3820 "Offloaded checksum of transmitted buffer"); 3821 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 3822 CTLFLAG_RD, &stats->tx_tso_offloaded, 3823 "Segmentation offload of transmitted buffer"); 3824 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 3825 CTLFLAG_RD, &stats->tx_task_rescheduled, 3826 "Times the transmit interrupt task rescheduled itself"); 3827} 3828 3829static void 3830vtnet_setup_sysctl(struct vtnet_softc *sc) 3831{ 3832 device_t dev; 3833 struct sysctl_ctx_list *ctx; 3834 struct sysctl_oid *tree; 3835 struct sysctl_oid_list *child; 3836 3837 dev = sc->vtnet_dev; 3838 ctx = device_get_sysctl_ctx(dev); 3839 tree = device_get_sysctl_tree(dev); 3840 child = SYSCTL_CHILDREN(tree); 3841 3842 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 3843 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 3844 "Maximum number of supported virtqueue pairs"); 3845 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 3846 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 3847 "Number of active virtqueue pairs"); 3848 3849 vtnet_setup_stat_sysctl(ctx, child, sc); 3850} 3851 3852static int 3853vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 3854{ 3855 3856 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 3857} 3858 3859static void 3860vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 3861{ 3862 3863 virtqueue_disable_intr(rxq->vtnrx_vq); 3864} 3865 3866static int 3867vtnet_txq_enable_intr(struct vtnet_txq *txq) 3868{ 3869 struct virtqueue *vq; 3870 3871 vq = txq->vtntx_vq; 3872 3873 if (vtnet_txq_below_threshold(txq) != 0) 3874 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG)); 3875 3876 /* 3877 * The free count is above our threshold. Keep the Tx interrupt 3878 * disabled until the queue is fuller. 3879 */ 3880 return (0); 3881} 3882 3883static void 3884vtnet_txq_disable_intr(struct vtnet_txq *txq) 3885{ 3886 3887 virtqueue_disable_intr(txq->vtntx_vq); 3888} 3889 3890static void 3891vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 3892{ 3893 int i; 3894 3895 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3896 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]); 3897} 3898 3899static void 3900vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 3901{ 3902 int i; 3903 3904 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3905 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 3906} 3907 3908static void 3909vtnet_enable_interrupts(struct vtnet_softc *sc) 3910{ 3911 3912 vtnet_enable_rx_interrupts(sc); 3913 vtnet_enable_tx_interrupts(sc); 3914} 3915 3916static void 3917vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 3918{ 3919 int i; 3920 3921 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3922 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 3923} 3924 3925static void 3926vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 3927{ 3928 int i; 3929 3930 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3931 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 3932} 3933 3934static void 3935vtnet_disable_interrupts(struct vtnet_softc *sc) 3936{ 3937 3938 vtnet_disable_rx_interrupts(sc); 3939 vtnet_disable_tx_interrupts(sc); 3940} 3941 3942static int 3943vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 3944{ 3945 char path[64]; 3946 3947 snprintf(path, sizeof(path), 3948 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 3949 TUNABLE_INT_FETCH(path, &def); 3950 3951 return (def); 3952} 3953