1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice unmodified, this list of conditions, and the following 12 * disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 /* Driver for VirtIO network devices. */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/eventhandler.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/sockio.h> 39 #include <sys/mbuf.h> 40 #include <sys/malloc.h> 41 #include <sys/module.h> 42 #include <sys/socket.h> 43 #include <sys/sysctl.h> 44 #include <sys/random.h> 45 #include <sys/sglist.h> 46 #include <sys/lock.h> 47 #include <sys/mutex.h> 48 #include <sys/taskqueue.h> 49 #include <sys/smp.h> 50 #include <machine/smp.h> 51 52 #include <vm/uma.h> 53 54 #include <net/debugnet.h> 55 #include <net/ethernet.h> 56 #include <net/pfil.h> 57 #include <net/if.h> 58 #include <net/if_var.h> 59 #include <net/if_arp.h> 60 #include <net/if_dl.h> 61 #include <net/if_types.h> 62 #include <net/if_media.h> 63 #include <net/if_vlan_var.h> 64 65 #include <net/bpf.h> 66 67 #include <netinet/in_systm.h> 68 #include <netinet/in.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip6.h> 71 #include <netinet6/ip6_var.h> 72 #include <netinet/udp.h> 73 #include <netinet/tcp.h> 74 75 #include <machine/bus.h> 76 #include <machine/resource.h> 77 #include <sys/bus.h> 78 #include <sys/rman.h> 79 80 #include <dev/virtio/virtio.h> 81 #include <dev/virtio/virtqueue.h> 82 #include <dev/virtio/network/virtio_net.h> 83 #include <dev/virtio/network/if_vtnetvar.h> 84 #include "virtio_if.h" 85 86 #include "opt_inet.h" 87 #include "opt_inet6.h" 88 89 static int vtnet_modevent(module_t, int, void *); 90 91 static int vtnet_probe(device_t); 92 static int vtnet_attach(device_t); 93 static int vtnet_detach(device_t); 94 static int vtnet_suspend(device_t); 95 static int vtnet_resume(device_t); 96 static int vtnet_shutdown(device_t); 97 static int vtnet_attach_completed(device_t); 98 static int vtnet_config_change(device_t); 99 100 static void vtnet_negotiate_features(struct vtnet_softc *); 101 static void vtnet_setup_features(struct vtnet_softc *); 102 static int vtnet_init_rxq(struct vtnet_softc *, int); 103 static int vtnet_init_txq(struct vtnet_softc *, int); 104 static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); 105 static void vtnet_free_rxtx_queues(struct vtnet_softc *); 106 static int vtnet_alloc_rx_filters(struct vtnet_softc *); 107 static void vtnet_free_rx_filters(struct vtnet_softc *); 108 static int vtnet_alloc_virtqueues(struct vtnet_softc *); 109 static int vtnet_setup_interface(struct vtnet_softc *); 110 static int vtnet_change_mtu(struct vtnet_softc *, int); 111 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t); 112 static uint64_t vtnet_get_counter(struct ifnet *, ift_counter); 113 114 static int vtnet_rxq_populate(struct vtnet_rxq *); 115 static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); 116 static struct mbuf * 117 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); 118 static int vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *, 119 struct mbuf *, int); 120 static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); 121 static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); 122 static int vtnet_rxq_new_buf(struct vtnet_rxq *); 123 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, 124 struct virtio_net_hdr *); 125 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); 126 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); 127 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); 128 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, 129 struct virtio_net_hdr *); 130 static int vtnet_rxq_eof(struct vtnet_rxq *); 131 static void vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries); 132 static void vtnet_rx_vq_intr(void *); 133 static void vtnet_rxq_tq_intr(void *, int); 134 135 static int vtnet_txq_below_threshold(struct vtnet_txq *); 136 static int vtnet_txq_notify(struct vtnet_txq *); 137 static void vtnet_txq_free_mbufs(struct vtnet_txq *); 138 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, 139 int *, int *, int *); 140 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, 141 int, struct virtio_net_hdr *); 142 static struct mbuf * 143 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, 144 struct virtio_net_hdr *); 145 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, 146 struct vtnet_tx_header *); 147 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int); 148 #ifdef VTNET_LEGACY_TX 149 static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *); 150 static void vtnet_start(struct ifnet *); 151 #else 152 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); 153 static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *); 154 static void vtnet_txq_tq_deferred(void *, int); 155 #endif 156 static void vtnet_txq_start(struct vtnet_txq *); 157 static void vtnet_txq_tq_intr(void *, int); 158 static int vtnet_txq_eof(struct vtnet_txq *); 159 static void vtnet_tx_vq_intr(void *); 160 static void vtnet_tx_start_all(struct vtnet_softc *); 161 162 #ifndef VTNET_LEGACY_TX 163 static void vtnet_qflush(struct ifnet *); 164 #endif 165 166 static int vtnet_watchdog(struct vtnet_txq *); 167 static void vtnet_accum_stats(struct vtnet_softc *, 168 struct vtnet_rxq_stats *, struct vtnet_txq_stats *); 169 static void vtnet_tick(void *); 170 171 static void vtnet_start_taskqueues(struct vtnet_softc *); 172 static void vtnet_free_taskqueues(struct vtnet_softc *); 173 static void vtnet_drain_taskqueues(struct vtnet_softc *); 174 175 static void vtnet_drain_rxtx_queues(struct vtnet_softc *); 176 static void vtnet_stop_rendezvous(struct vtnet_softc *); 177 static void vtnet_stop(struct vtnet_softc *); 178 static int vtnet_virtio_reinit(struct vtnet_softc *); 179 static void vtnet_init_rx_filters(struct vtnet_softc *); 180 static int vtnet_init_rx_queues(struct vtnet_softc *); 181 static int vtnet_init_tx_queues(struct vtnet_softc *); 182 static int vtnet_init_rxtx_queues(struct vtnet_softc *); 183 static void vtnet_set_active_vq_pairs(struct vtnet_softc *); 184 static int vtnet_reinit(struct vtnet_softc *); 185 static void vtnet_init_locked(struct vtnet_softc *, int); 186 static void vtnet_init(void *); 187 188 static void vtnet_free_ctrl_vq(struct vtnet_softc *); 189 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 190 struct sglist *, int, int); 191 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 192 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); 193 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int); 194 static int vtnet_set_promisc(struct vtnet_softc *, int); 195 static int vtnet_set_allmulti(struct vtnet_softc *, int); 196 static void vtnet_attach_disable_promisc(struct vtnet_softc *); 197 static void vtnet_rx_filter(struct vtnet_softc *); 198 static void vtnet_rx_filter_mac(struct vtnet_softc *); 199 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 200 static void vtnet_rx_filter_vlan(struct vtnet_softc *); 201 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 202 static void vtnet_register_vlan(void *, struct ifnet *, uint16_t); 203 static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t); 204 205 static int vtnet_is_link_up(struct vtnet_softc *); 206 static void vtnet_update_link_status(struct vtnet_softc *); 207 static int vtnet_ifmedia_upd(struct ifnet *); 208 static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *); 209 static void vtnet_get_hwaddr(struct vtnet_softc *); 210 static void vtnet_set_hwaddr(struct vtnet_softc *); 211 static void vtnet_vlan_tag_remove(struct mbuf *); 212 static void vtnet_set_rx_process_limit(struct vtnet_softc *); 213 static void vtnet_set_tx_intr_threshold(struct vtnet_softc *); 214 215 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, 216 struct sysctl_oid_list *, struct vtnet_rxq *); 217 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, 218 struct sysctl_oid_list *, struct vtnet_txq *); 219 static void vtnet_setup_queue_sysctl(struct vtnet_softc *); 220 static void vtnet_setup_sysctl(struct vtnet_softc *); 221 222 static int vtnet_rxq_enable_intr(struct vtnet_rxq *); 223 static void vtnet_rxq_disable_intr(struct vtnet_rxq *); 224 static int vtnet_txq_enable_intr(struct vtnet_txq *); 225 static void vtnet_txq_disable_intr(struct vtnet_txq *); 226 static void vtnet_enable_rx_interrupts(struct vtnet_softc *); 227 static void vtnet_enable_tx_interrupts(struct vtnet_softc *); 228 static void vtnet_enable_interrupts(struct vtnet_softc *); 229 static void vtnet_disable_rx_interrupts(struct vtnet_softc *); 230 static void vtnet_disable_tx_interrupts(struct vtnet_softc *); 231 static void vtnet_disable_interrupts(struct vtnet_softc *); 232 233 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); 234 235 DEBUGNET_DEFINE(vtnet); 236 237 /* Tunables. */ 238 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 239 "VNET driver parameters"); 240 static int vtnet_csum_disable = 0; 241 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable); 242 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN, 243 &vtnet_csum_disable, 0, "Disables receive and send checksum offload"); 244 static int vtnet_tso_disable = 0; 245 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable); 246 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, &vtnet_tso_disable, 247 0, "Disables TCP Segmentation Offload"); 248 static int vtnet_lro_disable = 0; 249 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable); 250 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, &vtnet_lro_disable, 251 0, "Disables TCP Large Receive Offload"); 252 static int vtnet_mq_disable = 0; 253 TUNABLE_INT("hw.vtnet.mq_disable", &vtnet_mq_disable); 254 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, &vtnet_mq_disable, 255 0, "Disables Multi Queue support"); 256 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS; 257 TUNABLE_INT("hw.vtnet.mq_max_pairs", &vtnet_mq_max_pairs); 258 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN, 259 &vtnet_mq_max_pairs, 0, "Sets the maximum number of Multi Queue pairs"); 260 static int vtnet_rx_process_limit = 512; 261 TUNABLE_INT("hw.vtnet.rx_process_limit", &vtnet_rx_process_limit); 262 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, 263 &vtnet_rx_process_limit, 0, 264 "Limits the number RX segments processed in a single pass"); 265 266 static uma_zone_t vtnet_tx_header_zone; 267 268 static struct virtio_feature_desc vtnet_feature_desc[] = { 269 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 270 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 271 { VIRTIO_NET_F_MAC, "MacAddress" }, 272 { VIRTIO_NET_F_GSO, "TxAllGSO" }, 273 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" }, 274 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" }, 275 { VIRTIO_NET_F_GUEST_ECN, "RxECN" }, 276 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 277 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 278 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 279 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 280 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 281 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 282 { VIRTIO_NET_F_STATUS, "Status" }, 283 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" }, 284 { VIRTIO_NET_F_CTRL_RX, "RxMode" }, 285 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" }, 286 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" }, 287 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 288 { VIRTIO_NET_F_MQ, "Multiqueue" }, 289 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" }, 290 { 0, NULL } 291 }; 292 293 static device_method_t vtnet_methods[] = { 294 /* Device methods. */ 295 DEVMETHOD(device_probe, vtnet_probe), 296 DEVMETHOD(device_attach, vtnet_attach), 297 DEVMETHOD(device_detach, vtnet_detach), 298 DEVMETHOD(device_suspend, vtnet_suspend), 299 DEVMETHOD(device_resume, vtnet_resume), 300 DEVMETHOD(device_shutdown, vtnet_shutdown), 301 302 /* VirtIO methods. */ 303 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), 304 DEVMETHOD(virtio_config_change, vtnet_config_change), 305 306 DEVMETHOD_END 307 }; 308 309 #ifdef DEV_NETMAP 310 #include <dev/netmap/if_vtnet_netmap.h> 311 #endif /* DEV_NETMAP */ 312 313 static driver_t vtnet_driver = { 314 "vtnet", 315 vtnet_methods, 316 sizeof(struct vtnet_softc) 317 }; 318 static devclass_t vtnet_devclass; 319 320 DRIVER_MODULE(vtnet, virtio_mmio, vtnet_driver, vtnet_devclass, 321 vtnet_modevent, 0); 322 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass, 323 vtnet_modevent, 0); 324 MODULE_VERSION(vtnet, 1); 325 MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 326 #ifdef DEV_NETMAP 327 MODULE_DEPEND(vtnet, netmap, 1, 1, 1); 328 #endif /* DEV_NETMAP */ 329 330 VIRTIO_SIMPLE_PNPTABLE(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter"); 331 VIRTIO_SIMPLE_PNPINFO(virtio_mmio, vtnet); 332 VIRTIO_SIMPLE_PNPINFO(virtio_pci, vtnet); 333 334 static int 335 vtnet_modevent(module_t mod, int type, void *unused) 336 { 337 int error = 0; 338 static int loaded = 0; 339 340 switch (type) { 341 case MOD_LOAD: 342 if (loaded++ == 0) { 343 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", 344 sizeof(struct vtnet_tx_header), 345 NULL, NULL, NULL, NULL, 0, 0); 346 #ifdef DEBUGNET 347 /* 348 * We need to allocate from this zone in the transmit path, so ensure 349 * that we have at least one item per header available. 350 * XXX add a separate zone like we do for mbufs? otherwise we may alloc 351 * buckets 352 */ 353 uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 354 uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 355 #endif 356 } 357 break; 358 case MOD_QUIESCE: 359 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) 360 error = EBUSY; 361 break; 362 case MOD_UNLOAD: 363 if (--loaded == 0) { 364 uma_zdestroy(vtnet_tx_header_zone); 365 vtnet_tx_header_zone = NULL; 366 } 367 break; 368 case MOD_SHUTDOWN: 369 break; 370 default: 371 error = EOPNOTSUPP; 372 break; 373 } 374 375 return (error); 376 } 377 378 static int 379 vtnet_probe(device_t dev) 380 { 381 return (VIRTIO_SIMPLE_PROBE(dev, vtnet)); 382 } 383 384 static int 385 vtnet_attach(device_t dev) 386 { 387 struct vtnet_softc *sc; 388 int error; 389 390 sc = device_get_softc(dev); 391 sc->vtnet_dev = dev; 392 393 /* Register our feature descriptions. */ 394 virtio_set_feature_desc(dev, vtnet_feature_desc); 395 396 VTNET_CORE_LOCK_INIT(sc); 397 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); 398 399 vtnet_setup_sysctl(sc); 400 vtnet_setup_features(sc); 401 402 error = vtnet_alloc_rx_filters(sc); 403 if (error) { 404 device_printf(dev, "cannot allocate Rx filters\n"); 405 goto fail; 406 } 407 408 error = vtnet_alloc_rxtx_queues(sc); 409 if (error) { 410 device_printf(dev, "cannot allocate queues\n"); 411 goto fail; 412 } 413 414 error = vtnet_alloc_virtqueues(sc); 415 if (error) { 416 device_printf(dev, "cannot allocate virtqueues\n"); 417 goto fail; 418 } 419 420 error = vtnet_setup_interface(sc); 421 if (error) { 422 device_printf(dev, "cannot setup interface\n"); 423 goto fail; 424 } 425 426 error = virtio_setup_intr(dev, INTR_TYPE_NET); 427 if (error) { 428 device_printf(dev, "cannot setup virtqueue interrupts\n"); 429 /* BMV: This will crash if during boot! */ 430 ether_ifdetach(sc->vtnet_ifp); 431 goto fail; 432 } 433 434 #ifdef DEV_NETMAP 435 vtnet_netmap_attach(sc); 436 #endif /* DEV_NETMAP */ 437 438 vtnet_start_taskqueues(sc); 439 440 fail: 441 if (error) 442 vtnet_detach(dev); 443 444 return (error); 445 } 446 447 static int 448 vtnet_detach(device_t dev) 449 { 450 struct vtnet_softc *sc; 451 struct ifnet *ifp; 452 453 sc = device_get_softc(dev); 454 ifp = sc->vtnet_ifp; 455 456 if (device_is_attached(dev)) { 457 VTNET_CORE_LOCK(sc); 458 vtnet_stop(sc); 459 VTNET_CORE_UNLOCK(sc); 460 461 callout_drain(&sc->vtnet_tick_ch); 462 vtnet_drain_taskqueues(sc); 463 464 ether_ifdetach(ifp); 465 } 466 467 #ifdef DEV_NETMAP 468 netmap_detach(ifp); 469 #endif /* DEV_NETMAP */ 470 471 vtnet_free_taskqueues(sc); 472 473 if (sc->vtnet_vlan_attach != NULL) { 474 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 475 sc->vtnet_vlan_attach = NULL; 476 } 477 if (sc->vtnet_vlan_detach != NULL) { 478 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach); 479 sc->vtnet_vlan_detach = NULL; 480 } 481 482 ifmedia_removeall(&sc->vtnet_media); 483 484 if (ifp != NULL) { 485 if_free(ifp); 486 sc->vtnet_ifp = NULL; 487 } 488 489 vtnet_free_rxtx_queues(sc); 490 vtnet_free_rx_filters(sc); 491 492 if (sc->vtnet_ctrl_vq != NULL) 493 vtnet_free_ctrl_vq(sc); 494 495 VTNET_CORE_LOCK_DESTROY(sc); 496 497 return (0); 498 } 499 500 static int 501 vtnet_suspend(device_t dev) 502 { 503 struct vtnet_softc *sc; 504 505 sc = device_get_softc(dev); 506 507 VTNET_CORE_LOCK(sc); 508 vtnet_stop(sc); 509 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 510 VTNET_CORE_UNLOCK(sc); 511 512 return (0); 513 } 514 515 static int 516 vtnet_resume(device_t dev) 517 { 518 struct vtnet_softc *sc; 519 struct ifnet *ifp; 520 521 sc = device_get_softc(dev); 522 ifp = sc->vtnet_ifp; 523 524 VTNET_CORE_LOCK(sc); 525 if (ifp->if_flags & IFF_UP) 526 vtnet_init_locked(sc, 0); 527 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 528 VTNET_CORE_UNLOCK(sc); 529 530 return (0); 531 } 532 533 static int 534 vtnet_shutdown(device_t dev) 535 { 536 537 /* 538 * Suspend already does all of what we need to 539 * do here; we just never expect to be resumed. 540 */ 541 return (vtnet_suspend(dev)); 542 } 543 544 static int 545 vtnet_attach_completed(device_t dev) 546 { 547 548 vtnet_attach_disable_promisc(device_get_softc(dev)); 549 550 return (0); 551 } 552 553 static int 554 vtnet_config_change(device_t dev) 555 { 556 struct vtnet_softc *sc; 557 558 sc = device_get_softc(dev); 559 560 VTNET_CORE_LOCK(sc); 561 vtnet_update_link_status(sc); 562 if (sc->vtnet_link_active != 0) 563 vtnet_tx_start_all(sc); 564 VTNET_CORE_UNLOCK(sc); 565 566 return (0); 567 } 568 569 static void 570 vtnet_negotiate_features(struct vtnet_softc *sc) 571 { 572 device_t dev; 573 uint64_t mask, features; 574 575 dev = sc->vtnet_dev; 576 mask = 0; 577 578 /* 579 * TSO and LRO are only available when their corresponding checksum 580 * offload feature is also negotiated. 581 */ 582 if (vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable)) { 583 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM; 584 mask |= VTNET_TSO_FEATURES | VTNET_LRO_FEATURES; 585 } 586 if (vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) 587 mask |= VTNET_TSO_FEATURES; 588 if (vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) 589 mask |= VTNET_LRO_FEATURES; 590 #ifndef VTNET_LEGACY_TX 591 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 592 mask |= VIRTIO_NET_F_MQ; 593 #else 594 mask |= VIRTIO_NET_F_MQ; 595 #endif 596 597 features = VTNET_FEATURES & ~mask; 598 sc->vtnet_features = virtio_negotiate_features(dev, features); 599 600 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 601 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 602 /* 603 * LRO without mergeable buffers requires special care. This 604 * is not ideal because every receive buffer must be large 605 * enough to hold the maximum TCP packet, the Ethernet header, 606 * and the header. This requires up to 34 descriptors with 607 * MCLBYTES clusters. If we do not have indirect descriptors, 608 * LRO is disabled since the virtqueue will not contain very 609 * many receive buffers. 610 */ 611 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 612 device_printf(dev, 613 "LRO disabled due to both mergeable buffers and " 614 "indirect descriptors not negotiated\n"); 615 616 features &= ~VTNET_LRO_FEATURES; 617 sc->vtnet_features = 618 virtio_negotiate_features(dev, features); 619 } else 620 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 621 } 622 } 623 624 static void 625 vtnet_setup_features(struct vtnet_softc *sc) 626 { 627 device_t dev; 628 629 dev = sc->vtnet_dev; 630 631 vtnet_negotiate_features(sc); 632 633 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 634 sc->vtnet_flags |= VTNET_FLAG_INDIRECT; 635 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 636 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 637 638 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 639 /* This feature should always be negotiated. */ 640 sc->vtnet_flags |= VTNET_FLAG_MAC; 641 } 642 643 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 644 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 645 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 646 } else 647 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 648 649 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 650 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS; 651 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 652 sc->vtnet_rx_nsegs = VTNET_MAX_RX_SEGS; 653 else 654 sc->vtnet_rx_nsegs = VTNET_MIN_RX_SEGS; 655 656 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 657 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 658 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 659 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS; 660 else 661 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS; 662 663 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 664 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 665 666 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 667 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 668 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 669 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 670 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 671 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 672 } 673 674 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ) && 675 sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 676 sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev, 677 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 678 } else 679 sc->vtnet_max_vq_pairs = 1; 680 681 if (sc->vtnet_max_vq_pairs > 1) { 682 /* 683 * Limit the maximum number of queue pairs to the lower of 684 * the number of CPUs and the configured maximum. 685 * The actual number of queues that get used may be less. 686 */ 687 int max; 688 689 max = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 690 if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN) { 691 if (max > mp_ncpus) 692 max = mp_ncpus; 693 if (max > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) 694 max = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX; 695 if (max > 1) { 696 sc->vtnet_requested_vq_pairs = max; 697 sc->vtnet_flags |= VTNET_FLAG_MULTIQ; 698 } 699 } 700 } 701 } 702 703 static int 704 vtnet_init_rxq(struct vtnet_softc *sc, int id) 705 { 706 struct vtnet_rxq *rxq; 707 708 rxq = &sc->vtnet_rxqs[id]; 709 710 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 711 device_get_nameunit(sc->vtnet_dev), id); 712 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 713 714 rxq->vtnrx_sc = sc; 715 rxq->vtnrx_id = id; 716 717 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 718 if (rxq->vtnrx_sg == NULL) 719 return (ENOMEM); 720 721 NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 722 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 723 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 724 725 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 726 } 727 728 static int 729 vtnet_init_txq(struct vtnet_softc *sc, int id) 730 { 731 struct vtnet_txq *txq; 732 733 txq = &sc->vtnet_txqs[id]; 734 735 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 736 device_get_nameunit(sc->vtnet_dev), id); 737 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 738 739 txq->vtntx_sc = sc; 740 txq->vtntx_id = id; 741 742 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 743 if (txq->vtntx_sg == NULL) 744 return (ENOMEM); 745 746 #ifndef VTNET_LEGACY_TX 747 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 748 M_NOWAIT, &txq->vtntx_mtx); 749 if (txq->vtntx_br == NULL) 750 return (ENOMEM); 751 752 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 753 #endif 754 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 755 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 756 taskqueue_thread_enqueue, &txq->vtntx_tq); 757 if (txq->vtntx_tq == NULL) 758 return (ENOMEM); 759 760 return (0); 761 } 762 763 static int 764 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 765 { 766 int i, npairs, error; 767 768 npairs = sc->vtnet_max_vq_pairs; 769 770 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 771 M_NOWAIT | M_ZERO); 772 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 773 M_NOWAIT | M_ZERO); 774 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 775 return (ENOMEM); 776 777 for (i = 0; i < npairs; i++) { 778 error = vtnet_init_rxq(sc, i); 779 if (error) 780 return (error); 781 error = vtnet_init_txq(sc, i); 782 if (error) 783 return (error); 784 } 785 786 vtnet_setup_queue_sysctl(sc); 787 788 return (0); 789 } 790 791 static void 792 vtnet_destroy_rxq(struct vtnet_rxq *rxq) 793 { 794 795 rxq->vtnrx_sc = NULL; 796 rxq->vtnrx_id = -1; 797 798 if (rxq->vtnrx_sg != NULL) { 799 sglist_free(rxq->vtnrx_sg); 800 rxq->vtnrx_sg = NULL; 801 } 802 803 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 804 mtx_destroy(&rxq->vtnrx_mtx); 805 } 806 807 static void 808 vtnet_destroy_txq(struct vtnet_txq *txq) 809 { 810 811 txq->vtntx_sc = NULL; 812 txq->vtntx_id = -1; 813 814 if (txq->vtntx_sg != NULL) { 815 sglist_free(txq->vtntx_sg); 816 txq->vtntx_sg = NULL; 817 } 818 819 #ifndef VTNET_LEGACY_TX 820 if (txq->vtntx_br != NULL) { 821 buf_ring_free(txq->vtntx_br, M_DEVBUF); 822 txq->vtntx_br = NULL; 823 } 824 #endif 825 826 if (mtx_initialized(&txq->vtntx_mtx) != 0) 827 mtx_destroy(&txq->vtntx_mtx); 828 } 829 830 static void 831 vtnet_free_rxtx_queues(struct vtnet_softc *sc) 832 { 833 int i; 834 835 if (sc->vtnet_rxqs != NULL) { 836 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 837 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 838 free(sc->vtnet_rxqs, M_DEVBUF); 839 sc->vtnet_rxqs = NULL; 840 } 841 842 if (sc->vtnet_txqs != NULL) { 843 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 844 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 845 free(sc->vtnet_txqs, M_DEVBUF); 846 sc->vtnet_txqs = NULL; 847 } 848 } 849 850 static int 851 vtnet_alloc_rx_filters(struct vtnet_softc *sc) 852 { 853 854 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 855 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 856 M_DEVBUF, M_NOWAIT | M_ZERO); 857 if (sc->vtnet_mac_filter == NULL) 858 return (ENOMEM); 859 } 860 861 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 862 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 863 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 864 if (sc->vtnet_vlan_filter == NULL) 865 return (ENOMEM); 866 } 867 868 return (0); 869 } 870 871 static void 872 vtnet_free_rx_filters(struct vtnet_softc *sc) 873 { 874 875 if (sc->vtnet_mac_filter != NULL) { 876 free(sc->vtnet_mac_filter, M_DEVBUF); 877 sc->vtnet_mac_filter = NULL; 878 } 879 880 if (sc->vtnet_vlan_filter != NULL) { 881 free(sc->vtnet_vlan_filter, M_DEVBUF); 882 sc->vtnet_vlan_filter = NULL; 883 } 884 } 885 886 static int 887 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 888 { 889 device_t dev; 890 struct vq_alloc_info *info; 891 struct vtnet_rxq *rxq; 892 struct vtnet_txq *txq; 893 int i, idx, flags, nvqs, error; 894 895 dev = sc->vtnet_dev; 896 flags = 0; 897 898 nvqs = sc->vtnet_max_vq_pairs * 2; 899 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 900 nvqs++; 901 902 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); 903 if (info == NULL) 904 return (ENOMEM); 905 906 for (i = 0, idx = 0; i < sc->vtnet_max_vq_pairs; i++, idx+=2) { 907 rxq = &sc->vtnet_rxqs[i]; 908 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, 909 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, 910 "%s-%d rx", device_get_nameunit(dev), rxq->vtnrx_id); 911 912 txq = &sc->vtnet_txqs[i]; 913 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs, 914 vtnet_tx_vq_intr, txq, &txq->vtntx_vq, 915 "%s-%d tx", device_get_nameunit(dev), txq->vtntx_id); 916 } 917 918 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 919 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, 920 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); 921 } 922 923 /* 924 * Enable interrupt binding if this is multiqueue. This only matters 925 * when per-vq MSIX is available. 926 */ 927 if (sc->vtnet_flags & VTNET_FLAG_MULTIQ) 928 flags |= 0; 929 930 error = virtio_alloc_virtqueues(dev, flags, nvqs, info); 931 free(info, M_TEMP); 932 933 return (error); 934 } 935 936 static int 937 vtnet_setup_interface(struct vtnet_softc *sc) 938 { 939 device_t dev; 940 struct pfil_head_args pa; 941 struct ifnet *ifp; 942 943 dev = sc->vtnet_dev; 944 945 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER); 946 if (ifp == NULL) { 947 device_printf(dev, "cannot allocate ifnet structure\n"); 948 return (ENOSPC); 949 } 950 951 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 952 ifp->if_baudrate = IF_Gbps(10); /* Approx. */ 953 ifp->if_softc = sc; 954 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST | 955 IFF_KNOWSEPOCH; 956 ifp->if_init = vtnet_init; 957 ifp->if_ioctl = vtnet_ioctl; 958 ifp->if_get_counter = vtnet_get_counter; 959 #ifndef VTNET_LEGACY_TX 960 ifp->if_transmit = vtnet_txq_mq_start; 961 ifp->if_qflush = vtnet_qflush; 962 #else 963 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 964 ifp->if_start = vtnet_start; 965 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1); 966 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1; 967 IFQ_SET_READY(&ifp->if_snd); 968 #endif 969 970 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd, 971 vtnet_ifmedia_sts); 972 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL); 973 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE); 974 975 /* Read (or generate) the MAC address for the adapter. */ 976 vtnet_get_hwaddr(sc); 977 978 ether_ifattach(ifp, sc->vtnet_hwaddr); 979 980 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 981 ifp->if_capabilities |= IFCAP_LINKSTATE; 982 983 /* Tell the upper layer(s) we support long frames. */ 984 ifp->if_hdrlen = sizeof(struct ether_vlan_header); 985 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU; 986 987 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 988 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6; 989 990 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO)) { 991 ifp->if_capabilities |= IFCAP_TSO4 | IFCAP_TSO6; 992 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 993 } else { 994 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 995 ifp->if_capabilities |= IFCAP_TSO4; 996 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 997 ifp->if_capabilities |= IFCAP_TSO6; 998 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 999 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 1000 } 1001 1002 if (ifp->if_capabilities & IFCAP_TSO) 1003 ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 1004 } 1005 1006 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) { 1007 ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6; 1008 1009 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) || 1010 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6)) 1011 ifp->if_capabilities |= IFCAP_LRO; 1012 } 1013 1014 if (ifp->if_capabilities & IFCAP_HWCSUM) { 1015 /* 1016 * VirtIO does not support VLAN tagging, but we can fake 1017 * it by inserting and removing the 802.1Q header during 1018 * transmit and receive. We are then able to do checksum 1019 * offloading of VLAN frames. 1020 */ 1021 ifp->if_capabilities |= 1022 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; 1023 } 1024 1025 ifp->if_capenable = ifp->if_capabilities; 1026 1027 /* 1028 * Capabilities after here are not enabled by default. 1029 */ 1030 1031 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 1032 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 1033 1034 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1035 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1036 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1037 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1038 } 1039 1040 vtnet_set_rx_process_limit(sc); 1041 vtnet_set_tx_intr_threshold(sc); 1042 1043 DEBUGNET_SET(ifp, vtnet); 1044 1045 pa.pa_version = PFIL_VERSION; 1046 pa.pa_flags = PFIL_IN; 1047 pa.pa_type = PFIL_TYPE_ETHERNET; 1048 pa.pa_headname = ifp->if_xname; 1049 sc->vtnet_pfil = pfil_head_register(&pa); 1050 1051 return (0); 1052 } 1053 1054 static int 1055 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu) 1056 { 1057 struct ifnet *ifp; 1058 int frame_size, clsize; 1059 1060 ifp = sc->vtnet_ifp; 1061 1062 if (new_mtu < ETHERMIN || new_mtu > VTNET_MAX_MTU) 1063 return (EINVAL); 1064 1065 frame_size = sc->vtnet_hdr_size + sizeof(struct ether_vlan_header) + 1066 new_mtu; 1067 1068 /* 1069 * Based on the new MTU (and hence frame size) determine which 1070 * cluster size is most appropriate for the receive queues. 1071 */ 1072 if (frame_size <= MCLBYTES) { 1073 clsize = MCLBYTES; 1074 } else if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1075 /* Avoid going past 9K jumbos. */ 1076 if (frame_size > MJUM9BYTES) 1077 return (EINVAL); 1078 clsize = MJUM9BYTES; 1079 } else 1080 clsize = MJUMPAGESIZE; 1081 1082 ifp->if_mtu = new_mtu; 1083 sc->vtnet_rx_new_clsize = clsize; 1084 1085 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1086 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1087 vtnet_init_locked(sc, 0); 1088 } 1089 1090 return (0); 1091 } 1092 1093 static int 1094 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1095 { 1096 struct vtnet_softc *sc; 1097 struct ifreq *ifr; 1098 int reinit, mask, error; 1099 1100 sc = ifp->if_softc; 1101 ifr = (struct ifreq *) data; 1102 error = 0; 1103 1104 switch (cmd) { 1105 case SIOCSIFMTU: 1106 if (ifp->if_mtu != ifr->ifr_mtu) { 1107 VTNET_CORE_LOCK(sc); 1108 error = vtnet_change_mtu(sc, ifr->ifr_mtu); 1109 VTNET_CORE_UNLOCK(sc); 1110 } 1111 break; 1112 1113 case SIOCSIFFLAGS: 1114 VTNET_CORE_LOCK(sc); 1115 if ((ifp->if_flags & IFF_UP) == 0) { 1116 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1117 vtnet_stop(sc); 1118 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1119 if ((ifp->if_flags ^ sc->vtnet_if_flags) & 1120 (IFF_PROMISC | IFF_ALLMULTI)) { 1121 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1122 vtnet_rx_filter(sc); 1123 else { 1124 ifp->if_flags |= IFF_PROMISC; 1125 if ((ifp->if_flags ^ sc->vtnet_if_flags) 1126 & IFF_ALLMULTI) 1127 error = ENOTSUP; 1128 } 1129 } 1130 } else 1131 vtnet_init_locked(sc, 0); 1132 1133 if (error == 0) 1134 sc->vtnet_if_flags = ifp->if_flags; 1135 VTNET_CORE_UNLOCK(sc); 1136 break; 1137 1138 case SIOCADDMULTI: 1139 case SIOCDELMULTI: 1140 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) 1141 break; 1142 VTNET_CORE_LOCK(sc); 1143 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1144 vtnet_rx_filter_mac(sc); 1145 VTNET_CORE_UNLOCK(sc); 1146 break; 1147 1148 case SIOCSIFMEDIA: 1149 case SIOCGIFMEDIA: 1150 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1151 break; 1152 1153 case SIOCSIFCAP: 1154 VTNET_CORE_LOCK(sc); 1155 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1156 1157 if (mask & IFCAP_TXCSUM) 1158 ifp->if_capenable ^= IFCAP_TXCSUM; 1159 if (mask & IFCAP_TXCSUM_IPV6) 1160 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; 1161 if (mask & IFCAP_TSO4) 1162 ifp->if_capenable ^= IFCAP_TSO4; 1163 if (mask & IFCAP_TSO6) 1164 ifp->if_capenable ^= IFCAP_TSO6; 1165 1166 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO | 1167 IFCAP_VLAN_HWFILTER)) { 1168 /* These Rx features require us to renegotiate. */ 1169 reinit = 1; 1170 1171 if (mask & IFCAP_RXCSUM) 1172 ifp->if_capenable ^= IFCAP_RXCSUM; 1173 if (mask & IFCAP_RXCSUM_IPV6) 1174 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; 1175 if (mask & IFCAP_LRO) 1176 ifp->if_capenable ^= IFCAP_LRO; 1177 if (mask & IFCAP_VLAN_HWFILTER) 1178 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 1179 } else 1180 reinit = 0; 1181 1182 if (mask & IFCAP_VLAN_HWTSO) 1183 ifp->if_capenable ^= IFCAP_VLAN_HWTSO; 1184 if (mask & IFCAP_VLAN_HWTAGGING) 1185 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 1186 1187 if (reinit && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { 1188 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1189 vtnet_init_locked(sc, 0); 1190 } 1191 1192 VTNET_CORE_UNLOCK(sc); 1193 VLAN_CAPABILITIES(ifp); 1194 1195 break; 1196 1197 default: 1198 error = ether_ioctl(ifp, cmd, data); 1199 break; 1200 } 1201 1202 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1203 1204 return (error); 1205 } 1206 1207 static int 1208 vtnet_rxq_populate(struct vtnet_rxq *rxq) 1209 { 1210 struct virtqueue *vq; 1211 int nbufs, error; 1212 1213 #ifdef DEV_NETMAP 1214 error = vtnet_netmap_rxq_populate(rxq); 1215 if (error >= 0) 1216 return (error); 1217 #endif /* DEV_NETMAP */ 1218 1219 vq = rxq->vtnrx_vq; 1220 error = ENOSPC; 1221 1222 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1223 error = vtnet_rxq_new_buf(rxq); 1224 if (error) 1225 break; 1226 } 1227 1228 if (nbufs > 0) { 1229 virtqueue_notify(vq); 1230 /* 1231 * EMSGSIZE signifies the virtqueue did not have enough 1232 * entries available to hold the last mbuf. This is not 1233 * an error. 1234 */ 1235 if (error == EMSGSIZE) 1236 error = 0; 1237 } 1238 1239 return (error); 1240 } 1241 1242 static void 1243 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1244 { 1245 struct virtqueue *vq; 1246 struct mbuf *m; 1247 int last; 1248 #ifdef DEV_NETMAP 1249 struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp), 1250 rxq->vtnrx_id, NR_RX); 1251 #else /* !DEV_NETMAP */ 1252 void *kring = NULL; 1253 #endif /* !DEV_NETMAP */ 1254 1255 vq = rxq->vtnrx_vq; 1256 last = 0; 1257 1258 while ((m = virtqueue_drain(vq, &last)) != NULL) { 1259 if (kring == NULL) 1260 m_freem(m); 1261 } 1262 1263 KASSERT(virtqueue_empty(vq), 1264 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1265 } 1266 1267 static struct mbuf * 1268 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1269 { 1270 struct mbuf *m_head, *m_tail, *m; 1271 int i, clsize; 1272 1273 clsize = sc->vtnet_rx_clsize; 1274 1275 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1276 ("%s: chained mbuf %d request without LRO_NOMRG", __func__, nbufs)); 1277 1278 m_head = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, clsize); 1279 if (m_head == NULL) 1280 goto fail; 1281 1282 m_head->m_len = clsize; 1283 m_tail = m_head; 1284 1285 /* Allocate the rest of the chain. */ 1286 for (i = 1; i < nbufs; i++) { 1287 m = m_getjcl(M_NOWAIT, MT_DATA, 0, clsize); 1288 if (m == NULL) 1289 goto fail; 1290 1291 m->m_len = clsize; 1292 m_tail->m_next = m; 1293 m_tail = m; 1294 } 1295 1296 if (m_tailp != NULL) 1297 *m_tailp = m_tail; 1298 1299 return (m_head); 1300 1301 fail: 1302 sc->vtnet_stats.mbuf_alloc_failed++; 1303 m_freem(m_head); 1304 1305 return (NULL); 1306 } 1307 1308 /* 1309 * Slow path for when LRO without mergeable buffers is negotiated. 1310 */ 1311 static int 1312 vtnet_rxq_replace_lro_nomgr_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1313 int len0) 1314 { 1315 struct vtnet_softc *sc; 1316 struct mbuf *m, *m_prev; 1317 struct mbuf *m_new, *m_tail; 1318 int len, clsize, nreplace, error; 1319 1320 sc = rxq->vtnrx_sc; 1321 clsize = sc->vtnet_rx_clsize; 1322 1323 m_prev = NULL; 1324 m_tail = NULL; 1325 nreplace = 0; 1326 1327 m = m0; 1328 len = len0; 1329 1330 /* 1331 * Since these mbuf chains are so large, we avoid allocating an 1332 * entire replacement chain if possible. When the received frame 1333 * did not consume the entire chain, the unused mbufs are moved 1334 * to the replacement chain. 1335 */ 1336 while (len > 0) { 1337 /* 1338 * Something is seriously wrong if we received a frame 1339 * larger than the chain. Drop it. 1340 */ 1341 if (m == NULL) { 1342 sc->vtnet_stats.rx_frame_too_large++; 1343 return (EMSGSIZE); 1344 } 1345 1346 /* We always allocate the same cluster size. */ 1347 KASSERT(m->m_len == clsize, 1348 ("%s: mbuf size %d is not the cluster size %d", 1349 __func__, m->m_len, clsize)); 1350 1351 m->m_len = MIN(m->m_len, len); 1352 len -= m->m_len; 1353 1354 m_prev = m; 1355 m = m->m_next; 1356 nreplace++; 1357 } 1358 1359 KASSERT(nreplace <= sc->vtnet_rx_nmbufs, 1360 ("%s: too many replacement mbufs %d max %d", __func__, nreplace, 1361 sc->vtnet_rx_nmbufs)); 1362 1363 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1364 if (m_new == NULL) { 1365 m_prev->m_len = clsize; 1366 return (ENOBUFS); 1367 } 1368 1369 /* 1370 * Move any unused mbufs from the received chain onto the end 1371 * of the new chain. 1372 */ 1373 if (m_prev->m_next != NULL) { 1374 m_tail->m_next = m_prev->m_next; 1375 m_prev->m_next = NULL; 1376 } 1377 1378 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1379 if (error) { 1380 /* 1381 * BAD! We could not enqueue the replacement mbuf chain. We 1382 * must restore the m0 chain to the original state if it was 1383 * modified so we can subsequently discard it. 1384 * 1385 * NOTE: The replacement is suppose to be an identical copy 1386 * to the one just dequeued so this is an unexpected error. 1387 */ 1388 sc->vtnet_stats.rx_enq_replacement_failed++; 1389 1390 if (m_tail->m_next != NULL) { 1391 m_prev->m_next = m_tail->m_next; 1392 m_tail->m_next = NULL; 1393 } 1394 1395 m_prev->m_len = clsize; 1396 m_freem(m_new); 1397 } 1398 1399 return (error); 1400 } 1401 1402 static int 1403 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1404 { 1405 struct vtnet_softc *sc; 1406 struct mbuf *m_new; 1407 int error; 1408 1409 sc = rxq->vtnrx_sc; 1410 1411 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1412 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1413 1414 if (m->m_next == NULL) { 1415 /* Fast-path for the common case of just one mbuf. */ 1416 if (m->m_len < len) 1417 return (EINVAL); 1418 1419 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1420 if (m_new == NULL) 1421 return (ENOBUFS); 1422 1423 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1424 if (error) { 1425 /* 1426 * The new mbuf is suppose to be an identical 1427 * copy of the one just dequeued so this is an 1428 * unexpected error. 1429 */ 1430 m_freem(m_new); 1431 sc->vtnet_stats.rx_enq_replacement_failed++; 1432 } else 1433 m->m_len = len; 1434 } else 1435 error = vtnet_rxq_replace_lro_nomgr_buf(rxq, m, len); 1436 1437 return (error); 1438 } 1439 1440 static int 1441 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1442 { 1443 struct vtnet_softc *sc; 1444 struct sglist *sg; 1445 struct vtnet_rx_header *rxhdr; 1446 uint8_t *mdata; 1447 int offset, error; 1448 1449 sc = rxq->vtnrx_sc; 1450 sg = rxq->vtnrx_sg; 1451 mdata = mtod(m, uint8_t *); 1452 1453 VTNET_RXQ_LOCK_ASSERT(rxq); 1454 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG || m->m_next == NULL, 1455 ("%s: chained mbuf without LRO_NOMRG", __func__)); 1456 KASSERT(m->m_len == sc->vtnet_rx_clsize, 1457 ("%s: unexpected cluster size %d/%d", __func__, m->m_len, 1458 sc->vtnet_rx_clsize)); 1459 1460 sglist_reset(sg); 1461 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1462 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1463 rxhdr = (struct vtnet_rx_header *) mdata; 1464 sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1465 offset = sizeof(struct vtnet_rx_header); 1466 } else 1467 offset = 0; 1468 1469 sglist_append(sg, mdata + offset, m->m_len - offset); 1470 if (m->m_next != NULL) { 1471 error = sglist_append_mbuf(sg, m->m_next); 1472 MPASS(error == 0); 1473 } 1474 1475 error = virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg); 1476 1477 return (error); 1478 } 1479 1480 static int 1481 vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1482 { 1483 struct vtnet_softc *sc; 1484 struct mbuf *m; 1485 int error; 1486 1487 sc = rxq->vtnrx_sc; 1488 1489 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1490 if (m == NULL) 1491 return (ENOBUFS); 1492 1493 error = vtnet_rxq_enqueue_buf(rxq, m); 1494 if (error) 1495 m_freem(m); 1496 1497 return (error); 1498 } 1499 1500 /* 1501 * Use the checksum offset in the VirtIO header to set the 1502 * correct CSUM_* flags. 1503 */ 1504 static int 1505 vtnet_rxq_csum_by_offset(struct vtnet_rxq *rxq, struct mbuf *m, 1506 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1507 { 1508 struct vtnet_softc *sc; 1509 #if defined(INET) || defined(INET6) 1510 int offset = hdr->csum_start + hdr->csum_offset; 1511 #endif 1512 1513 sc = rxq->vtnrx_sc; 1514 1515 /* Only do a basic sanity check on the offset. */ 1516 switch (eth_type) { 1517 #if defined(INET) 1518 case ETHERTYPE_IP: 1519 if (__predict_false(offset < ip_start + sizeof(struct ip))) 1520 return (1); 1521 break; 1522 #endif 1523 #if defined(INET6) 1524 case ETHERTYPE_IPV6: 1525 if (__predict_false(offset < ip_start + sizeof(struct ip6_hdr))) 1526 return (1); 1527 break; 1528 #endif 1529 default: 1530 sc->vtnet_stats.rx_csum_bad_ethtype++; 1531 return (1); 1532 } 1533 1534 /* 1535 * Use the offset to determine the appropriate CSUM_* flags. This is 1536 * a bit dirty, but we can get by with it since the checksum offsets 1537 * happen to be different. We assume the host host does not do IPv4 1538 * header checksum offloading. 1539 */ 1540 switch (hdr->csum_offset) { 1541 case offsetof(struct udphdr, uh_sum): 1542 case offsetof(struct tcphdr, th_sum): 1543 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1544 m->m_pkthdr.csum_data = 0xFFFF; 1545 break; 1546 default: 1547 sc->vtnet_stats.rx_csum_bad_offset++; 1548 return (1); 1549 } 1550 1551 return (0); 1552 } 1553 1554 static int 1555 vtnet_rxq_csum_by_parse(struct vtnet_rxq *rxq, struct mbuf *m, 1556 uint16_t eth_type, int ip_start, struct virtio_net_hdr *hdr) 1557 { 1558 struct vtnet_softc *sc; 1559 int offset, proto; 1560 1561 sc = rxq->vtnrx_sc; 1562 1563 switch (eth_type) { 1564 #if defined(INET) 1565 case ETHERTYPE_IP: { 1566 struct ip *ip; 1567 if (__predict_false(m->m_len < ip_start + sizeof(struct ip))) 1568 return (1); 1569 ip = (struct ip *)(m->m_data + ip_start); 1570 proto = ip->ip_p; 1571 offset = ip_start + (ip->ip_hl << 2); 1572 break; 1573 } 1574 #endif 1575 #if defined(INET6) 1576 case ETHERTYPE_IPV6: 1577 if (__predict_false(m->m_len < ip_start + 1578 sizeof(struct ip6_hdr))) 1579 return (1); 1580 offset = ip6_lasthdr(m, ip_start, IPPROTO_IPV6, &proto); 1581 if (__predict_false(offset < 0)) 1582 return (1); 1583 break; 1584 #endif 1585 default: 1586 sc->vtnet_stats.rx_csum_bad_ethtype++; 1587 return (1); 1588 } 1589 1590 switch (proto) { 1591 case IPPROTO_TCP: 1592 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) 1593 return (1); 1594 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1595 m->m_pkthdr.csum_data = 0xFFFF; 1596 break; 1597 case IPPROTO_UDP: 1598 if (__predict_false(m->m_len < offset + sizeof(struct udphdr))) 1599 return (1); 1600 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1601 m->m_pkthdr.csum_data = 0xFFFF; 1602 break; 1603 default: 1604 /* 1605 * For the remaining protocols, FreeBSD does not support 1606 * checksum offloading, so the checksum will be recomputed. 1607 */ 1608 #if 0 1609 if_printf(sc->vtnet_ifp, "cksum offload of unsupported " 1610 "protocol eth_type=%#x proto=%d csum_start=%d " 1611 "csum_offset=%d\n", __func__, eth_type, proto, 1612 hdr->csum_start, hdr->csum_offset); 1613 #endif 1614 break; 1615 } 1616 1617 return (0); 1618 } 1619 1620 /* 1621 * Set the appropriate CSUM_* flags. Unfortunately, the information 1622 * provided is not directly useful to us. The VirtIO header gives the 1623 * offset of the checksum, which is all Linux needs, but this is not 1624 * how FreeBSD does things. We are forced to peek inside the packet 1625 * a bit. 1626 * 1627 * It would be nice if VirtIO gave us the L4 protocol or if FreeBSD 1628 * could accept the offsets and let the stack figure it out. 1629 */ 1630 static int 1631 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1632 struct virtio_net_hdr *hdr) 1633 { 1634 struct ether_header *eh; 1635 struct ether_vlan_header *evh; 1636 uint16_t eth_type; 1637 int offset, error; 1638 1639 eh = mtod(m, struct ether_header *); 1640 eth_type = ntohs(eh->ether_type); 1641 if (eth_type == ETHERTYPE_VLAN) { 1642 /* BMV: We should handle nested VLAN tags too. */ 1643 evh = mtod(m, struct ether_vlan_header *); 1644 eth_type = ntohs(evh->evl_proto); 1645 offset = sizeof(struct ether_vlan_header); 1646 } else 1647 offset = sizeof(struct ether_header); 1648 1649 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1650 error = vtnet_rxq_csum_by_offset(rxq, m, eth_type, offset, hdr); 1651 else 1652 error = vtnet_rxq_csum_by_parse(rxq, m, eth_type, offset, hdr); 1653 1654 return (error); 1655 } 1656 1657 static void 1658 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1659 { 1660 struct mbuf *m; 1661 1662 while (--nbufs > 0) { 1663 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1664 if (m == NULL) 1665 break; 1666 vtnet_rxq_discard_buf(rxq, m); 1667 } 1668 } 1669 1670 static void 1671 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1672 { 1673 int error; 1674 1675 /* 1676 * Requeue the discarded mbuf. This should always be successful 1677 * since it was just dequeued. 1678 */ 1679 error = vtnet_rxq_enqueue_buf(rxq, m); 1680 KASSERT(error == 0, 1681 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1682 } 1683 1684 static int 1685 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1686 { 1687 struct vtnet_softc *sc; 1688 struct virtqueue *vq; 1689 struct mbuf *m, *m_tail; 1690 int len; 1691 1692 sc = rxq->vtnrx_sc; 1693 vq = rxq->vtnrx_vq; 1694 m_tail = m_head; 1695 1696 while (--nbufs > 0) { 1697 m = virtqueue_dequeue(vq, &len); 1698 if (m == NULL) { 1699 rxq->vtnrx_stats.vrxs_ierrors++; 1700 goto fail; 1701 } 1702 1703 if (vtnet_rxq_new_buf(rxq) != 0) { 1704 rxq->vtnrx_stats.vrxs_iqdrops++; 1705 vtnet_rxq_discard_buf(rxq, m); 1706 if (nbufs > 1) 1707 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1708 goto fail; 1709 } 1710 1711 if (m->m_len < len) 1712 len = m->m_len; 1713 1714 m->m_len = len; 1715 m->m_flags &= ~M_PKTHDR; 1716 1717 m_head->m_pkthdr.len += len; 1718 m_tail->m_next = m; 1719 m_tail = m; 1720 } 1721 1722 return (0); 1723 1724 fail: 1725 sc->vtnet_stats.rx_mergeable_failed++; 1726 m_freem(m_head); 1727 1728 return (1); 1729 } 1730 1731 static void 1732 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 1733 struct virtio_net_hdr *hdr) 1734 { 1735 struct vtnet_softc *sc; 1736 struct ifnet *ifp; 1737 struct ether_header *eh; 1738 1739 sc = rxq->vtnrx_sc; 1740 ifp = sc->vtnet_ifp; 1741 1742 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { 1743 eh = mtod(m, struct ether_header *); 1744 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1745 vtnet_vlan_tag_remove(m); 1746 /* 1747 * With the 802.1Q header removed, update the 1748 * checksum starting location accordingly. 1749 */ 1750 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1751 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 1752 } 1753 } 1754 1755 m->m_pkthdr.flowid = rxq->vtnrx_id; 1756 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 1757 1758 /* 1759 * BMV: FreeBSD does not have the UNNECESSARY and PARTIAL checksum 1760 * distinction that Linux does. Need to reevaluate if performing 1761 * offloading for the NEEDS_CSUM case is really appropriate. 1762 */ 1763 if (hdr->flags & (VIRTIO_NET_HDR_F_NEEDS_CSUM | 1764 VIRTIO_NET_HDR_F_DATA_VALID)) { 1765 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 1766 rxq->vtnrx_stats.vrxs_csum++; 1767 else 1768 rxq->vtnrx_stats.vrxs_csum_failed++; 1769 } 1770 1771 rxq->vtnrx_stats.vrxs_ipackets++; 1772 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 1773 1774 VTNET_RXQ_UNLOCK(rxq); 1775 (*ifp->if_input)(ifp, m); 1776 VTNET_RXQ_LOCK(rxq); 1777 } 1778 1779 static int 1780 vtnet_rxq_eof(struct vtnet_rxq *rxq) 1781 { 1782 struct virtio_net_hdr lhdr, *hdr; 1783 struct vtnet_softc *sc; 1784 struct ifnet *ifp; 1785 struct virtqueue *vq; 1786 struct mbuf *m, *mr; 1787 struct virtio_net_hdr_mrg_rxbuf *mhdr; 1788 int len, deq, nbufs, adjsz, count; 1789 pfil_return_t pfil; 1790 bool pfil_done; 1791 1792 sc = rxq->vtnrx_sc; 1793 vq = rxq->vtnrx_vq; 1794 ifp = sc->vtnet_ifp; 1795 hdr = &lhdr; 1796 deq = 0; 1797 count = sc->vtnet_rx_process_limit; 1798 1799 VTNET_RXQ_LOCK_ASSERT(rxq); 1800 1801 while (count-- > 0) { 1802 m = virtqueue_dequeue(vq, &len); 1803 if (m == NULL) 1804 break; 1805 deq++; 1806 1807 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 1808 rxq->vtnrx_stats.vrxs_ierrors++; 1809 vtnet_rxq_discard_buf(rxq, m); 1810 continue; 1811 } 1812 1813 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) { 1814 nbufs = 1; 1815 adjsz = sizeof(struct vtnet_rx_header); 1816 /* 1817 * Account for our pad inserted between the header 1818 * and the actual start of the frame. 1819 */ 1820 len += VTNET_RX_HEADER_PAD; 1821 } else { 1822 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 1823 nbufs = mhdr->num_buffers; 1824 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 1825 } 1826 1827 /* 1828 * If we have enough data in first mbuf, run it through 1829 * pfil as a memory buffer before dequeueing the rest. 1830 */ 1831 if (PFIL_HOOKED_IN(sc->vtnet_pfil) && 1832 len - adjsz >= ETHER_HDR_LEN + max_protohdr) { 1833 pfil = pfil_run_hooks(sc->vtnet_pfil, 1834 m->m_data + adjsz, ifp, 1835 (len - adjsz) | PFIL_MEMPTR | PFIL_IN, NULL); 1836 switch (pfil) { 1837 case PFIL_REALLOCED: 1838 mr = pfil_mem2mbuf(m->m_data + adjsz); 1839 vtnet_rxq_input(rxq, mr, hdr); 1840 /* FALLTHROUGH */ 1841 case PFIL_DROPPED: 1842 case PFIL_CONSUMED: 1843 vtnet_rxq_discard_buf(rxq, m); 1844 if (nbufs > 1) 1845 vtnet_rxq_discard_merged_bufs(rxq, 1846 nbufs); 1847 continue; 1848 default: 1849 KASSERT(pfil == PFIL_PASS, 1850 ("Filter returned %d!\n", pfil)); 1851 }; 1852 pfil_done = true; 1853 } else 1854 pfil_done = false; 1855 1856 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 1857 rxq->vtnrx_stats.vrxs_iqdrops++; 1858 vtnet_rxq_discard_buf(rxq, m); 1859 if (nbufs > 1) 1860 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1861 continue; 1862 } 1863 1864 m->m_pkthdr.len = len; 1865 m->m_pkthdr.rcvif = ifp; 1866 m->m_pkthdr.csum_flags = 0; 1867 1868 if (nbufs > 1) { 1869 /* Dequeue the rest of chain. */ 1870 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 1871 continue; 1872 } 1873 1874 /* 1875 * Save copy of header before we strip it. For both mergeable 1876 * and non-mergeable, the header is at the beginning of the 1877 * mbuf data. We no longer need num_buffers, so always use a 1878 * regular header. 1879 * 1880 * BMV: Is this memcpy() expensive? We know the mbuf data is 1881 * still valid even after the m_adj(). 1882 */ 1883 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr)); 1884 m_adj(m, adjsz); 1885 1886 if (PFIL_HOOKED_IN(sc->vtnet_pfil) && pfil_done == false) { 1887 pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN, 1888 NULL); 1889 switch (pfil) { 1890 case PFIL_DROPPED: 1891 case PFIL_CONSUMED: 1892 continue; 1893 default: 1894 KASSERT(pfil == PFIL_PASS, 1895 ("Filter returned %d!\n", pfil)); 1896 } 1897 } 1898 1899 vtnet_rxq_input(rxq, m, hdr); 1900 1901 /* Must recheck after dropping the Rx lock. */ 1902 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) 1903 break; 1904 } 1905 1906 if (deq > 0) 1907 virtqueue_notify(vq); 1908 1909 return (count > 0 ? 0 : EAGAIN); 1910 } 1911 1912 static void 1913 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries) 1914 { 1915 struct vtnet_softc *sc; 1916 struct ifnet *ifp; 1917 int more; 1918 #ifdef DEV_NETMAP 1919 int nmirq; 1920 #endif /* DEV_NETMAP */ 1921 1922 sc = rxq->vtnrx_sc; 1923 ifp = sc->vtnet_ifp; 1924 1925 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 1926 /* 1927 * Ignore this interrupt. Either this is a spurious interrupt 1928 * or multiqueue without per-VQ MSIX so every queue needs to 1929 * be polled (a brain dead configuration we could try harder 1930 * to avoid). 1931 */ 1932 vtnet_rxq_disable_intr(rxq); 1933 return; 1934 } 1935 1936 VTNET_RXQ_LOCK(rxq); 1937 1938 #ifdef DEV_NETMAP 1939 /* 1940 * We call netmap_rx_irq() under lock to prevent concurrent calls. 1941 * This is not necessary to serialize the access to the RX vq, but 1942 * rather to avoid races that may happen if this interface is 1943 * attached to a VALE switch, which would cause received packets 1944 * to stall in the RX queue (nm_kr_tryget() could find the kring 1945 * busy when called from netmap_bwrap_intr_notify()). 1946 */ 1947 nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more); 1948 if (nmirq != NM_IRQ_PASS) { 1949 VTNET_RXQ_UNLOCK(rxq); 1950 if (nmirq == NM_IRQ_RESCHED) { 1951 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1952 } 1953 return; 1954 } 1955 #endif /* DEV_NETMAP */ 1956 1957 again: 1958 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 1959 VTNET_RXQ_UNLOCK(rxq); 1960 return; 1961 } 1962 1963 more = vtnet_rxq_eof(rxq); 1964 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 1965 if (!more) 1966 vtnet_rxq_disable_intr(rxq); 1967 /* 1968 * This is an occasional condition or race (when !more), 1969 * so retry a few times before scheduling the taskqueue. 1970 */ 1971 if (tries-- > 0) 1972 goto again; 1973 1974 rxq->vtnrx_stats.vrxs_rescheduled++; 1975 VTNET_RXQ_UNLOCK(rxq); 1976 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 1977 } else 1978 VTNET_RXQ_UNLOCK(rxq); 1979 } 1980 1981 static void 1982 vtnet_rx_vq_intr(void *xrxq) 1983 { 1984 struct vtnet_rxq *rxq; 1985 1986 rxq = xrxq; 1987 vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES); 1988 } 1989 1990 static void 1991 vtnet_rxq_tq_intr(void *xrxq, int pending) 1992 { 1993 struct vtnet_rxq *rxq; 1994 1995 rxq = xrxq; 1996 vtnet_rx_vq_process(rxq, 0); 1997 } 1998 1999 static int 2000 vtnet_txq_below_threshold(struct vtnet_txq *txq) 2001 { 2002 struct vtnet_softc *sc; 2003 struct virtqueue *vq; 2004 2005 sc = txq->vtntx_sc; 2006 vq = txq->vtntx_vq; 2007 2008 return (virtqueue_nfree(vq) <= sc->vtnet_tx_intr_thresh); 2009 } 2010 2011 static int 2012 vtnet_txq_notify(struct vtnet_txq *txq) 2013 { 2014 struct virtqueue *vq; 2015 2016 vq = txq->vtntx_vq; 2017 2018 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2019 virtqueue_notify(vq); 2020 2021 if (vtnet_txq_enable_intr(txq) == 0) 2022 return (0); 2023 2024 /* 2025 * Drain frames that were completed since last checked. If this 2026 * causes the queue to go above the threshold, the caller should 2027 * continue transmitting. 2028 */ 2029 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) { 2030 virtqueue_disable_intr(vq); 2031 return (1); 2032 } 2033 2034 return (0); 2035 } 2036 2037 static void 2038 vtnet_txq_free_mbufs(struct vtnet_txq *txq) 2039 { 2040 struct virtqueue *vq; 2041 struct vtnet_tx_header *txhdr; 2042 int last; 2043 #ifdef DEV_NETMAP 2044 struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp), 2045 txq->vtntx_id, NR_TX); 2046 #else /* !DEV_NETMAP */ 2047 void *kring = NULL; 2048 #endif /* !DEV_NETMAP */ 2049 2050 vq = txq->vtntx_vq; 2051 last = 0; 2052 2053 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 2054 if (kring == NULL) { 2055 m_freem(txhdr->vth_mbuf); 2056 uma_zfree(vtnet_tx_header_zone, txhdr); 2057 } 2058 } 2059 2060 KASSERT(virtqueue_empty(vq), 2061 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 2062 } 2063 2064 /* 2065 * BMV: Much of this can go away once we finally have offsets in 2066 * the mbuf packet header. Bug andre@. 2067 */ 2068 static int 2069 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, 2070 int *etype, int *proto, int *start) 2071 { 2072 struct vtnet_softc *sc; 2073 struct ether_vlan_header *evh; 2074 int offset; 2075 2076 sc = txq->vtntx_sc; 2077 2078 evh = mtod(m, struct ether_vlan_header *); 2079 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 2080 /* BMV: We should handle nested VLAN tags too. */ 2081 *etype = ntohs(evh->evl_proto); 2082 offset = sizeof(struct ether_vlan_header); 2083 } else { 2084 *etype = ntohs(evh->evl_encap_proto); 2085 offset = sizeof(struct ether_header); 2086 } 2087 2088 switch (*etype) { 2089 #if defined(INET) 2090 case ETHERTYPE_IP: { 2091 struct ip *ip, iphdr; 2092 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 2093 m_copydata(m, offset, sizeof(struct ip), 2094 (caddr_t) &iphdr); 2095 ip = &iphdr; 2096 } else 2097 ip = (struct ip *)(m->m_data + offset); 2098 *proto = ip->ip_p; 2099 *start = offset + (ip->ip_hl << 2); 2100 break; 2101 } 2102 #endif 2103 #if defined(INET6) 2104 case ETHERTYPE_IPV6: 2105 *proto = -1; 2106 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 2107 /* Assert the network stack sent us a valid packet. */ 2108 KASSERT(*start > offset, 2109 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 2110 *start, offset, *proto)); 2111 break; 2112 #endif 2113 default: 2114 sc->vtnet_stats.tx_csum_bad_ethtype++; 2115 return (EINVAL); 2116 } 2117 2118 return (0); 2119 } 2120 2121 static int 2122 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 2123 int offset, struct virtio_net_hdr *hdr) 2124 { 2125 static struct timeval lastecn; 2126 static int curecn; 2127 struct vtnet_softc *sc; 2128 struct tcphdr *tcp, tcphdr; 2129 2130 sc = txq->vtntx_sc; 2131 2132 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 2133 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 2134 tcp = &tcphdr; 2135 } else 2136 tcp = (struct tcphdr *)(m->m_data + offset); 2137 2138 hdr->hdr_len = offset + (tcp->th_off << 2); 2139 hdr->gso_size = m->m_pkthdr.tso_segsz; 2140 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 2141 VIRTIO_NET_HDR_GSO_TCPV6; 2142 2143 if (tcp->th_flags & TH_CWR) { 2144 /* 2145 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In FreeBSD, 2146 * ECN support is not on a per-interface basis, but globally via 2147 * the net.inet.tcp.ecn.enable sysctl knob. The default is off. 2148 */ 2149 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2150 if (ppsratecheck(&lastecn, &curecn, 1)) 2151 if_printf(sc->vtnet_ifp, 2152 "TSO with ECN not negotiated with host\n"); 2153 return (ENOTSUP); 2154 } 2155 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2156 } 2157 2158 txq->vtntx_stats.vtxs_tso++; 2159 2160 return (0); 2161 } 2162 2163 static struct mbuf * 2164 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2165 struct virtio_net_hdr *hdr) 2166 { 2167 struct vtnet_softc *sc; 2168 int flags, etype, csum_start, proto, error; 2169 2170 sc = txq->vtntx_sc; 2171 flags = m->m_pkthdr.csum_flags; 2172 2173 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2174 if (error) 2175 goto drop; 2176 2177 if ((etype == ETHERTYPE_IP && flags & VTNET_CSUM_OFFLOAD) || 2178 (etype == ETHERTYPE_IPV6 && flags & VTNET_CSUM_OFFLOAD_IPV6)) { 2179 /* 2180 * We could compare the IP protocol vs the CSUM_ flag too, 2181 * but that really should not be necessary. 2182 */ 2183 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2184 hdr->csum_start = csum_start; 2185 hdr->csum_offset = m->m_pkthdr.csum_data; 2186 txq->vtntx_stats.vtxs_csum++; 2187 } 2188 2189 if (flags & CSUM_TSO) { 2190 if (__predict_false(proto != IPPROTO_TCP)) { 2191 /* Likely failed to correctly parse the mbuf. */ 2192 sc->vtnet_stats.tx_tso_not_tcp++; 2193 goto drop; 2194 } 2195 2196 KASSERT(hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM, 2197 ("%s: mbuf %p TSO without checksum offload %#x", 2198 __func__, m, flags)); 2199 2200 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2201 if (error) 2202 goto drop; 2203 } 2204 2205 return (m); 2206 2207 drop: 2208 m_freem(m); 2209 return (NULL); 2210 } 2211 2212 static int 2213 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2214 struct vtnet_tx_header *txhdr) 2215 { 2216 struct vtnet_softc *sc; 2217 struct virtqueue *vq; 2218 struct sglist *sg; 2219 struct mbuf *m; 2220 int error; 2221 2222 sc = txq->vtntx_sc; 2223 vq = txq->vtntx_vq; 2224 sg = txq->vtntx_sg; 2225 m = *m_head; 2226 2227 sglist_reset(sg); 2228 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2229 KASSERT(error == 0 && sg->sg_nseg == 1, 2230 ("%s: error %d adding header to sglist", __func__, error)); 2231 2232 error = sglist_append_mbuf(sg, m); 2233 if (error) { 2234 m = m_defrag(m, M_NOWAIT); 2235 if (m == NULL) 2236 goto fail; 2237 2238 *m_head = m; 2239 sc->vtnet_stats.tx_defragged++; 2240 2241 error = sglist_append_mbuf(sg, m); 2242 if (error) 2243 goto fail; 2244 } 2245 2246 txhdr->vth_mbuf = m; 2247 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2248 2249 return (error); 2250 2251 fail: 2252 sc->vtnet_stats.tx_defrag_failed++; 2253 m_freem(*m_head); 2254 *m_head = NULL; 2255 2256 return (ENOBUFS); 2257 } 2258 2259 static int 2260 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags) 2261 { 2262 struct vtnet_tx_header *txhdr; 2263 struct virtio_net_hdr *hdr; 2264 struct mbuf *m; 2265 int error; 2266 2267 m = *m_head; 2268 M_ASSERTPKTHDR(m); 2269 2270 txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO); 2271 if (txhdr == NULL) { 2272 m_freem(m); 2273 *m_head = NULL; 2274 return (ENOMEM); 2275 } 2276 2277 /* 2278 * Always use the non-mergeable header, regardless if the feature 2279 * was negotiated. For transmit, num_buffers is always zero. The 2280 * vtnet_hdr_size is used to enqueue the correct header size. 2281 */ 2282 hdr = &txhdr->vth_uhdr.hdr; 2283 2284 if (m->m_flags & M_VLANTAG) { 2285 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2286 if ((*m_head = m) == NULL) { 2287 error = ENOBUFS; 2288 goto fail; 2289 } 2290 m->m_flags &= ~M_VLANTAG; 2291 } 2292 2293 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2294 m = vtnet_txq_offload(txq, m, hdr); 2295 if ((*m_head = m) == NULL) { 2296 error = ENOBUFS; 2297 goto fail; 2298 } 2299 } 2300 2301 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2302 if (error == 0) 2303 return (0); 2304 2305 fail: 2306 uma_zfree(vtnet_tx_header_zone, txhdr); 2307 2308 return (error); 2309 } 2310 2311 #ifdef VTNET_LEGACY_TX 2312 2313 static void 2314 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp) 2315 { 2316 struct vtnet_softc *sc; 2317 struct virtqueue *vq; 2318 struct mbuf *m0; 2319 int tries, enq; 2320 2321 sc = txq->vtntx_sc; 2322 vq = txq->vtntx_vq; 2323 tries = 0; 2324 2325 VTNET_TXQ_LOCK_ASSERT(txq); 2326 2327 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2328 sc->vtnet_link_active == 0) 2329 return; 2330 2331 vtnet_txq_eof(txq); 2332 2333 again: 2334 enq = 0; 2335 2336 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { 2337 if (virtqueue_full(vq)) 2338 break; 2339 2340 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0); 2341 if (m0 == NULL) 2342 break; 2343 2344 if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) { 2345 if (m0 != NULL) 2346 IFQ_DRV_PREPEND(&ifp->if_snd, m0); 2347 break; 2348 } 2349 2350 enq++; 2351 ETHER_BPF_MTAP(ifp, m0); 2352 } 2353 2354 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2355 if (tries++ < VTNET_NOTIFY_RETRIES) 2356 goto again; 2357 2358 txq->vtntx_stats.vtxs_rescheduled++; 2359 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2360 } 2361 } 2362 2363 static void 2364 vtnet_start(struct ifnet *ifp) 2365 { 2366 struct vtnet_softc *sc; 2367 struct vtnet_txq *txq; 2368 2369 sc = ifp->if_softc; 2370 txq = &sc->vtnet_txqs[0]; 2371 2372 VTNET_TXQ_LOCK(txq); 2373 vtnet_start_locked(txq, ifp); 2374 VTNET_TXQ_UNLOCK(txq); 2375 } 2376 2377 #else /* !VTNET_LEGACY_TX */ 2378 2379 static int 2380 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2381 { 2382 struct vtnet_softc *sc; 2383 struct virtqueue *vq; 2384 struct buf_ring *br; 2385 struct ifnet *ifp; 2386 int enq, tries, error; 2387 2388 sc = txq->vtntx_sc; 2389 vq = txq->vtntx_vq; 2390 br = txq->vtntx_br; 2391 ifp = sc->vtnet_ifp; 2392 tries = 0; 2393 error = 0; 2394 2395 VTNET_TXQ_LOCK_ASSERT(txq); 2396 2397 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2398 sc->vtnet_link_active == 0) { 2399 if (m != NULL) 2400 error = drbr_enqueue(ifp, br, m); 2401 return (error); 2402 } 2403 2404 if (m != NULL) { 2405 error = drbr_enqueue(ifp, br, m); 2406 if (error) 2407 return (error); 2408 } 2409 2410 vtnet_txq_eof(txq); 2411 2412 again: 2413 enq = 0; 2414 2415 while ((m = drbr_peek(ifp, br)) != NULL) { 2416 if (virtqueue_full(vq)) { 2417 drbr_putback(ifp, br, m); 2418 break; 2419 } 2420 2421 if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) { 2422 if (m != NULL) 2423 drbr_putback(ifp, br, m); 2424 else 2425 drbr_advance(ifp, br); 2426 break; 2427 } 2428 drbr_advance(ifp, br); 2429 2430 enq++; 2431 ETHER_BPF_MTAP(ifp, m); 2432 } 2433 2434 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2435 if (tries++ < VTNET_NOTIFY_RETRIES) 2436 goto again; 2437 2438 txq->vtntx_stats.vtxs_rescheduled++; 2439 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2440 } 2441 2442 return (0); 2443 } 2444 2445 static int 2446 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m) 2447 { 2448 struct vtnet_softc *sc; 2449 struct vtnet_txq *txq; 2450 int i, npairs, error; 2451 2452 sc = ifp->if_softc; 2453 npairs = sc->vtnet_act_vq_pairs; 2454 2455 /* check if flowid is set */ 2456 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2457 i = m->m_pkthdr.flowid % npairs; 2458 else 2459 i = curcpu % npairs; 2460 2461 txq = &sc->vtnet_txqs[i]; 2462 2463 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2464 error = vtnet_txq_mq_start_locked(txq, m); 2465 VTNET_TXQ_UNLOCK(txq); 2466 } else { 2467 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2468 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2469 } 2470 2471 return (error); 2472 } 2473 2474 static void 2475 vtnet_txq_tq_deferred(void *xtxq, int pending) 2476 { 2477 struct vtnet_softc *sc; 2478 struct vtnet_txq *txq; 2479 2480 txq = xtxq; 2481 sc = txq->vtntx_sc; 2482 2483 VTNET_TXQ_LOCK(txq); 2484 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2485 vtnet_txq_mq_start_locked(txq, NULL); 2486 VTNET_TXQ_UNLOCK(txq); 2487 } 2488 2489 #endif /* VTNET_LEGACY_TX */ 2490 2491 static void 2492 vtnet_txq_start(struct vtnet_txq *txq) 2493 { 2494 struct vtnet_softc *sc; 2495 struct ifnet *ifp; 2496 2497 sc = txq->vtntx_sc; 2498 ifp = sc->vtnet_ifp; 2499 2500 #ifdef VTNET_LEGACY_TX 2501 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 2502 vtnet_start_locked(txq, ifp); 2503 #else 2504 if (!drbr_empty(ifp, txq->vtntx_br)) 2505 vtnet_txq_mq_start_locked(txq, NULL); 2506 #endif 2507 } 2508 2509 static void 2510 vtnet_txq_tq_intr(void *xtxq, int pending) 2511 { 2512 struct vtnet_softc *sc; 2513 struct vtnet_txq *txq; 2514 struct ifnet *ifp; 2515 2516 txq = xtxq; 2517 sc = txq->vtntx_sc; 2518 ifp = sc->vtnet_ifp; 2519 2520 VTNET_TXQ_LOCK(txq); 2521 2522 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2523 VTNET_TXQ_UNLOCK(txq); 2524 return; 2525 } 2526 2527 vtnet_txq_eof(txq); 2528 vtnet_txq_start(txq); 2529 2530 VTNET_TXQ_UNLOCK(txq); 2531 } 2532 2533 static int 2534 vtnet_txq_eof(struct vtnet_txq *txq) 2535 { 2536 struct virtqueue *vq; 2537 struct vtnet_tx_header *txhdr; 2538 struct mbuf *m; 2539 int deq; 2540 2541 vq = txq->vtntx_vq; 2542 deq = 0; 2543 VTNET_TXQ_LOCK_ASSERT(txq); 2544 2545 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2546 m = txhdr->vth_mbuf; 2547 deq++; 2548 2549 txq->vtntx_stats.vtxs_opackets++; 2550 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2551 if (m->m_flags & M_MCAST) 2552 txq->vtntx_stats.vtxs_omcasts++; 2553 2554 m_freem(m); 2555 uma_zfree(vtnet_tx_header_zone, txhdr); 2556 } 2557 2558 if (virtqueue_empty(vq)) 2559 txq->vtntx_watchdog = 0; 2560 2561 return (deq); 2562 } 2563 2564 static void 2565 vtnet_tx_vq_intr(void *xtxq) 2566 { 2567 struct vtnet_softc *sc; 2568 struct vtnet_txq *txq; 2569 struct ifnet *ifp; 2570 2571 txq = xtxq; 2572 sc = txq->vtntx_sc; 2573 ifp = sc->vtnet_ifp; 2574 2575 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2576 /* 2577 * Ignore this interrupt. Either this is a spurious interrupt 2578 * or multiqueue without per-VQ MSIX so every queue needs to 2579 * be polled (a brain dead configuration we could try harder 2580 * to avoid). 2581 */ 2582 vtnet_txq_disable_intr(txq); 2583 return; 2584 } 2585 2586 #ifdef DEV_NETMAP 2587 if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS) 2588 return; 2589 #endif /* DEV_NETMAP */ 2590 2591 VTNET_TXQ_LOCK(txq); 2592 2593 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2594 VTNET_TXQ_UNLOCK(txq); 2595 return; 2596 } 2597 2598 vtnet_txq_eof(txq); 2599 vtnet_txq_start(txq); 2600 2601 VTNET_TXQ_UNLOCK(txq); 2602 } 2603 2604 static void 2605 vtnet_tx_start_all(struct vtnet_softc *sc) 2606 { 2607 struct vtnet_txq *txq; 2608 int i; 2609 2610 VTNET_CORE_LOCK_ASSERT(sc); 2611 2612 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2613 txq = &sc->vtnet_txqs[i]; 2614 2615 VTNET_TXQ_LOCK(txq); 2616 vtnet_txq_start(txq); 2617 VTNET_TXQ_UNLOCK(txq); 2618 } 2619 } 2620 2621 #ifndef VTNET_LEGACY_TX 2622 static void 2623 vtnet_qflush(struct ifnet *ifp) 2624 { 2625 struct vtnet_softc *sc; 2626 struct vtnet_txq *txq; 2627 struct mbuf *m; 2628 int i; 2629 2630 sc = ifp->if_softc; 2631 2632 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2633 txq = &sc->vtnet_txqs[i]; 2634 2635 VTNET_TXQ_LOCK(txq); 2636 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2637 m_freem(m); 2638 VTNET_TXQ_UNLOCK(txq); 2639 } 2640 2641 if_qflush(ifp); 2642 } 2643 #endif 2644 2645 static int 2646 vtnet_watchdog(struct vtnet_txq *txq) 2647 { 2648 struct ifnet *ifp; 2649 2650 ifp = txq->vtntx_sc->vtnet_ifp; 2651 2652 VTNET_TXQ_LOCK(txq); 2653 if (txq->vtntx_watchdog == 1) { 2654 /* 2655 * Only drain completed frames if the watchdog is about to 2656 * expire. If any frames were drained, there may be enough 2657 * free descriptors now available to transmit queued frames. 2658 * In that case, the timer will immediately be decremented 2659 * below, but the timeout is generous enough that should not 2660 * be a problem. 2661 */ 2662 if (vtnet_txq_eof(txq) != 0) 2663 vtnet_txq_start(txq); 2664 } 2665 2666 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 2667 VTNET_TXQ_UNLOCK(txq); 2668 return (0); 2669 } 2670 VTNET_TXQ_UNLOCK(txq); 2671 2672 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id); 2673 return (1); 2674 } 2675 2676 static void 2677 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc, 2678 struct vtnet_txq_stats *txacc) 2679 { 2680 2681 bzero(rxacc, sizeof(struct vtnet_rxq_stats)); 2682 bzero(txacc, sizeof(struct vtnet_txq_stats)); 2683 2684 for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2685 struct vtnet_rxq_stats *rxst; 2686 struct vtnet_txq_stats *txst; 2687 2688 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 2689 rxacc->vrxs_ipackets += rxst->vrxs_ipackets; 2690 rxacc->vrxs_ibytes += rxst->vrxs_ibytes; 2691 rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops; 2692 rxacc->vrxs_csum += rxst->vrxs_csum; 2693 rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed; 2694 rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled; 2695 2696 txst = &sc->vtnet_txqs[i].vtntx_stats; 2697 txacc->vtxs_opackets += txst->vtxs_opackets; 2698 txacc->vtxs_obytes += txst->vtxs_obytes; 2699 txacc->vtxs_csum += txst->vtxs_csum; 2700 txacc->vtxs_tso += txst->vtxs_tso; 2701 txacc->vtxs_rescheduled += txst->vtxs_rescheduled; 2702 } 2703 } 2704 2705 static uint64_t 2706 vtnet_get_counter(if_t ifp, ift_counter cnt) 2707 { 2708 struct vtnet_softc *sc; 2709 struct vtnet_rxq_stats rxaccum; 2710 struct vtnet_txq_stats txaccum; 2711 2712 sc = if_getsoftc(ifp); 2713 vtnet_accum_stats(sc, &rxaccum, &txaccum); 2714 2715 switch (cnt) { 2716 case IFCOUNTER_IPACKETS: 2717 return (rxaccum.vrxs_ipackets); 2718 case IFCOUNTER_IQDROPS: 2719 return (rxaccum.vrxs_iqdrops); 2720 case IFCOUNTER_IERRORS: 2721 return (rxaccum.vrxs_ierrors); 2722 case IFCOUNTER_OPACKETS: 2723 return (txaccum.vtxs_opackets); 2724 #ifndef VTNET_LEGACY_TX 2725 case IFCOUNTER_OBYTES: 2726 return (txaccum.vtxs_obytes); 2727 case IFCOUNTER_OMCASTS: 2728 return (txaccum.vtxs_omcasts); 2729 #endif 2730 default: 2731 return (if_get_counter_default(ifp, cnt)); 2732 } 2733 } 2734 2735 static void 2736 vtnet_tick(void *xsc) 2737 { 2738 struct vtnet_softc *sc; 2739 struct ifnet *ifp; 2740 int i, timedout; 2741 2742 sc = xsc; 2743 ifp = sc->vtnet_ifp; 2744 timedout = 0; 2745 2746 VTNET_CORE_LOCK_ASSERT(sc); 2747 2748 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 2749 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 2750 2751 if (timedout != 0) { 2752 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2753 vtnet_init_locked(sc, 0); 2754 } else 2755 callout_schedule(&sc->vtnet_tick_ch, hz); 2756 } 2757 2758 static void 2759 vtnet_start_taskqueues(struct vtnet_softc *sc) 2760 { 2761 device_t dev; 2762 struct vtnet_rxq *rxq; 2763 struct vtnet_txq *txq; 2764 int i, error; 2765 2766 dev = sc->vtnet_dev; 2767 2768 /* 2769 * Errors here are very difficult to recover from - we cannot 2770 * easily fail because, if this is during boot, we will hang 2771 * when freeing any successfully started taskqueues because 2772 * the scheduler isn't up yet. 2773 * 2774 * Most drivers just ignore the return value - it only fails 2775 * with ENOMEM so an error is not likely. 2776 */ 2777 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2778 rxq = &sc->vtnet_rxqs[i]; 2779 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 2780 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 2781 if (error) { 2782 device_printf(dev, "failed to start rx taskq %d\n", 2783 rxq->vtnrx_id); 2784 } 2785 2786 txq = &sc->vtnet_txqs[i]; 2787 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 2788 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 2789 if (error) { 2790 device_printf(dev, "failed to start tx taskq %d\n", 2791 txq->vtntx_id); 2792 } 2793 } 2794 } 2795 2796 static void 2797 vtnet_free_taskqueues(struct vtnet_softc *sc) 2798 { 2799 struct vtnet_rxq *rxq; 2800 struct vtnet_txq *txq; 2801 int i; 2802 2803 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2804 rxq = &sc->vtnet_rxqs[i]; 2805 if (rxq->vtnrx_tq != NULL) { 2806 taskqueue_free(rxq->vtnrx_tq); 2807 rxq->vtnrx_tq = NULL; 2808 } 2809 2810 txq = &sc->vtnet_txqs[i]; 2811 if (txq->vtntx_tq != NULL) { 2812 taskqueue_free(txq->vtntx_tq); 2813 txq->vtntx_tq = NULL; 2814 } 2815 } 2816 } 2817 2818 static void 2819 vtnet_drain_taskqueues(struct vtnet_softc *sc) 2820 { 2821 struct vtnet_rxq *rxq; 2822 struct vtnet_txq *txq; 2823 int i; 2824 2825 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2826 rxq = &sc->vtnet_rxqs[i]; 2827 if (rxq->vtnrx_tq != NULL) 2828 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2829 2830 txq = &sc->vtnet_txqs[i]; 2831 if (txq->vtntx_tq != NULL) { 2832 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 2833 #ifndef VTNET_LEGACY_TX 2834 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 2835 #endif 2836 } 2837 } 2838 } 2839 2840 static void 2841 vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 2842 { 2843 struct vtnet_rxq *rxq; 2844 struct vtnet_txq *txq; 2845 int i; 2846 2847 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2848 rxq = &sc->vtnet_rxqs[i]; 2849 vtnet_rxq_free_mbufs(rxq); 2850 2851 txq = &sc->vtnet_txqs[i]; 2852 vtnet_txq_free_mbufs(txq); 2853 } 2854 } 2855 2856 static void 2857 vtnet_stop_rendezvous(struct vtnet_softc *sc) 2858 { 2859 struct vtnet_rxq *rxq; 2860 struct vtnet_txq *txq; 2861 int i; 2862 2863 /* 2864 * Lock and unlock the per-queue mutex so we known the stop 2865 * state is visible. Doing only the active queues should be 2866 * sufficient, but it does not cost much extra to do all the 2867 * queues. Note we hold the core mutex here too. 2868 */ 2869 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2870 rxq = &sc->vtnet_rxqs[i]; 2871 VTNET_RXQ_LOCK(rxq); 2872 VTNET_RXQ_UNLOCK(rxq); 2873 2874 txq = &sc->vtnet_txqs[i]; 2875 VTNET_TXQ_LOCK(txq); 2876 VTNET_TXQ_UNLOCK(txq); 2877 } 2878 } 2879 2880 static void 2881 vtnet_stop(struct vtnet_softc *sc) 2882 { 2883 device_t dev; 2884 struct ifnet *ifp; 2885 2886 dev = sc->vtnet_dev; 2887 ifp = sc->vtnet_ifp; 2888 2889 VTNET_CORE_LOCK_ASSERT(sc); 2890 2891 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 2892 sc->vtnet_link_active = 0; 2893 callout_stop(&sc->vtnet_tick_ch); 2894 2895 /* Only advisory. */ 2896 vtnet_disable_interrupts(sc); 2897 2898 #ifdef DEV_NETMAP 2899 /* Stop any pending txsync/rxsync and disable them. */ 2900 netmap_disable_all_rings(ifp); 2901 #endif /* DEV_NETMAP */ 2902 2903 /* 2904 * Stop the host adapter. This resets it to the pre-initialized 2905 * state. It will not generate any interrupts until after it is 2906 * reinitialized. 2907 */ 2908 virtio_stop(dev); 2909 vtnet_stop_rendezvous(sc); 2910 2911 /* Free any mbufs left in the virtqueues. */ 2912 vtnet_drain_rxtx_queues(sc); 2913 } 2914 2915 static int 2916 vtnet_virtio_reinit(struct vtnet_softc *sc) 2917 { 2918 device_t dev; 2919 struct ifnet *ifp; 2920 uint64_t features; 2921 int mask, error; 2922 2923 dev = sc->vtnet_dev; 2924 ifp = sc->vtnet_ifp; 2925 features = sc->vtnet_features; 2926 2927 mask = 0; 2928 #if defined(INET) 2929 mask |= IFCAP_RXCSUM; 2930 #endif 2931 #if defined (INET6) 2932 mask |= IFCAP_RXCSUM_IPV6; 2933 #endif 2934 2935 /* 2936 * Re-negotiate with the host, removing any disabled receive 2937 * features. Transmit features are disabled only on our side 2938 * via if_capenable and if_hwassist. 2939 */ 2940 2941 if (ifp->if_capabilities & mask) { 2942 /* 2943 * We require both IPv4 and IPv6 offloading to be enabled 2944 * in order to negotiated it: VirtIO does not distinguish 2945 * between the two. 2946 */ 2947 if ((ifp->if_capenable & mask) != mask) 2948 features &= ~VIRTIO_NET_F_GUEST_CSUM; 2949 } 2950 2951 if (ifp->if_capabilities & IFCAP_LRO) { 2952 if ((ifp->if_capenable & IFCAP_LRO) == 0) 2953 features &= ~VTNET_LRO_FEATURES; 2954 } 2955 2956 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) { 2957 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) 2958 features &= ~VIRTIO_NET_F_CTRL_VLAN; 2959 } 2960 2961 error = virtio_reinit(dev, features); 2962 if (error) 2963 device_printf(dev, "virtio reinit error %d\n", error); 2964 2965 return (error); 2966 } 2967 2968 static void 2969 vtnet_init_rx_filters(struct vtnet_softc *sc) 2970 { 2971 struct ifnet *ifp; 2972 2973 ifp = sc->vtnet_ifp; 2974 2975 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 2976 /* Restore promiscuous and all-multicast modes. */ 2977 vtnet_rx_filter(sc); 2978 /* Restore filtered MAC addresses. */ 2979 vtnet_rx_filter_mac(sc); 2980 } 2981 2982 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) 2983 vtnet_rx_filter_vlan(sc); 2984 } 2985 2986 static int 2987 vtnet_init_rx_queues(struct vtnet_softc *sc) 2988 { 2989 device_t dev; 2990 struct vtnet_rxq *rxq; 2991 int i, clsize, error; 2992 2993 dev = sc->vtnet_dev; 2994 2995 /* 2996 * Use the new cluster size if one has been set (via a MTU 2997 * change). Otherwise, use the standard 2K clusters. 2998 * 2999 * BMV: It might make sense to use page sized clusters as 3000 * the default (depending on the features negotiated). 3001 */ 3002 if (sc->vtnet_rx_new_clsize != 0) { 3003 clsize = sc->vtnet_rx_new_clsize; 3004 sc->vtnet_rx_new_clsize = 0; 3005 } else 3006 clsize = MCLBYTES; 3007 3008 sc->vtnet_rx_clsize = clsize; 3009 sc->vtnet_rx_nmbufs = VTNET_NEEDED_RX_MBUFS(sc, clsize); 3010 3011 KASSERT(sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS || 3012 sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 3013 ("%s: too many rx mbufs %d for %d segments", __func__, 3014 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 3015 3016 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3017 rxq = &sc->vtnet_rxqs[i]; 3018 3019 /* Hold the lock to satisfy asserts. */ 3020 VTNET_RXQ_LOCK(rxq); 3021 error = vtnet_rxq_populate(rxq); 3022 VTNET_RXQ_UNLOCK(rxq); 3023 3024 if (error) { 3025 device_printf(dev, 3026 "cannot allocate mbufs for Rx queue %d\n", i); 3027 return (error); 3028 } 3029 } 3030 3031 return (0); 3032 } 3033 3034 static int 3035 vtnet_init_tx_queues(struct vtnet_softc *sc) 3036 { 3037 struct vtnet_txq *txq; 3038 int i; 3039 3040 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3041 txq = &sc->vtnet_txqs[i]; 3042 txq->vtntx_watchdog = 0; 3043 #ifdef DEV_NETMAP 3044 netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0); 3045 #endif /* DEV_NETMAP */ 3046 } 3047 3048 return (0); 3049 } 3050 3051 static int 3052 vtnet_init_rxtx_queues(struct vtnet_softc *sc) 3053 { 3054 int error; 3055 3056 error = vtnet_init_rx_queues(sc); 3057 if (error) 3058 return (error); 3059 3060 error = vtnet_init_tx_queues(sc); 3061 if (error) 3062 return (error); 3063 3064 return (0); 3065 } 3066 3067 static void 3068 vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 3069 { 3070 device_t dev; 3071 int npairs; 3072 3073 dev = sc->vtnet_dev; 3074 3075 if ((sc->vtnet_flags & VTNET_FLAG_MULTIQ) == 0) { 3076 sc->vtnet_act_vq_pairs = 1; 3077 return; 3078 } 3079 3080 npairs = sc->vtnet_requested_vq_pairs; 3081 3082 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 3083 device_printf(dev, 3084 "cannot set active queue pairs to %d\n", npairs); 3085 npairs = 1; 3086 } 3087 3088 sc->vtnet_act_vq_pairs = npairs; 3089 } 3090 3091 static int 3092 vtnet_reinit(struct vtnet_softc *sc) 3093 { 3094 struct ifnet *ifp; 3095 int error; 3096 3097 ifp = sc->vtnet_ifp; 3098 3099 /* Use the current MAC address. */ 3100 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 3101 vtnet_set_hwaddr(sc); 3102 3103 vtnet_set_active_vq_pairs(sc); 3104 3105 ifp->if_hwassist = 0; 3106 if (ifp->if_capenable & IFCAP_TXCSUM) 3107 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 3108 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) 3109 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6; 3110 if (ifp->if_capenable & IFCAP_TSO4) 3111 ifp->if_hwassist |= CSUM_IP_TSO; 3112 if (ifp->if_capenable & IFCAP_TSO6) 3113 ifp->if_hwassist |= CSUM_IP6_TSO; 3114 3115 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 3116 vtnet_init_rx_filters(sc); 3117 3118 error = vtnet_init_rxtx_queues(sc); 3119 if (error) 3120 return (error); 3121 3122 vtnet_enable_interrupts(sc); 3123 ifp->if_drv_flags |= IFF_DRV_RUNNING; 3124 3125 return (0); 3126 } 3127 3128 static void 3129 vtnet_init_locked(struct vtnet_softc *sc, int init_mode) 3130 { 3131 device_t dev; 3132 struct ifnet *ifp; 3133 3134 dev = sc->vtnet_dev; 3135 ifp = sc->vtnet_ifp; 3136 3137 VTNET_CORE_LOCK_ASSERT(sc); 3138 3139 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 3140 return; 3141 3142 vtnet_stop(sc); 3143 3144 #ifdef DEV_NETMAP 3145 /* Once stopped we can update the netmap flags, if necessary. */ 3146 switch (init_mode) { 3147 case VTNET_INIT_NETMAP_ENTER: 3148 nm_set_native_flags(NA(ifp)); 3149 break; 3150 case VTNET_INIT_NETMAP_EXIT: 3151 nm_clear_native_flags(NA(ifp)); 3152 break; 3153 } 3154 #endif /* DEV_NETMAP */ 3155 3156 /* Reinitialize with the host. */ 3157 if (vtnet_virtio_reinit(sc) != 0) 3158 goto fail; 3159 3160 if (vtnet_reinit(sc) != 0) 3161 goto fail; 3162 3163 virtio_reinit_complete(dev); 3164 3165 vtnet_update_link_status(sc); 3166 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3167 3168 #ifdef DEV_NETMAP 3169 /* Re-enable txsync/rxsync. */ 3170 netmap_enable_all_rings(ifp); 3171 #endif /* DEV_NETMAP */ 3172 3173 return; 3174 3175 fail: 3176 vtnet_stop(sc); 3177 } 3178 3179 static void 3180 vtnet_init(void *xsc) 3181 { 3182 struct vtnet_softc *sc; 3183 3184 sc = xsc; 3185 3186 VTNET_CORE_LOCK(sc); 3187 vtnet_init_locked(sc, 0); 3188 VTNET_CORE_UNLOCK(sc); 3189 } 3190 3191 static void 3192 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3193 { 3194 struct virtqueue *vq; 3195 3196 vq = sc->vtnet_ctrl_vq; 3197 3198 /* 3199 * The control virtqueue is only polled and therefore it should 3200 * already be empty. 3201 */ 3202 KASSERT(virtqueue_empty(vq), 3203 ("%s: ctrl vq %p not empty", __func__, vq)); 3204 } 3205 3206 static void 3207 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3208 struct sglist *sg, int readable, int writable) 3209 { 3210 struct virtqueue *vq; 3211 3212 vq = sc->vtnet_ctrl_vq; 3213 3214 VTNET_CORE_LOCK_ASSERT(sc); 3215 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ, 3216 ("%s: CTRL_VQ feature not negotiated", __func__)); 3217 3218 if (!virtqueue_empty(vq)) 3219 return; 3220 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0) 3221 return; 3222 3223 /* 3224 * Poll for the response, but the command is likely already 3225 * done when we return from the notify. 3226 */ 3227 virtqueue_notify(vq); 3228 virtqueue_poll(vq, NULL); 3229 } 3230 3231 static int 3232 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3233 { 3234 struct virtio_net_ctrl_hdr hdr __aligned(2); 3235 struct sglist_seg segs[3]; 3236 struct sglist sg; 3237 uint8_t ack; 3238 int error; 3239 3240 hdr.class = VIRTIO_NET_CTRL_MAC; 3241 hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3242 ack = VIRTIO_NET_ERR; 3243 3244 sglist_init(&sg, 3, segs); 3245 error = 0; 3246 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3247 error |= sglist_append(&sg, hwaddr, ETHER_ADDR_LEN); 3248 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3249 KASSERT(error == 0 && sg.sg_nseg == 3, 3250 ("%s: error %d adding set MAC msg to sglist", __func__, error)); 3251 3252 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3253 3254 return (ack == VIRTIO_NET_OK ? 0 : EIO); 3255 } 3256 3257 static int 3258 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3259 { 3260 struct sglist_seg segs[3]; 3261 struct sglist sg; 3262 struct { 3263 struct virtio_net_ctrl_hdr hdr; 3264 uint8_t pad1; 3265 struct virtio_net_ctrl_mq mq; 3266 uint8_t pad2; 3267 uint8_t ack; 3268 } s __aligned(2); 3269 int error; 3270 3271 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3272 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3273 s.mq.virtqueue_pairs = npairs; 3274 s.ack = VIRTIO_NET_ERR; 3275 3276 sglist_init(&sg, 3, segs); 3277 error = 0; 3278 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3279 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3280 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3281 KASSERT(error == 0 && sg.sg_nseg == 3, 3282 ("%s: error %d adding MQ message to sglist", __func__, error)); 3283 3284 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3285 3286 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3287 } 3288 3289 static int 3290 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on) 3291 { 3292 struct sglist_seg segs[3]; 3293 struct sglist sg; 3294 struct { 3295 struct virtio_net_ctrl_hdr hdr; 3296 uint8_t pad1; 3297 uint8_t onoff; 3298 uint8_t pad2; 3299 uint8_t ack; 3300 } s __aligned(2); 3301 int error; 3302 3303 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3304 ("%s: CTRL_RX feature not negotiated", __func__)); 3305 3306 s.hdr.class = VIRTIO_NET_CTRL_RX; 3307 s.hdr.cmd = cmd; 3308 s.onoff = !!on; 3309 s.ack = VIRTIO_NET_ERR; 3310 3311 sglist_init(&sg, 3, segs); 3312 error = 0; 3313 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3314 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3315 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3316 KASSERT(error == 0 && sg.sg_nseg == 3, 3317 ("%s: error %d adding Rx message to sglist", __func__, error)); 3318 3319 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3320 3321 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3322 } 3323 3324 static int 3325 vtnet_set_promisc(struct vtnet_softc *sc, int on) 3326 { 3327 3328 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3329 } 3330 3331 static int 3332 vtnet_set_allmulti(struct vtnet_softc *sc, int on) 3333 { 3334 3335 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3336 } 3337 3338 /* 3339 * The device defaults to promiscuous mode for backwards compatibility. 3340 * Turn it off at attach time if possible. 3341 */ 3342 static void 3343 vtnet_attach_disable_promisc(struct vtnet_softc *sc) 3344 { 3345 struct ifnet *ifp; 3346 3347 ifp = sc->vtnet_ifp; 3348 3349 VTNET_CORE_LOCK(sc); 3350 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) { 3351 ifp->if_flags |= IFF_PROMISC; 3352 } else if (vtnet_set_promisc(sc, 0) != 0) { 3353 ifp->if_flags |= IFF_PROMISC; 3354 device_printf(sc->vtnet_dev, 3355 "cannot disable default promiscuous mode\n"); 3356 } 3357 VTNET_CORE_UNLOCK(sc); 3358 } 3359 3360 static void 3361 vtnet_rx_filter(struct vtnet_softc *sc) 3362 { 3363 device_t dev; 3364 struct ifnet *ifp; 3365 3366 dev = sc->vtnet_dev; 3367 ifp = sc->vtnet_ifp; 3368 3369 VTNET_CORE_LOCK_ASSERT(sc); 3370 3371 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) 3372 device_printf(dev, "cannot %s promiscuous mode\n", 3373 ifp->if_flags & IFF_PROMISC ? "enable" : "disable"); 3374 3375 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) 3376 device_printf(dev, "cannot %s all-multicast mode\n", 3377 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable"); 3378 } 3379 3380 static u_int 3381 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt) 3382 { 3383 struct vtnet_softc *sc = arg; 3384 3385 if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3386 return (0); 3387 3388 if (ucnt < VTNET_MAX_MAC_ENTRIES) 3389 bcopy(LLADDR(sdl), 3390 &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt], 3391 ETHER_ADDR_LEN); 3392 3393 return (1); 3394 } 3395 3396 static u_int 3397 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt) 3398 { 3399 struct vtnet_mac_filter *filter = arg; 3400 3401 if (mcnt < VTNET_MAX_MAC_ENTRIES) 3402 bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt], 3403 ETHER_ADDR_LEN); 3404 3405 return (1); 3406 } 3407 3408 static void 3409 vtnet_rx_filter_mac(struct vtnet_softc *sc) 3410 { 3411 struct virtio_net_ctrl_hdr hdr __aligned(2); 3412 struct vtnet_mac_filter *filter; 3413 struct sglist_seg segs[4]; 3414 struct sglist sg; 3415 struct ifnet *ifp; 3416 bool promisc, allmulti; 3417 u_int ucnt, mcnt; 3418 int error; 3419 uint8_t ack; 3420 3421 ifp = sc->vtnet_ifp; 3422 filter = sc->vtnet_mac_filter; 3423 3424 VTNET_CORE_LOCK_ASSERT(sc); 3425 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX, 3426 ("%s: CTRL_RX feature not negotiated", __func__)); 3427 3428 /* Unicast MAC addresses: */ 3429 ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc); 3430 promisc = (ucnt > VTNET_MAX_MAC_ENTRIES); 3431 3432 if (promisc) { 3433 filter->vmf_unicast.nentries = 0; 3434 if_printf(ifp, "more than %d MAC addresses assigned, " 3435 "falling back to promiscuous mode\n", 3436 VTNET_MAX_MAC_ENTRIES); 3437 } else 3438 filter->vmf_unicast.nentries = ucnt; 3439 3440 /* Multicast MAC addresses: */ 3441 mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter); 3442 allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES); 3443 3444 if (allmulti) { 3445 filter->vmf_multicast.nentries = 0; 3446 if_printf(ifp, "more than %d multicast MAC addresses " 3447 "assigned, falling back to all-multicast mode\n", 3448 VTNET_MAX_MAC_ENTRIES); 3449 } else 3450 filter->vmf_multicast.nentries = mcnt; 3451 3452 if (promisc && allmulti) 3453 goto out; 3454 3455 hdr.class = VIRTIO_NET_CTRL_MAC; 3456 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3457 ack = VIRTIO_NET_ERR; 3458 3459 sglist_init(&sg, 4, segs); 3460 error = 0; 3461 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3462 error |= sglist_append(&sg, &filter->vmf_unicast, 3463 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN); 3464 error |= sglist_append(&sg, &filter->vmf_multicast, 3465 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN); 3466 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3467 KASSERT(error == 0 && sg.sg_nseg == 4, 3468 ("%s: error %d adding MAC filter msg to sglist", __func__, error)); 3469 3470 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3471 3472 if (ack != VIRTIO_NET_OK) 3473 if_printf(ifp, "error setting host MAC filter table\n"); 3474 3475 out: 3476 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0) 3477 if_printf(ifp, "cannot enable promiscuous mode\n"); 3478 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0) 3479 if_printf(ifp, "cannot enable all-multicast mode\n"); 3480 } 3481 3482 static int 3483 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3484 { 3485 struct sglist_seg segs[3]; 3486 struct sglist sg; 3487 struct { 3488 struct virtio_net_ctrl_hdr hdr; 3489 uint8_t pad1; 3490 uint16_t tag; 3491 uint8_t pad2; 3492 uint8_t ack; 3493 } s __aligned(2); 3494 int error; 3495 3496 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3497 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3498 s.tag = tag; 3499 s.ack = VIRTIO_NET_ERR; 3500 3501 sglist_init(&sg, 3, segs); 3502 error = 0; 3503 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3504 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3505 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3506 KASSERT(error == 0 && sg.sg_nseg == 3, 3507 ("%s: error %d adding VLAN message to sglist", __func__, error)); 3508 3509 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3510 3511 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3512 } 3513 3514 static void 3515 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3516 { 3517 uint32_t w; 3518 uint16_t tag; 3519 int i, bit; 3520 3521 VTNET_CORE_LOCK_ASSERT(sc); 3522 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER, 3523 ("%s: VLAN_FILTER feature not negotiated", __func__)); 3524 3525 /* Enable the filter for each configured VLAN. */ 3526 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3527 w = sc->vtnet_vlan_filter[i]; 3528 3529 while ((bit = ffs(w) - 1) != -1) { 3530 w &= ~(1 << bit); 3531 tag = sizeof(w) * CHAR_BIT * i + bit; 3532 3533 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3534 device_printf(sc->vtnet_dev, 3535 "cannot enable VLAN %d filter\n", tag); 3536 } 3537 } 3538 } 3539 } 3540 3541 static void 3542 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3543 { 3544 struct ifnet *ifp; 3545 int idx, bit; 3546 3547 ifp = sc->vtnet_ifp; 3548 idx = (tag >> 5) & 0x7F; 3549 bit = tag & 0x1F; 3550 3551 if (tag == 0 || tag > 4095) 3552 return; 3553 3554 VTNET_CORE_LOCK(sc); 3555 3556 if (add) 3557 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3558 else 3559 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3560 3561 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && 3562 ifp->if_drv_flags & IFF_DRV_RUNNING && 3563 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3564 device_printf(sc->vtnet_dev, 3565 "cannot %s VLAN %d %s the host filter table\n", 3566 add ? "add" : "remove", tag, add ? "to" : "from"); 3567 } 3568 3569 VTNET_CORE_UNLOCK(sc); 3570 } 3571 3572 static void 3573 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3574 { 3575 3576 if (ifp->if_softc != arg) 3577 return; 3578 3579 vtnet_update_vlan_filter(arg, 1, tag); 3580 } 3581 3582 static void 3583 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3584 { 3585 3586 if (ifp->if_softc != arg) 3587 return; 3588 3589 vtnet_update_vlan_filter(arg, 0, tag); 3590 } 3591 3592 static int 3593 vtnet_is_link_up(struct vtnet_softc *sc) 3594 { 3595 device_t dev; 3596 struct ifnet *ifp; 3597 uint16_t status; 3598 3599 dev = sc->vtnet_dev; 3600 ifp = sc->vtnet_ifp; 3601 3602 if ((ifp->if_capabilities & IFCAP_LINKSTATE) == 0) 3603 status = VIRTIO_NET_S_LINK_UP; 3604 else 3605 status = virtio_read_dev_config_2(dev, 3606 offsetof(struct virtio_net_config, status)); 3607 3608 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3609 } 3610 3611 static void 3612 vtnet_update_link_status(struct vtnet_softc *sc) 3613 { 3614 struct ifnet *ifp; 3615 int link; 3616 3617 ifp = sc->vtnet_ifp; 3618 3619 VTNET_CORE_LOCK_ASSERT(sc); 3620 link = vtnet_is_link_up(sc); 3621 3622 /* Notify if the link status has changed. */ 3623 if (link != 0 && sc->vtnet_link_active == 0) { 3624 sc->vtnet_link_active = 1; 3625 if_link_state_change(ifp, LINK_STATE_UP); 3626 } else if (link == 0 && sc->vtnet_link_active != 0) { 3627 sc->vtnet_link_active = 0; 3628 if_link_state_change(ifp, LINK_STATE_DOWN); 3629 } 3630 } 3631 3632 static int 3633 vtnet_ifmedia_upd(struct ifnet *ifp) 3634 { 3635 struct vtnet_softc *sc; 3636 struct ifmedia *ifm; 3637 3638 sc = ifp->if_softc; 3639 ifm = &sc->vtnet_media; 3640 3641 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER) 3642 return (EINVAL); 3643 3644 return (0); 3645 } 3646 3647 static void 3648 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 3649 { 3650 struct vtnet_softc *sc; 3651 3652 sc = ifp->if_softc; 3653 3654 ifmr->ifm_status = IFM_AVALID; 3655 ifmr->ifm_active = IFM_ETHER; 3656 3657 VTNET_CORE_LOCK(sc); 3658 if (vtnet_is_link_up(sc) != 0) { 3659 ifmr->ifm_status |= IFM_ACTIVE; 3660 ifmr->ifm_active |= VTNET_MEDIATYPE; 3661 } else 3662 ifmr->ifm_active |= IFM_NONE; 3663 VTNET_CORE_UNLOCK(sc); 3664 } 3665 3666 static void 3667 vtnet_set_hwaddr(struct vtnet_softc *sc) 3668 { 3669 device_t dev; 3670 int i; 3671 3672 dev = sc->vtnet_dev; 3673 3674 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 3675 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0) 3676 device_printf(dev, "unable to set MAC address\n"); 3677 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) { 3678 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3679 virtio_write_dev_config_1(dev, 3680 offsetof(struct virtio_net_config, mac) + i, 3681 sc->vtnet_hwaddr[i]); 3682 } 3683 } 3684 } 3685 3686 static void 3687 vtnet_get_hwaddr(struct vtnet_softc *sc) 3688 { 3689 device_t dev; 3690 int i; 3691 3692 dev = sc->vtnet_dev; 3693 3694 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) { 3695 /* 3696 * Generate a random locally administered unicast address. 3697 * 3698 * It would be nice to generate the same MAC address across 3699 * reboots, but it seems all the hosts currently available 3700 * support the MAC feature, so this isn't too important. 3701 */ 3702 sc->vtnet_hwaddr[0] = 0xB2; 3703 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 3704 vtnet_set_hwaddr(sc); 3705 return; 3706 } 3707 3708 for (i = 0; i < ETHER_ADDR_LEN; i++) { 3709 sc->vtnet_hwaddr[i] = virtio_read_dev_config_1(dev, 3710 offsetof(struct virtio_net_config, mac) + i); 3711 } 3712 } 3713 3714 static void 3715 vtnet_vlan_tag_remove(struct mbuf *m) 3716 { 3717 struct ether_vlan_header *evh; 3718 3719 evh = mtod(m, struct ether_vlan_header *); 3720 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 3721 m->m_flags |= M_VLANTAG; 3722 3723 /* Strip the 802.1Q header. */ 3724 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 3725 ETHER_HDR_LEN - ETHER_TYPE_LEN); 3726 m_adj(m, ETHER_VLAN_ENCAP_LEN); 3727 } 3728 3729 static void 3730 vtnet_set_rx_process_limit(struct vtnet_softc *sc) 3731 { 3732 int limit; 3733 3734 limit = vtnet_tunable_int(sc, "rx_process_limit", 3735 vtnet_rx_process_limit); 3736 if (limit < 0) 3737 limit = INT_MAX; 3738 sc->vtnet_rx_process_limit = limit; 3739 } 3740 3741 static void 3742 vtnet_set_tx_intr_threshold(struct vtnet_softc *sc) 3743 { 3744 int size, thresh; 3745 3746 size = virtqueue_size(sc->vtnet_txqs[0].vtntx_vq); 3747 3748 /* 3749 * The Tx interrupt is disabled until the queue free count falls 3750 * below our threshold. Completed frames are drained from the Tx 3751 * virtqueue before transmitting new frames and in the watchdog 3752 * callout, so the frequency of Tx interrupts is greatly reduced, 3753 * at the cost of not freeing mbufs as quickly as they otherwise 3754 * would be. 3755 * 3756 * N.B. We assume all the Tx queues are the same size. 3757 */ 3758 thresh = size / 4; 3759 3760 /* 3761 * Without indirect descriptors, leave enough room for the most 3762 * segments we handle. 3763 */ 3764 if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 && 3765 thresh < sc->vtnet_tx_nsegs) 3766 thresh = sc->vtnet_tx_nsegs; 3767 3768 sc->vtnet_tx_intr_thresh = thresh; 3769 } 3770 3771 static void 3772 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 3773 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 3774 { 3775 struct sysctl_oid *node; 3776 struct sysctl_oid_list *list; 3777 struct vtnet_rxq_stats *stats; 3778 char namebuf[16]; 3779 3780 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 3781 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3782 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue"); 3783 list = SYSCTL_CHILDREN(node); 3784 3785 stats = &rxq->vtnrx_stats; 3786 3787 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, 3788 &stats->vrxs_ipackets, "Receive packets"); 3789 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, 3790 &stats->vrxs_ibytes, "Receive bytes"); 3791 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, 3792 &stats->vrxs_iqdrops, "Receive drops"); 3793 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, 3794 &stats->vrxs_ierrors, "Receive errors"); 3795 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3796 &stats->vrxs_csum, "Receive checksum offloaded"); 3797 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, 3798 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 3799 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3800 &stats->vrxs_rescheduled, 3801 "Receive interrupt handler rescheduled"); 3802 } 3803 3804 static void 3805 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 3806 struct sysctl_oid_list *child, struct vtnet_txq *txq) 3807 { 3808 struct sysctl_oid *node; 3809 struct sysctl_oid_list *list; 3810 struct vtnet_txq_stats *stats; 3811 char namebuf[16]; 3812 3813 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 3814 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 3815 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue"); 3816 list = SYSCTL_CHILDREN(node); 3817 3818 stats = &txq->vtntx_stats; 3819 3820 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, 3821 &stats->vtxs_opackets, "Transmit packets"); 3822 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, 3823 &stats->vtxs_obytes, "Transmit bytes"); 3824 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, 3825 &stats->vtxs_omcasts, "Transmit multicasts"); 3826 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 3827 &stats->vtxs_csum, "Transmit checksum offloaded"); 3828 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, 3829 &stats->vtxs_tso, "Transmit segmentation offloaded"); 3830 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 3831 &stats->vtxs_rescheduled, 3832 "Transmit interrupt handler rescheduled"); 3833 } 3834 3835 static void 3836 vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 3837 { 3838 device_t dev; 3839 struct sysctl_ctx_list *ctx; 3840 struct sysctl_oid *tree; 3841 struct sysctl_oid_list *child; 3842 int i; 3843 3844 dev = sc->vtnet_dev; 3845 ctx = device_get_sysctl_ctx(dev); 3846 tree = device_get_sysctl_tree(dev); 3847 child = SYSCTL_CHILDREN(tree); 3848 3849 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3850 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 3851 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 3852 } 3853 } 3854 3855 static void 3856 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 3857 struct sysctl_oid_list *child, struct vtnet_softc *sc) 3858 { 3859 struct vtnet_statistics *stats; 3860 struct vtnet_rxq_stats rxaccum; 3861 struct vtnet_txq_stats txaccum; 3862 3863 vtnet_accum_stats(sc, &rxaccum, &txaccum); 3864 3865 stats = &sc->vtnet_stats; 3866 stats->rx_csum_offloaded = rxaccum.vrxs_csum; 3867 stats->rx_csum_failed = rxaccum.vrxs_csum_failed; 3868 stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 3869 stats->tx_csum_offloaded = txaccum.vtxs_csum; 3870 stats->tx_tso_offloaded = txaccum.vtxs_tso; 3871 stats->tx_task_rescheduled = txaccum.vtxs_rescheduled; 3872 3873 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 3874 CTLFLAG_RD, &stats->mbuf_alloc_failed, 3875 "Mbuf cluster allocation failures"); 3876 3877 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 3878 CTLFLAG_RD, &stats->rx_frame_too_large, 3879 "Received frame larger than the mbuf chain"); 3880 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 3881 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 3882 "Enqueuing the replacement receive mbuf failed"); 3883 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 3884 CTLFLAG_RD, &stats->rx_mergeable_failed, 3885 "Mergeable buffers receive failures"); 3886 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 3887 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 3888 "Received checksum offloaded buffer with unsupported " 3889 "Ethernet type"); 3890 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 3891 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 3892 "Received checksum offloaded buffer with incorrect IP protocol"); 3893 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 3894 CTLFLAG_RD, &stats->rx_csum_bad_offset, 3895 "Received checksum offloaded buffer with incorrect offset"); 3896 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", 3897 CTLFLAG_RD, &stats->rx_csum_bad_proto, 3898 "Received checksum offloaded buffer with incorrect protocol"); 3899 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 3900 CTLFLAG_RD, &stats->rx_csum_failed, 3901 "Received buffer checksum offload failed"); 3902 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 3903 CTLFLAG_RD, &stats->rx_csum_offloaded, 3904 "Received buffer checksum offload succeeded"); 3905 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 3906 CTLFLAG_RD, &stats->rx_task_rescheduled, 3907 "Times the receive interrupt task rescheduled itself"); 3908 3909 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype", 3910 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 3911 "Aborted transmit of checksum offloaded buffer with unknown " 3912 "Ethernet type"); 3913 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype", 3914 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 3915 "Aborted transmit of TSO buffer with unknown Ethernet type"); 3916 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 3917 CTLFLAG_RD, &stats->tx_tso_not_tcp, 3918 "Aborted transmit of TSO buffer with non TCP protocol"); 3919 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 3920 CTLFLAG_RD, &stats->tx_defragged, 3921 "Transmit mbufs defragged"); 3922 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 3923 CTLFLAG_RD, &stats->tx_defrag_failed, 3924 "Aborted transmit of buffer because defrag failed"); 3925 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 3926 CTLFLAG_RD, &stats->tx_csum_offloaded, 3927 "Offloaded checksum of transmitted buffer"); 3928 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 3929 CTLFLAG_RD, &stats->tx_tso_offloaded, 3930 "Segmentation offload of transmitted buffer"); 3931 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 3932 CTLFLAG_RD, &stats->tx_task_rescheduled, 3933 "Times the transmit interrupt task rescheduled itself"); 3934 } 3935 3936 static void 3937 vtnet_setup_sysctl(struct vtnet_softc *sc) 3938 { 3939 device_t dev; 3940 struct sysctl_ctx_list *ctx; 3941 struct sysctl_oid *tree; 3942 struct sysctl_oid_list *child; 3943 3944 dev = sc->vtnet_dev; 3945 ctx = device_get_sysctl_ctx(dev); 3946 tree = device_get_sysctl_tree(dev); 3947 child = SYSCTL_CHILDREN(tree); 3948 3949 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 3950 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 3951 "Maximum number of supported virtqueue pairs"); 3952 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "requested_vq_pairs", 3953 CTLFLAG_RD, &sc->vtnet_requested_vq_pairs, 0, 3954 "Requested number of virtqueue pairs"); 3955 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 3956 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 3957 "Number of active virtqueue pairs"); 3958 3959 vtnet_setup_stat_sysctl(ctx, child, sc); 3960 } 3961 3962 static int 3963 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 3964 { 3965 3966 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 3967 } 3968 3969 static void 3970 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 3971 { 3972 3973 virtqueue_disable_intr(rxq->vtnrx_vq); 3974 } 3975 3976 static int 3977 vtnet_txq_enable_intr(struct vtnet_txq *txq) 3978 { 3979 struct virtqueue *vq; 3980 3981 vq = txq->vtntx_vq; 3982 3983 if (vtnet_txq_below_threshold(txq) != 0) 3984 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG)); 3985 3986 /* 3987 * The free count is above our threshold. Keep the Tx interrupt 3988 * disabled until the queue is fuller. 3989 */ 3990 return (0); 3991 } 3992 3993 static void 3994 vtnet_txq_disable_intr(struct vtnet_txq *txq) 3995 { 3996 3997 virtqueue_disable_intr(txq->vtntx_vq); 3998 } 3999 4000 static void 4001 vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 4002 { 4003 int i; 4004 4005 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4006 vtnet_rxq_enable_intr(&sc->vtnet_rxqs[i]); 4007 } 4008 4009 static void 4010 vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 4011 { 4012 int i; 4013 4014 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4015 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 4016 } 4017 4018 static void 4019 vtnet_enable_interrupts(struct vtnet_softc *sc) 4020 { 4021 4022 vtnet_enable_rx_interrupts(sc); 4023 vtnet_enable_tx_interrupts(sc); 4024 } 4025 4026 static void 4027 vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 4028 { 4029 int i; 4030 4031 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4032 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 4033 } 4034 4035 static void 4036 vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 4037 { 4038 int i; 4039 4040 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4041 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 4042 } 4043 4044 static void 4045 vtnet_disable_interrupts(struct vtnet_softc *sc) 4046 { 4047 4048 vtnet_disable_rx_interrupts(sc); 4049 vtnet_disable_tx_interrupts(sc); 4050 } 4051 4052 static int 4053 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 4054 { 4055 char path[64]; 4056 4057 snprintf(path, sizeof(path), 4058 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 4059 TUNABLE_INT_FETCH(path, &def); 4060 4061 return (def); 4062 } 4063 4064 #ifdef DEBUGNET 4065 static void 4066 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize) 4067 { 4068 struct vtnet_softc *sc; 4069 4070 sc = if_getsoftc(ifp); 4071 4072 VTNET_CORE_LOCK(sc); 4073 *nrxr = sc->vtnet_max_vq_pairs; 4074 *ncl = DEBUGNET_MAX_IN_FLIGHT; 4075 *clsize = sc->vtnet_rx_clsize; 4076 VTNET_CORE_UNLOCK(sc); 4077 } 4078 4079 static void 4080 vtnet_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused) 4081 { 4082 } 4083 4084 static int 4085 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m) 4086 { 4087 struct vtnet_softc *sc; 4088 struct vtnet_txq *txq; 4089 int error; 4090 4091 sc = if_getsoftc(ifp); 4092 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4093 IFF_DRV_RUNNING) 4094 return (EBUSY); 4095 4096 txq = &sc->vtnet_txqs[0]; 4097 error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE); 4098 if (error == 0) 4099 (void)vtnet_txq_notify(txq); 4100 return (error); 4101 } 4102 4103 static int 4104 vtnet_debugnet_poll(struct ifnet *ifp, int count) 4105 { 4106 struct vtnet_softc *sc; 4107 int i; 4108 4109 sc = if_getsoftc(ifp); 4110 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4111 IFF_DRV_RUNNING) 4112 return (EBUSY); 4113 4114 (void)vtnet_txq_eof(&sc->vtnet_txqs[0]); 4115 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4116 (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]); 4117 return (0); 4118 } 4119 #endif /* DEBUGNET */ 4120