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