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