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