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