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