1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice unmodified, this list of conditions, and the following 12 * disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, 21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT 22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF 26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 /* Driver for VirtIO network devices. */ 30 31 #include <sys/cdefs.h> 32 __FBSDID("$FreeBSD$"); 33 34 #include <sys/param.h> 35 #include <sys/eventhandler.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/sockio.h> 39 #include <sys/mbuf.h> 40 #include <sys/malloc.h> 41 #include <sys/module.h> 42 #include <sys/socket.h> 43 #include <sys/sysctl.h> 44 #include <sys/random.h> 45 #include <sys/sglist.h> 46 #include <sys/lock.h> 47 #include <sys/mutex.h> 48 #include <sys/taskqueue.h> 49 #include <sys/smp.h> 50 #include <machine/smp.h> 51 52 #include <vm/uma.h> 53 54 #include <net/debugnet.h> 55 #include <net/ethernet.h> 56 #include <net/pfil.h> 57 #include <net/if.h> 58 #include <net/if_var.h> 59 #include <net/if_arp.h> 60 #include <net/if_dl.h> 61 #include <net/if_types.h> 62 #include <net/if_media.h> 63 #include <net/if_vlan_var.h> 64 65 #include <net/bpf.h> 66 67 #include <netinet/in_systm.h> 68 #include <netinet/in.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip6.h> 71 #include <netinet6/ip6_var.h> 72 #include <netinet/udp.h> 73 #include <netinet/tcp.h> 74 #include <netinet/tcp_lro.h> 75 76 #include <machine/bus.h> 77 #include <machine/resource.h> 78 #include <sys/bus.h> 79 #include <sys/rman.h> 80 81 #include <dev/virtio/virtio.h> 82 #include <dev/virtio/virtqueue.h> 83 #include <dev/virtio/network/virtio_net.h> 84 #include <dev/virtio/network/if_vtnetvar.h> 85 #include "virtio_if.h" 86 87 #include "opt_inet.h" 88 #include "opt_inet6.h" 89 90 #if defined(INET) || defined(INET6) 91 #include <machine/in_cksum.h> 92 #endif 93 94 static int vtnet_modevent(module_t, int, void *); 95 96 static int vtnet_probe(device_t); 97 static int vtnet_attach(device_t); 98 static int vtnet_detach(device_t); 99 static int vtnet_suspend(device_t); 100 static int vtnet_resume(device_t); 101 static int vtnet_shutdown(device_t); 102 static int vtnet_attach_completed(device_t); 103 static int vtnet_config_change(device_t); 104 105 static int vtnet_negotiate_features(struct vtnet_softc *); 106 static int vtnet_setup_features(struct vtnet_softc *); 107 static int vtnet_init_rxq(struct vtnet_softc *, int); 108 static int vtnet_init_txq(struct vtnet_softc *, int); 109 static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); 110 static void vtnet_free_rxtx_queues(struct vtnet_softc *); 111 static int vtnet_alloc_rx_filters(struct vtnet_softc *); 112 static void vtnet_free_rx_filters(struct vtnet_softc *); 113 static int vtnet_alloc_virtqueues(struct vtnet_softc *); 114 static int vtnet_alloc_interface(struct vtnet_softc *); 115 static int vtnet_setup_interface(struct vtnet_softc *); 116 static int vtnet_ioctl_mtu(struct vtnet_softc *, u_int); 117 static int vtnet_ioctl_ifflags(struct vtnet_softc *); 118 static int vtnet_ioctl_multi(struct vtnet_softc *); 119 static int vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *); 120 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t); 121 static uint64_t vtnet_get_counter(struct ifnet *, ift_counter); 122 123 static int vtnet_rxq_populate(struct vtnet_rxq *); 124 static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); 125 static struct mbuf * 126 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); 127 static int vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *, 128 struct mbuf *, int); 129 static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); 130 static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); 131 static int vtnet_rxq_new_buf(struct vtnet_rxq *); 132 static int vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *, 133 uint16_t, int, struct virtio_net_hdr *); 134 static int vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *, 135 uint16_t, int, struct virtio_net_hdr *); 136 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, 137 struct virtio_net_hdr *); 138 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); 139 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); 140 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); 141 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, 142 struct virtio_net_hdr *); 143 static int vtnet_rxq_eof(struct vtnet_rxq *); 144 static void vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries); 145 static void vtnet_rx_vq_intr(void *); 146 static void vtnet_rxq_tq_intr(void *, int); 147 148 static int vtnet_txq_intr_threshold(struct vtnet_txq *); 149 static int vtnet_txq_below_threshold(struct vtnet_txq *); 150 static int vtnet_txq_notify(struct vtnet_txq *); 151 static void vtnet_txq_free_mbufs(struct vtnet_txq *); 152 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, 153 int *, int *, int *); 154 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, 155 int, struct virtio_net_hdr *); 156 static struct mbuf * 157 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, 158 struct virtio_net_hdr *); 159 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, 160 struct vtnet_tx_header *); 161 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int); 162 #ifdef VTNET_LEGACY_TX 163 static void vtnet_start_locked(struct vtnet_txq *, struct ifnet *); 164 static void vtnet_start(struct ifnet *); 165 #else 166 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); 167 static int vtnet_txq_mq_start(struct ifnet *, struct mbuf *); 168 static void vtnet_txq_tq_deferred(void *, int); 169 #endif 170 static void vtnet_txq_start(struct vtnet_txq *); 171 static void vtnet_txq_tq_intr(void *, int); 172 static int vtnet_txq_eof(struct vtnet_txq *); 173 static void vtnet_tx_vq_intr(void *); 174 static void vtnet_tx_start_all(struct vtnet_softc *); 175 176 #ifndef VTNET_LEGACY_TX 177 static void vtnet_qflush(struct ifnet *); 178 #endif 179 180 static int vtnet_watchdog(struct vtnet_txq *); 181 static void vtnet_accum_stats(struct vtnet_softc *, 182 struct vtnet_rxq_stats *, struct vtnet_txq_stats *); 183 static void vtnet_tick(void *); 184 185 static void vtnet_start_taskqueues(struct vtnet_softc *); 186 static void vtnet_free_taskqueues(struct vtnet_softc *); 187 static void vtnet_drain_taskqueues(struct vtnet_softc *); 188 189 static void vtnet_drain_rxtx_queues(struct vtnet_softc *); 190 static void vtnet_stop_rendezvous(struct vtnet_softc *); 191 static void vtnet_stop(struct vtnet_softc *); 192 static int vtnet_virtio_reinit(struct vtnet_softc *); 193 static void vtnet_init_rx_filters(struct vtnet_softc *); 194 static int vtnet_init_rx_queues(struct vtnet_softc *); 195 static int vtnet_init_tx_queues(struct vtnet_softc *); 196 static int vtnet_init_rxtx_queues(struct vtnet_softc *); 197 static void vtnet_set_active_vq_pairs(struct vtnet_softc *); 198 static void vtnet_update_rx_offloads(struct vtnet_softc *); 199 static int vtnet_reinit(struct vtnet_softc *); 200 static void vtnet_init_locked(struct vtnet_softc *, int); 201 static void vtnet_init(void *); 202 203 static void vtnet_free_ctrl_vq(struct vtnet_softc *); 204 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 205 struct sglist *, int, int); 206 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 207 static int vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t); 208 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); 209 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool); 210 static int vtnet_set_promisc(struct vtnet_softc *, bool); 211 static int vtnet_set_allmulti(struct vtnet_softc *, bool); 212 static void vtnet_rx_filter(struct vtnet_softc *); 213 static void vtnet_rx_filter_mac(struct vtnet_softc *); 214 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 215 static void vtnet_rx_filter_vlan(struct vtnet_softc *); 216 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 217 static void vtnet_register_vlan(void *, struct ifnet *, uint16_t); 218 static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t); 219 220 static void vtnet_update_speed_duplex(struct vtnet_softc *); 221 static int vtnet_is_link_up(struct vtnet_softc *); 222 static void vtnet_update_link_status(struct vtnet_softc *); 223 static int vtnet_ifmedia_upd(struct ifnet *); 224 static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *); 225 static void vtnet_get_macaddr(struct vtnet_softc *); 226 static void vtnet_set_macaddr(struct vtnet_softc *); 227 static void vtnet_attached_set_macaddr(struct vtnet_softc *); 228 static void vtnet_vlan_tag_remove(struct mbuf *); 229 static void vtnet_set_rx_process_limit(struct vtnet_softc *); 230 231 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, 232 struct sysctl_oid_list *, struct vtnet_rxq *); 233 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, 234 struct sysctl_oid_list *, struct vtnet_txq *); 235 static void vtnet_setup_queue_sysctl(struct vtnet_softc *); 236 static void vtnet_load_tunables(struct vtnet_softc *); 237 static void vtnet_setup_sysctl(struct vtnet_softc *); 238 239 static int vtnet_rxq_enable_intr(struct vtnet_rxq *); 240 static void vtnet_rxq_disable_intr(struct vtnet_rxq *); 241 static int vtnet_txq_enable_intr(struct vtnet_txq *); 242 static void vtnet_txq_disable_intr(struct vtnet_txq *); 243 static void vtnet_enable_rx_interrupts(struct vtnet_softc *); 244 static void vtnet_enable_tx_interrupts(struct vtnet_softc *); 245 static void vtnet_enable_interrupts(struct vtnet_softc *); 246 static void vtnet_disable_rx_interrupts(struct vtnet_softc *); 247 static void vtnet_disable_tx_interrupts(struct vtnet_softc *); 248 static void vtnet_disable_interrupts(struct vtnet_softc *); 249 250 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); 251 252 DEBUGNET_DEFINE(vtnet); 253 254 #define vtnet_htog16(_sc, _val) virtio_htog16(vtnet_modern(_sc), _val) 255 #define vtnet_htog32(_sc, _val) virtio_htog32(vtnet_modern(_sc), _val) 256 #define vtnet_htog64(_sc, _val) virtio_htog64(vtnet_modern(_sc), _val) 257 #define vtnet_gtoh16(_sc, _val) virtio_gtoh16(vtnet_modern(_sc), _val) 258 #define vtnet_gtoh32(_sc, _val) virtio_gtoh32(vtnet_modern(_sc), _val) 259 #define vtnet_gtoh64(_sc, _val) virtio_gtoh64(vtnet_modern(_sc), _val) 260 261 /* Tunables. */ 262 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 263 "VirtIO Net driver parameters"); 264 265 static int vtnet_csum_disable = 0; 266 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN, 267 &vtnet_csum_disable, 0, "Disables receive and send checksum offload"); 268 269 static int vtnet_fixup_needs_csum = 0; 270 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN, 271 &vtnet_fixup_needs_csum, 0, 272 "Calculate valid checksum for NEEDS_CSUM packets"); 273 274 static int vtnet_tso_disable = 0; 275 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, 276 &vtnet_tso_disable, 0, "Disables TSO"); 277 278 static int vtnet_lro_disable = 0; 279 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, 280 &vtnet_lro_disable, 0, "Disables hardware LRO"); 281 282 static int vtnet_mq_disable = 0; 283 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, 284 &vtnet_mq_disable, 0, "Disables multiqueue support"); 285 286 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS; 287 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN, 288 &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs"); 289 290 static int vtnet_tso_maxlen = IP_MAXPACKET; 291 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN, 292 &vtnet_tso_maxlen, 0, "TSO burst limit"); 293 294 static int vtnet_rx_process_limit = 1024; 295 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, 296 &vtnet_rx_process_limit, 0, 297 "Number of RX segments processed in one pass"); 298 299 static int vtnet_lro_entry_count = 128; 300 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN, 301 &vtnet_lro_entry_count, 0, "Software LRO entry count"); 302 303 /* Enable sorted LRO, and the depth of the mbuf queue. */ 304 static int vtnet_lro_mbufq_depth = 0; 305 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN, 306 &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue"); 307 308 static uma_zone_t vtnet_tx_header_zone; 309 310 static struct virtio_feature_desc vtnet_feature_desc[] = { 311 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 312 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 313 { VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, "CtrlRxOffloads" }, 314 { VIRTIO_NET_F_MAC, "MAC" }, 315 { VIRTIO_NET_F_GSO, "TxGSO" }, 316 { VIRTIO_NET_F_GUEST_TSO4, "RxLROv4" }, 317 { VIRTIO_NET_F_GUEST_TSO6, "RxLROv6" }, 318 { VIRTIO_NET_F_GUEST_ECN, "RxLROECN" }, 319 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 320 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 321 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 322 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 323 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 324 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 325 { VIRTIO_NET_F_STATUS, "Status" }, 326 { VIRTIO_NET_F_CTRL_VQ, "CtrlVq" }, 327 { VIRTIO_NET_F_CTRL_RX, "CtrlRxMode" }, 328 { VIRTIO_NET_F_CTRL_VLAN, "CtrlVLANFilter" }, 329 { VIRTIO_NET_F_CTRL_RX_EXTRA, "CtrlRxModeExtra" }, 330 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 331 { VIRTIO_NET_F_MQ, "Multiqueue" }, 332 { VIRTIO_NET_F_CTRL_MAC_ADDR, "CtrlMacAddr" }, 333 { VIRTIO_NET_F_SPEED_DUPLEX, "SpeedDuplex" }, 334 335 { 0, NULL } 336 }; 337 338 static device_method_t vtnet_methods[] = { 339 /* Device methods. */ 340 DEVMETHOD(device_probe, vtnet_probe), 341 DEVMETHOD(device_attach, vtnet_attach), 342 DEVMETHOD(device_detach, vtnet_detach), 343 DEVMETHOD(device_suspend, vtnet_suspend), 344 DEVMETHOD(device_resume, vtnet_resume), 345 DEVMETHOD(device_shutdown, vtnet_shutdown), 346 347 /* VirtIO methods. */ 348 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), 349 DEVMETHOD(virtio_config_change, vtnet_config_change), 350 351 DEVMETHOD_END 352 }; 353 354 #ifdef DEV_NETMAP 355 #include <dev/netmap/if_vtnet_netmap.h> 356 #endif 357 358 static driver_t vtnet_driver = { 359 .name = "vtnet", 360 .methods = vtnet_methods, 361 .size = sizeof(struct vtnet_softc) 362 }; 363 static devclass_t vtnet_devclass; 364 365 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_devclass, 366 vtnet_modevent, 0); 367 MODULE_VERSION(vtnet, 1); 368 MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 369 #ifdef DEV_NETMAP 370 MODULE_DEPEND(vtnet, netmap, 1, 1, 1); 371 #endif 372 373 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter"); 374 375 static int 376 vtnet_modevent(module_t mod __unused, int type, void *unused __unused) 377 { 378 int error = 0; 379 static int loaded = 0; 380 381 switch (type) { 382 case MOD_LOAD: 383 if (loaded++ == 0) { 384 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", 385 sizeof(struct vtnet_tx_header), 386 NULL, NULL, NULL, NULL, 0, 0); 387 #ifdef DEBUGNET 388 /* 389 * We need to allocate from this zone in the transmit path, so ensure 390 * that we have at least one item per header available. 391 * XXX add a separate zone like we do for mbufs? otherwise we may alloc 392 * buckets 393 */ 394 uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 395 uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 396 #endif 397 } 398 break; 399 case MOD_QUIESCE: 400 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) 401 error = EBUSY; 402 break; 403 case MOD_UNLOAD: 404 if (--loaded == 0) { 405 uma_zdestroy(vtnet_tx_header_zone); 406 vtnet_tx_header_zone = NULL; 407 } 408 break; 409 case MOD_SHUTDOWN: 410 break; 411 default: 412 error = EOPNOTSUPP; 413 break; 414 } 415 416 return (error); 417 } 418 419 static int 420 vtnet_probe(device_t dev) 421 { 422 return (VIRTIO_SIMPLE_PROBE(dev, vtnet)); 423 } 424 425 static int 426 vtnet_attach(device_t dev) 427 { 428 struct vtnet_softc *sc; 429 int error; 430 431 sc = device_get_softc(dev); 432 sc->vtnet_dev = dev; 433 virtio_set_feature_desc(dev, vtnet_feature_desc); 434 435 VTNET_CORE_LOCK_INIT(sc); 436 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); 437 vtnet_load_tunables(sc); 438 439 error = vtnet_alloc_interface(sc); 440 if (error) { 441 device_printf(dev, "cannot allocate interface\n"); 442 goto fail; 443 } 444 445 vtnet_setup_sysctl(sc); 446 447 error = vtnet_setup_features(sc); 448 if (error) { 449 device_printf(dev, "cannot setup features\n"); 450 goto fail; 451 } 452 453 error = vtnet_alloc_rx_filters(sc); 454 if (error) { 455 device_printf(dev, "cannot allocate Rx filters\n"); 456 goto fail; 457 } 458 459 error = vtnet_alloc_rxtx_queues(sc); 460 if (error) { 461 device_printf(dev, "cannot allocate queues\n"); 462 goto fail; 463 } 464 465 error = vtnet_alloc_virtqueues(sc); 466 if (error) { 467 device_printf(dev, "cannot allocate virtqueues\n"); 468 goto fail; 469 } 470 471 error = vtnet_setup_interface(sc); 472 if (error) { 473 device_printf(dev, "cannot setup interface\n"); 474 goto fail; 475 } 476 477 error = virtio_setup_intr(dev, INTR_TYPE_NET); 478 if (error) { 479 device_printf(dev, "cannot setup interrupts\n"); 480 ether_ifdetach(sc->vtnet_ifp); 481 goto fail; 482 } 483 484 #ifdef DEV_NETMAP 485 vtnet_netmap_attach(sc); 486 #endif 487 vtnet_start_taskqueues(sc); 488 489 fail: 490 if (error) 491 vtnet_detach(dev); 492 493 return (error); 494 } 495 496 static int 497 vtnet_detach(device_t dev) 498 { 499 struct vtnet_softc *sc; 500 struct ifnet *ifp; 501 502 sc = device_get_softc(dev); 503 ifp = sc->vtnet_ifp; 504 505 if (device_is_attached(dev)) { 506 VTNET_CORE_LOCK(sc); 507 vtnet_stop(sc); 508 VTNET_CORE_UNLOCK(sc); 509 510 callout_drain(&sc->vtnet_tick_ch); 511 vtnet_drain_taskqueues(sc); 512 513 ether_ifdetach(ifp); 514 } 515 516 #ifdef DEV_NETMAP 517 netmap_detach(ifp); 518 #endif 519 520 vtnet_free_taskqueues(sc); 521 522 if (sc->vtnet_vlan_attach != NULL) { 523 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 524 sc->vtnet_vlan_attach = NULL; 525 } 526 if (sc->vtnet_vlan_detach != NULL) { 527 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach); 528 sc->vtnet_vlan_detach = NULL; 529 } 530 531 ifmedia_removeall(&sc->vtnet_media); 532 533 if (ifp != NULL) { 534 if_free(ifp); 535 sc->vtnet_ifp = NULL; 536 } 537 538 vtnet_free_rxtx_queues(sc); 539 vtnet_free_rx_filters(sc); 540 541 if (sc->vtnet_ctrl_vq != NULL) 542 vtnet_free_ctrl_vq(sc); 543 544 VTNET_CORE_LOCK_DESTROY(sc); 545 546 return (0); 547 } 548 549 static int 550 vtnet_suspend(device_t dev) 551 { 552 struct vtnet_softc *sc; 553 554 sc = device_get_softc(dev); 555 556 VTNET_CORE_LOCK(sc); 557 vtnet_stop(sc); 558 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 559 VTNET_CORE_UNLOCK(sc); 560 561 return (0); 562 } 563 564 static int 565 vtnet_resume(device_t dev) 566 { 567 struct vtnet_softc *sc; 568 struct ifnet *ifp; 569 570 sc = device_get_softc(dev); 571 ifp = sc->vtnet_ifp; 572 573 VTNET_CORE_LOCK(sc); 574 if (ifp->if_flags & IFF_UP) 575 vtnet_init_locked(sc, 0); 576 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 577 VTNET_CORE_UNLOCK(sc); 578 579 return (0); 580 } 581 582 static int 583 vtnet_shutdown(device_t dev) 584 { 585 /* 586 * Suspend already does all of what we need to 587 * do here; we just never expect to be resumed. 588 */ 589 return (vtnet_suspend(dev)); 590 } 591 592 static int 593 vtnet_attach_completed(device_t dev) 594 { 595 struct vtnet_softc *sc; 596 597 sc = device_get_softc(dev); 598 599 VTNET_CORE_LOCK(sc); 600 vtnet_attached_set_macaddr(sc); 601 VTNET_CORE_UNLOCK(sc); 602 603 return (0); 604 } 605 606 static int 607 vtnet_config_change(device_t dev) 608 { 609 struct vtnet_softc *sc; 610 611 sc = device_get_softc(dev); 612 613 VTNET_CORE_LOCK(sc); 614 vtnet_update_link_status(sc); 615 if (sc->vtnet_link_active != 0) 616 vtnet_tx_start_all(sc); 617 VTNET_CORE_UNLOCK(sc); 618 619 return (0); 620 } 621 622 static int 623 vtnet_negotiate_features(struct vtnet_softc *sc) 624 { 625 device_t dev; 626 uint64_t features, negotiated_features; 627 int no_csum; 628 629 dev = sc->vtnet_dev; 630 features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES : 631 VTNET_LEGACY_FEATURES; 632 633 /* 634 * TSO and LRO are only available when their corresponding checksum 635 * offload feature is also negotiated. 636 */ 637 no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable); 638 if (no_csum) 639 features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM); 640 if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) 641 features &= ~VTNET_TSO_FEATURES; 642 if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) 643 features &= ~VTNET_LRO_FEATURES; 644 645 #ifndef VTNET_LEGACY_TX 646 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 647 features &= ~VIRTIO_NET_F_MQ; 648 #else 649 features &= ~VIRTIO_NET_F_MQ; 650 #endif 651 652 negotiated_features = virtio_negotiate_features(dev, features); 653 654 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) { 655 uint16_t mtu; 656 657 mtu = virtio_read_dev_config_2(dev, 658 offsetof(struct virtio_net_config, mtu)); 659 if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) { 660 device_printf(dev, "Invalid MTU value: %d. " 661 "MTU feature disabled.\n", mtu); 662 features &= ~VIRTIO_NET_F_MTU; 663 negotiated_features = 664 virtio_negotiate_features(dev, features); 665 } 666 } 667 668 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) { 669 uint16_t npairs; 670 671 npairs = virtio_read_dev_config_2(dev, 672 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 673 if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 674 npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) { 675 device_printf(dev, "Invalid max_virtqueue_pairs value: " 676 "%d. Multiqueue feature disabled.\n", npairs); 677 features &= ~VIRTIO_NET_F_MQ; 678 negotiated_features = 679 virtio_negotiate_features(dev, features); 680 } 681 } 682 683 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 684 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 685 /* 686 * LRO without mergeable buffers requires special care. This 687 * is not ideal because every receive buffer must be large 688 * enough to hold the maximum TCP packet, the Ethernet header, 689 * and the header. This requires up to 34 descriptors with 690 * MCLBYTES clusters. If we do not have indirect descriptors, 691 * LRO is disabled since the virtqueue will not contain very 692 * many receive buffers. 693 */ 694 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 695 device_printf(dev, 696 "Host LRO disabled since both mergeable buffers " 697 "and indirect descriptors were not negotiated\n"); 698 features &= ~VTNET_LRO_FEATURES; 699 negotiated_features = 700 virtio_negotiate_features(dev, features); 701 } else 702 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 703 } 704 705 sc->vtnet_features = negotiated_features; 706 sc->vtnet_negotiated_features = negotiated_features; 707 708 return (virtio_finalize_features(dev)); 709 } 710 711 static int 712 vtnet_setup_features(struct vtnet_softc *sc) 713 { 714 device_t dev; 715 int error; 716 717 dev = sc->vtnet_dev; 718 719 error = vtnet_negotiate_features(sc); 720 if (error) 721 return (error); 722 723 if (virtio_with_feature(dev, VIRTIO_F_VERSION_1)) 724 sc->vtnet_flags |= VTNET_FLAG_MODERN; 725 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 726 sc->vtnet_flags |= VTNET_FLAG_INDIRECT; 727 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 728 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 729 730 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 731 /* This feature should always be negotiated. */ 732 sc->vtnet_flags |= VTNET_FLAG_MAC; 733 } 734 735 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) { 736 sc->vtnet_max_mtu = virtio_read_dev_config_2(dev, 737 offsetof(struct virtio_net_config, mtu)); 738 } else 739 sc->vtnet_max_mtu = VTNET_MAX_MTU; 740 741 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 742 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 743 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 744 } else if (vtnet_modern(sc)) { 745 /* This is identical to the mergeable header. */ 746 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1); 747 } else 748 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 749 750 if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 751 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE; 752 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 753 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG; 754 else 755 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE; 756 757 /* 758 * Favor "hardware" LRO if negotiated, but support software LRO as 759 * a fallback; there is usually little benefit (or worse) with both. 760 */ 761 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 && 762 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0) 763 sc->vtnet_flags |= VTNET_FLAG_SW_LRO; 764 765 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 766 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 767 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 768 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX; 769 else 770 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN; 771 772 sc->vtnet_req_vq_pairs = 1; 773 sc->vtnet_max_vq_pairs = 1; 774 775 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 776 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 777 778 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 779 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 780 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 781 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 782 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 783 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 784 785 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) { 786 sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev, 787 offsetof(struct virtio_net_config, 788 max_virtqueue_pairs)); 789 } 790 } 791 792 if (sc->vtnet_max_vq_pairs > 1) { 793 int req; 794 795 /* 796 * Limit the maximum number of requested queue pairs to the 797 * number of CPUs and the configured maximum. 798 */ 799 req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 800 if (req < 0) 801 req = 1; 802 if (req == 0) 803 req = mp_ncpus; 804 if (req > sc->vtnet_max_vq_pairs) 805 req = sc->vtnet_max_vq_pairs; 806 if (req > mp_ncpus) 807 req = mp_ncpus; 808 if (req > 1) { 809 sc->vtnet_req_vq_pairs = req; 810 sc->vtnet_flags |= VTNET_FLAG_MQ; 811 } 812 } 813 814 return (0); 815 } 816 817 static int 818 vtnet_init_rxq(struct vtnet_softc *sc, int id) 819 { 820 struct vtnet_rxq *rxq; 821 822 rxq = &sc->vtnet_rxqs[id]; 823 824 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 825 device_get_nameunit(sc->vtnet_dev), id); 826 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 827 828 rxq->vtnrx_sc = sc; 829 rxq->vtnrx_id = id; 830 831 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 832 if (rxq->vtnrx_sg == NULL) 833 return (ENOMEM); 834 835 #if defined(INET) || defined(INET6) 836 if (vtnet_software_lro(sc)) { 837 if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp, 838 sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0) 839 return (ENOMEM); 840 } 841 #endif 842 843 NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 844 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 845 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 846 847 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 848 } 849 850 static int 851 vtnet_init_txq(struct vtnet_softc *sc, int id) 852 { 853 struct vtnet_txq *txq; 854 855 txq = &sc->vtnet_txqs[id]; 856 857 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 858 device_get_nameunit(sc->vtnet_dev), id); 859 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 860 861 txq->vtntx_sc = sc; 862 txq->vtntx_id = id; 863 864 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 865 if (txq->vtntx_sg == NULL) 866 return (ENOMEM); 867 868 #ifndef VTNET_LEGACY_TX 869 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 870 M_NOWAIT, &txq->vtntx_mtx); 871 if (txq->vtntx_br == NULL) 872 return (ENOMEM); 873 874 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 875 #endif 876 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 877 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 878 taskqueue_thread_enqueue, &txq->vtntx_tq); 879 if (txq->vtntx_tq == NULL) 880 return (ENOMEM); 881 882 return (0); 883 } 884 885 static int 886 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 887 { 888 int i, npairs, error; 889 890 npairs = sc->vtnet_max_vq_pairs; 891 892 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 893 M_NOWAIT | M_ZERO); 894 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 895 M_NOWAIT | M_ZERO); 896 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 897 return (ENOMEM); 898 899 for (i = 0; i < npairs; i++) { 900 error = vtnet_init_rxq(sc, i); 901 if (error) 902 return (error); 903 error = vtnet_init_txq(sc, i); 904 if (error) 905 return (error); 906 } 907 908 vtnet_set_rx_process_limit(sc); 909 vtnet_setup_queue_sysctl(sc); 910 911 return (0); 912 } 913 914 static void 915 vtnet_destroy_rxq(struct vtnet_rxq *rxq) 916 { 917 918 rxq->vtnrx_sc = NULL; 919 rxq->vtnrx_id = -1; 920 921 #if defined(INET) || defined(INET6) 922 tcp_lro_free(&rxq->vtnrx_lro); 923 #endif 924 925 if (rxq->vtnrx_sg != NULL) { 926 sglist_free(rxq->vtnrx_sg); 927 rxq->vtnrx_sg = NULL; 928 } 929 930 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 931 mtx_destroy(&rxq->vtnrx_mtx); 932 } 933 934 static void 935 vtnet_destroy_txq(struct vtnet_txq *txq) 936 { 937 938 txq->vtntx_sc = NULL; 939 txq->vtntx_id = -1; 940 941 if (txq->vtntx_sg != NULL) { 942 sglist_free(txq->vtntx_sg); 943 txq->vtntx_sg = NULL; 944 } 945 946 #ifndef VTNET_LEGACY_TX 947 if (txq->vtntx_br != NULL) { 948 buf_ring_free(txq->vtntx_br, M_DEVBUF); 949 txq->vtntx_br = NULL; 950 } 951 #endif 952 953 if (mtx_initialized(&txq->vtntx_mtx) != 0) 954 mtx_destroy(&txq->vtntx_mtx); 955 } 956 957 static void 958 vtnet_free_rxtx_queues(struct vtnet_softc *sc) 959 { 960 int i; 961 962 if (sc->vtnet_rxqs != NULL) { 963 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 964 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 965 free(sc->vtnet_rxqs, M_DEVBUF); 966 sc->vtnet_rxqs = NULL; 967 } 968 969 if (sc->vtnet_txqs != NULL) { 970 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 971 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 972 free(sc->vtnet_txqs, M_DEVBUF); 973 sc->vtnet_txqs = NULL; 974 } 975 } 976 977 static int 978 vtnet_alloc_rx_filters(struct vtnet_softc *sc) 979 { 980 981 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 982 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 983 M_DEVBUF, M_NOWAIT | M_ZERO); 984 if (sc->vtnet_mac_filter == NULL) 985 return (ENOMEM); 986 } 987 988 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 989 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 990 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 991 if (sc->vtnet_vlan_filter == NULL) 992 return (ENOMEM); 993 } 994 995 return (0); 996 } 997 998 static void 999 vtnet_free_rx_filters(struct vtnet_softc *sc) 1000 { 1001 1002 if (sc->vtnet_mac_filter != NULL) { 1003 free(sc->vtnet_mac_filter, M_DEVBUF); 1004 sc->vtnet_mac_filter = NULL; 1005 } 1006 1007 if (sc->vtnet_vlan_filter != NULL) { 1008 free(sc->vtnet_vlan_filter, M_DEVBUF); 1009 sc->vtnet_vlan_filter = NULL; 1010 } 1011 } 1012 1013 static int 1014 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 1015 { 1016 device_t dev; 1017 struct vq_alloc_info *info; 1018 struct vtnet_rxq *rxq; 1019 struct vtnet_txq *txq; 1020 int i, idx, flags, nvqs, error; 1021 1022 dev = sc->vtnet_dev; 1023 flags = 0; 1024 1025 nvqs = sc->vtnet_max_vq_pairs * 2; 1026 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 1027 nvqs++; 1028 1029 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); 1030 if (info == NULL) 1031 return (ENOMEM); 1032 1033 for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) { 1034 rxq = &sc->vtnet_rxqs[i]; 1035 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, 1036 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, 1037 "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id); 1038 1039 txq = &sc->vtnet_txqs[i]; 1040 VQ_ALLOC_INFO_INIT(&info[idx+1], sc->vtnet_tx_nsegs, 1041 vtnet_tx_vq_intr, txq, &txq->vtntx_vq, 1042 "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id); 1043 } 1044 1045 /* These queues will not be used so allocate the minimum resources. */ 1046 for (/**/; i < sc->vtnet_max_vq_pairs; i++, idx += 2) { 1047 rxq = &sc->vtnet_rxqs[i]; 1048 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq, 1049 "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id); 1050 1051 txq = &sc->vtnet_txqs[i]; 1052 VQ_ALLOC_INFO_INIT(&info[idx+1], 0, NULL, txq, &txq->vtntx_vq, 1053 "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id); 1054 } 1055 1056 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 1057 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, 1058 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); 1059 } 1060 1061 /* 1062 * TODO: Enable interrupt binding if this is multiqueue. This will 1063 * only matter when per-virtqueue MSIX is available. 1064 */ 1065 if (sc->vtnet_flags & VTNET_FLAG_MQ) 1066 flags |= 0; 1067 1068 error = virtio_alloc_virtqueues(dev, flags, nvqs, info); 1069 free(info, M_TEMP); 1070 1071 return (error); 1072 } 1073 1074 static int 1075 vtnet_alloc_interface(struct vtnet_softc *sc) 1076 { 1077 device_t dev; 1078 struct ifnet *ifp; 1079 1080 dev = sc->vtnet_dev; 1081 1082 ifp = if_alloc(IFT_ETHER); 1083 if (ifp == NULL) 1084 return (ENOMEM); 1085 1086 sc->vtnet_ifp = ifp; 1087 ifp->if_softc = sc; 1088 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 1089 1090 return (0); 1091 } 1092 1093 static int 1094 vtnet_setup_interface(struct vtnet_softc *sc) 1095 { 1096 device_t dev; 1097 struct pfil_head_args pa; 1098 struct ifnet *ifp; 1099 1100 dev = sc->vtnet_dev; 1101 ifp = sc->vtnet_ifp; 1102 1103 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST | 1104 IFF_KNOWSEPOCH; 1105 ifp->if_baudrate = IF_Gbps(10); 1106 ifp->if_init = vtnet_init; 1107 ifp->if_ioctl = vtnet_ioctl; 1108 ifp->if_get_counter = vtnet_get_counter; 1109 #ifndef VTNET_LEGACY_TX 1110 ifp->if_transmit = vtnet_txq_mq_start; 1111 ifp->if_qflush = vtnet_qflush; 1112 #else 1113 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 1114 ifp->if_start = vtnet_start; 1115 IFQ_SET_MAXLEN(&ifp->if_snd, virtqueue_size(vq) - 1); 1116 ifp->if_snd.ifq_drv_maxlen = virtqueue_size(vq) - 1; 1117 IFQ_SET_READY(&ifp->if_snd); 1118 #endif 1119 1120 vtnet_get_macaddr(sc); 1121 1122 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 1123 ifp->if_capabilities |= IFCAP_LINKSTATE; 1124 1125 ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts); 1126 ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL); 1127 ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO); 1128 1129 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 1130 int gso; 1131 1132 ifp->if_capabilities |= IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6; 1133 1134 gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO); 1135 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 1136 ifp->if_capabilities |= IFCAP_TSO4; 1137 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 1138 ifp->if_capabilities |= IFCAP_TSO6; 1139 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 1140 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 1141 1142 if (ifp->if_capabilities & (IFCAP_TSO4 | IFCAP_TSO6)) { 1143 int tso_maxlen; 1144 1145 ifp->if_capabilities |= IFCAP_VLAN_HWTSO; 1146 1147 tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen", 1148 vtnet_tso_maxlen); 1149 ifp->if_hw_tsomax = tso_maxlen - 1150 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); 1151 ifp->if_hw_tsomaxsegcount = sc->vtnet_tx_nsegs - 1; 1152 ifp->if_hw_tsomaxsegsize = PAGE_SIZE; 1153 } 1154 } 1155 1156 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) { 1157 ifp->if_capabilities |= IFCAP_RXCSUM; 1158 #ifdef notyet 1159 /* BMV: Rx checksums not distinguished between IPv4 and IPv6. */ 1160 ifp->if_capabilities |= IFCAP_RXCSUM_IPV6; 1161 #endif 1162 1163 if (vtnet_tunable_int(sc, "fixup_needs_csum", 1164 vtnet_fixup_needs_csum) != 0) 1165 sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM; 1166 1167 /* Support either "hardware" or software LRO. */ 1168 ifp->if_capabilities |= IFCAP_LRO; 1169 } 1170 1171 if (ifp->if_capabilities & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) { 1172 /* 1173 * VirtIO does not support VLAN tagging, but we can fake 1174 * it by inserting and removing the 802.1Q header during 1175 * transmit and receive. We are then able to do checksum 1176 * offloading of VLAN frames. 1177 */ 1178 ifp->if_capabilities |= 1179 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM; 1180 } 1181 1182 if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO) 1183 ifp->if_capabilities |= IFCAP_JUMBO_MTU; 1184 ifp->if_capabilities |= IFCAP_VLAN_MTU; 1185 1186 /* 1187 * Capabilities after here are not enabled by default. 1188 */ 1189 ifp->if_capenable = ifp->if_capabilities; 1190 1191 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 1192 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER; 1193 1194 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1195 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1196 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1197 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1198 } 1199 1200 ether_ifattach(ifp, sc->vtnet_hwaddr); 1201 1202 /* Tell the upper layer(s) we support long frames. */ 1203 ifp->if_hdrlen = sizeof(struct ether_vlan_header); 1204 1205 DEBUGNET_SET(ifp, vtnet); 1206 1207 pa.pa_version = PFIL_VERSION; 1208 pa.pa_flags = PFIL_IN; 1209 pa.pa_type = PFIL_TYPE_ETHERNET; 1210 pa.pa_headname = ifp->if_xname; 1211 sc->vtnet_pfil = pfil_head_register(&pa); 1212 1213 return (0); 1214 } 1215 1216 static int 1217 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu) 1218 { 1219 int framesz; 1220 1221 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 1222 return (MJUMPAGESIZE); 1223 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 1224 return (MCLBYTES); 1225 1226 /* 1227 * Try to scale the receive mbuf cluster size from the MTU. We 1228 * could also use the VQ size to influence the selected size, 1229 * but that would only matter for very small queues. 1230 */ 1231 if (vtnet_modern(sc)) { 1232 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1)); 1233 framesz = sizeof(struct virtio_net_hdr_v1); 1234 } else 1235 framesz = sizeof(struct vtnet_rx_header); 1236 framesz += sizeof(struct ether_vlan_header) + mtu; 1237 1238 if (framesz <= MCLBYTES) 1239 return (MCLBYTES); 1240 else if (framesz <= MJUMPAGESIZE) 1241 return (MJUMPAGESIZE); 1242 else if (framesz <= MJUM9BYTES) 1243 return (MJUM9BYTES); 1244 1245 /* Sane default; avoid 16KB clusters. */ 1246 return (MCLBYTES); 1247 } 1248 1249 static int 1250 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu) 1251 { 1252 struct ifnet *ifp; 1253 int clustersz; 1254 1255 ifp = sc->vtnet_ifp; 1256 VTNET_CORE_LOCK_ASSERT(sc); 1257 1258 if (ifp->if_mtu == mtu) 1259 return (0); 1260 else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu) 1261 return (EINVAL); 1262 1263 ifp->if_mtu = mtu; 1264 clustersz = vtnet_rx_cluster_size(sc, mtu); 1265 1266 if (clustersz != sc->vtnet_rx_clustersz && 1267 ifp->if_drv_flags & IFF_DRV_RUNNING) { 1268 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1269 vtnet_init_locked(sc, 0); 1270 } 1271 1272 return (0); 1273 } 1274 1275 static int 1276 vtnet_ioctl_ifflags(struct vtnet_softc *sc) 1277 { 1278 struct ifnet *ifp; 1279 int drv_running; 1280 1281 ifp = sc->vtnet_ifp; 1282 drv_running = (ifp->if_drv_flags & IFF_DRV_RUNNING) != 0; 1283 1284 VTNET_CORE_LOCK_ASSERT(sc); 1285 1286 if ((ifp->if_flags & IFF_UP) == 0) { 1287 if (drv_running) 1288 vtnet_stop(sc); 1289 goto out; 1290 } 1291 1292 if (!drv_running) { 1293 vtnet_init_locked(sc, 0); 1294 goto out; 1295 } 1296 1297 if ((ifp->if_flags ^ sc->vtnet_if_flags) & 1298 (IFF_PROMISC | IFF_ALLMULTI)) { 1299 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) == 0) 1300 return (ENOTSUP); 1301 vtnet_rx_filter(sc); 1302 } 1303 1304 out: 1305 sc->vtnet_if_flags = ifp->if_flags; 1306 return (0); 1307 } 1308 1309 static int 1310 vtnet_ioctl_multi(struct vtnet_softc *sc) 1311 { 1312 struct ifnet *ifp; 1313 1314 ifp = sc->vtnet_ifp; 1315 1316 VTNET_CORE_LOCK_ASSERT(sc); 1317 1318 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX && 1319 ifp->if_drv_flags & IFF_DRV_RUNNING) 1320 vtnet_rx_filter_mac(sc); 1321 1322 return (0); 1323 } 1324 1325 static int 1326 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr) 1327 { 1328 struct ifnet *ifp; 1329 int mask, reinit, update; 1330 1331 ifp = sc->vtnet_ifp; 1332 mask = (ifr->ifr_reqcap & ifp->if_capabilities) ^ ifp->if_capenable; 1333 reinit = update = 0; 1334 1335 VTNET_CORE_LOCK_ASSERT(sc); 1336 1337 if (mask & IFCAP_TXCSUM) 1338 ifp->if_capenable ^= IFCAP_TXCSUM; 1339 if (mask & IFCAP_TXCSUM_IPV6) 1340 ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; 1341 if (mask & IFCAP_TSO4) 1342 ifp->if_capenable ^= IFCAP_TSO4; 1343 if (mask & IFCAP_TSO6) 1344 ifp->if_capenable ^= IFCAP_TSO6; 1345 1346 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) { 1347 /* 1348 * These Rx features require the negotiated features to 1349 * be updated. Avoid a full reinit if possible. 1350 */ 1351 if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) 1352 update = 1; 1353 else 1354 reinit = 1; 1355 1356 /* BMV: Avoid needless renegotiation for just software LRO. */ 1357 if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) == 1358 IFCAP_LRO && vtnet_software_lro(sc)) 1359 reinit = update = 0; 1360 1361 if (mask & IFCAP_RXCSUM) 1362 ifp->if_capenable ^= IFCAP_RXCSUM; 1363 if (mask & IFCAP_RXCSUM_IPV6) 1364 ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; 1365 if (mask & IFCAP_LRO) 1366 ifp->if_capenable ^= IFCAP_LRO; 1367 1368 /* 1369 * VirtIO does not distinguish between IPv4 and IPv6 checksums 1370 * so treat them as a pair. Guest TSO (LRO) requires receive 1371 * checksums. 1372 */ 1373 if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 1374 ifp->if_capenable |= IFCAP_RXCSUM; 1375 #ifdef notyet 1376 ifp->if_capenable |= IFCAP_RXCSUM_IPV6; 1377 #endif 1378 } else 1379 ifp->if_capenable &= 1380 ~(IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO); 1381 } 1382 1383 if (mask & IFCAP_VLAN_HWFILTER) { 1384 /* These Rx features require renegotiation. */ 1385 reinit = 1; 1386 1387 if (mask & IFCAP_VLAN_HWFILTER) 1388 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; 1389 } 1390 1391 if (mask & IFCAP_VLAN_HWTSO) 1392 ifp->if_capenable ^= IFCAP_VLAN_HWTSO; 1393 if (mask & IFCAP_VLAN_HWTAGGING) 1394 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; 1395 1396 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1397 if (reinit) { 1398 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1399 vtnet_init_locked(sc, 0); 1400 } else if (update) 1401 vtnet_update_rx_offloads(sc); 1402 } 1403 1404 return (0); 1405 } 1406 1407 static int 1408 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1409 { 1410 struct vtnet_softc *sc; 1411 struct ifreq *ifr; 1412 int error; 1413 1414 sc = ifp->if_softc; 1415 ifr = (struct ifreq *) data; 1416 error = 0; 1417 1418 switch (cmd) { 1419 case SIOCSIFMTU: 1420 VTNET_CORE_LOCK(sc); 1421 error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu); 1422 VTNET_CORE_UNLOCK(sc); 1423 break; 1424 1425 case SIOCSIFFLAGS: 1426 VTNET_CORE_LOCK(sc); 1427 error = vtnet_ioctl_ifflags(sc); 1428 VTNET_CORE_UNLOCK(sc); 1429 break; 1430 1431 case SIOCADDMULTI: 1432 case SIOCDELMULTI: 1433 VTNET_CORE_LOCK(sc); 1434 error = vtnet_ioctl_multi(sc); 1435 VTNET_CORE_UNLOCK(sc); 1436 break; 1437 1438 case SIOCSIFMEDIA: 1439 case SIOCGIFMEDIA: 1440 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1441 break; 1442 1443 case SIOCSIFCAP: 1444 VTNET_CORE_LOCK(sc); 1445 error = vtnet_ioctl_ifcap(sc, ifr); 1446 VTNET_CORE_UNLOCK(sc); 1447 VLAN_CAPABILITIES(ifp); 1448 break; 1449 1450 default: 1451 error = ether_ioctl(ifp, cmd, data); 1452 break; 1453 } 1454 1455 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1456 1457 return (error); 1458 } 1459 1460 static int 1461 vtnet_rxq_populate(struct vtnet_rxq *rxq) 1462 { 1463 struct virtqueue *vq; 1464 int nbufs, error; 1465 1466 #ifdef DEV_NETMAP 1467 error = vtnet_netmap_rxq_populate(rxq); 1468 if (error >= 0) 1469 return (error); 1470 #endif /* DEV_NETMAP */ 1471 1472 vq = rxq->vtnrx_vq; 1473 error = ENOSPC; 1474 1475 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1476 error = vtnet_rxq_new_buf(rxq); 1477 if (error) 1478 break; 1479 } 1480 1481 if (nbufs > 0) { 1482 virtqueue_notify(vq); 1483 /* 1484 * EMSGSIZE signifies the virtqueue did not have enough 1485 * entries available to hold the last mbuf. This is not 1486 * an error. 1487 */ 1488 if (error == EMSGSIZE) 1489 error = 0; 1490 } 1491 1492 return (error); 1493 } 1494 1495 static void 1496 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1497 { 1498 struct virtqueue *vq; 1499 struct mbuf *m; 1500 int last; 1501 #ifdef DEV_NETMAP 1502 struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp), 1503 rxq->vtnrx_id, NR_RX); 1504 #else /* !DEV_NETMAP */ 1505 void *kring = NULL; 1506 #endif /* !DEV_NETMAP */ 1507 1508 vq = rxq->vtnrx_vq; 1509 last = 0; 1510 1511 while ((m = virtqueue_drain(vq, &last)) != NULL) { 1512 if (kring == NULL) 1513 m_freem(m); 1514 } 1515 1516 KASSERT(virtqueue_empty(vq), 1517 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1518 } 1519 1520 static struct mbuf * 1521 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1522 { 1523 struct mbuf *m_head, *m_tail, *m; 1524 int i, size; 1525 1526 m_head = NULL; 1527 size = sc->vtnet_rx_clustersz; 1528 1529 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1530 ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs)); 1531 1532 for (i = 0; i < nbufs; i++) { 1533 m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size); 1534 if (m == NULL) { 1535 sc->vtnet_stats.mbuf_alloc_failed++; 1536 m_freem(m_head); 1537 return (NULL); 1538 } 1539 1540 m->m_len = size; 1541 if (m_head != NULL) { 1542 m_tail->m_next = m; 1543 m_tail = m; 1544 } else 1545 m_head = m_tail = m; 1546 } 1547 1548 if (m_tailp != NULL) 1549 *m_tailp = m_tail; 1550 1551 return (m_head); 1552 } 1553 1554 /* 1555 * Slow path for when LRO without mergeable buffers is negotiated. 1556 */ 1557 static int 1558 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1559 int len0) 1560 { 1561 struct vtnet_softc *sc; 1562 struct mbuf *m, *m_prev, *m_new, *m_tail; 1563 int len, clustersz, nreplace, error; 1564 1565 sc = rxq->vtnrx_sc; 1566 clustersz = sc->vtnet_rx_clustersz; 1567 1568 m_prev = NULL; 1569 m_tail = NULL; 1570 nreplace = 0; 1571 1572 m = m0; 1573 len = len0; 1574 1575 /* 1576 * Since these mbuf chains are so large, avoid allocating a complete 1577 * replacement when the received frame did not consume the entire 1578 * chain. Unused mbufs are moved to the tail of the replacement mbuf. 1579 */ 1580 while (len > 0) { 1581 if (m == NULL) { 1582 sc->vtnet_stats.rx_frame_too_large++; 1583 return (EMSGSIZE); 1584 } 1585 1586 /* 1587 * Every mbuf should have the expected cluster size since that 1588 * is also used to allocate the replacements. 1589 */ 1590 KASSERT(m->m_len == clustersz, 1591 ("%s: mbuf size %d not expected cluster size %d", __func__, 1592 m->m_len, clustersz)); 1593 1594 m->m_len = MIN(m->m_len, len); 1595 len -= m->m_len; 1596 1597 m_prev = m; 1598 m = m->m_next; 1599 nreplace++; 1600 } 1601 1602 KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs, 1603 ("%s: invalid replacement mbuf count %d max %d", __func__, 1604 nreplace, sc->vtnet_rx_nmbufs)); 1605 1606 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1607 if (m_new == NULL) { 1608 m_prev->m_len = clustersz; 1609 return (ENOBUFS); 1610 } 1611 1612 /* 1613 * Move any unused mbufs from the received mbuf chain onto the 1614 * end of the replacement chain. 1615 */ 1616 if (m_prev->m_next != NULL) { 1617 m_tail->m_next = m_prev->m_next; 1618 m_prev->m_next = NULL; 1619 } 1620 1621 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1622 if (error) { 1623 /* 1624 * The replacement is suppose to be an copy of the one 1625 * dequeued so this is a very unexpected error. 1626 * 1627 * Restore the m0 chain to the original state if it was 1628 * modified so we can then discard it. 1629 */ 1630 if (m_tail->m_next != NULL) { 1631 m_prev->m_next = m_tail->m_next; 1632 m_tail->m_next = NULL; 1633 } 1634 m_prev->m_len = clustersz; 1635 sc->vtnet_stats.rx_enq_replacement_failed++; 1636 m_freem(m_new); 1637 } 1638 1639 return (error); 1640 } 1641 1642 static int 1643 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1644 { 1645 struct vtnet_softc *sc; 1646 struct mbuf *m_new; 1647 int error; 1648 1649 sc = rxq->vtnrx_sc; 1650 1651 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 1652 return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len)); 1653 1654 MPASS(m->m_next == NULL); 1655 if (m->m_len < len) 1656 return (EMSGSIZE); 1657 1658 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1659 if (m_new == NULL) 1660 return (ENOBUFS); 1661 1662 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1663 if (error) { 1664 sc->vtnet_stats.rx_enq_replacement_failed++; 1665 m_freem(m_new); 1666 } else 1667 m->m_len = len; 1668 1669 return (error); 1670 } 1671 1672 static int 1673 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1674 { 1675 struct vtnet_softc *sc; 1676 struct sglist *sg; 1677 int header_inlined, error; 1678 1679 sc = rxq->vtnrx_sc; 1680 sg = rxq->vtnrx_sg; 1681 1682 KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1683 ("%s: mbuf chain without LRO_NOMRG", __func__)); 1684 VTNET_RXQ_LOCK_ASSERT(rxq); 1685 1686 sglist_reset(sg); 1687 header_inlined = vtnet_modern(sc) || 1688 (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */ 1689 1690 if (header_inlined) 1691 error = sglist_append_mbuf(sg, m); 1692 else { 1693 struct vtnet_rx_header *rxhdr = 1694 mtod(m, struct vtnet_rx_header *); 1695 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1696 1697 /* Append the header and remaining mbuf data. */ 1698 error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1699 if (error) 1700 return (error); 1701 error = sglist_append(sg, &rxhdr[1], 1702 m->m_len - sizeof(struct vtnet_rx_header)); 1703 if (error) 1704 return (error); 1705 1706 if (m->m_next != NULL) 1707 error = sglist_append_mbuf(sg, m->m_next); 1708 } 1709 1710 if (error) 1711 return (error); 1712 1713 return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg)); 1714 } 1715 1716 static int 1717 vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1718 { 1719 struct vtnet_softc *sc; 1720 struct mbuf *m; 1721 int error; 1722 1723 sc = rxq->vtnrx_sc; 1724 1725 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1726 if (m == NULL) 1727 return (ENOBUFS); 1728 1729 error = vtnet_rxq_enqueue_buf(rxq, m); 1730 if (error) 1731 m_freem(m); 1732 1733 return (error); 1734 } 1735 1736 static int 1737 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype, 1738 int hoff, struct virtio_net_hdr *hdr) 1739 { 1740 struct vtnet_softc *sc; 1741 int error; 1742 1743 sc = rxq->vtnrx_sc; 1744 1745 /* 1746 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does 1747 * not have an analogous CSUM flag. The checksum has been validated, 1748 * but is incomplete (TCP/UDP pseudo header). 1749 * 1750 * The packet is likely from another VM on the same host that itself 1751 * performed checksum offloading so Tx/Rx is basically a memcpy and 1752 * the checksum has little value. 1753 * 1754 * Default to receiving the packet as-is for performance reasons, but 1755 * this can cause issues if the packet is to be forwarded because it 1756 * does not contain a valid checksum. This patch may be helpful: 1757 * https://reviews.freebsd.org/D6611. In the meantime, have the driver 1758 * compute the checksum if requested. 1759 * 1760 * BMV: Need to add an CSUM_PARTIAL flag? 1761 */ 1762 if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) { 1763 error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr); 1764 return (error); 1765 } 1766 1767 /* 1768 * Compute the checksum in the driver so the packet will contain a 1769 * valid checksum. The checksum is at csum_offset from csum_start. 1770 */ 1771 switch (etype) { 1772 #if defined(INET) || defined(INET6) 1773 case ETHERTYPE_IP: 1774 case ETHERTYPE_IPV6: { 1775 int csum_off, csum_end; 1776 uint16_t csum; 1777 1778 csum_off = hdr->csum_start + hdr->csum_offset; 1779 csum_end = csum_off + sizeof(uint16_t); 1780 1781 /* Assume checksum will be in the first mbuf. */ 1782 if (m->m_len < csum_end || m->m_pkthdr.len < csum_end) 1783 return (1); 1784 1785 /* 1786 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the 1787 * checksum and write it at the specified offset. We could 1788 * try to verify the packet: csum_start should probably 1789 * correspond to the start of the TCP/UDP header. 1790 * 1791 * BMV: Need to properly handle UDP with zero checksum. Is 1792 * the IPv4 header checksum implicitly validated? 1793 */ 1794 csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start); 1795 *(uint16_t *)(mtodo(m, csum_off)) = csum; 1796 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1797 m->m_pkthdr.csum_data = 0xFFFF; 1798 break; 1799 } 1800 #endif 1801 default: 1802 sc->vtnet_stats.rx_csum_bad_ethtype++; 1803 return (1); 1804 } 1805 1806 return (0); 1807 } 1808 1809 static int 1810 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m, 1811 uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused) 1812 { 1813 struct vtnet_softc *sc; 1814 int protocol; 1815 1816 sc = rxq->vtnrx_sc; 1817 1818 switch (etype) { 1819 #if defined(INET) 1820 case ETHERTYPE_IP: 1821 if (__predict_false(m->m_len < hoff + sizeof(struct ip))) 1822 protocol = IPPROTO_DONE; 1823 else { 1824 struct ip *ip = (struct ip *)(m->m_data + hoff); 1825 protocol = ip->ip_p; 1826 } 1827 break; 1828 #endif 1829 #if defined(INET6) 1830 case ETHERTYPE_IPV6: 1831 if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr)) 1832 || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0) 1833 protocol = IPPROTO_DONE; 1834 break; 1835 #endif 1836 default: 1837 protocol = IPPROTO_DONE; 1838 break; 1839 } 1840 1841 switch (protocol) { 1842 case IPPROTO_TCP: 1843 case IPPROTO_UDP: 1844 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1845 m->m_pkthdr.csum_data = 0xFFFF; 1846 break; 1847 default: 1848 /* 1849 * FreeBSD does not support checksum offloading of this 1850 * protocol. Let the stack re-verify the checksum later 1851 * if the protocol is supported. 1852 */ 1853 #if 0 1854 if_printf(sc->vtnet_ifp, 1855 "%s: checksum offload of unsupported protocol " 1856 "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n", 1857 __func__, etype, protocol, hdr->csum_start, 1858 hdr->csum_offset); 1859 #endif 1860 break; 1861 } 1862 1863 return (0); 1864 } 1865 1866 static int 1867 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1868 struct virtio_net_hdr *hdr) 1869 { 1870 const struct ether_header *eh; 1871 int hoff; 1872 uint16_t etype; 1873 1874 eh = mtod(m, const struct ether_header *); 1875 etype = ntohs(eh->ether_type); 1876 if (etype == ETHERTYPE_VLAN) { 1877 /* TODO BMV: Handle QinQ. */ 1878 const struct ether_vlan_header *evh = 1879 mtod(m, const struct ether_vlan_header *); 1880 etype = ntohs(evh->evl_proto); 1881 hoff = sizeof(struct ether_vlan_header); 1882 } else 1883 hoff = sizeof(struct ether_header); 1884 1885 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1886 return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr)); 1887 else /* VIRTIO_NET_HDR_F_DATA_VALID */ 1888 return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr)); 1889 } 1890 1891 static void 1892 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1893 { 1894 struct mbuf *m; 1895 1896 while (--nbufs > 0) { 1897 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1898 if (m == NULL) 1899 break; 1900 vtnet_rxq_discard_buf(rxq, m); 1901 } 1902 } 1903 1904 static void 1905 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1906 { 1907 int error; 1908 1909 /* 1910 * Requeue the discarded mbuf. This should always be successful 1911 * since it was just dequeued. 1912 */ 1913 error = vtnet_rxq_enqueue_buf(rxq, m); 1914 KASSERT(error == 0, 1915 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1916 } 1917 1918 static int 1919 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1920 { 1921 struct vtnet_softc *sc; 1922 struct virtqueue *vq; 1923 struct mbuf *m_tail; 1924 1925 sc = rxq->vtnrx_sc; 1926 vq = rxq->vtnrx_vq; 1927 m_tail = m_head; 1928 1929 while (--nbufs > 0) { 1930 struct mbuf *m; 1931 uint32_t len; 1932 1933 m = virtqueue_dequeue(vq, &len); 1934 if (m == NULL) { 1935 rxq->vtnrx_stats.vrxs_ierrors++; 1936 goto fail; 1937 } 1938 1939 if (vtnet_rxq_new_buf(rxq) != 0) { 1940 rxq->vtnrx_stats.vrxs_iqdrops++; 1941 vtnet_rxq_discard_buf(rxq, m); 1942 if (nbufs > 1) 1943 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1944 goto fail; 1945 } 1946 1947 if (m->m_len < len) 1948 len = m->m_len; 1949 1950 m->m_len = len; 1951 m->m_flags &= ~M_PKTHDR; 1952 1953 m_head->m_pkthdr.len += len; 1954 m_tail->m_next = m; 1955 m_tail = m; 1956 } 1957 1958 return (0); 1959 1960 fail: 1961 sc->vtnet_stats.rx_mergeable_failed++; 1962 m_freem(m_head); 1963 1964 return (1); 1965 } 1966 1967 #if defined(INET) || defined(INET6) 1968 static int 1969 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m) 1970 { 1971 struct lro_ctrl *lro; 1972 1973 lro = &rxq->vtnrx_lro; 1974 1975 if (lro->lro_mbuf_max != 0) { 1976 tcp_lro_queue_mbuf(lro, m); 1977 return (0); 1978 } 1979 1980 return (tcp_lro_rx(lro, m, 0)); 1981 } 1982 #endif 1983 1984 static void 1985 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 1986 struct virtio_net_hdr *hdr) 1987 { 1988 struct vtnet_softc *sc; 1989 struct ifnet *ifp; 1990 1991 sc = rxq->vtnrx_sc; 1992 ifp = sc->vtnet_ifp; 1993 1994 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) { 1995 struct ether_header *eh = mtod(m, struct ether_header *); 1996 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1997 vtnet_vlan_tag_remove(m); 1998 /* 1999 * With the 802.1Q header removed, update the 2000 * checksum starting location accordingly. 2001 */ 2002 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 2003 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 2004 } 2005 } 2006 2007 m->m_pkthdr.flowid = rxq->vtnrx_id; 2008 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 2009 2010 if (hdr->flags & 2011 (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) { 2012 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 2013 rxq->vtnrx_stats.vrxs_csum++; 2014 else 2015 rxq->vtnrx_stats.vrxs_csum_failed++; 2016 } 2017 2018 if (hdr->gso_size != 0) { 2019 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2020 case VIRTIO_NET_HDR_GSO_TCPV4: 2021 case VIRTIO_NET_HDR_GSO_TCPV6: 2022 m->m_pkthdr.lro_nsegs = 2023 howmany(m->m_pkthdr.len, hdr->gso_size); 2024 rxq->vtnrx_stats.vrxs_host_lro++; 2025 break; 2026 } 2027 } 2028 2029 rxq->vtnrx_stats.vrxs_ipackets++; 2030 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 2031 2032 #if defined(INET) || defined(INET6) 2033 if (vtnet_software_lro(sc) && ifp->if_capenable & IFCAP_LRO) { 2034 if (vtnet_lro_rx(rxq, m) == 0) 2035 return; 2036 } 2037 #endif 2038 2039 (*ifp->if_input)(ifp, m); 2040 } 2041 2042 static int 2043 vtnet_rxq_eof(struct vtnet_rxq *rxq) 2044 { 2045 struct virtio_net_hdr lhdr, *hdr; 2046 struct vtnet_softc *sc; 2047 struct ifnet *ifp; 2048 struct virtqueue *vq; 2049 int deq, count; 2050 2051 sc = rxq->vtnrx_sc; 2052 vq = rxq->vtnrx_vq; 2053 ifp = sc->vtnet_ifp; 2054 deq = 0; 2055 count = sc->vtnet_rx_process_limit; 2056 2057 VTNET_RXQ_LOCK_ASSERT(rxq); 2058 2059 while (count-- > 0) { 2060 struct mbuf *m; 2061 uint32_t len, nbufs, adjsz; 2062 2063 m = virtqueue_dequeue(vq, &len); 2064 if (m == NULL) 2065 break; 2066 deq++; 2067 2068 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 2069 rxq->vtnrx_stats.vrxs_ierrors++; 2070 vtnet_rxq_discard_buf(rxq, m); 2071 continue; 2072 } 2073 2074 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) { 2075 struct virtio_net_hdr_mrg_rxbuf *mhdr = 2076 mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 2077 nbufs = vtnet_htog16(sc, mhdr->num_buffers); 2078 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 2079 } else if (vtnet_modern(sc)) { 2080 nbufs = 1; /* num_buffers is always 1 */ 2081 adjsz = sizeof(struct virtio_net_hdr_v1); 2082 } else { 2083 nbufs = 1; 2084 adjsz = sizeof(struct vtnet_rx_header); 2085 /* 2086 * Account for our gap between the header and start of 2087 * data to keep the segments separated. 2088 */ 2089 len += VTNET_RX_HEADER_PAD; 2090 } 2091 2092 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 2093 rxq->vtnrx_stats.vrxs_iqdrops++; 2094 vtnet_rxq_discard_buf(rxq, m); 2095 if (nbufs > 1) 2096 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 2097 continue; 2098 } 2099 2100 m->m_pkthdr.len = len; 2101 m->m_pkthdr.rcvif = ifp; 2102 m->m_pkthdr.csum_flags = 0; 2103 2104 if (nbufs > 1) { 2105 /* Dequeue the rest of chain. */ 2106 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 2107 continue; 2108 } 2109 2110 /* 2111 * Save an endian swapped version of the header prior to it 2112 * being stripped. The header is always at the start of the 2113 * mbuf data. num_buffers was already saved (and not needed) 2114 * so use the standard header. 2115 */ 2116 hdr = mtod(m, struct virtio_net_hdr *); 2117 lhdr.flags = hdr->flags; 2118 lhdr.gso_type = hdr->gso_type; 2119 lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len); 2120 lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size); 2121 lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start); 2122 lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset); 2123 m_adj(m, adjsz); 2124 2125 if (PFIL_HOOKED_IN(sc->vtnet_pfil)) { 2126 pfil_return_t pfil; 2127 2128 pfil = pfil_run_hooks(sc->vtnet_pfil, &m, ifp, PFIL_IN, 2129 NULL); 2130 switch (pfil) { 2131 case PFIL_REALLOCED: 2132 m = pfil_mem2mbuf(m->m_data); 2133 break; 2134 case PFIL_DROPPED: 2135 case PFIL_CONSUMED: 2136 continue; 2137 default: 2138 KASSERT(pfil == PFIL_PASS, 2139 ("Filter returned %d!", pfil)); 2140 } 2141 } 2142 2143 vtnet_rxq_input(rxq, m, &lhdr); 2144 } 2145 2146 if (deq > 0) { 2147 #if defined(INET) || defined(INET6) 2148 tcp_lro_flush_all(&rxq->vtnrx_lro); 2149 #endif 2150 virtqueue_notify(vq); 2151 } 2152 2153 return (count > 0 ? 0 : EAGAIN); 2154 } 2155 2156 static void 2157 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries) 2158 { 2159 struct vtnet_softc *sc; 2160 struct ifnet *ifp; 2161 u_int more; 2162 #ifdef DEV_NETMAP 2163 int nmirq; 2164 #endif /* DEV_NETMAP */ 2165 2166 sc = rxq->vtnrx_sc; 2167 ifp = sc->vtnet_ifp; 2168 2169 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 2170 /* 2171 * Ignore this interrupt. Either this is a spurious interrupt 2172 * or multiqueue without per-VQ MSIX so every queue needs to 2173 * be polled (a brain dead configuration we could try harder 2174 * to avoid). 2175 */ 2176 vtnet_rxq_disable_intr(rxq); 2177 return; 2178 } 2179 2180 VTNET_RXQ_LOCK(rxq); 2181 2182 #ifdef DEV_NETMAP 2183 /* 2184 * We call netmap_rx_irq() under lock to prevent concurrent calls. 2185 * This is not necessary to serialize the access to the RX vq, but 2186 * rather to avoid races that may happen if this interface is 2187 * attached to a VALE switch, which would cause received packets 2188 * to stall in the RX queue (nm_kr_tryget() could find the kring 2189 * busy when called from netmap_bwrap_intr_notify()). 2190 */ 2191 nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more); 2192 if (nmirq != NM_IRQ_PASS) { 2193 VTNET_RXQ_UNLOCK(rxq); 2194 if (nmirq == NM_IRQ_RESCHED) { 2195 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2196 } 2197 return; 2198 } 2199 #endif /* DEV_NETMAP */ 2200 2201 again: 2202 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2203 VTNET_RXQ_UNLOCK(rxq); 2204 return; 2205 } 2206 2207 more = vtnet_rxq_eof(rxq); 2208 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 2209 if (!more) 2210 vtnet_rxq_disable_intr(rxq); 2211 /* 2212 * This is an occasional condition or race (when !more), 2213 * so retry a few times before scheduling the taskqueue. 2214 */ 2215 if (tries-- > 0) 2216 goto again; 2217 2218 rxq->vtnrx_stats.vrxs_rescheduled++; 2219 VTNET_RXQ_UNLOCK(rxq); 2220 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2221 } else 2222 VTNET_RXQ_UNLOCK(rxq); 2223 } 2224 2225 static void 2226 vtnet_rx_vq_intr(void *xrxq) 2227 { 2228 struct vtnet_rxq *rxq; 2229 2230 rxq = xrxq; 2231 vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES); 2232 } 2233 2234 static void 2235 vtnet_rxq_tq_intr(void *xrxq, int pending __unused) 2236 { 2237 struct vtnet_rxq *rxq; 2238 2239 rxq = xrxq; 2240 vtnet_rx_vq_process(rxq, 0); 2241 } 2242 2243 static int 2244 vtnet_txq_intr_threshold(struct vtnet_txq *txq) 2245 { 2246 struct vtnet_softc *sc; 2247 int threshold; 2248 2249 sc = txq->vtntx_sc; 2250 2251 /* 2252 * The Tx interrupt is disabled until the queue free count falls 2253 * below our threshold. Completed frames are drained from the Tx 2254 * virtqueue before transmitting new frames and in the watchdog 2255 * callout, so the frequency of Tx interrupts is greatly reduced, 2256 * at the cost of not freeing mbufs as quickly as they otherwise 2257 * would be. 2258 */ 2259 threshold = virtqueue_size(txq->vtntx_vq) / 4; 2260 2261 /* 2262 * Without indirect descriptors, leave enough room for the most 2263 * segments we handle. 2264 */ 2265 if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 && 2266 threshold < sc->vtnet_tx_nsegs) 2267 threshold = sc->vtnet_tx_nsegs; 2268 2269 return (threshold); 2270 } 2271 2272 static int 2273 vtnet_txq_below_threshold(struct vtnet_txq *txq) 2274 { 2275 struct virtqueue *vq; 2276 2277 vq = txq->vtntx_vq; 2278 2279 return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold); 2280 } 2281 2282 static int 2283 vtnet_txq_notify(struct vtnet_txq *txq) 2284 { 2285 struct virtqueue *vq; 2286 2287 vq = txq->vtntx_vq; 2288 2289 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2290 virtqueue_notify(vq); 2291 2292 if (vtnet_txq_enable_intr(txq) == 0) 2293 return (0); 2294 2295 /* 2296 * Drain frames that were completed since last checked. If this 2297 * causes the queue to go above the threshold, the caller should 2298 * continue transmitting. 2299 */ 2300 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) { 2301 virtqueue_disable_intr(vq); 2302 return (1); 2303 } 2304 2305 return (0); 2306 } 2307 2308 static void 2309 vtnet_txq_free_mbufs(struct vtnet_txq *txq) 2310 { 2311 struct virtqueue *vq; 2312 struct vtnet_tx_header *txhdr; 2313 int last; 2314 #ifdef DEV_NETMAP 2315 struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp), 2316 txq->vtntx_id, NR_TX); 2317 #else /* !DEV_NETMAP */ 2318 void *kring = NULL; 2319 #endif /* !DEV_NETMAP */ 2320 2321 vq = txq->vtntx_vq; 2322 last = 0; 2323 2324 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 2325 if (kring == NULL) { 2326 m_freem(txhdr->vth_mbuf); 2327 uma_zfree(vtnet_tx_header_zone, txhdr); 2328 } 2329 } 2330 2331 KASSERT(virtqueue_empty(vq), 2332 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 2333 } 2334 2335 /* 2336 * BMV: This can go away once we finally have offsets in the mbuf header. 2337 */ 2338 static int 2339 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype, 2340 int *proto, int *start) 2341 { 2342 struct vtnet_softc *sc; 2343 struct ether_vlan_header *evh; 2344 int offset; 2345 2346 sc = txq->vtntx_sc; 2347 2348 evh = mtod(m, struct ether_vlan_header *); 2349 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 2350 /* BMV: We should handle nested VLAN tags too. */ 2351 *etype = ntohs(evh->evl_proto); 2352 offset = sizeof(struct ether_vlan_header); 2353 } else { 2354 *etype = ntohs(evh->evl_encap_proto); 2355 offset = sizeof(struct ether_header); 2356 } 2357 2358 switch (*etype) { 2359 #if defined(INET) 2360 case ETHERTYPE_IP: { 2361 struct ip *ip, iphdr; 2362 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 2363 m_copydata(m, offset, sizeof(struct ip), 2364 (caddr_t) &iphdr); 2365 ip = &iphdr; 2366 } else 2367 ip = (struct ip *)(m->m_data + offset); 2368 *proto = ip->ip_p; 2369 *start = offset + (ip->ip_hl << 2); 2370 break; 2371 } 2372 #endif 2373 #if defined(INET6) 2374 case ETHERTYPE_IPV6: 2375 *proto = -1; 2376 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 2377 /* Assert the network stack sent us a valid packet. */ 2378 KASSERT(*start > offset, 2379 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 2380 *start, offset, *proto)); 2381 break; 2382 #endif 2383 default: 2384 sc->vtnet_stats.tx_csum_unknown_ethtype++; 2385 return (EINVAL); 2386 } 2387 2388 return (0); 2389 } 2390 2391 static int 2392 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 2393 int offset, struct virtio_net_hdr *hdr) 2394 { 2395 static struct timeval lastecn; 2396 static int curecn; 2397 struct vtnet_softc *sc; 2398 struct tcphdr *tcp, tcphdr; 2399 2400 sc = txq->vtntx_sc; 2401 2402 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 2403 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 2404 tcp = &tcphdr; 2405 } else 2406 tcp = (struct tcphdr *)(m->m_data + offset); 2407 2408 hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2)); 2409 hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz); 2410 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 2411 VIRTIO_NET_HDR_GSO_TCPV6; 2412 2413 if (__predict_false(tcp->th_flags & TH_CWR)) { 2414 /* 2415 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In 2416 * FreeBSD, ECN support is not on a per-interface basis, 2417 * but globally via the net.inet.tcp.ecn.enable sysctl 2418 * knob. The default is off. 2419 */ 2420 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2421 if (ppsratecheck(&lastecn, &curecn, 1)) 2422 if_printf(sc->vtnet_ifp, 2423 "TSO with ECN not negotiated with host\n"); 2424 return (ENOTSUP); 2425 } 2426 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2427 } 2428 2429 txq->vtntx_stats.vtxs_tso++; 2430 2431 return (0); 2432 } 2433 2434 static struct mbuf * 2435 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2436 struct virtio_net_hdr *hdr) 2437 { 2438 struct vtnet_softc *sc; 2439 int flags, etype, csum_start, proto, error; 2440 2441 sc = txq->vtntx_sc; 2442 flags = m->m_pkthdr.csum_flags; 2443 2444 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2445 if (error) 2446 goto drop; 2447 2448 if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) { 2449 /* Sanity check the parsed mbuf matches the offload flags. */ 2450 if (__predict_false((flags & VTNET_CSUM_OFFLOAD && 2451 etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6 2452 && etype != ETHERTYPE_IPV6))) { 2453 sc->vtnet_stats.tx_csum_proto_mismatch++; 2454 goto drop; 2455 } 2456 2457 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2458 hdr->csum_start = vtnet_gtoh16(sc, csum_start); 2459 hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data); 2460 txq->vtntx_stats.vtxs_csum++; 2461 } 2462 2463 if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) { 2464 /* 2465 * Sanity check the parsed mbuf IP protocol is TCP, and 2466 * VirtIO TSO reqires the checksum offloading above. 2467 */ 2468 if (__predict_false(proto != IPPROTO_TCP)) { 2469 sc->vtnet_stats.tx_tso_not_tcp++; 2470 goto drop; 2471 } else if (__predict_false((hdr->flags & 2472 VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) { 2473 sc->vtnet_stats.tx_tso_without_csum++; 2474 goto drop; 2475 } 2476 2477 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2478 if (error) 2479 goto drop; 2480 } 2481 2482 return (m); 2483 2484 drop: 2485 m_freem(m); 2486 return (NULL); 2487 } 2488 2489 static int 2490 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2491 struct vtnet_tx_header *txhdr) 2492 { 2493 struct vtnet_softc *sc; 2494 struct virtqueue *vq; 2495 struct sglist *sg; 2496 struct mbuf *m; 2497 int error; 2498 2499 sc = txq->vtntx_sc; 2500 vq = txq->vtntx_vq; 2501 sg = txq->vtntx_sg; 2502 m = *m_head; 2503 2504 sglist_reset(sg); 2505 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2506 if (error != 0 || sg->sg_nseg != 1) { 2507 KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d", 2508 __func__, error, sg->sg_nseg)); 2509 goto fail; 2510 } 2511 2512 error = sglist_append_mbuf(sg, m); 2513 if (error) { 2514 m = m_defrag(m, M_NOWAIT); 2515 if (m == NULL) 2516 goto fail; 2517 2518 *m_head = m; 2519 sc->vtnet_stats.tx_defragged++; 2520 2521 error = sglist_append_mbuf(sg, m); 2522 if (error) 2523 goto fail; 2524 } 2525 2526 txhdr->vth_mbuf = m; 2527 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2528 2529 return (error); 2530 2531 fail: 2532 sc->vtnet_stats.tx_defrag_failed++; 2533 m_freem(*m_head); 2534 *m_head = NULL; 2535 2536 return (ENOBUFS); 2537 } 2538 2539 static int 2540 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags) 2541 { 2542 struct vtnet_tx_header *txhdr; 2543 struct virtio_net_hdr *hdr; 2544 struct mbuf *m; 2545 int error; 2546 2547 m = *m_head; 2548 M_ASSERTPKTHDR(m); 2549 2550 txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO); 2551 if (txhdr == NULL) { 2552 m_freem(m); 2553 *m_head = NULL; 2554 return (ENOMEM); 2555 } 2556 2557 /* 2558 * Always use the non-mergeable header, regardless if mergable headers 2559 * were negotiated, because for transmit num_buffers is always zero. 2560 * The vtnet_hdr_size is used to enqueue the right header size segment. 2561 */ 2562 hdr = &txhdr->vth_uhdr.hdr; 2563 2564 if (m->m_flags & M_VLANTAG) { 2565 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2566 if ((*m_head = m) == NULL) { 2567 error = ENOBUFS; 2568 goto fail; 2569 } 2570 m->m_flags &= ~M_VLANTAG; 2571 } 2572 2573 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2574 m = vtnet_txq_offload(txq, m, hdr); 2575 if ((*m_head = m) == NULL) { 2576 error = ENOBUFS; 2577 goto fail; 2578 } 2579 } 2580 2581 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2582 fail: 2583 if (error) 2584 uma_zfree(vtnet_tx_header_zone, txhdr); 2585 2586 return (error); 2587 } 2588 2589 #ifdef VTNET_LEGACY_TX 2590 2591 static void 2592 vtnet_start_locked(struct vtnet_txq *txq, struct ifnet *ifp) 2593 { 2594 struct vtnet_softc *sc; 2595 struct virtqueue *vq; 2596 struct mbuf *m0; 2597 int tries, enq; 2598 2599 sc = txq->vtntx_sc; 2600 vq = txq->vtntx_vq; 2601 tries = 0; 2602 2603 VTNET_TXQ_LOCK_ASSERT(txq); 2604 2605 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2606 sc->vtnet_link_active == 0) 2607 return; 2608 2609 vtnet_txq_eof(txq); 2610 2611 again: 2612 enq = 0; 2613 2614 while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) { 2615 if (virtqueue_full(vq)) 2616 break; 2617 2618 IFQ_DRV_DEQUEUE(&ifp->if_snd, m0); 2619 if (m0 == NULL) 2620 break; 2621 2622 if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) { 2623 if (m0 != NULL) 2624 IFQ_DRV_PREPEND(&ifp->if_snd, m0); 2625 break; 2626 } 2627 2628 enq++; 2629 ETHER_BPF_MTAP(ifp, m0); 2630 } 2631 2632 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2633 if (tries++ < VTNET_NOTIFY_RETRIES) 2634 goto again; 2635 2636 txq->vtntx_stats.vtxs_rescheduled++; 2637 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2638 } 2639 } 2640 2641 static void 2642 vtnet_start(struct ifnet *ifp) 2643 { 2644 struct vtnet_softc *sc; 2645 struct vtnet_txq *txq; 2646 2647 sc = ifp->if_softc; 2648 txq = &sc->vtnet_txqs[0]; 2649 2650 VTNET_TXQ_LOCK(txq); 2651 vtnet_start_locked(txq, ifp); 2652 VTNET_TXQ_UNLOCK(txq); 2653 } 2654 2655 #else /* !VTNET_LEGACY_TX */ 2656 2657 static int 2658 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2659 { 2660 struct vtnet_softc *sc; 2661 struct virtqueue *vq; 2662 struct buf_ring *br; 2663 struct ifnet *ifp; 2664 int enq, tries, error; 2665 2666 sc = txq->vtntx_sc; 2667 vq = txq->vtntx_vq; 2668 br = txq->vtntx_br; 2669 ifp = sc->vtnet_ifp; 2670 tries = 0; 2671 error = 0; 2672 2673 VTNET_TXQ_LOCK_ASSERT(txq); 2674 2675 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || 2676 sc->vtnet_link_active == 0) { 2677 if (m != NULL) 2678 error = drbr_enqueue(ifp, br, m); 2679 return (error); 2680 } 2681 2682 if (m != NULL) { 2683 error = drbr_enqueue(ifp, br, m); 2684 if (error) 2685 return (error); 2686 } 2687 2688 vtnet_txq_eof(txq); 2689 2690 again: 2691 enq = 0; 2692 2693 while ((m = drbr_peek(ifp, br)) != NULL) { 2694 if (virtqueue_full(vq)) { 2695 drbr_putback(ifp, br, m); 2696 break; 2697 } 2698 2699 if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) { 2700 if (m != NULL) 2701 drbr_putback(ifp, br, m); 2702 else 2703 drbr_advance(ifp, br); 2704 break; 2705 } 2706 drbr_advance(ifp, br); 2707 2708 enq++; 2709 ETHER_BPF_MTAP(ifp, m); 2710 } 2711 2712 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2713 if (tries++ < VTNET_NOTIFY_RETRIES) 2714 goto again; 2715 2716 txq->vtntx_stats.vtxs_rescheduled++; 2717 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2718 } 2719 2720 return (0); 2721 } 2722 2723 static int 2724 vtnet_txq_mq_start(struct ifnet *ifp, struct mbuf *m) 2725 { 2726 struct vtnet_softc *sc; 2727 struct vtnet_txq *txq; 2728 int i, npairs, error; 2729 2730 sc = ifp->if_softc; 2731 npairs = sc->vtnet_act_vq_pairs; 2732 2733 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2734 i = m->m_pkthdr.flowid % npairs; 2735 else 2736 i = curcpu % npairs; 2737 2738 txq = &sc->vtnet_txqs[i]; 2739 2740 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2741 error = vtnet_txq_mq_start_locked(txq, m); 2742 VTNET_TXQ_UNLOCK(txq); 2743 } else { 2744 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2745 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2746 } 2747 2748 return (error); 2749 } 2750 2751 static void 2752 vtnet_txq_tq_deferred(void *xtxq, int pending __unused) 2753 { 2754 struct vtnet_softc *sc; 2755 struct vtnet_txq *txq; 2756 2757 txq = xtxq; 2758 sc = txq->vtntx_sc; 2759 2760 VTNET_TXQ_LOCK(txq); 2761 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2762 vtnet_txq_mq_start_locked(txq, NULL); 2763 VTNET_TXQ_UNLOCK(txq); 2764 } 2765 2766 #endif /* VTNET_LEGACY_TX */ 2767 2768 static void 2769 vtnet_txq_start(struct vtnet_txq *txq) 2770 { 2771 struct vtnet_softc *sc; 2772 struct ifnet *ifp; 2773 2774 sc = txq->vtntx_sc; 2775 ifp = sc->vtnet_ifp; 2776 2777 #ifdef VTNET_LEGACY_TX 2778 if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 2779 vtnet_start_locked(txq, ifp); 2780 #else 2781 if (!drbr_empty(ifp, txq->vtntx_br)) 2782 vtnet_txq_mq_start_locked(txq, NULL); 2783 #endif 2784 } 2785 2786 static void 2787 vtnet_txq_tq_intr(void *xtxq, int pending __unused) 2788 { 2789 struct vtnet_softc *sc; 2790 struct vtnet_txq *txq; 2791 struct ifnet *ifp; 2792 2793 txq = xtxq; 2794 sc = txq->vtntx_sc; 2795 ifp = sc->vtnet_ifp; 2796 2797 VTNET_TXQ_LOCK(txq); 2798 2799 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2800 VTNET_TXQ_UNLOCK(txq); 2801 return; 2802 } 2803 2804 vtnet_txq_eof(txq); 2805 vtnet_txq_start(txq); 2806 2807 VTNET_TXQ_UNLOCK(txq); 2808 } 2809 2810 static int 2811 vtnet_txq_eof(struct vtnet_txq *txq) 2812 { 2813 struct virtqueue *vq; 2814 struct vtnet_tx_header *txhdr; 2815 struct mbuf *m; 2816 int deq; 2817 2818 vq = txq->vtntx_vq; 2819 deq = 0; 2820 VTNET_TXQ_LOCK_ASSERT(txq); 2821 2822 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2823 m = txhdr->vth_mbuf; 2824 deq++; 2825 2826 txq->vtntx_stats.vtxs_opackets++; 2827 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2828 if (m->m_flags & M_MCAST) 2829 txq->vtntx_stats.vtxs_omcasts++; 2830 2831 m_freem(m); 2832 uma_zfree(vtnet_tx_header_zone, txhdr); 2833 } 2834 2835 if (virtqueue_empty(vq)) 2836 txq->vtntx_watchdog = 0; 2837 2838 return (deq); 2839 } 2840 2841 static void 2842 vtnet_tx_vq_intr(void *xtxq) 2843 { 2844 struct vtnet_softc *sc; 2845 struct vtnet_txq *txq; 2846 struct ifnet *ifp; 2847 2848 txq = xtxq; 2849 sc = txq->vtntx_sc; 2850 ifp = sc->vtnet_ifp; 2851 2852 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2853 /* 2854 * Ignore this interrupt. Either this is a spurious interrupt 2855 * or multiqueue without per-VQ MSIX so every queue needs to 2856 * be polled (a brain dead configuration we could try harder 2857 * to avoid). 2858 */ 2859 vtnet_txq_disable_intr(txq); 2860 return; 2861 } 2862 2863 #ifdef DEV_NETMAP 2864 if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS) 2865 return; 2866 #endif /* DEV_NETMAP */ 2867 2868 VTNET_TXQ_LOCK(txq); 2869 2870 if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { 2871 VTNET_TXQ_UNLOCK(txq); 2872 return; 2873 } 2874 2875 vtnet_txq_eof(txq); 2876 vtnet_txq_start(txq); 2877 2878 VTNET_TXQ_UNLOCK(txq); 2879 } 2880 2881 static void 2882 vtnet_tx_start_all(struct vtnet_softc *sc) 2883 { 2884 struct vtnet_txq *txq; 2885 int i; 2886 2887 VTNET_CORE_LOCK_ASSERT(sc); 2888 2889 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2890 txq = &sc->vtnet_txqs[i]; 2891 2892 VTNET_TXQ_LOCK(txq); 2893 vtnet_txq_start(txq); 2894 VTNET_TXQ_UNLOCK(txq); 2895 } 2896 } 2897 2898 #ifndef VTNET_LEGACY_TX 2899 static void 2900 vtnet_qflush(struct ifnet *ifp) 2901 { 2902 struct vtnet_softc *sc; 2903 struct vtnet_txq *txq; 2904 struct mbuf *m; 2905 int i; 2906 2907 sc = ifp->if_softc; 2908 2909 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2910 txq = &sc->vtnet_txqs[i]; 2911 2912 VTNET_TXQ_LOCK(txq); 2913 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2914 m_freem(m); 2915 VTNET_TXQ_UNLOCK(txq); 2916 } 2917 2918 if_qflush(ifp); 2919 } 2920 #endif 2921 2922 static int 2923 vtnet_watchdog(struct vtnet_txq *txq) 2924 { 2925 struct ifnet *ifp; 2926 2927 ifp = txq->vtntx_sc->vtnet_ifp; 2928 2929 VTNET_TXQ_LOCK(txq); 2930 if (txq->vtntx_watchdog == 1) { 2931 /* 2932 * Only drain completed frames if the watchdog is about to 2933 * expire. If any frames were drained, there may be enough 2934 * free descriptors now available to transmit queued frames. 2935 * In that case, the timer will immediately be decremented 2936 * below, but the timeout is generous enough that should not 2937 * be a problem. 2938 */ 2939 if (vtnet_txq_eof(txq) != 0) 2940 vtnet_txq_start(txq); 2941 } 2942 2943 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 2944 VTNET_TXQ_UNLOCK(txq); 2945 return (0); 2946 } 2947 VTNET_TXQ_UNLOCK(txq); 2948 2949 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id); 2950 return (1); 2951 } 2952 2953 static void 2954 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc, 2955 struct vtnet_txq_stats *txacc) 2956 { 2957 2958 bzero(rxacc, sizeof(struct vtnet_rxq_stats)); 2959 bzero(txacc, sizeof(struct vtnet_txq_stats)); 2960 2961 for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2962 struct vtnet_rxq_stats *rxst; 2963 struct vtnet_txq_stats *txst; 2964 2965 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 2966 rxacc->vrxs_ipackets += rxst->vrxs_ipackets; 2967 rxacc->vrxs_ibytes += rxst->vrxs_ibytes; 2968 rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops; 2969 rxacc->vrxs_csum += rxst->vrxs_csum; 2970 rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed; 2971 rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled; 2972 2973 txst = &sc->vtnet_txqs[i].vtntx_stats; 2974 txacc->vtxs_opackets += txst->vtxs_opackets; 2975 txacc->vtxs_obytes += txst->vtxs_obytes; 2976 txacc->vtxs_csum += txst->vtxs_csum; 2977 txacc->vtxs_tso += txst->vtxs_tso; 2978 txacc->vtxs_rescheduled += txst->vtxs_rescheduled; 2979 } 2980 } 2981 2982 static uint64_t 2983 vtnet_get_counter(if_t ifp, ift_counter cnt) 2984 { 2985 struct vtnet_softc *sc; 2986 struct vtnet_rxq_stats rxaccum; 2987 struct vtnet_txq_stats txaccum; 2988 2989 sc = if_getsoftc(ifp); 2990 vtnet_accum_stats(sc, &rxaccum, &txaccum); 2991 2992 switch (cnt) { 2993 case IFCOUNTER_IPACKETS: 2994 return (rxaccum.vrxs_ipackets); 2995 case IFCOUNTER_IQDROPS: 2996 return (rxaccum.vrxs_iqdrops); 2997 case IFCOUNTER_IERRORS: 2998 return (rxaccum.vrxs_ierrors); 2999 case IFCOUNTER_OPACKETS: 3000 return (txaccum.vtxs_opackets); 3001 #ifndef VTNET_LEGACY_TX 3002 case IFCOUNTER_OBYTES: 3003 return (txaccum.vtxs_obytes); 3004 case IFCOUNTER_OMCASTS: 3005 return (txaccum.vtxs_omcasts); 3006 #endif 3007 default: 3008 return (if_get_counter_default(ifp, cnt)); 3009 } 3010 } 3011 3012 static void 3013 vtnet_tick(void *xsc) 3014 { 3015 struct vtnet_softc *sc; 3016 struct ifnet *ifp; 3017 int i, timedout; 3018 3019 sc = xsc; 3020 ifp = sc->vtnet_ifp; 3021 timedout = 0; 3022 3023 VTNET_CORE_LOCK_ASSERT(sc); 3024 3025 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3026 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 3027 3028 if (timedout != 0) { 3029 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 3030 vtnet_init_locked(sc, 0); 3031 } else 3032 callout_schedule(&sc->vtnet_tick_ch, hz); 3033 } 3034 3035 static void 3036 vtnet_start_taskqueues(struct vtnet_softc *sc) 3037 { 3038 device_t dev; 3039 struct vtnet_rxq *rxq; 3040 struct vtnet_txq *txq; 3041 int i, error; 3042 3043 dev = sc->vtnet_dev; 3044 3045 /* 3046 * Errors here are very difficult to recover from - we cannot 3047 * easily fail because, if this is during boot, we will hang 3048 * when freeing any successfully started taskqueues because 3049 * the scheduler isn't up yet. 3050 * 3051 * Most drivers just ignore the return value - it only fails 3052 * with ENOMEM so an error is not likely. 3053 */ 3054 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) { 3055 rxq = &sc->vtnet_rxqs[i]; 3056 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 3057 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 3058 if (error) { 3059 device_printf(dev, "failed to start rx taskq %d\n", 3060 rxq->vtnrx_id); 3061 } 3062 3063 txq = &sc->vtnet_txqs[i]; 3064 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 3065 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 3066 if (error) { 3067 device_printf(dev, "failed to start tx taskq %d\n", 3068 txq->vtntx_id); 3069 } 3070 } 3071 } 3072 3073 static void 3074 vtnet_free_taskqueues(struct vtnet_softc *sc) 3075 { 3076 struct vtnet_rxq *rxq; 3077 struct vtnet_txq *txq; 3078 int i; 3079 3080 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3081 rxq = &sc->vtnet_rxqs[i]; 3082 if (rxq->vtnrx_tq != NULL) { 3083 taskqueue_free(rxq->vtnrx_tq); 3084 rxq->vtnrx_tq = NULL; 3085 } 3086 3087 txq = &sc->vtnet_txqs[i]; 3088 if (txq->vtntx_tq != NULL) { 3089 taskqueue_free(txq->vtntx_tq); 3090 txq->vtntx_tq = NULL; 3091 } 3092 } 3093 } 3094 3095 static void 3096 vtnet_drain_taskqueues(struct vtnet_softc *sc) 3097 { 3098 struct vtnet_rxq *rxq; 3099 struct vtnet_txq *txq; 3100 int i; 3101 3102 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3103 rxq = &sc->vtnet_rxqs[i]; 3104 if (rxq->vtnrx_tq != NULL) 3105 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 3106 3107 txq = &sc->vtnet_txqs[i]; 3108 if (txq->vtntx_tq != NULL) { 3109 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 3110 #ifndef VTNET_LEGACY_TX 3111 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 3112 #endif 3113 } 3114 } 3115 } 3116 3117 static void 3118 vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 3119 { 3120 struct vtnet_rxq *rxq; 3121 struct vtnet_txq *txq; 3122 int i; 3123 3124 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3125 rxq = &sc->vtnet_rxqs[i]; 3126 vtnet_rxq_free_mbufs(rxq); 3127 3128 txq = &sc->vtnet_txqs[i]; 3129 vtnet_txq_free_mbufs(txq); 3130 } 3131 } 3132 3133 static void 3134 vtnet_stop_rendezvous(struct vtnet_softc *sc) 3135 { 3136 struct vtnet_rxq *rxq; 3137 struct vtnet_txq *txq; 3138 int i; 3139 3140 VTNET_CORE_LOCK_ASSERT(sc); 3141 3142 /* 3143 * Lock and unlock the per-queue mutex so we known the stop 3144 * state is visible. Doing only the active queues should be 3145 * sufficient, but it does not cost much extra to do all the 3146 * queues. 3147 */ 3148 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3149 rxq = &sc->vtnet_rxqs[i]; 3150 VTNET_RXQ_LOCK(rxq); 3151 VTNET_RXQ_UNLOCK(rxq); 3152 3153 txq = &sc->vtnet_txqs[i]; 3154 VTNET_TXQ_LOCK(txq); 3155 VTNET_TXQ_UNLOCK(txq); 3156 } 3157 } 3158 3159 static void 3160 vtnet_stop(struct vtnet_softc *sc) 3161 { 3162 device_t dev; 3163 struct ifnet *ifp; 3164 3165 dev = sc->vtnet_dev; 3166 ifp = sc->vtnet_ifp; 3167 3168 VTNET_CORE_LOCK_ASSERT(sc); 3169 3170 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 3171 sc->vtnet_link_active = 0; 3172 callout_stop(&sc->vtnet_tick_ch); 3173 3174 /* Only advisory. */ 3175 vtnet_disable_interrupts(sc); 3176 3177 #ifdef DEV_NETMAP 3178 /* Stop any pending txsync/rxsync and disable them. */ 3179 netmap_disable_all_rings(ifp); 3180 #endif /* DEV_NETMAP */ 3181 3182 /* 3183 * Stop the host adapter. This resets it to the pre-initialized 3184 * state. It will not generate any interrupts until after it is 3185 * reinitialized. 3186 */ 3187 virtio_stop(dev); 3188 vtnet_stop_rendezvous(sc); 3189 3190 vtnet_drain_rxtx_queues(sc); 3191 sc->vtnet_act_vq_pairs = 1; 3192 } 3193 3194 static int 3195 vtnet_virtio_reinit(struct vtnet_softc *sc) 3196 { 3197 device_t dev; 3198 struct ifnet *ifp; 3199 uint64_t features; 3200 int error; 3201 3202 dev = sc->vtnet_dev; 3203 ifp = sc->vtnet_ifp; 3204 features = sc->vtnet_negotiated_features; 3205 3206 /* 3207 * Re-negotiate with the host, removing any disabled receive 3208 * features. Transmit features are disabled only on our side 3209 * via if_capenable and if_hwassist. 3210 */ 3211 3212 if ((ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0) 3213 features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES); 3214 3215 if ((ifp->if_capenable & IFCAP_LRO) == 0) 3216 features &= ~VTNET_LRO_FEATURES; 3217 3218 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0) 3219 features &= ~VIRTIO_NET_F_CTRL_VLAN; 3220 3221 error = virtio_reinit(dev, features); 3222 if (error) { 3223 device_printf(dev, "virtio reinit error %d\n", error); 3224 return (error); 3225 } 3226 3227 sc->vtnet_features = features; 3228 virtio_reinit_complete(dev); 3229 3230 return (0); 3231 } 3232 3233 static void 3234 vtnet_init_rx_filters(struct vtnet_softc *sc) 3235 { 3236 struct ifnet *ifp; 3237 3238 ifp = sc->vtnet_ifp; 3239 3240 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 3241 vtnet_rx_filter(sc); 3242 vtnet_rx_filter_mac(sc); 3243 } 3244 3245 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) 3246 vtnet_rx_filter_vlan(sc); 3247 } 3248 3249 static int 3250 vtnet_init_rx_queues(struct vtnet_softc *sc) 3251 { 3252 device_t dev; 3253 struct ifnet *ifp; 3254 struct vtnet_rxq *rxq; 3255 int i, clustersz, error; 3256 3257 dev = sc->vtnet_dev; 3258 ifp = sc->vtnet_ifp; 3259 3260 clustersz = vtnet_rx_cluster_size(sc, ifp->if_mtu); 3261 sc->vtnet_rx_clustersz = clustersz; 3262 3263 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) { 3264 sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) + 3265 VTNET_MAX_RX_SIZE, clustersz); 3266 KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 3267 ("%s: too many rx mbufs %d for %d segments", __func__, 3268 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 3269 } else 3270 sc->vtnet_rx_nmbufs = 1; 3271 3272 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3273 rxq = &sc->vtnet_rxqs[i]; 3274 3275 /* Hold the lock to satisfy asserts. */ 3276 VTNET_RXQ_LOCK(rxq); 3277 error = vtnet_rxq_populate(rxq); 3278 VTNET_RXQ_UNLOCK(rxq); 3279 3280 if (error) { 3281 device_printf(dev, "cannot populate Rx queue %d\n", i); 3282 return (error); 3283 } 3284 } 3285 3286 return (0); 3287 } 3288 3289 static int 3290 vtnet_init_tx_queues(struct vtnet_softc *sc) 3291 { 3292 struct vtnet_txq *txq; 3293 int i; 3294 3295 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3296 txq = &sc->vtnet_txqs[i]; 3297 txq->vtntx_watchdog = 0; 3298 txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq); 3299 #ifdef DEV_NETMAP 3300 netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0); 3301 #endif /* DEV_NETMAP */ 3302 } 3303 3304 return (0); 3305 } 3306 3307 static int 3308 vtnet_init_rxtx_queues(struct vtnet_softc *sc) 3309 { 3310 int error; 3311 3312 error = vtnet_init_rx_queues(sc); 3313 if (error) 3314 return (error); 3315 3316 error = vtnet_init_tx_queues(sc); 3317 if (error) 3318 return (error); 3319 3320 return (0); 3321 } 3322 3323 static void 3324 vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 3325 { 3326 device_t dev; 3327 int npairs; 3328 3329 dev = sc->vtnet_dev; 3330 3331 if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) { 3332 sc->vtnet_act_vq_pairs = 1; 3333 return; 3334 } 3335 3336 npairs = sc->vtnet_req_vq_pairs; 3337 3338 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 3339 device_printf(dev, "cannot set active queue pairs to %d, " 3340 "falling back to 1 queue pair\n", npairs); 3341 npairs = 1; 3342 } 3343 3344 sc->vtnet_act_vq_pairs = npairs; 3345 } 3346 3347 static void 3348 vtnet_update_rx_offloads(struct vtnet_softc *sc) 3349 { 3350 struct ifnet *ifp; 3351 uint64_t features; 3352 int error; 3353 3354 ifp = sc->vtnet_ifp; 3355 features = sc->vtnet_features; 3356 3357 VTNET_CORE_LOCK_ASSERT(sc); 3358 3359 if (ifp->if_capabilities & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 3360 if (ifp->if_capenable & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) 3361 features |= VIRTIO_NET_F_GUEST_CSUM; 3362 else 3363 features &= ~VIRTIO_NET_F_GUEST_CSUM; 3364 } 3365 3366 if (ifp->if_capabilities & IFCAP_LRO && !vtnet_software_lro(sc)) { 3367 if (ifp->if_capenable & IFCAP_LRO) 3368 features |= VTNET_LRO_FEATURES; 3369 else 3370 features &= ~VTNET_LRO_FEATURES; 3371 } 3372 3373 error = vtnet_ctrl_guest_offloads(sc, 3374 features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 | 3375 VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN | 3376 VIRTIO_NET_F_GUEST_UFO)); 3377 if (error) { 3378 device_printf(sc->vtnet_dev, 3379 "%s: cannot update Rx features\n", __func__); 3380 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 3381 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 3382 vtnet_init_locked(sc, 0); 3383 } 3384 } else 3385 sc->vtnet_features = features; 3386 } 3387 3388 static int 3389 vtnet_reinit(struct vtnet_softc *sc) 3390 { 3391 device_t dev; 3392 struct ifnet *ifp; 3393 int error; 3394 3395 dev = sc->vtnet_dev; 3396 ifp = sc->vtnet_ifp; 3397 3398 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 3399 3400 error = vtnet_virtio_reinit(sc); 3401 if (error) 3402 return (error); 3403 3404 vtnet_set_macaddr(sc); 3405 vtnet_set_active_vq_pairs(sc); 3406 3407 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 3408 vtnet_init_rx_filters(sc); 3409 3410 ifp->if_hwassist = 0; 3411 if (ifp->if_capenable & IFCAP_TXCSUM) 3412 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD; 3413 if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) 3414 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD_IPV6; 3415 if (ifp->if_capenable & IFCAP_TSO4) 3416 ifp->if_hwassist |= CSUM_IP_TSO; 3417 if (ifp->if_capenable & IFCAP_TSO6) 3418 ifp->if_hwassist |= CSUM_IP6_TSO; 3419 3420 error = vtnet_init_rxtx_queues(sc); 3421 if (error) 3422 return (error); 3423 3424 return (0); 3425 } 3426 3427 static void 3428 vtnet_init_locked(struct vtnet_softc *sc, int init_mode) 3429 { 3430 device_t dev; 3431 struct ifnet *ifp; 3432 3433 dev = sc->vtnet_dev; 3434 ifp = sc->vtnet_ifp; 3435 3436 VTNET_CORE_LOCK_ASSERT(sc); 3437 3438 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 3439 return; 3440 3441 vtnet_stop(sc); 3442 3443 #ifdef DEV_NETMAP 3444 /* Once stopped we can update the netmap flags, if necessary. */ 3445 switch (init_mode) { 3446 case VTNET_INIT_NETMAP_ENTER: 3447 nm_set_native_flags(NA(ifp)); 3448 break; 3449 case VTNET_INIT_NETMAP_EXIT: 3450 nm_clear_native_flags(NA(ifp)); 3451 break; 3452 } 3453 #endif /* DEV_NETMAP */ 3454 3455 if (vtnet_reinit(sc) != 0) { 3456 vtnet_stop(sc); 3457 return; 3458 } 3459 3460 ifp->if_drv_flags |= IFF_DRV_RUNNING; 3461 vtnet_update_link_status(sc); 3462 vtnet_enable_interrupts(sc); 3463 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3464 3465 #ifdef DEV_NETMAP 3466 /* Re-enable txsync/rxsync. */ 3467 netmap_enable_all_rings(ifp); 3468 #endif /* DEV_NETMAP */ 3469 } 3470 3471 static void 3472 vtnet_init(void *xsc) 3473 { 3474 struct vtnet_softc *sc; 3475 3476 sc = xsc; 3477 3478 VTNET_CORE_LOCK(sc); 3479 vtnet_init_locked(sc, 0); 3480 VTNET_CORE_UNLOCK(sc); 3481 } 3482 3483 static void 3484 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3485 { 3486 3487 /* 3488 * The control virtqueue is only polled and therefore it should 3489 * already be empty. 3490 */ 3491 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq), 3492 ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq)); 3493 } 3494 3495 static void 3496 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3497 struct sglist *sg, int readable, int writable) 3498 { 3499 struct virtqueue *vq; 3500 3501 vq = sc->vtnet_ctrl_vq; 3502 3503 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ); 3504 VTNET_CORE_LOCK_ASSERT(sc); 3505 3506 if (!virtqueue_empty(vq)) 3507 return; 3508 3509 /* 3510 * Poll for the response, but the command is likely completed before 3511 * returning from the notify. 3512 */ 3513 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0) { 3514 virtqueue_notify(vq); 3515 virtqueue_poll(vq, NULL); 3516 } 3517 } 3518 3519 static int 3520 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3521 { 3522 struct sglist_seg segs[3]; 3523 struct sglist sg; 3524 struct { 3525 struct virtio_net_ctrl_hdr hdr __aligned(2); 3526 uint8_t pad1; 3527 uint8_t addr[ETHER_ADDR_LEN] __aligned(8); 3528 uint8_t pad2; 3529 uint8_t ack; 3530 } s; 3531 int error; 3532 3533 error = 0; 3534 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC); 3535 3536 s.hdr.class = VIRTIO_NET_CTRL_MAC; 3537 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3538 bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN); 3539 s.ack = VIRTIO_NET_ERR; 3540 3541 sglist_init(&sg, nitems(segs), segs); 3542 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3543 error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN); 3544 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3545 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3546 3547 if (error == 0) 3548 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3549 3550 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3551 } 3552 3553 static int 3554 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads) 3555 { 3556 struct sglist_seg segs[3]; 3557 struct sglist sg; 3558 struct { 3559 struct virtio_net_ctrl_hdr hdr __aligned(2); 3560 uint8_t pad1; 3561 uint64_t offloads __aligned(8); 3562 uint8_t pad2; 3563 uint8_t ack; 3564 } s; 3565 int error; 3566 3567 error = 0; 3568 MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 3569 3570 s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS; 3571 s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET; 3572 s.offloads = vtnet_gtoh64(sc, offloads); 3573 s.ack = VIRTIO_NET_ERR; 3574 3575 sglist_init(&sg, nitems(segs), segs); 3576 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3577 error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t)); 3578 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3579 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3580 3581 if (error == 0) 3582 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3583 3584 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3585 } 3586 3587 static int 3588 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3589 { 3590 struct sglist_seg segs[3]; 3591 struct sglist sg; 3592 struct { 3593 struct virtio_net_ctrl_hdr hdr __aligned(2); 3594 uint8_t pad1; 3595 struct virtio_net_ctrl_mq mq __aligned(2); 3596 uint8_t pad2; 3597 uint8_t ack; 3598 } s; 3599 int error; 3600 3601 error = 0; 3602 MPASS(sc->vtnet_flags & VTNET_FLAG_MQ); 3603 3604 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3605 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3606 s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs); 3607 s.ack = VIRTIO_NET_ERR; 3608 3609 sglist_init(&sg, nitems(segs), segs); 3610 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3611 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3612 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3613 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3614 3615 if (error == 0) 3616 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3617 3618 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3619 } 3620 3621 static int 3622 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on) 3623 { 3624 struct sglist_seg segs[3]; 3625 struct sglist sg; 3626 struct { 3627 struct virtio_net_ctrl_hdr hdr __aligned(2); 3628 uint8_t pad1; 3629 uint8_t onoff; 3630 uint8_t pad2; 3631 uint8_t ack; 3632 } s; 3633 int error; 3634 3635 error = 0; 3636 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX); 3637 3638 s.hdr.class = VIRTIO_NET_CTRL_RX; 3639 s.hdr.cmd = cmd; 3640 s.onoff = on; 3641 s.ack = VIRTIO_NET_ERR; 3642 3643 sglist_init(&sg, nitems(segs), segs); 3644 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3645 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3646 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3647 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3648 3649 if (error == 0) 3650 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3651 3652 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3653 } 3654 3655 static int 3656 vtnet_set_promisc(struct vtnet_softc *sc, bool on) 3657 { 3658 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3659 } 3660 3661 static int 3662 vtnet_set_allmulti(struct vtnet_softc *sc, bool on) 3663 { 3664 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3665 } 3666 3667 static void 3668 vtnet_rx_filter(struct vtnet_softc *sc) 3669 { 3670 device_t dev; 3671 struct ifnet *ifp; 3672 3673 dev = sc->vtnet_dev; 3674 ifp = sc->vtnet_ifp; 3675 3676 VTNET_CORE_LOCK_ASSERT(sc); 3677 3678 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0) { 3679 device_printf(dev, "cannot %s promiscuous mode\n", 3680 ifp->if_flags & IFF_PROMISC ? "enable" : "disable"); 3681 } 3682 3683 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0) { 3684 device_printf(dev, "cannot %s all-multicast mode\n", 3685 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable"); 3686 } 3687 } 3688 3689 static u_int 3690 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt) 3691 { 3692 struct vtnet_softc *sc = arg; 3693 3694 if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3695 return (0); 3696 3697 if (ucnt < VTNET_MAX_MAC_ENTRIES) 3698 bcopy(LLADDR(sdl), 3699 &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt], 3700 ETHER_ADDR_LEN); 3701 3702 return (1); 3703 } 3704 3705 static u_int 3706 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt) 3707 { 3708 struct vtnet_mac_filter *filter = arg; 3709 3710 if (mcnt < VTNET_MAX_MAC_ENTRIES) 3711 bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt], 3712 ETHER_ADDR_LEN); 3713 3714 return (1); 3715 } 3716 3717 static void 3718 vtnet_rx_filter_mac(struct vtnet_softc *sc) 3719 { 3720 struct virtio_net_ctrl_hdr hdr __aligned(2); 3721 struct vtnet_mac_filter *filter; 3722 struct sglist_seg segs[4]; 3723 struct sglist sg; 3724 struct ifnet *ifp; 3725 bool promisc, allmulti; 3726 u_int ucnt, mcnt; 3727 int error; 3728 uint8_t ack; 3729 3730 ifp = sc->vtnet_ifp; 3731 filter = sc->vtnet_mac_filter; 3732 error = 0; 3733 3734 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX); 3735 VTNET_CORE_LOCK_ASSERT(sc); 3736 3737 /* Unicast MAC addresses: */ 3738 ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc); 3739 promisc = (ucnt > VTNET_MAX_MAC_ENTRIES); 3740 3741 if (promisc) { 3742 ucnt = 0; 3743 if_printf(ifp, "more than %d MAC addresses assigned, " 3744 "falling back to promiscuous mode\n", 3745 VTNET_MAX_MAC_ENTRIES); 3746 } 3747 3748 /* Multicast MAC addresses: */ 3749 mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter); 3750 allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES); 3751 3752 if (allmulti) { 3753 mcnt = 0; 3754 if_printf(ifp, "more than %d multicast MAC addresses " 3755 "assigned, falling back to all-multicast mode\n", 3756 VTNET_MAX_MAC_ENTRIES); 3757 } 3758 3759 if (promisc && allmulti) 3760 goto out; 3761 3762 filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt); 3763 filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt); 3764 3765 hdr.class = VIRTIO_NET_CTRL_MAC; 3766 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3767 ack = VIRTIO_NET_ERR; 3768 3769 sglist_init(&sg, nitems(segs), segs); 3770 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3771 error |= sglist_append(&sg, &filter->vmf_unicast, 3772 sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN); 3773 error |= sglist_append(&sg, &filter->vmf_multicast, 3774 sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN); 3775 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3776 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3777 3778 if (error == 0) 3779 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3780 if (ack != VIRTIO_NET_OK) 3781 if_printf(ifp, "error setting host MAC filter table\n"); 3782 3783 out: 3784 if (promisc != 0 && vtnet_set_promisc(sc, true) != 0) 3785 if_printf(ifp, "cannot enable promiscuous mode\n"); 3786 if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0) 3787 if_printf(ifp, "cannot enable all-multicast mode\n"); 3788 } 3789 3790 static int 3791 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3792 { 3793 struct sglist_seg segs[3]; 3794 struct sglist sg; 3795 struct { 3796 struct virtio_net_ctrl_hdr hdr __aligned(2); 3797 uint8_t pad1; 3798 uint16_t tag __aligned(2); 3799 uint8_t pad2; 3800 uint8_t ack; 3801 } s; 3802 int error; 3803 3804 error = 0; 3805 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER); 3806 3807 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3808 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3809 s.tag = vtnet_gtoh16(sc, tag); 3810 s.ack = VIRTIO_NET_ERR; 3811 3812 sglist_init(&sg, nitems(segs), segs); 3813 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3814 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3815 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3816 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3817 3818 if (error == 0) 3819 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3820 3821 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3822 } 3823 3824 static void 3825 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3826 { 3827 int i, bit; 3828 uint32_t w; 3829 uint16_t tag; 3830 3831 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER); 3832 VTNET_CORE_LOCK_ASSERT(sc); 3833 3834 /* Enable the filter for each configured VLAN. */ 3835 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3836 w = sc->vtnet_vlan_filter[i]; 3837 3838 while ((bit = ffs(w) - 1) != -1) { 3839 w &= ~(1 << bit); 3840 tag = sizeof(w) * CHAR_BIT * i + bit; 3841 3842 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3843 device_printf(sc->vtnet_dev, 3844 "cannot enable VLAN %d filter\n", tag); 3845 } 3846 } 3847 } 3848 } 3849 3850 static void 3851 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3852 { 3853 struct ifnet *ifp; 3854 int idx, bit; 3855 3856 ifp = sc->vtnet_ifp; 3857 idx = (tag >> 5) & 0x7F; 3858 bit = tag & 0x1F; 3859 3860 if (tag == 0 || tag > 4095) 3861 return; 3862 3863 VTNET_CORE_LOCK(sc); 3864 3865 if (add) 3866 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3867 else 3868 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3869 3870 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER && 3871 ifp->if_drv_flags & IFF_DRV_RUNNING && 3872 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3873 device_printf(sc->vtnet_dev, 3874 "cannot %s VLAN %d %s the host filter table\n", 3875 add ? "add" : "remove", tag, add ? "to" : "from"); 3876 } 3877 3878 VTNET_CORE_UNLOCK(sc); 3879 } 3880 3881 static void 3882 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3883 { 3884 3885 if (ifp->if_softc != arg) 3886 return; 3887 3888 vtnet_update_vlan_filter(arg, 1, tag); 3889 } 3890 3891 static void 3892 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag) 3893 { 3894 3895 if (ifp->if_softc != arg) 3896 return; 3897 3898 vtnet_update_vlan_filter(arg, 0, tag); 3899 } 3900 3901 static void 3902 vtnet_update_speed_duplex(struct vtnet_softc *sc) 3903 { 3904 struct ifnet *ifp; 3905 uint32_t speed; 3906 3907 ifp = sc->vtnet_ifp; 3908 3909 if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0) 3910 return; 3911 3912 /* BMV: Ignore duplex. */ 3913 speed = virtio_read_dev_config_4(sc->vtnet_dev, 3914 offsetof(struct virtio_net_config, speed)); 3915 if (speed != UINT32_MAX) 3916 ifp->if_baudrate = IF_Mbps(speed); 3917 } 3918 3919 static int 3920 vtnet_is_link_up(struct vtnet_softc *sc) 3921 { 3922 uint16_t status; 3923 3924 if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0) 3925 return (1); 3926 3927 status = virtio_read_dev_config_2(sc->vtnet_dev, 3928 offsetof(struct virtio_net_config, status)); 3929 3930 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3931 } 3932 3933 static void 3934 vtnet_update_link_status(struct vtnet_softc *sc) 3935 { 3936 struct ifnet *ifp; 3937 int link; 3938 3939 ifp = sc->vtnet_ifp; 3940 VTNET_CORE_LOCK_ASSERT(sc); 3941 link = vtnet_is_link_up(sc); 3942 3943 /* Notify if the link status has changed. */ 3944 if (link != 0 && sc->vtnet_link_active == 0) { 3945 vtnet_update_speed_duplex(sc); 3946 sc->vtnet_link_active = 1; 3947 if_link_state_change(ifp, LINK_STATE_UP); 3948 } else if (link == 0 && sc->vtnet_link_active != 0) { 3949 sc->vtnet_link_active = 0; 3950 if_link_state_change(ifp, LINK_STATE_DOWN); 3951 } 3952 } 3953 3954 static int 3955 vtnet_ifmedia_upd(struct ifnet *ifp __unused) 3956 { 3957 return (EOPNOTSUPP); 3958 } 3959 3960 static void 3961 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 3962 { 3963 struct vtnet_softc *sc; 3964 3965 sc = ifp->if_softc; 3966 3967 ifmr->ifm_status = IFM_AVALID; 3968 ifmr->ifm_active = IFM_ETHER; 3969 3970 VTNET_CORE_LOCK(sc); 3971 if (vtnet_is_link_up(sc) != 0) { 3972 ifmr->ifm_status |= IFM_ACTIVE; 3973 ifmr->ifm_active |= IFM_10G_T | IFM_FDX; 3974 } else 3975 ifmr->ifm_active |= IFM_NONE; 3976 VTNET_CORE_UNLOCK(sc); 3977 } 3978 3979 static void 3980 vtnet_get_macaddr(struct vtnet_softc *sc) 3981 { 3982 3983 if (sc->vtnet_flags & VTNET_FLAG_MAC) { 3984 virtio_read_device_config_array(sc->vtnet_dev, 3985 offsetof(struct virtio_net_config, mac), 3986 &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN); 3987 } else { 3988 /* Generate a random locally administered unicast address. */ 3989 sc->vtnet_hwaddr[0] = 0xB2; 3990 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 3991 } 3992 } 3993 3994 static void 3995 vtnet_set_macaddr(struct vtnet_softc *sc) 3996 { 3997 device_t dev; 3998 int error; 3999 4000 dev = sc->vtnet_dev; 4001 4002 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 4003 error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr); 4004 if (error) 4005 device_printf(dev, "unable to set MAC address\n"); 4006 return; 4007 } 4008 4009 /* MAC in config is read-only in modern VirtIO. */ 4010 if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) { 4011 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4012 virtio_write_dev_config_1(dev, 4013 offsetof(struct virtio_net_config, mac) + i, 4014 sc->vtnet_hwaddr[i]); 4015 } 4016 } 4017 } 4018 4019 static void 4020 vtnet_attached_set_macaddr(struct vtnet_softc *sc) 4021 { 4022 4023 /* Assign MAC address if it was generated. */ 4024 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) 4025 vtnet_set_macaddr(sc); 4026 } 4027 4028 static void 4029 vtnet_vlan_tag_remove(struct mbuf *m) 4030 { 4031 struct ether_vlan_header *evh; 4032 4033 evh = mtod(m, struct ether_vlan_header *); 4034 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 4035 m->m_flags |= M_VLANTAG; 4036 4037 /* Strip the 802.1Q header. */ 4038 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 4039 ETHER_HDR_LEN - ETHER_TYPE_LEN); 4040 m_adj(m, ETHER_VLAN_ENCAP_LEN); 4041 } 4042 4043 static void 4044 vtnet_set_rx_process_limit(struct vtnet_softc *sc) 4045 { 4046 int limit; 4047 4048 limit = vtnet_tunable_int(sc, "rx_process_limit", 4049 vtnet_rx_process_limit); 4050 if (limit < 0) 4051 limit = INT_MAX; 4052 sc->vtnet_rx_process_limit = limit; 4053 } 4054 4055 static void 4056 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 4057 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 4058 { 4059 struct sysctl_oid *node; 4060 struct sysctl_oid_list *list; 4061 struct vtnet_rxq_stats *stats; 4062 char namebuf[16]; 4063 4064 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 4065 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 4066 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue"); 4067 list = SYSCTL_CHILDREN(node); 4068 4069 stats = &rxq->vtnrx_stats; 4070 4071 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, 4072 &stats->vrxs_ipackets, "Receive packets"); 4073 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, 4074 &stats->vrxs_ibytes, "Receive bytes"); 4075 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, 4076 &stats->vrxs_iqdrops, "Receive drops"); 4077 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, 4078 &stats->vrxs_ierrors, "Receive errors"); 4079 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 4080 &stats->vrxs_csum, "Receive checksum offloaded"); 4081 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, 4082 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 4083 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD, 4084 &stats->vrxs_host_lro, "Receive host segmentation offloaded"); 4085 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 4086 &stats->vrxs_rescheduled, 4087 "Receive interrupt handler rescheduled"); 4088 } 4089 4090 static void 4091 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 4092 struct sysctl_oid_list *child, struct vtnet_txq *txq) 4093 { 4094 struct sysctl_oid *node; 4095 struct sysctl_oid_list *list; 4096 struct vtnet_txq_stats *stats; 4097 char namebuf[16]; 4098 4099 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 4100 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 4101 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue"); 4102 list = SYSCTL_CHILDREN(node); 4103 4104 stats = &txq->vtntx_stats; 4105 4106 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, 4107 &stats->vtxs_opackets, "Transmit packets"); 4108 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, 4109 &stats->vtxs_obytes, "Transmit bytes"); 4110 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, 4111 &stats->vtxs_omcasts, "Transmit multicasts"); 4112 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 4113 &stats->vtxs_csum, "Transmit checksum offloaded"); 4114 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, 4115 &stats->vtxs_tso, "Transmit TCP segmentation offloaded"); 4116 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 4117 &stats->vtxs_rescheduled, 4118 "Transmit interrupt handler rescheduled"); 4119 } 4120 4121 static void 4122 vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 4123 { 4124 device_t dev; 4125 struct sysctl_ctx_list *ctx; 4126 struct sysctl_oid *tree; 4127 struct sysctl_oid_list *child; 4128 int i; 4129 4130 dev = sc->vtnet_dev; 4131 ctx = device_get_sysctl_ctx(dev); 4132 tree = device_get_sysctl_tree(dev); 4133 child = SYSCTL_CHILDREN(tree); 4134 4135 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) { 4136 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 4137 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 4138 } 4139 } 4140 4141 static void 4142 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 4143 struct sysctl_oid_list *child, struct vtnet_softc *sc) 4144 { 4145 struct vtnet_statistics *stats; 4146 struct vtnet_rxq_stats rxaccum; 4147 struct vtnet_txq_stats txaccum; 4148 4149 vtnet_accum_stats(sc, &rxaccum, &txaccum); 4150 4151 stats = &sc->vtnet_stats; 4152 stats->rx_csum_offloaded = rxaccum.vrxs_csum; 4153 stats->rx_csum_failed = rxaccum.vrxs_csum_failed; 4154 stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 4155 stats->tx_csum_offloaded = txaccum.vtxs_csum; 4156 stats->tx_tso_offloaded = txaccum.vtxs_tso; 4157 stats->tx_task_rescheduled = txaccum.vtxs_rescheduled; 4158 4159 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 4160 CTLFLAG_RD, &stats->mbuf_alloc_failed, 4161 "Mbuf cluster allocation failures"); 4162 4163 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 4164 CTLFLAG_RD, &stats->rx_frame_too_large, 4165 "Received frame larger than the mbuf chain"); 4166 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 4167 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 4168 "Enqueuing the replacement receive mbuf failed"); 4169 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 4170 CTLFLAG_RD, &stats->rx_mergeable_failed, 4171 "Mergeable buffers receive failures"); 4172 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 4173 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 4174 "Received checksum offloaded buffer with unsupported " 4175 "Ethernet type"); 4176 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 4177 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 4178 "Received checksum offloaded buffer with incorrect IP protocol"); 4179 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 4180 CTLFLAG_RD, &stats->rx_csum_bad_offset, 4181 "Received checksum offloaded buffer with incorrect offset"); 4182 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", 4183 CTLFLAG_RD, &stats->rx_csum_bad_proto, 4184 "Received checksum offloaded buffer with incorrect protocol"); 4185 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 4186 CTLFLAG_RD, &stats->rx_csum_failed, 4187 "Received buffer checksum offload failed"); 4188 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 4189 CTLFLAG_RD, &stats->rx_csum_offloaded, 4190 "Received buffer checksum offload succeeded"); 4191 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 4192 CTLFLAG_RD, &stats->rx_task_rescheduled, 4193 "Times the receive interrupt task rescheduled itself"); 4194 4195 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype", 4196 CTLFLAG_RD, &stats->tx_csum_unknown_ethtype, 4197 "Aborted transmit of checksum offloaded buffer with unknown " 4198 "Ethernet type"); 4199 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch", 4200 CTLFLAG_RD, &stats->tx_csum_proto_mismatch, 4201 "Aborted transmit of checksum offloaded buffer because mismatched " 4202 "protocols"); 4203 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 4204 CTLFLAG_RD, &stats->tx_tso_not_tcp, 4205 "Aborted transmit of TSO buffer with non TCP protocol"); 4206 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum", 4207 CTLFLAG_RD, &stats->tx_tso_without_csum, 4208 "Aborted transmit of TSO buffer without TCP checksum offload"); 4209 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 4210 CTLFLAG_RD, &stats->tx_defragged, 4211 "Transmit mbufs defragged"); 4212 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 4213 CTLFLAG_RD, &stats->tx_defrag_failed, 4214 "Aborted transmit of buffer because defrag failed"); 4215 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 4216 CTLFLAG_RD, &stats->tx_csum_offloaded, 4217 "Offloaded checksum of transmitted buffer"); 4218 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 4219 CTLFLAG_RD, &stats->tx_tso_offloaded, 4220 "Segmentation offload of transmitted buffer"); 4221 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 4222 CTLFLAG_RD, &stats->tx_task_rescheduled, 4223 "Times the transmit interrupt task rescheduled itself"); 4224 } 4225 4226 static void 4227 vtnet_setup_sysctl(struct vtnet_softc *sc) 4228 { 4229 device_t dev; 4230 struct sysctl_ctx_list *ctx; 4231 struct sysctl_oid *tree; 4232 struct sysctl_oid_list *child; 4233 4234 dev = sc->vtnet_dev; 4235 ctx = device_get_sysctl_ctx(dev); 4236 tree = device_get_sysctl_tree(dev); 4237 child = SYSCTL_CHILDREN(tree); 4238 4239 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 4240 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 4241 "Number of maximum supported virtqueue pairs"); 4242 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs", 4243 CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0, 4244 "Number of requested virtqueue pairs"); 4245 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 4246 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 4247 "Number of active virtqueue pairs"); 4248 4249 vtnet_setup_stat_sysctl(ctx, child, sc); 4250 } 4251 4252 static void 4253 vtnet_load_tunables(struct vtnet_softc *sc) 4254 { 4255 4256 sc->vtnet_lro_entry_count = vtnet_tunable_int(sc, 4257 "lro_entry_count", vtnet_lro_entry_count); 4258 if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES) 4259 sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES; 4260 4261 sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc, 4262 "lro_mbufq_depth", vtnet_lro_mbufq_depth); 4263 } 4264 4265 static int 4266 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 4267 { 4268 4269 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 4270 } 4271 4272 static void 4273 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 4274 { 4275 4276 virtqueue_disable_intr(rxq->vtnrx_vq); 4277 } 4278 4279 static int 4280 vtnet_txq_enable_intr(struct vtnet_txq *txq) 4281 { 4282 struct virtqueue *vq; 4283 4284 vq = txq->vtntx_vq; 4285 4286 if (vtnet_txq_below_threshold(txq) != 0) 4287 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG)); 4288 4289 /* 4290 * The free count is above our threshold. Keep the Tx interrupt 4291 * disabled until the queue is fuller. 4292 */ 4293 return (0); 4294 } 4295 4296 static void 4297 vtnet_txq_disable_intr(struct vtnet_txq *txq) 4298 { 4299 4300 virtqueue_disable_intr(txq->vtntx_vq); 4301 } 4302 4303 static void 4304 vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 4305 { 4306 struct vtnet_rxq *rxq; 4307 int i; 4308 4309 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 4310 rxq = &sc->vtnet_rxqs[i]; 4311 if (vtnet_rxq_enable_intr(rxq) != 0) 4312 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 4313 } 4314 } 4315 4316 static void 4317 vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 4318 { 4319 int i; 4320 4321 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4322 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 4323 } 4324 4325 static void 4326 vtnet_enable_interrupts(struct vtnet_softc *sc) 4327 { 4328 4329 vtnet_enable_rx_interrupts(sc); 4330 vtnet_enable_tx_interrupts(sc); 4331 } 4332 4333 static void 4334 vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 4335 { 4336 int i; 4337 4338 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4339 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 4340 } 4341 4342 static void 4343 vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 4344 { 4345 int i; 4346 4347 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4348 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 4349 } 4350 4351 static void 4352 vtnet_disable_interrupts(struct vtnet_softc *sc) 4353 { 4354 4355 vtnet_disable_rx_interrupts(sc); 4356 vtnet_disable_tx_interrupts(sc); 4357 } 4358 4359 static int 4360 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 4361 { 4362 char path[64]; 4363 4364 snprintf(path, sizeof(path), 4365 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 4366 TUNABLE_INT_FETCH(path, &def); 4367 4368 return (def); 4369 } 4370 4371 #ifdef DEBUGNET 4372 static void 4373 vtnet_debugnet_init(struct ifnet *ifp, int *nrxr, int *ncl, int *clsize) 4374 { 4375 struct vtnet_softc *sc; 4376 4377 sc = if_getsoftc(ifp); 4378 4379 VTNET_CORE_LOCK(sc); 4380 *nrxr = sc->vtnet_req_vq_pairs; 4381 *ncl = DEBUGNET_MAX_IN_FLIGHT; 4382 *clsize = sc->vtnet_rx_clustersz; 4383 VTNET_CORE_UNLOCK(sc); 4384 } 4385 4386 static void 4387 vtnet_debugnet_event(struct ifnet *ifp __unused, enum debugnet_ev event __unused) 4388 { 4389 } 4390 4391 static int 4392 vtnet_debugnet_transmit(struct ifnet *ifp, struct mbuf *m) 4393 { 4394 struct vtnet_softc *sc; 4395 struct vtnet_txq *txq; 4396 int error; 4397 4398 sc = if_getsoftc(ifp); 4399 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4400 IFF_DRV_RUNNING) 4401 return (EBUSY); 4402 4403 txq = &sc->vtnet_txqs[0]; 4404 error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE); 4405 if (error == 0) 4406 (void)vtnet_txq_notify(txq); 4407 return (error); 4408 } 4409 4410 static int 4411 vtnet_debugnet_poll(struct ifnet *ifp, int count) 4412 { 4413 struct vtnet_softc *sc; 4414 int i; 4415 4416 sc = if_getsoftc(ifp); 4417 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4418 IFF_DRV_RUNNING) 4419 return (EBUSY); 4420 4421 (void)vtnet_txq_eof(&sc->vtnet_txqs[0]); 4422 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4423 (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]); 4424 return (0); 4425 } 4426 #endif /* DEBUGNET */ 4427