1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 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 #include <sys/param.h> 33 #include <sys/eventhandler.h> 34 #include <sys/systm.h> 35 #include <sys/kernel.h> 36 #include <sys/sockio.h> 37 #include <sys/malloc.h> 38 #include <sys/mbuf.h> 39 #include <sys/module.h> 40 #include <sys/msan.h> 41 #include <sys/socket.h> 42 #include <sys/sysctl.h> 43 #include <sys/random.h> 44 #include <sys/sglist.h> 45 #include <sys/lock.h> 46 #include <sys/mutex.h> 47 #include <sys/taskqueue.h> 48 #include <sys/smp.h> 49 #include <machine/smp.h> 50 51 #include <vm/uma.h> 52 53 #include <net/debugnet.h> 54 #include <net/ethernet.h> 55 #include <net/pfil.h> 56 #include <net/if.h> 57 #include <net/if_var.h> 58 #include <net/if_arp.h> 59 #include <net/if_dl.h> 60 #include <net/if_types.h> 61 #include <net/if_media.h> 62 #include <net/if_vlan_var.h> 63 64 #include <net/bpf.h> 65 66 #include <netinet/in_systm.h> 67 #include <netinet/in.h> 68 #include <netinet/ip.h> 69 #include <netinet/ip6.h> 70 #include <netinet6/ip6_var.h> 71 #include <netinet/udp.h> 72 #include <netinet/tcp.h> 73 #include <netinet/tcp_lro.h> 74 75 #include <machine/bus.h> 76 #include <machine/resource.h> 77 #include <sys/bus.h> 78 #include <sys/rman.h> 79 80 #include <dev/virtio/virtio.h> 81 #include <dev/virtio/virtqueue.h> 82 #include <dev/virtio/network/virtio_net.h> 83 #include <dev/virtio/network/if_vtnetvar.h> 84 #include "virtio_if.h" 85 86 #include "opt_inet.h" 87 #include "opt_inet6.h" 88 89 #if defined(INET) || defined(INET6) 90 #include <machine/in_cksum.h> 91 #endif 92 93 static int vtnet_modevent(module_t, int, void *); 94 95 static int vtnet_probe(device_t); 96 static int vtnet_attach(device_t); 97 static int vtnet_detach(device_t); 98 static int vtnet_suspend(device_t); 99 static int vtnet_resume(device_t); 100 static int vtnet_shutdown(device_t); 101 static int vtnet_attach_completed(device_t); 102 static int vtnet_config_change(device_t); 103 104 static int vtnet_negotiate_features(struct vtnet_softc *); 105 static int vtnet_setup_features(struct vtnet_softc *); 106 static int vtnet_init_rxq(struct vtnet_softc *, int); 107 static int vtnet_init_txq(struct vtnet_softc *, int); 108 static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); 109 static void vtnet_free_rxtx_queues(struct vtnet_softc *); 110 static int vtnet_alloc_rx_filters(struct vtnet_softc *); 111 static void vtnet_free_rx_filters(struct vtnet_softc *); 112 static int vtnet_alloc_virtqueues(struct vtnet_softc *); 113 static int vtnet_alloc_interface(struct vtnet_softc *); 114 static int vtnet_setup_interface(struct vtnet_softc *); 115 static int vtnet_ioctl_mtu(struct vtnet_softc *, u_int); 116 static int vtnet_ioctl_ifflags(struct vtnet_softc *); 117 static int vtnet_ioctl_multi(struct vtnet_softc *); 118 static int vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *); 119 static int vtnet_ioctl(if_t, u_long, caddr_t); 120 static uint64_t vtnet_get_counter(if_t, ift_counter); 121 122 static int vtnet_rxq_populate(struct vtnet_rxq *); 123 static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); 124 static struct mbuf * 125 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); 126 static int vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *, 127 struct mbuf *, int); 128 static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); 129 static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); 130 static int vtnet_rxq_new_buf(struct vtnet_rxq *); 131 static int vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *, 132 uint16_t, int, struct virtio_net_hdr *); 133 static int vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *, 134 uint16_t, int, struct virtio_net_hdr *); 135 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, 136 struct virtio_net_hdr *); 137 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); 138 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); 139 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); 140 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, 141 struct virtio_net_hdr *); 142 static int vtnet_rxq_eof(struct vtnet_rxq *); 143 static void vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries); 144 static void vtnet_rx_vq_intr(void *); 145 static void vtnet_rxq_tq_intr(void *, int); 146 147 static int vtnet_txq_intr_threshold(struct vtnet_txq *); 148 static int vtnet_txq_below_threshold(struct vtnet_txq *); 149 static int vtnet_txq_notify(struct vtnet_txq *); 150 static void vtnet_txq_free_mbufs(struct vtnet_txq *); 151 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, 152 int *, int *, int *); 153 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, 154 int, struct virtio_net_hdr *); 155 static struct mbuf * 156 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, 157 struct virtio_net_hdr *); 158 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, 159 struct vtnet_tx_header *); 160 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int); 161 #ifdef VTNET_LEGACY_TX 162 static void vtnet_start_locked(struct vtnet_txq *, if_t); 163 static void vtnet_start(if_t); 164 #else 165 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); 166 static int vtnet_txq_mq_start(if_t, struct mbuf *); 167 static void vtnet_txq_tq_deferred(void *, int); 168 #endif 169 static void vtnet_txq_start(struct vtnet_txq *); 170 static void vtnet_txq_tq_intr(void *, int); 171 static int vtnet_txq_eof(struct vtnet_txq *); 172 static void vtnet_tx_vq_intr(void *); 173 static void vtnet_tx_start_all(struct vtnet_softc *); 174 175 #ifndef VTNET_LEGACY_TX 176 static void vtnet_qflush(if_t); 177 #endif 178 179 static int vtnet_watchdog(struct vtnet_txq *); 180 static void vtnet_accum_stats(struct vtnet_softc *, 181 struct vtnet_rxq_stats *, struct vtnet_txq_stats *); 182 static void vtnet_tick(void *); 183 184 static void vtnet_start_taskqueues(struct vtnet_softc *); 185 static void vtnet_free_taskqueues(struct vtnet_softc *); 186 static void vtnet_drain_taskqueues(struct vtnet_softc *); 187 188 static void vtnet_drain_rxtx_queues(struct vtnet_softc *); 189 static void vtnet_stop_rendezvous(struct vtnet_softc *); 190 static void vtnet_stop(struct vtnet_softc *); 191 static int vtnet_virtio_reinit(struct vtnet_softc *); 192 static void vtnet_init_rx_filters(struct vtnet_softc *); 193 static int vtnet_init_rx_queues(struct vtnet_softc *); 194 static int vtnet_init_tx_queues(struct vtnet_softc *); 195 static int vtnet_init_rxtx_queues(struct vtnet_softc *); 196 static void vtnet_set_active_vq_pairs(struct vtnet_softc *); 197 static void vtnet_update_rx_offloads(struct vtnet_softc *); 198 static int vtnet_reinit(struct vtnet_softc *); 199 static void vtnet_init_locked(struct vtnet_softc *, int); 200 static void vtnet_init(void *); 201 202 static void vtnet_free_ctrl_vq(struct vtnet_softc *); 203 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 204 struct sglist *, int, int); 205 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 206 static int vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t); 207 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); 208 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool); 209 static int vtnet_set_promisc(struct vtnet_softc *, bool); 210 static int vtnet_set_allmulti(struct vtnet_softc *, bool); 211 static void vtnet_rx_filter(struct vtnet_softc *); 212 static void vtnet_rx_filter_mac(struct vtnet_softc *); 213 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 214 static void vtnet_rx_filter_vlan(struct vtnet_softc *); 215 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 216 static void vtnet_register_vlan(void *, if_t, uint16_t); 217 static void vtnet_unregister_vlan(void *, if_t, uint16_t); 218 219 static void vtnet_update_speed_duplex(struct vtnet_softc *); 220 static int vtnet_is_link_up(struct vtnet_softc *); 221 static void vtnet_update_link_status(struct vtnet_softc *); 222 static int vtnet_ifmedia_upd(if_t); 223 static void vtnet_ifmedia_sts(if_t, struct ifmediareq *); 224 static void vtnet_get_macaddr(struct vtnet_softc *); 225 static void vtnet_set_macaddr(struct vtnet_softc *); 226 static void vtnet_attached_set_macaddr(struct vtnet_softc *); 227 static void vtnet_vlan_tag_remove(struct mbuf *); 228 static void vtnet_set_rx_process_limit(struct vtnet_softc *); 229 230 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, 231 struct sysctl_oid_list *, struct vtnet_rxq *); 232 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, 233 struct sysctl_oid_list *, struct vtnet_txq *); 234 static void vtnet_setup_queue_sysctl(struct vtnet_softc *); 235 static void vtnet_load_tunables(struct vtnet_softc *); 236 static void vtnet_setup_sysctl(struct vtnet_softc *); 237 238 static int vtnet_rxq_enable_intr(struct vtnet_rxq *); 239 static void vtnet_rxq_disable_intr(struct vtnet_rxq *); 240 static int vtnet_txq_enable_intr(struct vtnet_txq *); 241 static void vtnet_txq_disable_intr(struct vtnet_txq *); 242 static void vtnet_enable_rx_interrupts(struct vtnet_softc *); 243 static void vtnet_enable_tx_interrupts(struct vtnet_softc *); 244 static void vtnet_enable_interrupts(struct vtnet_softc *); 245 static void vtnet_disable_rx_interrupts(struct vtnet_softc *); 246 static void vtnet_disable_tx_interrupts(struct vtnet_softc *); 247 static void vtnet_disable_interrupts(struct vtnet_softc *); 248 249 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); 250 251 DEBUGNET_DEFINE(vtnet); 252 253 #define vtnet_htog16(_sc, _val) virtio_htog16(vtnet_modern(_sc), _val) 254 #define vtnet_htog32(_sc, _val) virtio_htog32(vtnet_modern(_sc), _val) 255 #define vtnet_htog64(_sc, _val) virtio_htog64(vtnet_modern(_sc), _val) 256 #define vtnet_gtoh16(_sc, _val) virtio_gtoh16(vtnet_modern(_sc), _val) 257 #define vtnet_gtoh32(_sc, _val) virtio_gtoh32(vtnet_modern(_sc), _val) 258 #define vtnet_gtoh64(_sc, _val) virtio_gtoh64(vtnet_modern(_sc), _val) 259 260 /* Tunables. */ 261 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 262 "VirtIO Net driver parameters"); 263 264 static int vtnet_csum_disable = 0; 265 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN, 266 &vtnet_csum_disable, 0, "Disables receive and send checksum offload"); 267 268 static int vtnet_fixup_needs_csum = 0; 269 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN, 270 &vtnet_fixup_needs_csum, 0, 271 "Calculate valid checksum for NEEDS_CSUM packets"); 272 273 static int vtnet_tso_disable = 0; 274 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, 275 &vtnet_tso_disable, 0, "Disables TSO"); 276 277 static int vtnet_lro_disable = 0; 278 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, 279 &vtnet_lro_disable, 0, "Disables hardware LRO"); 280 281 static int vtnet_mq_disable = 0; 282 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, 283 &vtnet_mq_disable, 0, "Disables multiqueue support"); 284 285 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS; 286 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN, 287 &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs"); 288 289 static int vtnet_tso_maxlen = IP_MAXPACKET; 290 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN, 291 &vtnet_tso_maxlen, 0, "TSO burst limit"); 292 293 static int vtnet_rx_process_limit = 1024; 294 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, 295 &vtnet_rx_process_limit, 0, 296 "Number of RX segments processed in one pass"); 297 298 static int vtnet_lro_entry_count = 128; 299 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN, 300 &vtnet_lro_entry_count, 0, "Software LRO entry count"); 301 302 /* Enable sorted LRO, and the depth of the mbuf queue. */ 303 static int vtnet_lro_mbufq_depth = 0; 304 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN, 305 &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue"); 306 307 static uma_zone_t vtnet_tx_header_zone; 308 309 static struct virtio_feature_desc vtnet_feature_desc[] = { 310 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 311 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 312 { VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, "CtrlRxOffloads" }, 313 { VIRTIO_NET_F_MAC, "MAC" }, 314 { VIRTIO_NET_F_GSO, "TxGSO" }, 315 { VIRTIO_NET_F_GUEST_TSO4, "RxLROv4" }, 316 { VIRTIO_NET_F_GUEST_TSO6, "RxLROv6" }, 317 { VIRTIO_NET_F_GUEST_ECN, "RxLROECN" }, 318 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 319 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 320 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 321 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 322 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 323 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 324 { VIRTIO_NET_F_STATUS, "Status" }, 325 { VIRTIO_NET_F_CTRL_VQ, "CtrlVq" }, 326 { VIRTIO_NET_F_CTRL_RX, "CtrlRxMode" }, 327 { VIRTIO_NET_F_CTRL_VLAN, "CtrlVLANFilter" }, 328 { VIRTIO_NET_F_CTRL_RX_EXTRA, "CtrlRxModeExtra" }, 329 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 330 { VIRTIO_NET_F_MQ, "Multiqueue" }, 331 { VIRTIO_NET_F_CTRL_MAC_ADDR, "CtrlMacAddr" }, 332 { VIRTIO_NET_F_SPEED_DUPLEX, "SpeedDuplex" }, 333 334 { 0, NULL } 335 }; 336 337 static device_method_t vtnet_methods[] = { 338 /* Device methods. */ 339 DEVMETHOD(device_probe, vtnet_probe), 340 DEVMETHOD(device_attach, vtnet_attach), 341 DEVMETHOD(device_detach, vtnet_detach), 342 DEVMETHOD(device_suspend, vtnet_suspend), 343 DEVMETHOD(device_resume, vtnet_resume), 344 DEVMETHOD(device_shutdown, vtnet_shutdown), 345 346 /* VirtIO methods. */ 347 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), 348 DEVMETHOD(virtio_config_change, vtnet_config_change), 349 350 DEVMETHOD_END 351 }; 352 353 #ifdef DEV_NETMAP 354 #include <dev/netmap/if_vtnet_netmap.h> 355 #endif 356 357 static driver_t vtnet_driver = { 358 .name = "vtnet", 359 .methods = vtnet_methods, 360 .size = sizeof(struct vtnet_softc) 361 }; 362 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL); 363 MODULE_VERSION(vtnet, 1); 364 MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 365 #ifdef DEV_NETMAP 366 MODULE_DEPEND(vtnet, netmap, 1, 1, 1); 367 #endif 368 369 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter"); 370 371 static int 372 vtnet_modevent(module_t mod __unused, int type, void *unused __unused) 373 { 374 int error = 0; 375 static int loaded = 0; 376 377 switch (type) { 378 case MOD_LOAD: 379 if (loaded++ == 0) { 380 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", 381 sizeof(struct vtnet_tx_header), 382 NULL, NULL, NULL, NULL, 0, 0); 383 #ifdef DEBUGNET 384 /* 385 * We need to allocate from this zone in the transmit path, so ensure 386 * that we have at least one item per header available. 387 * XXX add a separate zone like we do for mbufs? otherwise we may alloc 388 * buckets 389 */ 390 uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 391 uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 392 #endif 393 } 394 break; 395 case MOD_QUIESCE: 396 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) 397 error = EBUSY; 398 break; 399 case MOD_UNLOAD: 400 if (--loaded == 0) { 401 uma_zdestroy(vtnet_tx_header_zone); 402 vtnet_tx_header_zone = NULL; 403 } 404 break; 405 case MOD_SHUTDOWN: 406 break; 407 default: 408 error = EOPNOTSUPP; 409 break; 410 } 411 412 return (error); 413 } 414 415 static int 416 vtnet_probe(device_t dev) 417 { 418 return (VIRTIO_SIMPLE_PROBE(dev, vtnet)); 419 } 420 421 static int 422 vtnet_attach(device_t dev) 423 { 424 struct vtnet_softc *sc; 425 int error; 426 427 sc = device_get_softc(dev); 428 sc->vtnet_dev = dev; 429 virtio_set_feature_desc(dev, vtnet_feature_desc); 430 431 VTNET_CORE_LOCK_INIT(sc); 432 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); 433 vtnet_load_tunables(sc); 434 435 error = vtnet_alloc_interface(sc); 436 if (error) { 437 device_printf(dev, "cannot allocate interface\n"); 438 goto fail; 439 } 440 441 vtnet_setup_sysctl(sc); 442 443 error = vtnet_setup_features(sc); 444 if (error) { 445 device_printf(dev, "cannot setup features\n"); 446 goto fail; 447 } 448 449 error = vtnet_alloc_rx_filters(sc); 450 if (error) { 451 device_printf(dev, "cannot allocate Rx filters\n"); 452 goto fail; 453 } 454 455 error = vtnet_alloc_rxtx_queues(sc); 456 if (error) { 457 device_printf(dev, "cannot allocate queues\n"); 458 goto fail; 459 } 460 461 error = vtnet_alloc_virtqueues(sc); 462 if (error) { 463 device_printf(dev, "cannot allocate virtqueues\n"); 464 goto fail; 465 } 466 467 error = vtnet_setup_interface(sc); 468 if (error) { 469 device_printf(dev, "cannot setup interface\n"); 470 goto fail; 471 } 472 473 error = virtio_setup_intr(dev, INTR_TYPE_NET); 474 if (error) { 475 device_printf(dev, "cannot setup interrupts\n"); 476 ether_ifdetach(sc->vtnet_ifp); 477 goto fail; 478 } 479 480 #ifdef DEV_NETMAP 481 vtnet_netmap_attach(sc); 482 #endif 483 vtnet_start_taskqueues(sc); 484 485 fail: 486 if (error) 487 vtnet_detach(dev); 488 489 return (error); 490 } 491 492 static int 493 vtnet_detach(device_t dev) 494 { 495 struct vtnet_softc *sc; 496 if_t ifp; 497 498 sc = device_get_softc(dev); 499 ifp = sc->vtnet_ifp; 500 501 if (device_is_attached(dev)) { 502 VTNET_CORE_LOCK(sc); 503 vtnet_stop(sc); 504 VTNET_CORE_UNLOCK(sc); 505 506 callout_drain(&sc->vtnet_tick_ch); 507 vtnet_drain_taskqueues(sc); 508 509 ether_ifdetach(ifp); 510 } 511 512 #ifdef DEV_NETMAP 513 netmap_detach(ifp); 514 #endif 515 516 if (sc->vtnet_pfil != NULL) { 517 pfil_head_unregister(sc->vtnet_pfil); 518 sc->vtnet_pfil = NULL; 519 } 520 521 vtnet_free_taskqueues(sc); 522 523 if (sc->vtnet_vlan_attach != NULL) { 524 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 525 sc->vtnet_vlan_attach = NULL; 526 } 527 if (sc->vtnet_vlan_detach != NULL) { 528 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach); 529 sc->vtnet_vlan_detach = NULL; 530 } 531 532 ifmedia_removeall(&sc->vtnet_media); 533 534 if (ifp != NULL) { 535 if_free(ifp); 536 sc->vtnet_ifp = NULL; 537 } 538 539 vtnet_free_rxtx_queues(sc); 540 vtnet_free_rx_filters(sc); 541 542 if (sc->vtnet_ctrl_vq != NULL) 543 vtnet_free_ctrl_vq(sc); 544 545 VTNET_CORE_LOCK_DESTROY(sc); 546 547 return (0); 548 } 549 550 static int 551 vtnet_suspend(device_t dev) 552 { 553 struct vtnet_softc *sc; 554 555 sc = device_get_softc(dev); 556 557 VTNET_CORE_LOCK(sc); 558 vtnet_stop(sc); 559 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 560 VTNET_CORE_UNLOCK(sc); 561 562 return (0); 563 } 564 565 static int 566 vtnet_resume(device_t dev) 567 { 568 struct vtnet_softc *sc; 569 if_t ifp; 570 571 sc = device_get_softc(dev); 572 ifp = sc->vtnet_ifp; 573 574 VTNET_CORE_LOCK(sc); 575 if (if_getflags(ifp) & IFF_UP) 576 vtnet_init_locked(sc, 0); 577 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 578 VTNET_CORE_UNLOCK(sc); 579 580 return (0); 581 } 582 583 static int 584 vtnet_shutdown(device_t dev) 585 { 586 /* 587 * Suspend already does all of what we need to 588 * do here; we just never expect to be resumed. 589 */ 590 return (vtnet_suspend(dev)); 591 } 592 593 static int 594 vtnet_attach_completed(device_t dev) 595 { 596 struct vtnet_softc *sc; 597 598 sc = device_get_softc(dev); 599 600 VTNET_CORE_LOCK(sc); 601 vtnet_attached_set_macaddr(sc); 602 VTNET_CORE_UNLOCK(sc); 603 604 return (0); 605 } 606 607 static int 608 vtnet_config_change(device_t dev) 609 { 610 struct vtnet_softc *sc; 611 612 sc = device_get_softc(dev); 613 614 VTNET_CORE_LOCK(sc); 615 vtnet_update_link_status(sc); 616 if (sc->vtnet_link_active != 0) 617 vtnet_tx_start_all(sc); 618 VTNET_CORE_UNLOCK(sc); 619 620 return (0); 621 } 622 623 static int 624 vtnet_negotiate_features(struct vtnet_softc *sc) 625 { 626 device_t dev; 627 uint64_t features, negotiated_features; 628 int no_csum; 629 630 dev = sc->vtnet_dev; 631 features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES : 632 VTNET_LEGACY_FEATURES; 633 634 /* 635 * TSO and LRO are only available when their corresponding checksum 636 * offload feature is also negotiated. 637 */ 638 no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable); 639 if (no_csum) 640 features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM); 641 if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) 642 features &= ~VTNET_TSO_FEATURES; 643 if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) 644 features &= ~VTNET_LRO_FEATURES; 645 646 #ifndef VTNET_LEGACY_TX 647 if (vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 648 features &= ~VIRTIO_NET_F_MQ; 649 #else 650 features &= ~VIRTIO_NET_F_MQ; 651 #endif 652 653 negotiated_features = virtio_negotiate_features(dev, features); 654 655 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) { 656 uint16_t mtu; 657 658 mtu = virtio_read_dev_config_2(dev, 659 offsetof(struct virtio_net_config, mtu)); 660 if (mtu < VTNET_MIN_MTU /* || mtu > VTNET_MAX_MTU */) { 661 device_printf(dev, "Invalid MTU value: %d. " 662 "MTU feature disabled.\n", mtu); 663 features &= ~VIRTIO_NET_F_MTU; 664 negotiated_features = 665 virtio_negotiate_features(dev, features); 666 } 667 } 668 669 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) { 670 uint16_t npairs; 671 672 npairs = virtio_read_dev_config_2(dev, 673 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 674 if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 675 npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) { 676 device_printf(dev, "Invalid max_virtqueue_pairs value: " 677 "%d. Multiqueue feature disabled.\n", npairs); 678 features &= ~VIRTIO_NET_F_MQ; 679 negotiated_features = 680 virtio_negotiate_features(dev, features); 681 } 682 } 683 684 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 685 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 686 /* 687 * LRO without mergeable buffers requires special care. This 688 * is not ideal because every receive buffer must be large 689 * enough to hold the maximum TCP packet, the Ethernet header, 690 * and the header. This requires up to 34 descriptors with 691 * MCLBYTES clusters. If we do not have indirect descriptors, 692 * LRO is disabled since the virtqueue will not contain very 693 * many receive buffers. 694 */ 695 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 696 device_printf(dev, 697 "Host LRO disabled since both mergeable buffers " 698 "and indirect descriptors were not negotiated\n"); 699 features &= ~VTNET_LRO_FEATURES; 700 negotiated_features = 701 virtio_negotiate_features(dev, features); 702 } else 703 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 704 } 705 706 sc->vtnet_features = negotiated_features; 707 sc->vtnet_negotiated_features = negotiated_features; 708 709 return (virtio_finalize_features(dev)); 710 } 711 712 static int 713 vtnet_setup_features(struct vtnet_softc *sc) 714 { 715 device_t dev; 716 int error; 717 718 dev = sc->vtnet_dev; 719 720 error = vtnet_negotiate_features(sc); 721 if (error) 722 return (error); 723 724 if (virtio_with_feature(dev, VIRTIO_F_VERSION_1)) 725 sc->vtnet_flags |= VTNET_FLAG_MODERN; 726 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 727 sc->vtnet_flags |= VTNET_FLAG_INDIRECT; 728 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 729 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 730 731 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 732 /* This feature should always be negotiated. */ 733 sc->vtnet_flags |= VTNET_FLAG_MAC; 734 } 735 736 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) { 737 sc->vtnet_max_mtu = virtio_read_dev_config_2(dev, 738 offsetof(struct virtio_net_config, mtu)); 739 } else 740 sc->vtnet_max_mtu = VTNET_MAX_MTU; 741 742 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 743 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 744 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 745 } else if (vtnet_modern(sc)) { 746 /* This is identical to the mergeable header. */ 747 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1); 748 } else 749 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 750 751 if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 752 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE; 753 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 754 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG; 755 else 756 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE; 757 758 /* 759 * Favor "hardware" LRO if negotiated, but support software LRO as 760 * a fallback; there is usually little benefit (or worse) with both. 761 */ 762 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 && 763 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0) 764 sc->vtnet_flags |= VTNET_FLAG_SW_LRO; 765 766 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 767 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 768 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 769 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX; 770 else 771 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN; 772 773 sc->vtnet_req_vq_pairs = 1; 774 sc->vtnet_max_vq_pairs = 1; 775 776 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 777 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 778 779 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 780 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 781 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 782 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 783 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 784 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 785 786 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) { 787 sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev, 788 offsetof(struct virtio_net_config, 789 max_virtqueue_pairs)); 790 } 791 } 792 793 if (sc->vtnet_max_vq_pairs > 1) { 794 int req; 795 796 /* 797 * Limit the maximum number of requested queue pairs to the 798 * number of CPUs and the configured maximum. 799 */ 800 req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 801 if (req < 0) 802 req = 1; 803 if (req == 0) 804 req = mp_ncpus; 805 if (req > sc->vtnet_max_vq_pairs) 806 req = sc->vtnet_max_vq_pairs; 807 if (req > mp_ncpus) 808 req = mp_ncpus; 809 if (req > 1) { 810 sc->vtnet_req_vq_pairs = req; 811 sc->vtnet_flags |= VTNET_FLAG_MQ; 812 } 813 } 814 815 return (0); 816 } 817 818 static int 819 vtnet_init_rxq(struct vtnet_softc *sc, int id) 820 { 821 struct vtnet_rxq *rxq; 822 823 rxq = &sc->vtnet_rxqs[id]; 824 825 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 826 device_get_nameunit(sc->vtnet_dev), id); 827 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 828 829 rxq->vtnrx_sc = sc; 830 rxq->vtnrx_id = id; 831 832 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 833 if (rxq->vtnrx_sg == NULL) 834 return (ENOMEM); 835 836 #if defined(INET) || defined(INET6) 837 if (vtnet_software_lro(sc)) { 838 if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp, 839 sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0) 840 return (ENOMEM); 841 } 842 #endif 843 844 NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 845 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 846 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 847 848 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 849 } 850 851 static int 852 vtnet_init_txq(struct vtnet_softc *sc, int id) 853 { 854 struct vtnet_txq *txq; 855 856 txq = &sc->vtnet_txqs[id]; 857 858 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 859 device_get_nameunit(sc->vtnet_dev), id); 860 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 861 862 txq->vtntx_sc = sc; 863 txq->vtntx_id = id; 864 865 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 866 if (txq->vtntx_sg == NULL) 867 return (ENOMEM); 868 869 #ifndef VTNET_LEGACY_TX 870 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 871 M_NOWAIT, &txq->vtntx_mtx); 872 if (txq->vtntx_br == NULL) 873 return (ENOMEM); 874 875 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 876 #endif 877 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 878 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 879 taskqueue_thread_enqueue, &txq->vtntx_tq); 880 if (txq->vtntx_tq == NULL) 881 return (ENOMEM); 882 883 return (0); 884 } 885 886 static int 887 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 888 { 889 int i, npairs, error; 890 891 npairs = sc->vtnet_max_vq_pairs; 892 893 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 894 M_NOWAIT | M_ZERO); 895 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 896 M_NOWAIT | M_ZERO); 897 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 898 return (ENOMEM); 899 900 for (i = 0; i < npairs; i++) { 901 error = vtnet_init_rxq(sc, i); 902 if (error) 903 return (error); 904 error = vtnet_init_txq(sc, i); 905 if (error) 906 return (error); 907 } 908 909 vtnet_set_rx_process_limit(sc); 910 vtnet_setup_queue_sysctl(sc); 911 912 return (0); 913 } 914 915 static void 916 vtnet_destroy_rxq(struct vtnet_rxq *rxq) 917 { 918 919 rxq->vtnrx_sc = NULL; 920 rxq->vtnrx_id = -1; 921 922 #if defined(INET) || defined(INET6) 923 tcp_lro_free(&rxq->vtnrx_lro); 924 #endif 925 926 if (rxq->vtnrx_sg != NULL) { 927 sglist_free(rxq->vtnrx_sg); 928 rxq->vtnrx_sg = NULL; 929 } 930 931 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 932 mtx_destroy(&rxq->vtnrx_mtx); 933 } 934 935 static void 936 vtnet_destroy_txq(struct vtnet_txq *txq) 937 { 938 939 txq->vtntx_sc = NULL; 940 txq->vtntx_id = -1; 941 942 if (txq->vtntx_sg != NULL) { 943 sglist_free(txq->vtntx_sg); 944 txq->vtntx_sg = NULL; 945 } 946 947 #ifndef VTNET_LEGACY_TX 948 if (txq->vtntx_br != NULL) { 949 buf_ring_free(txq->vtntx_br, M_DEVBUF); 950 txq->vtntx_br = NULL; 951 } 952 #endif 953 954 if (mtx_initialized(&txq->vtntx_mtx) != 0) 955 mtx_destroy(&txq->vtntx_mtx); 956 } 957 958 static void 959 vtnet_free_rxtx_queues(struct vtnet_softc *sc) 960 { 961 int i; 962 963 if (sc->vtnet_rxqs != NULL) { 964 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 965 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 966 free(sc->vtnet_rxqs, M_DEVBUF); 967 sc->vtnet_rxqs = NULL; 968 } 969 970 if (sc->vtnet_txqs != NULL) { 971 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 972 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 973 free(sc->vtnet_txqs, M_DEVBUF); 974 sc->vtnet_txqs = NULL; 975 } 976 } 977 978 static int 979 vtnet_alloc_rx_filters(struct vtnet_softc *sc) 980 { 981 982 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 983 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 984 M_DEVBUF, M_NOWAIT | M_ZERO); 985 if (sc->vtnet_mac_filter == NULL) 986 return (ENOMEM); 987 } 988 989 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 990 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 991 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 992 if (sc->vtnet_vlan_filter == NULL) 993 return (ENOMEM); 994 } 995 996 return (0); 997 } 998 999 static void 1000 vtnet_free_rx_filters(struct vtnet_softc *sc) 1001 { 1002 1003 if (sc->vtnet_mac_filter != NULL) { 1004 free(sc->vtnet_mac_filter, M_DEVBUF); 1005 sc->vtnet_mac_filter = NULL; 1006 } 1007 1008 if (sc->vtnet_vlan_filter != NULL) { 1009 free(sc->vtnet_vlan_filter, M_DEVBUF); 1010 sc->vtnet_vlan_filter = NULL; 1011 } 1012 } 1013 1014 static int 1015 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 1016 { 1017 device_t dev; 1018 struct vq_alloc_info *info; 1019 struct vtnet_rxq *rxq; 1020 struct vtnet_txq *txq; 1021 int i, idx, nvqs, error; 1022 1023 dev = sc->vtnet_dev; 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 error = virtio_alloc_virtqueues(dev, nvqs, info); 1062 free(info, M_TEMP); 1063 1064 return (error); 1065 } 1066 1067 static int 1068 vtnet_alloc_interface(struct vtnet_softc *sc) 1069 { 1070 device_t dev; 1071 if_t ifp; 1072 1073 dev = sc->vtnet_dev; 1074 1075 ifp = if_alloc(IFT_ETHER); 1076 if (ifp == NULL) 1077 return (ENOMEM); 1078 1079 sc->vtnet_ifp = ifp; 1080 if_setsoftc(ifp, sc); 1081 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 1082 1083 return (0); 1084 } 1085 1086 static int 1087 vtnet_setup_interface(struct vtnet_softc *sc) 1088 { 1089 device_t dev; 1090 struct pfil_head_args pa; 1091 if_t ifp; 1092 1093 dev = sc->vtnet_dev; 1094 ifp = sc->vtnet_ifp; 1095 1096 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); 1097 if_setbaudrate(ifp, IF_Gbps(10)); 1098 if_setinitfn(ifp, vtnet_init); 1099 if_setioctlfn(ifp, vtnet_ioctl); 1100 if_setgetcounterfn(ifp, vtnet_get_counter); 1101 #ifndef VTNET_LEGACY_TX 1102 if_settransmitfn(ifp, vtnet_txq_mq_start); 1103 if_setqflushfn(ifp, vtnet_qflush); 1104 #else 1105 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 1106 if_setstartfn(ifp, vtnet_start); 1107 if_setsendqlen(ifp, virtqueue_size(vq) - 1); 1108 if_setsendqready(ifp); 1109 #endif 1110 1111 vtnet_get_macaddr(sc); 1112 1113 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 1114 if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0); 1115 1116 ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts); 1117 ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL); 1118 ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO); 1119 1120 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 1121 int gso; 1122 1123 if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0); 1124 1125 gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO); 1126 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 1127 if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0); 1128 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 1129 if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0); 1130 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 1131 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 1132 1133 if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) { 1134 int tso_maxlen; 1135 1136 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0); 1137 1138 tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen", 1139 vtnet_tso_maxlen); 1140 if_sethwtsomax(ifp, tso_maxlen - 1141 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN)); 1142 if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1); 1143 if_sethwtsomaxsegsize(ifp, PAGE_SIZE); 1144 } 1145 } 1146 1147 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) { 1148 if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0); 1149 #ifdef notyet 1150 /* BMV: Rx checksums not distinguished between IPv4 and IPv6. */ 1151 if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0); 1152 #endif 1153 1154 if (vtnet_tunable_int(sc, "fixup_needs_csum", 1155 vtnet_fixup_needs_csum) != 0) 1156 sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM; 1157 1158 /* Support either "hardware" or software LRO. */ 1159 if_setcapabilitiesbit(ifp, IFCAP_LRO, 0); 1160 } 1161 1162 if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) { 1163 /* 1164 * VirtIO does not support VLAN tagging, but we can fake 1165 * it by inserting and removing the 802.1Q header during 1166 * transmit and receive. We are then able to do checksum 1167 * offloading of VLAN frames. 1168 */ 1169 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0); 1170 } 1171 1172 if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO) 1173 if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0); 1174 if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0); 1175 1176 /* 1177 * Capabilities after here are not enabled by default. 1178 */ 1179 if_setcapenable(ifp, if_getcapabilities(ifp)); 1180 1181 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 1182 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0); 1183 1184 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1185 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1186 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1187 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1188 } 1189 1190 ether_ifattach(ifp, sc->vtnet_hwaddr); 1191 1192 /* Tell the upper layer(s) we support long frames. */ 1193 if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); 1194 1195 DEBUGNET_SET(ifp, vtnet); 1196 1197 pa.pa_version = PFIL_VERSION; 1198 pa.pa_flags = PFIL_IN; 1199 pa.pa_type = PFIL_TYPE_ETHERNET; 1200 pa.pa_headname = if_name(ifp); 1201 sc->vtnet_pfil = pfil_head_register(&pa); 1202 1203 return (0); 1204 } 1205 1206 static int 1207 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu) 1208 { 1209 int framesz; 1210 1211 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 1212 return (MJUMPAGESIZE); 1213 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 1214 return (MCLBYTES); 1215 1216 /* 1217 * Try to scale the receive mbuf cluster size from the MTU. We 1218 * could also use the VQ size to influence the selected size, 1219 * but that would only matter for very small queues. 1220 */ 1221 if (vtnet_modern(sc)) { 1222 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1)); 1223 framesz = sizeof(struct virtio_net_hdr_v1); 1224 } else 1225 framesz = sizeof(struct vtnet_rx_header); 1226 framesz += sizeof(struct ether_vlan_header) + mtu; 1227 1228 if (framesz <= MCLBYTES) 1229 return (MCLBYTES); 1230 else if (framesz <= MJUMPAGESIZE) 1231 return (MJUMPAGESIZE); 1232 else if (framesz <= MJUM9BYTES) 1233 return (MJUM9BYTES); 1234 1235 /* Sane default; avoid 16KB clusters. */ 1236 return (MCLBYTES); 1237 } 1238 1239 static int 1240 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu) 1241 { 1242 if_t ifp; 1243 int clustersz; 1244 1245 ifp = sc->vtnet_ifp; 1246 VTNET_CORE_LOCK_ASSERT(sc); 1247 1248 if (if_getmtu(ifp) == mtu) 1249 return (0); 1250 else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu) 1251 return (EINVAL); 1252 1253 if_setmtu(ifp, mtu); 1254 clustersz = vtnet_rx_cluster_size(sc, mtu); 1255 1256 if (clustersz != sc->vtnet_rx_clustersz && 1257 if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 1258 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 1259 vtnet_init_locked(sc, 0); 1260 } 1261 1262 return (0); 1263 } 1264 1265 static int 1266 vtnet_ioctl_ifflags(struct vtnet_softc *sc) 1267 { 1268 if_t ifp; 1269 int drv_running; 1270 1271 ifp = sc->vtnet_ifp; 1272 drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0; 1273 1274 VTNET_CORE_LOCK_ASSERT(sc); 1275 1276 if ((if_getflags(ifp) & IFF_UP) == 0) { 1277 if (drv_running) 1278 vtnet_stop(sc); 1279 goto out; 1280 } 1281 1282 if (!drv_running) { 1283 vtnet_init_locked(sc, 0); 1284 goto out; 1285 } 1286 1287 if ((if_getflags(ifp) ^ sc->vtnet_if_flags) & 1288 (IFF_PROMISC | IFF_ALLMULTI)) { 1289 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1290 vtnet_rx_filter(sc); 1291 else { 1292 /* 1293 * We don't support filtering out multicast, so 1294 * ALLMULTI is always set. 1295 */ 1296 if_setflagbits(ifp, IFF_ALLMULTI, 0); 1297 if_setflagbits(ifp, IFF_PROMISC, 0); 1298 } 1299 } 1300 1301 out: 1302 sc->vtnet_if_flags = if_getflags(ifp); 1303 return (0); 1304 } 1305 1306 static int 1307 vtnet_ioctl_multi(struct vtnet_softc *sc) 1308 { 1309 if_t ifp; 1310 1311 ifp = sc->vtnet_ifp; 1312 1313 VTNET_CORE_LOCK_ASSERT(sc); 1314 1315 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX && 1316 if_getdrvflags(ifp) & IFF_DRV_RUNNING) 1317 vtnet_rx_filter_mac(sc); 1318 1319 return (0); 1320 } 1321 1322 static int 1323 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr) 1324 { 1325 if_t ifp; 1326 int mask, reinit, update; 1327 1328 ifp = sc->vtnet_ifp; 1329 mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp); 1330 reinit = update = 0; 1331 1332 VTNET_CORE_LOCK_ASSERT(sc); 1333 1334 if (mask & IFCAP_TXCSUM) 1335 if_togglecapenable(ifp, IFCAP_TXCSUM); 1336 if (mask & IFCAP_TXCSUM_IPV6) 1337 if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6); 1338 if (mask & IFCAP_TSO4) 1339 if_togglecapenable(ifp, IFCAP_TSO4); 1340 if (mask & IFCAP_TSO6) 1341 if_togglecapenable(ifp, IFCAP_TSO6); 1342 1343 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) { 1344 /* 1345 * These Rx features require the negotiated features to 1346 * be updated. Avoid a full reinit if possible. 1347 */ 1348 if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) 1349 update = 1; 1350 else 1351 reinit = 1; 1352 1353 /* BMV: Avoid needless renegotiation for just software LRO. */ 1354 if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) == 1355 IFCAP_LRO && vtnet_software_lro(sc)) 1356 reinit = update = 0; 1357 1358 if (mask & IFCAP_RXCSUM) 1359 if_togglecapenable(ifp, IFCAP_RXCSUM); 1360 if (mask & IFCAP_RXCSUM_IPV6) 1361 if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6); 1362 if (mask & IFCAP_LRO) 1363 if_togglecapenable(ifp, IFCAP_LRO); 1364 1365 /* 1366 * VirtIO does not distinguish between IPv4 and IPv6 checksums 1367 * so treat them as a pair. Guest TSO (LRO) requires receive 1368 * checksums. 1369 */ 1370 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 1371 if_setcapenablebit(ifp, IFCAP_RXCSUM, 0); 1372 #ifdef notyet 1373 if_setcapenablebit(ifp, IFCAP_RXCSUM_IPV6, 0); 1374 #endif 1375 } else 1376 if_setcapenablebit(ifp, 0, 1377 (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)); 1378 } 1379 1380 if (mask & IFCAP_VLAN_HWFILTER) { 1381 /* These Rx features require renegotiation. */ 1382 reinit = 1; 1383 1384 if (mask & IFCAP_VLAN_HWFILTER) 1385 if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER); 1386 } 1387 1388 if (mask & IFCAP_VLAN_HWTSO) 1389 if_togglecapenable(ifp, IFCAP_VLAN_HWTSO); 1390 if (mask & IFCAP_VLAN_HWTAGGING) 1391 if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING); 1392 1393 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 1394 if (reinit) { 1395 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 1396 vtnet_init_locked(sc, 0); 1397 } else if (update) 1398 vtnet_update_rx_offloads(sc); 1399 } 1400 1401 return (0); 1402 } 1403 1404 static int 1405 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data) 1406 { 1407 struct vtnet_softc *sc; 1408 struct ifreq *ifr; 1409 int error; 1410 1411 sc = if_getsoftc(ifp); 1412 ifr = (struct ifreq *) data; 1413 error = 0; 1414 1415 switch (cmd) { 1416 case SIOCSIFMTU: 1417 VTNET_CORE_LOCK(sc); 1418 error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu); 1419 VTNET_CORE_UNLOCK(sc); 1420 break; 1421 1422 case SIOCSIFFLAGS: 1423 VTNET_CORE_LOCK(sc); 1424 error = vtnet_ioctl_ifflags(sc); 1425 VTNET_CORE_UNLOCK(sc); 1426 break; 1427 1428 case SIOCADDMULTI: 1429 case SIOCDELMULTI: 1430 VTNET_CORE_LOCK(sc); 1431 error = vtnet_ioctl_multi(sc); 1432 VTNET_CORE_UNLOCK(sc); 1433 break; 1434 1435 case SIOCSIFMEDIA: 1436 case SIOCGIFMEDIA: 1437 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1438 break; 1439 1440 case SIOCSIFCAP: 1441 VTNET_CORE_LOCK(sc); 1442 error = vtnet_ioctl_ifcap(sc, ifr); 1443 VTNET_CORE_UNLOCK(sc); 1444 VLAN_CAPABILITIES(ifp); 1445 break; 1446 1447 default: 1448 error = ether_ioctl(ifp, cmd, data); 1449 break; 1450 } 1451 1452 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1453 1454 return (error); 1455 } 1456 1457 static int 1458 vtnet_rxq_populate(struct vtnet_rxq *rxq) 1459 { 1460 struct virtqueue *vq; 1461 int nbufs, error; 1462 1463 #ifdef DEV_NETMAP 1464 error = vtnet_netmap_rxq_populate(rxq); 1465 if (error >= 0) 1466 return (error); 1467 #endif /* DEV_NETMAP */ 1468 1469 vq = rxq->vtnrx_vq; 1470 error = ENOSPC; 1471 1472 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1473 error = vtnet_rxq_new_buf(rxq); 1474 if (error) 1475 break; 1476 } 1477 1478 if (nbufs > 0) { 1479 virtqueue_notify(vq); 1480 /* 1481 * EMSGSIZE signifies the virtqueue did not have enough 1482 * entries available to hold the last mbuf. This is not 1483 * an error. 1484 */ 1485 if (error == EMSGSIZE) 1486 error = 0; 1487 } 1488 1489 return (error); 1490 } 1491 1492 static void 1493 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1494 { 1495 struct virtqueue *vq; 1496 struct mbuf *m; 1497 int last; 1498 #ifdef DEV_NETMAP 1499 struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp), 1500 rxq->vtnrx_id, NR_RX); 1501 #else /* !DEV_NETMAP */ 1502 void *kring = NULL; 1503 #endif /* !DEV_NETMAP */ 1504 1505 vq = rxq->vtnrx_vq; 1506 last = 0; 1507 1508 while ((m = virtqueue_drain(vq, &last)) != NULL) { 1509 if (kring == NULL) 1510 m_freem(m); 1511 } 1512 1513 KASSERT(virtqueue_empty(vq), 1514 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1515 } 1516 1517 static struct mbuf * 1518 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1519 { 1520 struct mbuf *m_head, *m_tail, *m; 1521 int i, size; 1522 1523 m_head = NULL; 1524 size = sc->vtnet_rx_clustersz; 1525 1526 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1527 ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs)); 1528 1529 for (i = 0; i < nbufs; i++) { 1530 m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size); 1531 if (m == NULL) { 1532 sc->vtnet_stats.mbuf_alloc_failed++; 1533 m_freem(m_head); 1534 return (NULL); 1535 } 1536 1537 m->m_len = size; 1538 if (m_head != NULL) { 1539 m_tail->m_next = m; 1540 m_tail = m; 1541 } else 1542 m_head = m_tail = m; 1543 } 1544 1545 if (m_tailp != NULL) 1546 *m_tailp = m_tail; 1547 1548 return (m_head); 1549 } 1550 1551 /* 1552 * Slow path for when LRO without mergeable buffers is negotiated. 1553 */ 1554 static int 1555 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1556 int len0) 1557 { 1558 struct vtnet_softc *sc; 1559 struct mbuf *m, *m_prev, *m_new, *m_tail; 1560 int len, clustersz, nreplace, error; 1561 1562 sc = rxq->vtnrx_sc; 1563 clustersz = sc->vtnet_rx_clustersz; 1564 1565 m_prev = NULL; 1566 m_tail = NULL; 1567 nreplace = 0; 1568 1569 m = m0; 1570 len = len0; 1571 1572 /* 1573 * Since these mbuf chains are so large, avoid allocating a complete 1574 * replacement when the received frame did not consume the entire 1575 * chain. Unused mbufs are moved to the tail of the replacement mbuf. 1576 */ 1577 while (len > 0) { 1578 if (m == NULL) { 1579 sc->vtnet_stats.rx_frame_too_large++; 1580 return (EMSGSIZE); 1581 } 1582 1583 /* 1584 * Every mbuf should have the expected cluster size since that 1585 * is also used to allocate the replacements. 1586 */ 1587 KASSERT(m->m_len == clustersz, 1588 ("%s: mbuf size %d not expected cluster size %d", __func__, 1589 m->m_len, clustersz)); 1590 1591 m->m_len = MIN(m->m_len, len); 1592 len -= m->m_len; 1593 1594 m_prev = m; 1595 m = m->m_next; 1596 nreplace++; 1597 } 1598 1599 KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs, 1600 ("%s: invalid replacement mbuf count %d max %d", __func__, 1601 nreplace, sc->vtnet_rx_nmbufs)); 1602 1603 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1604 if (m_new == NULL) { 1605 m_prev->m_len = clustersz; 1606 return (ENOBUFS); 1607 } 1608 1609 /* 1610 * Move any unused mbufs from the received mbuf chain onto the 1611 * end of the replacement chain. 1612 */ 1613 if (m_prev->m_next != NULL) { 1614 m_tail->m_next = m_prev->m_next; 1615 m_prev->m_next = NULL; 1616 } 1617 1618 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1619 if (error) { 1620 /* 1621 * The replacement is suppose to be an copy of the one 1622 * dequeued so this is a very unexpected error. 1623 * 1624 * Restore the m0 chain to the original state if it was 1625 * modified so we can then discard it. 1626 */ 1627 if (m_tail->m_next != NULL) { 1628 m_prev->m_next = m_tail->m_next; 1629 m_tail->m_next = NULL; 1630 } 1631 m_prev->m_len = clustersz; 1632 sc->vtnet_stats.rx_enq_replacement_failed++; 1633 m_freem(m_new); 1634 } 1635 1636 return (error); 1637 } 1638 1639 static int 1640 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1641 { 1642 struct vtnet_softc *sc; 1643 struct mbuf *m_new; 1644 int error; 1645 1646 sc = rxq->vtnrx_sc; 1647 1648 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 1649 return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len)); 1650 1651 MPASS(m->m_next == NULL); 1652 if (m->m_len < len) 1653 return (EMSGSIZE); 1654 1655 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1656 if (m_new == NULL) 1657 return (ENOBUFS); 1658 1659 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1660 if (error) { 1661 sc->vtnet_stats.rx_enq_replacement_failed++; 1662 m_freem(m_new); 1663 } else 1664 m->m_len = len; 1665 1666 return (error); 1667 } 1668 1669 static int 1670 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1671 { 1672 struct vtnet_softc *sc; 1673 struct sglist *sg; 1674 int header_inlined, error; 1675 1676 sc = rxq->vtnrx_sc; 1677 sg = rxq->vtnrx_sg; 1678 1679 KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1680 ("%s: mbuf chain without LRO_NOMRG", __func__)); 1681 VTNET_RXQ_LOCK_ASSERT(rxq); 1682 1683 sglist_reset(sg); 1684 header_inlined = vtnet_modern(sc) || 1685 (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */ 1686 1687 if (header_inlined) 1688 error = sglist_append_mbuf(sg, m); 1689 else { 1690 struct vtnet_rx_header *rxhdr = 1691 mtod(m, struct vtnet_rx_header *); 1692 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1693 1694 /* Append the header and remaining mbuf data. */ 1695 error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1696 if (error) 1697 return (error); 1698 error = sglist_append(sg, &rxhdr[1], 1699 m->m_len - sizeof(struct vtnet_rx_header)); 1700 if (error) 1701 return (error); 1702 1703 if (m->m_next != NULL) 1704 error = sglist_append_mbuf(sg, m->m_next); 1705 } 1706 1707 if (error) 1708 return (error); 1709 1710 return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg)); 1711 } 1712 1713 static int 1714 vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1715 { 1716 struct vtnet_softc *sc; 1717 struct mbuf *m; 1718 int error; 1719 1720 sc = rxq->vtnrx_sc; 1721 1722 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1723 if (m == NULL) 1724 return (ENOBUFS); 1725 1726 error = vtnet_rxq_enqueue_buf(rxq, m); 1727 if (error) 1728 m_freem(m); 1729 1730 return (error); 1731 } 1732 1733 static int 1734 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, uint16_t etype, 1735 int hoff, struct virtio_net_hdr *hdr) 1736 { 1737 struct vtnet_softc *sc; 1738 int error; 1739 1740 sc = rxq->vtnrx_sc; 1741 1742 /* 1743 * NEEDS_CSUM corresponds to Linux's CHECKSUM_PARTIAL, but FreeBSD does 1744 * not have an analogous CSUM flag. The checksum has been validated, 1745 * but is incomplete (TCP/UDP pseudo header). 1746 * 1747 * The packet is likely from another VM on the same host that itself 1748 * performed checksum offloading so Tx/Rx is basically a memcpy and 1749 * the checksum has little value. 1750 * 1751 * Default to receiving the packet as-is for performance reasons, but 1752 * this can cause issues if the packet is to be forwarded because it 1753 * does not contain a valid checksum. This patch may be helpful: 1754 * https://reviews.freebsd.org/D6611. In the meantime, have the driver 1755 * compute the checksum if requested. 1756 * 1757 * BMV: Need to add an CSUM_PARTIAL flag? 1758 */ 1759 if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) { 1760 error = vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr); 1761 return (error); 1762 } 1763 1764 /* 1765 * Compute the checksum in the driver so the packet will contain a 1766 * valid checksum. The checksum is at csum_offset from csum_start. 1767 */ 1768 switch (etype) { 1769 #if defined(INET) || defined(INET6) 1770 case ETHERTYPE_IP: 1771 case ETHERTYPE_IPV6: { 1772 int csum_off, csum_end; 1773 uint16_t csum; 1774 1775 csum_off = hdr->csum_start + hdr->csum_offset; 1776 csum_end = csum_off + sizeof(uint16_t); 1777 1778 /* Assume checksum will be in the first mbuf. */ 1779 if (m->m_len < csum_end || m->m_pkthdr.len < csum_end) 1780 return (1); 1781 1782 /* 1783 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the 1784 * checksum and write it at the specified offset. We could 1785 * try to verify the packet: csum_start should probably 1786 * correspond to the start of the TCP/UDP header. 1787 * 1788 * BMV: Need to properly handle UDP with zero checksum. Is 1789 * the IPv4 header checksum implicitly validated? 1790 */ 1791 csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start); 1792 *(uint16_t *)(mtodo(m, csum_off)) = csum; 1793 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1794 m->m_pkthdr.csum_data = 0xFFFF; 1795 break; 1796 } 1797 #endif 1798 default: 1799 sc->vtnet_stats.rx_csum_bad_ethtype++; 1800 return (1); 1801 } 1802 1803 return (0); 1804 } 1805 1806 static int 1807 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m, 1808 uint16_t etype, int hoff, struct virtio_net_hdr *hdr __unused) 1809 { 1810 #if 0 1811 struct vtnet_softc *sc; 1812 #endif 1813 int protocol; 1814 1815 #if 0 1816 sc = rxq->vtnrx_sc; 1817 #endif 1818 1819 switch (etype) { 1820 #if defined(INET) 1821 case ETHERTYPE_IP: 1822 if (__predict_false(m->m_len < hoff + sizeof(struct ip))) 1823 protocol = IPPROTO_DONE; 1824 else { 1825 struct ip *ip = (struct ip *)(m->m_data + hoff); 1826 protocol = ip->ip_p; 1827 } 1828 break; 1829 #endif 1830 #if defined(INET6) 1831 case ETHERTYPE_IPV6: 1832 if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr)) 1833 || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0) 1834 protocol = IPPROTO_DONE; 1835 break; 1836 #endif 1837 default: 1838 protocol = IPPROTO_DONE; 1839 break; 1840 } 1841 1842 switch (protocol) { 1843 case IPPROTO_TCP: 1844 case IPPROTO_UDP: 1845 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1846 m->m_pkthdr.csum_data = 0xFFFF; 1847 break; 1848 default: 1849 /* 1850 * FreeBSD does not support checksum offloading of this 1851 * protocol. Let the stack re-verify the checksum later 1852 * if the protocol is supported. 1853 */ 1854 #if 0 1855 if_printf(sc->vtnet_ifp, 1856 "%s: checksum offload of unsupported protocol " 1857 "etype=%#x protocol=%d csum_start=%d csum_offset=%d\n", 1858 __func__, etype, protocol, hdr->csum_start, 1859 hdr->csum_offset); 1860 #endif 1861 break; 1862 } 1863 1864 return (0); 1865 } 1866 1867 static int 1868 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1869 struct virtio_net_hdr *hdr) 1870 { 1871 const struct ether_header *eh; 1872 int hoff; 1873 uint16_t etype; 1874 1875 eh = mtod(m, const struct ether_header *); 1876 etype = ntohs(eh->ether_type); 1877 if (etype == ETHERTYPE_VLAN) { 1878 /* TODO BMV: Handle QinQ. */ 1879 const struct ether_vlan_header *evh = 1880 mtod(m, const struct ether_vlan_header *); 1881 etype = ntohs(evh->evl_proto); 1882 hoff = sizeof(struct ether_vlan_header); 1883 } else 1884 hoff = sizeof(struct ether_header); 1885 1886 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1887 return (vtnet_rxq_csum_needs_csum(rxq, m, etype, hoff, hdr)); 1888 else /* VIRTIO_NET_HDR_F_DATA_VALID */ 1889 return (vtnet_rxq_csum_data_valid(rxq, m, etype, hoff, hdr)); 1890 } 1891 1892 static void 1893 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1894 { 1895 struct mbuf *m; 1896 1897 while (--nbufs > 0) { 1898 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1899 if (m == NULL) 1900 break; 1901 vtnet_rxq_discard_buf(rxq, m); 1902 } 1903 } 1904 1905 static void 1906 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1907 { 1908 int error __diagused; 1909 1910 /* 1911 * Requeue the discarded mbuf. This should always be successful 1912 * since it was just dequeued. 1913 */ 1914 error = vtnet_rxq_enqueue_buf(rxq, m); 1915 KASSERT(error == 0, 1916 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1917 } 1918 1919 static int 1920 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1921 { 1922 struct vtnet_softc *sc; 1923 struct virtqueue *vq; 1924 struct mbuf *m_tail; 1925 1926 sc = rxq->vtnrx_sc; 1927 vq = rxq->vtnrx_vq; 1928 m_tail = m_head; 1929 1930 while (--nbufs > 0) { 1931 struct mbuf *m; 1932 uint32_t len; 1933 1934 m = virtqueue_dequeue(vq, &len); 1935 if (m == NULL) { 1936 rxq->vtnrx_stats.vrxs_ierrors++; 1937 goto fail; 1938 } 1939 1940 if (vtnet_rxq_new_buf(rxq) != 0) { 1941 rxq->vtnrx_stats.vrxs_iqdrops++; 1942 vtnet_rxq_discard_buf(rxq, m); 1943 if (nbufs > 1) 1944 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1945 goto fail; 1946 } 1947 1948 if (m->m_len < len) 1949 len = m->m_len; 1950 1951 m->m_len = len; 1952 m->m_flags &= ~M_PKTHDR; 1953 1954 m_head->m_pkthdr.len += len; 1955 m_tail->m_next = m; 1956 m_tail = m; 1957 } 1958 1959 return (0); 1960 1961 fail: 1962 sc->vtnet_stats.rx_mergeable_failed++; 1963 m_freem(m_head); 1964 1965 return (1); 1966 } 1967 1968 #if defined(INET) || defined(INET6) 1969 static int 1970 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m) 1971 { 1972 struct lro_ctrl *lro; 1973 1974 lro = &rxq->vtnrx_lro; 1975 1976 if (lro->lro_mbuf_max != 0) { 1977 tcp_lro_queue_mbuf(lro, m); 1978 return (0); 1979 } 1980 1981 return (tcp_lro_rx(lro, m, 0)); 1982 } 1983 #endif 1984 1985 static void 1986 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 1987 struct virtio_net_hdr *hdr) 1988 { 1989 struct vtnet_softc *sc; 1990 if_t ifp; 1991 1992 sc = rxq->vtnrx_sc; 1993 ifp = sc->vtnet_ifp; 1994 1995 if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) { 1996 struct ether_header *eh = mtod(m, struct ether_header *); 1997 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 1998 vtnet_vlan_tag_remove(m); 1999 /* 2000 * With the 802.1Q header removed, update the 2001 * checksum starting location accordingly. 2002 */ 2003 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 2004 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 2005 } 2006 } 2007 2008 m->m_pkthdr.flowid = rxq->vtnrx_id; 2009 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 2010 2011 if (hdr->flags & 2012 (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) { 2013 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 2014 rxq->vtnrx_stats.vrxs_csum++; 2015 else 2016 rxq->vtnrx_stats.vrxs_csum_failed++; 2017 } 2018 2019 if (hdr->gso_size != 0) { 2020 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2021 case VIRTIO_NET_HDR_GSO_TCPV4: 2022 case VIRTIO_NET_HDR_GSO_TCPV6: 2023 m->m_pkthdr.lro_nsegs = 2024 howmany(m->m_pkthdr.len, hdr->gso_size); 2025 rxq->vtnrx_stats.vrxs_host_lro++; 2026 break; 2027 } 2028 } 2029 2030 rxq->vtnrx_stats.vrxs_ipackets++; 2031 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 2032 2033 #if defined(INET) || defined(INET6) 2034 if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) { 2035 if (vtnet_lro_rx(rxq, m) == 0) 2036 return; 2037 } 2038 #endif 2039 2040 if_input(ifp, m); 2041 } 2042 2043 static int 2044 vtnet_rxq_eof(struct vtnet_rxq *rxq) 2045 { 2046 struct virtio_net_hdr lhdr, *hdr; 2047 struct vtnet_softc *sc; 2048 if_t ifp; 2049 struct virtqueue *vq; 2050 int deq, count; 2051 2052 sc = rxq->vtnrx_sc; 2053 vq = rxq->vtnrx_vq; 2054 ifp = sc->vtnet_ifp; 2055 deq = 0; 2056 count = sc->vtnet_rx_process_limit; 2057 2058 VTNET_RXQ_LOCK_ASSERT(rxq); 2059 2060 while (count-- > 0) { 2061 struct mbuf *m; 2062 uint32_t len, nbufs, adjsz; 2063 2064 m = virtqueue_dequeue(vq, &len); 2065 if (m == NULL) 2066 break; 2067 deq++; 2068 2069 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 2070 rxq->vtnrx_stats.vrxs_ierrors++; 2071 vtnet_rxq_discard_buf(rxq, m); 2072 continue; 2073 } 2074 2075 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) { 2076 struct virtio_net_hdr_mrg_rxbuf *mhdr = 2077 mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 2078 kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED); 2079 nbufs = vtnet_htog16(sc, mhdr->num_buffers); 2080 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 2081 } else if (vtnet_modern(sc)) { 2082 nbufs = 1; /* num_buffers is always 1 */ 2083 adjsz = sizeof(struct virtio_net_hdr_v1); 2084 } else { 2085 nbufs = 1; 2086 adjsz = sizeof(struct vtnet_rx_header); 2087 /* 2088 * Account for our gap between the header and start of 2089 * data to keep the segments separated. 2090 */ 2091 len += VTNET_RX_HEADER_PAD; 2092 } 2093 2094 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 2095 rxq->vtnrx_stats.vrxs_iqdrops++; 2096 vtnet_rxq_discard_buf(rxq, m); 2097 if (nbufs > 1) 2098 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 2099 continue; 2100 } 2101 2102 m->m_pkthdr.len = len; 2103 m->m_pkthdr.rcvif = ifp; 2104 m->m_pkthdr.csum_flags = 0; 2105 2106 if (nbufs > 1) { 2107 /* Dequeue the rest of chain. */ 2108 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 2109 continue; 2110 } 2111 2112 kmsan_mark_mbuf(m, KMSAN_STATE_INITED); 2113 2114 /* 2115 * Save an endian swapped version of the header prior to it 2116 * being stripped. The header is always at the start of the 2117 * mbuf data. num_buffers was already saved (and not needed) 2118 * so use the standard header. 2119 */ 2120 hdr = mtod(m, struct virtio_net_hdr *); 2121 lhdr.flags = hdr->flags; 2122 lhdr.gso_type = hdr->gso_type; 2123 lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len); 2124 lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size); 2125 lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start); 2126 lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset); 2127 m_adj(m, adjsz); 2128 2129 if (PFIL_HOOKED_IN(sc->vtnet_pfil)) { 2130 pfil_return_t pfil; 2131 2132 pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL); 2133 switch (pfil) { 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 if (vtnet_software_lro(sc)) 2149 tcp_lro_flush_all(&rxq->vtnrx_lro); 2150 #endif 2151 virtqueue_notify(vq); 2152 } 2153 2154 return (count > 0 ? 0 : EAGAIN); 2155 } 2156 2157 static void 2158 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries) 2159 { 2160 struct vtnet_softc *sc; 2161 if_t ifp; 2162 u_int more; 2163 #ifdef DEV_NETMAP 2164 int nmirq; 2165 #endif /* DEV_NETMAP */ 2166 2167 sc = rxq->vtnrx_sc; 2168 ifp = sc->vtnet_ifp; 2169 2170 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 2171 /* 2172 * Ignore this interrupt. Either this is a spurious interrupt 2173 * or multiqueue without per-VQ MSIX so every queue needs to 2174 * be polled (a brain dead configuration we could try harder 2175 * to avoid). 2176 */ 2177 vtnet_rxq_disable_intr(rxq); 2178 return; 2179 } 2180 2181 VTNET_RXQ_LOCK(rxq); 2182 2183 #ifdef DEV_NETMAP 2184 /* 2185 * We call netmap_rx_irq() under lock to prevent concurrent calls. 2186 * This is not necessary to serialize the access to the RX vq, but 2187 * rather to avoid races that may happen if this interface is 2188 * attached to a VALE switch, which would cause received packets 2189 * to stall in the RX queue (nm_kr_tryget() could find the kring 2190 * busy when called from netmap_bwrap_intr_notify()). 2191 */ 2192 nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more); 2193 if (nmirq != NM_IRQ_PASS) { 2194 VTNET_RXQ_UNLOCK(rxq); 2195 if (nmirq == NM_IRQ_RESCHED) { 2196 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2197 } 2198 return; 2199 } 2200 #endif /* DEV_NETMAP */ 2201 2202 again: 2203 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) { 2204 VTNET_RXQ_UNLOCK(rxq); 2205 return; 2206 } 2207 2208 more = vtnet_rxq_eof(rxq); 2209 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 2210 if (!more) 2211 vtnet_rxq_disable_intr(rxq); 2212 /* 2213 * This is an occasional condition or race (when !more), 2214 * so retry a few times before scheduling the taskqueue. 2215 */ 2216 if (tries-- > 0) 2217 goto again; 2218 2219 rxq->vtnrx_stats.vrxs_rescheduled++; 2220 VTNET_RXQ_UNLOCK(rxq); 2221 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2222 } else 2223 VTNET_RXQ_UNLOCK(rxq); 2224 } 2225 2226 static void 2227 vtnet_rx_vq_intr(void *xrxq) 2228 { 2229 struct vtnet_rxq *rxq; 2230 2231 rxq = xrxq; 2232 vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES); 2233 } 2234 2235 static void 2236 vtnet_rxq_tq_intr(void *xrxq, int pending __unused) 2237 { 2238 struct vtnet_rxq *rxq; 2239 2240 rxq = xrxq; 2241 vtnet_rx_vq_process(rxq, 0); 2242 } 2243 2244 static int 2245 vtnet_txq_intr_threshold(struct vtnet_txq *txq) 2246 { 2247 struct vtnet_softc *sc; 2248 int threshold; 2249 2250 sc = txq->vtntx_sc; 2251 2252 /* 2253 * The Tx interrupt is disabled until the queue free count falls 2254 * below our threshold. Completed frames are drained from the Tx 2255 * virtqueue before transmitting new frames and in the watchdog 2256 * callout, so the frequency of Tx interrupts is greatly reduced, 2257 * at the cost of not freeing mbufs as quickly as they otherwise 2258 * would be. 2259 */ 2260 threshold = virtqueue_size(txq->vtntx_vq) / 4; 2261 2262 /* 2263 * Without indirect descriptors, leave enough room for the most 2264 * segments we handle. 2265 */ 2266 if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 && 2267 threshold < sc->vtnet_tx_nsegs) 2268 threshold = sc->vtnet_tx_nsegs; 2269 2270 return (threshold); 2271 } 2272 2273 static int 2274 vtnet_txq_below_threshold(struct vtnet_txq *txq) 2275 { 2276 struct virtqueue *vq; 2277 2278 vq = txq->vtntx_vq; 2279 2280 return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold); 2281 } 2282 2283 static int 2284 vtnet_txq_notify(struct vtnet_txq *txq) 2285 { 2286 struct virtqueue *vq; 2287 2288 vq = txq->vtntx_vq; 2289 2290 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2291 virtqueue_notify(vq); 2292 2293 if (vtnet_txq_enable_intr(txq) == 0) 2294 return (0); 2295 2296 /* 2297 * Drain frames that were completed since last checked. If this 2298 * causes the queue to go above the threshold, the caller should 2299 * continue transmitting. 2300 */ 2301 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) { 2302 virtqueue_disable_intr(vq); 2303 return (1); 2304 } 2305 2306 return (0); 2307 } 2308 2309 static void 2310 vtnet_txq_free_mbufs(struct vtnet_txq *txq) 2311 { 2312 struct virtqueue *vq; 2313 struct vtnet_tx_header *txhdr; 2314 int last; 2315 #ifdef DEV_NETMAP 2316 struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp), 2317 txq->vtntx_id, NR_TX); 2318 #else /* !DEV_NETMAP */ 2319 void *kring = NULL; 2320 #endif /* !DEV_NETMAP */ 2321 2322 vq = txq->vtntx_vq; 2323 last = 0; 2324 2325 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 2326 if (kring == NULL) { 2327 m_freem(txhdr->vth_mbuf); 2328 uma_zfree(vtnet_tx_header_zone, txhdr); 2329 } 2330 } 2331 2332 KASSERT(virtqueue_empty(vq), 2333 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 2334 } 2335 2336 /* 2337 * BMV: This can go away once we finally have offsets in the mbuf header. 2338 */ 2339 static int 2340 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype, 2341 int *proto, int *start) 2342 { 2343 struct vtnet_softc *sc; 2344 struct ether_vlan_header *evh; 2345 #if defined(INET) || defined(INET6) 2346 int offset; 2347 #endif 2348 2349 sc = txq->vtntx_sc; 2350 2351 evh = mtod(m, struct ether_vlan_header *); 2352 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 2353 /* BMV: We should handle nested VLAN tags too. */ 2354 *etype = ntohs(evh->evl_proto); 2355 #if defined(INET) || defined(INET6) 2356 offset = sizeof(struct ether_vlan_header); 2357 #endif 2358 } else { 2359 *etype = ntohs(evh->evl_encap_proto); 2360 #if defined(INET) || defined(INET6) 2361 offset = sizeof(struct ether_header); 2362 #endif 2363 } 2364 2365 switch (*etype) { 2366 #if defined(INET) 2367 case ETHERTYPE_IP: { 2368 struct ip *ip, iphdr; 2369 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 2370 m_copydata(m, offset, sizeof(struct ip), 2371 (caddr_t) &iphdr); 2372 ip = &iphdr; 2373 } else 2374 ip = (struct ip *)(m->m_data + offset); 2375 *proto = ip->ip_p; 2376 *start = offset + (ip->ip_hl << 2); 2377 break; 2378 } 2379 #endif 2380 #if defined(INET6) 2381 case ETHERTYPE_IPV6: 2382 *proto = -1; 2383 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 2384 /* Assert the network stack sent us a valid packet. */ 2385 KASSERT(*start > offset, 2386 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 2387 *start, offset, *proto)); 2388 break; 2389 #endif 2390 default: 2391 sc->vtnet_stats.tx_csum_unknown_ethtype++; 2392 return (EINVAL); 2393 } 2394 2395 return (0); 2396 } 2397 2398 static int 2399 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 2400 int offset, struct virtio_net_hdr *hdr) 2401 { 2402 static struct timeval lastecn; 2403 static int curecn; 2404 struct vtnet_softc *sc; 2405 struct tcphdr *tcp, tcphdr; 2406 2407 sc = txq->vtntx_sc; 2408 2409 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 2410 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 2411 tcp = &tcphdr; 2412 } else 2413 tcp = (struct tcphdr *)(m->m_data + offset); 2414 2415 hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2)); 2416 hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz); 2417 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 2418 VIRTIO_NET_HDR_GSO_TCPV6; 2419 2420 if (__predict_false(tcp->th_flags & TH_CWR)) { 2421 /* 2422 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In 2423 * FreeBSD, ECN support is not on a per-interface basis, 2424 * but globally via the net.inet.tcp.ecn.enable sysctl 2425 * knob. The default is off. 2426 */ 2427 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2428 if (ppsratecheck(&lastecn, &curecn, 1)) 2429 if_printf(sc->vtnet_ifp, 2430 "TSO with ECN not negotiated with host\n"); 2431 return (ENOTSUP); 2432 } 2433 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2434 } 2435 2436 txq->vtntx_stats.vtxs_tso++; 2437 2438 return (0); 2439 } 2440 2441 static struct mbuf * 2442 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2443 struct virtio_net_hdr *hdr) 2444 { 2445 struct vtnet_softc *sc; 2446 int flags, etype, csum_start, proto, error; 2447 2448 sc = txq->vtntx_sc; 2449 flags = m->m_pkthdr.csum_flags; 2450 2451 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2452 if (error) 2453 goto drop; 2454 2455 if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) { 2456 /* Sanity check the parsed mbuf matches the offload flags. */ 2457 if (__predict_false((flags & VTNET_CSUM_OFFLOAD && 2458 etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6 2459 && etype != ETHERTYPE_IPV6))) { 2460 sc->vtnet_stats.tx_csum_proto_mismatch++; 2461 goto drop; 2462 } 2463 2464 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2465 hdr->csum_start = vtnet_gtoh16(sc, csum_start); 2466 hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data); 2467 txq->vtntx_stats.vtxs_csum++; 2468 } 2469 2470 if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) { 2471 /* 2472 * Sanity check the parsed mbuf IP protocol is TCP, and 2473 * VirtIO TSO reqires the checksum offloading above. 2474 */ 2475 if (__predict_false(proto != IPPROTO_TCP)) { 2476 sc->vtnet_stats.tx_tso_not_tcp++; 2477 goto drop; 2478 } else if (__predict_false((hdr->flags & 2479 VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) { 2480 sc->vtnet_stats.tx_tso_without_csum++; 2481 goto drop; 2482 } 2483 2484 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2485 if (error) 2486 goto drop; 2487 } 2488 2489 return (m); 2490 2491 drop: 2492 m_freem(m); 2493 return (NULL); 2494 } 2495 2496 static int 2497 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2498 struct vtnet_tx_header *txhdr) 2499 { 2500 struct vtnet_softc *sc; 2501 struct virtqueue *vq; 2502 struct sglist *sg; 2503 struct mbuf *m; 2504 int error; 2505 2506 sc = txq->vtntx_sc; 2507 vq = txq->vtntx_vq; 2508 sg = txq->vtntx_sg; 2509 m = *m_head; 2510 2511 sglist_reset(sg); 2512 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2513 if (error != 0 || sg->sg_nseg != 1) { 2514 KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d", 2515 __func__, error, sg->sg_nseg)); 2516 goto fail; 2517 } 2518 2519 error = sglist_append_mbuf(sg, m); 2520 if (error) { 2521 m = m_defrag(m, M_NOWAIT); 2522 if (m == NULL) 2523 goto fail; 2524 2525 *m_head = m; 2526 sc->vtnet_stats.tx_defragged++; 2527 2528 error = sglist_append_mbuf(sg, m); 2529 if (error) 2530 goto fail; 2531 } 2532 2533 txhdr->vth_mbuf = m; 2534 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2535 2536 return (error); 2537 2538 fail: 2539 sc->vtnet_stats.tx_defrag_failed++; 2540 m_freem(*m_head); 2541 *m_head = NULL; 2542 2543 return (ENOBUFS); 2544 } 2545 2546 static int 2547 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags) 2548 { 2549 struct vtnet_tx_header *txhdr; 2550 struct virtio_net_hdr *hdr; 2551 struct mbuf *m; 2552 int error; 2553 2554 m = *m_head; 2555 M_ASSERTPKTHDR(m); 2556 2557 txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO); 2558 if (txhdr == NULL) { 2559 m_freem(m); 2560 *m_head = NULL; 2561 return (ENOMEM); 2562 } 2563 2564 /* 2565 * Always use the non-mergeable header, regardless if mergable headers 2566 * were negotiated, because for transmit num_buffers is always zero. 2567 * The vtnet_hdr_size is used to enqueue the right header size segment. 2568 */ 2569 hdr = &txhdr->vth_uhdr.hdr; 2570 2571 if (m->m_flags & M_VLANTAG) { 2572 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2573 if ((*m_head = m) == NULL) { 2574 error = ENOBUFS; 2575 goto fail; 2576 } 2577 m->m_flags &= ~M_VLANTAG; 2578 } 2579 2580 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2581 m = vtnet_txq_offload(txq, m, hdr); 2582 if ((*m_head = m) == NULL) { 2583 error = ENOBUFS; 2584 goto fail; 2585 } 2586 } 2587 2588 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2589 fail: 2590 if (error) 2591 uma_zfree(vtnet_tx_header_zone, txhdr); 2592 2593 return (error); 2594 } 2595 2596 #ifdef VTNET_LEGACY_TX 2597 2598 static void 2599 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp) 2600 { 2601 struct vtnet_softc *sc; 2602 struct virtqueue *vq; 2603 struct mbuf *m0; 2604 int tries, enq; 2605 2606 sc = txq->vtntx_sc; 2607 vq = txq->vtntx_vq; 2608 tries = 0; 2609 2610 VTNET_TXQ_LOCK_ASSERT(txq); 2611 2612 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 || 2613 sc->vtnet_link_active == 0) 2614 return; 2615 2616 vtnet_txq_eof(txq); 2617 2618 again: 2619 enq = 0; 2620 2621 while (!if_sendq_empty(ifp)) { 2622 if (virtqueue_full(vq)) 2623 break; 2624 2625 m0 = if_dequeue(ifp); 2626 if (m0 == NULL) 2627 break; 2628 2629 if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) { 2630 if (m0 != NULL) 2631 if_sendq_prepend(ifp, m0); 2632 break; 2633 } 2634 2635 enq++; 2636 ETHER_BPF_MTAP(ifp, m0); 2637 } 2638 2639 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2640 if (tries++ < VTNET_NOTIFY_RETRIES) 2641 goto again; 2642 2643 txq->vtntx_stats.vtxs_rescheduled++; 2644 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2645 } 2646 } 2647 2648 static void 2649 vtnet_start(if_t ifp) 2650 { 2651 struct vtnet_softc *sc; 2652 struct vtnet_txq *txq; 2653 2654 sc = if_getsoftc(ifp); 2655 txq = &sc->vtnet_txqs[0]; 2656 2657 VTNET_TXQ_LOCK(txq); 2658 vtnet_start_locked(txq, ifp); 2659 VTNET_TXQ_UNLOCK(txq); 2660 } 2661 2662 #else /* !VTNET_LEGACY_TX */ 2663 2664 static int 2665 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2666 { 2667 struct vtnet_softc *sc; 2668 struct virtqueue *vq; 2669 struct buf_ring *br; 2670 if_t ifp; 2671 int enq, tries, error; 2672 2673 sc = txq->vtntx_sc; 2674 vq = txq->vtntx_vq; 2675 br = txq->vtntx_br; 2676 ifp = sc->vtnet_ifp; 2677 tries = 0; 2678 error = 0; 2679 2680 VTNET_TXQ_LOCK_ASSERT(txq); 2681 2682 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 || 2683 sc->vtnet_link_active == 0) { 2684 if (m != NULL) 2685 error = drbr_enqueue(ifp, br, m); 2686 return (error); 2687 } 2688 2689 if (m != NULL) { 2690 error = drbr_enqueue(ifp, br, m); 2691 if (error) 2692 return (error); 2693 } 2694 2695 vtnet_txq_eof(txq); 2696 2697 again: 2698 enq = 0; 2699 2700 while ((m = drbr_peek(ifp, br)) != NULL) { 2701 if (virtqueue_full(vq)) { 2702 drbr_putback(ifp, br, m); 2703 break; 2704 } 2705 2706 if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) { 2707 if (m != NULL) 2708 drbr_putback(ifp, br, m); 2709 else 2710 drbr_advance(ifp, br); 2711 break; 2712 } 2713 drbr_advance(ifp, br); 2714 2715 enq++; 2716 ETHER_BPF_MTAP(ifp, m); 2717 } 2718 2719 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2720 if (tries++ < VTNET_NOTIFY_RETRIES) 2721 goto again; 2722 2723 txq->vtntx_stats.vtxs_rescheduled++; 2724 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2725 } 2726 2727 return (0); 2728 } 2729 2730 static int 2731 vtnet_txq_mq_start(if_t ifp, struct mbuf *m) 2732 { 2733 struct vtnet_softc *sc; 2734 struct vtnet_txq *txq; 2735 int i, npairs, error; 2736 2737 sc = if_getsoftc(ifp); 2738 npairs = sc->vtnet_act_vq_pairs; 2739 2740 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2741 i = m->m_pkthdr.flowid % npairs; 2742 else 2743 i = curcpu % npairs; 2744 2745 txq = &sc->vtnet_txqs[i]; 2746 2747 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2748 error = vtnet_txq_mq_start_locked(txq, m); 2749 VTNET_TXQ_UNLOCK(txq); 2750 } else { 2751 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2752 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2753 } 2754 2755 return (error); 2756 } 2757 2758 static void 2759 vtnet_txq_tq_deferred(void *xtxq, int pending __unused) 2760 { 2761 struct vtnet_softc *sc; 2762 struct vtnet_txq *txq; 2763 2764 txq = xtxq; 2765 sc = txq->vtntx_sc; 2766 2767 VTNET_TXQ_LOCK(txq); 2768 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2769 vtnet_txq_mq_start_locked(txq, NULL); 2770 VTNET_TXQ_UNLOCK(txq); 2771 } 2772 2773 #endif /* VTNET_LEGACY_TX */ 2774 2775 static void 2776 vtnet_txq_start(struct vtnet_txq *txq) 2777 { 2778 struct vtnet_softc *sc; 2779 if_t ifp; 2780 2781 sc = txq->vtntx_sc; 2782 ifp = sc->vtnet_ifp; 2783 2784 #ifdef VTNET_LEGACY_TX 2785 if (!if_sendq_empty(ifp)) 2786 vtnet_start_locked(txq, ifp); 2787 #else 2788 if (!drbr_empty(ifp, txq->vtntx_br)) 2789 vtnet_txq_mq_start_locked(txq, NULL); 2790 #endif 2791 } 2792 2793 static void 2794 vtnet_txq_tq_intr(void *xtxq, int pending __unused) 2795 { 2796 struct vtnet_softc *sc; 2797 struct vtnet_txq *txq; 2798 if_t ifp; 2799 2800 txq = xtxq; 2801 sc = txq->vtntx_sc; 2802 ifp = sc->vtnet_ifp; 2803 2804 VTNET_TXQ_LOCK(txq); 2805 2806 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) { 2807 VTNET_TXQ_UNLOCK(txq); 2808 return; 2809 } 2810 2811 vtnet_txq_eof(txq); 2812 vtnet_txq_start(txq); 2813 2814 VTNET_TXQ_UNLOCK(txq); 2815 } 2816 2817 static int 2818 vtnet_txq_eof(struct vtnet_txq *txq) 2819 { 2820 struct virtqueue *vq; 2821 struct vtnet_tx_header *txhdr; 2822 struct mbuf *m; 2823 int deq; 2824 2825 vq = txq->vtntx_vq; 2826 deq = 0; 2827 VTNET_TXQ_LOCK_ASSERT(txq); 2828 2829 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2830 m = txhdr->vth_mbuf; 2831 deq++; 2832 2833 txq->vtntx_stats.vtxs_opackets++; 2834 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2835 if (m->m_flags & M_MCAST) 2836 txq->vtntx_stats.vtxs_omcasts++; 2837 2838 m_freem(m); 2839 uma_zfree(vtnet_tx_header_zone, txhdr); 2840 } 2841 2842 if (virtqueue_empty(vq)) 2843 txq->vtntx_watchdog = 0; 2844 2845 return (deq); 2846 } 2847 2848 static void 2849 vtnet_tx_vq_intr(void *xtxq) 2850 { 2851 struct vtnet_softc *sc; 2852 struct vtnet_txq *txq; 2853 if_t ifp; 2854 2855 txq = xtxq; 2856 sc = txq->vtntx_sc; 2857 ifp = sc->vtnet_ifp; 2858 2859 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2860 /* 2861 * Ignore this interrupt. Either this is a spurious interrupt 2862 * or multiqueue without per-VQ MSIX so every queue needs to 2863 * be polled (a brain dead configuration we could try harder 2864 * to avoid). 2865 */ 2866 vtnet_txq_disable_intr(txq); 2867 return; 2868 } 2869 2870 #ifdef DEV_NETMAP 2871 if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS) 2872 return; 2873 #endif /* DEV_NETMAP */ 2874 2875 VTNET_TXQ_LOCK(txq); 2876 2877 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) { 2878 VTNET_TXQ_UNLOCK(txq); 2879 return; 2880 } 2881 2882 vtnet_txq_eof(txq); 2883 vtnet_txq_start(txq); 2884 2885 VTNET_TXQ_UNLOCK(txq); 2886 } 2887 2888 static void 2889 vtnet_tx_start_all(struct vtnet_softc *sc) 2890 { 2891 struct vtnet_txq *txq; 2892 int i; 2893 2894 VTNET_CORE_LOCK_ASSERT(sc); 2895 2896 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2897 txq = &sc->vtnet_txqs[i]; 2898 2899 VTNET_TXQ_LOCK(txq); 2900 vtnet_txq_start(txq); 2901 VTNET_TXQ_UNLOCK(txq); 2902 } 2903 } 2904 2905 #ifndef VTNET_LEGACY_TX 2906 static void 2907 vtnet_qflush(if_t ifp) 2908 { 2909 struct vtnet_softc *sc; 2910 struct vtnet_txq *txq; 2911 struct mbuf *m; 2912 int i; 2913 2914 sc = if_getsoftc(ifp); 2915 2916 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2917 txq = &sc->vtnet_txqs[i]; 2918 2919 VTNET_TXQ_LOCK(txq); 2920 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2921 m_freem(m); 2922 VTNET_TXQ_UNLOCK(txq); 2923 } 2924 2925 if_qflush(ifp); 2926 } 2927 #endif 2928 2929 static int 2930 vtnet_watchdog(struct vtnet_txq *txq) 2931 { 2932 if_t ifp; 2933 2934 ifp = txq->vtntx_sc->vtnet_ifp; 2935 2936 VTNET_TXQ_LOCK(txq); 2937 if (txq->vtntx_watchdog == 1) { 2938 /* 2939 * Only drain completed frames if the watchdog is about to 2940 * expire. If any frames were drained, there may be enough 2941 * free descriptors now available to transmit queued frames. 2942 * In that case, the timer will immediately be decremented 2943 * below, but the timeout is generous enough that should not 2944 * be a problem. 2945 */ 2946 if (vtnet_txq_eof(txq) != 0) 2947 vtnet_txq_start(txq); 2948 } 2949 2950 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 2951 VTNET_TXQ_UNLOCK(txq); 2952 return (0); 2953 } 2954 VTNET_TXQ_UNLOCK(txq); 2955 2956 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id); 2957 return (1); 2958 } 2959 2960 static void 2961 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc, 2962 struct vtnet_txq_stats *txacc) 2963 { 2964 2965 bzero(rxacc, sizeof(struct vtnet_rxq_stats)); 2966 bzero(txacc, sizeof(struct vtnet_txq_stats)); 2967 2968 for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) { 2969 struct vtnet_rxq_stats *rxst; 2970 struct vtnet_txq_stats *txst; 2971 2972 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 2973 rxacc->vrxs_ipackets += rxst->vrxs_ipackets; 2974 rxacc->vrxs_ibytes += rxst->vrxs_ibytes; 2975 rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops; 2976 rxacc->vrxs_csum += rxst->vrxs_csum; 2977 rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed; 2978 rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled; 2979 2980 txst = &sc->vtnet_txqs[i].vtntx_stats; 2981 txacc->vtxs_opackets += txst->vtxs_opackets; 2982 txacc->vtxs_obytes += txst->vtxs_obytes; 2983 txacc->vtxs_csum += txst->vtxs_csum; 2984 txacc->vtxs_tso += txst->vtxs_tso; 2985 txacc->vtxs_rescheduled += txst->vtxs_rescheduled; 2986 } 2987 } 2988 2989 static uint64_t 2990 vtnet_get_counter(if_t ifp, ift_counter cnt) 2991 { 2992 struct vtnet_softc *sc; 2993 struct vtnet_rxq_stats rxaccum; 2994 struct vtnet_txq_stats txaccum; 2995 2996 sc = if_getsoftc(ifp); 2997 vtnet_accum_stats(sc, &rxaccum, &txaccum); 2998 2999 switch (cnt) { 3000 case IFCOUNTER_IPACKETS: 3001 return (rxaccum.vrxs_ipackets); 3002 case IFCOUNTER_IQDROPS: 3003 return (rxaccum.vrxs_iqdrops); 3004 case IFCOUNTER_IERRORS: 3005 return (rxaccum.vrxs_ierrors); 3006 case IFCOUNTER_OPACKETS: 3007 return (txaccum.vtxs_opackets); 3008 #ifndef VTNET_LEGACY_TX 3009 case IFCOUNTER_OBYTES: 3010 return (txaccum.vtxs_obytes); 3011 case IFCOUNTER_OMCASTS: 3012 return (txaccum.vtxs_omcasts); 3013 #endif 3014 default: 3015 return (if_get_counter_default(ifp, cnt)); 3016 } 3017 } 3018 3019 static void 3020 vtnet_tick(void *xsc) 3021 { 3022 struct vtnet_softc *sc; 3023 if_t ifp; 3024 int i, timedout; 3025 3026 sc = xsc; 3027 ifp = sc->vtnet_ifp; 3028 timedout = 0; 3029 3030 VTNET_CORE_LOCK_ASSERT(sc); 3031 3032 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3033 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 3034 3035 if (timedout != 0) { 3036 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 3037 vtnet_init_locked(sc, 0); 3038 } else 3039 callout_schedule(&sc->vtnet_tick_ch, hz); 3040 } 3041 3042 static void 3043 vtnet_start_taskqueues(struct vtnet_softc *sc) 3044 { 3045 device_t dev; 3046 struct vtnet_rxq *rxq; 3047 struct vtnet_txq *txq; 3048 int i, error; 3049 3050 dev = sc->vtnet_dev; 3051 3052 /* 3053 * Errors here are very difficult to recover from - we cannot 3054 * easily fail because, if this is during boot, we will hang 3055 * when freeing any successfully started taskqueues because 3056 * the scheduler isn't up yet. 3057 * 3058 * Most drivers just ignore the return value - it only fails 3059 * with ENOMEM so an error is not likely. 3060 */ 3061 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) { 3062 rxq = &sc->vtnet_rxqs[i]; 3063 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 3064 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 3065 if (error) { 3066 device_printf(dev, "failed to start rx taskq %d\n", 3067 rxq->vtnrx_id); 3068 } 3069 3070 txq = &sc->vtnet_txqs[i]; 3071 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 3072 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 3073 if (error) { 3074 device_printf(dev, "failed to start tx taskq %d\n", 3075 txq->vtntx_id); 3076 } 3077 } 3078 } 3079 3080 static void 3081 vtnet_free_taskqueues(struct vtnet_softc *sc) 3082 { 3083 struct vtnet_rxq *rxq; 3084 struct vtnet_txq *txq; 3085 int i; 3086 3087 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3088 rxq = &sc->vtnet_rxqs[i]; 3089 if (rxq->vtnrx_tq != NULL) { 3090 taskqueue_free(rxq->vtnrx_tq); 3091 rxq->vtnrx_tq = NULL; 3092 } 3093 3094 txq = &sc->vtnet_txqs[i]; 3095 if (txq->vtntx_tq != NULL) { 3096 taskqueue_free(txq->vtntx_tq); 3097 txq->vtntx_tq = NULL; 3098 } 3099 } 3100 } 3101 3102 static void 3103 vtnet_drain_taskqueues(struct vtnet_softc *sc) 3104 { 3105 struct vtnet_rxq *rxq; 3106 struct vtnet_txq *txq; 3107 int i; 3108 3109 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3110 rxq = &sc->vtnet_rxqs[i]; 3111 if (rxq->vtnrx_tq != NULL) 3112 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 3113 3114 txq = &sc->vtnet_txqs[i]; 3115 if (txq->vtntx_tq != NULL) { 3116 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 3117 #ifndef VTNET_LEGACY_TX 3118 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 3119 #endif 3120 } 3121 } 3122 } 3123 3124 static void 3125 vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 3126 { 3127 struct vtnet_rxq *rxq; 3128 struct vtnet_txq *txq; 3129 int i; 3130 3131 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3132 rxq = &sc->vtnet_rxqs[i]; 3133 vtnet_rxq_free_mbufs(rxq); 3134 3135 txq = &sc->vtnet_txqs[i]; 3136 vtnet_txq_free_mbufs(txq); 3137 } 3138 } 3139 3140 static void 3141 vtnet_stop_rendezvous(struct vtnet_softc *sc) 3142 { 3143 struct vtnet_rxq *rxq; 3144 struct vtnet_txq *txq; 3145 int i; 3146 3147 VTNET_CORE_LOCK_ASSERT(sc); 3148 3149 /* 3150 * Lock and unlock the per-queue mutex so we known the stop 3151 * state is visible. Doing only the active queues should be 3152 * sufficient, but it does not cost much extra to do all the 3153 * queues. 3154 */ 3155 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3156 rxq = &sc->vtnet_rxqs[i]; 3157 VTNET_RXQ_LOCK(rxq); 3158 VTNET_RXQ_UNLOCK(rxq); 3159 3160 txq = &sc->vtnet_txqs[i]; 3161 VTNET_TXQ_LOCK(txq); 3162 VTNET_TXQ_UNLOCK(txq); 3163 } 3164 } 3165 3166 static void 3167 vtnet_stop(struct vtnet_softc *sc) 3168 { 3169 device_t dev; 3170 if_t ifp; 3171 3172 dev = sc->vtnet_dev; 3173 ifp = sc->vtnet_ifp; 3174 3175 VTNET_CORE_LOCK_ASSERT(sc); 3176 3177 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 3178 sc->vtnet_link_active = 0; 3179 callout_stop(&sc->vtnet_tick_ch); 3180 3181 /* Only advisory. */ 3182 vtnet_disable_interrupts(sc); 3183 3184 #ifdef DEV_NETMAP 3185 /* Stop any pending txsync/rxsync and disable them. */ 3186 netmap_disable_all_rings(ifp); 3187 #endif /* DEV_NETMAP */ 3188 3189 /* 3190 * Stop the host adapter. This resets it to the pre-initialized 3191 * state. It will not generate any interrupts until after it is 3192 * reinitialized. 3193 */ 3194 virtio_stop(dev); 3195 vtnet_stop_rendezvous(sc); 3196 3197 vtnet_drain_rxtx_queues(sc); 3198 sc->vtnet_act_vq_pairs = 1; 3199 } 3200 3201 static int 3202 vtnet_virtio_reinit(struct vtnet_softc *sc) 3203 { 3204 device_t dev; 3205 if_t ifp; 3206 uint64_t features; 3207 int error; 3208 3209 dev = sc->vtnet_dev; 3210 ifp = sc->vtnet_ifp; 3211 features = sc->vtnet_negotiated_features; 3212 3213 /* 3214 * Re-negotiate with the host, removing any disabled receive 3215 * features. Transmit features are disabled only on our side 3216 * via if_capenable and if_hwassist. 3217 */ 3218 3219 if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0) 3220 features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES); 3221 3222 if ((if_getcapenable(ifp) & IFCAP_LRO) == 0) 3223 features &= ~VTNET_LRO_FEATURES; 3224 3225 if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0) 3226 features &= ~VIRTIO_NET_F_CTRL_VLAN; 3227 3228 error = virtio_reinit(dev, features); 3229 if (error) { 3230 device_printf(dev, "virtio reinit error %d\n", error); 3231 return (error); 3232 } 3233 3234 sc->vtnet_features = features; 3235 virtio_reinit_complete(dev); 3236 3237 return (0); 3238 } 3239 3240 static void 3241 vtnet_init_rx_filters(struct vtnet_softc *sc) 3242 { 3243 if_t ifp; 3244 3245 ifp = sc->vtnet_ifp; 3246 3247 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 3248 vtnet_rx_filter(sc); 3249 vtnet_rx_filter_mac(sc); 3250 } 3251 3252 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) 3253 vtnet_rx_filter_vlan(sc); 3254 } 3255 3256 static int 3257 vtnet_init_rx_queues(struct vtnet_softc *sc) 3258 { 3259 device_t dev; 3260 if_t ifp; 3261 struct vtnet_rxq *rxq; 3262 int i, clustersz, error; 3263 3264 dev = sc->vtnet_dev; 3265 ifp = sc->vtnet_ifp; 3266 3267 clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp)); 3268 sc->vtnet_rx_clustersz = clustersz; 3269 3270 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) { 3271 sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) + 3272 VTNET_MAX_RX_SIZE, clustersz); 3273 KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 3274 ("%s: too many rx mbufs %d for %d segments", __func__, 3275 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 3276 } else 3277 sc->vtnet_rx_nmbufs = 1; 3278 3279 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3280 rxq = &sc->vtnet_rxqs[i]; 3281 3282 /* Hold the lock to satisfy asserts. */ 3283 VTNET_RXQ_LOCK(rxq); 3284 error = vtnet_rxq_populate(rxq); 3285 VTNET_RXQ_UNLOCK(rxq); 3286 3287 if (error) { 3288 device_printf(dev, "cannot populate Rx queue %d\n", i); 3289 return (error); 3290 } 3291 } 3292 3293 return (0); 3294 } 3295 3296 static int 3297 vtnet_init_tx_queues(struct vtnet_softc *sc) 3298 { 3299 struct vtnet_txq *txq; 3300 int i; 3301 3302 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3303 txq = &sc->vtnet_txqs[i]; 3304 txq->vtntx_watchdog = 0; 3305 txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq); 3306 #ifdef DEV_NETMAP 3307 netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0); 3308 #endif /* DEV_NETMAP */ 3309 } 3310 3311 return (0); 3312 } 3313 3314 static int 3315 vtnet_init_rxtx_queues(struct vtnet_softc *sc) 3316 { 3317 int error; 3318 3319 error = vtnet_init_rx_queues(sc); 3320 if (error) 3321 return (error); 3322 3323 error = vtnet_init_tx_queues(sc); 3324 if (error) 3325 return (error); 3326 3327 return (0); 3328 } 3329 3330 static void 3331 vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 3332 { 3333 device_t dev; 3334 int npairs; 3335 3336 dev = sc->vtnet_dev; 3337 3338 if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) { 3339 sc->vtnet_act_vq_pairs = 1; 3340 return; 3341 } 3342 3343 npairs = sc->vtnet_req_vq_pairs; 3344 3345 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 3346 device_printf(dev, "cannot set active queue pairs to %d, " 3347 "falling back to 1 queue pair\n", npairs); 3348 npairs = 1; 3349 } 3350 3351 sc->vtnet_act_vq_pairs = npairs; 3352 } 3353 3354 static void 3355 vtnet_update_rx_offloads(struct vtnet_softc *sc) 3356 { 3357 if_t ifp; 3358 uint64_t features; 3359 int error; 3360 3361 ifp = sc->vtnet_ifp; 3362 features = sc->vtnet_features; 3363 3364 VTNET_CORE_LOCK_ASSERT(sc); 3365 3366 if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 3367 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) 3368 features |= VIRTIO_NET_F_GUEST_CSUM; 3369 else 3370 features &= ~VIRTIO_NET_F_GUEST_CSUM; 3371 } 3372 3373 if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) { 3374 if (if_getcapenable(ifp) & IFCAP_LRO) 3375 features |= VTNET_LRO_FEATURES; 3376 else 3377 features &= ~VTNET_LRO_FEATURES; 3378 } 3379 3380 error = vtnet_ctrl_guest_offloads(sc, 3381 features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 | 3382 VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN | 3383 VIRTIO_NET_F_GUEST_UFO)); 3384 if (error) { 3385 device_printf(sc->vtnet_dev, 3386 "%s: cannot update Rx features\n", __func__); 3387 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 3388 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 3389 vtnet_init_locked(sc, 0); 3390 } 3391 } else 3392 sc->vtnet_features = features; 3393 } 3394 3395 static int 3396 vtnet_reinit(struct vtnet_softc *sc) 3397 { 3398 if_t ifp; 3399 int error; 3400 3401 ifp = sc->vtnet_ifp; 3402 3403 bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 3404 3405 error = vtnet_virtio_reinit(sc); 3406 if (error) 3407 return (error); 3408 3409 vtnet_set_macaddr(sc); 3410 vtnet_set_active_vq_pairs(sc); 3411 3412 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 3413 vtnet_init_rx_filters(sc); 3414 3415 if_sethwassist(ifp, 0); 3416 if (if_getcapenable(ifp) & IFCAP_TXCSUM) 3417 if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0); 3418 if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6) 3419 if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0); 3420 if (if_getcapenable(ifp) & IFCAP_TSO4) 3421 if_sethwassistbits(ifp, CSUM_IP_TSO, 0); 3422 if (if_getcapenable(ifp) & IFCAP_TSO6) 3423 if_sethwassistbits(ifp, CSUM_IP6_TSO, 0); 3424 3425 error = vtnet_init_rxtx_queues(sc); 3426 if (error) 3427 return (error); 3428 3429 return (0); 3430 } 3431 3432 static void 3433 vtnet_init_locked(struct vtnet_softc *sc, int init_mode) 3434 { 3435 if_t ifp; 3436 3437 ifp = sc->vtnet_ifp; 3438 3439 VTNET_CORE_LOCK_ASSERT(sc); 3440 3441 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) 3442 return; 3443 3444 vtnet_stop(sc); 3445 3446 #ifdef DEV_NETMAP 3447 /* Once stopped we can update the netmap flags, if necessary. */ 3448 switch (init_mode) { 3449 case VTNET_INIT_NETMAP_ENTER: 3450 nm_set_native_flags(NA(ifp)); 3451 break; 3452 case VTNET_INIT_NETMAP_EXIT: 3453 nm_clear_native_flags(NA(ifp)); 3454 break; 3455 } 3456 #endif /* DEV_NETMAP */ 3457 3458 if (vtnet_reinit(sc) != 0) { 3459 vtnet_stop(sc); 3460 return; 3461 } 3462 3463 if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0); 3464 vtnet_update_link_status(sc); 3465 vtnet_enable_interrupts(sc); 3466 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3467 3468 #ifdef DEV_NETMAP 3469 /* Re-enable txsync/rxsync. */ 3470 netmap_enable_all_rings(ifp); 3471 #endif /* DEV_NETMAP */ 3472 } 3473 3474 static void 3475 vtnet_init(void *xsc) 3476 { 3477 struct vtnet_softc *sc; 3478 3479 sc = xsc; 3480 3481 VTNET_CORE_LOCK(sc); 3482 vtnet_init_locked(sc, 0); 3483 VTNET_CORE_UNLOCK(sc); 3484 } 3485 3486 static void 3487 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3488 { 3489 3490 /* 3491 * The control virtqueue is only polled and therefore it should 3492 * already be empty. 3493 */ 3494 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq), 3495 ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq)); 3496 } 3497 3498 static void 3499 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3500 struct sglist *sg, int readable, int writable) 3501 { 3502 struct virtqueue *vq; 3503 3504 vq = sc->vtnet_ctrl_vq; 3505 3506 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ); 3507 VTNET_CORE_LOCK_ASSERT(sc); 3508 3509 if (!virtqueue_empty(vq)) 3510 return; 3511 3512 /* 3513 * Poll for the response, but the command is likely completed before 3514 * returning from the notify. 3515 */ 3516 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0) { 3517 virtqueue_notify(vq); 3518 virtqueue_poll(vq, NULL); 3519 } 3520 } 3521 3522 static int 3523 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3524 { 3525 struct sglist_seg segs[3]; 3526 struct sglist sg; 3527 struct { 3528 struct virtio_net_ctrl_hdr hdr __aligned(2); 3529 uint8_t pad1; 3530 uint8_t addr[ETHER_ADDR_LEN] __aligned(8); 3531 uint8_t pad2; 3532 uint8_t ack; 3533 } s; 3534 int error; 3535 3536 error = 0; 3537 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC); 3538 3539 s.hdr.class = VIRTIO_NET_CTRL_MAC; 3540 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3541 bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN); 3542 s.ack = VIRTIO_NET_ERR; 3543 3544 sglist_init(&sg, nitems(segs), segs); 3545 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3546 error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN); 3547 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3548 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3549 3550 if (error == 0) 3551 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3552 3553 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3554 } 3555 3556 static int 3557 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads) 3558 { 3559 struct sglist_seg segs[3]; 3560 struct sglist sg; 3561 struct { 3562 struct virtio_net_ctrl_hdr hdr __aligned(2); 3563 uint8_t pad1; 3564 uint64_t offloads __aligned(8); 3565 uint8_t pad2; 3566 uint8_t ack; 3567 } s; 3568 int error; 3569 3570 error = 0; 3571 MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 3572 3573 s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS; 3574 s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET; 3575 s.offloads = vtnet_gtoh64(sc, offloads); 3576 s.ack = VIRTIO_NET_ERR; 3577 3578 sglist_init(&sg, nitems(segs), segs); 3579 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3580 error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t)); 3581 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3582 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3583 3584 if (error == 0) 3585 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3586 3587 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3588 } 3589 3590 static int 3591 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3592 { 3593 struct sglist_seg segs[3]; 3594 struct sglist sg; 3595 struct { 3596 struct virtio_net_ctrl_hdr hdr __aligned(2); 3597 uint8_t pad1; 3598 struct virtio_net_ctrl_mq mq __aligned(2); 3599 uint8_t pad2; 3600 uint8_t ack; 3601 } s; 3602 int error; 3603 3604 error = 0; 3605 MPASS(sc->vtnet_flags & VTNET_FLAG_MQ); 3606 3607 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3608 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3609 s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs); 3610 s.ack = VIRTIO_NET_ERR; 3611 3612 sglist_init(&sg, nitems(segs), segs); 3613 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3614 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3615 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3616 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3617 3618 if (error == 0) 3619 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3620 3621 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3622 } 3623 3624 static int 3625 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on) 3626 { 3627 struct sglist_seg segs[3]; 3628 struct sglist sg; 3629 struct { 3630 struct virtio_net_ctrl_hdr hdr __aligned(2); 3631 uint8_t pad1; 3632 uint8_t onoff; 3633 uint8_t pad2; 3634 uint8_t ack; 3635 } s; 3636 int error; 3637 3638 error = 0; 3639 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX); 3640 3641 s.hdr.class = VIRTIO_NET_CTRL_RX; 3642 s.hdr.cmd = cmd; 3643 s.onoff = on; 3644 s.ack = VIRTIO_NET_ERR; 3645 3646 sglist_init(&sg, nitems(segs), segs); 3647 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3648 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3649 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3650 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3651 3652 if (error == 0) 3653 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3654 3655 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3656 } 3657 3658 static int 3659 vtnet_set_promisc(struct vtnet_softc *sc, bool on) 3660 { 3661 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3662 } 3663 3664 static int 3665 vtnet_set_allmulti(struct vtnet_softc *sc, bool on) 3666 { 3667 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3668 } 3669 3670 static void 3671 vtnet_rx_filter(struct vtnet_softc *sc) 3672 { 3673 device_t dev; 3674 if_t ifp; 3675 3676 dev = sc->vtnet_dev; 3677 ifp = sc->vtnet_ifp; 3678 3679 VTNET_CORE_LOCK_ASSERT(sc); 3680 3681 if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) { 3682 device_printf(dev, "cannot %s promiscuous mode\n", 3683 if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable"); 3684 } 3685 3686 if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) { 3687 device_printf(dev, "cannot %s all-multicast mode\n", 3688 if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable"); 3689 } 3690 } 3691 3692 static u_int 3693 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt) 3694 { 3695 struct vtnet_softc *sc = arg; 3696 3697 if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3698 return (0); 3699 3700 if (ucnt < VTNET_MAX_MAC_ENTRIES) 3701 bcopy(LLADDR(sdl), 3702 &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt], 3703 ETHER_ADDR_LEN); 3704 3705 return (1); 3706 } 3707 3708 static u_int 3709 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt) 3710 { 3711 struct vtnet_mac_filter *filter = arg; 3712 3713 if (mcnt < VTNET_MAX_MAC_ENTRIES) 3714 bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt], 3715 ETHER_ADDR_LEN); 3716 3717 return (1); 3718 } 3719 3720 static void 3721 vtnet_rx_filter_mac(struct vtnet_softc *sc) 3722 { 3723 struct virtio_net_ctrl_hdr hdr __aligned(2); 3724 struct vtnet_mac_filter *filter; 3725 struct sglist_seg segs[4]; 3726 struct sglist sg; 3727 if_t ifp; 3728 bool promisc, allmulti; 3729 u_int ucnt, mcnt; 3730 int error; 3731 uint8_t ack; 3732 3733 ifp = sc->vtnet_ifp; 3734 filter = sc->vtnet_mac_filter; 3735 error = 0; 3736 3737 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX); 3738 VTNET_CORE_LOCK_ASSERT(sc); 3739 3740 /* Unicast MAC addresses: */ 3741 ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc); 3742 promisc = (ucnt > VTNET_MAX_MAC_ENTRIES); 3743 3744 if (promisc) { 3745 ucnt = 0; 3746 if_printf(ifp, "more than %d MAC addresses assigned, " 3747 "falling back to promiscuous mode\n", 3748 VTNET_MAX_MAC_ENTRIES); 3749 } 3750 3751 /* Multicast MAC addresses: */ 3752 mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter); 3753 allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES); 3754 3755 if (allmulti) { 3756 mcnt = 0; 3757 if_printf(ifp, "more than %d multicast MAC addresses " 3758 "assigned, falling back to all-multicast mode\n", 3759 VTNET_MAX_MAC_ENTRIES); 3760 } 3761 3762 if (promisc && allmulti) 3763 goto out; 3764 3765 filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt); 3766 filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt); 3767 3768 hdr.class = VIRTIO_NET_CTRL_MAC; 3769 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3770 ack = VIRTIO_NET_ERR; 3771 3772 sglist_init(&sg, nitems(segs), segs); 3773 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3774 error |= sglist_append(&sg, &filter->vmf_unicast, 3775 sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN); 3776 error |= sglist_append(&sg, &filter->vmf_multicast, 3777 sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN); 3778 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3779 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3780 3781 if (error == 0) 3782 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3783 if (ack != VIRTIO_NET_OK) 3784 if_printf(ifp, "error setting host MAC filter table\n"); 3785 3786 out: 3787 if (promisc != 0 && vtnet_set_promisc(sc, true) != 0) 3788 if_printf(ifp, "cannot enable promiscuous mode\n"); 3789 if (allmulti != 0 && vtnet_set_allmulti(sc, true) != 0) 3790 if_printf(ifp, "cannot enable all-multicast mode\n"); 3791 } 3792 3793 static int 3794 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3795 { 3796 struct sglist_seg segs[3]; 3797 struct sglist sg; 3798 struct { 3799 struct virtio_net_ctrl_hdr hdr __aligned(2); 3800 uint8_t pad1; 3801 uint16_t tag __aligned(2); 3802 uint8_t pad2; 3803 uint8_t ack; 3804 } s; 3805 int error; 3806 3807 error = 0; 3808 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER); 3809 3810 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3811 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3812 s.tag = vtnet_gtoh16(sc, tag); 3813 s.ack = VIRTIO_NET_ERR; 3814 3815 sglist_init(&sg, nitems(segs), segs); 3816 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3817 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3818 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3819 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3820 3821 if (error == 0) 3822 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3823 3824 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3825 } 3826 3827 static void 3828 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3829 { 3830 int i, bit; 3831 uint32_t w; 3832 uint16_t tag; 3833 3834 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER); 3835 VTNET_CORE_LOCK_ASSERT(sc); 3836 3837 /* Enable the filter for each configured VLAN. */ 3838 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3839 w = sc->vtnet_vlan_filter[i]; 3840 3841 while ((bit = ffs(w) - 1) != -1) { 3842 w &= ~(1 << bit); 3843 tag = sizeof(w) * CHAR_BIT * i + bit; 3844 3845 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3846 device_printf(sc->vtnet_dev, 3847 "cannot enable VLAN %d filter\n", tag); 3848 } 3849 } 3850 } 3851 } 3852 3853 static void 3854 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3855 { 3856 if_t ifp; 3857 int idx, bit; 3858 3859 ifp = sc->vtnet_ifp; 3860 idx = (tag >> 5) & 0x7F; 3861 bit = tag & 0x1F; 3862 3863 if (tag == 0 || tag > 4095) 3864 return; 3865 3866 VTNET_CORE_LOCK(sc); 3867 3868 if (add) 3869 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3870 else 3871 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3872 3873 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER && 3874 if_getdrvflags(ifp) & IFF_DRV_RUNNING && 3875 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3876 device_printf(sc->vtnet_dev, 3877 "cannot %s VLAN %d %s the host filter table\n", 3878 add ? "add" : "remove", tag, add ? "to" : "from"); 3879 } 3880 3881 VTNET_CORE_UNLOCK(sc); 3882 } 3883 3884 static void 3885 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag) 3886 { 3887 3888 if (if_getsoftc(ifp) != arg) 3889 return; 3890 3891 vtnet_update_vlan_filter(arg, 1, tag); 3892 } 3893 3894 static void 3895 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag) 3896 { 3897 3898 if (if_getsoftc(ifp) != arg) 3899 return; 3900 3901 vtnet_update_vlan_filter(arg, 0, tag); 3902 } 3903 3904 static void 3905 vtnet_update_speed_duplex(struct vtnet_softc *sc) 3906 { 3907 if_t ifp; 3908 uint32_t speed; 3909 3910 ifp = sc->vtnet_ifp; 3911 3912 if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0) 3913 return; 3914 3915 /* BMV: Ignore duplex. */ 3916 speed = virtio_read_dev_config_4(sc->vtnet_dev, 3917 offsetof(struct virtio_net_config, speed)); 3918 if (speed != UINT32_MAX) 3919 if_setbaudrate(ifp, IF_Mbps(speed)); 3920 } 3921 3922 static int 3923 vtnet_is_link_up(struct vtnet_softc *sc) 3924 { 3925 uint16_t status; 3926 3927 if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0) 3928 return (1); 3929 3930 status = virtio_read_dev_config_2(sc->vtnet_dev, 3931 offsetof(struct virtio_net_config, status)); 3932 3933 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3934 } 3935 3936 static void 3937 vtnet_update_link_status(struct vtnet_softc *sc) 3938 { 3939 if_t ifp; 3940 int link; 3941 3942 ifp = sc->vtnet_ifp; 3943 VTNET_CORE_LOCK_ASSERT(sc); 3944 link = vtnet_is_link_up(sc); 3945 3946 /* Notify if the link status has changed. */ 3947 if (link != 0 && sc->vtnet_link_active == 0) { 3948 vtnet_update_speed_duplex(sc); 3949 sc->vtnet_link_active = 1; 3950 if_link_state_change(ifp, LINK_STATE_UP); 3951 } else if (link == 0 && sc->vtnet_link_active != 0) { 3952 sc->vtnet_link_active = 0; 3953 if_link_state_change(ifp, LINK_STATE_DOWN); 3954 } 3955 } 3956 3957 static int 3958 vtnet_ifmedia_upd(if_t ifp __unused) 3959 { 3960 return (EOPNOTSUPP); 3961 } 3962 3963 static void 3964 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr) 3965 { 3966 struct vtnet_softc *sc; 3967 3968 sc = if_getsoftc(ifp); 3969 3970 ifmr->ifm_status = IFM_AVALID; 3971 ifmr->ifm_active = IFM_ETHER; 3972 3973 VTNET_CORE_LOCK(sc); 3974 if (vtnet_is_link_up(sc) != 0) { 3975 ifmr->ifm_status |= IFM_ACTIVE; 3976 ifmr->ifm_active |= IFM_10G_T | IFM_FDX; 3977 } else 3978 ifmr->ifm_active |= IFM_NONE; 3979 VTNET_CORE_UNLOCK(sc); 3980 } 3981 3982 static void 3983 vtnet_get_macaddr(struct vtnet_softc *sc) 3984 { 3985 3986 if (sc->vtnet_flags & VTNET_FLAG_MAC) { 3987 virtio_read_device_config_array(sc->vtnet_dev, 3988 offsetof(struct virtio_net_config, mac), 3989 &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN); 3990 } else { 3991 /* Generate a random locally administered unicast address. */ 3992 sc->vtnet_hwaddr[0] = 0xB2; 3993 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 3994 } 3995 } 3996 3997 static void 3998 vtnet_set_macaddr(struct vtnet_softc *sc) 3999 { 4000 device_t dev; 4001 int error; 4002 4003 dev = sc->vtnet_dev; 4004 4005 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 4006 error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr); 4007 if (error) 4008 device_printf(dev, "unable to set MAC address\n"); 4009 return; 4010 } 4011 4012 /* MAC in config is read-only in modern VirtIO. */ 4013 if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) { 4014 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4015 virtio_write_dev_config_1(dev, 4016 offsetof(struct virtio_net_config, mac) + i, 4017 sc->vtnet_hwaddr[i]); 4018 } 4019 } 4020 } 4021 4022 static void 4023 vtnet_attached_set_macaddr(struct vtnet_softc *sc) 4024 { 4025 4026 /* Assign MAC address if it was generated. */ 4027 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) 4028 vtnet_set_macaddr(sc); 4029 } 4030 4031 static void 4032 vtnet_vlan_tag_remove(struct mbuf *m) 4033 { 4034 struct ether_vlan_header *evh; 4035 4036 evh = mtod(m, struct ether_vlan_header *); 4037 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 4038 m->m_flags |= M_VLANTAG; 4039 4040 /* Strip the 802.1Q header. */ 4041 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 4042 ETHER_HDR_LEN - ETHER_TYPE_LEN); 4043 m_adj(m, ETHER_VLAN_ENCAP_LEN); 4044 } 4045 4046 static void 4047 vtnet_set_rx_process_limit(struct vtnet_softc *sc) 4048 { 4049 int limit; 4050 4051 limit = vtnet_tunable_int(sc, "rx_process_limit", 4052 vtnet_rx_process_limit); 4053 if (limit < 0) 4054 limit = INT_MAX; 4055 sc->vtnet_rx_process_limit = limit; 4056 } 4057 4058 static void 4059 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 4060 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 4061 { 4062 struct sysctl_oid *node; 4063 struct sysctl_oid_list *list; 4064 struct vtnet_rxq_stats *stats; 4065 char namebuf[16]; 4066 4067 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 4068 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 4069 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue"); 4070 list = SYSCTL_CHILDREN(node); 4071 4072 stats = &rxq->vtnrx_stats; 4073 4074 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", CTLFLAG_RD, 4075 &stats->vrxs_ipackets, "Receive packets"); 4076 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", CTLFLAG_RD, 4077 &stats->vrxs_ibytes, "Receive bytes"); 4078 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", CTLFLAG_RD, 4079 &stats->vrxs_iqdrops, "Receive drops"); 4080 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", CTLFLAG_RD, 4081 &stats->vrxs_ierrors, "Receive errors"); 4082 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 4083 &stats->vrxs_csum, "Receive checksum offloaded"); 4084 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", CTLFLAG_RD, 4085 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 4086 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", CTLFLAG_RD, 4087 &stats->vrxs_host_lro, "Receive host segmentation offloaded"); 4088 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 4089 &stats->vrxs_rescheduled, 4090 "Receive interrupt handler rescheduled"); 4091 } 4092 4093 static void 4094 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 4095 struct sysctl_oid_list *child, struct vtnet_txq *txq) 4096 { 4097 struct sysctl_oid *node; 4098 struct sysctl_oid_list *list; 4099 struct vtnet_txq_stats *stats; 4100 char namebuf[16]; 4101 4102 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 4103 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 4104 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue"); 4105 list = SYSCTL_CHILDREN(node); 4106 4107 stats = &txq->vtntx_stats; 4108 4109 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", CTLFLAG_RD, 4110 &stats->vtxs_opackets, "Transmit packets"); 4111 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", CTLFLAG_RD, 4112 &stats->vtxs_obytes, "Transmit bytes"); 4113 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", CTLFLAG_RD, 4114 &stats->vtxs_omcasts, "Transmit multicasts"); 4115 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", CTLFLAG_RD, 4116 &stats->vtxs_csum, "Transmit checksum offloaded"); 4117 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", CTLFLAG_RD, 4118 &stats->vtxs_tso, "Transmit TCP segmentation offloaded"); 4119 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", CTLFLAG_RD, 4120 &stats->vtxs_rescheduled, 4121 "Transmit interrupt handler rescheduled"); 4122 } 4123 4124 static void 4125 vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 4126 { 4127 device_t dev; 4128 struct sysctl_ctx_list *ctx; 4129 struct sysctl_oid *tree; 4130 struct sysctl_oid_list *child; 4131 int i; 4132 4133 dev = sc->vtnet_dev; 4134 ctx = device_get_sysctl_ctx(dev); 4135 tree = device_get_sysctl_tree(dev); 4136 child = SYSCTL_CHILDREN(tree); 4137 4138 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) { 4139 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 4140 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 4141 } 4142 } 4143 4144 static void 4145 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 4146 struct sysctl_oid_list *child, struct vtnet_softc *sc) 4147 { 4148 struct vtnet_statistics *stats; 4149 struct vtnet_rxq_stats rxaccum; 4150 struct vtnet_txq_stats txaccum; 4151 4152 vtnet_accum_stats(sc, &rxaccum, &txaccum); 4153 4154 stats = &sc->vtnet_stats; 4155 stats->rx_csum_offloaded = rxaccum.vrxs_csum; 4156 stats->rx_csum_failed = rxaccum.vrxs_csum_failed; 4157 stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 4158 stats->tx_csum_offloaded = txaccum.vtxs_csum; 4159 stats->tx_tso_offloaded = txaccum.vtxs_tso; 4160 stats->tx_task_rescheduled = txaccum.vtxs_rescheduled; 4161 4162 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 4163 CTLFLAG_RD, &stats->mbuf_alloc_failed, 4164 "Mbuf cluster allocation failures"); 4165 4166 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 4167 CTLFLAG_RD, &stats->rx_frame_too_large, 4168 "Received frame larger than the mbuf chain"); 4169 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 4170 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 4171 "Enqueuing the replacement receive mbuf failed"); 4172 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 4173 CTLFLAG_RD, &stats->rx_mergeable_failed, 4174 "Mergeable buffers receive failures"); 4175 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 4176 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 4177 "Received checksum offloaded buffer with unsupported " 4178 "Ethernet type"); 4179 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 4180 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 4181 "Received checksum offloaded buffer with incorrect IP protocol"); 4182 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 4183 CTLFLAG_RD, &stats->rx_csum_bad_offset, 4184 "Received checksum offloaded buffer with incorrect offset"); 4185 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_proto", 4186 CTLFLAG_RD, &stats->rx_csum_bad_proto, 4187 "Received checksum offloaded buffer with incorrect protocol"); 4188 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed", 4189 CTLFLAG_RD, &stats->rx_csum_failed, 4190 "Received buffer checksum offload failed"); 4191 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded", 4192 CTLFLAG_RD, &stats->rx_csum_offloaded, 4193 "Received buffer checksum offload succeeded"); 4194 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled", 4195 CTLFLAG_RD, &stats->rx_task_rescheduled, 4196 "Times the receive interrupt task rescheduled itself"); 4197 4198 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype", 4199 CTLFLAG_RD, &stats->tx_csum_unknown_ethtype, 4200 "Aborted transmit of checksum offloaded buffer with unknown " 4201 "Ethernet type"); 4202 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch", 4203 CTLFLAG_RD, &stats->tx_csum_proto_mismatch, 4204 "Aborted transmit of checksum offloaded buffer because mismatched " 4205 "protocols"); 4206 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 4207 CTLFLAG_RD, &stats->tx_tso_not_tcp, 4208 "Aborted transmit of TSO buffer with non TCP protocol"); 4209 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum", 4210 CTLFLAG_RD, &stats->tx_tso_without_csum, 4211 "Aborted transmit of TSO buffer without TCP checksum offload"); 4212 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 4213 CTLFLAG_RD, &stats->tx_defragged, 4214 "Transmit mbufs defragged"); 4215 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 4216 CTLFLAG_RD, &stats->tx_defrag_failed, 4217 "Aborted transmit of buffer because defrag failed"); 4218 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded", 4219 CTLFLAG_RD, &stats->tx_csum_offloaded, 4220 "Offloaded checksum of transmitted buffer"); 4221 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded", 4222 CTLFLAG_RD, &stats->tx_tso_offloaded, 4223 "Segmentation offload of transmitted buffer"); 4224 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled", 4225 CTLFLAG_RD, &stats->tx_task_rescheduled, 4226 "Times the transmit interrupt task rescheduled itself"); 4227 } 4228 4229 static void 4230 vtnet_setup_sysctl(struct vtnet_softc *sc) 4231 { 4232 device_t dev; 4233 struct sysctl_ctx_list *ctx; 4234 struct sysctl_oid *tree; 4235 struct sysctl_oid_list *child; 4236 4237 dev = sc->vtnet_dev; 4238 ctx = device_get_sysctl_ctx(dev); 4239 tree = device_get_sysctl_tree(dev); 4240 child = SYSCTL_CHILDREN(tree); 4241 4242 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 4243 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 4244 "Number of maximum supported virtqueue pairs"); 4245 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs", 4246 CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0, 4247 "Number of requested virtqueue pairs"); 4248 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 4249 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 4250 "Number of active virtqueue pairs"); 4251 4252 vtnet_setup_stat_sysctl(ctx, child, sc); 4253 } 4254 4255 static void 4256 vtnet_load_tunables(struct vtnet_softc *sc) 4257 { 4258 4259 sc->vtnet_lro_entry_count = vtnet_tunable_int(sc, 4260 "lro_entry_count", vtnet_lro_entry_count); 4261 if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES) 4262 sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES; 4263 4264 sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc, 4265 "lro_mbufq_depth", vtnet_lro_mbufq_depth); 4266 } 4267 4268 static int 4269 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 4270 { 4271 4272 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 4273 } 4274 4275 static void 4276 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 4277 { 4278 4279 virtqueue_disable_intr(rxq->vtnrx_vq); 4280 } 4281 4282 static int 4283 vtnet_txq_enable_intr(struct vtnet_txq *txq) 4284 { 4285 struct virtqueue *vq; 4286 4287 vq = txq->vtntx_vq; 4288 4289 if (vtnet_txq_below_threshold(txq) != 0) 4290 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG)); 4291 4292 /* 4293 * The free count is above our threshold. Keep the Tx interrupt 4294 * disabled until the queue is fuller. 4295 */ 4296 return (0); 4297 } 4298 4299 static void 4300 vtnet_txq_disable_intr(struct vtnet_txq *txq) 4301 { 4302 4303 virtqueue_disable_intr(txq->vtntx_vq); 4304 } 4305 4306 static void 4307 vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 4308 { 4309 struct vtnet_rxq *rxq; 4310 int i; 4311 4312 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 4313 rxq = &sc->vtnet_rxqs[i]; 4314 if (vtnet_rxq_enable_intr(rxq) != 0) 4315 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 4316 } 4317 } 4318 4319 static void 4320 vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 4321 { 4322 int i; 4323 4324 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4325 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 4326 } 4327 4328 static void 4329 vtnet_enable_interrupts(struct vtnet_softc *sc) 4330 { 4331 4332 vtnet_enable_rx_interrupts(sc); 4333 vtnet_enable_tx_interrupts(sc); 4334 } 4335 4336 static void 4337 vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 4338 { 4339 int i; 4340 4341 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4342 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 4343 } 4344 4345 static void 4346 vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 4347 { 4348 int i; 4349 4350 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4351 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 4352 } 4353 4354 static void 4355 vtnet_disable_interrupts(struct vtnet_softc *sc) 4356 { 4357 4358 vtnet_disable_rx_interrupts(sc); 4359 vtnet_disable_tx_interrupts(sc); 4360 } 4361 4362 static int 4363 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 4364 { 4365 char path[64]; 4366 4367 snprintf(path, sizeof(path), 4368 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 4369 TUNABLE_INT_FETCH(path, &def); 4370 4371 return (def); 4372 } 4373 4374 #ifdef DEBUGNET 4375 static void 4376 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize) 4377 { 4378 struct vtnet_softc *sc; 4379 4380 sc = if_getsoftc(ifp); 4381 4382 VTNET_CORE_LOCK(sc); 4383 *nrxr = sc->vtnet_req_vq_pairs; 4384 *ncl = DEBUGNET_MAX_IN_FLIGHT; 4385 *clsize = sc->vtnet_rx_clustersz; 4386 VTNET_CORE_UNLOCK(sc); 4387 } 4388 4389 static void 4390 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event) 4391 { 4392 struct vtnet_softc *sc; 4393 static bool sw_lro_enabled = false; 4394 4395 /* 4396 * Disable software LRO, since it would require entering the network 4397 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll(). 4398 */ 4399 sc = if_getsoftc(ifp); 4400 switch (event) { 4401 case DEBUGNET_START: 4402 sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0; 4403 if (sw_lro_enabled) 4404 sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO; 4405 break; 4406 case DEBUGNET_END: 4407 if (sw_lro_enabled) 4408 sc->vtnet_flags |= VTNET_FLAG_SW_LRO; 4409 break; 4410 } 4411 } 4412 4413 static int 4414 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m) 4415 { 4416 struct vtnet_softc *sc; 4417 struct vtnet_txq *txq; 4418 int error; 4419 4420 sc = if_getsoftc(ifp); 4421 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4422 IFF_DRV_RUNNING) 4423 return (EBUSY); 4424 4425 txq = &sc->vtnet_txqs[0]; 4426 error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE); 4427 if (error == 0) 4428 (void)vtnet_txq_notify(txq); 4429 return (error); 4430 } 4431 4432 static int 4433 vtnet_debugnet_poll(if_t ifp, int count) 4434 { 4435 struct vtnet_softc *sc; 4436 int i; 4437 4438 sc = if_getsoftc(ifp); 4439 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4440 IFF_DRV_RUNNING) 4441 return (EBUSY); 4442 4443 (void)vtnet_txq_eof(&sc->vtnet_txqs[0]); 4444 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4445 (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]); 4446 return (0); 4447 } 4448 #endif /* DEBUGNET */ 4449