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 "opt_inet.h" 32 #include "opt_inet6.h" 33 34 #include <sys/param.h> 35 #include <sys/eventhandler.h> 36 #include <sys/systm.h> 37 #include <sys/kernel.h> 38 #include <sys/sockio.h> 39 #include <sys/malloc.h> 40 #include <sys/mbuf.h> 41 #include <sys/module.h> 42 #include <sys/msan.h> 43 #include <sys/sbuf.h> 44 #include <sys/socket.h> 45 #include <sys/sysctl.h> 46 #include <sys/random.h> 47 #include <sys/sglist.h> 48 #include <sys/lock.h> 49 #include <sys/mutex.h> 50 #include <sys/taskqueue.h> 51 #include <sys/smp.h> 52 #include <machine/smp.h> 53 54 #include <vm/uma.h> 55 56 #include <net/debugnet.h> 57 #include <net/ethernet.h> 58 #include <net/pfil.h> 59 #include <net/if.h> 60 #include <net/if_var.h> 61 #include <net/if_arp.h> 62 #include <net/if_dl.h> 63 #include <net/if_types.h> 64 #include <net/if_media.h> 65 #include <net/if_vlan_var.h> 66 67 #include <net/bpf.h> 68 69 #include <netinet/in_systm.h> 70 #include <netinet/in.h> 71 #include <netinet/ip.h> 72 #include <netinet/ip6.h> 73 #include <netinet6/ip6_var.h> 74 #include <netinet/udp.h> 75 #include <netinet/tcp.h> 76 #include <netinet/tcp_lro.h> 77 78 #include <machine/bus.h> 79 #include <machine/resource.h> 80 #include <sys/bus.h> 81 #include <sys/rman.h> 82 83 #include <dev/virtio/virtio.h> 84 #include <dev/virtio/virtqueue.h> 85 #include <dev/virtio/network/virtio_net.h> 86 #include <dev/virtio/network/if_vtnetvar.h> 87 #include "virtio_if.h" 88 89 #if defined(INET) || defined(INET6) 90 #include <machine/in_cksum.h> 91 #endif 92 93 #ifdef __NO_STRICT_ALIGNMENT 94 #define VTNET_ETHER_ALIGN 0 95 #else /* Strict alignment */ 96 #define VTNET_ETHER_ALIGN ETHER_ALIGN 97 #endif 98 99 static int vtnet_modevent(module_t, int, void *); 100 101 static int vtnet_probe(device_t); 102 static int vtnet_attach(device_t); 103 static int vtnet_detach(device_t); 104 static int vtnet_suspend(device_t); 105 static int vtnet_resume(device_t); 106 static int vtnet_shutdown(device_t); 107 static int vtnet_attach_completed(device_t); 108 static int vtnet_config_change(device_t); 109 110 static int vtnet_negotiate_features(struct vtnet_softc *); 111 static int vtnet_setup_features(struct vtnet_softc *); 112 static int vtnet_init_rxq(struct vtnet_softc *, int); 113 static int vtnet_init_txq(struct vtnet_softc *, int); 114 static int vtnet_alloc_rxtx_queues(struct vtnet_softc *); 115 static void vtnet_free_rxtx_queues(struct vtnet_softc *); 116 static int vtnet_alloc_rx_filters(struct vtnet_softc *); 117 static void vtnet_free_rx_filters(struct vtnet_softc *); 118 static int vtnet_alloc_virtqueues(struct vtnet_softc *); 119 static void vtnet_alloc_interface(struct vtnet_softc *); 120 static int vtnet_setup_interface(struct vtnet_softc *); 121 static int vtnet_ioctl_mtu(struct vtnet_softc *, u_int); 122 static int vtnet_ioctl_ifflags(struct vtnet_softc *); 123 static int vtnet_ioctl_multi(struct vtnet_softc *); 124 static int vtnet_ioctl_ifcap(struct vtnet_softc *, struct ifreq *); 125 static int vtnet_ioctl(if_t, u_long, caddr_t); 126 static uint64_t vtnet_get_counter(if_t, ift_counter); 127 128 static int vtnet_rxq_populate(struct vtnet_rxq *); 129 static void vtnet_rxq_free_mbufs(struct vtnet_rxq *); 130 static struct mbuf * 131 vtnet_rx_alloc_buf(struct vtnet_softc *, int , struct mbuf **); 132 static int vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *, 133 struct mbuf *, int); 134 static int vtnet_rxq_replace_buf(struct vtnet_rxq *, struct mbuf *, int); 135 static int vtnet_rxq_enqueue_buf(struct vtnet_rxq *, struct mbuf *); 136 static int vtnet_rxq_new_buf(struct vtnet_rxq *); 137 #if defined(INET) || defined(INET6) 138 static int vtnet_rxq_csum_needs_csum(struct vtnet_rxq *, struct mbuf *, 139 bool, int, struct virtio_net_hdr *); 140 static void vtnet_rxq_csum_data_valid(struct vtnet_rxq *, struct mbuf *, 141 int); 142 static int vtnet_rxq_csum(struct vtnet_rxq *, struct mbuf *, 143 struct virtio_net_hdr *); 144 #endif 145 static void vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *, int); 146 static void vtnet_rxq_discard_buf(struct vtnet_rxq *, struct mbuf *); 147 static int vtnet_rxq_merged_eof(struct vtnet_rxq *, struct mbuf *, int); 148 static void vtnet_rxq_input(struct vtnet_rxq *, struct mbuf *, 149 struct virtio_net_hdr *); 150 static int vtnet_rxq_eof(struct vtnet_rxq *); 151 static void vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries); 152 static void vtnet_rx_vq_intr(void *); 153 static void vtnet_rxq_tq_intr(void *, int); 154 155 static int vtnet_txq_intr_threshold(struct vtnet_txq *); 156 static int vtnet_txq_below_threshold(struct vtnet_txq *); 157 static int vtnet_txq_notify(struct vtnet_txq *); 158 static void vtnet_txq_free_mbufs(struct vtnet_txq *); 159 static int vtnet_txq_offload_ctx(struct vtnet_txq *, struct mbuf *, 160 int *, int *, int *); 161 static int vtnet_txq_offload_tso(struct vtnet_txq *, struct mbuf *, int, 162 int, struct virtio_net_hdr *); 163 static struct mbuf * 164 vtnet_txq_offload(struct vtnet_txq *, struct mbuf *, 165 struct virtio_net_hdr *); 166 static int vtnet_txq_enqueue_buf(struct vtnet_txq *, struct mbuf **, 167 struct vtnet_tx_header *); 168 static int vtnet_txq_encap(struct vtnet_txq *, struct mbuf **, int); 169 170 /* Required for ALTQ */ 171 static void vtnet_start_locked(struct vtnet_txq *, if_t); 172 static void vtnet_start(if_t); 173 174 /* Required for MQ */ 175 static int vtnet_txq_mq_start_locked(struct vtnet_txq *, struct mbuf *); 176 static int vtnet_txq_mq_start(if_t, struct mbuf *); 177 static void vtnet_txq_tq_deferred(void *, int); 178 static void vtnet_qflush(if_t); 179 180 181 static void vtnet_txq_start(struct vtnet_txq *); 182 static void vtnet_txq_tq_intr(void *, int); 183 static int vtnet_txq_eof(struct vtnet_txq *); 184 static void vtnet_tx_vq_intr(void *); 185 static void vtnet_tx_start_all(struct vtnet_softc *); 186 187 static int vtnet_watchdog(struct vtnet_txq *); 188 static void vtnet_accum_stats(struct vtnet_softc *, 189 struct vtnet_rxq_stats *, struct vtnet_txq_stats *); 190 static void vtnet_tick(void *); 191 192 static void vtnet_start_taskqueues(struct vtnet_softc *); 193 static void vtnet_free_taskqueues(struct vtnet_softc *); 194 static void vtnet_drain_taskqueues(struct vtnet_softc *); 195 196 static void vtnet_drain_rxtx_queues(struct vtnet_softc *); 197 static void vtnet_stop_rendezvous(struct vtnet_softc *); 198 static void vtnet_stop(struct vtnet_softc *); 199 static int vtnet_virtio_reinit(struct vtnet_softc *); 200 static void vtnet_init_rx_filters(struct vtnet_softc *); 201 static int vtnet_init_rx_queues(struct vtnet_softc *); 202 static int vtnet_init_tx_queues(struct vtnet_softc *); 203 static int vtnet_init_rxtx_queues(struct vtnet_softc *); 204 static void vtnet_set_active_vq_pairs(struct vtnet_softc *); 205 static void vtnet_update_rx_offloads(struct vtnet_softc *); 206 static int vtnet_reinit(struct vtnet_softc *); 207 static void vtnet_init_locked(struct vtnet_softc *, int); 208 static void vtnet_init(void *); 209 210 static void vtnet_free_ctrl_vq(struct vtnet_softc *); 211 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *, 212 struct sglist *, int, int); 213 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *); 214 static int vtnet_ctrl_guest_offloads(struct vtnet_softc *, uint64_t); 215 static int vtnet_ctrl_mq_cmd(struct vtnet_softc *, uint16_t); 216 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, uint8_t, bool); 217 static int vtnet_set_promisc(struct vtnet_softc *, bool); 218 static int vtnet_set_allmulti(struct vtnet_softc *, bool); 219 static void vtnet_rx_filter(struct vtnet_softc *); 220 static void vtnet_rx_filter_mac(struct vtnet_softc *); 221 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t); 222 static void vtnet_rx_filter_vlan(struct vtnet_softc *); 223 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t); 224 static void vtnet_register_vlan(void *, if_t, uint16_t); 225 static void vtnet_unregister_vlan(void *, if_t, uint16_t); 226 227 static void vtnet_update_speed_duplex(struct vtnet_softc *); 228 static int vtnet_is_link_up(struct vtnet_softc *); 229 static void vtnet_update_link_status(struct vtnet_softc *); 230 static int vtnet_ifmedia_upd(if_t); 231 static void vtnet_ifmedia_sts(if_t, struct ifmediareq *); 232 static void vtnet_get_macaddr(struct vtnet_softc *); 233 static void vtnet_set_macaddr(struct vtnet_softc *); 234 static void vtnet_attached_set_macaddr(struct vtnet_softc *); 235 static void vtnet_vlan_tag_remove(struct mbuf *); 236 static void vtnet_set_rx_process_limit(struct vtnet_softc *); 237 238 static void vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *, 239 struct sysctl_oid_list *, struct vtnet_rxq *); 240 static void vtnet_setup_txq_sysctl(struct sysctl_ctx_list *, 241 struct sysctl_oid_list *, struct vtnet_txq *); 242 static void vtnet_setup_queue_sysctl(struct vtnet_softc *); 243 static void vtnet_load_tunables(struct vtnet_softc *); 244 static void vtnet_setup_sysctl(struct vtnet_softc *); 245 246 static int vtnet_rxq_enable_intr(struct vtnet_rxq *); 247 static void vtnet_rxq_disable_intr(struct vtnet_rxq *); 248 static int vtnet_txq_enable_intr(struct vtnet_txq *); 249 static void vtnet_txq_disable_intr(struct vtnet_txq *); 250 static void vtnet_enable_rx_interrupts(struct vtnet_softc *); 251 static void vtnet_enable_tx_interrupts(struct vtnet_softc *); 252 static void vtnet_enable_interrupts(struct vtnet_softc *); 253 static void vtnet_disable_rx_interrupts(struct vtnet_softc *); 254 static void vtnet_disable_tx_interrupts(struct vtnet_softc *); 255 static void vtnet_disable_interrupts(struct vtnet_softc *); 256 257 static int vtnet_tunable_int(struct vtnet_softc *, const char *, int); 258 259 DEBUGNET_DEFINE(vtnet); 260 261 #define vtnet_htog16(_sc, _val) virtio_htog16(vtnet_modern(_sc), _val) 262 #define vtnet_htog32(_sc, _val) virtio_htog32(vtnet_modern(_sc), _val) 263 #define vtnet_htog64(_sc, _val) virtio_htog64(vtnet_modern(_sc), _val) 264 #define vtnet_gtoh16(_sc, _val) virtio_gtoh16(vtnet_modern(_sc), _val) 265 #define vtnet_gtoh32(_sc, _val) virtio_gtoh32(vtnet_modern(_sc), _val) 266 #define vtnet_gtoh64(_sc, _val) virtio_gtoh64(vtnet_modern(_sc), _val) 267 268 /* Tunables. */ 269 static SYSCTL_NODE(_hw, OID_AUTO, vtnet, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 270 "VirtIO Net driver parameters"); 271 272 static int vtnet_csum_disable = 0; 273 SYSCTL_INT(_hw_vtnet, OID_AUTO, csum_disable, CTLFLAG_RDTUN, 274 &vtnet_csum_disable, 0, "Disables receive and send checksum offload"); 275 276 static int vtnet_fixup_needs_csum = 0; 277 SYSCTL_INT(_hw_vtnet, OID_AUTO, fixup_needs_csum, CTLFLAG_RDTUN, 278 &vtnet_fixup_needs_csum, 0, 279 "Calculate valid checksum for NEEDS_CSUM packets"); 280 281 static int vtnet_tso_disable = 0; 282 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_disable, CTLFLAG_RDTUN, 283 &vtnet_tso_disable, 0, "Disables TSO"); 284 285 static int vtnet_lro_disable = 1; 286 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_disable, CTLFLAG_RDTUN, 287 &vtnet_lro_disable, 0, "Disables hardware LRO"); 288 289 static int vtnet_mq_disable = 0; 290 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_disable, CTLFLAG_RDTUN, 291 &vtnet_mq_disable, 0, "Disables multiqueue support"); 292 293 static int vtnet_mq_max_pairs = VTNET_MAX_QUEUE_PAIRS; 294 SYSCTL_INT(_hw_vtnet, OID_AUTO, mq_max_pairs, CTLFLAG_RDTUN, 295 &vtnet_mq_max_pairs, 0, "Maximum number of multiqueue pairs"); 296 297 static int vtnet_tso_maxlen = IP_MAXPACKET; 298 SYSCTL_INT(_hw_vtnet, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN, 299 &vtnet_tso_maxlen, 0, "TSO burst limit"); 300 301 static int vtnet_rx_process_limit = 1024; 302 SYSCTL_INT(_hw_vtnet, OID_AUTO, rx_process_limit, CTLFLAG_RDTUN, 303 &vtnet_rx_process_limit, 0, 304 "Number of RX segments processed in one pass"); 305 306 static int vtnet_lro_entry_count = 128; 307 SYSCTL_INT(_hw_vtnet, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN, 308 &vtnet_lro_entry_count, 0, "Software LRO entry count"); 309 310 /* Enable sorted LRO, and the depth of the mbuf queue. */ 311 static int vtnet_lro_mbufq_depth = 0; 312 SYSCTL_UINT(_hw_vtnet, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN, 313 &vtnet_lro_mbufq_depth, 0, "Depth of software LRO mbuf queue"); 314 315 /* Deactivate ALTQ Support */ 316 static int vtnet_altq_disable = 0; 317 SYSCTL_INT(_hw_vtnet, OID_AUTO, altq_disable, CTLFLAG_RDTUN, 318 &vtnet_altq_disable, 0, "Disables ALTQ Support"); 319 320 /* 321 * For the driver to be considered as having altq enabled, 322 * it must be compiled with an ALTQ capable kernel, 323 * and the tunable hw.vtnet.altq_disable must be zero 324 */ 325 #define VTNET_ALTQ_ENABLED (VTNET_ALTQ_CAPABLE && (!vtnet_altq_disable)) 326 327 328 static uma_zone_t vtnet_tx_header_zone; 329 330 static struct virtio_feature_desc vtnet_feature_desc[] = { 331 { VIRTIO_NET_F_CSUM, "TxChecksum" }, 332 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" }, 333 { VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, "CtrlRxOffloads" }, 334 { VIRTIO_NET_F_MAC, "MAC" }, 335 { VIRTIO_NET_F_GSO, "TxGSO" }, 336 { VIRTIO_NET_F_GUEST_TSO4, "RxLROv4" }, 337 { VIRTIO_NET_F_GUEST_TSO6, "RxLROv6" }, 338 { VIRTIO_NET_F_GUEST_ECN, "RxLROECN" }, 339 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" }, 340 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" }, 341 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" }, 342 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" }, 343 { VIRTIO_NET_F_HOST_UFO, "TxUFO" }, 344 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" }, 345 { VIRTIO_NET_F_STATUS, "Status" }, 346 { VIRTIO_NET_F_CTRL_VQ, "CtrlVq" }, 347 { VIRTIO_NET_F_CTRL_RX, "CtrlRxMode" }, 348 { VIRTIO_NET_F_CTRL_VLAN, "CtrlVLANFilter" }, 349 { VIRTIO_NET_F_CTRL_RX_EXTRA, "CtrlRxModeExtra" }, 350 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" }, 351 { VIRTIO_NET_F_MQ, "Multiqueue" }, 352 { VIRTIO_NET_F_CTRL_MAC_ADDR, "CtrlMacAddr" }, 353 { VIRTIO_NET_F_SPEED_DUPLEX, "SpeedDuplex" }, 354 355 { 0, NULL } 356 }; 357 358 static device_method_t vtnet_methods[] = { 359 /* Device methods. */ 360 DEVMETHOD(device_probe, vtnet_probe), 361 DEVMETHOD(device_attach, vtnet_attach), 362 DEVMETHOD(device_detach, vtnet_detach), 363 DEVMETHOD(device_suspend, vtnet_suspend), 364 DEVMETHOD(device_resume, vtnet_resume), 365 DEVMETHOD(device_shutdown, vtnet_shutdown), 366 367 /* VirtIO methods. */ 368 DEVMETHOD(virtio_attach_completed, vtnet_attach_completed), 369 DEVMETHOD(virtio_config_change, vtnet_config_change), 370 371 DEVMETHOD_END 372 }; 373 374 #ifdef DEV_NETMAP 375 #include <dev/netmap/if_vtnet_netmap.h> 376 #endif 377 378 static driver_t vtnet_driver = { 379 .name = "vtnet", 380 .methods = vtnet_methods, 381 .size = sizeof(struct vtnet_softc) 382 }; 383 VIRTIO_DRIVER_MODULE(vtnet, vtnet_driver, vtnet_modevent, NULL); 384 MODULE_VERSION(vtnet, 1); 385 MODULE_DEPEND(vtnet, virtio, 1, 1, 1); 386 #ifdef DEV_NETMAP 387 MODULE_DEPEND(vtnet, netmap, 1, 1, 1); 388 #endif 389 390 VIRTIO_SIMPLE_PNPINFO(vtnet, VIRTIO_ID_NETWORK, "VirtIO Networking Adapter"); 391 392 static int 393 vtnet_modevent(module_t mod __unused, int type, void *unused __unused) 394 { 395 int error = 0; 396 static int loaded = 0; 397 398 switch (type) { 399 case MOD_LOAD: 400 if (loaded++ == 0) { 401 vtnet_tx_header_zone = uma_zcreate("vtnet_tx_hdr", 402 sizeof(struct vtnet_tx_header), 403 NULL, NULL, NULL, NULL, 0, 0); 404 #ifdef DEBUGNET 405 /* 406 * We need to allocate from this zone in the transmit path, so ensure 407 * that we have at least one item per header available. 408 * XXX add a separate zone like we do for mbufs? otherwise we may alloc 409 * buckets 410 */ 411 uma_zone_reserve(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 412 uma_prealloc(vtnet_tx_header_zone, DEBUGNET_MAX_IN_FLIGHT * 2); 413 #endif 414 } 415 break; 416 case MOD_QUIESCE: 417 if (uma_zone_get_cur(vtnet_tx_header_zone) > 0) 418 error = EBUSY; 419 break; 420 case MOD_UNLOAD: 421 if (--loaded == 0) { 422 uma_zdestroy(vtnet_tx_header_zone); 423 vtnet_tx_header_zone = NULL; 424 } 425 break; 426 case MOD_SHUTDOWN: 427 break; 428 default: 429 error = EOPNOTSUPP; 430 break; 431 } 432 433 return (error); 434 } 435 436 static int 437 vtnet_probe(device_t dev) 438 { 439 return (VIRTIO_SIMPLE_PROBE(dev, vtnet)); 440 } 441 442 static int 443 vtnet_attach(device_t dev) 444 { 445 struct vtnet_softc *sc; 446 int error; 447 448 sc = device_get_softc(dev); 449 sc->vtnet_dev = dev; 450 virtio_set_feature_desc(dev, vtnet_feature_desc); 451 452 VTNET_CORE_LOCK_INIT(sc); 453 callout_init_mtx(&sc->vtnet_tick_ch, VTNET_CORE_MTX(sc), 0); 454 vtnet_load_tunables(sc); 455 456 vtnet_alloc_interface(sc); 457 vtnet_setup_sysctl(sc); 458 459 error = vtnet_setup_features(sc); 460 if (error) { 461 device_printf(dev, "cannot setup features\n"); 462 goto fail; 463 } 464 465 error = vtnet_alloc_rx_filters(sc); 466 if (error) { 467 device_printf(dev, "cannot allocate Rx filters\n"); 468 goto fail; 469 } 470 471 error = vtnet_alloc_rxtx_queues(sc); 472 if (error) { 473 device_printf(dev, "cannot allocate queues\n"); 474 goto fail; 475 } 476 477 error = vtnet_alloc_virtqueues(sc); 478 if (error) { 479 device_printf(dev, "cannot allocate virtqueues\n"); 480 goto fail; 481 } 482 483 error = vtnet_setup_interface(sc); 484 if (error) { 485 device_printf(dev, "cannot setup interface\n"); 486 goto fail; 487 } 488 489 error = virtio_setup_intr(dev, INTR_TYPE_NET); 490 if (error) { 491 device_printf(dev, "cannot setup interrupts\n"); 492 ether_ifdetach(sc->vtnet_ifp); 493 goto fail; 494 } 495 496 #ifdef DEV_NETMAP 497 vtnet_netmap_attach(sc); 498 #endif 499 vtnet_start_taskqueues(sc); 500 501 fail: 502 if (error) 503 vtnet_detach(dev); 504 505 return (error); 506 } 507 508 static int 509 vtnet_detach(device_t dev) 510 { 511 struct vtnet_softc *sc; 512 if_t ifp; 513 514 sc = device_get_softc(dev); 515 ifp = sc->vtnet_ifp; 516 517 if (device_is_attached(dev)) { 518 VTNET_CORE_LOCK(sc); 519 vtnet_stop(sc); 520 VTNET_CORE_UNLOCK(sc); 521 522 callout_drain(&sc->vtnet_tick_ch); 523 vtnet_drain_taskqueues(sc); 524 525 ether_ifdetach(ifp); 526 } 527 528 #ifdef DEV_NETMAP 529 netmap_detach(ifp); 530 #endif 531 532 if (sc->vtnet_pfil != NULL) { 533 pfil_head_unregister(sc->vtnet_pfil); 534 sc->vtnet_pfil = NULL; 535 } 536 537 vtnet_free_taskqueues(sc); 538 539 if (sc->vtnet_vlan_attach != NULL) { 540 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach); 541 sc->vtnet_vlan_attach = NULL; 542 } 543 if (sc->vtnet_vlan_detach != NULL) { 544 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach); 545 sc->vtnet_vlan_detach = NULL; 546 } 547 548 ifmedia_removeall(&sc->vtnet_media); 549 550 if (ifp != NULL) { 551 if_free(ifp); 552 sc->vtnet_ifp = NULL; 553 } 554 555 vtnet_free_rxtx_queues(sc); 556 vtnet_free_rx_filters(sc); 557 558 if (sc->vtnet_ctrl_vq != NULL) 559 vtnet_free_ctrl_vq(sc); 560 561 VTNET_CORE_LOCK_DESTROY(sc); 562 563 return (0); 564 } 565 566 static int 567 vtnet_suspend(device_t dev) 568 { 569 struct vtnet_softc *sc; 570 571 sc = device_get_softc(dev); 572 573 VTNET_CORE_LOCK(sc); 574 vtnet_stop(sc); 575 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED; 576 VTNET_CORE_UNLOCK(sc); 577 578 return (0); 579 } 580 581 static int 582 vtnet_resume(device_t dev) 583 { 584 struct vtnet_softc *sc; 585 if_t ifp; 586 587 sc = device_get_softc(dev); 588 ifp = sc->vtnet_ifp; 589 590 VTNET_CORE_LOCK(sc); 591 if (if_getflags(ifp) & IFF_UP) 592 vtnet_init_locked(sc, 0); 593 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED; 594 VTNET_CORE_UNLOCK(sc); 595 596 return (0); 597 } 598 599 static int 600 vtnet_shutdown(device_t dev) 601 { 602 /* 603 * Suspend already does all of what we need to 604 * do here; we just never expect to be resumed. 605 */ 606 return (vtnet_suspend(dev)); 607 } 608 609 static int 610 vtnet_attach_completed(device_t dev) 611 { 612 struct vtnet_softc *sc; 613 614 sc = device_get_softc(dev); 615 616 VTNET_CORE_LOCK(sc); 617 vtnet_attached_set_macaddr(sc); 618 VTNET_CORE_UNLOCK(sc); 619 620 return (0); 621 } 622 623 static int 624 vtnet_config_change(device_t dev) 625 { 626 struct vtnet_softc *sc; 627 628 sc = device_get_softc(dev); 629 630 VTNET_CORE_LOCK(sc); 631 vtnet_update_link_status(sc); 632 if (sc->vtnet_link_active != 0) 633 vtnet_tx_start_all(sc); 634 VTNET_CORE_UNLOCK(sc); 635 636 return (0); 637 } 638 639 static int 640 vtnet_negotiate_features(struct vtnet_softc *sc) 641 { 642 device_t dev; 643 uint64_t features, negotiated_features; 644 int no_csum; 645 646 dev = sc->vtnet_dev; 647 features = virtio_bus_is_modern(dev) ? VTNET_MODERN_FEATURES : 648 VTNET_LEGACY_FEATURES; 649 650 /* 651 * TSO and LRO are only available when their corresponding checksum 652 * offload feature is also negotiated. 653 */ 654 no_csum = vtnet_tunable_int(sc, "csum_disable", vtnet_csum_disable); 655 if (no_csum) 656 features &= ~(VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM); 657 if (no_csum || vtnet_tunable_int(sc, "tso_disable", vtnet_tso_disable)) 658 features &= ~VTNET_TSO_FEATURES; 659 if (no_csum || vtnet_tunable_int(sc, "lro_disable", vtnet_lro_disable)) 660 features &= ~VTNET_LRO_FEATURES; 661 662 /* Deactivate MQ Feature flag, if driver has ALTQ enabled, or MQ is explicitly disabled */ 663 if (VTNET_ALTQ_ENABLED || vtnet_tunable_int(sc, "mq_disable", vtnet_mq_disable)) 664 features &= ~VIRTIO_NET_F_MQ; 665 666 negotiated_features = virtio_negotiate_features(dev, features); 667 668 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) { 669 uint16_t mtu; 670 671 mtu = virtio_read_dev_config_2(dev, 672 offsetof(struct virtio_net_config, mtu)); 673 if (mtu < VTNET_MIN_MTU) { 674 device_printf(dev, "Invalid MTU value: %d. " 675 "MTU feature disabled.\n", mtu); 676 features &= ~VIRTIO_NET_F_MTU; 677 negotiated_features = 678 virtio_negotiate_features(dev, features); 679 } 680 } 681 682 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) { 683 uint16_t npairs; 684 685 npairs = virtio_read_dev_config_2(dev, 686 offsetof(struct virtio_net_config, max_virtqueue_pairs)); 687 if (npairs < VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN || 688 npairs > VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX) { 689 device_printf(dev, "Invalid max_virtqueue_pairs value: " 690 "%d. Multiqueue feature disabled.\n", npairs); 691 features &= ~VIRTIO_NET_F_MQ; 692 negotiated_features = 693 virtio_negotiate_features(dev, features); 694 } 695 } 696 697 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) && 698 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) { 699 /* 700 * LRO without mergeable buffers requires special care. This 701 * is not ideal because every receive buffer must be large 702 * enough to hold the maximum TCP packet, the Ethernet header, 703 * and the header. This requires up to 34 descriptors with 704 * MCLBYTES clusters. If we do not have indirect descriptors, 705 * LRO is disabled since the virtqueue will not contain very 706 * many receive buffers. 707 */ 708 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) { 709 device_printf(dev, 710 "Host LRO disabled since both mergeable buffers " 711 "and indirect descriptors were not negotiated\n"); 712 features &= ~VTNET_LRO_FEATURES; 713 negotiated_features = 714 virtio_negotiate_features(dev, features); 715 } else 716 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG; 717 } 718 719 sc->vtnet_features = negotiated_features; 720 sc->vtnet_negotiated_features = negotiated_features; 721 722 return (virtio_finalize_features(dev)); 723 } 724 725 static int 726 vtnet_setup_features(struct vtnet_softc *sc) 727 { 728 device_t dev; 729 int error; 730 731 dev = sc->vtnet_dev; 732 733 error = vtnet_negotiate_features(sc); 734 if (error) 735 return (error); 736 737 if (virtio_with_feature(dev, VIRTIO_F_VERSION_1)) 738 sc->vtnet_flags |= VTNET_FLAG_MODERN; 739 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) 740 sc->vtnet_flags |= VTNET_FLAG_INDIRECT; 741 if (virtio_with_feature(dev, VIRTIO_RING_F_EVENT_IDX)) 742 sc->vtnet_flags |= VTNET_FLAG_EVENT_IDX; 743 744 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) { 745 /* This feature should always be negotiated. */ 746 sc->vtnet_flags |= VTNET_FLAG_MAC; 747 } 748 749 if (virtio_with_feature(dev, VIRTIO_NET_F_MTU)) { 750 sc->vtnet_max_mtu = virtio_read_dev_config_2(dev, 751 offsetof(struct virtio_net_config, mtu)); 752 } else 753 sc->vtnet_max_mtu = VTNET_MAX_MTU; 754 755 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) { 756 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS; 757 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf); 758 } else if (vtnet_modern(sc)) { 759 /* This is identical to the mergeable header. */ 760 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_v1); 761 } else 762 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr); 763 764 if (vtnet_modern(sc) || sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 765 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_INLINE; 766 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 767 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_LRO_NOMRG; 768 else 769 sc->vtnet_rx_nsegs = VTNET_RX_SEGS_HDR_SEPARATE; 770 771 /* 772 * Favor "hardware" LRO if negotiated, but support software LRO as 773 * a fallback; there is usually little benefit (or worse) with both. 774 */ 775 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) == 0 && 776 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6) == 0) 777 sc->vtnet_flags |= VTNET_FLAG_SW_LRO; 778 779 if (virtio_with_feature(dev, VIRTIO_NET_F_GSO) || 780 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) || 781 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 782 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MAX; 783 else 784 sc->vtnet_tx_nsegs = VTNET_TX_SEGS_MIN; 785 786 sc->vtnet_req_vq_pairs = 1; 787 sc->vtnet_max_vq_pairs = 1; 788 789 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) { 790 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ; 791 792 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX)) 793 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX; 794 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN)) 795 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER; 796 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR)) 797 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC; 798 799 if (virtio_with_feature(dev, VIRTIO_NET_F_MQ)) { 800 sc->vtnet_max_vq_pairs = virtio_read_dev_config_2(dev, 801 offsetof(struct virtio_net_config, 802 max_virtqueue_pairs)); 803 } 804 } 805 806 if (sc->vtnet_max_vq_pairs > 1) { 807 int req; 808 809 /* 810 * Limit the maximum number of requested queue pairs to the 811 * number of CPUs and the configured maximum. 812 */ 813 req = vtnet_tunable_int(sc, "mq_max_pairs", vtnet_mq_max_pairs); 814 if (req < 0) 815 req = 1; 816 if (req == 0) 817 req = mp_ncpus; 818 if (req > sc->vtnet_max_vq_pairs) 819 req = sc->vtnet_max_vq_pairs; 820 if (req > mp_ncpus) 821 req = mp_ncpus; 822 if (req > 1) { 823 sc->vtnet_req_vq_pairs = req; 824 sc->vtnet_flags |= VTNET_FLAG_MQ; 825 } 826 } 827 828 return (0); 829 } 830 831 static int 832 vtnet_init_rxq(struct vtnet_softc *sc, int id) 833 { 834 struct vtnet_rxq *rxq; 835 836 rxq = &sc->vtnet_rxqs[id]; 837 838 snprintf(rxq->vtnrx_name, sizeof(rxq->vtnrx_name), "%s-rx%d", 839 device_get_nameunit(sc->vtnet_dev), id); 840 mtx_init(&rxq->vtnrx_mtx, rxq->vtnrx_name, NULL, MTX_DEF); 841 842 rxq->vtnrx_sc = sc; 843 rxq->vtnrx_id = id; 844 845 rxq->vtnrx_sg = sglist_alloc(sc->vtnet_rx_nsegs, M_NOWAIT); 846 if (rxq->vtnrx_sg == NULL) 847 return (ENOMEM); 848 849 #if defined(INET) || defined(INET6) 850 if (vtnet_software_lro(sc)) { 851 if (tcp_lro_init_args(&rxq->vtnrx_lro, sc->vtnet_ifp, 852 sc->vtnet_lro_entry_count, sc->vtnet_lro_mbufq_depth) != 0) 853 return (ENOMEM); 854 } 855 #endif 856 857 NET_TASK_INIT(&rxq->vtnrx_intrtask, 0, vtnet_rxq_tq_intr, rxq); 858 rxq->vtnrx_tq = taskqueue_create(rxq->vtnrx_name, M_NOWAIT, 859 taskqueue_thread_enqueue, &rxq->vtnrx_tq); 860 861 return (rxq->vtnrx_tq == NULL ? ENOMEM : 0); 862 } 863 864 static int 865 vtnet_init_txq(struct vtnet_softc *sc, int id) 866 { 867 struct vtnet_txq *txq; 868 869 txq = &sc->vtnet_txqs[id]; 870 871 snprintf(txq->vtntx_name, sizeof(txq->vtntx_name), "%s-tx%d", 872 device_get_nameunit(sc->vtnet_dev), id); 873 mtx_init(&txq->vtntx_mtx, txq->vtntx_name, NULL, MTX_DEF); 874 875 txq->vtntx_sc = sc; 876 txq->vtntx_id = id; 877 878 txq->vtntx_sg = sglist_alloc(sc->vtnet_tx_nsegs, M_NOWAIT); 879 if (txq->vtntx_sg == NULL) 880 return (ENOMEM); 881 882 if (!VTNET_ALTQ_ENABLED) { 883 txq->vtntx_br = buf_ring_alloc(VTNET_DEFAULT_BUFRING_SIZE, M_DEVBUF, 884 M_NOWAIT, &txq->vtntx_mtx); 885 if (txq->vtntx_br == NULL) 886 return (ENOMEM); 887 888 TASK_INIT(&txq->vtntx_defrtask, 0, vtnet_txq_tq_deferred, txq); 889 } 890 TASK_INIT(&txq->vtntx_intrtask, 0, vtnet_txq_tq_intr, txq); 891 txq->vtntx_tq = taskqueue_create(txq->vtntx_name, M_NOWAIT, 892 taskqueue_thread_enqueue, &txq->vtntx_tq); 893 if (txq->vtntx_tq == NULL) 894 return (ENOMEM); 895 896 return (0); 897 } 898 899 static int 900 vtnet_alloc_rxtx_queues(struct vtnet_softc *sc) 901 { 902 int i, npairs, error; 903 904 npairs = sc->vtnet_max_vq_pairs; 905 906 sc->vtnet_rxqs = malloc(sizeof(struct vtnet_rxq) * npairs, M_DEVBUF, 907 M_NOWAIT | M_ZERO); 908 sc->vtnet_txqs = malloc(sizeof(struct vtnet_txq) * npairs, M_DEVBUF, 909 M_NOWAIT | M_ZERO); 910 if (sc->vtnet_rxqs == NULL || sc->vtnet_txqs == NULL) 911 return (ENOMEM); 912 913 for (i = 0; i < npairs; i++) { 914 error = vtnet_init_rxq(sc, i); 915 if (error) 916 return (error); 917 error = vtnet_init_txq(sc, i); 918 if (error) 919 return (error); 920 } 921 922 vtnet_set_rx_process_limit(sc); 923 vtnet_setup_queue_sysctl(sc); 924 925 return (0); 926 } 927 928 static void 929 vtnet_destroy_rxq(struct vtnet_rxq *rxq) 930 { 931 932 rxq->vtnrx_sc = NULL; 933 rxq->vtnrx_id = -1; 934 935 #if defined(INET) || defined(INET6) 936 tcp_lro_free(&rxq->vtnrx_lro); 937 #endif 938 939 if (rxq->vtnrx_sg != NULL) { 940 sglist_free(rxq->vtnrx_sg); 941 rxq->vtnrx_sg = NULL; 942 } 943 944 if (mtx_initialized(&rxq->vtnrx_mtx) != 0) 945 mtx_destroy(&rxq->vtnrx_mtx); 946 } 947 948 static void 949 vtnet_destroy_txq(struct vtnet_txq *txq) 950 { 951 952 txq->vtntx_sc = NULL; 953 txq->vtntx_id = -1; 954 955 if (txq->vtntx_sg != NULL) { 956 sglist_free(txq->vtntx_sg); 957 txq->vtntx_sg = NULL; 958 } 959 960 if (!VTNET_ALTQ_ENABLED) { 961 if (txq->vtntx_br != NULL) { 962 buf_ring_free(txq->vtntx_br, M_DEVBUF); 963 txq->vtntx_br = NULL; 964 } 965 } 966 967 if (mtx_initialized(&txq->vtntx_mtx) != 0) 968 mtx_destroy(&txq->vtntx_mtx); 969 } 970 971 static void 972 vtnet_free_rxtx_queues(struct vtnet_softc *sc) 973 { 974 int i; 975 976 if (sc->vtnet_rxqs != NULL) { 977 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 978 vtnet_destroy_rxq(&sc->vtnet_rxqs[i]); 979 free(sc->vtnet_rxqs, M_DEVBUF); 980 sc->vtnet_rxqs = NULL; 981 } 982 983 if (sc->vtnet_txqs != NULL) { 984 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 985 vtnet_destroy_txq(&sc->vtnet_txqs[i]); 986 free(sc->vtnet_txqs, M_DEVBUF); 987 sc->vtnet_txqs = NULL; 988 } 989 } 990 991 static int 992 vtnet_alloc_rx_filters(struct vtnet_softc *sc) 993 { 994 995 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 996 sc->vtnet_mac_filter = malloc(sizeof(struct vtnet_mac_filter), 997 M_DEVBUF, M_NOWAIT | M_ZERO); 998 if (sc->vtnet_mac_filter == NULL) 999 return (ENOMEM); 1000 } 1001 1002 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 1003 sc->vtnet_vlan_filter = malloc(sizeof(uint32_t) * 1004 VTNET_VLAN_FILTER_NWORDS, M_DEVBUF, M_NOWAIT | M_ZERO); 1005 if (sc->vtnet_vlan_filter == NULL) 1006 return (ENOMEM); 1007 } 1008 1009 return (0); 1010 } 1011 1012 static void 1013 vtnet_free_rx_filters(struct vtnet_softc *sc) 1014 { 1015 1016 if (sc->vtnet_mac_filter != NULL) { 1017 free(sc->vtnet_mac_filter, M_DEVBUF); 1018 sc->vtnet_mac_filter = NULL; 1019 } 1020 1021 if (sc->vtnet_vlan_filter != NULL) { 1022 free(sc->vtnet_vlan_filter, M_DEVBUF); 1023 sc->vtnet_vlan_filter = NULL; 1024 } 1025 } 1026 1027 static int 1028 vtnet_alloc_virtqueues(struct vtnet_softc *sc) 1029 { 1030 device_t dev; 1031 struct vq_alloc_info *info; 1032 struct vtnet_rxq *rxq; 1033 struct vtnet_txq *txq; 1034 int i, idx, nvqs, error; 1035 1036 dev = sc->vtnet_dev; 1037 1038 nvqs = sc->vtnet_max_vq_pairs * 2; 1039 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 1040 nvqs++; 1041 1042 info = malloc(sizeof(struct vq_alloc_info) * nvqs, M_TEMP, M_NOWAIT); 1043 if (info == NULL) 1044 return (ENOMEM); 1045 1046 for (i = 0, idx = 0; i < sc->vtnet_req_vq_pairs; i++, idx += 2) { 1047 rxq = &sc->vtnet_rxqs[i]; 1048 VQ_ALLOC_INFO_INIT(&info[idx], sc->vtnet_rx_nsegs, 1049 vtnet_rx_vq_intr, rxq, &rxq->vtnrx_vq, 1050 "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id); 1051 1052 txq = &sc->vtnet_txqs[i]; 1053 VQ_ALLOC_INFO_INIT(&info[idx + 1], sc->vtnet_tx_nsegs, 1054 vtnet_tx_vq_intr, txq, &txq->vtntx_vq, 1055 "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id); 1056 } 1057 1058 /* These queues will not be used so allocate the minimum resources. */ 1059 for (; i < sc->vtnet_max_vq_pairs; i++, idx += 2) { 1060 rxq = &sc->vtnet_rxqs[i]; 1061 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, rxq, &rxq->vtnrx_vq, 1062 "%s-rx%d", device_get_nameunit(dev), rxq->vtnrx_id); 1063 1064 txq = &sc->vtnet_txqs[i]; 1065 VQ_ALLOC_INFO_INIT(&info[idx + 1], 0, NULL, txq, &txq->vtntx_vq, 1066 "%s-tx%d", device_get_nameunit(dev), txq->vtntx_id); 1067 } 1068 1069 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) { 1070 VQ_ALLOC_INFO_INIT(&info[idx], 0, NULL, NULL, 1071 &sc->vtnet_ctrl_vq, "%s ctrl", device_get_nameunit(dev)); 1072 } 1073 1074 error = virtio_alloc_virtqueues(dev, nvqs, info); 1075 free(info, M_TEMP); 1076 1077 return (error); 1078 } 1079 1080 static void 1081 vtnet_alloc_interface(struct vtnet_softc *sc) 1082 { 1083 device_t dev; 1084 if_t ifp; 1085 1086 dev = sc->vtnet_dev; 1087 1088 ifp = if_alloc(IFT_ETHER); 1089 sc->vtnet_ifp = ifp; 1090 if_setsoftc(ifp, sc); 1091 if_initname(ifp, device_get_name(dev), device_get_unit(dev)); 1092 } 1093 1094 static int 1095 vtnet_setup_interface(struct vtnet_softc *sc) 1096 { 1097 device_t dev; 1098 struct pfil_head_args pa; 1099 if_t ifp; 1100 1101 dev = sc->vtnet_dev; 1102 ifp = sc->vtnet_ifp; 1103 1104 if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST); 1105 if_setbaudrate(ifp, IF_Gbps(10)); 1106 if_setinitfn(ifp, vtnet_init); 1107 if_setioctlfn(ifp, vtnet_ioctl); 1108 if_setgetcounterfn(ifp, vtnet_get_counter); 1109 1110 if (!VTNET_ALTQ_ENABLED) { 1111 if_settransmitfn(ifp, vtnet_txq_mq_start); 1112 if_setqflushfn(ifp, vtnet_qflush); 1113 } else { 1114 struct virtqueue *vq = sc->vtnet_txqs[0].vtntx_vq; 1115 if_setstartfn(ifp, vtnet_start); 1116 if_setsendqlen(ifp, virtqueue_size(vq) - 1); 1117 if_setsendqready(ifp); 1118 } 1119 1120 vtnet_get_macaddr(sc); 1121 1122 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) 1123 if_setcapabilitiesbit(ifp, IFCAP_LINKSTATE, 0); 1124 1125 ifmedia_init(&sc->vtnet_media, 0, vtnet_ifmedia_upd, vtnet_ifmedia_sts); 1126 ifmedia_add(&sc->vtnet_media, IFM_ETHER | IFM_AUTO, 0, NULL); 1127 ifmedia_set(&sc->vtnet_media, IFM_ETHER | IFM_AUTO); 1128 1129 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) { 1130 int gso; 1131 1132 if_setcapabilitiesbit(ifp, IFCAP_TXCSUM | IFCAP_TXCSUM_IPV6, 0); 1133 1134 gso = virtio_with_feature(dev, VIRTIO_NET_F_GSO); 1135 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4)) 1136 if_setcapabilitiesbit(ifp, IFCAP_TSO4, 0); 1137 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6)) 1138 if_setcapabilitiesbit(ifp, IFCAP_TSO6, 0); 1139 if (gso || virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN)) 1140 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN; 1141 1142 if (if_getcapabilities(ifp) & (IFCAP_TSO4 | IFCAP_TSO6)) { 1143 int tso_maxlen; 1144 1145 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTSO, 0); 1146 1147 tso_maxlen = vtnet_tunable_int(sc, "tso_maxlen", 1148 vtnet_tso_maxlen); 1149 if_sethwtsomax(ifp, tso_maxlen - 1150 (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN)); 1151 if_sethwtsomaxsegcount(ifp, sc->vtnet_tx_nsegs - 1); 1152 if_sethwtsomaxsegsize(ifp, PAGE_SIZE); 1153 } 1154 } 1155 1156 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) { 1157 /* BMV: Rx checksums not distinguished between IPv4 and IPv6. */ 1158 if_setcapabilitiesbit(ifp, IFCAP_RXCSUM, 0); 1159 if_setcapabilitiesbit(ifp, IFCAP_RXCSUM_IPV6, 0); 1160 1161 if (vtnet_tunable_int(sc, "fixup_needs_csum", 1162 vtnet_fixup_needs_csum) != 0) 1163 sc->vtnet_flags |= VTNET_FLAG_FIXUP_NEEDS_CSUM; 1164 1165 /* Support either "hardware" or software LRO. */ 1166 if_setcapabilitiesbit(ifp, IFCAP_LRO, 0); 1167 } 1168 1169 if (if_getcapabilities(ifp) & (IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6)) { 1170 /* 1171 * VirtIO does not support VLAN tagging, but we can fake 1172 * it by inserting and removing the 802.1Q header during 1173 * transmit and receive. We are then able to do checksum 1174 * offloading of VLAN frames. 1175 */ 1176 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM, 0); 1177 } 1178 1179 if (sc->vtnet_max_mtu >= ETHERMTU_JUMBO) 1180 if_setcapabilitiesbit(ifp, IFCAP_JUMBO_MTU, 0); 1181 if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU, 0); 1182 if_setcapabilitiesbit(ifp, IFCAP_HWSTATS, 0); 1183 1184 /* 1185 * Capabilities after here are not enabled by default. 1186 */ 1187 if_setcapenable(ifp, if_getcapabilities(ifp)); 1188 1189 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) { 1190 if_setcapabilitiesbit(ifp, IFCAP_VLAN_HWFILTER, 0); 1191 1192 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config, 1193 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST); 1194 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, 1195 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST); 1196 } 1197 1198 ether_ifattach(ifp, sc->vtnet_hwaddr); 1199 1200 /* Tell the upper layer(s) we support long frames. */ 1201 if_setifheaderlen(ifp, sizeof(struct ether_vlan_header)); 1202 1203 DEBUGNET_SET(ifp, vtnet); 1204 1205 pa.pa_version = PFIL_VERSION; 1206 pa.pa_flags = PFIL_IN; 1207 pa.pa_type = PFIL_TYPE_ETHERNET; 1208 pa.pa_headname = if_name(ifp); 1209 sc->vtnet_pfil = pfil_head_register(&pa); 1210 1211 return (0); 1212 } 1213 1214 static int 1215 vtnet_rx_cluster_size(struct vtnet_softc *sc, int mtu) 1216 { 1217 int framesz; 1218 1219 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) 1220 return (MJUMPAGESIZE); 1221 else if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 1222 return (MCLBYTES); 1223 1224 /* 1225 * Try to scale the receive mbuf cluster size from the MTU. We 1226 * could also use the VQ size to influence the selected size, 1227 * but that would only matter for very small queues. 1228 */ 1229 if (vtnet_modern(sc)) { 1230 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr_v1)); 1231 framesz = sizeof(struct virtio_net_hdr_v1); 1232 } else 1233 framesz = sizeof(struct vtnet_rx_header); 1234 framesz += sizeof(struct ether_vlan_header) + mtu; 1235 /* 1236 * Account for the offsetting we'll do elsewhere so we allocate the 1237 * right size for the mtu. 1238 */ 1239 if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) { 1240 framesz += VTNET_ETHER_ALIGN; 1241 } 1242 1243 if (framesz <= MCLBYTES) 1244 return (MCLBYTES); 1245 else if (framesz <= MJUMPAGESIZE) 1246 return (MJUMPAGESIZE); 1247 else if (framesz <= MJUM9BYTES) 1248 return (MJUM9BYTES); 1249 1250 /* Sane default; avoid 16KB clusters. */ 1251 return (MCLBYTES); 1252 } 1253 1254 static int 1255 vtnet_ioctl_mtu(struct vtnet_softc *sc, u_int mtu) 1256 { 1257 if_t ifp; 1258 int clustersz; 1259 1260 ifp = sc->vtnet_ifp; 1261 VTNET_CORE_LOCK_ASSERT(sc); 1262 1263 if (if_getmtu(ifp) == mtu) 1264 return (0); 1265 else if (mtu < ETHERMIN || mtu > sc->vtnet_max_mtu) 1266 return (EINVAL); 1267 1268 if_setmtu(ifp, mtu); 1269 clustersz = vtnet_rx_cluster_size(sc, mtu); 1270 1271 if (clustersz != sc->vtnet_rx_clustersz && 1272 if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 1273 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 1274 vtnet_init_locked(sc, 0); 1275 } 1276 1277 return (0); 1278 } 1279 1280 static int 1281 vtnet_ioctl_ifflags(struct vtnet_softc *sc) 1282 { 1283 if_t ifp; 1284 int drv_running; 1285 1286 ifp = sc->vtnet_ifp; 1287 drv_running = (if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0; 1288 1289 VTNET_CORE_LOCK_ASSERT(sc); 1290 1291 if ((if_getflags(ifp) & IFF_UP) == 0) { 1292 if (drv_running) 1293 vtnet_stop(sc); 1294 goto out; 1295 } 1296 1297 if (!drv_running) { 1298 vtnet_init_locked(sc, 0); 1299 goto out; 1300 } 1301 1302 if ((if_getflags(ifp) ^ sc->vtnet_if_flags) & 1303 (IFF_PROMISC | IFF_ALLMULTI)) { 1304 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) 1305 vtnet_rx_filter(sc); 1306 else { 1307 /* 1308 * We don't support filtering out multicast, so 1309 * ALLMULTI is always set. 1310 */ 1311 if_setflagbits(ifp, IFF_ALLMULTI, 0); 1312 if_setflagbits(ifp, IFF_PROMISC, 0); 1313 } 1314 } 1315 1316 out: 1317 sc->vtnet_if_flags = if_getflags(ifp); 1318 return (0); 1319 } 1320 1321 static int 1322 vtnet_ioctl_multi(struct vtnet_softc *sc) 1323 { 1324 if_t ifp; 1325 1326 ifp = sc->vtnet_ifp; 1327 1328 VTNET_CORE_LOCK_ASSERT(sc); 1329 1330 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX && 1331 if_getdrvflags(ifp) & IFF_DRV_RUNNING) 1332 vtnet_rx_filter_mac(sc); 1333 1334 return (0); 1335 } 1336 1337 static int 1338 vtnet_ioctl_ifcap(struct vtnet_softc *sc, struct ifreq *ifr) 1339 { 1340 if_t ifp; 1341 int mask, reinit, update; 1342 1343 ifp = sc->vtnet_ifp; 1344 mask = (ifr->ifr_reqcap & if_getcapabilities(ifp)) ^ if_getcapenable(ifp); 1345 reinit = update = 0; 1346 1347 VTNET_CORE_LOCK_ASSERT(sc); 1348 1349 if (mask & IFCAP_TXCSUM) { 1350 if (if_getcapenable(ifp) & IFCAP_TXCSUM && 1351 if_getcapenable(ifp) & IFCAP_TSO4) { 1352 /* Disable tso4, because txcsum will be disabled. */ 1353 if_setcapenablebit(ifp, 0, IFCAP_TSO4); 1354 if_sethwassistbits(ifp, 0, CSUM_IP_TSO); 1355 mask &= ~IFCAP_TSO4; 1356 } 1357 if_togglecapenable(ifp, IFCAP_TXCSUM); 1358 if_togglehwassist(ifp, VTNET_CSUM_OFFLOAD); 1359 } 1360 if (mask & IFCAP_TXCSUM_IPV6) { 1361 if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6 && 1362 if_getcapenable(ifp) & IFCAP_TSO6) { 1363 /* Disable tso6, because txcsum6 will be disabled. */ 1364 if_setcapenablebit(ifp, 0, IFCAP_TSO6); 1365 if_sethwassistbits(ifp, 0, CSUM_IP6_TSO); 1366 mask &= ~IFCAP_TSO6; 1367 } 1368 if_togglecapenable(ifp, IFCAP_TXCSUM_IPV6); 1369 if_togglehwassist(ifp, VTNET_CSUM_OFFLOAD_IPV6); 1370 } 1371 if (mask & IFCAP_TSO4) { 1372 if (if_getcapenable(ifp) & (IFCAP_TXCSUM | IFCAP_TSO4)) { 1373 /* tso4 can only be enabled, if txcsum is enabled. */ 1374 if_togglecapenable(ifp, IFCAP_TSO4); 1375 if_togglehwassist(ifp, CSUM_IP_TSO); 1376 } 1377 } 1378 if (mask & IFCAP_TSO6) { 1379 if (if_getcapenable(ifp) & (IFCAP_TXCSUM_IPV6 | IFCAP_TSO6)) { 1380 /* tso6 can only be enabled, if txcsum6 is enabled. */ 1381 if_togglecapenable(ifp, IFCAP_TSO6); 1382 if_togglehwassist(ifp, CSUM_IP6_TSO); 1383 } 1384 } 1385 1386 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) { 1387 /* 1388 * These Rx features require the negotiated features to 1389 * be updated. Avoid a full reinit if possible. 1390 */ 1391 if (sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS) 1392 update = 1; 1393 else 1394 reinit = 1; 1395 1396 /* BMV: Avoid needless renegotiation for just software LRO. */ 1397 if ((mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO)) == 1398 IFCAP_LRO && vtnet_software_lro(sc)) 1399 reinit = update = 0; 1400 /* 1401 * VirtIO does not distinguish between receive checksum offload 1402 * for IPv4 and IPv6 packets, so treat them as a pair. 1403 */ 1404 if (mask & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 1405 if_togglecapenable(ifp, IFCAP_RXCSUM); 1406 if_togglecapenable(ifp, IFCAP_RXCSUM_IPV6); 1407 } 1408 if (mask & IFCAP_LRO) 1409 if_togglecapenable(ifp, IFCAP_LRO); 1410 /* Both SW and HW TCP LRO require receive checksum offload. */ 1411 if ((if_getcapenable(ifp) & 1412 (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0) 1413 if_setcapenablebit(ifp, 0, IFCAP_LRO); 1414 } 1415 1416 if (mask & IFCAP_VLAN_HWFILTER) { 1417 /* These Rx features require renegotiation. */ 1418 reinit = 1; 1419 1420 if (mask & IFCAP_VLAN_HWFILTER) 1421 if_togglecapenable(ifp, IFCAP_VLAN_HWFILTER); 1422 } 1423 1424 if (mask & IFCAP_VLAN_HWTSO) 1425 if_togglecapenable(ifp, IFCAP_VLAN_HWTSO); 1426 if (mask & IFCAP_VLAN_HWTAGGING) 1427 if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING); 1428 1429 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 1430 if (reinit) { 1431 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 1432 vtnet_init_locked(sc, 0); 1433 } else if (update) 1434 vtnet_update_rx_offloads(sc); 1435 } 1436 1437 return (0); 1438 } 1439 1440 static int 1441 vtnet_ioctl(if_t ifp, u_long cmd, caddr_t data) 1442 { 1443 struct vtnet_softc *sc; 1444 struct ifreq *ifr; 1445 int error; 1446 1447 sc = if_getsoftc(ifp); 1448 ifr = (struct ifreq *) data; 1449 error = 0; 1450 1451 switch (cmd) { 1452 case SIOCSIFMTU: 1453 VTNET_CORE_LOCK(sc); 1454 error = vtnet_ioctl_mtu(sc, ifr->ifr_mtu); 1455 VTNET_CORE_UNLOCK(sc); 1456 break; 1457 1458 case SIOCSIFFLAGS: 1459 VTNET_CORE_LOCK(sc); 1460 error = vtnet_ioctl_ifflags(sc); 1461 VTNET_CORE_UNLOCK(sc); 1462 break; 1463 1464 case SIOCADDMULTI: 1465 case SIOCDELMULTI: 1466 VTNET_CORE_LOCK(sc); 1467 error = vtnet_ioctl_multi(sc); 1468 VTNET_CORE_UNLOCK(sc); 1469 break; 1470 1471 case SIOCSIFMEDIA: 1472 case SIOCGIFMEDIA: 1473 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd); 1474 break; 1475 1476 case SIOCSIFCAP: 1477 VTNET_CORE_LOCK(sc); 1478 error = vtnet_ioctl_ifcap(sc, ifr); 1479 VTNET_CORE_UNLOCK(sc); 1480 VLAN_CAPABILITIES(ifp); 1481 break; 1482 1483 default: 1484 error = ether_ioctl(ifp, cmd, data); 1485 break; 1486 } 1487 1488 VTNET_CORE_LOCK_ASSERT_NOTOWNED(sc); 1489 1490 return (error); 1491 } 1492 1493 static int 1494 vtnet_rxq_populate(struct vtnet_rxq *rxq) 1495 { 1496 struct virtqueue *vq; 1497 int nbufs, error; 1498 1499 #ifdef DEV_NETMAP 1500 error = vtnet_netmap_rxq_populate(rxq); 1501 if (error >= 0) 1502 return (error); 1503 #endif /* DEV_NETMAP */ 1504 1505 vq = rxq->vtnrx_vq; 1506 error = ENOSPC; 1507 1508 for (nbufs = 0; !virtqueue_full(vq); nbufs++) { 1509 error = vtnet_rxq_new_buf(rxq); 1510 if (error) 1511 break; 1512 } 1513 1514 if (nbufs > 0) { 1515 virtqueue_notify(vq); 1516 /* 1517 * EMSGSIZE signifies the virtqueue did not have enough 1518 * entries available to hold the last mbuf. This is not 1519 * an error. 1520 */ 1521 if (error == EMSGSIZE) 1522 error = 0; 1523 } 1524 1525 return (error); 1526 } 1527 1528 static void 1529 vtnet_rxq_free_mbufs(struct vtnet_rxq *rxq) 1530 { 1531 struct virtqueue *vq; 1532 struct mbuf *m; 1533 int last; 1534 #ifdef DEV_NETMAP 1535 struct netmap_kring *kring = netmap_kring_on(NA(rxq->vtnrx_sc->vtnet_ifp), 1536 rxq->vtnrx_id, NR_RX); 1537 #else /* !DEV_NETMAP */ 1538 void *kring = NULL; 1539 #endif /* !DEV_NETMAP */ 1540 1541 vq = rxq->vtnrx_vq; 1542 last = 0; 1543 1544 while ((m = virtqueue_drain(vq, &last)) != NULL) { 1545 if (kring == NULL) 1546 m_freem(m); 1547 } 1548 1549 KASSERT(virtqueue_empty(vq), 1550 ("%s: mbufs remaining in rx queue %p", __func__, rxq)); 1551 } 1552 1553 static struct mbuf * 1554 vtnet_rx_alloc_buf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp) 1555 { 1556 struct mbuf *m_head, *m_tail, *m; 1557 int i, size; 1558 1559 m_head = NULL; 1560 size = sc->vtnet_rx_clustersz; 1561 1562 KASSERT(nbufs == 1 || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1563 ("%s: mbuf %d chain requested without LRO_NOMRG", __func__, nbufs)); 1564 1565 for (i = 0; i < nbufs; i++) { 1566 m = m_getjcl(M_NOWAIT, MT_DATA, i == 0 ? M_PKTHDR : 0, size); 1567 if (m == NULL) { 1568 sc->vtnet_stats.mbuf_alloc_failed++; 1569 m_freem(m_head); 1570 return (NULL); 1571 } 1572 1573 m->m_len = size; 1574 /* 1575 * Need to offset the mbuf if the header we're going to add 1576 * will misalign. 1577 */ 1578 if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) { 1579 m_adj(m, VTNET_ETHER_ALIGN); 1580 } 1581 if (m_head != NULL) { 1582 m_tail->m_next = m; 1583 m_tail = m; 1584 } else 1585 m_head = m_tail = m; 1586 } 1587 1588 if (m_tailp != NULL) 1589 *m_tailp = m_tail; 1590 1591 return (m_head); 1592 } 1593 1594 /* 1595 * Slow path for when LRO without mergeable buffers is negotiated. 1596 */ 1597 static int 1598 vtnet_rxq_replace_lro_nomrg_buf(struct vtnet_rxq *rxq, struct mbuf *m0, 1599 int len0) 1600 { 1601 struct vtnet_softc *sc; 1602 struct mbuf *m, *m_prev, *m_new, *m_tail; 1603 int len, clustersz, nreplace, error; 1604 1605 sc = rxq->vtnrx_sc; 1606 clustersz = sc->vtnet_rx_clustersz; 1607 /* 1608 * Need to offset the mbuf if the header we're going to add will 1609 * misalign, account for that here. 1610 */ 1611 if (VTNET_ETHER_ALIGN != 0 && sc->vtnet_hdr_size % 4 == 0) 1612 clustersz -= VTNET_ETHER_ALIGN; 1613 1614 m_prev = NULL; 1615 m_tail = NULL; 1616 nreplace = 0; 1617 1618 m = m0; 1619 len = len0; 1620 1621 /* 1622 * Since these mbuf chains are so large, avoid allocating a complete 1623 * replacement when the received frame did not consume the entire 1624 * chain. Unused mbufs are moved to the tail of the replacement mbuf. 1625 */ 1626 while (len > 0) { 1627 if (m == NULL) { 1628 sc->vtnet_stats.rx_frame_too_large++; 1629 return (EMSGSIZE); 1630 } 1631 1632 /* 1633 * Every mbuf should have the expected cluster size since that 1634 * is also used to allocate the replacements. 1635 */ 1636 KASSERT(m->m_len == clustersz, 1637 ("%s: mbuf size %d not expected cluster size %d", __func__, 1638 m->m_len, clustersz)); 1639 1640 m->m_len = MIN(m->m_len, len); 1641 len -= m->m_len; 1642 1643 m_prev = m; 1644 m = m->m_next; 1645 nreplace++; 1646 } 1647 1648 KASSERT(nreplace > 0 && nreplace <= sc->vtnet_rx_nmbufs, 1649 ("%s: invalid replacement mbuf count %d max %d", __func__, 1650 nreplace, sc->vtnet_rx_nmbufs)); 1651 1652 m_new = vtnet_rx_alloc_buf(sc, nreplace, &m_tail); 1653 if (m_new == NULL) { 1654 m_prev->m_len = clustersz; 1655 return (ENOBUFS); 1656 } 1657 1658 /* 1659 * Move any unused mbufs from the received mbuf chain onto the 1660 * end of the replacement chain. 1661 */ 1662 if (m_prev->m_next != NULL) { 1663 m_tail->m_next = m_prev->m_next; 1664 m_prev->m_next = NULL; 1665 } 1666 1667 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1668 if (error) { 1669 /* 1670 * The replacement is suppose to be an copy of the one 1671 * dequeued so this is a very unexpected error. 1672 * 1673 * Restore the m0 chain to the original state if it was 1674 * modified so we can then discard it. 1675 */ 1676 if (m_tail->m_next != NULL) { 1677 m_prev->m_next = m_tail->m_next; 1678 m_tail->m_next = NULL; 1679 } 1680 m_prev->m_len = clustersz; 1681 sc->vtnet_stats.rx_enq_replacement_failed++; 1682 m_freem(m_new); 1683 } 1684 1685 return (error); 1686 } 1687 1688 static int 1689 vtnet_rxq_replace_buf(struct vtnet_rxq *rxq, struct mbuf *m, int len) 1690 { 1691 struct vtnet_softc *sc; 1692 struct mbuf *m_new; 1693 int error; 1694 1695 sc = rxq->vtnrx_sc; 1696 1697 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) 1698 return (vtnet_rxq_replace_lro_nomrg_buf(rxq, m, len)); 1699 1700 MPASS(m->m_next == NULL); 1701 if (m->m_len < len) 1702 return (EMSGSIZE); 1703 1704 m_new = vtnet_rx_alloc_buf(sc, 1, NULL); 1705 if (m_new == NULL) 1706 return (ENOBUFS); 1707 1708 error = vtnet_rxq_enqueue_buf(rxq, m_new); 1709 if (error) { 1710 sc->vtnet_stats.rx_enq_replacement_failed++; 1711 m_freem(m_new); 1712 } else 1713 m->m_len = len; 1714 1715 return (error); 1716 } 1717 1718 static int 1719 vtnet_rxq_enqueue_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1720 { 1721 struct vtnet_softc *sc; 1722 struct sglist *sg; 1723 int header_inlined, error; 1724 1725 sc = rxq->vtnrx_sc; 1726 sg = rxq->vtnrx_sg; 1727 1728 KASSERT(m->m_next == NULL || sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG, 1729 ("%s: mbuf chain without LRO_NOMRG", __func__)); 1730 VTNET_RXQ_LOCK_ASSERT(rxq); 1731 1732 sglist_reset(sg); 1733 header_inlined = vtnet_modern(sc) || 1734 (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) != 0; /* TODO: ANY_LAYOUT */ 1735 1736 /* 1737 * Note: The mbuf has been already adjusted when we allocate it if we 1738 * have to do strict alignment. 1739 */ 1740 if (header_inlined) 1741 error = sglist_append_mbuf(sg, m); 1742 else { 1743 struct vtnet_rx_header *rxhdr = 1744 mtod(m, struct vtnet_rx_header *); 1745 MPASS(sc->vtnet_hdr_size == sizeof(struct virtio_net_hdr)); 1746 1747 /* Append the header and remaining mbuf data. */ 1748 error = sglist_append(sg, &rxhdr->vrh_hdr, sc->vtnet_hdr_size); 1749 if (error) 1750 return (error); 1751 error = sglist_append(sg, &rxhdr[1], 1752 m->m_len - sizeof(struct vtnet_rx_header)); 1753 if (error) 1754 return (error); 1755 1756 if (m->m_next != NULL) 1757 error = sglist_append_mbuf(sg, m->m_next); 1758 } 1759 1760 if (error) 1761 return (error); 1762 1763 return (virtqueue_enqueue(rxq->vtnrx_vq, m, sg, 0, sg->sg_nseg)); 1764 } 1765 1766 static int 1767 vtnet_rxq_new_buf(struct vtnet_rxq *rxq) 1768 { 1769 struct vtnet_softc *sc; 1770 struct mbuf *m; 1771 int error; 1772 1773 sc = rxq->vtnrx_sc; 1774 1775 m = vtnet_rx_alloc_buf(sc, sc->vtnet_rx_nmbufs, NULL); 1776 if (m == NULL) 1777 return (ENOBUFS); 1778 1779 error = vtnet_rxq_enqueue_buf(rxq, m); 1780 if (error) 1781 m_freem(m); 1782 1783 return (error); 1784 } 1785 1786 #if defined(INET) || defined(INET6) 1787 static int 1788 vtnet_rxq_csum_needs_csum(struct vtnet_rxq *rxq, struct mbuf *m, bool isipv6, 1789 int protocol, struct virtio_net_hdr *hdr) 1790 { 1791 struct vtnet_softc *sc; 1792 1793 /* 1794 * The packet is likely from another VM on the same host or from the 1795 * host that itself performed checksum offloading so Tx/Rx is basically 1796 * a memcpy and the checksum has little value so far. 1797 */ 1798 1799 KASSERT(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP, 1800 ("%s: unsupported IP protocol %d", __func__, protocol)); 1801 1802 /* 1803 * If the user don't want us to fix it up here by computing the 1804 * checksum, just forward the order to compute the checksum by setting 1805 * the corresponding mbuf flag (e.g., CSUM_TCP). 1806 */ 1807 sc = rxq->vtnrx_sc; 1808 if ((sc->vtnet_flags & VTNET_FLAG_FIXUP_NEEDS_CSUM) == 0) { 1809 switch (protocol) { 1810 case IPPROTO_TCP: 1811 m->m_pkthdr.csum_flags |= 1812 (isipv6 ? CSUM_TCP_IPV6 : CSUM_TCP); 1813 break; 1814 case IPPROTO_UDP: 1815 m->m_pkthdr.csum_flags |= 1816 (isipv6 ? CSUM_UDP_IPV6 : CSUM_UDP); 1817 break; 1818 } 1819 m->m_pkthdr.csum_data = hdr->csum_offset; 1820 return (0); 1821 } 1822 1823 /* 1824 * Compute the checksum in the driver so the packet will contain a 1825 * valid checksum. The checksum is at csum_offset from csum_start. 1826 */ 1827 int csum_off, csum_end; 1828 uint16_t csum; 1829 1830 csum_off = hdr->csum_start + hdr->csum_offset; 1831 csum_end = csum_off + sizeof(uint16_t); 1832 1833 /* Assume checksum will be in the first mbuf. */ 1834 if (m->m_len < csum_end || m->m_pkthdr.len < csum_end) { 1835 sc->vtnet_stats.rx_csum_bad_offset++; 1836 return (1); 1837 } 1838 1839 /* 1840 * Like in_delayed_cksum()/in6_delayed_cksum(), compute the 1841 * checksum and write it at the specified offset. We could 1842 * try to verify the packet: csum_start should probably 1843 * correspond to the start of the TCP/UDP header. 1844 * 1845 * BMV: Need to properly handle UDP with zero checksum. Is 1846 * the IPv4 header checksum implicitly validated? 1847 */ 1848 csum = in_cksum_skip(m, m->m_pkthdr.len, hdr->csum_start); 1849 *(uint16_t *)(mtodo(m, csum_off)) = csum; 1850 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1851 m->m_pkthdr.csum_data = 0xFFFF; 1852 1853 return (0); 1854 } 1855 1856 static void 1857 vtnet_rxq_csum_data_valid(struct vtnet_rxq *rxq, struct mbuf *m, int protocol) 1858 { 1859 KASSERT(protocol == IPPROTO_TCP || protocol == IPPROTO_UDP, 1860 ("%s: unsupported IP protocol %d", __func__, protocol)); 1861 1862 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR; 1863 m->m_pkthdr.csum_data = 0xFFFF; 1864 } 1865 1866 static int 1867 vtnet_rxq_csum(struct vtnet_rxq *rxq, struct mbuf *m, 1868 struct virtio_net_hdr *hdr) 1869 { 1870 const struct ether_header *eh; 1871 struct vtnet_softc *sc; 1872 int hoff, protocol; 1873 uint16_t etype; 1874 bool isipv6; 1875 1876 KASSERT(hdr->flags & 1877 (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID), 1878 ("%s: missing checksum offloading flag %x", __func__, hdr->flags)); 1879 1880 eh = mtod(m, const struct ether_header *); 1881 etype = ntohs(eh->ether_type); 1882 if (etype == ETHERTYPE_VLAN) { 1883 /* TODO BMV: Handle QinQ. */ 1884 const struct ether_vlan_header *evh = 1885 mtod(m, const struct ether_vlan_header *); 1886 etype = ntohs(evh->evl_proto); 1887 hoff = sizeof(struct ether_vlan_header); 1888 } else 1889 hoff = sizeof(struct ether_header); 1890 1891 sc = rxq->vtnrx_sc; 1892 1893 /* Check whether ethernet type is IP or IPv6, and get protocol. */ 1894 switch (etype) { 1895 #if defined(INET) 1896 case ETHERTYPE_IP: 1897 if (__predict_false(m->m_len < hoff + sizeof(struct ip))) { 1898 sc->vtnet_stats.rx_csum_inaccessible_ipproto++; 1899 return (1); 1900 } else { 1901 struct ip *ip = (struct ip *)(m->m_data + hoff); 1902 protocol = ip->ip_p; 1903 } 1904 isipv6 = false; 1905 break; 1906 #endif 1907 #if defined(INET6) 1908 case ETHERTYPE_IPV6: 1909 if (__predict_false(m->m_len < hoff + sizeof(struct ip6_hdr)) 1910 || ip6_lasthdr(m, hoff, IPPROTO_IPV6, &protocol) < 0) { 1911 sc->vtnet_stats.rx_csum_inaccessible_ipproto++; 1912 return (1); 1913 } 1914 isipv6 = true; 1915 break; 1916 #endif 1917 default: 1918 sc->vtnet_stats.rx_csum_bad_ethtype++; 1919 return (1); 1920 } 1921 1922 /* Check whether protocol is TCP or UDP. */ 1923 switch (protocol) { 1924 case IPPROTO_TCP: 1925 case IPPROTO_UDP: 1926 break; 1927 default: 1928 /* 1929 * FreeBSD does not support checksum offloading of this 1930 * protocol here. 1931 */ 1932 sc->vtnet_stats.rx_csum_bad_ipproto++; 1933 return (1); 1934 } 1935 1936 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 1937 return (vtnet_rxq_csum_needs_csum(rxq, m, isipv6, protocol, 1938 hdr)); 1939 else /* VIRTIO_NET_HDR_F_DATA_VALID */ 1940 vtnet_rxq_csum_data_valid(rxq, m, protocol); 1941 1942 return (0); 1943 } 1944 #endif 1945 1946 static void 1947 vtnet_rxq_discard_merged_bufs(struct vtnet_rxq *rxq, int nbufs) 1948 { 1949 struct mbuf *m; 1950 1951 while (--nbufs > 0) { 1952 m = virtqueue_dequeue(rxq->vtnrx_vq, NULL); 1953 if (m == NULL) 1954 break; 1955 vtnet_rxq_discard_buf(rxq, m); 1956 } 1957 } 1958 1959 static void 1960 vtnet_rxq_discard_buf(struct vtnet_rxq *rxq, struct mbuf *m) 1961 { 1962 int error __diagused; 1963 1964 /* 1965 * Requeue the discarded mbuf. This should always be successful 1966 * since it was just dequeued. 1967 */ 1968 error = vtnet_rxq_enqueue_buf(rxq, m); 1969 KASSERT(error == 0, 1970 ("%s: cannot requeue discarded mbuf %d", __func__, error)); 1971 } 1972 1973 static int 1974 vtnet_rxq_merged_eof(struct vtnet_rxq *rxq, struct mbuf *m_head, int nbufs) 1975 { 1976 struct vtnet_softc *sc; 1977 struct virtqueue *vq; 1978 struct mbuf *m_tail; 1979 1980 sc = rxq->vtnrx_sc; 1981 vq = rxq->vtnrx_vq; 1982 m_tail = m_head; 1983 1984 while (--nbufs > 0) { 1985 struct mbuf *m; 1986 uint32_t len; 1987 1988 m = virtqueue_dequeue(vq, &len); 1989 if (m == NULL) { 1990 rxq->vtnrx_stats.vrxs_ierrors++; 1991 goto fail; 1992 } 1993 1994 if (vtnet_rxq_new_buf(rxq) != 0) { 1995 rxq->vtnrx_stats.vrxs_iqdrops++; 1996 vtnet_rxq_discard_buf(rxq, m); 1997 if (nbufs > 1) 1998 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 1999 goto fail; 2000 } 2001 2002 if (m->m_len < len) 2003 len = m->m_len; 2004 2005 m->m_len = len; 2006 m->m_flags &= ~M_PKTHDR; 2007 2008 m_head->m_pkthdr.len += len; 2009 m_tail->m_next = m; 2010 m_tail = m; 2011 } 2012 2013 return (0); 2014 2015 fail: 2016 sc->vtnet_stats.rx_mergeable_failed++; 2017 m_freem(m_head); 2018 2019 return (1); 2020 } 2021 2022 #if defined(INET) || defined(INET6) 2023 static int 2024 vtnet_lro_rx(struct vtnet_rxq *rxq, struct mbuf *m) 2025 { 2026 struct lro_ctrl *lro; 2027 2028 lro = &rxq->vtnrx_lro; 2029 2030 if (lro->lro_mbuf_max != 0) { 2031 tcp_lro_queue_mbuf(lro, m); 2032 return (0); 2033 } 2034 2035 return (tcp_lro_rx(lro, m, 0)); 2036 } 2037 #endif 2038 2039 static void 2040 vtnet_rxq_input(struct vtnet_rxq *rxq, struct mbuf *m, 2041 struct virtio_net_hdr *hdr) 2042 { 2043 struct vtnet_softc *sc; 2044 if_t ifp; 2045 2046 sc = rxq->vtnrx_sc; 2047 ifp = sc->vtnet_ifp; 2048 2049 if (if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) { 2050 struct ether_header *eh = mtod(m, struct ether_header *); 2051 if (eh->ether_type == htons(ETHERTYPE_VLAN)) { 2052 vtnet_vlan_tag_remove(m); 2053 /* 2054 * With the 802.1Q header removed, update the 2055 * checksum starting location accordingly. 2056 */ 2057 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) 2058 hdr->csum_start -= ETHER_VLAN_ENCAP_LEN; 2059 } 2060 } 2061 2062 m->m_pkthdr.flowid = rxq->vtnrx_id; 2063 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE); 2064 2065 if (hdr->flags & 2066 (VIRTIO_NET_HDR_F_NEEDS_CSUM | VIRTIO_NET_HDR_F_DATA_VALID)) { 2067 #if defined(INET) || defined(INET6) 2068 if (vtnet_rxq_csum(rxq, m, hdr) == 0) 2069 rxq->vtnrx_stats.vrxs_csum++; 2070 else 2071 rxq->vtnrx_stats.vrxs_csum_failed++; 2072 #else 2073 sc->vtnet_stats.rx_csum_bad_ethtype++; 2074 rxq->vtnrx_stats.vrxs_csum_failed++; 2075 #endif 2076 } 2077 2078 if (hdr->gso_size != 0) { 2079 switch (hdr->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 2080 case VIRTIO_NET_HDR_GSO_TCPV4: 2081 case VIRTIO_NET_HDR_GSO_TCPV6: 2082 m->m_pkthdr.lro_nsegs = 2083 howmany(m->m_pkthdr.len, hdr->gso_size); 2084 rxq->vtnrx_stats.vrxs_host_lro++; 2085 break; 2086 } 2087 } 2088 2089 rxq->vtnrx_stats.vrxs_ipackets++; 2090 rxq->vtnrx_stats.vrxs_ibytes += m->m_pkthdr.len; 2091 2092 #if defined(INET) || defined(INET6) 2093 if (vtnet_software_lro(sc) && if_getcapenable(ifp) & IFCAP_LRO) { 2094 if (vtnet_lro_rx(rxq, m) == 0) 2095 return; 2096 } 2097 #endif 2098 2099 if_input(ifp, m); 2100 } 2101 2102 static int 2103 vtnet_rxq_eof(struct vtnet_rxq *rxq) 2104 { 2105 struct virtio_net_hdr lhdr, *hdr; 2106 struct vtnet_softc *sc; 2107 if_t ifp; 2108 struct virtqueue *vq; 2109 int deq, count; 2110 2111 sc = rxq->vtnrx_sc; 2112 vq = rxq->vtnrx_vq; 2113 ifp = sc->vtnet_ifp; 2114 deq = 0; 2115 count = sc->vtnet_rx_process_limit; 2116 2117 VTNET_RXQ_LOCK_ASSERT(rxq); 2118 2119 CURVNET_SET(if_getvnet(ifp)); 2120 while (count-- > 0) { 2121 struct mbuf *m; 2122 uint32_t len, nbufs, adjsz; 2123 2124 m = virtqueue_dequeue(vq, &len); 2125 if (m == NULL) 2126 break; 2127 deq++; 2128 2129 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) { 2130 rxq->vtnrx_stats.vrxs_ierrors++; 2131 vtnet_rxq_discard_buf(rxq, m); 2132 continue; 2133 } 2134 2135 if (sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) { 2136 struct virtio_net_hdr_mrg_rxbuf *mhdr = 2137 mtod(m, struct virtio_net_hdr_mrg_rxbuf *); 2138 kmsan_mark(mhdr, sizeof(*mhdr), KMSAN_STATE_INITED); 2139 nbufs = vtnet_htog16(sc, mhdr->num_buffers); 2140 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf); 2141 } else if (vtnet_modern(sc)) { 2142 nbufs = 1; /* num_buffers is always 1 */ 2143 adjsz = sizeof(struct virtio_net_hdr_v1); 2144 } else { 2145 nbufs = 1; 2146 adjsz = sizeof(struct vtnet_rx_header); 2147 /* 2148 * Account for our gap between the header and start of 2149 * data to keep the segments separated. 2150 */ 2151 len += VTNET_RX_HEADER_PAD; 2152 } 2153 2154 if (vtnet_rxq_replace_buf(rxq, m, len) != 0) { 2155 rxq->vtnrx_stats.vrxs_iqdrops++; 2156 vtnet_rxq_discard_buf(rxq, m); 2157 if (nbufs > 1) 2158 vtnet_rxq_discard_merged_bufs(rxq, nbufs); 2159 continue; 2160 } 2161 2162 m->m_pkthdr.len = len; 2163 m->m_pkthdr.rcvif = ifp; 2164 m->m_pkthdr.csum_flags = 0; 2165 2166 if (nbufs > 1) { 2167 /* Dequeue the rest of chain. */ 2168 if (vtnet_rxq_merged_eof(rxq, m, nbufs) != 0) 2169 continue; 2170 } 2171 2172 kmsan_mark_mbuf(m, KMSAN_STATE_INITED); 2173 2174 /* 2175 * Save an endian swapped version of the header prior to it 2176 * being stripped. The header is always at the start of the 2177 * mbuf data. num_buffers was already saved (and not needed) 2178 * so use the standard header. 2179 */ 2180 hdr = mtod(m, struct virtio_net_hdr *); 2181 lhdr.flags = hdr->flags; 2182 lhdr.gso_type = hdr->gso_type; 2183 lhdr.hdr_len = vtnet_htog16(sc, hdr->hdr_len); 2184 lhdr.gso_size = vtnet_htog16(sc, hdr->gso_size); 2185 lhdr.csum_start = vtnet_htog16(sc, hdr->csum_start); 2186 lhdr.csum_offset = vtnet_htog16(sc, hdr->csum_offset); 2187 m_adj(m, adjsz); 2188 2189 if (PFIL_HOOKED_IN(sc->vtnet_pfil)) { 2190 pfil_return_t pfil; 2191 2192 pfil = pfil_mbuf_in(sc->vtnet_pfil, &m, ifp, NULL); 2193 switch (pfil) { 2194 case PFIL_DROPPED: 2195 case PFIL_CONSUMED: 2196 continue; 2197 default: 2198 KASSERT(pfil == PFIL_PASS, 2199 ("Filter returned %d!", pfil)); 2200 } 2201 } 2202 2203 vtnet_rxq_input(rxq, m, &lhdr); 2204 } 2205 2206 if (deq > 0) { 2207 #if defined(INET) || defined(INET6) 2208 if (vtnet_software_lro(sc)) 2209 tcp_lro_flush_all(&rxq->vtnrx_lro); 2210 #endif 2211 virtqueue_notify(vq); 2212 } 2213 CURVNET_RESTORE(); 2214 2215 return (count > 0 ? 0 : EAGAIN); 2216 } 2217 2218 static void 2219 vtnet_rx_vq_process(struct vtnet_rxq *rxq, int tries) 2220 { 2221 struct vtnet_softc *sc; 2222 if_t ifp; 2223 u_int more; 2224 #ifdef DEV_NETMAP 2225 int nmirq; 2226 #endif /* DEV_NETMAP */ 2227 2228 sc = rxq->vtnrx_sc; 2229 ifp = sc->vtnet_ifp; 2230 2231 if (__predict_false(rxq->vtnrx_id >= sc->vtnet_act_vq_pairs)) { 2232 /* 2233 * Ignore this interrupt. Either this is a spurious interrupt 2234 * or multiqueue without per-VQ MSIX so every queue needs to 2235 * be polled (a brain dead configuration we could try harder 2236 * to avoid). 2237 */ 2238 vtnet_rxq_disable_intr(rxq); 2239 return; 2240 } 2241 2242 VTNET_RXQ_LOCK(rxq); 2243 2244 #ifdef DEV_NETMAP 2245 /* 2246 * We call netmap_rx_irq() under lock to prevent concurrent calls. 2247 * This is not necessary to serialize the access to the RX vq, but 2248 * rather to avoid races that may happen if this interface is 2249 * attached to a VALE switch, which would cause received packets 2250 * to stall in the RX queue (nm_kr_tryget() could find the kring 2251 * busy when called from netmap_bwrap_intr_notify()). 2252 */ 2253 nmirq = netmap_rx_irq(ifp, rxq->vtnrx_id, &more); 2254 if (nmirq != NM_IRQ_PASS) { 2255 VTNET_RXQ_UNLOCK(rxq); 2256 if (nmirq == NM_IRQ_RESCHED) { 2257 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2258 } 2259 return; 2260 } 2261 #endif /* DEV_NETMAP */ 2262 2263 again: 2264 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) { 2265 VTNET_RXQ_UNLOCK(rxq); 2266 return; 2267 } 2268 2269 more = vtnet_rxq_eof(rxq); 2270 if (more || vtnet_rxq_enable_intr(rxq) != 0) { 2271 if (!more) 2272 vtnet_rxq_disable_intr(rxq); 2273 /* 2274 * This is an occasional condition or race (when !more), 2275 * so retry a few times before scheduling the taskqueue. 2276 */ 2277 if (tries-- > 0) 2278 goto again; 2279 2280 rxq->vtnrx_stats.vrxs_rescheduled++; 2281 VTNET_RXQ_UNLOCK(rxq); 2282 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 2283 } else 2284 VTNET_RXQ_UNLOCK(rxq); 2285 } 2286 2287 static void 2288 vtnet_rx_vq_intr(void *xrxq) 2289 { 2290 struct vtnet_rxq *rxq; 2291 2292 rxq = xrxq; 2293 vtnet_rx_vq_process(rxq, VTNET_INTR_DISABLE_RETRIES); 2294 } 2295 2296 static void 2297 vtnet_rxq_tq_intr(void *xrxq, int pending __unused) 2298 { 2299 struct vtnet_rxq *rxq; 2300 2301 rxq = xrxq; 2302 vtnet_rx_vq_process(rxq, 0); 2303 } 2304 2305 static int 2306 vtnet_txq_intr_threshold(struct vtnet_txq *txq) 2307 { 2308 struct vtnet_softc *sc; 2309 int threshold; 2310 2311 sc = txq->vtntx_sc; 2312 2313 /* 2314 * The Tx interrupt is disabled until the queue free count falls 2315 * below our threshold. Completed frames are drained from the Tx 2316 * virtqueue before transmitting new frames and in the watchdog 2317 * callout, so the frequency of Tx interrupts is greatly reduced, 2318 * at the cost of not freeing mbufs as quickly as they otherwise 2319 * would be. 2320 */ 2321 threshold = virtqueue_size(txq->vtntx_vq) / 4; 2322 2323 /* 2324 * Without indirect descriptors, leave enough room for the most 2325 * segments we handle. 2326 */ 2327 if ((sc->vtnet_flags & VTNET_FLAG_INDIRECT) == 0 && 2328 threshold < sc->vtnet_tx_nsegs) 2329 threshold = sc->vtnet_tx_nsegs; 2330 2331 return (threshold); 2332 } 2333 2334 static int 2335 vtnet_txq_below_threshold(struct vtnet_txq *txq) 2336 { 2337 struct virtqueue *vq; 2338 2339 vq = txq->vtntx_vq; 2340 2341 return (virtqueue_nfree(vq) <= txq->vtntx_intr_threshold); 2342 } 2343 2344 static int 2345 vtnet_txq_notify(struct vtnet_txq *txq) 2346 { 2347 struct virtqueue *vq; 2348 2349 vq = txq->vtntx_vq; 2350 2351 txq->vtntx_watchdog = VTNET_TX_TIMEOUT; 2352 virtqueue_notify(vq); 2353 2354 if (vtnet_txq_enable_intr(txq) == 0) 2355 return (0); 2356 2357 /* 2358 * Drain frames that were completed since last checked. If this 2359 * causes the queue to go above the threshold, the caller should 2360 * continue transmitting. 2361 */ 2362 if (vtnet_txq_eof(txq) != 0 && vtnet_txq_below_threshold(txq) == 0) { 2363 virtqueue_disable_intr(vq); 2364 return (1); 2365 } 2366 2367 return (0); 2368 } 2369 2370 static void 2371 vtnet_txq_free_mbufs(struct vtnet_txq *txq) 2372 { 2373 struct virtqueue *vq; 2374 struct vtnet_tx_header *txhdr; 2375 int last; 2376 #ifdef DEV_NETMAP 2377 struct netmap_kring *kring = netmap_kring_on(NA(txq->vtntx_sc->vtnet_ifp), 2378 txq->vtntx_id, NR_TX); 2379 #else /* !DEV_NETMAP */ 2380 void *kring = NULL; 2381 #endif /* !DEV_NETMAP */ 2382 2383 vq = txq->vtntx_vq; 2384 last = 0; 2385 2386 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) { 2387 if (kring == NULL) { 2388 m_freem(txhdr->vth_mbuf); 2389 uma_zfree(vtnet_tx_header_zone, txhdr); 2390 } 2391 } 2392 2393 KASSERT(virtqueue_empty(vq), 2394 ("%s: mbufs remaining in tx queue %p", __func__, txq)); 2395 } 2396 2397 /* 2398 * BMV: This can go away once we finally have offsets in the mbuf header. 2399 */ 2400 static int 2401 vtnet_txq_offload_ctx(struct vtnet_txq *txq, struct mbuf *m, int *etype, 2402 int *proto, int *start) 2403 { 2404 struct vtnet_softc *sc; 2405 struct ether_vlan_header *evh; 2406 #if defined(INET) || defined(INET6) 2407 int offset; 2408 #endif 2409 2410 sc = txq->vtntx_sc; 2411 2412 evh = mtod(m, struct ether_vlan_header *); 2413 if (evh->evl_encap_proto == htons(ETHERTYPE_VLAN)) { 2414 /* BMV: We should handle nested VLAN tags too. */ 2415 *etype = ntohs(evh->evl_proto); 2416 #if defined(INET) || defined(INET6) 2417 offset = sizeof(struct ether_vlan_header); 2418 #endif 2419 } else { 2420 *etype = ntohs(evh->evl_encap_proto); 2421 #if defined(INET) || defined(INET6) 2422 offset = sizeof(struct ether_header); 2423 #endif 2424 } 2425 2426 switch (*etype) { 2427 #if defined(INET) 2428 case ETHERTYPE_IP: { 2429 struct ip *ip, iphdr; 2430 if (__predict_false(m->m_len < offset + sizeof(struct ip))) { 2431 m_copydata(m, offset, sizeof(struct ip), 2432 (caddr_t) &iphdr); 2433 ip = &iphdr; 2434 } else 2435 ip = (struct ip *)(m->m_data + offset); 2436 *proto = ip->ip_p; 2437 *start = offset + (ip->ip_hl << 2); 2438 break; 2439 } 2440 #endif 2441 #if defined(INET6) 2442 case ETHERTYPE_IPV6: 2443 *proto = -1; 2444 *start = ip6_lasthdr(m, offset, IPPROTO_IPV6, proto); 2445 /* Assert the network stack sent us a valid packet. */ 2446 KASSERT(*start > offset, 2447 ("%s: mbuf %p start %d offset %d proto %d", __func__, m, 2448 *start, offset, *proto)); 2449 break; 2450 #endif 2451 default: 2452 sc->vtnet_stats.tx_csum_unknown_ethtype++; 2453 return (EINVAL); 2454 } 2455 2456 return (0); 2457 } 2458 2459 static int 2460 vtnet_txq_offload_tso(struct vtnet_txq *txq, struct mbuf *m, int eth_type, 2461 int offset, struct virtio_net_hdr *hdr) 2462 { 2463 static struct timeval lastecn; 2464 static int curecn; 2465 struct vtnet_softc *sc; 2466 struct tcphdr *tcp, tcphdr; 2467 2468 sc = txq->vtntx_sc; 2469 2470 if (__predict_false(m->m_len < offset + sizeof(struct tcphdr))) { 2471 m_copydata(m, offset, sizeof(struct tcphdr), (caddr_t) &tcphdr); 2472 tcp = &tcphdr; 2473 } else 2474 tcp = (struct tcphdr *)(m->m_data + offset); 2475 2476 hdr->hdr_len = vtnet_gtoh16(sc, offset + (tcp->th_off << 2)); 2477 hdr->gso_size = vtnet_gtoh16(sc, m->m_pkthdr.tso_segsz); 2478 hdr->gso_type = eth_type == ETHERTYPE_IP ? VIRTIO_NET_HDR_GSO_TCPV4 : 2479 VIRTIO_NET_HDR_GSO_TCPV6; 2480 2481 if (__predict_false(tcp_get_flags(tcp) & TH_CWR)) { 2482 /* 2483 * Drop if VIRTIO_NET_F_HOST_ECN was not negotiated. In 2484 * FreeBSD, ECN support is not on a per-interface basis, 2485 * but globally via the net.inet.tcp.ecn.enable sysctl 2486 * knob. The default is off. 2487 */ 2488 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) { 2489 if (ppsratecheck(&lastecn, &curecn, 1)) 2490 if_printf(sc->vtnet_ifp, 2491 "TSO with ECN not negotiated with host\n"); 2492 return (ENOTSUP); 2493 } 2494 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN; 2495 } 2496 2497 txq->vtntx_stats.vtxs_tso++; 2498 2499 return (0); 2500 } 2501 2502 static struct mbuf * 2503 vtnet_txq_offload(struct vtnet_txq *txq, struct mbuf *m, 2504 struct virtio_net_hdr *hdr) 2505 { 2506 struct vtnet_softc *sc; 2507 int flags, etype, csum_start, proto, error; 2508 2509 sc = txq->vtntx_sc; 2510 flags = m->m_pkthdr.csum_flags; 2511 2512 error = vtnet_txq_offload_ctx(txq, m, &etype, &proto, &csum_start); 2513 if (error) 2514 goto drop; 2515 2516 if (flags & (VTNET_CSUM_OFFLOAD | VTNET_CSUM_OFFLOAD_IPV6)) { 2517 /* Sanity check the parsed mbuf matches the offload flags. */ 2518 if (__predict_false((flags & VTNET_CSUM_OFFLOAD && 2519 etype != ETHERTYPE_IP) || (flags & VTNET_CSUM_OFFLOAD_IPV6 2520 && etype != ETHERTYPE_IPV6))) { 2521 sc->vtnet_stats.tx_csum_proto_mismatch++; 2522 goto drop; 2523 } 2524 2525 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM; 2526 hdr->csum_start = vtnet_gtoh16(sc, csum_start); 2527 hdr->csum_offset = vtnet_gtoh16(sc, m->m_pkthdr.csum_data); 2528 txq->vtntx_stats.vtxs_csum++; 2529 } 2530 2531 if (flags & (CSUM_IP_TSO | CSUM_IP6_TSO)) { 2532 /* 2533 * Sanity check the parsed mbuf IP protocol is TCP, and 2534 * VirtIO TSO reqires the checksum offloading above. 2535 */ 2536 if (__predict_false(proto != IPPROTO_TCP)) { 2537 sc->vtnet_stats.tx_tso_not_tcp++; 2538 goto drop; 2539 } else if (__predict_false((hdr->flags & 2540 VIRTIO_NET_HDR_F_NEEDS_CSUM) == 0)) { 2541 sc->vtnet_stats.tx_tso_without_csum++; 2542 goto drop; 2543 } 2544 2545 error = vtnet_txq_offload_tso(txq, m, etype, csum_start, hdr); 2546 if (error) 2547 goto drop; 2548 } 2549 2550 return (m); 2551 2552 drop: 2553 m_freem(m); 2554 return (NULL); 2555 } 2556 2557 static int 2558 vtnet_txq_enqueue_buf(struct vtnet_txq *txq, struct mbuf **m_head, 2559 struct vtnet_tx_header *txhdr) 2560 { 2561 struct vtnet_softc *sc; 2562 struct virtqueue *vq; 2563 struct sglist *sg; 2564 struct mbuf *m; 2565 int error; 2566 2567 sc = txq->vtntx_sc; 2568 vq = txq->vtntx_vq; 2569 sg = txq->vtntx_sg; 2570 m = *m_head; 2571 2572 sglist_reset(sg); 2573 error = sglist_append(sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size); 2574 if (error != 0 || sg->sg_nseg != 1) { 2575 KASSERT(0, ("%s: cannot add header to sglist error %d nseg %d", 2576 __func__, error, sg->sg_nseg)); 2577 goto fail; 2578 } 2579 2580 error = sglist_append_mbuf(sg, m); 2581 if (error) { 2582 m = m_defrag(m, M_NOWAIT); 2583 if (m == NULL) { 2584 sc->vtnet_stats.tx_defrag_failed++; 2585 goto fail; 2586 } 2587 2588 *m_head = m; 2589 sc->vtnet_stats.tx_defragged++; 2590 2591 error = sglist_append_mbuf(sg, m); 2592 if (error) 2593 goto fail; 2594 } 2595 2596 txhdr->vth_mbuf = m; 2597 error = virtqueue_enqueue(vq, txhdr, sg, sg->sg_nseg, 0); 2598 2599 return (error); 2600 2601 fail: 2602 m_freem(*m_head); 2603 *m_head = NULL; 2604 2605 return (ENOBUFS); 2606 } 2607 2608 static int 2609 vtnet_txq_encap(struct vtnet_txq *txq, struct mbuf **m_head, int flags) 2610 { 2611 struct vtnet_tx_header *txhdr; 2612 struct virtio_net_hdr *hdr; 2613 struct mbuf *m; 2614 int error; 2615 2616 m = *m_head; 2617 M_ASSERTPKTHDR(m); 2618 2619 txhdr = uma_zalloc(vtnet_tx_header_zone, flags | M_ZERO); 2620 if (txhdr == NULL) { 2621 m_freem(m); 2622 *m_head = NULL; 2623 return (ENOMEM); 2624 } 2625 2626 /* 2627 * Always use the non-mergeable header, regardless if mergable headers 2628 * were negotiated, because for transmit num_buffers is always zero. 2629 * The vtnet_hdr_size is used to enqueue the right header size segment. 2630 */ 2631 hdr = &txhdr->vth_uhdr.hdr; 2632 2633 if (m->m_flags & M_VLANTAG) { 2634 m = ether_vlanencap(m, m->m_pkthdr.ether_vtag); 2635 if ((*m_head = m) == NULL) { 2636 error = ENOBUFS; 2637 goto fail; 2638 } 2639 m->m_flags &= ~M_VLANTAG; 2640 } 2641 2642 if (m->m_pkthdr.csum_flags & VTNET_CSUM_ALL_OFFLOAD) { 2643 m = vtnet_txq_offload(txq, m, hdr); 2644 if ((*m_head = m) == NULL) { 2645 error = ENOBUFS; 2646 goto fail; 2647 } 2648 } 2649 2650 error = vtnet_txq_enqueue_buf(txq, m_head, txhdr); 2651 fail: 2652 if (error) 2653 uma_zfree(vtnet_tx_header_zone, txhdr); 2654 2655 return (error); 2656 } 2657 2658 2659 static void 2660 vtnet_start_locked(struct vtnet_txq *txq, if_t ifp) 2661 { 2662 struct vtnet_softc *sc; 2663 struct virtqueue *vq; 2664 struct mbuf *m0; 2665 int tries, enq; 2666 2667 sc = txq->vtntx_sc; 2668 vq = txq->vtntx_vq; 2669 tries = 0; 2670 2671 VTNET_TXQ_LOCK_ASSERT(txq); 2672 2673 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 || 2674 sc->vtnet_link_active == 0) 2675 return; 2676 2677 vtnet_txq_eof(txq); 2678 2679 again: 2680 enq = 0; 2681 2682 while (!if_sendq_empty(ifp)) { 2683 if (virtqueue_full(vq)) 2684 break; 2685 2686 m0 = if_dequeue(ifp); 2687 if (m0 == NULL) 2688 break; 2689 2690 if (vtnet_txq_encap(txq, &m0, M_NOWAIT) != 0) { 2691 if (m0 != NULL) 2692 if_sendq_prepend(ifp, m0); 2693 break; 2694 } 2695 2696 enq++; 2697 ETHER_BPF_MTAP(ifp, m0); 2698 } 2699 2700 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2701 if (tries++ < VTNET_NOTIFY_RETRIES) 2702 goto again; 2703 2704 txq->vtntx_stats.vtxs_rescheduled++; 2705 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2706 } 2707 } 2708 2709 static void 2710 vtnet_start(if_t ifp) 2711 { 2712 struct vtnet_softc *sc; 2713 struct vtnet_txq *txq; 2714 2715 sc = if_getsoftc(ifp); 2716 txq = &sc->vtnet_txqs[0]; 2717 2718 VTNET_TXQ_LOCK(txq); 2719 vtnet_start_locked(txq, ifp); 2720 VTNET_TXQ_UNLOCK(txq); 2721 } 2722 2723 2724 static int 2725 vtnet_txq_mq_start_locked(struct vtnet_txq *txq, struct mbuf *m) 2726 { 2727 struct vtnet_softc *sc; 2728 struct virtqueue *vq; 2729 struct buf_ring *br; 2730 if_t ifp; 2731 int enq, tries, error; 2732 2733 sc = txq->vtntx_sc; 2734 vq = txq->vtntx_vq; 2735 br = txq->vtntx_br; 2736 ifp = sc->vtnet_ifp; 2737 tries = 0; 2738 error = 0; 2739 2740 VTNET_TXQ_LOCK_ASSERT(txq); 2741 2742 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0 || 2743 sc->vtnet_link_active == 0) { 2744 if (m != NULL) 2745 error = drbr_enqueue(ifp, br, m); 2746 return (error); 2747 } 2748 2749 if (m != NULL) { 2750 error = drbr_enqueue(ifp, br, m); 2751 if (error) 2752 return (error); 2753 } 2754 2755 vtnet_txq_eof(txq); 2756 2757 again: 2758 enq = 0; 2759 2760 while ((m = drbr_peek(ifp, br)) != NULL) { 2761 if (virtqueue_full(vq)) { 2762 drbr_putback(ifp, br, m); 2763 break; 2764 } 2765 2766 if (vtnet_txq_encap(txq, &m, M_NOWAIT) != 0) { 2767 if (m != NULL) 2768 drbr_putback(ifp, br, m); 2769 else 2770 drbr_advance(ifp, br); 2771 break; 2772 } 2773 drbr_advance(ifp, br); 2774 2775 enq++; 2776 ETHER_BPF_MTAP(ifp, m); 2777 } 2778 2779 if (enq > 0 && vtnet_txq_notify(txq) != 0) { 2780 if (tries++ < VTNET_NOTIFY_RETRIES) 2781 goto again; 2782 2783 txq->vtntx_stats.vtxs_rescheduled++; 2784 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_intrtask); 2785 } 2786 2787 return (0); 2788 } 2789 2790 static int 2791 vtnet_txq_mq_start(if_t ifp, struct mbuf *m) 2792 { 2793 struct vtnet_softc *sc; 2794 struct vtnet_txq *txq; 2795 int i, npairs, error; 2796 2797 sc = if_getsoftc(ifp); 2798 npairs = sc->vtnet_act_vq_pairs; 2799 2800 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) 2801 i = m->m_pkthdr.flowid % npairs; 2802 else 2803 i = curcpu % npairs; 2804 2805 txq = &sc->vtnet_txqs[i]; 2806 2807 if (VTNET_TXQ_TRYLOCK(txq) != 0) { 2808 error = vtnet_txq_mq_start_locked(txq, m); 2809 VTNET_TXQ_UNLOCK(txq); 2810 } else { 2811 error = drbr_enqueue(ifp, txq->vtntx_br, m); 2812 taskqueue_enqueue(txq->vtntx_tq, &txq->vtntx_defrtask); 2813 } 2814 2815 return (error); 2816 } 2817 2818 static void 2819 vtnet_txq_tq_deferred(void *xtxq, int pending __unused) 2820 { 2821 struct vtnet_softc *sc; 2822 struct vtnet_txq *txq; 2823 2824 txq = xtxq; 2825 sc = txq->vtntx_sc; 2826 2827 VTNET_TXQ_LOCK(txq); 2828 if (!drbr_empty(sc->vtnet_ifp, txq->vtntx_br)) 2829 vtnet_txq_mq_start_locked(txq, NULL); 2830 VTNET_TXQ_UNLOCK(txq); 2831 } 2832 2833 2834 static void 2835 vtnet_txq_start(struct vtnet_txq *txq) 2836 { 2837 struct vtnet_softc *sc; 2838 if_t ifp; 2839 2840 sc = txq->vtntx_sc; 2841 ifp = sc->vtnet_ifp; 2842 2843 if (!VTNET_ALTQ_ENABLED) { 2844 if (!drbr_empty(ifp, txq->vtntx_br)) 2845 vtnet_txq_mq_start_locked(txq, NULL); 2846 } else { 2847 if (!if_sendq_empty(ifp)) 2848 vtnet_start_locked(txq, ifp); 2849 2850 } 2851 } 2852 2853 static void 2854 vtnet_txq_tq_intr(void *xtxq, int pending __unused) 2855 { 2856 struct vtnet_softc *sc; 2857 struct vtnet_txq *txq; 2858 if_t ifp; 2859 2860 txq = xtxq; 2861 sc = txq->vtntx_sc; 2862 ifp = sc->vtnet_ifp; 2863 2864 VTNET_TXQ_LOCK(txq); 2865 2866 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) { 2867 VTNET_TXQ_UNLOCK(txq); 2868 return; 2869 } 2870 2871 vtnet_txq_eof(txq); 2872 vtnet_txq_start(txq); 2873 2874 VTNET_TXQ_UNLOCK(txq); 2875 } 2876 2877 static int 2878 vtnet_txq_eof(struct vtnet_txq *txq) 2879 { 2880 struct virtqueue *vq; 2881 struct vtnet_tx_header *txhdr; 2882 struct mbuf *m; 2883 int deq; 2884 2885 vq = txq->vtntx_vq; 2886 deq = 0; 2887 VTNET_TXQ_LOCK_ASSERT(txq); 2888 2889 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) { 2890 m = txhdr->vth_mbuf; 2891 deq++; 2892 2893 txq->vtntx_stats.vtxs_opackets++; 2894 txq->vtntx_stats.vtxs_obytes += m->m_pkthdr.len; 2895 if (m->m_flags & M_MCAST) 2896 txq->vtntx_stats.vtxs_omcasts++; 2897 2898 m_freem(m); 2899 uma_zfree(vtnet_tx_header_zone, txhdr); 2900 } 2901 2902 if (virtqueue_empty(vq)) 2903 txq->vtntx_watchdog = 0; 2904 2905 return (deq); 2906 } 2907 2908 static void 2909 vtnet_tx_vq_intr(void *xtxq) 2910 { 2911 struct vtnet_softc *sc; 2912 struct vtnet_txq *txq; 2913 if_t ifp; 2914 2915 txq = xtxq; 2916 sc = txq->vtntx_sc; 2917 ifp = sc->vtnet_ifp; 2918 2919 if (__predict_false(txq->vtntx_id >= sc->vtnet_act_vq_pairs)) { 2920 /* 2921 * Ignore this interrupt. Either this is a spurious interrupt 2922 * or multiqueue without per-VQ MSIX so every queue needs to 2923 * be polled (a brain dead configuration we could try harder 2924 * to avoid). 2925 */ 2926 vtnet_txq_disable_intr(txq); 2927 return; 2928 } 2929 2930 #ifdef DEV_NETMAP 2931 if (netmap_tx_irq(ifp, txq->vtntx_id) != NM_IRQ_PASS) 2932 return; 2933 #endif /* DEV_NETMAP */ 2934 2935 VTNET_TXQ_LOCK(txq); 2936 2937 if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0) { 2938 VTNET_TXQ_UNLOCK(txq); 2939 return; 2940 } 2941 2942 vtnet_txq_eof(txq); 2943 vtnet_txq_start(txq); 2944 2945 VTNET_TXQ_UNLOCK(txq); 2946 } 2947 2948 static void 2949 vtnet_tx_start_all(struct vtnet_softc *sc) 2950 { 2951 struct vtnet_txq *txq; 2952 int i; 2953 2954 VTNET_CORE_LOCK_ASSERT(sc); 2955 2956 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2957 txq = &sc->vtnet_txqs[i]; 2958 2959 VTNET_TXQ_LOCK(txq); 2960 vtnet_txq_start(txq); 2961 VTNET_TXQ_UNLOCK(txq); 2962 } 2963 } 2964 2965 static void 2966 vtnet_qflush(if_t ifp) 2967 { 2968 struct vtnet_softc *sc; 2969 struct vtnet_txq *txq; 2970 struct mbuf *m; 2971 int i; 2972 2973 sc = if_getsoftc(ifp); 2974 2975 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 2976 txq = &sc->vtnet_txqs[i]; 2977 2978 VTNET_TXQ_LOCK(txq); 2979 while ((m = buf_ring_dequeue_sc(txq->vtntx_br)) != NULL) 2980 m_freem(m); 2981 VTNET_TXQ_UNLOCK(txq); 2982 } 2983 2984 if_qflush(ifp); 2985 } 2986 2987 static int 2988 vtnet_watchdog(struct vtnet_txq *txq) 2989 { 2990 if_t ifp; 2991 2992 ifp = txq->vtntx_sc->vtnet_ifp; 2993 2994 VTNET_TXQ_LOCK(txq); 2995 if (txq->vtntx_watchdog == 1) { 2996 /* 2997 * Only drain completed frames if the watchdog is about to 2998 * expire. If any frames were drained, there may be enough 2999 * free descriptors now available to transmit queued frames. 3000 * In that case, the timer will immediately be decremented 3001 * below, but the timeout is generous enough that should not 3002 * be a problem. 3003 */ 3004 if (vtnet_txq_eof(txq) != 0) 3005 vtnet_txq_start(txq); 3006 } 3007 3008 if (txq->vtntx_watchdog == 0 || --txq->vtntx_watchdog) { 3009 VTNET_TXQ_UNLOCK(txq); 3010 return (0); 3011 } 3012 VTNET_TXQ_UNLOCK(txq); 3013 3014 if_printf(ifp, "watchdog timeout on queue %d\n", txq->vtntx_id); 3015 return (1); 3016 } 3017 3018 static void 3019 vtnet_accum_stats(struct vtnet_softc *sc, struct vtnet_rxq_stats *rxacc, 3020 struct vtnet_txq_stats *txacc) 3021 { 3022 3023 bzero(rxacc, sizeof(struct vtnet_rxq_stats)); 3024 bzero(txacc, sizeof(struct vtnet_txq_stats)); 3025 3026 for (int i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3027 struct vtnet_rxq_stats *rxst; 3028 struct vtnet_txq_stats *txst; 3029 3030 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 3031 rxacc->vrxs_ipackets += rxst->vrxs_ipackets; 3032 rxacc->vrxs_ibytes += rxst->vrxs_ibytes; 3033 rxacc->vrxs_iqdrops += rxst->vrxs_iqdrops; 3034 rxacc->vrxs_csum += rxst->vrxs_csum; 3035 rxacc->vrxs_csum_failed += rxst->vrxs_csum_failed; 3036 rxacc->vrxs_rescheduled += rxst->vrxs_rescheduled; 3037 3038 txst = &sc->vtnet_txqs[i].vtntx_stats; 3039 txacc->vtxs_opackets += txst->vtxs_opackets; 3040 txacc->vtxs_obytes += txst->vtxs_obytes; 3041 txacc->vtxs_csum += txst->vtxs_csum; 3042 txacc->vtxs_tso += txst->vtxs_tso; 3043 txacc->vtxs_rescheduled += txst->vtxs_rescheduled; 3044 } 3045 } 3046 3047 static uint64_t 3048 vtnet_get_counter(if_t ifp, ift_counter cnt) 3049 { 3050 struct vtnet_softc *sc; 3051 struct vtnet_rxq_stats rxaccum; 3052 struct vtnet_txq_stats txaccum; 3053 3054 sc = if_getsoftc(ifp); 3055 vtnet_accum_stats(sc, &rxaccum, &txaccum); 3056 3057 switch (cnt) { 3058 case IFCOUNTER_IPACKETS: 3059 return (rxaccum.vrxs_ipackets); 3060 case IFCOUNTER_IQDROPS: 3061 return (rxaccum.vrxs_iqdrops); 3062 case IFCOUNTER_IERRORS: 3063 return (rxaccum.vrxs_ierrors); 3064 case IFCOUNTER_IBYTES: 3065 return (rxaccum.vrxs_ibytes); 3066 case IFCOUNTER_OPACKETS: 3067 return (txaccum.vtxs_opackets); 3068 case IFCOUNTER_OBYTES: 3069 return (txaccum.vtxs_obytes); 3070 case IFCOUNTER_OMCASTS: 3071 return (txaccum.vtxs_omcasts); 3072 default: 3073 return (if_get_counter_default(ifp, cnt)); 3074 } 3075 } 3076 3077 static void 3078 vtnet_tick(void *xsc) 3079 { 3080 struct vtnet_softc *sc; 3081 if_t ifp; 3082 int i, timedout; 3083 3084 sc = xsc; 3085 ifp = sc->vtnet_ifp; 3086 timedout = 0; 3087 3088 VTNET_CORE_LOCK_ASSERT(sc); 3089 3090 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 3091 timedout |= vtnet_watchdog(&sc->vtnet_txqs[i]); 3092 3093 if (timedout != 0) { 3094 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 3095 vtnet_init_locked(sc, 0); 3096 } else 3097 callout_schedule(&sc->vtnet_tick_ch, hz); 3098 } 3099 3100 static void 3101 vtnet_start_taskqueues(struct vtnet_softc *sc) 3102 { 3103 device_t dev; 3104 struct vtnet_rxq *rxq; 3105 struct vtnet_txq *txq; 3106 int i, error; 3107 3108 dev = sc->vtnet_dev; 3109 3110 /* 3111 * Errors here are very difficult to recover from - we cannot 3112 * easily fail because, if this is during boot, we will hang 3113 * when freeing any successfully started taskqueues because 3114 * the scheduler isn't up yet. 3115 * 3116 * Most drivers just ignore the return value - it only fails 3117 * with ENOMEM so an error is not likely. 3118 */ 3119 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) { 3120 rxq = &sc->vtnet_rxqs[i]; 3121 error = taskqueue_start_threads(&rxq->vtnrx_tq, 1, PI_NET, 3122 "%s rxq %d", device_get_nameunit(dev), rxq->vtnrx_id); 3123 if (error) { 3124 device_printf(dev, "failed to start rx taskq %d\n", 3125 rxq->vtnrx_id); 3126 } 3127 3128 txq = &sc->vtnet_txqs[i]; 3129 error = taskqueue_start_threads(&txq->vtntx_tq, 1, PI_NET, 3130 "%s txq %d", device_get_nameunit(dev), txq->vtntx_id); 3131 if (error) { 3132 device_printf(dev, "failed to start tx taskq %d\n", 3133 txq->vtntx_id); 3134 } 3135 } 3136 } 3137 3138 static void 3139 vtnet_free_taskqueues(struct vtnet_softc *sc) 3140 { 3141 struct vtnet_rxq *rxq; 3142 struct vtnet_txq *txq; 3143 int i; 3144 3145 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3146 rxq = &sc->vtnet_rxqs[i]; 3147 if (rxq->vtnrx_tq != NULL) { 3148 taskqueue_free(rxq->vtnrx_tq); 3149 rxq->vtnrx_tq = NULL; 3150 } 3151 3152 txq = &sc->vtnet_txqs[i]; 3153 if (txq->vtntx_tq != NULL) { 3154 taskqueue_free(txq->vtntx_tq); 3155 txq->vtntx_tq = NULL; 3156 } 3157 } 3158 } 3159 3160 static void 3161 vtnet_drain_taskqueues(struct vtnet_softc *sc) 3162 { 3163 struct vtnet_rxq *rxq; 3164 struct vtnet_txq *txq; 3165 int i; 3166 3167 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3168 rxq = &sc->vtnet_rxqs[i]; 3169 if (rxq->vtnrx_tq != NULL) 3170 taskqueue_drain(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 3171 3172 txq = &sc->vtnet_txqs[i]; 3173 if (txq->vtntx_tq != NULL) { 3174 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_intrtask); 3175 if (!VTNET_ALTQ_ENABLED) 3176 taskqueue_drain(txq->vtntx_tq, &txq->vtntx_defrtask); 3177 } 3178 } 3179 } 3180 3181 static void 3182 vtnet_drain_rxtx_queues(struct vtnet_softc *sc) 3183 { 3184 struct vtnet_rxq *rxq; 3185 struct vtnet_txq *txq; 3186 int i; 3187 3188 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3189 rxq = &sc->vtnet_rxqs[i]; 3190 vtnet_rxq_free_mbufs(rxq); 3191 3192 txq = &sc->vtnet_txqs[i]; 3193 vtnet_txq_free_mbufs(txq); 3194 } 3195 } 3196 3197 static void 3198 vtnet_stop_rendezvous(struct vtnet_softc *sc) 3199 { 3200 struct vtnet_rxq *rxq; 3201 struct vtnet_txq *txq; 3202 int i; 3203 3204 VTNET_CORE_LOCK_ASSERT(sc); 3205 3206 /* 3207 * Lock and unlock the per-queue mutex so we known the stop 3208 * state is visible. Doing only the active queues should be 3209 * sufficient, but it does not cost much extra to do all the 3210 * queues. 3211 */ 3212 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 3213 rxq = &sc->vtnet_rxqs[i]; 3214 VTNET_RXQ_LOCK(rxq); 3215 VTNET_RXQ_UNLOCK(rxq); 3216 3217 txq = &sc->vtnet_txqs[i]; 3218 VTNET_TXQ_LOCK(txq); 3219 VTNET_TXQ_UNLOCK(txq); 3220 } 3221 } 3222 3223 static void 3224 vtnet_stop(struct vtnet_softc *sc) 3225 { 3226 device_t dev; 3227 if_t ifp; 3228 3229 dev = sc->vtnet_dev; 3230 ifp = sc->vtnet_ifp; 3231 3232 VTNET_CORE_LOCK_ASSERT(sc); 3233 3234 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 3235 sc->vtnet_link_active = 0; 3236 callout_stop(&sc->vtnet_tick_ch); 3237 3238 /* Only advisory. */ 3239 vtnet_disable_interrupts(sc); 3240 3241 #ifdef DEV_NETMAP 3242 /* Stop any pending txsync/rxsync and disable them. */ 3243 netmap_disable_all_rings(ifp); 3244 #endif /* DEV_NETMAP */ 3245 3246 /* 3247 * Stop the host adapter. This resets it to the pre-initialized 3248 * state. It will not generate any interrupts until after it is 3249 * reinitialized. 3250 */ 3251 virtio_stop(dev); 3252 vtnet_stop_rendezvous(sc); 3253 3254 vtnet_drain_rxtx_queues(sc); 3255 sc->vtnet_act_vq_pairs = 1; 3256 } 3257 3258 static int 3259 vtnet_virtio_reinit(struct vtnet_softc *sc) 3260 { 3261 device_t dev; 3262 if_t ifp; 3263 uint64_t features; 3264 int error; 3265 3266 dev = sc->vtnet_dev; 3267 ifp = sc->vtnet_ifp; 3268 features = sc->vtnet_negotiated_features; 3269 3270 /* 3271 * Re-negotiate with the host, removing any disabled receive 3272 * features. Transmit features are disabled only on our side 3273 * via if_capenable and if_hwassist. 3274 */ 3275 3276 if ((if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) == 0) 3277 features &= ~(VIRTIO_NET_F_GUEST_CSUM | VTNET_LRO_FEATURES); 3278 3279 if ((if_getcapenable(ifp) & IFCAP_LRO) == 0) 3280 features &= ~VTNET_LRO_FEATURES; 3281 3282 if ((if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) == 0) 3283 features &= ~VIRTIO_NET_F_CTRL_VLAN; 3284 3285 error = virtio_reinit(dev, features); 3286 if (error) { 3287 device_printf(dev, "virtio reinit error %d\n", error); 3288 return (error); 3289 } 3290 3291 sc->vtnet_features = features; 3292 virtio_reinit_complete(dev); 3293 3294 return (0); 3295 } 3296 3297 static void 3298 vtnet_init_rx_filters(struct vtnet_softc *sc) 3299 { 3300 if_t ifp; 3301 3302 ifp = sc->vtnet_ifp; 3303 3304 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) { 3305 vtnet_rx_filter(sc); 3306 vtnet_rx_filter_mac(sc); 3307 } 3308 3309 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER) 3310 vtnet_rx_filter_vlan(sc); 3311 } 3312 3313 static int 3314 vtnet_init_rx_queues(struct vtnet_softc *sc) 3315 { 3316 device_t dev; 3317 if_t ifp; 3318 struct vtnet_rxq *rxq; 3319 int i, clustersz, error; 3320 3321 dev = sc->vtnet_dev; 3322 ifp = sc->vtnet_ifp; 3323 3324 clustersz = vtnet_rx_cluster_size(sc, if_getmtu(ifp)); 3325 sc->vtnet_rx_clustersz = clustersz; 3326 3327 if (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) { 3328 sc->vtnet_rx_nmbufs = howmany(sizeof(struct vtnet_rx_header) + 3329 VTNET_MAX_RX_SIZE, clustersz); 3330 KASSERT(sc->vtnet_rx_nmbufs < sc->vtnet_rx_nsegs, 3331 ("%s: too many rx mbufs %d for %d segments", __func__, 3332 sc->vtnet_rx_nmbufs, sc->vtnet_rx_nsegs)); 3333 } else 3334 sc->vtnet_rx_nmbufs = 1; 3335 3336 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3337 rxq = &sc->vtnet_rxqs[i]; 3338 3339 /* Hold the lock to satisfy asserts. */ 3340 VTNET_RXQ_LOCK(rxq); 3341 error = vtnet_rxq_populate(rxq); 3342 VTNET_RXQ_UNLOCK(rxq); 3343 3344 if (error) { 3345 device_printf(dev, "cannot populate Rx queue %d\n", i); 3346 return (error); 3347 } 3348 } 3349 3350 return (0); 3351 } 3352 3353 static int 3354 vtnet_init_tx_queues(struct vtnet_softc *sc) 3355 { 3356 struct vtnet_txq *txq; 3357 int i; 3358 3359 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 3360 txq = &sc->vtnet_txqs[i]; 3361 txq->vtntx_watchdog = 0; 3362 txq->vtntx_intr_threshold = vtnet_txq_intr_threshold(txq); 3363 #ifdef DEV_NETMAP 3364 netmap_reset(NA(sc->vtnet_ifp), NR_TX, i, 0); 3365 #endif /* DEV_NETMAP */ 3366 } 3367 3368 return (0); 3369 } 3370 3371 static int 3372 vtnet_init_rxtx_queues(struct vtnet_softc *sc) 3373 { 3374 int error; 3375 3376 error = vtnet_init_rx_queues(sc); 3377 if (error) 3378 return (error); 3379 3380 error = vtnet_init_tx_queues(sc); 3381 if (error) 3382 return (error); 3383 3384 return (0); 3385 } 3386 3387 static void 3388 vtnet_set_active_vq_pairs(struct vtnet_softc *sc) 3389 { 3390 device_t dev; 3391 int npairs; 3392 3393 dev = sc->vtnet_dev; 3394 3395 if ((sc->vtnet_flags & VTNET_FLAG_MQ) == 0) { 3396 sc->vtnet_act_vq_pairs = 1; 3397 return; 3398 } 3399 3400 npairs = sc->vtnet_req_vq_pairs; 3401 3402 if (vtnet_ctrl_mq_cmd(sc, npairs) != 0) { 3403 device_printf(dev, "cannot set active queue pairs to %d, " 3404 "falling back to 1 queue pair\n", npairs); 3405 npairs = 1; 3406 } 3407 3408 sc->vtnet_act_vq_pairs = npairs; 3409 } 3410 3411 static void 3412 vtnet_update_rx_offloads(struct vtnet_softc *sc) 3413 { 3414 if_t ifp; 3415 uint64_t features; 3416 int error; 3417 3418 ifp = sc->vtnet_ifp; 3419 features = sc->vtnet_features; 3420 3421 VTNET_CORE_LOCK_ASSERT(sc); 3422 3423 if (if_getcapabilities(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) { 3424 if (if_getcapenable(ifp) & (IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6)) 3425 features |= VIRTIO_NET_F_GUEST_CSUM; 3426 else 3427 features &= ~VIRTIO_NET_F_GUEST_CSUM; 3428 } 3429 3430 if (if_getcapabilities(ifp) & IFCAP_LRO && !vtnet_software_lro(sc)) { 3431 if (if_getcapenable(ifp) & IFCAP_LRO) 3432 features |= VTNET_LRO_FEATURES; 3433 else 3434 features &= ~VTNET_LRO_FEATURES; 3435 } 3436 3437 error = vtnet_ctrl_guest_offloads(sc, 3438 features & (VIRTIO_NET_F_GUEST_CSUM | VIRTIO_NET_F_GUEST_TSO4 | 3439 VIRTIO_NET_F_GUEST_TSO6 | VIRTIO_NET_F_GUEST_ECN | 3440 VIRTIO_NET_F_GUEST_UFO)); 3441 if (error) { 3442 device_printf(sc->vtnet_dev, 3443 "%s: cannot update Rx features\n", __func__); 3444 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) { 3445 if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING); 3446 vtnet_init_locked(sc, 0); 3447 } 3448 } else 3449 sc->vtnet_features = features; 3450 } 3451 3452 static int 3453 vtnet_reinit(struct vtnet_softc *sc) 3454 { 3455 if_t ifp; 3456 int error; 3457 3458 ifp = sc->vtnet_ifp; 3459 3460 bcopy(if_getlladdr(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN); 3461 3462 error = vtnet_virtio_reinit(sc); 3463 if (error) 3464 return (error); 3465 3466 vtnet_set_macaddr(sc); 3467 vtnet_set_active_vq_pairs(sc); 3468 3469 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) 3470 vtnet_init_rx_filters(sc); 3471 3472 if_sethwassist(ifp, 0); 3473 if (if_getcapenable(ifp) & IFCAP_TXCSUM) 3474 if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD, 0); 3475 if (if_getcapenable(ifp) & IFCAP_TXCSUM_IPV6) 3476 if_sethwassistbits(ifp, VTNET_CSUM_OFFLOAD_IPV6, 0); 3477 if (if_getcapenable(ifp) & IFCAP_TSO4) 3478 if_sethwassistbits(ifp, CSUM_IP_TSO, 0); 3479 if (if_getcapenable(ifp) & IFCAP_TSO6) 3480 if_sethwassistbits(ifp, CSUM_IP6_TSO, 0); 3481 3482 error = vtnet_init_rxtx_queues(sc); 3483 if (error) 3484 return (error); 3485 3486 return (0); 3487 } 3488 3489 static void 3490 vtnet_init_locked(struct vtnet_softc *sc, int init_mode) 3491 { 3492 if_t ifp; 3493 3494 ifp = sc->vtnet_ifp; 3495 3496 VTNET_CORE_LOCK_ASSERT(sc); 3497 3498 if (if_getdrvflags(ifp) & IFF_DRV_RUNNING) 3499 return; 3500 3501 vtnet_stop(sc); 3502 3503 #ifdef DEV_NETMAP 3504 /* Once stopped we can update the netmap flags, if necessary. */ 3505 switch (init_mode) { 3506 case VTNET_INIT_NETMAP_ENTER: 3507 nm_set_native_flags(NA(ifp)); 3508 break; 3509 case VTNET_INIT_NETMAP_EXIT: 3510 nm_clear_native_flags(NA(ifp)); 3511 break; 3512 } 3513 #endif /* DEV_NETMAP */ 3514 3515 if (vtnet_reinit(sc) != 0) { 3516 vtnet_stop(sc); 3517 return; 3518 } 3519 3520 if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0); 3521 vtnet_update_link_status(sc); 3522 vtnet_enable_interrupts(sc); 3523 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc); 3524 3525 #ifdef DEV_NETMAP 3526 /* Re-enable txsync/rxsync. */ 3527 netmap_enable_all_rings(ifp); 3528 #endif /* DEV_NETMAP */ 3529 } 3530 3531 static void 3532 vtnet_init(void *xsc) 3533 { 3534 struct vtnet_softc *sc; 3535 3536 sc = xsc; 3537 3538 VTNET_CORE_LOCK(sc); 3539 vtnet_init_locked(sc, 0); 3540 VTNET_CORE_UNLOCK(sc); 3541 } 3542 3543 static void 3544 vtnet_free_ctrl_vq(struct vtnet_softc *sc) 3545 { 3546 3547 /* 3548 * The control virtqueue is only polled and therefore it should 3549 * already be empty. 3550 */ 3551 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq), 3552 ("%s: ctrl vq %p not empty", __func__, sc->vtnet_ctrl_vq)); 3553 } 3554 3555 static void 3556 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie, 3557 struct sglist *sg, int readable, int writable) 3558 { 3559 struct virtqueue *vq; 3560 3561 vq = sc->vtnet_ctrl_vq; 3562 3563 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ); 3564 VTNET_CORE_LOCK_ASSERT(sc); 3565 3566 if (!virtqueue_empty(vq)) 3567 return; 3568 3569 /* 3570 * Poll for the response, but the command is likely completed before 3571 * returning from the notify. 3572 */ 3573 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) == 0) { 3574 virtqueue_notify(vq); 3575 virtqueue_poll(vq, NULL); 3576 } 3577 } 3578 3579 static int 3580 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr) 3581 { 3582 struct sglist_seg segs[3]; 3583 struct sglist sg; 3584 struct { 3585 struct virtio_net_ctrl_hdr hdr __aligned(2); 3586 uint8_t pad1; 3587 uint8_t addr[ETHER_ADDR_LEN] __aligned(8); 3588 uint8_t pad2; 3589 uint8_t ack; 3590 } s; 3591 int error; 3592 3593 error = 0; 3594 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_MAC); 3595 3596 s.hdr.class = VIRTIO_NET_CTRL_MAC; 3597 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET; 3598 bcopy(hwaddr, &s.addr[0], ETHER_ADDR_LEN); 3599 s.ack = VIRTIO_NET_ERR; 3600 3601 sglist_init(&sg, nitems(segs), segs); 3602 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3603 error |= sglist_append(&sg, &s.addr[0], ETHER_ADDR_LEN); 3604 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3605 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3606 3607 if (error == 0) 3608 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3609 3610 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3611 } 3612 3613 static int 3614 vtnet_ctrl_guest_offloads(struct vtnet_softc *sc, uint64_t offloads) 3615 { 3616 struct sglist_seg segs[3]; 3617 struct sglist sg; 3618 struct { 3619 struct virtio_net_ctrl_hdr hdr __aligned(2); 3620 uint8_t pad1; 3621 uint64_t offloads __aligned(8); 3622 uint8_t pad2; 3623 uint8_t ack; 3624 } s; 3625 int error; 3626 3627 error = 0; 3628 MPASS(sc->vtnet_features & VIRTIO_NET_F_CTRL_GUEST_OFFLOADS); 3629 3630 s.hdr.class = VIRTIO_NET_CTRL_GUEST_OFFLOADS; 3631 s.hdr.cmd = VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET; 3632 s.offloads = vtnet_gtoh64(sc, offloads); 3633 s.ack = VIRTIO_NET_ERR; 3634 3635 sglist_init(&sg, nitems(segs), segs); 3636 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3637 error |= sglist_append(&sg, &s.offloads, sizeof(uint64_t)); 3638 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3639 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3640 3641 if (error == 0) 3642 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3643 3644 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3645 } 3646 3647 static int 3648 vtnet_ctrl_mq_cmd(struct vtnet_softc *sc, uint16_t npairs) 3649 { 3650 struct sglist_seg segs[3]; 3651 struct sglist sg; 3652 struct { 3653 struct virtio_net_ctrl_hdr hdr __aligned(2); 3654 uint8_t pad1; 3655 struct virtio_net_ctrl_mq mq __aligned(2); 3656 uint8_t pad2; 3657 uint8_t ack; 3658 } s; 3659 int error; 3660 3661 error = 0; 3662 MPASS(sc->vtnet_flags & VTNET_FLAG_MQ); 3663 3664 s.hdr.class = VIRTIO_NET_CTRL_MQ; 3665 s.hdr.cmd = VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET; 3666 s.mq.virtqueue_pairs = vtnet_gtoh16(sc, npairs); 3667 s.ack = VIRTIO_NET_ERR; 3668 3669 sglist_init(&sg, nitems(segs), segs); 3670 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3671 error |= sglist_append(&sg, &s.mq, sizeof(struct virtio_net_ctrl_mq)); 3672 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3673 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3674 3675 if (error == 0) 3676 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3677 3678 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3679 } 3680 3681 static int 3682 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, uint8_t cmd, bool on) 3683 { 3684 struct sglist_seg segs[3]; 3685 struct sglist sg; 3686 struct { 3687 struct virtio_net_ctrl_hdr hdr __aligned(2); 3688 uint8_t pad1; 3689 uint8_t onoff; 3690 uint8_t pad2; 3691 uint8_t ack; 3692 } s; 3693 int error; 3694 3695 error = 0; 3696 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX); 3697 3698 s.hdr.class = VIRTIO_NET_CTRL_RX; 3699 s.hdr.cmd = cmd; 3700 s.onoff = on; 3701 s.ack = VIRTIO_NET_ERR; 3702 3703 sglist_init(&sg, nitems(segs), segs); 3704 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3705 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t)); 3706 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3707 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3708 3709 if (error == 0) 3710 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3711 3712 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3713 } 3714 3715 static int 3716 vtnet_set_promisc(struct vtnet_softc *sc, bool on) 3717 { 3718 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on)); 3719 } 3720 3721 static int 3722 vtnet_set_allmulti(struct vtnet_softc *sc, bool on) 3723 { 3724 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on)); 3725 } 3726 3727 static void 3728 vtnet_rx_filter(struct vtnet_softc *sc) 3729 { 3730 device_t dev; 3731 if_t ifp; 3732 3733 dev = sc->vtnet_dev; 3734 ifp = sc->vtnet_ifp; 3735 3736 VTNET_CORE_LOCK_ASSERT(sc); 3737 3738 if (vtnet_set_promisc(sc, if_getflags(ifp) & IFF_PROMISC) != 0) { 3739 device_printf(dev, "cannot %s promiscuous mode\n", 3740 if_getflags(ifp) & IFF_PROMISC ? "enable" : "disable"); 3741 } 3742 3743 if (vtnet_set_allmulti(sc, if_getflags(ifp) & IFF_ALLMULTI) != 0) { 3744 device_printf(dev, "cannot %s all-multicast mode\n", 3745 if_getflags(ifp) & IFF_ALLMULTI ? "enable" : "disable"); 3746 } 3747 } 3748 3749 static u_int 3750 vtnet_copy_ifaddr(void *arg, struct sockaddr_dl *sdl, u_int ucnt) 3751 { 3752 struct vtnet_softc *sc = arg; 3753 3754 if (memcmp(LLADDR(sdl), sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0) 3755 return (0); 3756 3757 if (ucnt < VTNET_MAX_MAC_ENTRIES) 3758 bcopy(LLADDR(sdl), 3759 &sc->vtnet_mac_filter->vmf_unicast.macs[ucnt], 3760 ETHER_ADDR_LEN); 3761 3762 return (1); 3763 } 3764 3765 static u_int 3766 vtnet_copy_maddr(void *arg, struct sockaddr_dl *sdl, u_int mcnt) 3767 { 3768 struct vtnet_mac_filter *filter = arg; 3769 3770 if (mcnt < VTNET_MAX_MAC_ENTRIES) 3771 bcopy(LLADDR(sdl), &filter->vmf_multicast.macs[mcnt], 3772 ETHER_ADDR_LEN); 3773 3774 return (1); 3775 } 3776 3777 static void 3778 vtnet_rx_filter_mac(struct vtnet_softc *sc) 3779 { 3780 struct virtio_net_ctrl_hdr hdr __aligned(2); 3781 struct vtnet_mac_filter *filter; 3782 struct sglist_seg segs[4]; 3783 struct sglist sg; 3784 if_t ifp; 3785 bool promisc, allmulti; 3786 u_int ucnt, mcnt; 3787 int error; 3788 uint8_t ack; 3789 3790 ifp = sc->vtnet_ifp; 3791 filter = sc->vtnet_mac_filter; 3792 error = 0; 3793 3794 MPASS(sc->vtnet_flags & VTNET_FLAG_CTRL_RX); 3795 VTNET_CORE_LOCK_ASSERT(sc); 3796 3797 /* Unicast MAC addresses: */ 3798 ucnt = if_foreach_lladdr(ifp, vtnet_copy_ifaddr, sc); 3799 promisc = (ucnt > VTNET_MAX_MAC_ENTRIES); 3800 3801 if (promisc) { 3802 ucnt = 0; 3803 if_printf(ifp, "more than %d MAC addresses assigned, " 3804 "falling back to promiscuous mode\n", 3805 VTNET_MAX_MAC_ENTRIES); 3806 } 3807 3808 /* Multicast MAC addresses: */ 3809 mcnt = if_foreach_llmaddr(ifp, vtnet_copy_maddr, filter); 3810 allmulti = (mcnt > VTNET_MAX_MAC_ENTRIES); 3811 3812 if (allmulti) { 3813 mcnt = 0; 3814 if_printf(ifp, "more than %d multicast MAC addresses " 3815 "assigned, falling back to all-multicast mode\n", 3816 VTNET_MAX_MAC_ENTRIES); 3817 } 3818 3819 if (promisc && allmulti) 3820 goto out; 3821 3822 filter->vmf_unicast.nentries = vtnet_gtoh32(sc, ucnt); 3823 filter->vmf_multicast.nentries = vtnet_gtoh32(sc, mcnt); 3824 3825 hdr.class = VIRTIO_NET_CTRL_MAC; 3826 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET; 3827 ack = VIRTIO_NET_ERR; 3828 3829 sglist_init(&sg, nitems(segs), segs); 3830 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr)); 3831 error |= sglist_append(&sg, &filter->vmf_unicast, 3832 sizeof(uint32_t) + ucnt * ETHER_ADDR_LEN); 3833 error |= sglist_append(&sg, &filter->vmf_multicast, 3834 sizeof(uint32_t) + mcnt * ETHER_ADDR_LEN); 3835 error |= sglist_append(&sg, &ack, sizeof(uint8_t)); 3836 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3837 3838 if (error == 0) 3839 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1); 3840 if (ack != VIRTIO_NET_OK) 3841 if_printf(ifp, "error setting host MAC filter table\n"); 3842 3843 out: 3844 if (promisc && vtnet_set_promisc(sc, true) != 0) 3845 if_printf(ifp, "cannot enable promiscuous mode\n"); 3846 if (allmulti && vtnet_set_allmulti(sc, true) != 0) 3847 if_printf(ifp, "cannot enable all-multicast mode\n"); 3848 } 3849 3850 static int 3851 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3852 { 3853 struct sglist_seg segs[3]; 3854 struct sglist sg; 3855 struct { 3856 struct virtio_net_ctrl_hdr hdr __aligned(2); 3857 uint8_t pad1; 3858 uint16_t tag __aligned(2); 3859 uint8_t pad2; 3860 uint8_t ack; 3861 } s; 3862 int error; 3863 3864 error = 0; 3865 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER); 3866 3867 s.hdr.class = VIRTIO_NET_CTRL_VLAN; 3868 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL; 3869 s.tag = vtnet_gtoh16(sc, tag); 3870 s.ack = VIRTIO_NET_ERR; 3871 3872 sglist_init(&sg, nitems(segs), segs); 3873 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr)); 3874 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t)); 3875 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t)); 3876 MPASS(error == 0 && sg.sg_nseg == nitems(segs)); 3877 3878 if (error == 0) 3879 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1); 3880 3881 return (s.ack == VIRTIO_NET_OK ? 0 : EIO); 3882 } 3883 3884 static void 3885 vtnet_rx_filter_vlan(struct vtnet_softc *sc) 3886 { 3887 int i, bit; 3888 uint32_t w; 3889 uint16_t tag; 3890 3891 MPASS(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER); 3892 VTNET_CORE_LOCK_ASSERT(sc); 3893 3894 /* Enable the filter for each configured VLAN. */ 3895 for (i = 0; i < VTNET_VLAN_FILTER_NWORDS; i++) { 3896 w = sc->vtnet_vlan_filter[i]; 3897 3898 while ((bit = ffs(w) - 1) != -1) { 3899 w &= ~(1 << bit); 3900 tag = sizeof(w) * CHAR_BIT * i + bit; 3901 3902 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) { 3903 device_printf(sc->vtnet_dev, 3904 "cannot enable VLAN %d filter\n", tag); 3905 } 3906 } 3907 } 3908 } 3909 3910 static void 3911 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag) 3912 { 3913 if_t ifp; 3914 int idx, bit; 3915 3916 ifp = sc->vtnet_ifp; 3917 idx = (tag >> 5) & 0x7F; 3918 bit = tag & 0x1F; 3919 3920 if (tag == 0 || tag > 4095) 3921 return; 3922 3923 VTNET_CORE_LOCK(sc); 3924 3925 if (add) 3926 sc->vtnet_vlan_filter[idx] |= (1 << bit); 3927 else 3928 sc->vtnet_vlan_filter[idx] &= ~(1 << bit); 3929 3930 if (if_getcapenable(ifp) & IFCAP_VLAN_HWFILTER && 3931 if_getdrvflags(ifp) & IFF_DRV_RUNNING && 3932 vtnet_exec_vlan_filter(sc, add, tag) != 0) { 3933 device_printf(sc->vtnet_dev, 3934 "cannot %s VLAN %d %s the host filter table\n", 3935 add ? "add" : "remove", tag, add ? "to" : "from"); 3936 } 3937 3938 VTNET_CORE_UNLOCK(sc); 3939 } 3940 3941 static void 3942 vtnet_register_vlan(void *arg, if_t ifp, uint16_t tag) 3943 { 3944 3945 if (if_getsoftc(ifp) != arg) 3946 return; 3947 3948 vtnet_update_vlan_filter(arg, 1, tag); 3949 } 3950 3951 static void 3952 vtnet_unregister_vlan(void *arg, if_t ifp, uint16_t tag) 3953 { 3954 3955 if (if_getsoftc(ifp) != arg) 3956 return; 3957 3958 vtnet_update_vlan_filter(arg, 0, tag); 3959 } 3960 3961 static void 3962 vtnet_update_speed_duplex(struct vtnet_softc *sc) 3963 { 3964 if_t ifp; 3965 uint32_t speed; 3966 3967 ifp = sc->vtnet_ifp; 3968 3969 if ((sc->vtnet_features & VIRTIO_NET_F_SPEED_DUPLEX) == 0) 3970 return; 3971 3972 /* BMV: Ignore duplex. */ 3973 speed = virtio_read_dev_config_4(sc->vtnet_dev, 3974 offsetof(struct virtio_net_config, speed)); 3975 if (speed != UINT32_MAX) 3976 if_setbaudrate(ifp, IF_Mbps(speed)); 3977 } 3978 3979 static int 3980 vtnet_is_link_up(struct vtnet_softc *sc) 3981 { 3982 uint16_t status; 3983 3984 if ((sc->vtnet_features & VIRTIO_NET_F_STATUS) == 0) 3985 return (1); 3986 3987 status = virtio_read_dev_config_2(sc->vtnet_dev, 3988 offsetof(struct virtio_net_config, status)); 3989 3990 return ((status & VIRTIO_NET_S_LINK_UP) != 0); 3991 } 3992 3993 static void 3994 vtnet_update_link_status(struct vtnet_softc *sc) 3995 { 3996 if_t ifp; 3997 int link; 3998 3999 ifp = sc->vtnet_ifp; 4000 VTNET_CORE_LOCK_ASSERT(sc); 4001 link = vtnet_is_link_up(sc); 4002 4003 /* Notify if the link status has changed. */ 4004 if (link != 0 && sc->vtnet_link_active == 0) { 4005 vtnet_update_speed_duplex(sc); 4006 sc->vtnet_link_active = 1; 4007 if_link_state_change(ifp, LINK_STATE_UP); 4008 } else if (link == 0 && sc->vtnet_link_active != 0) { 4009 sc->vtnet_link_active = 0; 4010 if_link_state_change(ifp, LINK_STATE_DOWN); 4011 } 4012 } 4013 4014 static int 4015 vtnet_ifmedia_upd(if_t ifp __unused) 4016 { 4017 return (EOPNOTSUPP); 4018 } 4019 4020 static void 4021 vtnet_ifmedia_sts(if_t ifp, struct ifmediareq *ifmr) 4022 { 4023 struct vtnet_softc *sc; 4024 4025 sc = if_getsoftc(ifp); 4026 4027 ifmr->ifm_status = IFM_AVALID; 4028 ifmr->ifm_active = IFM_ETHER; 4029 4030 VTNET_CORE_LOCK(sc); 4031 if (vtnet_is_link_up(sc) != 0) { 4032 ifmr->ifm_status |= IFM_ACTIVE; 4033 ifmr->ifm_active |= IFM_10G_T | IFM_FDX; 4034 } else 4035 ifmr->ifm_active |= IFM_NONE; 4036 VTNET_CORE_UNLOCK(sc); 4037 } 4038 4039 static void 4040 vtnet_get_macaddr(struct vtnet_softc *sc) 4041 { 4042 4043 if (sc->vtnet_flags & VTNET_FLAG_MAC) { 4044 virtio_read_device_config_array(sc->vtnet_dev, 4045 offsetof(struct virtio_net_config, mac), 4046 &sc->vtnet_hwaddr[0], sizeof(uint8_t), ETHER_ADDR_LEN); 4047 } else { 4048 /* Generate a random locally administered unicast address. */ 4049 sc->vtnet_hwaddr[0] = 0xB2; 4050 arc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1, 0); 4051 } 4052 } 4053 4054 static void 4055 vtnet_set_macaddr(struct vtnet_softc *sc) 4056 { 4057 device_t dev; 4058 int error; 4059 4060 dev = sc->vtnet_dev; 4061 4062 if (sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) { 4063 error = vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr); 4064 if (error) 4065 device_printf(dev, "unable to set MAC address\n"); 4066 return; 4067 } 4068 4069 /* MAC in config is read-only in modern VirtIO. */ 4070 if (!vtnet_modern(sc) && sc->vtnet_flags & VTNET_FLAG_MAC) { 4071 for (int i = 0; i < ETHER_ADDR_LEN; i++) { 4072 virtio_write_dev_config_1(dev, 4073 offsetof(struct virtio_net_config, mac) + i, 4074 sc->vtnet_hwaddr[i]); 4075 } 4076 } 4077 } 4078 4079 static void 4080 vtnet_attached_set_macaddr(struct vtnet_softc *sc) 4081 { 4082 4083 /* Assign MAC address if it was generated. */ 4084 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) 4085 vtnet_set_macaddr(sc); 4086 } 4087 4088 static void 4089 vtnet_vlan_tag_remove(struct mbuf *m) 4090 { 4091 struct ether_vlan_header *evh; 4092 4093 evh = mtod(m, struct ether_vlan_header *); 4094 m->m_pkthdr.ether_vtag = ntohs(evh->evl_tag); 4095 m->m_flags |= M_VLANTAG; 4096 4097 /* Strip the 802.1Q header. */ 4098 bcopy((char *) evh, (char *) evh + ETHER_VLAN_ENCAP_LEN, 4099 ETHER_HDR_LEN - ETHER_TYPE_LEN); 4100 m_adj(m, ETHER_VLAN_ENCAP_LEN); 4101 } 4102 4103 static void 4104 vtnet_set_rx_process_limit(struct vtnet_softc *sc) 4105 { 4106 int limit; 4107 4108 limit = vtnet_tunable_int(sc, "rx_process_limit", 4109 vtnet_rx_process_limit); 4110 if (limit < 0) 4111 limit = INT_MAX; 4112 sc->vtnet_rx_process_limit = limit; 4113 } 4114 4115 static void 4116 vtnet_setup_rxq_sysctl(struct sysctl_ctx_list *ctx, 4117 struct sysctl_oid_list *child, struct vtnet_rxq *rxq) 4118 { 4119 struct sysctl_oid *node; 4120 struct sysctl_oid_list *list; 4121 struct vtnet_rxq_stats *stats; 4122 char namebuf[16]; 4123 4124 snprintf(namebuf, sizeof(namebuf), "rxq%d", rxq->vtnrx_id); 4125 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 4126 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Receive Queue"); 4127 list = SYSCTL_CHILDREN(node); 4128 4129 stats = &rxq->vtnrx_stats; 4130 4131 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ipackets", 4132 CTLFLAG_RD | CTLFLAG_STATS, 4133 &stats->vrxs_ipackets, "Receive packets"); 4134 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ibytes", 4135 CTLFLAG_RD | CTLFLAG_STATS, 4136 &stats->vrxs_ibytes, "Receive bytes"); 4137 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "iqdrops", 4138 CTLFLAG_RD | CTLFLAG_STATS, 4139 &stats->vrxs_iqdrops, "Receive drops"); 4140 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "ierrors", 4141 CTLFLAG_RD | CTLFLAG_STATS, 4142 &stats->vrxs_ierrors, "Receive errors"); 4143 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", 4144 CTLFLAG_RD | CTLFLAG_STATS, 4145 &stats->vrxs_csum, "Receive checksum offloaded"); 4146 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum_failed", 4147 CTLFLAG_RD | CTLFLAG_STATS, 4148 &stats->vrxs_csum_failed, "Receive checksum offload failed"); 4149 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "host_lro", 4150 CTLFLAG_RD | CTLFLAG_STATS, 4151 &stats->vrxs_host_lro, "Receive host segmentation offloaded"); 4152 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", 4153 CTLFLAG_RD | CTLFLAG_STATS, 4154 &stats->vrxs_rescheduled, 4155 "Receive interrupt handler rescheduled"); 4156 } 4157 4158 static void 4159 vtnet_setup_txq_sysctl(struct sysctl_ctx_list *ctx, 4160 struct sysctl_oid_list *child, struct vtnet_txq *txq) 4161 { 4162 struct sysctl_oid *node; 4163 struct sysctl_oid_list *list; 4164 struct vtnet_txq_stats *stats; 4165 char namebuf[16]; 4166 4167 snprintf(namebuf, sizeof(namebuf), "txq%d", txq->vtntx_id); 4168 node = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, namebuf, 4169 CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Transmit Queue"); 4170 list = SYSCTL_CHILDREN(node); 4171 4172 stats = &txq->vtntx_stats; 4173 4174 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "opackets", 4175 CTLFLAG_RD | CTLFLAG_STATS, 4176 &stats->vtxs_opackets, "Transmit packets"); 4177 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "obytes", 4178 CTLFLAG_RD | CTLFLAG_STATS, 4179 &stats->vtxs_obytes, "Transmit bytes"); 4180 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "omcasts", 4181 CTLFLAG_RD | CTLFLAG_STATS, 4182 &stats->vtxs_omcasts, "Transmit multicasts"); 4183 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "csum", 4184 CTLFLAG_RD | CTLFLAG_STATS, 4185 &stats->vtxs_csum, "Transmit checksum offloaded"); 4186 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "tso", 4187 CTLFLAG_RD | CTLFLAG_STATS, 4188 &stats->vtxs_tso, "Transmit TCP segmentation offloaded"); 4189 SYSCTL_ADD_UQUAD(ctx, list, OID_AUTO, "rescheduled", 4190 CTLFLAG_RD | CTLFLAG_STATS, 4191 &stats->vtxs_rescheduled, 4192 "Transmit interrupt handler rescheduled"); 4193 } 4194 4195 static void 4196 vtnet_setup_queue_sysctl(struct vtnet_softc *sc) 4197 { 4198 device_t dev; 4199 struct sysctl_ctx_list *ctx; 4200 struct sysctl_oid *tree; 4201 struct sysctl_oid_list *child; 4202 int i; 4203 4204 dev = sc->vtnet_dev; 4205 ctx = device_get_sysctl_ctx(dev); 4206 tree = device_get_sysctl_tree(dev); 4207 child = SYSCTL_CHILDREN(tree); 4208 4209 for (i = 0; i < sc->vtnet_req_vq_pairs; i++) { 4210 vtnet_setup_rxq_sysctl(ctx, child, &sc->vtnet_rxqs[i]); 4211 vtnet_setup_txq_sysctl(ctx, child, &sc->vtnet_txqs[i]); 4212 } 4213 } 4214 4215 static int 4216 vtnet_sysctl_rx_csum_failed(SYSCTL_HANDLER_ARGS) 4217 { 4218 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4219 struct vtnet_statistics *stats = &sc->vtnet_stats; 4220 struct vtnet_rxq_stats *rxst; 4221 int i; 4222 4223 stats->rx_csum_failed = 0; 4224 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 4225 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 4226 stats->rx_csum_failed += rxst->vrxs_csum_failed; 4227 } 4228 return (sysctl_handle_64(oidp, NULL, stats->rx_csum_failed, req)); 4229 } 4230 4231 static int 4232 vtnet_sysctl_rx_csum_offloaded(SYSCTL_HANDLER_ARGS) 4233 { 4234 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4235 struct vtnet_statistics *stats = &sc->vtnet_stats; 4236 struct vtnet_rxq_stats *rxst; 4237 int i; 4238 4239 stats->rx_csum_offloaded = 0; 4240 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 4241 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 4242 stats->rx_csum_offloaded += rxst->vrxs_csum; 4243 } 4244 return (sysctl_handle_64(oidp, NULL, stats->rx_csum_offloaded, req)); 4245 } 4246 4247 static int 4248 vtnet_sysctl_rx_task_rescheduled(SYSCTL_HANDLER_ARGS) 4249 { 4250 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4251 struct vtnet_statistics *stats = &sc->vtnet_stats; 4252 struct vtnet_rxq_stats *rxst; 4253 int i; 4254 4255 stats->rx_task_rescheduled = 0; 4256 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 4257 rxst = &sc->vtnet_rxqs[i].vtnrx_stats; 4258 stats->rx_task_rescheduled += rxst->vrxs_rescheduled; 4259 } 4260 return (sysctl_handle_64(oidp, NULL, stats->rx_task_rescheduled, req)); 4261 } 4262 4263 static int 4264 vtnet_sysctl_tx_csum_offloaded(SYSCTL_HANDLER_ARGS) 4265 { 4266 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4267 struct vtnet_statistics *stats = &sc->vtnet_stats; 4268 struct vtnet_txq_stats *txst; 4269 int i; 4270 4271 stats->tx_csum_offloaded = 0; 4272 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 4273 txst = &sc->vtnet_txqs[i].vtntx_stats; 4274 stats->tx_csum_offloaded += txst->vtxs_csum; 4275 } 4276 return (sysctl_handle_64(oidp, NULL, stats->tx_csum_offloaded, req)); 4277 } 4278 4279 static int 4280 vtnet_sysctl_tx_tso_offloaded(SYSCTL_HANDLER_ARGS) 4281 { 4282 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4283 struct vtnet_statistics *stats = &sc->vtnet_stats; 4284 struct vtnet_txq_stats *txst; 4285 int i; 4286 4287 stats->tx_tso_offloaded = 0; 4288 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 4289 txst = &sc->vtnet_txqs[i].vtntx_stats; 4290 stats->tx_tso_offloaded += txst->vtxs_tso; 4291 } 4292 return (sysctl_handle_64(oidp, NULL, stats->tx_tso_offloaded, req)); 4293 } 4294 4295 static int 4296 vtnet_sysctl_tx_task_rescheduled(SYSCTL_HANDLER_ARGS) 4297 { 4298 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4299 struct vtnet_statistics *stats = &sc->vtnet_stats; 4300 struct vtnet_txq_stats *txst; 4301 int i; 4302 4303 stats->tx_task_rescheduled = 0; 4304 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) { 4305 txst = &sc->vtnet_txqs[i].vtntx_stats; 4306 stats->tx_task_rescheduled += txst->vtxs_rescheduled; 4307 } 4308 return (sysctl_handle_64(oidp, NULL, stats->tx_task_rescheduled, req)); 4309 } 4310 4311 static void 4312 vtnet_setup_stat_sysctl(struct sysctl_ctx_list *ctx, 4313 struct sysctl_oid_list *child, struct vtnet_softc *sc) 4314 { 4315 struct vtnet_statistics *stats; 4316 struct vtnet_rxq_stats rxaccum; 4317 struct vtnet_txq_stats txaccum; 4318 4319 vtnet_accum_stats(sc, &rxaccum, &txaccum); 4320 4321 stats = &sc->vtnet_stats; 4322 stats->rx_csum_offloaded = rxaccum.vrxs_csum; 4323 stats->rx_csum_failed = rxaccum.vrxs_csum_failed; 4324 stats->rx_task_rescheduled = rxaccum.vrxs_rescheduled; 4325 stats->tx_csum_offloaded = txaccum.vtxs_csum; 4326 stats->tx_tso_offloaded = txaccum.vtxs_tso; 4327 stats->tx_task_rescheduled = txaccum.vtxs_rescheduled; 4328 4329 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed", 4330 CTLFLAG_RD | CTLFLAG_STATS, &stats->mbuf_alloc_failed, 4331 "Mbuf cluster allocation failures"); 4332 4333 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large", 4334 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_frame_too_large, 4335 "Received frame larger than the mbuf chain"); 4336 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed", 4337 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_enq_replacement_failed, 4338 "Enqueuing the replacement receive mbuf failed"); 4339 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed", 4340 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_mergeable_failed, 4341 "Mergeable buffers receive failures"); 4342 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype", 4343 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ethtype, 4344 "Received checksum offloaded buffer with unsupported " 4345 "Ethernet type"); 4346 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto", 4347 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_ipproto, 4348 "Received checksum offloaded buffer with incorrect IP protocol"); 4349 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset", 4350 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_bad_offset, 4351 "Received checksum offloaded buffer with incorrect offset"); 4352 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_inaccessible_ipproto", 4353 CTLFLAG_RD | CTLFLAG_STATS, &stats->rx_csum_inaccessible_ipproto, 4354 "Received checksum offloaded buffer with inaccessible IP protocol"); 4355 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_failed", 4356 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS, 4357 sc, 0, vtnet_sysctl_rx_csum_failed, "QU", 4358 "Received buffer checksum offload failed"); 4359 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_csum_offloaded", 4360 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS, 4361 sc, 0, vtnet_sysctl_rx_csum_offloaded, "QU", 4362 "Received buffer checksum offload succeeded"); 4363 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rx_task_rescheduled", 4364 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS, 4365 sc, 0, vtnet_sysctl_rx_task_rescheduled, "QU", 4366 "Times the receive interrupt task rescheduled itself"); 4367 4368 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_unknown_ethtype", 4369 CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_unknown_ethtype, 4370 "Aborted transmit of checksum offloaded buffer with unknown " 4371 "Ethernet type"); 4372 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_proto_mismatch", 4373 CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_csum_proto_mismatch, 4374 "Aborted transmit of checksum offloaded buffer because mismatched " 4375 "protocols"); 4376 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_not_tcp", 4377 CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_not_tcp, 4378 "Aborted transmit of TSO buffer with non TCP protocol"); 4379 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_without_csum", 4380 CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_tso_without_csum, 4381 "Aborted transmit of TSO buffer without TCP checksum offload"); 4382 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged", 4383 CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defragged, 4384 "Transmit mbufs defragged"); 4385 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed", 4386 CTLFLAG_RD | CTLFLAG_STATS, &stats->tx_defrag_failed, 4387 "Aborted transmit of buffer because defrag failed"); 4388 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_csum_offloaded", 4389 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS, 4390 sc, 0, vtnet_sysctl_tx_csum_offloaded, "QU", 4391 "Offloaded checksum of transmitted buffer"); 4392 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_tso_offloaded", 4393 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS, 4394 sc, 0, vtnet_sysctl_tx_tso_offloaded, "QU", 4395 "Segmentation offload of transmitted buffer"); 4396 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_task_rescheduled", 4397 CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_STATS, 4398 sc, 0, vtnet_sysctl_tx_task_rescheduled, "QU", 4399 "Times the transmit interrupt task rescheduled itself"); 4400 } 4401 4402 static int 4403 vtnet_sysctl_features(SYSCTL_HANDLER_ARGS) 4404 { 4405 struct sbuf sb; 4406 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4407 int error; 4408 4409 sbuf_new_for_sysctl(&sb, NULL, 0, req); 4410 sbuf_printf(&sb, "%b", (uint32_t)sc->vtnet_features, 4411 VIRTIO_NET_FEATURE_BITS); 4412 error = sbuf_finish(&sb); 4413 sbuf_delete(&sb); 4414 return (error); 4415 } 4416 4417 static int 4418 vtnet_sysctl_flags(SYSCTL_HANDLER_ARGS) 4419 { 4420 struct sbuf sb; 4421 struct vtnet_softc *sc = (struct vtnet_softc *)arg1; 4422 int error; 4423 4424 sbuf_new_for_sysctl(&sb, NULL, 0, req); 4425 sbuf_printf(&sb, "%b", sc->vtnet_flags, VTNET_FLAGS_BITS); 4426 error = sbuf_finish(&sb); 4427 sbuf_delete(&sb); 4428 return (error); 4429 } 4430 4431 static void 4432 vtnet_setup_sysctl(struct vtnet_softc *sc) 4433 { 4434 device_t dev; 4435 struct sysctl_ctx_list *ctx; 4436 struct sysctl_oid *tree; 4437 struct sysctl_oid_list *child; 4438 4439 dev = sc->vtnet_dev; 4440 ctx = device_get_sysctl_ctx(dev); 4441 tree = device_get_sysctl_tree(dev); 4442 child = SYSCTL_CHILDREN(tree); 4443 4444 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "max_vq_pairs", 4445 CTLFLAG_RD, &sc->vtnet_max_vq_pairs, 0, 4446 "Number of maximum supported virtqueue pairs"); 4447 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "req_vq_pairs", 4448 CTLFLAG_RD, &sc->vtnet_req_vq_pairs, 0, 4449 "Number of requested virtqueue pairs"); 4450 SYSCTL_ADD_INT(ctx, child, OID_AUTO, "act_vq_pairs", 4451 CTLFLAG_RD, &sc->vtnet_act_vq_pairs, 0, 4452 "Number of active virtqueue pairs"); 4453 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "features", 4454 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0, 4455 vtnet_sysctl_features, "A", "Features"); 4456 SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "flags", 4457 CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0, 4458 vtnet_sysctl_flags, "A", "Flags"); 4459 4460 vtnet_setup_stat_sysctl(ctx, child, sc); 4461 } 4462 4463 static void 4464 vtnet_load_tunables(struct vtnet_softc *sc) 4465 { 4466 4467 sc->vtnet_lro_entry_count = vtnet_tunable_int(sc, 4468 "lro_entry_count", vtnet_lro_entry_count); 4469 if (sc->vtnet_lro_entry_count < TCP_LRO_ENTRIES) 4470 sc->vtnet_lro_entry_count = TCP_LRO_ENTRIES; 4471 4472 sc->vtnet_lro_mbufq_depth = vtnet_tunable_int(sc, 4473 "lro_mbufq_depth", vtnet_lro_mbufq_depth); 4474 } 4475 4476 static int 4477 vtnet_rxq_enable_intr(struct vtnet_rxq *rxq) 4478 { 4479 4480 return (virtqueue_enable_intr(rxq->vtnrx_vq)); 4481 } 4482 4483 static void 4484 vtnet_rxq_disable_intr(struct vtnet_rxq *rxq) 4485 { 4486 4487 virtqueue_disable_intr(rxq->vtnrx_vq); 4488 } 4489 4490 static int 4491 vtnet_txq_enable_intr(struct vtnet_txq *txq) 4492 { 4493 struct virtqueue *vq; 4494 4495 vq = txq->vtntx_vq; 4496 4497 if (vtnet_txq_below_threshold(txq) != 0) 4498 return (virtqueue_postpone_intr(vq, VQ_POSTPONE_LONG)); 4499 4500 /* 4501 * The free count is above our threshold. Keep the Tx interrupt 4502 * disabled until the queue is fuller. 4503 */ 4504 return (0); 4505 } 4506 4507 static void 4508 vtnet_txq_disable_intr(struct vtnet_txq *txq) 4509 { 4510 4511 virtqueue_disable_intr(txq->vtntx_vq); 4512 } 4513 4514 static void 4515 vtnet_enable_rx_interrupts(struct vtnet_softc *sc) 4516 { 4517 struct vtnet_rxq *rxq; 4518 int i; 4519 4520 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) { 4521 rxq = &sc->vtnet_rxqs[i]; 4522 if (vtnet_rxq_enable_intr(rxq) != 0) 4523 taskqueue_enqueue(rxq->vtnrx_tq, &rxq->vtnrx_intrtask); 4524 } 4525 } 4526 4527 static void 4528 vtnet_enable_tx_interrupts(struct vtnet_softc *sc) 4529 { 4530 int i; 4531 4532 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4533 vtnet_txq_enable_intr(&sc->vtnet_txqs[i]); 4534 } 4535 4536 static void 4537 vtnet_enable_interrupts(struct vtnet_softc *sc) 4538 { 4539 4540 vtnet_enable_rx_interrupts(sc); 4541 vtnet_enable_tx_interrupts(sc); 4542 } 4543 4544 static void 4545 vtnet_disable_rx_interrupts(struct vtnet_softc *sc) 4546 { 4547 int i; 4548 4549 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4550 vtnet_rxq_disable_intr(&sc->vtnet_rxqs[i]); 4551 } 4552 4553 static void 4554 vtnet_disable_tx_interrupts(struct vtnet_softc *sc) 4555 { 4556 int i; 4557 4558 for (i = 0; i < sc->vtnet_max_vq_pairs; i++) 4559 vtnet_txq_disable_intr(&sc->vtnet_txqs[i]); 4560 } 4561 4562 static void 4563 vtnet_disable_interrupts(struct vtnet_softc *sc) 4564 { 4565 4566 vtnet_disable_rx_interrupts(sc); 4567 vtnet_disable_tx_interrupts(sc); 4568 } 4569 4570 static int 4571 vtnet_tunable_int(struct vtnet_softc *sc, const char *knob, int def) 4572 { 4573 char path[64]; 4574 4575 snprintf(path, sizeof(path), 4576 "hw.vtnet.%d.%s", device_get_unit(sc->vtnet_dev), knob); 4577 TUNABLE_INT_FETCH(path, &def); 4578 4579 return (def); 4580 } 4581 4582 #ifdef DEBUGNET 4583 static void 4584 vtnet_debugnet_init(if_t ifp, int *nrxr, int *ncl, int *clsize) 4585 { 4586 struct vtnet_softc *sc; 4587 4588 sc = if_getsoftc(ifp); 4589 4590 VTNET_CORE_LOCK(sc); 4591 *nrxr = sc->vtnet_req_vq_pairs; 4592 *ncl = DEBUGNET_MAX_IN_FLIGHT; 4593 *clsize = sc->vtnet_rx_clustersz; 4594 VTNET_CORE_UNLOCK(sc); 4595 } 4596 4597 static void 4598 vtnet_debugnet_event(if_t ifp __unused, enum debugnet_ev event) 4599 { 4600 struct vtnet_softc *sc; 4601 static bool sw_lro_enabled = false; 4602 4603 /* 4604 * Disable software LRO, since it would require entering the network 4605 * epoch when calling vtnet_txq_eof() in vtnet_debugnet_poll(). 4606 */ 4607 sc = if_getsoftc(ifp); 4608 switch (event) { 4609 case DEBUGNET_START: 4610 sw_lro_enabled = (sc->vtnet_flags & VTNET_FLAG_SW_LRO) != 0; 4611 if (sw_lro_enabled) 4612 sc->vtnet_flags &= ~VTNET_FLAG_SW_LRO; 4613 break; 4614 case DEBUGNET_END: 4615 if (sw_lro_enabled) 4616 sc->vtnet_flags |= VTNET_FLAG_SW_LRO; 4617 break; 4618 } 4619 } 4620 4621 static int 4622 vtnet_debugnet_transmit(if_t ifp, struct mbuf *m) 4623 { 4624 struct vtnet_softc *sc; 4625 struct vtnet_txq *txq; 4626 int error; 4627 4628 sc = if_getsoftc(ifp); 4629 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4630 IFF_DRV_RUNNING) 4631 return (EBUSY); 4632 4633 txq = &sc->vtnet_txqs[0]; 4634 error = vtnet_txq_encap(txq, &m, M_NOWAIT | M_USE_RESERVE); 4635 if (error == 0) 4636 (void)vtnet_txq_notify(txq); 4637 return (error); 4638 } 4639 4640 static int 4641 vtnet_debugnet_poll(if_t ifp, int count) 4642 { 4643 struct vtnet_softc *sc; 4644 int i; 4645 4646 sc = if_getsoftc(ifp); 4647 if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != 4648 IFF_DRV_RUNNING) 4649 return (EBUSY); 4650 4651 (void)vtnet_txq_eof(&sc->vtnet_txqs[0]); 4652 for (i = 0; i < sc->vtnet_act_vq_pairs; i++) 4653 (void)vtnet_rxq_eof(&sc->vtnet_rxqs[i]); 4654 return (0); 4655 } 4656 #endif /* DEBUGNET */ 4657