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