2 * Copyright (c) 2011, Bryan Venteicher <bryanv@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice unmodified, this list of conditions, and the following
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 /* Driver for VirtIO network devices. */
29 #include <sys/cdefs.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/sockio.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/socket.h>
39 #include <sys/sysctl.h>
40 #include <sys/taskqueue.h>
41 #include <sys/random.h>
42 #include <sys/sglist.h>
43 #include <sys/serialize.h>
47 #include <machine/limits.h>
49 #include <net/ethernet.h>
51 #include <net/if_arp.h>
52 #include <net/if_dl.h>
53 #include <net/if_types.h>
54 #include <net/if_media.h>
55 #include <net/vlan/if_vlan_var.h>
56 #include <net/vlan/if_vlan_ether.h>
57 #include <net/ifq_var.h>
61 #include <netinet/in_systm.h>
62 #include <netinet/in.h>
63 #include <netinet/ip.h>
64 #include <netinet/ip6.h>
65 #include <netinet/udp.h>
66 #include <netinet/tcp.h>
68 #include <dev/virtual/virtio/virtio/virtio.h>
69 #include <dev/virtual/virtio/virtio/virtqueue.h>
70 #include <dev/virtual/virtio/net/virtio_net.h>
71 #include <dev/virtual/virtio/net/if_vtnetvar.h>
73 #include "virtio_if.h"
75 MALLOC_DEFINE(M_VTNET, "VTNET_TX", "Outgoing VTNET TX frame header");
77 static int vtnet_modevent(module_t, int, void *);
79 static int vtnet_probe(device_t);
80 static int vtnet_attach(device_t);
81 static int vtnet_detach(device_t);
82 static int vtnet_suspend(device_t);
83 static int vtnet_resume(device_t);
84 static int vtnet_shutdown(device_t);
85 static int vtnet_config_change(device_t);
87 static void vtnet_negotiate_features(struct vtnet_softc *);
88 static int vtnet_alloc_virtqueues(struct vtnet_softc *);
89 static void vtnet_get_hwaddr(struct vtnet_softc *);
90 static void vtnet_set_hwaddr(struct vtnet_softc *);
91 static int vtnet_is_link_up(struct vtnet_softc *);
92 static void vtnet_update_link_status(struct vtnet_softc *);
94 static void vtnet_watchdog(struct vtnet_softc *);
96 static void vtnet_config_change_task(void *, int);
97 static int vtnet_setup_interface(struct vtnet_softc *);
98 static int vtnet_change_mtu(struct vtnet_softc *, int);
99 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
101 static int vtnet_init_rx_vq(struct vtnet_softc *);
102 static void vtnet_free_rx_mbufs(struct vtnet_softc *);
103 static void vtnet_free_tx_mbufs(struct vtnet_softc *);
104 static void vtnet_free_ctrl_vq(struct vtnet_softc *);
106 static struct mbuf * vtnet_alloc_rxbuf(struct vtnet_softc *, int,
108 static int vtnet_replace_rxbuf(struct vtnet_softc *,
110 static int vtnet_newbuf(struct vtnet_softc *);
111 static void vtnet_discard_merged_rxbuf(struct vtnet_softc *, int);
112 static void vtnet_discard_rxbuf(struct vtnet_softc *, struct mbuf *);
113 static int vtnet_enqueue_rxbuf(struct vtnet_softc *, struct mbuf *);
114 static void vtnet_vlan_tag_remove(struct mbuf *);
115 static int vtnet_rx_csum(struct vtnet_softc *, struct mbuf *,
116 struct virtio_net_hdr *);
117 static int vtnet_rxeof_merged(struct vtnet_softc *, struct mbuf *, int);
118 static int vtnet_rxeof(struct vtnet_softc *, int, int *);
119 static void vtnet_rx_intr_task(void *);
120 static int vtnet_rx_vq_intr(void *);
122 static void vtnet_txeof(struct vtnet_softc *);
123 static struct mbuf * vtnet_tx_offload(struct vtnet_softc *, struct mbuf *,
124 struct virtio_net_hdr *);
125 static int vtnet_enqueue_txbuf(struct vtnet_softc *, struct mbuf **,
126 struct vtnet_tx_header *);
127 static int vtnet_encap(struct vtnet_softc *, struct mbuf **);
128 static void vtnet_start_locked(struct ifnet *, struct ifaltq_subque *);
129 static void vtnet_start(struct ifnet *, struct ifaltq_subque *);
130 static void vtnet_tick(void *);
131 static void vtnet_tx_intr_task(void *);
132 static int vtnet_tx_vq_intr(void *);
134 static void vtnet_stop(struct vtnet_softc *);
135 static int vtnet_virtio_reinit(struct vtnet_softc *);
136 static void vtnet_init_locked(struct vtnet_softc *);
137 static void vtnet_init(void *);
139 static void vtnet_exec_ctrl_cmd(struct vtnet_softc *, void *,
140 struct sglist *, int, int);
142 static int vtnet_ctrl_mac_cmd(struct vtnet_softc *, uint8_t *);
143 static int vtnet_ctrl_rx_cmd(struct vtnet_softc *, int, int);
144 static int vtnet_set_promisc(struct vtnet_softc *, int);
145 static int vtnet_set_allmulti(struct vtnet_softc *, int);
146 static void vtnet_rx_filter(struct vtnet_softc *sc);
147 static void vtnet_rx_filter_mac(struct vtnet_softc *);
149 static int vtnet_exec_vlan_filter(struct vtnet_softc *, int, uint16_t);
150 static void vtnet_rx_filter_vlan(struct vtnet_softc *);
151 static void vtnet_update_vlan_filter(struct vtnet_softc *, int, uint16_t);
152 static void vtnet_register_vlan(void *, struct ifnet *, uint16_t);
153 static void vtnet_unregister_vlan(void *, struct ifnet *, uint16_t);
155 static int vtnet_ifmedia_upd(struct ifnet *);
156 static void vtnet_ifmedia_sts(struct ifnet *, struct ifmediareq *);
158 static void vtnet_add_statistics(struct vtnet_softc *);
160 static int vtnet_enable_rx_intr(struct vtnet_softc *);
161 static int vtnet_enable_tx_intr(struct vtnet_softc *);
162 static void vtnet_disable_rx_intr(struct vtnet_softc *);
163 static void vtnet_disable_tx_intr(struct vtnet_softc *);
166 static int vtnet_csum_disable = 0;
167 TUNABLE_INT("hw.vtnet.csum_disable", &vtnet_csum_disable);
168 static int vtnet_tso_disable = 1;
169 TUNABLE_INT("hw.vtnet.tso_disable", &vtnet_tso_disable);
170 static int vtnet_lro_disable = 0;
171 TUNABLE_INT("hw.vtnet.lro_disable", &vtnet_lro_disable);
174 * Reducing the number of transmit completed interrupts can
175 * improve performance. To do so, the define below keeps the
176 * Tx vq interrupt disabled and adds calls to vtnet_txeof()
177 * in the start and watchdog paths. The price to pay for this
178 * is the m_free'ing of transmitted mbufs may be delayed until
179 * the watchdog fires.
181 #define VTNET_TX_INTR_MODERATION
183 static struct virtio_feature_desc vtnet_feature_desc[] = {
184 { VIRTIO_NET_F_CSUM, "TxChecksum" },
185 { VIRTIO_NET_F_GUEST_CSUM, "RxChecksum" },
186 { VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, "DynOffload" },
187 { VIRTIO_NET_F_MAC, "MacAddress" },
188 { VIRTIO_NET_F_GSO, "TxAllGSO" },
189 { VIRTIO_NET_F_GUEST_TSO4, "RxTSOv4" },
190 { VIRTIO_NET_F_GUEST_TSO6, "RxTSOv6" },
191 { VIRTIO_NET_F_GUEST_ECN, "RxECN" },
192 { VIRTIO_NET_F_GUEST_UFO, "RxUFO" },
193 { VIRTIO_NET_F_HOST_TSO4, "TxTSOv4" },
194 { VIRTIO_NET_F_HOST_TSO6, "TxTSOv6" },
195 { VIRTIO_NET_F_HOST_ECN, "TxTSOECN" },
196 { VIRTIO_NET_F_HOST_UFO, "TxUFO" },
197 { VIRTIO_NET_F_MRG_RXBUF, "MrgRxBuf" },
198 { VIRTIO_NET_F_STATUS, "Status" },
199 { VIRTIO_NET_F_CTRL_VQ, "ControlVq" },
200 { VIRTIO_NET_F_CTRL_RX, "RxMode" },
201 { VIRTIO_NET_F_CTRL_VLAN, "VLanFilter" },
202 { VIRTIO_NET_F_CTRL_RX_EXTRA, "RxModeExtra" },
203 { VIRTIO_NET_F_GUEST_ANNOUNCE, "GuestAnnounce" },
204 { VIRTIO_NET_F_MQ, "RFS" },
205 { VIRTIO_NET_F_CTRL_MAC_ADDR, "SetMacAddress" },
209 static device_method_t vtnet_methods[] = {
210 /* Device methods. */
211 DEVMETHOD(device_probe, vtnet_probe),
212 DEVMETHOD(device_attach, vtnet_attach),
213 DEVMETHOD(device_detach, vtnet_detach),
214 DEVMETHOD(device_suspend, vtnet_suspend),
215 DEVMETHOD(device_resume, vtnet_resume),
216 DEVMETHOD(device_shutdown, vtnet_shutdown),
218 /* VirtIO methods. */
219 DEVMETHOD(virtio_config_change, vtnet_config_change),
224 static driver_t vtnet_driver = {
227 sizeof(struct vtnet_softc)
230 static devclass_t vtnet_devclass;
232 DRIVER_MODULE(vtnet, virtio_pci, vtnet_driver, vtnet_devclass,
234 MODULE_VERSION(vtnet, 1);
235 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
238 vtnet_modevent(module_t mod, int type, void *unused)
260 vtnet_probe(device_t dev)
262 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
265 device_set_desc(dev, "VirtIO Networking Adapter");
267 return (BUS_PROBE_DEFAULT);
271 vtnet_attach(device_t dev)
273 struct vtnet_softc *sc;
276 sc = device_get_softc(dev);
279 lwkt_serialize_init(&sc->vtnet_slz);
280 callout_init(&sc->vtnet_tick_ch);
282 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
284 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
285 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
287 vtnet_add_statistics(sc);
289 /* Register our feature descriptions. */
290 virtio_set_feature_desc(dev, vtnet_feature_desc);
291 vtnet_negotiate_features(sc);
293 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
294 sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
296 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
297 /* This feature should always be negotiated. */
298 sc->vtnet_flags |= VTNET_FLAG_MAC;
301 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
302 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
303 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
305 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
308 sc->vtnet_rx_mbuf_size = MCLBYTES;
309 sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
311 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
312 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
314 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
315 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
316 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
317 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
318 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
319 virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
320 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
323 /* Read (or generate) the MAC address for the adapter. */
324 vtnet_get_hwaddr(sc);
326 error = vtnet_alloc_virtqueues(sc);
328 device_printf(dev, "cannot allocate virtqueues\n");
332 error = vtnet_setup_interface(sc);
334 device_printf(dev, "cannot setup interface\n");
338 TASK_INIT(&sc->vtnet_cfgchg_task, 0, vtnet_config_change_task, sc);
340 error = virtio_setup_intr(dev, &sc->vtnet_slz);
342 device_printf(dev, "cannot setup virtqueue interrupts\n");
343 ether_ifdetach(sc->vtnet_ifp);
348 * Device defaults to promiscuous mode for backwards
349 * compatibility. Turn it off if possible.
351 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
352 lwkt_serialize_enter(&sc->vtnet_slz);
353 if (vtnet_set_promisc(sc, 0) != 0) {
354 sc->vtnet_ifp->if_flags |= IFF_PROMISC;
356 "cannot disable promiscuous mode\n");
358 lwkt_serialize_exit(&sc->vtnet_slz);
360 sc->vtnet_ifp->if_flags |= IFF_PROMISC;
370 vtnet_detach(device_t dev)
372 struct vtnet_softc *sc;
375 sc = device_get_softc(dev);
378 if (device_is_attached(dev)) {
379 lwkt_serialize_enter(&sc->vtnet_slz);
381 lwkt_serialize_exit(&sc->vtnet_slz);
383 callout_stop(&sc->vtnet_tick_ch);
384 taskqueue_drain(taskqueue_swi, &sc->vtnet_cfgchg_task);
389 if (sc->vtnet_vlan_attach != NULL) {
390 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
391 sc->vtnet_vlan_attach = NULL;
393 if (sc->vtnet_vlan_detach != NULL) {
394 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
395 sc->vtnet_vlan_detach = NULL;
400 sc->vtnet_ifp = NULL;
403 if (sc->vtnet_rx_vq != NULL)
404 vtnet_free_rx_mbufs(sc);
405 if (sc->vtnet_tx_vq != NULL)
406 vtnet_free_tx_mbufs(sc);
407 if (sc->vtnet_ctrl_vq != NULL)
408 vtnet_free_ctrl_vq(sc);
410 if (sc->vtnet_txhdrarea != NULL) {
411 contigfree(sc->vtnet_txhdrarea,
412 sc->vtnet_txhdrcount * sizeof(struct vtnet_tx_header),
414 sc->vtnet_txhdrarea = NULL;
416 if (sc->vtnet_macfilter != NULL) {
417 contigfree(sc->vtnet_macfilter,
418 sizeof(struct vtnet_mac_filter), M_DEVBUF);
419 sc->vtnet_macfilter = NULL;
422 ifmedia_removeall(&sc->vtnet_media);
428 vtnet_suspend(device_t dev)
430 struct vtnet_softc *sc;
432 sc = device_get_softc(dev);
434 lwkt_serialize_enter(&sc->vtnet_slz);
436 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
437 lwkt_serialize_exit(&sc->vtnet_slz);
443 vtnet_resume(device_t dev)
445 struct vtnet_softc *sc;
448 sc = device_get_softc(dev);
451 lwkt_serialize_enter(&sc->vtnet_slz);
452 if (ifp->if_flags & IFF_UP)
453 vtnet_init_locked(sc);
454 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
455 lwkt_serialize_exit(&sc->vtnet_slz);
461 vtnet_shutdown(device_t dev)
465 * Suspend already does all of what we need to
466 * do here; we just never expect to be resumed.
468 return (vtnet_suspend(dev));
472 vtnet_config_change(device_t dev)
474 struct vtnet_softc *sc;
476 sc = device_get_softc(dev);
478 taskqueue_enqueue(taskqueue_thread[mycpuid], &sc->vtnet_cfgchg_task);
484 vtnet_negotiate_features(struct vtnet_softc *sc)
487 uint64_t mask, features;
492 if (vtnet_csum_disable)
493 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
496 * TSO and LRO are only available when their corresponding checksum
497 * offload feature is also negotiated.
500 if (vtnet_csum_disable || vtnet_tso_disable)
501 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 |
502 VIRTIO_NET_F_HOST_ECN;
504 if (vtnet_csum_disable || vtnet_lro_disable)
505 mask |= VTNET_LRO_FEATURES;
507 features = VTNET_FEATURES & ~mask;
508 features |= VIRTIO_F_NOTIFY_ON_EMPTY;
509 features |= VIRTIO_F_ANY_LAYOUT;
510 sc->vtnet_features = virtio_negotiate_features(dev, features);
512 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
513 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
515 * LRO without mergeable buffers requires special care. This
516 * is not ideal because every receive buffer must be large
517 * enough to hold the maximum TCP packet, the Ethernet header,
518 * and the header. This requires up to 34 descriptors with
519 * MCLBYTES clusters. If we do not have indirect descriptors,
520 * LRO is disabled since the virtqueue will not contain very
521 * many receive buffers.
523 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
525 "LRO disabled due to both mergeable buffers and "
526 "indirect descriptors not negotiated\n");
528 features &= ~VTNET_LRO_FEATURES;
530 virtio_negotiate_features(dev, features);
532 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
537 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
540 struct vq_alloc_info vq_info[3];
547 * Indirect descriptors are not needed for the Rx
548 * virtqueue when mergeable buffers are negotiated.
549 * The header is placed inline with the data, not
550 * in a separate descriptor, and mbuf clusters are
551 * always physically contiguous.
553 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
554 sc->vtnet_rx_nsegs = sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG ?
555 VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS;
557 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
559 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
560 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
561 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
563 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
565 VQ_ALLOC_INFO_INIT(&vq_info[0], sc->vtnet_rx_nsegs,
566 vtnet_rx_vq_intr, sc, &sc->vtnet_rx_vq,
567 "%s receive", device_get_nameunit(dev));
569 VQ_ALLOC_INFO_INIT(&vq_info[1], sc->vtnet_tx_nsegs,
570 vtnet_tx_vq_intr, sc, &sc->vtnet_tx_vq,
571 "%s transmit", device_get_nameunit(dev));
573 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
576 VQ_ALLOC_INFO_INIT(&vq_info[2], 0, NULL, NULL,
577 &sc->vtnet_ctrl_vq, "%s control",
578 device_get_nameunit(dev));
581 return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info));
585 vtnet_setup_interface(struct vtnet_softc *sc)
593 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
595 device_printf(dev, "cannot allocate ifnet structure\n");
600 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
601 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
602 ifp->if_init = vtnet_init;
603 ifp->if_start = vtnet_start;
604 ifp->if_ioctl = vtnet_ioctl;
606 sc->vtnet_rx_size = virtqueue_size(sc->vtnet_rx_vq);
607 sc->vtnet_rx_process_limit = sc->vtnet_rx_size;
609 tx_size = virtqueue_size(sc->vtnet_tx_vq);
610 sc->vtnet_tx_size = tx_size;
611 sc->vtnet_txhdridx = 0;
612 if (sc->vtnet_flags & VTNET_FLAG_INDIRECT)
613 sc->vtnet_txhdrcount = sc->vtnet_tx_size;
615 sc->vtnet_txhdrcount = (sc->vtnet_tx_size / 2) + 1;
616 sc->vtnet_txhdrarea = contigmalloc(
617 sc->vtnet_txhdrcount * sizeof(struct vtnet_tx_header),
618 M_VTNET, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0);
619 if (sc->vtnet_txhdrarea == NULL) {
620 device_printf(dev, "cannot contigmalloc the tx headers\n");
623 sc->vtnet_macfilter = contigmalloc(
624 sizeof(struct vtnet_mac_filter),
625 M_DEVBUF, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0);
626 if (sc->vtnet_macfilter == NULL) {
628 "cannot contigmalloc the mac filter table\n");
631 ifq_set_maxlen(&ifp->if_snd, tx_size - 1);
632 ifq_set_ready(&ifp->if_snd);
634 ether_ifattach(ifp, sc->vtnet_hwaddr, NULL);
636 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)){
637 //ifp->if_capabilities |= IFCAP_LINKSTATE;
638 kprintf("add dynamic link state\n");
641 /* Tell the upper layer(s) we support long frames. */
642 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
643 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
645 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
646 ifp->if_capabilities |= IFCAP_TXCSUM;
648 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
649 ifp->if_capabilities |= IFCAP_TSO4;
650 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
651 ifp->if_capabilities |= IFCAP_TSO6;
652 if (ifp->if_capabilities & IFCAP_TSO)
653 ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
655 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
656 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
659 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
660 ifp->if_capabilities |= IFCAP_RXCSUM;
662 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
663 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
664 ifp->if_capabilities |= IFCAP_LRO;
667 if (ifp->if_capabilities & IFCAP_HWCSUM) {
669 * VirtIO does not support VLAN tagging, but we can fake
670 * it by inserting and removing the 802.1Q header during
671 * transmit and receive. We are then able to do checksum
672 * offloading of VLAN frames.
674 ifp->if_capabilities |=
675 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
678 ifp->if_capenable = ifp->if_capabilities;
681 * Capabilities after here are not enabled by default.
684 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
685 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
687 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
688 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
689 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
690 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
697 vtnet_set_hwaddr(struct vtnet_softc *sc)
703 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) &&
704 (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)) {
705 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
706 device_printf(dev, "unable to set MAC address\n");
707 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
708 virtio_write_device_config(dev,
709 offsetof(struct virtio_net_config, mac),
710 sc->vtnet_hwaddr, ETHER_ADDR_LEN);
715 vtnet_get_hwaddr(struct vtnet_softc *sc)
721 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
723 * Generate a random locally administered unicast address.
725 * It would be nice to generate the same MAC address across
726 * reboots, but it seems all the hosts currently available
727 * support the MAC feature, so this isn't too important.
729 sc->vtnet_hwaddr[0] = 0xB2;
730 karc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1);
731 vtnet_set_hwaddr(sc);
735 virtio_read_device_config(dev,
736 offsetof(struct virtio_net_config, mac),
737 sc->vtnet_hwaddr, ETHER_ADDR_LEN);
741 vtnet_is_link_up(struct vtnet_softc *sc)
750 ASSERT_SERIALIZED(&sc->vtnet_slz);
752 status = virtio_read_dev_config_2(dev,
753 offsetof(struct virtio_net_config, status));
755 return ((status & VIRTIO_NET_S_LINK_UP) != 0);
759 vtnet_update_link_status(struct vtnet_softc *sc)
763 struct ifaltq_subque *ifsq;
768 ifsq = ifq_get_subq_default(&ifp->if_snd);
770 link = vtnet_is_link_up(sc);
772 if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) {
773 sc->vtnet_flags |= VTNET_FLAG_LINK;
775 device_printf(dev, "Link is up\n");
776 ifp->if_link_state = LINK_STATE_UP;
777 if_link_state_change(ifp);
778 if (!ifsq_is_empty(ifsq))
779 vtnet_start_locked(ifp, ifsq);
780 } else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) {
781 sc->vtnet_flags &= ~VTNET_FLAG_LINK;
783 device_printf(dev, "Link is down\n");
785 ifp->if_link_state = LINK_STATE_DOWN;
786 if_link_state_change(ifp);
792 vtnet_watchdog(struct vtnet_softc *sc)
798 #ifdef VTNET_TX_INTR_MODERATION
802 if (sc->vtnet_watchdog_timer == 0 || --sc->vtnet_watchdog_timer)
805 if_printf(ifp, "watchdog timeout -- resetting\n");
807 virtqueue_dump(sc->vtnet_tx_vq);
810 ifp->if_flags &= ~IFF_RUNNING;
811 vtnet_init_locked(sc);
816 vtnet_config_change_task(void *arg, int pending)
818 struct vtnet_softc *sc;
822 lwkt_serialize_enter(&sc->vtnet_slz);
823 vtnet_update_link_status(sc);
824 lwkt_serialize_exit(&sc->vtnet_slz);
828 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data,struct ucred *cr)
830 struct vtnet_softc *sc;
832 int reinit, mask, error;
835 ifr = (struct ifreq *) data;
841 if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU)
843 else if (ifp->if_mtu != ifr->ifr_mtu) {
844 lwkt_serialize_enter(&sc->vtnet_slz);
845 error = vtnet_change_mtu(sc, ifr->ifr_mtu);
846 lwkt_serialize_exit(&sc->vtnet_slz);
851 lwkt_serialize_enter(&sc->vtnet_slz);
852 if ((ifp->if_flags & IFF_UP) == 0) {
853 if (ifp->if_flags & IFF_RUNNING)
855 } else if (ifp->if_flags & IFF_RUNNING) {
856 if ((ifp->if_flags ^ sc->vtnet_if_flags) &
857 (IFF_PROMISC | IFF_ALLMULTI)) {
858 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
864 vtnet_init_locked(sc);
867 sc->vtnet_if_flags = ifp->if_flags;
868 lwkt_serialize_exit(&sc->vtnet_slz);
873 lwkt_serialize_enter(&sc->vtnet_slz);
874 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) &&
875 (ifp->if_flags & IFF_RUNNING))
876 vtnet_rx_filter_mac(sc);
877 lwkt_serialize_exit(&sc->vtnet_slz);
882 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
886 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
888 lwkt_serialize_enter(&sc->vtnet_slz);
890 if (mask & IFCAP_TXCSUM) {
891 ifp->if_capenable ^= IFCAP_TXCSUM;
892 if (ifp->if_capenable & IFCAP_TXCSUM)
893 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
895 ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD;
898 if (mask & IFCAP_TSO4) {
899 ifp->if_capenable ^= IFCAP_TSO4;
900 if (ifp->if_capenable & IFCAP_TSO4)
901 ifp->if_hwassist |= CSUM_TSO;
903 ifp->if_hwassist &= ~CSUM_TSO;
906 if (mask & IFCAP_RXCSUM) {
907 ifp->if_capenable ^= IFCAP_RXCSUM;
911 if (mask & IFCAP_LRO) {
912 ifp->if_capenable ^= IFCAP_LRO;
916 if (mask & IFCAP_VLAN_HWFILTER) {
917 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
921 if (mask & IFCAP_VLAN_HWTSO)
922 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
924 if (mask & IFCAP_VLAN_HWTAGGING)
925 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
927 if (reinit && (ifp->if_flags & IFF_RUNNING)) {
928 ifp->if_flags &= ~IFF_RUNNING;
929 vtnet_init_locked(sc);
931 //VLAN_CAPABILITIES(ifp);
933 lwkt_serialize_exit(&sc->vtnet_slz);
937 error = ether_ioctl(ifp, cmd, data);
945 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
948 int new_frame_size, clsize;
952 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
953 new_frame_size = sizeof(struct vtnet_rx_header) +
954 sizeof(struct ether_vlan_header) + new_mtu;
956 if (new_frame_size > MJUM9BYTES)
959 if (new_frame_size <= MCLBYTES)
964 new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) +
965 sizeof(struct ether_vlan_header) + new_mtu;
967 if (new_frame_size <= MCLBYTES)
970 clsize = MJUMPAGESIZE;
973 sc->vtnet_rx_mbuf_size = clsize;
974 sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
975 KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS,
976 ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count));
978 ifp->if_mtu = new_mtu;
980 if (ifp->if_flags & IFF_RUNNING) {
981 ifp->if_flags &= ~IFF_RUNNING;
982 vtnet_init_locked(sc);
989 vtnet_init_rx_vq(struct vtnet_softc *sc)
991 struct virtqueue *vq;
994 vq = sc->vtnet_rx_vq;
998 while (!virtqueue_full(vq)) {
999 if ((error = vtnet_newbuf(sc)) != 0)
1005 virtqueue_notify(vq, &sc->vtnet_slz);
1008 * EMSGSIZE signifies the virtqueue did not have enough
1009 * entries available to hold the last mbuf. This is not
1010 * an error. We should not get ENOSPC since we check if
1011 * the virtqueue is full before attempting to add a
1014 if (error == EMSGSIZE)
1022 vtnet_free_rx_mbufs(struct vtnet_softc *sc)
1024 struct virtqueue *vq;
1028 vq = sc->vtnet_rx_vq;
1031 while ((m = virtqueue_drain(vq, &last)) != NULL)
1034 KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq"));
1038 vtnet_free_tx_mbufs(struct vtnet_softc *sc)
1040 struct virtqueue *vq;
1041 struct vtnet_tx_header *txhdr;
1044 vq = sc->vtnet_tx_vq;
1047 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1048 m_freem(txhdr->vth_mbuf);
1051 KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq"));
1055 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
1058 * The control virtqueue is only polled, therefore
1059 * it should already be empty.
1061 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
1062 ("Ctrl Vq not empty"));
1065 static struct mbuf *
1066 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1068 struct mbuf *m_head, *m_tail, *m;
1071 clsize = sc->vtnet_rx_mbuf_size;
1073 /*use getcl instead of getjcl. see if_mxge.c comment line 2398*/
1074 //m_head = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, clsize);
1075 m_head = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR );
1079 m_head->m_len = clsize;
1083 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1084 ("chained Rx mbuf requested without LRO_NOMRG"));
1086 for (i = 0; i < nbufs - 1; i++) {
1087 //m = m_getjcl(M_DONTWAIT, MT_DATA, 0, clsize);
1088 m = m_getcl(M_NOWAIT, MT_DATA, 0);
1098 if (m_tailp != NULL)
1104 sc->vtnet_stats.mbuf_alloc_failed++;
1111 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0)
1113 struct mbuf *m, *m_prev;
1114 struct mbuf *m_new, *m_tail;
1115 int len, clsize, nreplace, error;
1122 clsize = sc->vtnet_rx_mbuf_size;
1125 if (m->m_next != NULL)
1126 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1127 ("chained Rx mbuf without LRO_NOMRG"));
1130 * Since LRO_NOMRG mbuf chains are so large, we want to avoid
1131 * allocating an entire chain for each received frame. When
1132 * the received frame's length is less than that of the chain,
1133 * the unused mbufs are reassigned to the new chain.
1137 * Something is seriously wrong if we received
1138 * a frame larger than the mbuf chain. Drop it.
1141 sc->vtnet_stats.rx_frame_too_large++;
1145 KASSERT(m->m_len == clsize,
1146 ("mbuf length not expected cluster size: %d",
1149 m->m_len = MIN(m->m_len, len);
1157 KASSERT(m_prev != NULL, ("m_prev == NULL"));
1158 KASSERT(nreplace <= sc->vtnet_rx_mbuf_count,
1159 ("too many replacement mbufs: %d/%d", nreplace,
1160 sc->vtnet_rx_mbuf_count));
1162 m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail);
1163 if (m_new == NULL) {
1164 m_prev->m_len = clsize;
1169 * Move unused mbufs, if any, from the original chain
1170 * onto the end of the new chain.
1172 if (m_prev->m_next != NULL) {
1173 m_tail->m_next = m_prev->m_next;
1174 m_prev->m_next = NULL;
1177 error = vtnet_enqueue_rxbuf(sc, m_new);
1180 * BAD! We could not enqueue the replacement mbuf chain. We
1181 * must restore the m0 chain to the original state if it was
1182 * modified so we can subsequently discard it.
1184 * NOTE: The replacement is suppose to be an identical copy
1185 * to the one just dequeued so this is an unexpected error.
1187 sc->vtnet_stats.rx_enq_replacement_failed++;
1189 if (m_tail->m_next != NULL) {
1190 m_prev->m_next = m_tail->m_next;
1191 m_tail->m_next = NULL;
1194 m_prev->m_len = clsize;
1202 vtnet_newbuf(struct vtnet_softc *sc)
1207 m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL);
1211 error = vtnet_enqueue_rxbuf(sc, m);
1219 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs)
1221 struct virtqueue *vq;
1224 vq = sc->vtnet_rx_vq;
1226 while (--nbufs > 0) {
1227 if ((m = virtqueue_dequeue(vq, NULL)) == NULL)
1229 vtnet_discard_rxbuf(sc, m);
1234 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1239 * Requeue the discarded mbuf. This should always be
1240 * successful since it was just dequeued.
1242 error = vtnet_enqueue_rxbuf(sc, m);
1243 KASSERT(error == 0, ("cannot requeue discarded mbuf"));
1247 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1250 struct sglist_seg segs[VTNET_MAX_RX_SEGS];
1251 struct vtnet_rx_header *rxhdr;
1252 struct virtio_net_hdr *hdr;
1256 ASSERT_SERIALIZED(&sc->vtnet_slz);
1257 if ((sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) == 0)
1258 KASSERT(m->m_next == NULL, ("chained Rx mbuf"));
1260 sglist_init(&sg, sc->vtnet_rx_nsegs, segs);
1262 mdata = mtod(m, uint8_t *);
1265 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1266 rxhdr = (struct vtnet_rx_header *) mdata;
1267 hdr = &rxhdr->vrh_hdr;
1268 offset += sizeof(struct vtnet_rx_header);
1270 error = sglist_append(&sg, hdr, sc->vtnet_hdr_size);
1271 KASSERT(error == 0, ("cannot add header to sglist"));
1274 error = sglist_append(&sg, mdata + offset, m->m_len - offset);
1278 if (m->m_next != NULL) {
1279 error = sglist_append_mbuf(&sg, m->m_next);
1284 return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg));
1288 vtnet_vlan_tag_remove(struct mbuf *m)
1290 struct ether_vlan_header *evl;
1292 evl = mtod(m, struct ether_vlan_header *);
1294 m->m_pkthdr.ether_vlantag = ntohs(evl->evl_tag);
1295 m->m_flags |= M_VLANTAG;
1297 /* Strip the 802.1Q header. */
1298 bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN,
1299 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1300 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1304 * Alternative method of doing receive checksum offloading. Rather
1305 * than parsing the received frame down to the IP header, use the
1306 * csum_offset to determine which CSUM_* flags are appropriate. We
1307 * can get by with doing this only because the checksum offsets are
1308 * unique for the things we care about.
1311 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1312 struct virtio_net_hdr *hdr)
1314 struct ether_header *eh;
1315 struct ether_vlan_header *evh;
1320 csum_len = hdr->csum_start + hdr->csum_offset;
1322 if (csum_len < sizeof(struct ether_header) + sizeof(struct ip))
1324 if (m->m_len < csum_len)
1327 eh = mtod(m, struct ether_header *);
1328 eth_type = ntohs(eh->ether_type);
1329 if (eth_type == ETHERTYPE_VLAN) {
1330 evh = mtod(m, struct ether_vlan_header *);
1331 eth_type = ntohs(evh->evl_proto);
1334 if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) {
1335 sc->vtnet_stats.rx_csum_bad_ethtype++;
1339 /* Use the offset to determine the appropriate CSUM_* flags. */
1340 switch (hdr->csum_offset) {
1341 case offsetof(struct udphdr, uh_sum):
1342 if (m->m_len < hdr->csum_start + sizeof(struct udphdr))
1344 udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start);
1345 if (udp->uh_sum == 0)
1350 case offsetof(struct tcphdr, th_sum):
1351 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1352 m->m_pkthdr.csum_data = 0xFFFF;
1356 sc->vtnet_stats.rx_csum_bad_offset++;
1360 sc->vtnet_stats.rx_csum_offloaded++;
1366 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs)
1369 struct virtqueue *vq;
1370 struct mbuf *m, *m_tail;
1373 ifp = sc->vtnet_ifp;
1374 vq = sc->vtnet_rx_vq;
1377 while (--nbufs > 0) {
1378 m = virtqueue_dequeue(vq, &len);
1384 if (vtnet_newbuf(sc) != 0) {
1386 vtnet_discard_rxbuf(sc, m);
1388 vtnet_discard_merged_rxbuf(sc, nbufs);
1396 m->m_flags &= ~M_PKTHDR;
1398 m_head->m_pkthdr.len += len;
1406 sc->vtnet_stats.rx_mergeable_failed++;
1413 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp)
1415 struct virtio_net_hdr lhdr;
1417 struct virtqueue *vq;
1419 struct ether_header *eh;
1420 struct virtio_net_hdr *hdr;
1421 struct virtio_net_hdr_mrg_rxbuf *mhdr;
1422 int len, deq, nbufs, adjsz, rx_npkts;
1424 ifp = sc->vtnet_ifp;
1425 vq = sc->vtnet_rx_vq;
1430 ASSERT_SERIALIZED(&sc->vtnet_slz);
1432 while (--count >= 0) {
1433 m = virtqueue_dequeue(vq, &len);
1438 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1440 vtnet_discard_rxbuf(sc, m);
1444 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1446 adjsz = sizeof(struct vtnet_rx_header);
1448 * Account for our pad between the header and
1449 * the actual start of the frame.
1451 len += VTNET_RX_HEADER_PAD;
1453 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1454 nbufs = mhdr->num_buffers;
1455 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1458 if (vtnet_replace_rxbuf(sc, m, len) != 0) {
1460 vtnet_discard_rxbuf(sc, m);
1462 vtnet_discard_merged_rxbuf(sc, nbufs);
1466 m->m_pkthdr.len = len;
1467 m->m_pkthdr.rcvif = ifp;
1468 m->m_pkthdr.csum_flags = 0;
1471 if (vtnet_rxeof_merged(sc, m, nbufs) != 0)
1478 * Save copy of header before we strip it. For both mergeable
1479 * and non-mergeable, the VirtIO header is placed first in the
1480 * mbuf's data. We no longer need num_buffers, so always use a
1483 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1486 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1487 eh = mtod(m, struct ether_header *);
1488 if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1489 vtnet_vlan_tag_remove(m);
1492 * With the 802.1Q header removed, update the
1493 * checksum starting location accordingly.
1495 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1497 ETHER_VLAN_ENCAP_LEN;
1501 if (ifp->if_capenable & IFCAP_RXCSUM &&
1502 hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1503 if (vtnet_rx_csum(sc, m, hdr) != 0)
1504 sc->vtnet_stats.rx_csum_failed++;
1507 lwkt_serialize_exit(&sc->vtnet_slz);
1509 ifp->if_input(ifp, m, NULL, -1);
1510 lwkt_serialize_enter(&sc->vtnet_slz);
1513 * The interface may have been stopped while we were
1514 * passing the packet up the network stack.
1516 if ((ifp->if_flags & IFF_RUNNING) == 0)
1520 virtqueue_notify(vq, &sc->vtnet_slz);
1522 if (rx_npktsp != NULL)
1523 *rx_npktsp = rx_npkts;
1525 return (count > 0 ? 0 : EAGAIN);
1529 vtnet_rx_intr_task(void *arg)
1531 struct vtnet_softc *sc;
1536 ifp = sc->vtnet_ifp;
1539 // lwkt_serialize_enter(&sc->vtnet_slz);
1541 if ((ifp->if_flags & IFF_RUNNING) == 0) {
1542 vtnet_enable_rx_intr(sc);
1543 // lwkt_serialize_exit(&sc->vtnet_slz);
1547 more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL);
1548 if (!more && vtnet_enable_rx_intr(sc) != 0) {
1549 vtnet_disable_rx_intr(sc);
1553 // lwkt_serialize_exit(&sc->vtnet_slz);
1556 sc->vtnet_stats.rx_task_rescheduled++;
1562 vtnet_rx_vq_intr(void *xsc)
1564 struct vtnet_softc *sc;
1568 vtnet_disable_rx_intr(sc);
1569 vtnet_rx_intr_task(sc);
1575 vtnet_txeof(struct vtnet_softc *sc)
1577 struct virtqueue *vq;
1579 struct vtnet_tx_header *txhdr;
1582 vq = sc->vtnet_tx_vq;
1583 ifp = sc->vtnet_ifp;
1586 ASSERT_SERIALIZED(&sc->vtnet_slz);
1588 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
1591 m_freem(txhdr->vth_mbuf);
1595 ifq_clr_oactive(&ifp->if_snd);
1596 if (virtqueue_empty(vq))
1597 sc->vtnet_watchdog_timer = 0;
1601 static struct mbuf *
1602 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m,
1603 struct virtio_net_hdr *hdr)
1606 struct ether_header *eh;
1607 struct ether_vlan_header *evh;
1609 struct ip6_hdr *ip6;
1612 uint16_t eth_type, csum_start;
1613 uint8_t ip_proto, gso_type;
1615 ifp = sc->vtnet_ifp;
1618 ip_offset = sizeof(struct ether_header);
1619 if (m->m_len < ip_offset) {
1620 if ((m = m_pullup(m, ip_offset)) == NULL)
1624 eh = mtod(m, struct ether_header *);
1625 eth_type = ntohs(eh->ether_type);
1626 if (eth_type == ETHERTYPE_VLAN) {
1627 ip_offset = sizeof(struct ether_vlan_header);
1628 if (m->m_len < ip_offset) {
1629 if ((m = m_pullup(m, ip_offset)) == NULL)
1632 evh = mtod(m, struct ether_vlan_header *);
1633 eth_type = ntohs(evh->evl_proto);
1638 if (m->m_len < ip_offset + sizeof(struct ip)) {
1639 m = m_pullup(m, ip_offset + sizeof(struct ip));
1644 ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1645 ip_proto = ip->ip_p;
1646 csum_start = ip_offset + (ip->ip_hl << 2);
1647 gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1650 case ETHERTYPE_IPV6:
1651 if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) {
1652 m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr));
1657 ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1659 * XXX Assume no extension headers are present. Presently,
1660 * this will always be true in the case of TSO, and FreeBSD
1661 * does not perform checksum offloading of IPv6 yet.
1663 ip_proto = ip6->ip6_nxt;
1664 csum_start = ip_offset + sizeof(struct ip6_hdr);
1665 gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1672 if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) {
1673 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
1674 hdr->csum_start = csum_start;
1675 hdr->csum_offset = m->m_pkthdr.csum_data;
1677 sc->vtnet_stats.tx_csum_offloaded++;
1680 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1681 if (ip_proto != IPPROTO_TCP)
1684 if (m->m_len < csum_start + sizeof(struct tcphdr)) {
1685 m = m_pullup(m, csum_start + sizeof(struct tcphdr));
1690 tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start);
1691 hdr->gso_type = gso_type;
1692 hdr->hdr_len = csum_start + (tcp->th_off << 2);
1693 hdr->gso_size = m->m_pkthdr.tso_segsz;
1695 if (tcp->th_flags & TH_CWR) {
1697 * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN.
1698 * ECN support is only configurable globally with the
1699 * net.inet.tcp.ecn.enable sysctl knob.
1701 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
1702 if_printf(ifp, "TSO with ECN not supported "
1708 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1711 sc->vtnet_stats.tx_tso_offloaded++;
1718 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head,
1719 struct vtnet_tx_header *txhdr)
1722 struct sglist_seg segs[VTNET_MAX_TX_SEGS];
1723 struct virtqueue *vq;
1727 vq = sc->vtnet_tx_vq;
1730 sglist_init(&sg, sc->vtnet_tx_nsegs, segs);
1731 error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
1732 KASSERT(error == 0 && sg.sg_nseg == 1,
1733 ("%s: error %d adding header to sglist", __func__, error));
1735 error = sglist_append_mbuf(&sg, m);
1737 m = m_defrag(m, M_NOWAIT);
1742 sc->vtnet_stats.tx_defragged++;
1744 error = sglist_append_mbuf(&sg, m);
1749 txhdr->vth_mbuf = m;
1750 error = virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0);
1755 sc->vtnet_stats.tx_defrag_failed++;
1762 static struct mbuf *
1763 vtnet_vlan_tag_insert(struct mbuf *m)
1766 struct ether_vlan_header *evl;
1768 if (M_WRITABLE(m) == 0) {
1769 n = m_dup(m, M_NOWAIT);
1771 if ((m = n) == NULL)
1775 M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_NOWAIT);
1778 if (m->m_len < sizeof(struct ether_vlan_header)) {
1779 m = m_pullup(m, sizeof(struct ether_vlan_header));
1784 /* Insert 802.1Q header into the existing Ethernet header. */
1785 evl = mtod(m, struct ether_vlan_header *);
1786 bcopy((char *) evl + ETHER_VLAN_ENCAP_LEN,
1787 (char *) evl, ETHER_HDR_LEN - ETHER_TYPE_LEN);
1788 evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1789 evl->evl_tag = htons(m->m_pkthdr.ether_vlantag);
1790 m->m_flags &= ~M_VLANTAG;
1796 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head)
1798 struct vtnet_tx_header *txhdr;
1799 struct virtio_net_hdr *hdr;
1803 txhdr = &sc->vtnet_txhdrarea[sc->vtnet_txhdridx];
1804 memset(txhdr, 0, sizeof(struct vtnet_tx_header));
1807 * Always use the non-mergeable header to simplify things. When
1808 * the mergeable feature is negotiated, the num_buffers field
1809 * must be set to zero. We use vtnet_hdr_size later to enqueue
1810 * the correct header size to the host.
1812 hdr = &txhdr->vth_uhdr.hdr;
1817 if (m->m_flags & M_VLANTAG) {
1818 //m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
1819 m = vtnet_vlan_tag_insert(m);
1820 if ((*m_head = m) == NULL)
1822 m->m_flags &= ~M_VLANTAG;
1825 if (m->m_pkthdr.csum_flags != 0) {
1826 m = vtnet_tx_offload(sc, m, hdr);
1827 if ((*m_head = m) == NULL)
1831 error = vtnet_enqueue_txbuf(sc, m_head, txhdr);
1833 sc->vtnet_txhdridx =
1834 (sc->vtnet_txhdridx + 1) % sc->vtnet_txhdrcount;
1840 vtnet_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1842 struct vtnet_softc *sc;
1846 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1847 lwkt_serialize_enter(&sc->vtnet_slz);
1848 vtnet_start_locked(ifp, ifsq);
1849 lwkt_serialize_exit(&sc->vtnet_slz);
1853 vtnet_start_locked(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1855 struct vtnet_softc *sc;
1856 struct virtqueue *vq;
1861 vq = sc->vtnet_tx_vq;
1864 ASSERT_SERIALIZED(&sc->vtnet_slz);
1866 if ((ifp->if_flags & (IFF_RUNNING)) !=
1867 IFF_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0))
1870 #ifdef VTNET_TX_INTR_MODERATION
1871 if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2)
1875 while (!ifsq_is_empty(ifsq)) {
1876 if (virtqueue_full(vq)) {
1877 ifq_set_oactive(&ifp->if_snd);
1881 m0 = ifq_dequeue(&ifp->if_snd);
1885 if (vtnet_encap(sc, &m0) != 0) {
1888 ifq_prepend(&ifp->if_snd, m0);
1889 ifq_set_oactive(&ifp->if_snd);
1894 ETHER_BPF_MTAP(ifp, m0);
1898 virtqueue_notify(vq, &sc->vtnet_slz);
1899 sc->vtnet_watchdog_timer = VTNET_WATCHDOG_TIMEOUT;
1904 vtnet_tick(void *xsc)
1906 struct vtnet_softc *sc;
1911 ASSERT_SERIALIZED(&sc->vtnet_slz);
1913 virtqueue_dump(sc->vtnet_rx_vq);
1914 virtqueue_dump(sc->vtnet_tx_vq);
1918 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
1923 vtnet_tx_intr_task(void *arg)
1925 struct vtnet_softc *sc;
1927 struct ifaltq_subque *ifsq;
1930 ifp = sc->vtnet_ifp;
1931 ifsq = ifq_get_subq_default(&ifp->if_snd);
1934 // lwkt_serialize_enter(&sc->vtnet_slz);
1936 if ((ifp->if_flags & IFF_RUNNING) == 0) {
1937 vtnet_enable_tx_intr(sc);
1938 // lwkt_serialize_exit(&sc->vtnet_slz);
1944 if (!ifsq_is_empty(ifsq))
1945 vtnet_start_locked(ifp, ifsq);
1947 if (vtnet_enable_tx_intr(sc) != 0) {
1948 vtnet_disable_tx_intr(sc);
1949 sc->vtnet_stats.tx_task_rescheduled++;
1950 // lwkt_serialize_exit(&sc->vtnet_slz);
1954 // lwkt_serialize_exit(&sc->vtnet_slz);
1958 vtnet_tx_vq_intr(void *xsc)
1960 struct vtnet_softc *sc;
1964 vtnet_disable_tx_intr(sc);
1965 vtnet_tx_intr_task(sc);
1971 vtnet_stop(struct vtnet_softc *sc)
1976 dev = sc->vtnet_dev;
1977 ifp = sc->vtnet_ifp;
1979 ASSERT_SERIALIZED(&sc->vtnet_slz);
1981 sc->vtnet_watchdog_timer = 0;
1982 callout_stop(&sc->vtnet_tick_ch);
1983 ifq_clr_oactive(&ifp->if_snd);
1984 ifp->if_flags &= ~(IFF_RUNNING);
1986 vtnet_disable_rx_intr(sc);
1987 vtnet_disable_tx_intr(sc);
1990 * Stop the host VirtIO adapter. Note this will reset the host
1991 * adapter's state back to the pre-initialized state, so in
1992 * order to make the device usable again, we must drive it
1993 * through virtio_reinit() and virtio_reinit_complete().
1997 sc->vtnet_flags &= ~VTNET_FLAG_LINK;
1999 vtnet_free_rx_mbufs(sc);
2000 vtnet_free_tx_mbufs(sc);
2004 vtnet_virtio_reinit(struct vtnet_softc *sc)
2011 dev = sc->vtnet_dev;
2012 ifp = sc->vtnet_ifp;
2013 features = sc->vtnet_features;
2016 * Re-negotiate with the host, removing any disabled receive
2017 * features. Transmit features are disabled only on our side
2018 * via if_capenable and if_hwassist.
2021 if (ifp->if_capabilities & IFCAP_RXCSUM) {
2022 if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
2023 features &= ~VIRTIO_NET_F_GUEST_CSUM;
2026 if (ifp->if_capabilities & IFCAP_LRO) {
2027 if ((ifp->if_capenable & IFCAP_LRO) == 0)
2028 features &= ~VTNET_LRO_FEATURES;
2031 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2032 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2033 features &= ~VIRTIO_NET_F_CTRL_VLAN;
2036 error = virtio_reinit(dev, features);
2038 device_printf(dev, "virtio reinit error %d\n", error);
2044 vtnet_init_locked(struct vtnet_softc *sc)
2050 dev = sc->vtnet_dev;
2051 ifp = sc->vtnet_ifp;
2053 ASSERT_SERIALIZED(&sc->vtnet_slz);
2055 if (ifp->if_flags & IFF_RUNNING)
2058 /* Stop host's adapter, cancel any pending I/O. */
2061 /* Reinitialize the host device. */
2062 error = vtnet_virtio_reinit(sc);
2065 "reinitialization failed, stopping device...\n");
2070 /* Update host with assigned MAC address. */
2071 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2072 vtnet_set_hwaddr(sc);
2074 ifp->if_hwassist = 0;
2075 if (ifp->if_capenable & IFCAP_TXCSUM)
2076 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2077 if (ifp->if_capenable & IFCAP_TSO4)
2078 ifp->if_hwassist |= CSUM_TSO;
2080 error = vtnet_init_rx_vq(sc);
2083 "cannot allocate mbufs for Rx virtqueue\n");
2088 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
2089 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2090 /* Restore promiscuous and all-multicast modes. */
2091 vtnet_rx_filter(sc);
2093 /* Restore filtered MAC addresses. */
2094 vtnet_rx_filter_mac(sc);
2097 /* Restore VLAN filters. */
2098 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2099 vtnet_rx_filter_vlan(sc);
2103 vtnet_enable_rx_intr(sc);
2104 vtnet_enable_tx_intr(sc);
2107 ifp->if_flags |= IFF_RUNNING;
2108 ifq_clr_oactive(&ifp->if_snd);
2110 virtio_reinit_complete(dev);
2112 vtnet_update_link_status(sc);
2113 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2117 vtnet_init(void *xsc)
2119 struct vtnet_softc *sc;
2123 lwkt_serialize_enter(&sc->vtnet_slz);
2124 vtnet_init_locked(sc);
2125 lwkt_serialize_exit(&sc->vtnet_slz);
2129 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
2130 struct sglist *sg, int readable, int writable)
2132 struct virtqueue *vq;
2135 vq = sc->vtnet_ctrl_vq;
2137 ASSERT_SERIALIZED(&sc->vtnet_slz);
2138 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
2139 ("no control virtqueue"));
2140 KASSERT(virtqueue_empty(vq),
2141 ("control command already enqueued"));
2143 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
2146 virtqueue_notify(vq, &sc->vtnet_slz);
2149 * Poll until the command is complete. Previously, we would
2150 * sleep until the control virtqueue interrupt handler woke
2151 * us up, but dropping the VTNET_MTX leads to serialization
2154 * Furthermore, it appears QEMU/KVM only allocates three MSIX
2155 * vectors. Two of those vectors are needed for the Rx and Tx
2156 * virtqueues. We do not support sharing both a Vq and config
2157 * changed notification on the same MSIX vector.
2159 c = virtqueue_poll(vq, NULL);
2160 KASSERT(c == cookie, ("unexpected control command response"));
2164 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
2167 struct virtio_net_ctrl_hdr hdr __aligned(2);
2169 char aligned_hwaddr[ETHER_ADDR_LEN] __aligned(8);
2173 struct sglist_seg segs[3];
2177 s.hdr.class = VIRTIO_NET_CTRL_MAC;
2178 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
2179 s.ack = VIRTIO_NET_ERR;
2181 /* Copy the mac address into physically contiguous memory */
2182 memcpy(s.aligned_hwaddr, hwaddr, ETHER_ADDR_LEN);
2184 sglist_init(&sg, 3, segs);
2186 error |= sglist_append(&sg, &s.hdr,
2187 sizeof(struct virtio_net_ctrl_hdr));
2188 error |= sglist_append(&sg, s.aligned_hwaddr, ETHER_ADDR_LEN);
2189 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
2190 KASSERT(error == 0 && sg.sg_nseg == 3,
2191 ("%s: error %d adding set MAC msg to sglist", __func__, error));
2193 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
2195 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
2199 vtnet_rx_filter(struct vtnet_softc *sc)
2204 dev = sc->vtnet_dev;
2205 ifp = sc->vtnet_ifp;
2207 ASSERT_SERIALIZED(&sc->vtnet_slz);
2208 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2209 ("CTRL_RX feature not negotiated"));
2211 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
2212 device_printf(dev, "cannot %s promiscuous mode\n",
2213 ifp->if_flags & IFF_PROMISC ? "enable" : "disable");
2215 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
2216 device_printf(dev, "cannot %s all-multicast mode\n",
2217 ifp->if_flags & IFF_ALLMULTI ? "enable" : "disable");
2221 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
2223 struct sglist_seg segs[3];
2226 struct virtio_net_ctrl_hdr hdr __aligned(2);
2234 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2235 ("%s: CTRL_RX feature not negotiated", __func__));
2237 s.hdr.class = VIRTIO_NET_CTRL_RX;
2240 s.ack = VIRTIO_NET_ERR;
2242 sglist_init(&sg, 3, segs);
2244 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
2245 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
2246 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
2247 KASSERT(error == 0 && sg.sg_nseg == 3,
2248 ("%s: error %d adding Rx message to sglist", __func__, error));
2250 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
2252 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
2256 vtnet_set_promisc(struct vtnet_softc *sc, int on)
2259 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
2263 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
2266 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
2270 vtnet_rx_filter_mac(struct vtnet_softc *sc)
2272 struct virtio_net_ctrl_hdr hdr __aligned(2);
2273 struct vtnet_mac_filter *filter;
2274 struct sglist_seg segs[4];
2278 struct ifaddr_container *ifac;
2279 struct ifmultiaddr *ifma;
2280 int ucnt, mcnt, promisc, allmulti, error;
2283 ifp = sc->vtnet_ifp;
2289 ASSERT_SERIALIZED(&sc->vtnet_slz);
2290 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2291 ("%s: CTRL_RX feature not negotiated", __func__));
2293 /* Use the MAC filtering table allocated in vtnet_attach. */
2294 filter = sc->vtnet_macfilter;
2295 memset(filter, 0, sizeof(struct vtnet_mac_filter));
2297 /* Unicast MAC addresses: */
2298 //if_addr_rlock(ifp);
2299 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2301 if (ifa->ifa_addr->sa_family != AF_LINK)
2303 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2304 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
2306 else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
2311 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2312 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
2315 //if_addr_runlock(ifp);
2318 filter->vmf_unicast.nentries = 0;
2319 if_printf(ifp, "more than %d MAC addresses assigned, "
2320 "falling back to promiscuous mode\n",
2321 VTNET_MAX_MAC_ENTRIES);
2323 filter->vmf_unicast.nentries = ucnt;
2325 /* Multicast MAC addresses: */
2326 //if_maddr_rlock(ifp);
2327 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2328 if (ifma->ifma_addr->sa_family != AF_LINK)
2330 else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
2335 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
2336 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
2339 //if_maddr_runlock(ifp);
2341 if (allmulti != 0) {
2342 filter->vmf_multicast.nentries = 0;
2343 if_printf(ifp, "more than %d multicast MAC addresses "
2344 "assigned, falling back to all-multicast mode\n",
2345 VTNET_MAX_MAC_ENTRIES);
2347 filter->vmf_multicast.nentries = mcnt;
2349 if (promisc != 0 && allmulti != 0)
2352 hdr.class = VIRTIO_NET_CTRL_MAC;
2353 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
2354 ack = VIRTIO_NET_ERR;
2356 sglist_init(&sg, 4, segs);
2358 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2359 error |= sglist_append(&sg, &filter->vmf_unicast,
2360 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
2361 error |= sglist_append(&sg, &filter->vmf_multicast,
2362 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
2363 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2364 KASSERT(error == 0 && sg.sg_nseg == 4,
2365 ("%s: error %d adding MAC filter msg to sglist", __func__, error));
2367 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2369 if (ack != VIRTIO_NET_OK)
2370 if_printf(ifp, "error setting host MAC filter table\n");
2373 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
2374 if_printf(ifp, "cannot enable promiscuous mode\n");
2375 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
2376 if_printf(ifp, "cannot enable all-multicast mode\n");
2380 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2382 struct sglist_seg segs[3];
2385 struct virtio_net_ctrl_hdr hdr __aligned(2);
2393 s.hdr.class = VIRTIO_NET_CTRL_VLAN;
2394 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
2396 s.ack = VIRTIO_NET_ERR;
2398 sglist_init(&sg, 3, segs);
2400 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
2401 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
2402 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
2403 KASSERT(error == 0 && sg.sg_nseg == 3,
2404 ("%s: error %d adding VLAN message to sglist", __func__, error));
2406 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
2408 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
2412 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
2418 ASSERT_SERIALIZED(&sc->vtnet_slz);
2419 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2420 ("%s: VLAN_FILTER feature not negotiated", __func__));
2422 nvlans = sc->vtnet_nvlans;
2424 /* Enable the filter for each configured VLAN. */
2425 for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) {
2426 w = sc->vtnet_vlan_shadow[i];
2427 while ((bit = ffs(w) - 1) != -1) {
2429 tag = sizeof(w) * CHAR_BIT * i + bit;
2432 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
2433 device_printf(sc->vtnet_dev,
2434 "cannot enable VLAN %d filter\n", tag);
2439 KASSERT(nvlans == 0, ("VLAN count incorrect"));
2443 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2448 ifp = sc->vtnet_ifp;
2449 idx = (tag >> 5) & 0x7F;
2452 if (tag == 0 || tag > 4095)
2455 lwkt_serialize_enter(&sc->vtnet_slz);
2457 /* Update shadow VLAN table. */
2460 sc->vtnet_vlan_shadow[idx] |= (1 << bit);
2463 sc->vtnet_vlan_shadow[idx] &= ~(1 << bit);
2466 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
2467 vtnet_exec_vlan_filter(sc, add, tag) != 0) {
2468 device_printf(sc->vtnet_dev,
2469 "cannot %s VLAN %d %s the host filter table\n",
2470 add ? "add" : "remove", tag, add ? "to" : "from");
2473 lwkt_serialize_exit(&sc->vtnet_slz);
2477 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2480 if (ifp->if_softc != arg)
2483 vtnet_update_vlan_filter(arg, 1, tag);
2487 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2490 if (ifp->if_softc != arg)
2493 vtnet_update_vlan_filter(arg, 0, tag);
2497 vtnet_ifmedia_upd(struct ifnet *ifp)
2499 struct vtnet_softc *sc;
2500 struct ifmedia *ifm;
2503 ifm = &sc->vtnet_media;
2505 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
2512 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2514 struct vtnet_softc *sc;
2518 ifmr->ifm_status = IFM_AVALID;
2519 ifmr->ifm_active = IFM_ETHER;
2521 lwkt_serialize_enter(&sc->vtnet_slz);
2522 if (vtnet_is_link_up(sc) != 0) {
2523 ifmr->ifm_status |= IFM_ACTIVE;
2524 ifmr->ifm_active |= VTNET_MEDIATYPE;
2526 ifmr->ifm_active |= IFM_NONE;
2527 lwkt_serialize_exit(&sc->vtnet_slz);
2531 vtnet_add_statistics(struct vtnet_softc *sc)
2534 struct vtnet_statistics *stats;
2535 struct sysctl_ctx_list *ctx;
2536 struct sysctl_oid *tree;
2537 struct sysctl_oid_list *child;
2539 dev = sc->vtnet_dev;
2540 stats = &sc->vtnet_stats;
2541 ctx = device_get_sysctl_ctx(dev);
2542 tree = device_get_sysctl_tree(dev);
2543 child = SYSCTL_CHILDREN(tree);
2545 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
2546 CTLFLAG_RD, &stats->mbuf_alloc_failed, 0,
2547 "Mbuf cluster allocation failures");
2549 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
2550 CTLFLAG_RD, &stats->rx_frame_too_large, 0,
2551 "Received frame larger than the mbuf chain");
2552 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
2553 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 0,
2554 "Enqueuing the replacement receive mbuf failed");
2555 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
2556 CTLFLAG_RD, &stats->rx_mergeable_failed, 0,
2557 "Mergeable buffers receive failures");
2558 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
2559 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 0,
2560 "Received checksum offloaded buffer with unsupported "
2562 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
2563 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 0,
2564 "Received checksum offloaded buffer with incorrect IP protocol");
2565 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
2566 CTLFLAG_RD, &stats->rx_csum_bad_offset, 0,
2567 "Received checksum offloaded buffer with incorrect offset");
2568 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
2569 CTLFLAG_RD, &stats->rx_csum_failed, 0,
2570 "Received buffer checksum offload failed");
2571 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
2572 CTLFLAG_RD, &stats->rx_csum_offloaded, 0,
2573 "Received buffer checksum offload succeeded");
2574 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
2575 CTLFLAG_RD, &stats->rx_task_rescheduled, 0,
2576 "Times the receive interrupt task rescheduled itself");
2578 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
2579 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 0,
2580 "Aborted transmit of checksum offloaded buffer with unknown "
2582 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
2583 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 0,
2584 "Aborted transmit of TSO buffer with unknown Ethernet type");
2585 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
2586 CTLFLAG_RD, &stats->tx_defragged, 0,
2587 "Transmit mbufs defragged");
2588 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
2589 CTLFLAG_RD, &stats->tx_defrag_failed, 0,
2590 "Aborted transmit of buffer because defrag failed");
2591 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
2592 CTLFLAG_RD, &stats->tx_csum_offloaded, 0,
2593 "Offloaded checksum of transmitted buffer");
2594 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
2595 CTLFLAG_RD, &stats->tx_tso_offloaded, 0,
2596 "Segmentation offload of transmitted buffer");
2597 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
2598 CTLFLAG_RD, &stats->tx_task_rescheduled, 0,
2599 "Times the transmit interrupt task rescheduled itself");
2603 vtnet_enable_rx_intr(struct vtnet_softc *sc)
2606 return (virtqueue_enable_intr(sc->vtnet_rx_vq));
2610 vtnet_disable_rx_intr(struct vtnet_softc *sc)
2613 virtqueue_disable_intr(sc->vtnet_rx_vq);
2617 vtnet_enable_tx_intr(struct vtnet_softc *sc)
2620 #ifdef VTNET_TX_INTR_MODERATION
2623 return (virtqueue_enable_intr(sc->vtnet_tx_vq));
2628 vtnet_disable_tx_intr(struct vtnet_softc *sc)
2631 virtqueue_disable_intr(sc->vtnet_tx_vq);