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_probe(device_t);
78 static int vtnet_attach(device_t);
79 static int vtnet_detach(device_t);
80 static int vtnet_suspend(device_t);
81 static int vtnet_resume(device_t);
82 static int vtnet_shutdown(device_t);
83 static int vtnet_config_change(device_t);
85 static void vtnet_negotiate_features(struct vtnet_softc *);
86 static int vtnet_alloc_virtqueues(struct vtnet_softc *);
87 static void vtnet_get_hwaddr(struct vtnet_softc *);
88 static void vtnet_set_hwaddr(struct vtnet_softc *);
89 static int vtnet_is_link_up(struct vtnet_softc *);
90 static void vtnet_update_link_status(struct vtnet_softc *);
92 static void vtnet_watchdog(struct vtnet_softc *);
94 static void vtnet_config_change_task(void *, int);
95 static int vtnet_setup_interface(struct vtnet_softc *);
96 static int vtnet_change_mtu(struct vtnet_softc *, int);
97 static int vtnet_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
99 static int vtnet_init_rx_vq(struct vtnet_softc *);
100 static void vtnet_free_rx_mbufs(struct vtnet_softc *);
101 static void vtnet_free_tx_mbufs(struct vtnet_softc *);
102 static void vtnet_free_ctrl_vq(struct vtnet_softc *);
104 static struct mbuf * vtnet_alloc_rxbuf(struct vtnet_softc *, int,
106 static int vtnet_replace_rxbuf(struct vtnet_softc *,
108 static int vtnet_newbuf(struct vtnet_softc *);
109 static void vtnet_discard_merged_rxbuf(struct vtnet_softc *, int);
110 static void vtnet_discard_rxbuf(struct vtnet_softc *, struct mbuf *);
111 static int vtnet_enqueue_rxbuf(struct vtnet_softc *, struct mbuf *);
112 static void vtnet_vlan_tag_remove(struct mbuf *);
113 static int vtnet_rx_csum(struct vtnet_softc *, struct mbuf *,
114 struct virtio_net_hdr *);
115 static int vtnet_rxeof_merged(struct vtnet_softc *, struct mbuf *, int);
116 static int vtnet_rxeof(struct vtnet_softc *, int, int *);
117 static void vtnet_rx_intr_task(void *);
118 static int vtnet_rx_vq_intr(void *);
120 static void vtnet_enqueue_txhdr(struct vtnet_softc *,
121 struct vtnet_tx_header *);
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, NULL, NULL);
233 MODULE_VERSION(vtnet, 1);
234 MODULE_DEPEND(vtnet, virtio, 1, 1, 1);
237 vtnet_probe(device_t dev)
239 if (virtio_get_device_type(dev) != VIRTIO_ID_NETWORK)
242 device_set_desc(dev, "VirtIO Networking Adapter");
244 return (BUS_PROBE_DEFAULT);
248 vtnet_attach(device_t dev)
250 struct vtnet_softc *sc;
253 sc = device_get_softc(dev);
256 lwkt_serialize_init(&sc->vtnet_slz);
257 callout_init(&sc->vtnet_tick_ch);
259 ifmedia_init(&sc->vtnet_media, IFM_IMASK, vtnet_ifmedia_upd,
261 ifmedia_add(&sc->vtnet_media, VTNET_MEDIATYPE, 0, NULL);
262 ifmedia_set(&sc->vtnet_media, VTNET_MEDIATYPE);
264 vtnet_add_statistics(sc);
265 SLIST_INIT(&sc->vtnet_txhdr_free);
267 /* Register our feature descriptions. */
268 virtio_set_feature_desc(dev, vtnet_feature_desc);
269 vtnet_negotiate_features(sc);
271 if (virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC))
272 sc->vtnet_flags |= VTNET_FLAG_INDIRECT;
274 if (virtio_with_feature(dev, VIRTIO_NET_F_MAC)) {
275 /* This feature should always be negotiated. */
276 sc->vtnet_flags |= VTNET_FLAG_MAC;
279 if (virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF)) {
280 sc->vtnet_flags |= VTNET_FLAG_MRG_RXBUFS;
281 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
283 sc->vtnet_hdr_size = sizeof(struct virtio_net_hdr);
286 sc->vtnet_rx_mbuf_size = MCLBYTES;
287 sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
289 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VQ)) {
290 sc->vtnet_flags |= VTNET_FLAG_CTRL_VQ;
292 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
293 sc->vtnet_flags |= VTNET_FLAG_CTRL_RX;
294 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_VLAN))
295 sc->vtnet_flags |= VTNET_FLAG_VLAN_FILTER;
296 if (virtio_with_feature(dev, VIRTIO_NET_F_CTRL_MAC_ADDR) &&
297 virtio_with_feature(dev, VIRTIO_NET_F_CTRL_RX))
298 sc->vtnet_flags |= VTNET_FLAG_CTRL_MAC;
301 /* Read (or generate) the MAC address for the adapter. */
302 vtnet_get_hwaddr(sc);
304 error = vtnet_alloc_virtqueues(sc);
306 device_printf(dev, "cannot allocate virtqueues\n");
310 error = vtnet_setup_interface(sc);
312 device_printf(dev, "cannot setup interface\n");
316 TASK_INIT(&sc->vtnet_cfgchg_task, 0, vtnet_config_change_task, sc);
318 error = virtio_setup_intr(dev, &sc->vtnet_slz);
320 device_printf(dev, "cannot setup virtqueue interrupts\n");
321 ether_ifdetach(sc->vtnet_ifp);
326 * Device defaults to promiscuous mode for backwards
327 * compatibility. Turn it off if possible.
329 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
330 lwkt_serialize_enter(&sc->vtnet_slz);
331 if (vtnet_set_promisc(sc, 0) != 0) {
332 sc->vtnet_ifp->if_flags |= IFF_PROMISC;
334 "cannot disable promiscuous mode\n");
336 lwkt_serialize_exit(&sc->vtnet_slz);
338 sc->vtnet_ifp->if_flags |= IFF_PROMISC;
348 vtnet_detach(device_t dev)
350 struct vtnet_softc *sc;
353 sc = device_get_softc(dev);
356 if (device_is_attached(dev)) {
357 lwkt_serialize_enter(&sc->vtnet_slz);
359 lwkt_serialize_exit(&sc->vtnet_slz);
361 callout_stop(&sc->vtnet_tick_ch);
362 taskqueue_drain(taskqueue_swi, &sc->vtnet_cfgchg_task);
367 if (sc->vtnet_vlan_attach != NULL) {
368 EVENTHANDLER_DEREGISTER(vlan_config, sc->vtnet_vlan_attach);
369 sc->vtnet_vlan_attach = NULL;
371 if (sc->vtnet_vlan_detach != NULL) {
372 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vtnet_vlan_detach);
373 sc->vtnet_vlan_detach = NULL;
378 sc->vtnet_ifp = NULL;
381 if (sc->vtnet_rx_vq != NULL)
382 vtnet_free_rx_mbufs(sc);
383 if (sc->vtnet_tx_vq != NULL)
384 vtnet_free_tx_mbufs(sc);
385 if (sc->vtnet_ctrl_vq != NULL)
386 vtnet_free_ctrl_vq(sc);
388 if (sc->vtnet_txhdrarea != NULL) {
389 contigfree(sc->vtnet_txhdrarea,
390 sc->vtnet_txhdrcount * sizeof(struct vtnet_tx_header),
392 sc->vtnet_txhdrarea = NULL;
394 SLIST_INIT(&sc->vtnet_txhdr_free);
395 if (sc->vtnet_macfilter != NULL) {
396 contigfree(sc->vtnet_macfilter,
397 sizeof(struct vtnet_mac_filter), M_DEVBUF);
398 sc->vtnet_macfilter = NULL;
401 ifmedia_removeall(&sc->vtnet_media);
407 vtnet_suspend(device_t dev)
409 struct vtnet_softc *sc;
411 sc = device_get_softc(dev);
413 lwkt_serialize_enter(&sc->vtnet_slz);
415 sc->vtnet_flags |= VTNET_FLAG_SUSPENDED;
416 lwkt_serialize_exit(&sc->vtnet_slz);
422 vtnet_resume(device_t dev)
424 struct vtnet_softc *sc;
427 sc = device_get_softc(dev);
430 lwkt_serialize_enter(&sc->vtnet_slz);
431 if (ifp->if_flags & IFF_UP)
432 vtnet_init_locked(sc);
433 sc->vtnet_flags &= ~VTNET_FLAG_SUSPENDED;
434 lwkt_serialize_exit(&sc->vtnet_slz);
440 vtnet_shutdown(device_t dev)
444 * Suspend already does all of what we need to
445 * do here; we just never expect to be resumed.
447 return (vtnet_suspend(dev));
451 vtnet_config_change(device_t dev)
453 struct vtnet_softc *sc;
455 sc = device_get_softc(dev);
457 taskqueue_enqueue(taskqueue_thread[mycpuid], &sc->vtnet_cfgchg_task);
463 vtnet_negotiate_features(struct vtnet_softc *sc)
466 uint64_t mask, features;
471 if (vtnet_csum_disable)
472 mask |= VIRTIO_NET_F_CSUM | VIRTIO_NET_F_GUEST_CSUM;
475 * TSO and LRO are only available when their corresponding checksum
476 * offload feature is also negotiated.
479 if (vtnet_csum_disable || vtnet_tso_disable)
480 mask |= VIRTIO_NET_F_HOST_TSO4 | VIRTIO_NET_F_HOST_TSO6 |
481 VIRTIO_NET_F_HOST_ECN;
483 if (vtnet_csum_disable || vtnet_lro_disable)
484 mask |= VTNET_LRO_FEATURES;
486 features = VTNET_FEATURES & ~mask;
487 features |= VIRTIO_F_NOTIFY_ON_EMPTY;
488 features |= VIRTIO_F_ANY_LAYOUT;
489 sc->vtnet_features = virtio_negotiate_features(dev, features);
491 if (virtio_with_feature(dev, VTNET_LRO_FEATURES) &&
492 virtio_with_feature(dev, VIRTIO_NET_F_MRG_RXBUF) == 0) {
494 * LRO without mergeable buffers requires special care. This
495 * is not ideal because every receive buffer must be large
496 * enough to hold the maximum TCP packet, the Ethernet header,
497 * and the header. This requires up to 34 descriptors with
498 * MCLBYTES clusters. If we do not have indirect descriptors,
499 * LRO is disabled since the virtqueue will not contain very
500 * many receive buffers.
502 if (!virtio_with_feature(dev, VIRTIO_RING_F_INDIRECT_DESC)) {
504 "LRO disabled due to both mergeable buffers and "
505 "indirect descriptors not negotiated\n");
507 features &= ~VTNET_LRO_FEATURES;
509 virtio_negotiate_features(dev, features);
511 sc->vtnet_flags |= VTNET_FLAG_LRO_NOMRG;
516 vtnet_alloc_virtqueues(struct vtnet_softc *sc)
519 struct vq_alloc_info vq_info[3];
526 * Indirect descriptors are not needed for the Rx
527 * virtqueue when mergeable buffers are negotiated.
528 * The header is placed inline with the data, not
529 * in a separate descriptor, and mbuf clusters are
530 * always physically contiguous.
532 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
533 sc->vtnet_rx_nsegs = (sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) ?
534 VTNET_MAX_RX_SEGS : VTNET_MIN_RX_SEGS;
536 sc->vtnet_rx_nsegs = VTNET_MRG_RX_SEGS;
538 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4) ||
539 virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
540 sc->vtnet_tx_nsegs = VTNET_MAX_TX_SEGS;
542 sc->vtnet_tx_nsegs = VTNET_MIN_TX_SEGS;
544 VQ_ALLOC_INFO_INIT(&vq_info[0], sc->vtnet_rx_nsegs,
545 vtnet_rx_vq_intr, sc, &sc->vtnet_rx_vq,
546 "%s receive", device_get_nameunit(dev));
548 VQ_ALLOC_INFO_INIT(&vq_info[1], sc->vtnet_tx_nsegs,
549 vtnet_tx_vq_intr, sc, &sc->vtnet_tx_vq,
550 "%s transmit", device_get_nameunit(dev));
552 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
555 VQ_ALLOC_INFO_INIT(&vq_info[2], 0, NULL, NULL,
556 &sc->vtnet_ctrl_vq, "%s control",
557 device_get_nameunit(dev));
560 return (virtio_alloc_virtqueues(dev, 0, nvqs, vq_info));
564 vtnet_setup_interface(struct vtnet_softc *sc)
572 ifp = sc->vtnet_ifp = if_alloc(IFT_ETHER);
574 device_printf(dev, "cannot allocate ifnet structure\n");
579 if_initname(ifp, device_get_name(dev), device_get_unit(dev));
580 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
581 ifp->if_init = vtnet_init;
582 ifp->if_start = vtnet_start;
583 ifp->if_ioctl = vtnet_ioctl;
585 sc->vtnet_rx_size = virtqueue_size(sc->vtnet_rx_vq);
586 sc->vtnet_rx_process_limit = sc->vtnet_rx_size;
588 tx_size = virtqueue_size(sc->vtnet_tx_vq);
589 sc->vtnet_tx_size = tx_size;
590 /* Select size, such that we never run out of tx_header entries. */
591 if (sc->vtnet_flags & VTNET_FLAG_INDIRECT)
592 sc->vtnet_txhdrcount = sc->vtnet_tx_size;
594 sc->vtnet_txhdrcount = (sc->vtnet_tx_size / 2) + 1;
595 sc->vtnet_txhdrarea = contigmalloc(
596 sc->vtnet_txhdrcount * sizeof(struct vtnet_tx_header),
597 M_VTNET, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0);
598 if (sc->vtnet_txhdrarea == NULL) {
599 device_printf(dev, "cannot contigmalloc the tx headers\n");
602 for (i = 0; i < sc->vtnet_txhdrcount; i++)
603 vtnet_enqueue_txhdr(sc, &sc->vtnet_txhdrarea[i]);
604 sc->vtnet_macfilter = contigmalloc(
605 sizeof(struct vtnet_mac_filter),
606 M_DEVBUF, M_WAITOK, 0, BUS_SPACE_MAXADDR, 4, 0);
607 if (sc->vtnet_macfilter == NULL) {
609 "cannot contigmalloc the mac filter table\n");
612 ifq_set_maxlen(&ifp->if_snd, tx_size - 1);
613 ifq_set_ready(&ifp->if_snd);
615 ether_ifattach(ifp, sc->vtnet_hwaddr, NULL);
617 /* Tell the upper layer(s) we support long frames. */
618 ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
619 ifp->if_capabilities |= IFCAP_JUMBO_MTU | IFCAP_VLAN_MTU;
621 if (virtio_with_feature(dev, VIRTIO_NET_F_CSUM)) {
622 ifp->if_capabilities |= IFCAP_TXCSUM;
624 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO4))
625 ifp->if_capabilities |= IFCAP_TSO4;
626 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_TSO6))
627 ifp->if_capabilities |= IFCAP_TSO6;
628 if (ifp->if_capabilities & IFCAP_TSO)
629 ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
631 if (virtio_with_feature(dev, VIRTIO_NET_F_HOST_ECN))
632 sc->vtnet_flags |= VTNET_FLAG_TSO_ECN;
635 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_CSUM)) {
636 ifp->if_capabilities |= IFCAP_RXCSUM;
638 if (virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO4) ||
639 virtio_with_feature(dev, VIRTIO_NET_F_GUEST_TSO6))
640 ifp->if_capabilities |= IFCAP_LRO;
643 if (ifp->if_capabilities & IFCAP_HWCSUM) {
645 * VirtIO does not support VLAN tagging, but we can fake
646 * it by inserting and removing the 802.1Q header during
647 * transmit and receive. We are then able to do checksum
648 * offloading of VLAN frames.
650 ifp->if_capabilities |=
651 IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_HWCSUM;
654 ifp->if_capenable = ifp->if_capabilities;
657 * Capabilities after here are not enabled by default.
660 if (sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER) {
661 ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
663 sc->vtnet_vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
664 vtnet_register_vlan, sc, EVENTHANDLER_PRI_FIRST);
665 sc->vtnet_vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
666 vtnet_unregister_vlan, sc, EVENTHANDLER_PRI_FIRST);
673 vtnet_set_hwaddr(struct vtnet_softc *sc)
679 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_MAC) &&
680 (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)) {
681 if (vtnet_ctrl_mac_cmd(sc, sc->vtnet_hwaddr) != 0)
682 device_printf(dev, "unable to set MAC address\n");
683 } else if (sc->vtnet_flags & VTNET_FLAG_MAC) {
684 virtio_write_device_config(dev,
685 offsetof(struct virtio_net_config, mac),
686 sc->vtnet_hwaddr, ETHER_ADDR_LEN);
691 vtnet_get_hwaddr(struct vtnet_softc *sc)
697 if ((sc->vtnet_flags & VTNET_FLAG_MAC) == 0) {
699 * Generate a random locally administered unicast address.
701 * It would be nice to generate the same MAC address across
702 * reboots, but it seems all the hosts currently available
703 * support the MAC feature, so this isn't too important.
705 sc->vtnet_hwaddr[0] = 0xB2;
706 karc4rand(&sc->vtnet_hwaddr[1], ETHER_ADDR_LEN - 1);
707 vtnet_set_hwaddr(sc);
711 virtio_read_device_config(dev,
712 offsetof(struct virtio_net_config, mac),
713 sc->vtnet_hwaddr, ETHER_ADDR_LEN);
717 vtnet_is_link_up(struct vtnet_softc *sc)
726 ASSERT_SERIALIZED(&sc->vtnet_slz);
728 if (virtio_with_feature(dev, VIRTIO_NET_F_STATUS)) {
729 status = virtio_read_dev_config_2(dev,
730 offsetof(struct virtio_net_config, status));
732 status = VIRTIO_NET_S_LINK_UP;
735 return ((status & VIRTIO_NET_S_LINK_UP) != 0);
739 vtnet_update_link_status(struct vtnet_softc *sc)
743 struct ifaltq_subque *ifsq;
748 ifsq = ifq_get_subq_default(&ifp->if_snd);
750 link = vtnet_is_link_up(sc);
752 if (link && ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0)) {
753 sc->vtnet_flags |= VTNET_FLAG_LINK;
755 device_printf(dev, "Link is up\n");
756 ifp->if_link_state = LINK_STATE_UP;
757 if_link_state_change(ifp);
758 if (!ifsq_is_empty(ifsq))
759 vtnet_start_locked(ifp, ifsq);
760 } else if (!link && (sc->vtnet_flags & VTNET_FLAG_LINK)) {
761 sc->vtnet_flags &= ~VTNET_FLAG_LINK;
763 device_printf(dev, "Link is down\n");
765 ifp->if_link_state = LINK_STATE_DOWN;
766 if_link_state_change(ifp);
772 vtnet_watchdog(struct vtnet_softc *sc)
778 #ifdef VTNET_TX_INTR_MODERATION
782 if (sc->vtnet_watchdog_timer == 0 || --sc->vtnet_watchdog_timer)
785 if_printf(ifp, "watchdog timeout -- resetting\n");
787 virtqueue_dump(sc->vtnet_tx_vq);
790 ifp->if_flags &= ~IFF_RUNNING;
791 vtnet_init_locked(sc);
796 vtnet_config_change_task(void *arg, int pending)
798 struct vtnet_softc *sc;
802 lwkt_serialize_enter(&sc->vtnet_slz);
803 vtnet_update_link_status(sc);
804 lwkt_serialize_exit(&sc->vtnet_slz);
808 vtnet_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data,struct ucred *cr)
810 struct vtnet_softc *sc;
812 int reinit, mask, error;
815 ifr = (struct ifreq *) data;
821 if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > VTNET_MAX_MTU)
823 else if (ifp->if_mtu != ifr->ifr_mtu) {
824 lwkt_serialize_enter(&sc->vtnet_slz);
825 error = vtnet_change_mtu(sc, ifr->ifr_mtu);
826 lwkt_serialize_exit(&sc->vtnet_slz);
831 lwkt_serialize_enter(&sc->vtnet_slz);
832 if ((ifp->if_flags & IFF_UP) == 0) {
833 if (ifp->if_flags & IFF_RUNNING)
835 } else if (ifp->if_flags & IFF_RUNNING) {
836 if ((ifp->if_flags ^ sc->vtnet_if_flags) &
837 (IFF_PROMISC | IFF_ALLMULTI)) {
838 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX)
844 vtnet_init_locked(sc);
847 sc->vtnet_if_flags = ifp->if_flags;
848 lwkt_serialize_exit(&sc->vtnet_slz);
853 lwkt_serialize_enter(&sc->vtnet_slz);
854 if ((sc->vtnet_flags & VTNET_FLAG_CTRL_RX) &&
855 (ifp->if_flags & IFF_RUNNING))
856 vtnet_rx_filter_mac(sc);
857 lwkt_serialize_exit(&sc->vtnet_slz);
862 error = ifmedia_ioctl(ifp, ifr, &sc->vtnet_media, cmd);
866 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
868 lwkt_serialize_enter(&sc->vtnet_slz);
870 if (mask & IFCAP_TXCSUM) {
871 ifp->if_capenable ^= IFCAP_TXCSUM;
872 if (ifp->if_capenable & IFCAP_TXCSUM)
873 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
875 ifp->if_hwassist &= ~VTNET_CSUM_OFFLOAD;
878 if (mask & IFCAP_TSO4) {
879 ifp->if_capenable ^= IFCAP_TSO4;
880 if (ifp->if_capenable & IFCAP_TSO4)
881 ifp->if_hwassist |= CSUM_TSO;
883 ifp->if_hwassist &= ~CSUM_TSO;
886 if (mask & IFCAP_RXCSUM) {
887 ifp->if_capenable ^= IFCAP_RXCSUM;
891 if (mask & IFCAP_LRO) {
892 ifp->if_capenable ^= IFCAP_LRO;
896 if (mask & IFCAP_VLAN_HWFILTER) {
897 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
901 if (mask & IFCAP_VLAN_HWTSO)
902 ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
904 if (mask & IFCAP_VLAN_HWTAGGING)
905 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
907 if (reinit && (ifp->if_flags & IFF_RUNNING)) {
908 ifp->if_flags &= ~IFF_RUNNING;
909 vtnet_init_locked(sc);
911 //VLAN_CAPABILITIES(ifp);
913 lwkt_serialize_exit(&sc->vtnet_slz);
917 error = ether_ioctl(ifp, cmd, data);
925 vtnet_change_mtu(struct vtnet_softc *sc, int new_mtu)
928 int new_frame_size, clsize;
932 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
933 new_frame_size = sizeof(struct vtnet_rx_header) +
934 sizeof(struct ether_vlan_header) + new_mtu;
936 if (new_frame_size > MJUM9BYTES)
939 if (new_frame_size <= MCLBYTES)
944 new_frame_size = sizeof(struct virtio_net_hdr_mrg_rxbuf) +
945 sizeof(struct ether_vlan_header) + new_mtu;
947 if (new_frame_size <= MCLBYTES)
950 clsize = MJUMPAGESIZE;
953 sc->vtnet_rx_mbuf_size = clsize;
954 sc->vtnet_rx_mbuf_count = VTNET_NEEDED_RX_MBUFS(sc);
955 KASSERT(sc->vtnet_rx_mbuf_count < VTNET_MAX_RX_SEGS,
956 ("too many rx mbufs: %d", sc->vtnet_rx_mbuf_count));
958 ifp->if_mtu = new_mtu;
960 if (ifp->if_flags & IFF_RUNNING) {
961 ifp->if_flags &= ~IFF_RUNNING;
962 vtnet_init_locked(sc);
969 vtnet_init_rx_vq(struct vtnet_softc *sc)
971 struct virtqueue *vq;
974 vq = sc->vtnet_rx_vq;
978 while (!virtqueue_full(vq)) {
979 if ((error = vtnet_newbuf(sc)) != 0)
985 virtqueue_notify(vq, &sc->vtnet_slz);
988 * EMSGSIZE signifies the virtqueue did not have enough
989 * entries available to hold the last mbuf. This is not
990 * an error. We should not get ENOSPC since we check if
991 * the virtqueue is full before attempting to add a
994 if (error == EMSGSIZE)
1002 vtnet_free_rx_mbufs(struct vtnet_softc *sc)
1004 struct virtqueue *vq;
1008 vq = sc->vtnet_rx_vq;
1011 while ((m = virtqueue_drain(vq, &last)) != NULL)
1014 KASSERT(virtqueue_empty(vq), ("mbufs remaining in Rx Vq"));
1018 vtnet_free_tx_mbufs(struct vtnet_softc *sc)
1020 struct virtqueue *vq;
1021 struct vtnet_tx_header *txhdr;
1024 vq = sc->vtnet_tx_vq;
1027 while ((txhdr = virtqueue_drain(vq, &last)) != NULL) {
1028 m_freem(txhdr->vth_mbuf);
1029 vtnet_enqueue_txhdr(sc, txhdr);
1032 KASSERT(virtqueue_empty(vq), ("mbufs remaining in Tx Vq"));
1036 vtnet_free_ctrl_vq(struct vtnet_softc *sc)
1039 * The control virtqueue is only polled, therefore
1040 * it should already be empty.
1042 KASSERT(virtqueue_empty(sc->vtnet_ctrl_vq),
1043 ("Ctrl Vq not empty"));
1046 static struct mbuf *
1047 vtnet_alloc_rxbuf(struct vtnet_softc *sc, int nbufs, struct mbuf **m_tailp)
1049 struct mbuf *m_head, *m_tail, *m;
1052 clsize = sc->vtnet_rx_mbuf_size;
1054 /*use getcl instead of getjcl. see if_mxge.c comment line 2398*/
1055 //m_head = m_getjcl(M_DONTWAIT, MT_DATA, M_PKTHDR, clsize);
1056 m_head = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR );
1060 m_head->m_len = clsize;
1064 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1065 ("chained Rx mbuf requested without LRO_NOMRG"));
1067 for (i = 0; i < nbufs - 1; i++) {
1068 //m = m_getjcl(M_DONTWAIT, MT_DATA, 0, clsize);
1069 m = m_getcl(M_NOWAIT, MT_DATA, 0);
1079 if (m_tailp != NULL)
1085 sc->vtnet_stats.mbuf_alloc_failed++;
1092 vtnet_replace_rxbuf(struct vtnet_softc *sc, struct mbuf *m0, int len0)
1094 struct mbuf *m, *m_prev;
1095 struct mbuf *m_new, *m_tail;
1096 int len, clsize, nreplace, error;
1103 clsize = sc->vtnet_rx_mbuf_size;
1106 if (m->m_next != NULL)
1107 KASSERT(sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG,
1108 ("chained Rx mbuf without LRO_NOMRG"));
1111 * Since LRO_NOMRG mbuf chains are so large, we want to avoid
1112 * allocating an entire chain for each received frame. When
1113 * the received frame's length is less than that of the chain,
1114 * the unused mbufs are reassigned to the new chain.
1118 * Something is seriously wrong if we received
1119 * a frame larger than the mbuf chain. Drop it.
1122 sc->vtnet_stats.rx_frame_too_large++;
1126 KASSERT(m->m_len == clsize,
1127 ("mbuf length not expected cluster size: %d",
1130 m->m_len = MIN(m->m_len, len);
1138 KASSERT(m_prev != NULL, ("m_prev == NULL"));
1139 KASSERT(nreplace <= sc->vtnet_rx_mbuf_count,
1140 ("too many replacement mbufs: %d/%d", nreplace,
1141 sc->vtnet_rx_mbuf_count));
1143 m_new = vtnet_alloc_rxbuf(sc, nreplace, &m_tail);
1144 if (m_new == NULL) {
1145 m_prev->m_len = clsize;
1150 * Move unused mbufs, if any, from the original chain
1151 * onto the end of the new chain.
1153 if (m_prev->m_next != NULL) {
1154 m_tail->m_next = m_prev->m_next;
1155 m_prev->m_next = NULL;
1158 error = vtnet_enqueue_rxbuf(sc, m_new);
1161 * BAD! We could not enqueue the replacement mbuf chain. We
1162 * must restore the m0 chain to the original state if it was
1163 * modified so we can subsequently discard it.
1165 * NOTE: The replacement is suppose to be an identical copy
1166 * to the one just dequeued so this is an unexpected error.
1168 sc->vtnet_stats.rx_enq_replacement_failed++;
1170 if (m_tail->m_next != NULL) {
1171 m_prev->m_next = m_tail->m_next;
1172 m_tail->m_next = NULL;
1175 m_prev->m_len = clsize;
1183 vtnet_newbuf(struct vtnet_softc *sc)
1188 m = vtnet_alloc_rxbuf(sc, sc->vtnet_rx_mbuf_count, NULL);
1192 error = vtnet_enqueue_rxbuf(sc, m);
1200 vtnet_discard_merged_rxbuf(struct vtnet_softc *sc, int nbufs)
1202 struct virtqueue *vq;
1205 vq = sc->vtnet_rx_vq;
1207 while (--nbufs > 0) {
1208 if ((m = virtqueue_dequeue(vq, NULL)) == NULL)
1210 vtnet_discard_rxbuf(sc, m);
1215 vtnet_discard_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1220 * Requeue the discarded mbuf. This should always be
1221 * successful since it was just dequeued.
1223 error = vtnet_enqueue_rxbuf(sc, m);
1224 KASSERT(error == 0, ("cannot requeue discarded mbuf"));
1228 vtnet_enqueue_rxbuf(struct vtnet_softc *sc, struct mbuf *m)
1231 struct sglist_seg segs[VTNET_MAX_RX_SEGS];
1232 struct vtnet_rx_header *rxhdr;
1233 struct virtio_net_hdr *hdr;
1237 ASSERT_SERIALIZED(&sc->vtnet_slz);
1238 if ((sc->vtnet_flags & VTNET_FLAG_LRO_NOMRG) == 0)
1239 KASSERT(m->m_next == NULL, ("chained Rx mbuf"));
1241 sglist_init(&sg, sc->vtnet_rx_nsegs, segs);
1243 mdata = mtod(m, uint8_t *);
1246 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1247 rxhdr = (struct vtnet_rx_header *) mdata;
1248 hdr = &rxhdr->vrh_hdr;
1249 offset += sizeof(struct vtnet_rx_header);
1251 error = sglist_append(&sg, hdr, sc->vtnet_hdr_size);
1252 KASSERT(error == 0, ("cannot add header to sglist"));
1255 error = sglist_append(&sg, mdata + offset, m->m_len - offset);
1259 if (m->m_next != NULL) {
1260 error = sglist_append_mbuf(&sg, m->m_next);
1265 return (virtqueue_enqueue(sc->vtnet_rx_vq, m, &sg, 0, sg.sg_nseg));
1269 vtnet_vlan_tag_remove(struct mbuf *m)
1271 struct ether_vlan_header *evl;
1273 evl = mtod(m, struct ether_vlan_header *);
1275 m->m_pkthdr.ether_vlantag = ntohs(evl->evl_tag);
1276 m->m_flags |= M_VLANTAG;
1278 /* Strip the 802.1Q header. */
1279 bcopy((char *) evl, (char *) evl + ETHER_VLAN_ENCAP_LEN,
1280 ETHER_HDR_LEN - ETHER_TYPE_LEN);
1281 m_adj(m, ETHER_VLAN_ENCAP_LEN);
1285 * Alternative method of doing receive checksum offloading. Rather
1286 * than parsing the received frame down to the IP header, use the
1287 * csum_offset to determine which CSUM_* flags are appropriate. We
1288 * can get by with doing this only because the checksum offsets are
1289 * unique for the things we care about.
1292 vtnet_rx_csum(struct vtnet_softc *sc, struct mbuf *m,
1293 struct virtio_net_hdr *hdr)
1295 struct ether_header *eh;
1296 struct ether_vlan_header *evh;
1301 csum_len = hdr->csum_start + hdr->csum_offset;
1303 if (csum_len < sizeof(struct ether_header) + sizeof(struct ip))
1305 if (m->m_len < csum_len)
1308 eh = mtod(m, struct ether_header *);
1309 eth_type = ntohs(eh->ether_type);
1310 if (eth_type == ETHERTYPE_VLAN) {
1311 evh = mtod(m, struct ether_vlan_header *);
1312 eth_type = ntohs(evh->evl_proto);
1315 if (eth_type != ETHERTYPE_IP && eth_type != ETHERTYPE_IPV6) {
1316 sc->vtnet_stats.rx_csum_bad_ethtype++;
1320 /* Use the offset to determine the appropriate CSUM_* flags. */
1321 switch (hdr->csum_offset) {
1322 case offsetof(struct udphdr, uh_sum):
1323 if (m->m_len < hdr->csum_start + sizeof(struct udphdr))
1325 udp = (struct udphdr *)(mtod(m, uint8_t *) + hdr->csum_start);
1326 if (udp->uh_sum == 0)
1331 case offsetof(struct tcphdr, th_sum):
1332 m->m_pkthdr.csum_flags |= CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
1333 m->m_pkthdr.csum_data = 0xFFFF;
1337 sc->vtnet_stats.rx_csum_bad_offset++;
1341 sc->vtnet_stats.rx_csum_offloaded++;
1347 vtnet_rxeof_merged(struct vtnet_softc *sc, struct mbuf *m_head, int nbufs)
1350 struct virtqueue *vq;
1351 struct mbuf *m, *m_tail;
1354 ifp = sc->vtnet_ifp;
1355 vq = sc->vtnet_rx_vq;
1358 while (--nbufs > 0) {
1359 m = virtqueue_dequeue(vq, &len);
1365 if (vtnet_newbuf(sc) != 0) {
1367 vtnet_discard_rxbuf(sc, m);
1369 vtnet_discard_merged_rxbuf(sc, nbufs);
1377 m->m_flags &= ~M_PKTHDR;
1379 m_head->m_pkthdr.len += len;
1387 sc->vtnet_stats.rx_mergeable_failed++;
1394 vtnet_rxeof(struct vtnet_softc *sc, int count, int *rx_npktsp)
1396 struct virtio_net_hdr lhdr;
1398 struct virtqueue *vq;
1400 struct ether_header *eh;
1401 struct virtio_net_hdr *hdr;
1402 struct virtio_net_hdr_mrg_rxbuf *mhdr;
1403 int len, deq, nbufs, adjsz, rx_npkts;
1405 ifp = sc->vtnet_ifp;
1406 vq = sc->vtnet_rx_vq;
1411 ASSERT_SERIALIZED(&sc->vtnet_slz);
1413 while (--count >= 0) {
1414 m = virtqueue_dequeue(vq, &len);
1419 if (len < sc->vtnet_hdr_size + ETHER_HDR_LEN) {
1421 vtnet_discard_rxbuf(sc, m);
1425 if ((sc->vtnet_flags & VTNET_FLAG_MRG_RXBUFS) == 0) {
1427 adjsz = sizeof(struct vtnet_rx_header);
1429 * Account for our pad between the header and
1430 * the actual start of the frame.
1432 len += VTNET_RX_HEADER_PAD;
1434 mhdr = mtod(m, struct virtio_net_hdr_mrg_rxbuf *);
1435 nbufs = mhdr->num_buffers;
1436 adjsz = sizeof(struct virtio_net_hdr_mrg_rxbuf);
1439 if (vtnet_replace_rxbuf(sc, m, len) != 0) {
1441 vtnet_discard_rxbuf(sc, m);
1443 vtnet_discard_merged_rxbuf(sc, nbufs);
1447 m->m_pkthdr.len = len;
1448 m->m_pkthdr.rcvif = ifp;
1449 m->m_pkthdr.csum_flags = 0;
1452 if (vtnet_rxeof_merged(sc, m, nbufs) != 0)
1459 * Save copy of header before we strip it. For both mergeable
1460 * and non-mergeable, the VirtIO header is placed first in the
1461 * mbuf's data. We no longer need num_buffers, so always use a
1464 memcpy(hdr, mtod(m, void *), sizeof(struct virtio_net_hdr));
1467 if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING) {
1468 eh = mtod(m, struct ether_header *);
1469 if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1470 vtnet_vlan_tag_remove(m);
1473 * With the 802.1Q header removed, update the
1474 * checksum starting location accordingly.
1476 if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)
1478 ETHER_VLAN_ENCAP_LEN;
1482 if (ifp->if_capenable & IFCAP_RXCSUM &&
1483 hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
1484 if (vtnet_rx_csum(sc, m, hdr) != 0)
1485 sc->vtnet_stats.rx_csum_failed++;
1488 lwkt_serialize_exit(&sc->vtnet_slz);
1490 ifp->if_input(ifp, m, NULL, -1);
1491 lwkt_serialize_enter(&sc->vtnet_slz);
1494 * The interface may have been stopped while we were
1495 * passing the packet up the network stack.
1497 if ((ifp->if_flags & IFF_RUNNING) == 0)
1501 virtqueue_notify(vq, &sc->vtnet_slz);
1503 if (rx_npktsp != NULL)
1504 *rx_npktsp = rx_npkts;
1506 return (count > 0 ? 0 : EAGAIN);
1510 vtnet_rx_intr_task(void *arg)
1512 struct vtnet_softc *sc;
1517 ifp = sc->vtnet_ifp;
1520 // lwkt_serialize_enter(&sc->vtnet_slz);
1522 if ((ifp->if_flags & IFF_RUNNING) == 0) {
1523 vtnet_enable_rx_intr(sc);
1524 // lwkt_serialize_exit(&sc->vtnet_slz);
1528 more = vtnet_rxeof(sc, sc->vtnet_rx_process_limit, NULL);
1529 if (!more && vtnet_enable_rx_intr(sc) != 0) {
1530 vtnet_disable_rx_intr(sc);
1534 // lwkt_serialize_exit(&sc->vtnet_slz);
1537 sc->vtnet_stats.rx_task_rescheduled++;
1543 vtnet_rx_vq_intr(void *xsc)
1545 struct vtnet_softc *sc;
1549 vtnet_disable_rx_intr(sc);
1550 vtnet_rx_intr_task(sc);
1556 vtnet_enqueue_txhdr(struct vtnet_softc *sc, struct vtnet_tx_header *txhdr)
1558 bzero(txhdr, sizeof(*txhdr));
1559 SLIST_INSERT_HEAD(&sc->vtnet_txhdr_free, txhdr, link);
1563 vtnet_txeof(struct vtnet_softc *sc)
1565 struct virtqueue *vq;
1567 struct vtnet_tx_header *txhdr;
1570 vq = sc->vtnet_tx_vq;
1571 ifp = sc->vtnet_ifp;
1574 ASSERT_SERIALIZED(&sc->vtnet_slz);
1576 while ((txhdr = virtqueue_dequeue(vq, NULL)) != NULL) {
1579 m_freem(txhdr->vth_mbuf);
1580 vtnet_enqueue_txhdr(sc, txhdr);
1584 ifq_clr_oactive(&ifp->if_snd);
1585 if (virtqueue_empty(vq))
1586 sc->vtnet_watchdog_timer = 0;
1590 static struct mbuf *
1591 vtnet_tx_offload(struct vtnet_softc *sc, struct mbuf *m,
1592 struct virtio_net_hdr *hdr)
1595 struct ether_header *eh;
1596 struct ether_vlan_header *evh;
1598 struct ip6_hdr *ip6;
1601 uint16_t eth_type, csum_start;
1602 uint8_t ip_proto, gso_type;
1604 ifp = sc->vtnet_ifp;
1607 ip_offset = sizeof(struct ether_header);
1608 if (m->m_len < ip_offset) {
1609 if ((m = m_pullup(m, ip_offset)) == NULL)
1613 eh = mtod(m, struct ether_header *);
1614 eth_type = ntohs(eh->ether_type);
1615 if (eth_type == ETHERTYPE_VLAN) {
1616 ip_offset = sizeof(struct ether_vlan_header);
1617 if (m->m_len < ip_offset) {
1618 if ((m = m_pullup(m, ip_offset)) == NULL)
1621 evh = mtod(m, struct ether_vlan_header *);
1622 eth_type = ntohs(evh->evl_proto);
1627 if (m->m_len < ip_offset + sizeof(struct ip)) {
1628 m = m_pullup(m, ip_offset + sizeof(struct ip));
1633 ip = (struct ip *)(mtod(m, uint8_t *) + ip_offset);
1634 ip_proto = ip->ip_p;
1635 csum_start = ip_offset + (ip->ip_hl << 2);
1636 gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
1639 case ETHERTYPE_IPV6:
1640 if (m->m_len < ip_offset + sizeof(struct ip6_hdr)) {
1641 m = m_pullup(m, ip_offset + sizeof(struct ip6_hdr));
1646 ip6 = (struct ip6_hdr *)(mtod(m, uint8_t *) + ip_offset);
1648 * XXX Assume no extension headers are present. Presently,
1649 * this will always be true in the case of TSO, and FreeBSD
1650 * does not perform checksum offloading of IPv6 yet.
1652 ip_proto = ip6->ip6_nxt;
1653 csum_start = ip_offset + sizeof(struct ip6_hdr);
1654 gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
1661 if (m->m_pkthdr.csum_flags & VTNET_CSUM_OFFLOAD) {
1662 hdr->flags |= VIRTIO_NET_HDR_F_NEEDS_CSUM;
1663 hdr->csum_start = csum_start;
1664 hdr->csum_offset = m->m_pkthdr.csum_data;
1666 sc->vtnet_stats.tx_csum_offloaded++;
1669 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1670 if (ip_proto != IPPROTO_TCP)
1673 if (m->m_len < csum_start + sizeof(struct tcphdr)) {
1674 m = m_pullup(m, csum_start + sizeof(struct tcphdr));
1679 tcp = (struct tcphdr *)(mtod(m, uint8_t *) + csum_start);
1680 hdr->gso_type = gso_type;
1681 hdr->hdr_len = csum_start + (tcp->th_off << 2);
1682 hdr->gso_size = m->m_pkthdr.tso_segsz;
1684 if (tcp->th_flags & TH_CWR) {
1686 * Drop if we did not negotiate VIRTIO_NET_F_HOST_ECN.
1687 * ECN support is only configurable globally with the
1688 * net.inet.tcp.ecn.enable sysctl knob.
1690 if ((sc->vtnet_flags & VTNET_FLAG_TSO_ECN) == 0) {
1691 if_printf(ifp, "TSO with ECN not supported "
1697 hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
1700 sc->vtnet_stats.tx_tso_offloaded++;
1707 vtnet_enqueue_txbuf(struct vtnet_softc *sc, struct mbuf **m_head,
1708 struct vtnet_tx_header *txhdr)
1711 struct sglist_seg segs[VTNET_MAX_TX_SEGS];
1712 struct virtqueue *vq;
1716 vq = sc->vtnet_tx_vq;
1719 sglist_init(&sg, sc->vtnet_tx_nsegs, segs);
1720 error = sglist_append(&sg, &txhdr->vth_uhdr, sc->vtnet_hdr_size);
1721 KASSERT(error == 0 && sg.sg_nseg == 1,
1722 ("%s: error %d adding header to sglist", __func__, error));
1724 error = sglist_append_mbuf(&sg, m);
1726 m = m_defrag(m, M_NOWAIT);
1731 sc->vtnet_stats.tx_defragged++;
1733 error = sglist_append_mbuf(&sg, m);
1738 txhdr->vth_mbuf = m;
1739 error = virtqueue_enqueue(vq, txhdr, &sg, sg.sg_nseg, 0);
1744 sc->vtnet_stats.tx_defrag_failed++;
1751 static struct mbuf *
1752 vtnet_vlan_tag_insert(struct mbuf *m)
1755 struct ether_vlan_header *evl;
1757 if (M_WRITABLE(m) == 0) {
1758 n = m_dup(m, M_NOWAIT);
1760 if ((m = n) == NULL)
1764 M_PREPEND(m, ETHER_VLAN_ENCAP_LEN, M_NOWAIT);
1767 if (m->m_len < sizeof(struct ether_vlan_header)) {
1768 m = m_pullup(m, sizeof(struct ether_vlan_header));
1773 /* Insert 802.1Q header into the existing Ethernet header. */
1774 evl = mtod(m, struct ether_vlan_header *);
1775 bcopy((char *) evl + ETHER_VLAN_ENCAP_LEN,
1776 (char *) evl, ETHER_HDR_LEN - ETHER_TYPE_LEN);
1777 evl->evl_encap_proto = htons(ETHERTYPE_VLAN);
1778 evl->evl_tag = htons(m->m_pkthdr.ether_vlantag);
1779 m->m_flags &= ~M_VLANTAG;
1785 vtnet_encap(struct vtnet_softc *sc, struct mbuf **m_head)
1787 struct vtnet_tx_header *txhdr;
1788 struct virtio_net_hdr *hdr;
1792 txhdr = SLIST_FIRST(&sc->vtnet_txhdr_free);
1795 SLIST_REMOVE_HEAD(&sc->vtnet_txhdr_free, link);
1798 * Always use the non-mergeable header to simplify things. When
1799 * the mergeable feature is negotiated, the num_buffers field
1800 * must be set to zero. We use vtnet_hdr_size later to enqueue
1801 * the correct header size to the host.
1803 hdr = &txhdr->vth_uhdr.hdr;
1808 if (m->m_flags & M_VLANTAG) {
1809 //m = ether_vlanencap(m, m->m_pkthdr.ether_vtag);
1810 m = vtnet_vlan_tag_insert(m);
1811 if ((*m_head = m) == NULL)
1813 m->m_flags &= ~M_VLANTAG;
1816 if (m->m_pkthdr.csum_flags != 0) {
1817 m = vtnet_tx_offload(sc, m, hdr);
1818 if ((*m_head = m) == NULL)
1822 error = vtnet_enqueue_txbuf(sc, m_head, txhdr);
1825 vtnet_enqueue_txhdr(sc, txhdr);
1830 vtnet_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1832 struct vtnet_softc *sc;
1836 ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1837 lwkt_serialize_enter(&sc->vtnet_slz);
1838 vtnet_start_locked(ifp, ifsq);
1839 lwkt_serialize_exit(&sc->vtnet_slz);
1843 vtnet_start_locked(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1845 struct vtnet_softc *sc;
1846 struct virtqueue *vq;
1851 vq = sc->vtnet_tx_vq;
1854 ASSERT_SERIALIZED(&sc->vtnet_slz);
1856 if ((ifp->if_flags & (IFF_RUNNING)) !=
1857 IFF_RUNNING || ((sc->vtnet_flags & VTNET_FLAG_LINK) == 0))
1860 #ifdef VTNET_TX_INTR_MODERATION
1861 if (virtqueue_nused(vq) >= sc->vtnet_tx_size / 2)
1865 while (!ifsq_is_empty(ifsq)) {
1866 if (virtqueue_full(vq)) {
1867 ifq_set_oactive(&ifp->if_snd);
1871 m0 = ifq_dequeue(&ifp->if_snd);
1875 if (vtnet_encap(sc, &m0) != 0) {
1878 ifq_prepend(&ifp->if_snd, m0);
1879 ifq_set_oactive(&ifp->if_snd);
1884 ETHER_BPF_MTAP(ifp, m0);
1888 virtqueue_notify(vq, &sc->vtnet_slz);
1889 sc->vtnet_watchdog_timer = VTNET_WATCHDOG_TIMEOUT;
1894 vtnet_tick(void *xsc)
1896 struct vtnet_softc *sc;
1901 ASSERT_SERIALIZED(&sc->vtnet_slz);
1903 virtqueue_dump(sc->vtnet_rx_vq);
1904 virtqueue_dump(sc->vtnet_tx_vq);
1908 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
1913 vtnet_tx_intr_task(void *arg)
1915 struct vtnet_softc *sc;
1917 struct ifaltq_subque *ifsq;
1920 ifp = sc->vtnet_ifp;
1921 ifsq = ifq_get_subq_default(&ifp->if_snd);
1924 // lwkt_serialize_enter(&sc->vtnet_slz);
1926 if ((ifp->if_flags & IFF_RUNNING) == 0) {
1927 vtnet_enable_tx_intr(sc);
1928 // lwkt_serialize_exit(&sc->vtnet_slz);
1934 if (!ifsq_is_empty(ifsq))
1935 vtnet_start_locked(ifp, ifsq);
1937 if (vtnet_enable_tx_intr(sc) != 0) {
1938 vtnet_disable_tx_intr(sc);
1939 sc->vtnet_stats.tx_task_rescheduled++;
1940 // lwkt_serialize_exit(&sc->vtnet_slz);
1944 // lwkt_serialize_exit(&sc->vtnet_slz);
1948 vtnet_tx_vq_intr(void *xsc)
1950 struct vtnet_softc *sc;
1954 vtnet_disable_tx_intr(sc);
1955 vtnet_tx_intr_task(sc);
1961 vtnet_stop(struct vtnet_softc *sc)
1966 dev = sc->vtnet_dev;
1967 ifp = sc->vtnet_ifp;
1969 ASSERT_SERIALIZED(&sc->vtnet_slz);
1971 sc->vtnet_watchdog_timer = 0;
1972 callout_stop(&sc->vtnet_tick_ch);
1973 ifq_clr_oactive(&ifp->if_snd);
1974 ifp->if_flags &= ~(IFF_RUNNING);
1976 vtnet_disable_rx_intr(sc);
1977 vtnet_disable_tx_intr(sc);
1980 * Stop the host VirtIO adapter. Note this will reset the host
1981 * adapter's state back to the pre-initialized state, so in
1982 * order to make the device usable again, we must drive it
1983 * through virtio_reinit() and virtio_reinit_complete().
1987 sc->vtnet_flags &= ~VTNET_FLAG_LINK;
1989 vtnet_free_rx_mbufs(sc);
1990 vtnet_free_tx_mbufs(sc);
1994 vtnet_virtio_reinit(struct vtnet_softc *sc)
2001 dev = sc->vtnet_dev;
2002 ifp = sc->vtnet_ifp;
2003 features = sc->vtnet_features;
2006 * Re-negotiate with the host, removing any disabled receive
2007 * features. Transmit features are disabled only on our side
2008 * via if_capenable and if_hwassist.
2011 if (ifp->if_capabilities & IFCAP_RXCSUM) {
2012 if ((ifp->if_capenable & IFCAP_RXCSUM) == 0)
2013 features &= ~VIRTIO_NET_F_GUEST_CSUM;
2016 if (ifp->if_capabilities & IFCAP_LRO) {
2017 if ((ifp->if_capenable & IFCAP_LRO) == 0)
2018 features &= ~VTNET_LRO_FEATURES;
2021 if (ifp->if_capabilities & IFCAP_VLAN_HWFILTER) {
2022 if ((ifp->if_capenable & IFCAP_VLAN_HWFILTER) == 0)
2023 features &= ~VIRTIO_NET_F_CTRL_VLAN;
2026 error = virtio_reinit(dev, features);
2028 device_printf(dev, "virtio reinit error %d\n", error);
2034 vtnet_init_locked(struct vtnet_softc *sc)
2040 dev = sc->vtnet_dev;
2041 ifp = sc->vtnet_ifp;
2043 ASSERT_SERIALIZED(&sc->vtnet_slz);
2045 if (ifp->if_flags & IFF_RUNNING)
2048 /* Stop host's adapter, cancel any pending I/O. */
2051 /* Reinitialize the host device. */
2052 error = vtnet_virtio_reinit(sc);
2055 "reinitialization failed, stopping device...\n");
2060 /* Update host with assigned MAC address. */
2061 bcopy(IF_LLADDR(ifp), sc->vtnet_hwaddr, ETHER_ADDR_LEN);
2062 vtnet_set_hwaddr(sc);
2064 ifp->if_hwassist = 0;
2065 if (ifp->if_capenable & IFCAP_TXCSUM)
2066 ifp->if_hwassist |= VTNET_CSUM_OFFLOAD;
2067 if (ifp->if_capenable & IFCAP_TSO4)
2068 ifp->if_hwassist |= CSUM_TSO;
2070 error = vtnet_init_rx_vq(sc);
2073 "cannot allocate mbufs for Rx virtqueue\n");
2078 if (sc->vtnet_flags & VTNET_FLAG_CTRL_VQ) {
2079 if (sc->vtnet_flags & VTNET_FLAG_CTRL_RX) {
2080 /* Restore promiscuous and all-multicast modes. */
2081 vtnet_rx_filter(sc);
2083 /* Restore filtered MAC addresses. */
2084 vtnet_rx_filter_mac(sc);
2087 /* Restore VLAN filters. */
2088 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER)
2089 vtnet_rx_filter_vlan(sc);
2093 vtnet_enable_rx_intr(sc);
2094 vtnet_enable_tx_intr(sc);
2097 ifp->if_flags |= IFF_RUNNING;
2098 ifq_clr_oactive(&ifp->if_snd);
2100 virtio_reinit_complete(dev);
2102 vtnet_update_link_status(sc);
2103 callout_reset(&sc->vtnet_tick_ch, hz, vtnet_tick, sc);
2107 vtnet_init(void *xsc)
2109 struct vtnet_softc *sc;
2113 lwkt_serialize_enter(&sc->vtnet_slz);
2114 vtnet_init_locked(sc);
2115 lwkt_serialize_exit(&sc->vtnet_slz);
2119 vtnet_exec_ctrl_cmd(struct vtnet_softc *sc, void *cookie,
2120 struct sglist *sg, int readable, int writable)
2122 struct virtqueue *vq;
2125 vq = sc->vtnet_ctrl_vq;
2127 ASSERT_SERIALIZED(&sc->vtnet_slz);
2128 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_VQ,
2129 ("no control virtqueue"));
2130 KASSERT(virtqueue_empty(vq),
2131 ("control command already enqueued"));
2133 if (virtqueue_enqueue(vq, cookie, sg, readable, writable) != 0)
2136 virtqueue_notify(vq, &sc->vtnet_slz);
2139 * Poll until the command is complete. Previously, we would
2140 * sleep until the control virtqueue interrupt handler woke
2141 * us up, but dropping the VTNET_MTX leads to serialization
2144 * Furthermore, it appears QEMU/KVM only allocates three MSIX
2145 * vectors. Two of those vectors are needed for the Rx and Tx
2146 * virtqueues. We do not support sharing both a Vq and config
2147 * changed notification on the same MSIX vector.
2149 c = virtqueue_poll(vq, NULL);
2150 KASSERT(c == cookie, ("unexpected control command response"));
2154 vtnet_ctrl_mac_cmd(struct vtnet_softc *sc, uint8_t *hwaddr)
2157 struct virtio_net_ctrl_hdr hdr __aligned(2);
2159 char aligned_hwaddr[ETHER_ADDR_LEN] __aligned(8);
2163 struct sglist_seg segs[3];
2167 s.hdr.class = VIRTIO_NET_CTRL_MAC;
2168 s.hdr.cmd = VIRTIO_NET_CTRL_MAC_ADDR_SET;
2169 s.ack = VIRTIO_NET_ERR;
2171 /* Copy the mac address into physically contiguous memory */
2172 memcpy(s.aligned_hwaddr, hwaddr, ETHER_ADDR_LEN);
2174 sglist_init(&sg, 3, segs);
2176 error |= sglist_append(&sg, &s.hdr,
2177 sizeof(struct virtio_net_ctrl_hdr));
2178 error |= sglist_append(&sg, s.aligned_hwaddr, ETHER_ADDR_LEN);
2179 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
2180 KASSERT(error == 0 && sg.sg_nseg == 3,
2181 ("%s: error %d adding set MAC msg to sglist", __func__, error));
2183 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
2185 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
2189 vtnet_rx_filter(struct vtnet_softc *sc)
2194 dev = sc->vtnet_dev;
2195 ifp = sc->vtnet_ifp;
2197 ASSERT_SERIALIZED(&sc->vtnet_slz);
2198 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2199 ("CTRL_RX feature not negotiated"));
2201 if (vtnet_set_promisc(sc, ifp->if_flags & IFF_PROMISC) != 0)
2202 device_printf(dev, "cannot %s promiscuous mode\n",
2203 (ifp->if_flags & IFF_PROMISC) ? "enable" : "disable");
2205 if (vtnet_set_allmulti(sc, ifp->if_flags & IFF_ALLMULTI) != 0)
2206 device_printf(dev, "cannot %s all-multicast mode\n",
2207 (ifp->if_flags & IFF_ALLMULTI) ? "enable" : "disable");
2211 vtnet_ctrl_rx_cmd(struct vtnet_softc *sc, int cmd, int on)
2213 struct sglist_seg segs[3];
2216 struct virtio_net_ctrl_hdr hdr __aligned(2);
2224 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2225 ("%s: CTRL_RX feature not negotiated", __func__));
2227 s.hdr.class = VIRTIO_NET_CTRL_RX;
2230 s.ack = VIRTIO_NET_ERR;
2232 sglist_init(&sg, 3, segs);
2234 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
2235 error |= sglist_append(&sg, &s.onoff, sizeof(uint8_t));
2236 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
2237 KASSERT(error == 0 && sg.sg_nseg == 3,
2238 ("%s: error %d adding Rx message to sglist", __func__, error));
2240 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
2242 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
2246 vtnet_set_promisc(struct vtnet_softc *sc, int on)
2249 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_PROMISC, on));
2253 vtnet_set_allmulti(struct vtnet_softc *sc, int on)
2256 return (vtnet_ctrl_rx_cmd(sc, VIRTIO_NET_CTRL_RX_ALLMULTI, on));
2260 vtnet_rx_filter_mac(struct vtnet_softc *sc)
2262 struct virtio_net_ctrl_hdr hdr __aligned(2);
2263 struct vtnet_mac_filter *filter;
2264 struct sglist_seg segs[4];
2268 struct ifaddr_container *ifac;
2269 struct ifmultiaddr *ifma;
2270 int ucnt, mcnt, promisc, allmulti, error;
2273 ifp = sc->vtnet_ifp;
2279 ASSERT_SERIALIZED(&sc->vtnet_slz);
2280 KASSERT(sc->vtnet_flags & VTNET_FLAG_CTRL_RX,
2281 ("%s: CTRL_RX feature not negotiated", __func__));
2283 /* Use the MAC filtering table allocated in vtnet_attach. */
2284 filter = sc->vtnet_macfilter;
2285 memset(filter, 0, sizeof(struct vtnet_mac_filter));
2287 /* Unicast MAC addresses: */
2288 //if_addr_rlock(ifp);
2289 TAILQ_FOREACH(ifac, &ifp->if_addrheads[mycpuid], ifa_link) {
2291 if (ifa->ifa_addr->sa_family != AF_LINK)
2293 else if (memcmp(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2294 sc->vtnet_hwaddr, ETHER_ADDR_LEN) == 0)
2296 else if (ucnt == VTNET_MAX_MAC_ENTRIES) {
2301 bcopy(LLADDR((struct sockaddr_dl *)ifa->ifa_addr),
2302 &filter->vmf_unicast.macs[ucnt], ETHER_ADDR_LEN);
2305 //if_addr_runlock(ifp);
2308 filter->vmf_unicast.nentries = 0;
2309 if_printf(ifp, "more than %d MAC addresses assigned, "
2310 "falling back to promiscuous mode\n",
2311 VTNET_MAX_MAC_ENTRIES);
2313 filter->vmf_unicast.nentries = ucnt;
2315 /* Multicast MAC addresses: */
2316 //if_maddr_rlock(ifp);
2317 TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2318 if (ifma->ifma_addr->sa_family != AF_LINK)
2320 else if (mcnt == VTNET_MAX_MAC_ENTRIES) {
2325 bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
2326 &filter->vmf_multicast.macs[mcnt], ETHER_ADDR_LEN);
2329 //if_maddr_runlock(ifp);
2331 if (allmulti != 0) {
2332 filter->vmf_multicast.nentries = 0;
2333 if_printf(ifp, "more than %d multicast MAC addresses "
2334 "assigned, falling back to all-multicast mode\n",
2335 VTNET_MAX_MAC_ENTRIES);
2337 filter->vmf_multicast.nentries = mcnt;
2339 if (promisc != 0 && allmulti != 0)
2342 hdr.class = VIRTIO_NET_CTRL_MAC;
2343 hdr.cmd = VIRTIO_NET_CTRL_MAC_TABLE_SET;
2344 ack = VIRTIO_NET_ERR;
2346 sglist_init(&sg, 4, segs);
2348 error |= sglist_append(&sg, &hdr, sizeof(struct virtio_net_ctrl_hdr));
2349 error |= sglist_append(&sg, &filter->vmf_unicast,
2350 sizeof(uint32_t) + filter->vmf_unicast.nentries * ETHER_ADDR_LEN);
2351 error |= sglist_append(&sg, &filter->vmf_multicast,
2352 sizeof(uint32_t) + filter->vmf_multicast.nentries * ETHER_ADDR_LEN);
2353 error |= sglist_append(&sg, &ack, sizeof(uint8_t));
2354 KASSERT(error == 0 && sg.sg_nseg == 4,
2355 ("%s: error %d adding MAC filter msg to sglist", __func__, error));
2357 vtnet_exec_ctrl_cmd(sc, &ack, &sg, sg.sg_nseg - 1, 1);
2359 if (ack != VIRTIO_NET_OK)
2360 if_printf(ifp, "error setting host MAC filter table\n");
2363 if (promisc != 0 && vtnet_set_promisc(sc, 1) != 0)
2364 if_printf(ifp, "cannot enable promiscuous mode\n");
2365 if (allmulti != 0 && vtnet_set_allmulti(sc, 1) != 0)
2366 if_printf(ifp, "cannot enable all-multicast mode\n");
2370 vtnet_exec_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2372 struct sglist_seg segs[3];
2375 struct virtio_net_ctrl_hdr hdr __aligned(2);
2383 s.hdr.class = VIRTIO_NET_CTRL_VLAN;
2384 s.hdr.cmd = add ? VIRTIO_NET_CTRL_VLAN_ADD : VIRTIO_NET_CTRL_VLAN_DEL;
2386 s.ack = VIRTIO_NET_ERR;
2388 sglist_init(&sg, 3, segs);
2390 error |= sglist_append(&sg, &s.hdr, sizeof(struct virtio_net_ctrl_hdr));
2391 error |= sglist_append(&sg, &s.tag, sizeof(uint16_t));
2392 error |= sglist_append(&sg, &s.ack, sizeof(uint8_t));
2393 KASSERT(error == 0 && sg.sg_nseg == 3,
2394 ("%s: error %d adding VLAN message to sglist", __func__, error));
2396 vtnet_exec_ctrl_cmd(sc, &s.ack, &sg, sg.sg_nseg - 1, 1);
2398 return (s.ack == VIRTIO_NET_OK ? 0 : EIO);
2402 vtnet_rx_filter_vlan(struct vtnet_softc *sc)
2408 ASSERT_SERIALIZED(&sc->vtnet_slz);
2409 KASSERT(sc->vtnet_flags & VTNET_FLAG_VLAN_FILTER,
2410 ("%s: VLAN_FILTER feature not negotiated", __func__));
2412 nvlans = sc->vtnet_nvlans;
2414 /* Enable the filter for each configured VLAN. */
2415 for (i = 0; i < VTNET_VLAN_SHADOW_SIZE && nvlans > 0; i++) {
2416 w = sc->vtnet_vlan_shadow[i];
2417 while ((bit = ffs(w) - 1) != -1) {
2419 tag = sizeof(w) * CHAR_BIT * i + bit;
2422 if (vtnet_exec_vlan_filter(sc, 1, tag) != 0) {
2423 device_printf(sc->vtnet_dev,
2424 "cannot enable VLAN %d filter\n", tag);
2429 KASSERT(nvlans == 0, ("VLAN count incorrect"));
2433 vtnet_update_vlan_filter(struct vtnet_softc *sc, int add, uint16_t tag)
2438 ifp = sc->vtnet_ifp;
2439 idx = (tag >> 5) & 0x7F;
2442 if (tag == 0 || tag > 4095)
2445 lwkt_serialize_enter(&sc->vtnet_slz);
2447 /* Update shadow VLAN table. */
2450 sc->vtnet_vlan_shadow[idx] |= (1 << bit);
2453 sc->vtnet_vlan_shadow[idx] &= ~(1 << bit);
2456 if (ifp->if_capenable & IFCAP_VLAN_HWFILTER &&
2457 vtnet_exec_vlan_filter(sc, add, tag) != 0) {
2458 device_printf(sc->vtnet_dev,
2459 "cannot %s VLAN %d %s the host filter table\n",
2460 add ? "add" : "remove", tag, add ? "to" : "from");
2463 lwkt_serialize_exit(&sc->vtnet_slz);
2467 vtnet_register_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2470 if (ifp->if_softc != arg)
2473 vtnet_update_vlan_filter(arg, 1, tag);
2477 vtnet_unregister_vlan(void *arg, struct ifnet *ifp, uint16_t tag)
2480 if (ifp->if_softc != arg)
2483 vtnet_update_vlan_filter(arg, 0, tag);
2487 vtnet_ifmedia_upd(struct ifnet *ifp)
2489 struct vtnet_softc *sc;
2490 struct ifmedia *ifm;
2493 ifm = &sc->vtnet_media;
2495 if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
2502 vtnet_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2504 struct vtnet_softc *sc;
2508 ifmr->ifm_status = IFM_AVALID;
2509 ifmr->ifm_active = IFM_ETHER;
2511 lwkt_serialize_enter(&sc->vtnet_slz);
2512 if (vtnet_is_link_up(sc) != 0) {
2513 ifmr->ifm_status |= IFM_ACTIVE;
2514 ifmr->ifm_active |= VTNET_MEDIATYPE;
2516 ifmr->ifm_active |= IFM_NONE;
2517 lwkt_serialize_exit(&sc->vtnet_slz);
2521 vtnet_add_statistics(struct vtnet_softc *sc)
2524 struct vtnet_statistics *stats;
2525 struct sysctl_ctx_list *ctx;
2526 struct sysctl_oid *tree;
2527 struct sysctl_oid_list *child;
2529 dev = sc->vtnet_dev;
2530 stats = &sc->vtnet_stats;
2531 ctx = device_get_sysctl_ctx(dev);
2532 tree = device_get_sysctl_tree(dev);
2533 child = SYSCTL_CHILDREN(tree);
2535 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "mbuf_alloc_failed",
2536 CTLFLAG_RD, &stats->mbuf_alloc_failed, 0,
2537 "Mbuf cluster allocation failures");
2539 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_frame_too_large",
2540 CTLFLAG_RD, &stats->rx_frame_too_large, 0,
2541 "Received frame larger than the mbuf chain");
2542 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_enq_replacement_failed",
2543 CTLFLAG_RD, &stats->rx_enq_replacement_failed, 0,
2544 "Enqueuing the replacement receive mbuf failed");
2545 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_mergeable_failed",
2546 CTLFLAG_RD, &stats->rx_mergeable_failed, 0,
2547 "Mergeable buffers receive failures");
2548 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ethtype",
2549 CTLFLAG_RD, &stats->rx_csum_bad_ethtype, 0,
2550 "Received checksum offloaded buffer with unsupported "
2552 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_ipproto",
2553 CTLFLAG_RD, &stats->rx_csum_bad_ipproto, 0,
2554 "Received checksum offloaded buffer with incorrect IP protocol");
2555 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_bad_offset",
2556 CTLFLAG_RD, &stats->rx_csum_bad_offset, 0,
2557 "Received checksum offloaded buffer with incorrect offset");
2558 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_failed",
2559 CTLFLAG_RD, &stats->rx_csum_failed, 0,
2560 "Received buffer checksum offload failed");
2561 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_csum_offloaded",
2562 CTLFLAG_RD, &stats->rx_csum_offloaded, 0,
2563 "Received buffer checksum offload succeeded");
2564 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "rx_task_rescheduled",
2565 CTLFLAG_RD, &stats->rx_task_rescheduled, 0,
2566 "Times the receive interrupt task rescheduled itself");
2568 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_bad_ethtype",
2569 CTLFLAG_RD, &stats->tx_csum_bad_ethtype, 0,
2570 "Aborted transmit of checksum offloaded buffer with unknown "
2572 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_bad_ethtype",
2573 CTLFLAG_RD, &stats->tx_tso_bad_ethtype, 0,
2574 "Aborted transmit of TSO buffer with unknown Ethernet type");
2575 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defragged",
2576 CTLFLAG_RD, &stats->tx_defragged, 0,
2577 "Transmit mbufs defragged");
2578 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_defrag_failed",
2579 CTLFLAG_RD, &stats->tx_defrag_failed, 0,
2580 "Aborted transmit of buffer because defrag failed");
2581 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_csum_offloaded",
2582 CTLFLAG_RD, &stats->tx_csum_offloaded, 0,
2583 "Offloaded checksum of transmitted buffer");
2584 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_tso_offloaded",
2585 CTLFLAG_RD, &stats->tx_tso_offloaded, 0,
2586 "Segmentation offload of transmitted buffer");
2587 SYSCTL_ADD_UQUAD(ctx, child, OID_AUTO, "tx_task_rescheduled",
2588 CTLFLAG_RD, &stats->tx_task_rescheduled, 0,
2589 "Times the transmit interrupt task rescheduled itself");
2593 vtnet_enable_rx_intr(struct vtnet_softc *sc)
2596 return (virtqueue_enable_intr(sc->vtnet_rx_vq));
2600 vtnet_disable_rx_intr(struct vtnet_softc *sc)
2603 virtqueue_disable_intr(sc->vtnet_rx_vq);
2607 vtnet_enable_tx_intr(struct vtnet_softc *sc)
2610 #ifdef VTNET_TX_INTR_MODERATION
2613 return (virtqueue_enable_intr(sc->vtnet_tx_vq));
2618 vtnet_disable_tx_intr(struct vtnet_softc *sc)
2621 virtqueue_disable_intr(sc->vtnet_tx_vq);