2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2011 NetApp, Inc.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #ifndef WITHOUT_CAPSICUM
36 #include <sys/capsicum.h>
38 #include <sys/linker_set.h>
39 #include <sys/select.h>
41 #include <sys/ioctl.h>
42 #include <machine/atomic.h>
43 #include <net/ethernet.h>
44 #ifndef NETMAP_WITH_LIBS
45 #define NETMAP_WITH_LIBS
47 #include <net/netmap_user.h>
49 #ifndef WITHOUT_CAPSICUM
50 #include <capsicum_helpers.h>
64 #include <pthread_np.h>
72 #define VTNET_RINGSZ 1024
74 #define VTNET_MAXSEGS 256
77 * Host capabilities. Note that we only offer a few of these.
79 #define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */
80 #define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */
81 #define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */
82 #define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */
83 #define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */
84 #define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */
85 #define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */
86 #define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */
87 #define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */
88 #define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */
89 #define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */
90 #define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */
91 #define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */
92 #define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */
93 #define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */
94 #define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */
95 #define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */
96 #define VIRTIO_NET_F_GUEST_ANNOUNCE \
97 (1 << 21) /* guest can send gratuitous pkts */
99 #define VTNET_S_HOSTCAPS \
100 ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_STATUS | \
101 VIRTIO_F_NOTIFY_ON_EMPTY | VIRTIO_RING_F_INDIRECT_DESC)
104 * PCI config-space "registers"
106 struct virtio_net_config {
116 #define VTNET_CTLQ 2 /* NB: not yet supported */
121 * Fixed network header size
123 struct virtio_net_rxhdr {
125 uint8_t vrh_gso_type;
126 uint16_t vrh_hdr_len;
127 uint16_t vrh_gso_size;
128 uint16_t vrh_csum_start;
129 uint16_t vrh_csum_offset;
136 static int pci_vtnet_debug;
137 #define DPRINTF(params) if (pci_vtnet_debug) printf params
138 #define WPRINTF(params) printf params
143 struct pci_vtnet_softc {
144 struct virtio_softc vsc_vs;
145 struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
146 pthread_mutex_t vsc_mtx;
147 struct mevent *vsc_mevp;
150 struct nm_desc *vsc_nmd;
153 int resetting; /* protected by tx_mtx */
155 uint64_t vsc_features; /* negotiated features */
157 struct virtio_net_config vsc_config;
159 pthread_mutex_t rx_mtx;
161 int rx_merge; /* merged rx bufs in use */
164 pthread_mutex_t tx_mtx;
165 pthread_cond_t tx_cond;
168 void (*pci_vtnet_rx)(struct pci_vtnet_softc *sc);
169 void (*pci_vtnet_tx)(struct pci_vtnet_softc *sc, struct iovec *iov,
170 int iovcnt, int len);
173 static void pci_vtnet_reset(void *);
174 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
175 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
176 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
177 static void pci_vtnet_neg_features(void *, uint64_t);
179 static struct virtio_consts vtnet_vi_consts = {
180 "vtnet", /* our name */
181 VTNET_MAXQ - 1, /* we currently support 2 virtqueues */
182 sizeof(struct virtio_net_config), /* config reg size */
183 pci_vtnet_reset, /* reset */
184 NULL, /* device-wide qnotify -- not used */
185 pci_vtnet_cfgread, /* read PCI config */
186 pci_vtnet_cfgwrite, /* write PCI config */
187 pci_vtnet_neg_features, /* apply negotiated features */
188 VTNET_S_HOSTCAPS, /* our capabilities */
192 pci_vtnet_reset(void *vsc)
194 struct pci_vtnet_softc *sc = vsc;
196 DPRINTF(("vtnet: device reset requested !\n"));
198 /* Acquire the RX lock to block RX processing. */
199 pthread_mutex_lock(&sc->rx_mtx);
201 /* Set sc->resetting and give a chance to the TX thread to stop. */
202 pthread_mutex_lock(&sc->tx_mtx);
204 while (sc->tx_in_progress) {
205 pthread_mutex_unlock(&sc->tx_mtx);
207 pthread_mutex_lock(&sc->tx_mtx);
210 sc->vsc_rx_ready = 0;
212 sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
215 * Now reset rings, MSI-X vectors, and negotiated capabilities.
216 * Do that with the TX lock held, since we need to reset
219 vi_reset_dev(&sc->vsc_vs);
222 pthread_mutex_unlock(&sc->tx_mtx);
223 pthread_mutex_unlock(&sc->rx_mtx);
227 * Called to send a buffer chain out to the tap device
230 pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
233 static char pad[60]; /* all zero bytes */
235 if (sc->vsc_tapfd == -1)
239 * If the length is < 60, pad out to that and add the
240 * extra zero'd segment to the iov. It is guaranteed that
241 * there is always an extra iov available by the caller.
244 iov[iovcnt].iov_base = pad;
245 iov[iovcnt].iov_len = 60 - len;
248 (void) writev(sc->vsc_tapfd, iov, iovcnt);
252 * Called when there is read activity on the tap file descriptor.
253 * Each buffer posted by the guest is assumed to be able to contain
254 * an entire ethernet frame + rx header.
255 * MP note: the dummybuf is only used for discarding frames, so there
256 * is no need for it to be per-vtnet or locked.
258 static uint8_t dummybuf[2048];
260 static __inline struct iovec *
261 rx_iov_trim(struct iovec *iov, int *niov, int tlen)
265 /* XXX short-cut: assume first segment is >= tlen */
266 assert(iov[0].iov_len >= tlen);
268 iov[0].iov_len -= tlen;
269 if (iov[0].iov_len == 0) {
274 iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + tlen);
282 pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
284 struct iovec iov[VTNET_MAXSEGS], *riov;
285 struct vqueue_info *vq;
291 * Should never be called without a valid tap fd
293 assert(sc->vsc_tapfd != -1);
296 * But, will be called when the rx ring hasn't yet
299 if (!sc->vsc_rx_ready) {
301 * Drop the packet and try later.
303 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
308 * Check for available rx buffers
310 vq = &sc->vsc_queues[VTNET_RXQ];
311 if (!vq_has_descs(vq)) {
313 * Drop the packet and try later. Interrupt on
314 * empty, if that's negotiated.
316 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
323 * Get descriptor chain.
325 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
326 assert(n >= 1 && n <= VTNET_MAXSEGS);
329 * Get a pointer to the rx header, and use the
330 * data immediately following it for the packet buffer.
332 vrx = iov[0].iov_base;
333 riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
335 len = readv(sc->vsc_tapfd, riov, n);
337 if (len < 0 && errno == EWOULDBLOCK) {
339 * No more packets, but still some avail ring
340 * entries. Interrupt if needed/appropriate.
348 * The only valid field in the rx packet header is the
349 * number of buffers if merged rx bufs were negotiated.
351 memset(vrx, 0, sc->rx_vhdrlen);
354 struct virtio_net_rxhdr *vrxh;
361 * Release this chain and handle more chains.
363 vq_relchain(vq, idx, len + sc->rx_vhdrlen);
364 } while (vq_has_descs(vq));
366 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
371 pci_vtnet_netmap_writev(struct nm_desc *nmd, struct iovec *iov, int iovcnt)
376 for (r = nmd->cur_tx_ring; ; ) {
377 struct netmap_ring *ring = NETMAP_TXRING(nmd->nifp, r);
381 if (nm_ring_empty(ring)) {
383 if (r > nmd->last_tx_ring)
384 r = nmd->first_tx_ring;
385 if (r == nmd->cur_tx_ring)
390 idx = ring->slot[cur].buf_idx;
391 buf = NETMAP_BUF(ring, idx);
393 for (i = 0; i < iovcnt; i++) {
394 if (len + iov[i].iov_len > 2048)
396 memcpy(&buf[len], iov[i].iov_base, iov[i].iov_len);
397 len += iov[i].iov_len;
399 ring->slot[cur].len = len;
400 ring->head = ring->cur = nm_ring_next(ring, cur);
401 nmd->cur_tx_ring = r;
402 ioctl(nmd->fd, NIOCTXSYNC, NULL);
410 pci_vtnet_netmap_readv(struct nm_desc *nmd, struct iovec *iov, int iovcnt)
416 for (r = nmd->cur_rx_ring; ; ) {
417 struct netmap_ring *ring = NETMAP_RXRING(nmd->nifp, r);
422 if (nm_ring_empty(ring)) {
424 if (r > nmd->last_rx_ring)
425 r = nmd->first_rx_ring;
426 if (r == nmd->cur_rx_ring)
431 idx = ring->slot[cur].buf_idx;
432 buf = NETMAP_BUF(ring, idx);
433 left = ring->slot[cur].len;
435 for (i = 0; i < iovcnt && left > 0; i++) {
436 if (iov[i].iov_len > left)
437 iov[i].iov_len = left;
438 memcpy(iov[i].iov_base, &buf[len], iov[i].iov_len);
439 len += iov[i].iov_len;
440 left -= iov[i].iov_len;
442 ring->head = ring->cur = nm_ring_next(ring, cur);
443 nmd->cur_rx_ring = r;
444 ioctl(nmd->fd, NIOCRXSYNC, NULL);
447 for (; i < iovcnt; i++)
454 * Called to send a buffer chain out to the vale port
457 pci_vtnet_netmap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
460 static char pad[60]; /* all zero bytes */
462 if (sc->vsc_nmd == NULL)
466 * If the length is < 60, pad out to that and add the
467 * extra zero'd segment to the iov. It is guaranteed that
468 * there is always an extra iov available by the caller.
471 iov[iovcnt].iov_base = pad;
472 iov[iovcnt].iov_len = 60 - len;
475 (void) pci_vtnet_netmap_writev(sc->vsc_nmd, iov, iovcnt);
479 pci_vtnet_netmap_rx(struct pci_vtnet_softc *sc)
481 struct iovec iov[VTNET_MAXSEGS], *riov;
482 struct vqueue_info *vq;
488 * Should never be called without a valid netmap descriptor
490 assert(sc->vsc_nmd != NULL);
493 * But, will be called when the rx ring hasn't yet
496 if (!sc->vsc_rx_ready) {
498 * Drop the packet and try later.
500 (void) nm_nextpkt(sc->vsc_nmd, (void *)dummybuf);
505 * Check for available rx buffers
507 vq = &sc->vsc_queues[VTNET_RXQ];
508 if (!vq_has_descs(vq)) {
510 * Drop the packet and try later. Interrupt on
511 * empty, if that's negotiated.
513 (void) nm_nextpkt(sc->vsc_nmd, (void *)dummybuf);
520 * Get descriptor chain.
522 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
523 assert(n >= 1 && n <= VTNET_MAXSEGS);
526 * Get a pointer to the rx header, and use the
527 * data immediately following it for the packet buffer.
529 vrx = iov[0].iov_base;
530 riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
532 len = pci_vtnet_netmap_readv(sc->vsc_nmd, riov, n);
536 * No more packets, but still some avail ring
537 * entries. Interrupt if needed/appropriate.
545 * The only valid field in the rx packet header is the
546 * number of buffers if merged rx bufs were negotiated.
548 memset(vrx, 0, sc->rx_vhdrlen);
551 struct virtio_net_rxhdr *vrxh;
558 * Release this chain and handle more chains.
560 vq_relchain(vq, idx, len + sc->rx_vhdrlen);
561 } while (vq_has_descs(vq));
563 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
568 pci_vtnet_rx_callback(int fd, enum ev_type type, void *param)
570 struct pci_vtnet_softc *sc = param;
572 pthread_mutex_lock(&sc->rx_mtx);
573 sc->pci_vtnet_rx(sc);
574 pthread_mutex_unlock(&sc->rx_mtx);
579 pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
581 struct pci_vtnet_softc *sc = vsc;
584 * A qnotify means that the rx process can now begin
586 if (sc->vsc_rx_ready == 0) {
587 sc->vsc_rx_ready = 1;
588 vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
593 pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
595 struct iovec iov[VTNET_MAXSEGS + 1];
601 * Obtain chain of descriptors. The first one is
602 * really the header descriptor, so we need to sum
603 * up two lengths: packet length and transfer length.
605 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
606 assert(n >= 1 && n <= VTNET_MAXSEGS);
608 tlen = iov[0].iov_len;
609 for (i = 1; i < n; i++) {
610 plen += iov[i].iov_len;
611 tlen += iov[i].iov_len;
614 DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
615 sc->pci_vtnet_tx(sc, &iov[1], n - 1, plen);
617 /* chain is processed, release it and set tlen */
618 vq_relchain(vq, idx, tlen);
622 pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
624 struct pci_vtnet_softc *sc = vsc;
627 * Any ring entries to process?
629 if (!vq_has_descs(vq))
632 /* Signal the tx thread for processing */
633 pthread_mutex_lock(&sc->tx_mtx);
634 vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
635 if (sc->tx_in_progress == 0)
636 pthread_cond_signal(&sc->tx_cond);
637 pthread_mutex_unlock(&sc->tx_mtx);
641 * Thread which will handle processing of TX desc
644 pci_vtnet_tx_thread(void *param)
646 struct pci_vtnet_softc *sc = param;
647 struct vqueue_info *vq;
650 vq = &sc->vsc_queues[VTNET_TXQ];
653 * Let us wait till the tx queue pointers get initialised &
656 pthread_mutex_lock(&sc->tx_mtx);
657 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
661 /* note - tx mutex is locked here */
662 while (sc->resetting || !vq_has_descs(vq)) {
663 vq->vq_used->vu_flags &= ~VRING_USED_F_NO_NOTIFY;
665 if (!sc->resetting && vq_has_descs(vq))
668 sc->tx_in_progress = 0;
669 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
672 vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
673 sc->tx_in_progress = 1;
674 pthread_mutex_unlock(&sc->tx_mtx);
678 * Run through entries, placing them into
679 * iovecs and sending when an end-of-packet
682 pci_vtnet_proctx(sc, vq);
683 } while (vq_has_descs(vq));
686 * Generate an interrupt if needed.
690 pthread_mutex_lock(&sc->tx_mtx);
696 pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
699 DPRINTF(("vtnet: control qnotify!\n\r"));
704 pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
706 struct ether_addr *ea;
708 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
710 tmpstr = strsep(&mac_str,"=");
712 if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
713 ea = ether_aton(mac_str);
715 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
716 memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
717 fprintf(stderr, "Invalid MAC %s\n", mac_str);
720 memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
727 pci_vtnet_tap_setup(struct pci_vtnet_softc *sc, char *devname)
730 #ifndef WITHOUT_CAPSICUM
734 strcpy(tbuf, "/dev/");
735 strlcat(tbuf, devname, sizeof(tbuf));
737 sc->pci_vtnet_rx = pci_vtnet_tap_rx;
738 sc->pci_vtnet_tx = pci_vtnet_tap_tx;
740 sc->vsc_tapfd = open(tbuf, O_RDWR);
741 if (sc->vsc_tapfd == -1) {
742 WPRINTF(("open of tap device %s failed\n", tbuf));
747 * Set non-blocking and register for read
748 * notifications with the event loop
751 if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
752 WPRINTF(("tap device O_NONBLOCK failed\n"));
753 close(sc->vsc_tapfd);
757 #ifndef WITHOUT_CAPSICUM
758 cap_rights_init(&rights, CAP_EVENT, CAP_READ, CAP_WRITE);
759 if (caph_rights_limit(sc->vsc_tapfd, &rights) == -1)
760 errx(EX_OSERR, "Unable to apply rights for sandbox");
763 sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
765 pci_vtnet_rx_callback,
767 if (sc->vsc_mevp == NULL) {
768 WPRINTF(("Could not register event\n"));
769 close(sc->vsc_tapfd);
775 pci_vtnet_netmap_setup(struct pci_vtnet_softc *sc, char *ifname)
777 sc->pci_vtnet_rx = pci_vtnet_netmap_rx;
778 sc->pci_vtnet_tx = pci_vtnet_netmap_tx;
780 sc->vsc_nmd = nm_open(ifname, NULL, 0, 0);
781 if (sc->vsc_nmd == NULL) {
782 WPRINTF(("open of netmap device %s failed\n", ifname));
786 sc->vsc_mevp = mevent_add(sc->vsc_nmd->fd,
788 pci_vtnet_rx_callback,
790 if (sc->vsc_mevp == NULL) {
791 WPRINTF(("Could not register event\n"));
792 nm_close(sc->vsc_nmd);
798 pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
801 unsigned char digest[16];
803 char tname[MAXCOMLEN + 1];
804 struct pci_vtnet_softc *sc;
809 sc = calloc(1, sizeof(struct pci_vtnet_softc));
811 pthread_mutex_init(&sc->vsc_mtx, NULL);
813 vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
814 sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
816 sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
817 sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
818 sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
819 sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
821 sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
822 sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
826 * Attempt to open the tap device and read the MAC address
835 devname = vtopts = strdup(opts);
836 (void) strsep(&vtopts, ",");
838 if (vtopts != NULL) {
839 err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
847 if (strncmp(devname, "vale", 4) == 0)
848 pci_vtnet_netmap_setup(sc, devname);
849 if (strncmp(devname, "tap", 3) == 0 ||
850 strncmp(devname, "vmnet", 5) == 0)
851 pci_vtnet_tap_setup(sc, devname);
857 * The default MAC address is the standard NetApp OUI of 00-a0-98,
858 * followed by an MD5 of the PCI slot/func number and dev name
861 snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
862 pi->pi_func, vmname);
865 MD5Update(&mdctx, nstr, strlen(nstr));
866 MD5Final(digest, &mdctx);
868 sc->vsc_config.mac[0] = 0x00;
869 sc->vsc_config.mac[1] = 0xa0;
870 sc->vsc_config.mac[2] = 0x98;
871 sc->vsc_config.mac[3] = digest[0];
872 sc->vsc_config.mac[4] = digest[1];
873 sc->vsc_config.mac[5] = digest[2];
876 /* initialize config space */
877 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
878 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
879 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
880 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
881 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
883 /* Link is up if we managed to open tap device or vale port. */
884 sc->vsc_config.status = (opts == NULL || sc->vsc_tapfd >= 0 ||
885 sc->vsc_nmd != NULL);
887 /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
888 if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix()))
891 /* use BAR 0 to map config regs in IO space */
892 vi_set_io_bar(&sc->vsc_vs, 0);
897 sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
898 pthread_mutex_init(&sc->rx_mtx, NULL);
901 * Initialize tx semaphore & spawn TX processing thread.
902 * As of now, only one thread for TX desc processing is
905 sc->tx_in_progress = 0;
906 pthread_mutex_init(&sc->tx_mtx, NULL);
907 pthread_cond_init(&sc->tx_cond, NULL);
908 pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
909 snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
911 pthread_set_name_np(sc->tx_tid, tname);
917 pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
919 struct pci_vtnet_softc *sc = vsc;
923 assert(offset + size <= 6);
925 * The driver is allowed to change the MAC address
927 ptr = &sc->vsc_config.mac[offset];
928 memcpy(ptr, &value, size);
930 /* silently ignore other writes */
931 DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
938 pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
940 struct pci_vtnet_softc *sc = vsc;
943 ptr = (uint8_t *)&sc->vsc_config + offset;
944 memcpy(retval, ptr, size);
949 pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features)
951 struct pci_vtnet_softc *sc = vsc;
953 sc->vsc_features = negotiated_features;
955 if (!(sc->vsc_features & VIRTIO_NET_F_MRG_RXBUF)) {
957 /* non-merge rx header is 2 bytes shorter */
962 struct pci_devemu pci_de_vnet = {
963 .pe_emu = "virtio-net",
964 .pe_init = pci_vtnet_init,
965 .pe_barwrite = vi_pci_write,
966 .pe_barread = vi_pci_read
968 PCI_EMUL_SET(pci_de_vnet);