2 * Copyright (c) 2011 NetApp, Inc.
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, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/select.h>
36 #include <sys/ioctl.h>
37 #include <machine/atomic.h>
38 #include <net/ethernet.h>
51 #include <pthread_np.h>
58 #define VTNET_RINGSZ 1024
60 #define VTNET_MAXSEGS 32
63 * Host capabilities. Note that we only offer a few of these.
65 #define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */
66 #define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */
67 #define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */
68 #define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */
69 #define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */
70 #define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */
71 #define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */
72 #define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */
73 #define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */
74 #define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */
75 #define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */
76 #define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */
77 #define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */
78 #define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */
79 #define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */
80 #define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */
81 #define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */
82 #define VIRTIO_NET_F_GUEST_ANNOUNCE \
83 (1 << 21) /* guest can send gratuitous pkts */
85 #define VTNET_S_HOSTCAPS \
86 ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_STATUS | \
87 VIRTIO_F_NOTIFY_ON_EMPTY)
90 * PCI config-space "registers"
92 struct virtio_net_config {
102 #define VTNET_CTLQ 2 /* NB: not yet supported */
107 * Fixed network header size
109 struct virtio_net_rxhdr {
111 uint8_t vrh_gso_type;
112 uint16_t vrh_hdr_len;
113 uint16_t vrh_gso_size;
114 uint16_t vrh_csum_start;
115 uint16_t vrh_csum_offset;
122 static int pci_vtnet_debug;
123 #define DPRINTF(params) if (pci_vtnet_debug) printf params
124 #define WPRINTF(params) printf params
129 struct pci_vtnet_softc {
130 struct virtio_softc vsc_vs;
131 struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
132 pthread_mutex_t vsc_mtx;
133 struct mevent *vsc_mevp;
137 volatile int resetting; /* set and checked outside lock */
139 uint64_t vsc_features; /* negotiated features */
141 struct virtio_net_config vsc_config;
143 pthread_mutex_t rx_mtx;
146 int rx_merge; /* merged rx bufs in use */
149 pthread_mutex_t tx_mtx;
150 pthread_cond_t tx_cond;
154 static void pci_vtnet_reset(void *);
155 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
156 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
157 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
158 static void pci_vtnet_neg_features(void *, uint64_t);
160 static struct virtio_consts vtnet_vi_consts = {
161 "vtnet", /* our name */
162 VTNET_MAXQ - 1, /* we currently support 2 virtqueues */
163 sizeof(struct virtio_net_config), /* config reg size */
164 pci_vtnet_reset, /* reset */
165 NULL, /* device-wide qnotify -- not used */
166 pci_vtnet_cfgread, /* read PCI config */
167 pci_vtnet_cfgwrite, /* write PCI config */
168 pci_vtnet_neg_features, /* apply negotiated features */
169 VTNET_S_HOSTCAPS, /* our capabilities */
173 * If the transmit thread is active then stall until it is done.
176 pci_vtnet_txwait(struct pci_vtnet_softc *sc)
179 pthread_mutex_lock(&sc->tx_mtx);
180 while (sc->tx_in_progress) {
181 pthread_mutex_unlock(&sc->tx_mtx);
183 pthread_mutex_lock(&sc->tx_mtx);
185 pthread_mutex_unlock(&sc->tx_mtx);
189 * If the receive thread is active then stall until it is done.
192 pci_vtnet_rxwait(struct pci_vtnet_softc *sc)
195 pthread_mutex_lock(&sc->rx_mtx);
196 while (sc->rx_in_progress) {
197 pthread_mutex_unlock(&sc->rx_mtx);
199 pthread_mutex_lock(&sc->rx_mtx);
201 pthread_mutex_unlock(&sc->rx_mtx);
205 pci_vtnet_reset(void *vsc)
207 struct pci_vtnet_softc *sc = vsc;
209 DPRINTF(("vtnet: device reset requested !\n"));
214 * Wait for the transmit and receive threads to finish their
217 pci_vtnet_txwait(sc);
218 pci_vtnet_rxwait(sc);
220 sc->vsc_rx_ready = 0;
222 sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
224 /* now reset rings, MSI-X vectors, and negotiated capabilities */
225 vi_reset_dev(&sc->vsc_vs);
231 * Called to send a buffer chain out to the tap device
234 pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
237 static char pad[60]; /* all zero bytes */
239 if (sc->vsc_tapfd == -1)
243 * If the length is < 60, pad out to that and add the
244 * extra zero'd segment to the iov. It is guaranteed that
245 * there is always an extra iov available by the caller.
248 iov[iovcnt].iov_base = pad;
249 iov[iovcnt].iov_len = 60 - len;
252 (void) writev(sc->vsc_tapfd, iov, iovcnt);
256 * Called when there is read activity on the tap file descriptor.
257 * Each buffer posted by the guest is assumed to be able to contain
258 * an entire ethernet frame + rx header.
259 * MP note: the dummybuf is only used for discarding frames, so there
260 * is no need for it to be per-vtnet or locked.
262 static uint8_t dummybuf[2048];
264 static __inline struct iovec *
265 rx_iov_trim(struct iovec *iov, int *niov, int tlen)
269 /* XXX short-cut: assume first segment is >= tlen */
270 assert(iov[0].iov_len >= tlen);
272 iov[0].iov_len -= tlen;
273 if (iov[0].iov_len == 0) {
278 iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + tlen);
286 pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
288 struct iovec iov[VTNET_MAXSEGS], *riov;
289 struct vqueue_info *vq;
295 * Should never be called without a valid tap fd
297 assert(sc->vsc_tapfd != -1);
300 * But, will be called when the rx ring hasn't yet
301 * been set up or the guest is resetting the device.
303 if (!sc->vsc_rx_ready || sc->resetting) {
305 * Drop the packet and try later.
307 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
312 * Check for available rx buffers
314 vq = &sc->vsc_queues[VTNET_RXQ];
315 if (!vq_has_descs(vq)) {
317 * Drop the packet and try later. Interrupt on
318 * empty, if that's negotiated.
320 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
327 * Get descriptor chain.
329 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
330 assert(n >= 1 && n <= VTNET_MAXSEGS);
333 * Get a pointer to the rx header, and use the
334 * data immediately following it for the packet buffer.
336 vrx = iov[0].iov_base;
337 riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
339 len = readv(sc->vsc_tapfd, riov, n);
341 if (len < 0 && errno == EWOULDBLOCK) {
343 * No more packets, but still some avail ring
344 * entries. Interrupt if needed/appropriate.
352 * The only valid field in the rx packet header is the
353 * number of buffers if merged rx bufs were negotiated.
355 memset(vrx, 0, sc->rx_vhdrlen);
358 struct virtio_net_rxhdr *vrxh;
365 * Release this chain and handle more chains.
367 vq_relchain(vq, idx, len + sc->rx_vhdrlen);
368 } while (vq_has_descs(vq));
370 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
375 pci_vtnet_tap_callback(int fd, enum ev_type type, void *param)
377 struct pci_vtnet_softc *sc = param;
379 pthread_mutex_lock(&sc->rx_mtx);
380 sc->rx_in_progress = 1;
381 pci_vtnet_tap_rx(sc);
382 sc->rx_in_progress = 0;
383 pthread_mutex_unlock(&sc->rx_mtx);
388 pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
390 struct pci_vtnet_softc *sc = vsc;
393 * A qnotify means that the rx process can now begin
395 if (sc->vsc_rx_ready == 0) {
396 sc->vsc_rx_ready = 1;
397 vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
402 pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
404 struct iovec iov[VTNET_MAXSEGS + 1];
410 * Obtain chain of descriptors. The first one is
411 * really the header descriptor, so we need to sum
412 * up two lengths: packet length and transfer length.
414 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
415 assert(n >= 1 && n <= VTNET_MAXSEGS);
417 tlen = iov[0].iov_len;
418 for (i = 1; i < n; i++) {
419 plen += iov[i].iov_len;
420 tlen += iov[i].iov_len;
423 DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
424 pci_vtnet_tap_tx(sc, &iov[1], n - 1, plen);
426 /* chain is processed, release it and set tlen */
427 vq_relchain(vq, idx, tlen);
431 pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
433 struct pci_vtnet_softc *sc = vsc;
436 * Any ring entries to process?
438 if (!vq_has_descs(vq))
441 /* Signal the tx thread for processing */
442 pthread_mutex_lock(&sc->tx_mtx);
443 vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
444 if (sc->tx_in_progress == 0)
445 pthread_cond_signal(&sc->tx_cond);
446 pthread_mutex_unlock(&sc->tx_mtx);
450 * Thread which will handle processing of TX desc
453 pci_vtnet_tx_thread(void *param)
455 struct pci_vtnet_softc *sc = param;
456 struct vqueue_info *vq;
459 vq = &sc->vsc_queues[VTNET_TXQ];
462 * Let us wait till the tx queue pointers get initialised &
465 pthread_mutex_lock(&sc->tx_mtx);
466 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
470 /* note - tx mutex is locked here */
471 while (sc->resetting || !vq_has_descs(vq)) {
472 vq->vq_used->vu_flags &= ~VRING_USED_F_NO_NOTIFY;
474 if (!sc->resetting && vq_has_descs(vq))
477 sc->tx_in_progress = 0;
478 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
481 vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
482 sc->tx_in_progress = 1;
483 pthread_mutex_unlock(&sc->tx_mtx);
487 * Run through entries, placing them into
488 * iovecs and sending when an end-of-packet
491 pci_vtnet_proctx(sc, vq);
492 } while (vq_has_descs(vq));
495 * Generate an interrupt if needed.
499 pthread_mutex_lock(&sc->tx_mtx);
505 pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
508 DPRINTF(("vtnet: control qnotify!\n\r"));
513 pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
515 struct ether_addr *ea;
517 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
519 tmpstr = strsep(&mac_str,"=");
521 if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
522 ea = ether_aton(mac_str);
524 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
525 memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
526 fprintf(stderr, "Invalid MAC %s\n", mac_str);
529 memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
537 pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
540 unsigned char digest[16];
542 char tname[MAXCOMLEN + 1];
543 struct pci_vtnet_softc *sc;
548 sc = calloc(1, sizeof(struct pci_vtnet_softc));
550 pthread_mutex_init(&sc->vsc_mtx, NULL);
552 vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
553 sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
555 sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
556 sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
557 sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
558 sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
560 sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
561 sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
565 * Attempt to open the tap device and read the MAC address
574 devname = vtopts = strdup(opts);
575 (void) strsep(&vtopts, ",");
577 if (vtopts != NULL) {
578 err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
586 strcpy(tbuf, "/dev/");
587 strlcat(tbuf, devname, sizeof(tbuf));
591 sc->vsc_tapfd = open(tbuf, O_RDWR);
592 if (sc->vsc_tapfd == -1) {
593 WPRINTF(("open of tap device %s failed\n", tbuf));
596 * Set non-blocking and register for read
597 * notifications with the event loop
600 if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
601 WPRINTF(("tap device O_NONBLOCK failed\n"));
602 close(sc->vsc_tapfd);
606 sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
608 pci_vtnet_tap_callback,
610 if (sc->vsc_mevp == NULL) {
611 WPRINTF(("Could not register event\n"));
612 close(sc->vsc_tapfd);
619 * The default MAC address is the standard NetApp OUI of 00-a0-98,
620 * followed by an MD5 of the PCI slot/func number and dev name
623 snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
624 pi->pi_func, vmname);
627 MD5Update(&mdctx, nstr, strlen(nstr));
628 MD5Final(digest, &mdctx);
630 sc->vsc_config.mac[0] = 0x00;
631 sc->vsc_config.mac[1] = 0xa0;
632 sc->vsc_config.mac[2] = 0x98;
633 sc->vsc_config.mac[3] = digest[0];
634 sc->vsc_config.mac[4] = digest[1];
635 sc->vsc_config.mac[5] = digest[2];
638 /* initialize config space */
639 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
640 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
641 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
642 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
643 pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
645 /* Link is up if we managed to open tap device. */
646 sc->vsc_config.status = (opts == NULL || sc->vsc_tapfd >= 0);
648 /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
649 if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix()))
652 /* use BAR 0 to map config regs in IO space */
653 vi_set_io_bar(&sc->vsc_vs, 0);
658 sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
659 sc->rx_in_progress = 0;
660 pthread_mutex_init(&sc->rx_mtx, NULL);
663 * Initialize tx semaphore & spawn TX processing thread.
664 * As of now, only one thread for TX desc processing is
667 sc->tx_in_progress = 0;
668 pthread_mutex_init(&sc->tx_mtx, NULL);
669 pthread_cond_init(&sc->tx_cond, NULL);
670 pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
671 snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
673 pthread_set_name_np(sc->tx_tid, tname);
679 pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
681 struct pci_vtnet_softc *sc = vsc;
685 assert(offset + size <= 6);
687 * The driver is allowed to change the MAC address
689 ptr = &sc->vsc_config.mac[offset];
690 memcpy(ptr, &value, size);
692 /* silently ignore other writes */
693 DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
700 pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
702 struct pci_vtnet_softc *sc = vsc;
705 ptr = (uint8_t *)&sc->vsc_config + offset;
706 memcpy(retval, ptr, size);
711 pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features)
713 struct pci_vtnet_softc *sc = vsc;
715 sc->vsc_features = negotiated_features;
717 if (!(sc->vsc_features & VIRTIO_NET_F_MRG_RXBUF)) {
719 /* non-merge rx header is 2 bytes shorter */
724 struct pci_devemu pci_de_vnet = {
725 .pe_emu = "virtio-net",
726 .pe_init = pci_vtnet_init,
727 .pe_barwrite = vi_pci_write,
728 .pe_barread = vi_pci_read
730 PCI_EMUL_SET(pci_de_vnet);