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 <net/ethernet.h>
50 #include <pthread_np.h>
57 #define VTNET_RINGSZ 1024
59 #define VTNET_MAXSEGS 32
62 * Host capabilities. Note that we only offer a few of these.
64 #define VIRTIO_NET_F_CSUM (1 << 0) /* host handles partial cksum */
65 #define VIRTIO_NET_F_GUEST_CSUM (1 << 1) /* guest handles partial cksum */
66 #define VIRTIO_NET_F_MAC (1 << 5) /* host supplies MAC */
67 #define VIRTIO_NET_F_GSO_DEPREC (1 << 6) /* deprecated: host handles GSO */
68 #define VIRTIO_NET_F_GUEST_TSO4 (1 << 7) /* guest can rcv TSOv4 */
69 #define VIRTIO_NET_F_GUEST_TSO6 (1 << 8) /* guest can rcv TSOv6 */
70 #define VIRTIO_NET_F_GUEST_ECN (1 << 9) /* guest can rcv TSO with ECN */
71 #define VIRTIO_NET_F_GUEST_UFO (1 << 10) /* guest can rcv UFO */
72 #define VIRTIO_NET_F_HOST_TSO4 (1 << 11) /* host can rcv TSOv4 */
73 #define VIRTIO_NET_F_HOST_TSO6 (1 << 12) /* host can rcv TSOv6 */
74 #define VIRTIO_NET_F_HOST_ECN (1 << 13) /* host can rcv TSO with ECN */
75 #define VIRTIO_NET_F_HOST_UFO (1 << 14) /* host can rcv UFO */
76 #define VIRTIO_NET_F_MRG_RXBUF (1 << 15) /* host can merge RX buffers */
77 #define VIRTIO_NET_F_STATUS (1 << 16) /* config status field available */
78 #define VIRTIO_NET_F_CTRL_VQ (1 << 17) /* control channel available */
79 #define VIRTIO_NET_F_CTRL_RX (1 << 18) /* control channel RX mode support */
80 #define VIRTIO_NET_F_CTRL_VLAN (1 << 19) /* control channel VLAN filtering */
81 #define VIRTIO_NET_F_GUEST_ANNOUNCE \
82 (1 << 21) /* guest can send gratuitous pkts */
84 #define VTNET_S_HOSTCAPS \
85 ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_STATUS | \
86 VIRTIO_F_NOTIFY_ON_EMPTY)
89 * PCI config-space "registers"
91 struct virtio_net_config {
101 #define VTNET_CTLQ 2 /* NB: not yet supported */
106 * Fixed network header size
108 struct virtio_net_rxhdr {
110 uint8_t vrh_gso_type;
111 uint16_t vrh_hdr_len;
112 uint16_t vrh_gso_size;
113 uint16_t vrh_csum_start;
114 uint16_t vrh_csum_offset;
121 static int pci_vtnet_debug;
122 #define DPRINTF(params) if (pci_vtnet_debug) printf params
123 #define WPRINTF(params) printf params
128 struct pci_vtnet_softc {
129 struct virtio_softc vsc_vs;
130 struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
131 pthread_mutex_t vsc_mtx;
132 struct mevent *vsc_mevp;
136 volatile int resetting; /* set and checked outside lock */
138 uint64_t vsc_features; /* negotiated features */
140 struct virtio_net_config vsc_config;
142 pthread_mutex_t rx_mtx;
145 int rx_merge; /* merged rx bufs in use */
148 pthread_mutex_t tx_mtx;
149 pthread_cond_t tx_cond;
153 static void pci_vtnet_reset(void *);
154 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
155 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
156 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
157 static void pci_vtnet_neg_features(void *, uint64_t);
159 static struct virtio_consts vtnet_vi_consts = {
160 "vtnet", /* our name */
161 VTNET_MAXQ - 1, /* we currently support 2 virtqueues */
162 sizeof(struct virtio_net_config), /* config reg size */
163 pci_vtnet_reset, /* reset */
164 NULL, /* device-wide qnotify -- not used */
165 pci_vtnet_cfgread, /* read PCI config */
166 pci_vtnet_cfgwrite, /* write PCI config */
167 pci_vtnet_neg_features, /* apply negotiated features */
168 VTNET_S_HOSTCAPS, /* our capabilities */
172 * If the transmit thread is active then stall until it is done.
175 pci_vtnet_txwait(struct pci_vtnet_softc *sc)
178 pthread_mutex_lock(&sc->tx_mtx);
179 while (sc->tx_in_progress) {
180 pthread_mutex_unlock(&sc->tx_mtx);
182 pthread_mutex_lock(&sc->tx_mtx);
184 pthread_mutex_unlock(&sc->tx_mtx);
188 * If the receive thread is active then stall until it is done.
191 pci_vtnet_rxwait(struct pci_vtnet_softc *sc)
194 pthread_mutex_lock(&sc->rx_mtx);
195 while (sc->rx_in_progress) {
196 pthread_mutex_unlock(&sc->rx_mtx);
198 pthread_mutex_lock(&sc->rx_mtx);
200 pthread_mutex_unlock(&sc->rx_mtx);
204 pci_vtnet_reset(void *vsc)
206 struct pci_vtnet_softc *sc = vsc;
208 DPRINTF(("vtnet: device reset requested !\n"));
213 * Wait for the transmit and receive threads to finish their
216 pci_vtnet_txwait(sc);
217 pci_vtnet_rxwait(sc);
219 sc->vsc_rx_ready = 0;
221 sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
223 /* now reset rings, MSI-X vectors, and negotiated capabilities */
224 vi_reset_dev(&sc->vsc_vs);
230 * Called to send a buffer chain out to the tap device
233 pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
236 static char pad[60]; /* all zero bytes */
238 if (sc->vsc_tapfd == -1)
242 * If the length is < 60, pad out to that and add the
243 * extra zero'd segment to the iov. It is guaranteed that
244 * there is always an extra iov available by the caller.
247 iov[iovcnt].iov_base = pad;
248 iov[iovcnt].iov_len = 60 - len;
251 (void) writev(sc->vsc_tapfd, iov, iovcnt);
255 * Called when there is read activity on the tap file descriptor.
256 * Each buffer posted by the guest is assumed to be able to contain
257 * an entire ethernet frame + rx header.
258 * MP note: the dummybuf is only used for discarding frames, so there
259 * is no need for it to be per-vtnet or locked.
261 static uint8_t dummybuf[2048];
263 static __inline struct iovec *
264 rx_iov_trim(struct iovec *iov, int *niov, int tlen)
268 /* XXX short-cut: assume first segment is >= tlen */
269 assert(iov[0].iov_len >= tlen);
271 iov[0].iov_len -= tlen;
272 if (iov[0].iov_len == 0) {
277 iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + tlen);
285 pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
287 struct iovec iov[VTNET_MAXSEGS], *riov;
288 struct vqueue_info *vq;
294 * Should never be called without a valid tap fd
296 assert(sc->vsc_tapfd != -1);
299 * But, will be called when the rx ring hasn't yet
300 * been set up or the guest is resetting the device.
302 if (!sc->vsc_rx_ready || sc->resetting) {
304 * Drop the packet and try later.
306 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
311 * Check for available rx buffers
313 vq = &sc->vsc_queues[VTNET_RXQ];
314 if (!vq_has_descs(vq)) {
316 * Drop the packet and try later. Interrupt on
317 * empty, if that's negotiated.
319 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
326 * Get descriptor chain.
328 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
329 assert(n >= 1 && n <= VTNET_MAXSEGS);
332 * Get a pointer to the rx header, and use the
333 * data immediately following it for the packet buffer.
335 vrx = iov[0].iov_base;
336 riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
338 len = readv(sc->vsc_tapfd, riov, n);
340 if (len < 0 && errno == EWOULDBLOCK) {
342 * No more packets, but still some avail ring
343 * entries. Interrupt if needed/appropriate.
351 * The only valid field in the rx packet header is the
352 * number of buffers if merged rx bufs were negotiated.
354 memset(vrx, 0, sc->rx_vhdrlen);
357 struct virtio_net_rxhdr *vrxh;
364 * Release this chain and handle more chains.
366 vq_relchain(vq, idx, len + sc->rx_vhdrlen);
367 } while (vq_has_descs(vq));
369 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
374 pci_vtnet_tap_callback(int fd, enum ev_type type, void *param)
376 struct pci_vtnet_softc *sc = param;
378 pthread_mutex_lock(&sc->rx_mtx);
379 sc->rx_in_progress = 1;
380 pci_vtnet_tap_rx(sc);
381 sc->rx_in_progress = 0;
382 pthread_mutex_unlock(&sc->rx_mtx);
387 pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
389 struct pci_vtnet_softc *sc = vsc;
392 * A qnotify means that the rx process can now begin
394 if (sc->vsc_rx_ready == 0) {
395 sc->vsc_rx_ready = 1;
400 pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
402 struct iovec iov[VTNET_MAXSEGS + 1];
408 * Obtain chain of descriptors. The first one is
409 * really the header descriptor, so we need to sum
410 * up two lengths: packet length and transfer length.
412 n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
413 assert(n >= 1 && n <= VTNET_MAXSEGS);
415 tlen = iov[0].iov_len;
416 for (i = 1; i < n; i++) {
417 plen += iov[i].iov_len;
418 tlen += iov[i].iov_len;
421 DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
422 pci_vtnet_tap_tx(sc, &iov[1], n - 1, plen);
424 /* chain is processed, release it and set tlen */
425 vq_relchain(vq, idx, tlen);
429 pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
431 struct pci_vtnet_softc *sc = vsc;
434 * Any ring entries to process?
436 if (!vq_has_descs(vq))
439 /* Signal the tx thread for processing */
440 pthread_mutex_lock(&sc->tx_mtx);
441 if (sc->tx_in_progress == 0)
442 pthread_cond_signal(&sc->tx_cond);
443 pthread_mutex_unlock(&sc->tx_mtx);
447 * Thread which will handle processing of TX desc
450 pci_vtnet_tx_thread(void *param)
452 struct pci_vtnet_softc *sc = param;
453 struct vqueue_info *vq;
454 int have_work, error;
456 vq = &sc->vsc_queues[VTNET_TXQ];
459 * Let us wait till the tx queue pointers get initialised &
462 pthread_mutex_lock(&sc->tx_mtx);
463 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
467 /* note - tx mutex is locked here */
472 have_work = vq_has_descs(vq);
475 sc->tx_in_progress = 0;
476 error = pthread_cond_wait(&sc->tx_cond,
480 } while (!have_work);
481 sc->tx_in_progress = 1;
482 pthread_mutex_unlock(&sc->tx_mtx);
486 * Run through entries, placing them into
487 * iovecs and sending when an end-of-packet
490 pci_vtnet_proctx(sc, vq);
491 } while (vq_has_descs(vq));
494 * Generate an interrupt if needed.
498 pthread_mutex_lock(&sc->tx_mtx);
504 pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
507 DPRINTF(("vtnet: control qnotify!\n\r"));
512 pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
514 struct ether_addr *ea;
516 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
518 tmpstr = strsep(&mac_str,"=");
520 if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
521 ea = ether_aton(mac_str);
523 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
524 memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
525 fprintf(stderr, "Invalid MAC %s\n", mac_str);
528 memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
536 pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
539 unsigned char digest[16];
541 char tname[MAXCOMLEN + 1];
542 struct pci_vtnet_softc *sc;
547 sc = calloc(1, sizeof(struct pci_vtnet_softc));
549 pthread_mutex_init(&sc->vsc_mtx, NULL);
551 vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
552 sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
554 sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
555 sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
556 sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
557 sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
559 sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
560 sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
564 * Attempt to open the tap device and read the MAC address
573 devname = vtopts = strdup(opts);
574 (void) strsep(&vtopts, ",");
576 if (vtopts != NULL) {
577 err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
585 strcpy(tbuf, "/dev/");
586 strlcat(tbuf, devname, sizeof(tbuf));
590 sc->vsc_tapfd = open(tbuf, O_RDWR);
591 if (sc->vsc_tapfd == -1) {
592 WPRINTF(("open of tap device %s failed\n", tbuf));
595 * Set non-blocking and register for read
596 * notifications with the event loop
599 if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
600 WPRINTF(("tap device O_NONBLOCK failed\n"));
601 close(sc->vsc_tapfd);
605 sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
607 pci_vtnet_tap_callback,
609 if (sc->vsc_mevp == NULL) {
610 WPRINTF(("Could not register event\n"));
611 close(sc->vsc_tapfd);
618 * The default MAC address is the standard NetApp OUI of 00-a0-98,
619 * followed by an MD5 of the PCI slot/func number and dev name
622 snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
623 pi->pi_func, vmname);
626 MD5Update(&mdctx, nstr, strlen(nstr));
627 MD5Final(digest, &mdctx);
629 sc->vsc_config.mac[0] = 0x00;
630 sc->vsc_config.mac[1] = 0xa0;
631 sc->vsc_config.mac[2] = 0x98;
632 sc->vsc_config.mac[3] = digest[0];
633 sc->vsc_config.mac[4] = digest[1];
634 sc->vsc_config.mac[5] = digest[2];
637 /* initialize config space */
638 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
639 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
640 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
641 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
643 pci_lintr_request(pi);
646 sc->vsc_config.status = 1;
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);