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 uint32_t vsc_features;
139 struct virtio_net_config vsc_config;
141 pthread_mutex_t rx_mtx;
145 pthread_mutex_t tx_mtx;
146 pthread_cond_t tx_cond;
150 static void pci_vtnet_reset(void *);
151 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
152 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
153 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
155 static struct virtio_consts vtnet_vi_consts = {
156 "vtnet", /* our name */
157 VTNET_MAXQ - 1, /* we currently support 2 virtqueues */
158 sizeof(struct virtio_net_config), /* config reg size */
159 pci_vtnet_reset, /* reset */
160 NULL, /* device-wide qnotify -- not used */
161 pci_vtnet_cfgread, /* read PCI config */
162 pci_vtnet_cfgwrite, /* write PCI config */
163 VTNET_S_HOSTCAPS, /* our capabilities */
167 * If the transmit thread is active then stall until it is done.
170 pci_vtnet_txwait(struct pci_vtnet_softc *sc)
173 pthread_mutex_lock(&sc->tx_mtx);
174 while (sc->tx_in_progress) {
175 pthread_mutex_unlock(&sc->tx_mtx);
177 pthread_mutex_lock(&sc->tx_mtx);
179 pthread_mutex_unlock(&sc->tx_mtx);
183 * If the receive thread is active then stall until it is done.
186 pci_vtnet_rxwait(struct pci_vtnet_softc *sc)
189 pthread_mutex_lock(&sc->rx_mtx);
190 while (sc->rx_in_progress) {
191 pthread_mutex_unlock(&sc->rx_mtx);
193 pthread_mutex_lock(&sc->rx_mtx);
195 pthread_mutex_unlock(&sc->rx_mtx);
199 pci_vtnet_reset(void *vsc)
201 struct pci_vtnet_softc *sc = vsc;
203 DPRINTF(("vtnet: device reset requested !\n"));
208 * Wait for the transmit and receive threads to finish their
211 pci_vtnet_txwait(sc);
212 pci_vtnet_rxwait(sc);
214 sc->vsc_rx_ready = 0;
216 /* now reset rings, MSI-X vectors, and negotiated capabilities */
217 vi_reset_dev(&sc->vsc_vs);
223 * Called to send a buffer chain out to the tap device
226 pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
229 static char pad[60]; /* all zero bytes */
231 if (sc->vsc_tapfd == -1)
235 * If the length is < 60, pad out to that and add the
236 * extra zero'd segment to the iov. It is guaranteed that
237 * there is always an extra iov available by the caller.
240 iov[iovcnt].iov_base = pad;
241 iov[iovcnt].iov_len = 60 - len;
244 (void) writev(sc->vsc_tapfd, iov, iovcnt);
248 * Called when there is read activity on the tap file descriptor.
249 * Each buffer posted by the guest is assumed to be able to contain
250 * an entire ethernet frame + rx header.
251 * MP note: the dummybuf is only used for discarding frames, so there
252 * is no need for it to be per-vtnet or locked.
254 static uint8_t dummybuf[2048];
257 pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
259 struct vqueue_info *vq;
260 struct virtio_net_rxhdr *vrx;
266 * Should never be called without a valid tap fd
268 assert(sc->vsc_tapfd != -1);
271 * But, will be called when the rx ring hasn't yet
272 * been set up or the guest is resetting the device.
274 if (!sc->vsc_rx_ready || sc->resetting) {
276 * Drop the packet and try later.
278 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
283 * Check for available rx buffers
285 vq = &sc->vsc_queues[VTNET_RXQ];
287 if (!vq_has_descs(vq)) {
289 * Drop the packet and try later. Interrupt on
290 * empty, if that's negotiated.
292 (void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
299 * Get descriptor chain, which should have just
300 * one descriptor in it.
301 * ??? allow guests to use multiple descs?
303 assert(vq_getchain(vq, &iov, 1, NULL) == 1);
306 * Get a pointer to the rx header, and use the
307 * data immediately following it for the packet buffer.
310 buf = (uint8_t *)(vrx + 1);
312 len = read(sc->vsc_tapfd, buf,
313 iov.iov_len - sizeof(struct virtio_net_rxhdr));
315 if (len < 0 && errno == EWOULDBLOCK) {
317 * No more packets, but still some avail ring
318 * entries. Interrupt if needed/appropriate.
325 * The only valid field in the rx packet header is the
326 * number of buffers, which is always 1 without TSO
329 memset(vrx, 0, sizeof(struct virtio_net_rxhdr));
333 * Release this chain and handle more chains.
335 vq_relchain(vq, len + sizeof(struct virtio_net_rxhdr));
336 } while (vq_has_descs(vq));
338 /* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
343 pci_vtnet_tap_callback(int fd, enum ev_type type, void *param)
345 struct pci_vtnet_softc *sc = param;
347 pthread_mutex_lock(&sc->rx_mtx);
348 sc->rx_in_progress = 1;
349 pci_vtnet_tap_rx(sc);
350 sc->rx_in_progress = 0;
351 pthread_mutex_unlock(&sc->rx_mtx);
356 pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
358 struct pci_vtnet_softc *sc = vsc;
361 * A qnotify means that the rx process can now begin
363 if (sc->vsc_rx_ready == 0) {
364 sc->vsc_rx_ready = 1;
369 pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
371 struct iovec iov[VTNET_MAXSEGS + 1];
376 * Obtain chain of descriptors. The first one is
377 * really the header descriptor, so we need to sum
378 * up two lengths: packet length and transfer length.
380 n = vq_getchain(vq, iov, VTNET_MAXSEGS, NULL);
381 assert(n >= 1 && n <= VTNET_MAXSEGS);
383 tlen = iov[0].iov_len;
384 for (i = 1; i < n; i++) {
385 plen += iov[i].iov_len;
386 tlen += iov[i].iov_len;
389 DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
390 pci_vtnet_tap_tx(sc, &iov[1], n - 1, plen);
392 /* chain is processed, release it and set tlen */
393 vq_relchain(vq, tlen);
397 pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
399 struct pci_vtnet_softc *sc = vsc;
402 * Any ring entries to process?
404 if (!vq_has_descs(vq))
407 /* Signal the tx thread for processing */
408 pthread_mutex_lock(&sc->tx_mtx);
409 if (sc->tx_in_progress == 0)
410 pthread_cond_signal(&sc->tx_cond);
411 pthread_mutex_unlock(&sc->tx_mtx);
415 * Thread which will handle processing of TX desc
418 pci_vtnet_tx_thread(void *param)
420 struct pci_vtnet_softc *sc = param;
421 struct vqueue_info *vq;
422 int have_work, error;
424 vq = &sc->vsc_queues[VTNET_TXQ];
427 * Let us wait till the tx queue pointers get initialised &
430 pthread_mutex_lock(&sc->tx_mtx);
431 error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
435 /* note - tx mutex is locked here */
440 have_work = vq_has_descs(vq);
443 sc->tx_in_progress = 0;
444 error = pthread_cond_wait(&sc->tx_cond,
448 } while (!have_work);
449 sc->tx_in_progress = 1;
450 pthread_mutex_unlock(&sc->tx_mtx);
455 * Run through entries, placing them into
456 * iovecs and sending when an end-of-packet
459 pci_vtnet_proctx(sc, vq);
460 } while (vq_has_descs(vq));
463 * Generate an interrupt if needed.
467 pthread_mutex_lock(&sc->tx_mtx);
473 pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
476 DPRINTF(("vtnet: control qnotify!\n\r"));
481 pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
483 struct ether_addr *ea;
485 char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
487 tmpstr = strsep(&mac_str,"=");
489 if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
490 ea = ether_aton(mac_str);
492 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
493 memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
494 fprintf(stderr, "Invalid MAC %s\n", mac_str);
497 memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
505 pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
508 unsigned char digest[16];
510 char tname[MAXCOMLEN + 1];
511 struct pci_vtnet_softc *sc;
516 sc = malloc(sizeof(struct pci_vtnet_softc));
517 memset(sc, 0, sizeof(struct pci_vtnet_softc));
519 pthread_mutex_init(&sc->vsc_mtx, NULL);
521 vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
522 sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
524 sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
525 sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
526 sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
527 sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
529 sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
530 sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
534 * Attempt to open the tap device and read the MAC address
543 devname = vtopts = strdup(opts);
544 (void) strsep(&vtopts, ",");
546 if (vtopts != NULL) {
547 err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
555 strcpy(tbuf, "/dev/");
556 strlcat(tbuf, devname, sizeof(tbuf));
560 sc->vsc_tapfd = open(tbuf, O_RDWR);
561 if (sc->vsc_tapfd == -1) {
562 WPRINTF(("open of tap device %s failed\n", tbuf));
565 * Set non-blocking and register for read
566 * notifications with the event loop
569 if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
570 WPRINTF(("tap device O_NONBLOCK failed\n"));
571 close(sc->vsc_tapfd);
575 sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
577 pci_vtnet_tap_callback,
579 if (sc->vsc_mevp == NULL) {
580 WPRINTF(("Could not register event\n"));
581 close(sc->vsc_tapfd);
588 * The default MAC address is the standard NetApp OUI of 00-a0-98,
589 * followed by an MD5 of the PCI slot/func number and dev name
592 snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
593 pi->pi_func, vmname);
596 MD5Update(&mdctx, nstr, strlen(nstr));
597 MD5Final(digest, &mdctx);
599 sc->vsc_config.mac[0] = 0x00;
600 sc->vsc_config.mac[1] = 0xa0;
601 sc->vsc_config.mac[2] = 0x98;
602 sc->vsc_config.mac[3] = digest[0];
603 sc->vsc_config.mac[4] = digest[1];
604 sc->vsc_config.mac[5] = digest[2];
607 /* initialize config space */
608 pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
609 pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
610 pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
611 pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
613 pci_lintr_request(pi);
616 sc->vsc_config.status = 1;
618 /* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
619 if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix()))
622 /* use BAR 0 to map config regs in IO space */
623 vi_set_io_bar(&sc->vsc_vs, 0);
627 sc->rx_in_progress = 0;
628 pthread_mutex_init(&sc->rx_mtx, NULL);
631 * Initialize tx semaphore & spawn TX processing thread.
632 * As of now, only one thread for TX desc processing is
635 sc->tx_in_progress = 0;
636 pthread_mutex_init(&sc->tx_mtx, NULL);
637 pthread_cond_init(&sc->tx_cond, NULL);
638 pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
639 snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
641 pthread_set_name_np(sc->tx_tid, tname);
647 pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
649 struct pci_vtnet_softc *sc = vsc;
653 assert(offset + size <= 6);
655 * The driver is allowed to change the MAC address
657 ptr = &sc->vsc_config.mac[offset];
658 memcpy(ptr, &value, size);
660 DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
667 pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
669 struct pci_vtnet_softc *sc = vsc;
672 ptr = (uint8_t *)&sc->vsc_config + offset;
673 memcpy(retval, ptr, size);
677 struct pci_devemu pci_de_vnet = {
678 .pe_emu = "virtio-net",
679 .pe_init = pci_vtnet_init,
680 .pe_barwrite = vi_pci_write,
681 .pe_barread = vi_pci_read
683 PCI_EMUL_SET(pci_de_vnet);