xref: /freebsd/usr.sbin/bhyve/pci_virtio_net.c (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
1 /*-
2  * Copyright (c) 2011 NetApp, Inc.
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
13  *
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
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/linker_set.h>
34 #include <sys/select.h>
35 #include <sys/uio.h>
36 #include <sys/ioctl.h>
37 #include <machine/atomic.h>
38 #include <net/ethernet.h>
39 #ifndef NETMAP_WITH_LIBS
40 #define NETMAP_WITH_LIBS
41 #endif
42 #include <net/netmap_user.h>
43 
44 #include <errno.h>
45 #include <fcntl.h>
46 #include <stdio.h>
47 #include <stdlib.h>
48 #include <stdint.h>
49 #include <string.h>
50 #include <strings.h>
51 #include <unistd.h>
52 #include <assert.h>
53 #include <md5.h>
54 #include <pthread.h>
55 #include <pthread_np.h>
56 
57 #include "bhyverun.h"
58 #include "pci_emul.h"
59 #include "mevent.h"
60 #include "virtio.h"
61 
62 #define VTNET_RINGSZ	1024
63 
64 #define VTNET_MAXSEGS	256
65 
66 /*
67  * Host capabilities.  Note that we only offer a few of these.
68  */
69 #define	VIRTIO_NET_F_CSUM	(1 <<  0) /* host handles partial cksum */
70 #define	VIRTIO_NET_F_GUEST_CSUM	(1 <<  1) /* guest handles partial cksum */
71 #define	VIRTIO_NET_F_MAC	(1 <<  5) /* host supplies MAC */
72 #define	VIRTIO_NET_F_GSO_DEPREC	(1 <<  6) /* deprecated: host handles GSO */
73 #define	VIRTIO_NET_F_GUEST_TSO4	(1 <<  7) /* guest can rcv TSOv4 */
74 #define	VIRTIO_NET_F_GUEST_TSO6	(1 <<  8) /* guest can rcv TSOv6 */
75 #define	VIRTIO_NET_F_GUEST_ECN	(1 <<  9) /* guest can rcv TSO with ECN */
76 #define	VIRTIO_NET_F_GUEST_UFO	(1 << 10) /* guest can rcv UFO */
77 #define	VIRTIO_NET_F_HOST_TSO4	(1 << 11) /* host can rcv TSOv4 */
78 #define	VIRTIO_NET_F_HOST_TSO6	(1 << 12) /* host can rcv TSOv6 */
79 #define	VIRTIO_NET_F_HOST_ECN	(1 << 13) /* host can rcv TSO with ECN */
80 #define	VIRTIO_NET_F_HOST_UFO	(1 << 14) /* host can rcv UFO */
81 #define	VIRTIO_NET_F_MRG_RXBUF	(1 << 15) /* host can merge RX buffers */
82 #define	VIRTIO_NET_F_STATUS	(1 << 16) /* config status field available */
83 #define	VIRTIO_NET_F_CTRL_VQ	(1 << 17) /* control channel available */
84 #define	VIRTIO_NET_F_CTRL_RX	(1 << 18) /* control channel RX mode support */
85 #define	VIRTIO_NET_F_CTRL_VLAN	(1 << 19) /* control channel VLAN filtering */
86 #define	VIRTIO_NET_F_GUEST_ANNOUNCE \
87 				(1 << 21) /* guest can send gratuitous pkts */
88 
89 #define VTNET_S_HOSTCAPS      \
90   ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_STATUS | \
91     VIRTIO_F_NOTIFY_ON_EMPTY | VIRTIO_RING_F_INDIRECT_DESC)
92 
93 /*
94  * PCI config-space "registers"
95  */
96 struct virtio_net_config {
97 	uint8_t  mac[6];
98 	uint16_t status;
99 } __packed;
100 
101 /*
102  * Queue definitions.
103  */
104 #define VTNET_RXQ	0
105 #define VTNET_TXQ	1
106 #define VTNET_CTLQ	2	/* NB: not yet supported */
107 
108 #define VTNET_MAXQ	3
109 
110 /*
111  * Fixed network header size
112  */
113 struct virtio_net_rxhdr {
114 	uint8_t		vrh_flags;
115 	uint8_t		vrh_gso_type;
116 	uint16_t	vrh_hdr_len;
117 	uint16_t	vrh_gso_size;
118 	uint16_t	vrh_csum_start;
119 	uint16_t	vrh_csum_offset;
120 	uint16_t	vrh_bufs;
121 } __packed;
122 
123 /*
124  * Debug printf
125  */
126 static int pci_vtnet_debug;
127 #define DPRINTF(params) if (pci_vtnet_debug) printf params
128 #define WPRINTF(params) printf params
129 
130 /*
131  * Per-device softc
132  */
133 struct pci_vtnet_softc {
134 	struct virtio_softc vsc_vs;
135 	struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
136 	pthread_mutex_t vsc_mtx;
137 	struct mevent	*vsc_mevp;
138 
139 	int		vsc_tapfd;
140 	struct nm_desc	*vsc_nmd;
141 
142 	int		vsc_rx_ready;
143 	volatile int	resetting;	/* set and checked outside lock */
144 
145 	uint64_t	vsc_features;	/* negotiated features */
146 
147 	struct virtio_net_config vsc_config;
148 
149 	pthread_mutex_t	rx_mtx;
150 	int		rx_in_progress;
151 	int		rx_vhdrlen;
152 	int		rx_merge;	/* merged rx bufs in use */
153 
154 	pthread_t 	tx_tid;
155 	pthread_mutex_t	tx_mtx;
156 	pthread_cond_t	tx_cond;
157 	int		tx_in_progress;
158 
159 	void (*pci_vtnet_rx)(struct pci_vtnet_softc *sc);
160 	void (*pci_vtnet_tx)(struct pci_vtnet_softc *sc, struct iovec *iov,
161 			     int iovcnt, int len);
162 };
163 
164 static void pci_vtnet_reset(void *);
165 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
166 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
167 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
168 static void pci_vtnet_neg_features(void *, uint64_t);
169 
170 static struct virtio_consts vtnet_vi_consts = {
171 	"vtnet",		/* our name */
172 	VTNET_MAXQ - 1,		/* we currently support 2 virtqueues */
173 	sizeof(struct virtio_net_config), /* config reg size */
174 	pci_vtnet_reset,	/* reset */
175 	NULL,			/* device-wide qnotify -- not used */
176 	pci_vtnet_cfgread,	/* read PCI config */
177 	pci_vtnet_cfgwrite,	/* write PCI config */
178 	pci_vtnet_neg_features,	/* apply negotiated features */
179 	VTNET_S_HOSTCAPS,	/* our capabilities */
180 };
181 
182 /*
183  * If the transmit thread is active then stall until it is done.
184  */
185 static void
186 pci_vtnet_txwait(struct pci_vtnet_softc *sc)
187 {
188 
189 	pthread_mutex_lock(&sc->tx_mtx);
190 	while (sc->tx_in_progress) {
191 		pthread_mutex_unlock(&sc->tx_mtx);
192 		usleep(10000);
193 		pthread_mutex_lock(&sc->tx_mtx);
194 	}
195 	pthread_mutex_unlock(&sc->tx_mtx);
196 }
197 
198 /*
199  * If the receive thread is active then stall until it is done.
200  */
201 static void
202 pci_vtnet_rxwait(struct pci_vtnet_softc *sc)
203 {
204 
205 	pthread_mutex_lock(&sc->rx_mtx);
206 	while (sc->rx_in_progress) {
207 		pthread_mutex_unlock(&sc->rx_mtx);
208 		usleep(10000);
209 		pthread_mutex_lock(&sc->rx_mtx);
210 	}
211 	pthread_mutex_unlock(&sc->rx_mtx);
212 }
213 
214 static void
215 pci_vtnet_reset(void *vsc)
216 {
217 	struct pci_vtnet_softc *sc = vsc;
218 
219 	DPRINTF(("vtnet: device reset requested !\n"));
220 
221 	sc->resetting = 1;
222 
223 	/*
224 	 * Wait for the transmit and receive threads to finish their
225 	 * processing.
226 	 */
227 	pci_vtnet_txwait(sc);
228 	pci_vtnet_rxwait(sc);
229 
230 	sc->vsc_rx_ready = 0;
231 	sc->rx_merge = 1;
232 	sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
233 
234 	/* now reset rings, MSI-X vectors, and negotiated capabilities */
235 	vi_reset_dev(&sc->vsc_vs);
236 
237 	sc->resetting = 0;
238 }
239 
240 /*
241  * Called to send a buffer chain out to the tap device
242  */
243 static void
244 pci_vtnet_tap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
245 		 int len)
246 {
247 	static char pad[60]; /* all zero bytes */
248 
249 	if (sc->vsc_tapfd == -1)
250 		return;
251 
252 	/*
253 	 * If the length is < 60, pad out to that and add the
254 	 * extra zero'd segment to the iov. It is guaranteed that
255 	 * there is always an extra iov available by the caller.
256 	 */
257 	if (len < 60) {
258 		iov[iovcnt].iov_base = pad;
259 		iov[iovcnt].iov_len = 60 - len;
260 		iovcnt++;
261 	}
262 	(void) writev(sc->vsc_tapfd, iov, iovcnt);
263 }
264 
265 /*
266  *  Called when there is read activity on the tap file descriptor.
267  * Each buffer posted by the guest is assumed to be able to contain
268  * an entire ethernet frame + rx header.
269  *  MP note: the dummybuf is only used for discarding frames, so there
270  * is no need for it to be per-vtnet or locked.
271  */
272 static uint8_t dummybuf[2048];
273 
274 static __inline struct iovec *
275 rx_iov_trim(struct iovec *iov, int *niov, int tlen)
276 {
277 	struct iovec *riov;
278 
279 	/* XXX short-cut: assume first segment is >= tlen */
280 	assert(iov[0].iov_len >= tlen);
281 
282 	iov[0].iov_len -= tlen;
283 	if (iov[0].iov_len == 0) {
284 		assert(*niov > 1);
285 		*niov -= 1;
286 		riov = &iov[1];
287 	} else {
288 		iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + tlen);
289 		riov = &iov[0];
290 	}
291 
292 	return (riov);
293 }
294 
295 static void
296 pci_vtnet_tap_rx(struct pci_vtnet_softc *sc)
297 {
298 	struct iovec iov[VTNET_MAXSEGS], *riov;
299 	struct vqueue_info *vq;
300 	void *vrx;
301 	int len, n;
302 	uint16_t idx;
303 
304 	/*
305 	 * Should never be called without a valid tap fd
306 	 */
307 	assert(sc->vsc_tapfd != -1);
308 
309 	/*
310 	 * But, will be called when the rx ring hasn't yet
311 	 * been set up or the guest is resetting the device.
312 	 */
313 	if (!sc->vsc_rx_ready || sc->resetting) {
314 		/*
315 		 * Drop the packet and try later.
316 		 */
317 		(void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
318 		return;
319 	}
320 
321 	/*
322 	 * Check for available rx buffers
323 	 */
324 	vq = &sc->vsc_queues[VTNET_RXQ];
325 	if (!vq_has_descs(vq)) {
326 		/*
327 		 * Drop the packet and try later.  Interrupt on
328 		 * empty, if that's negotiated.
329 		 */
330 		(void) read(sc->vsc_tapfd, dummybuf, sizeof(dummybuf));
331 		vq_endchains(vq, 1);
332 		return;
333 	}
334 
335 	do {
336 		/*
337 		 * Get descriptor chain.
338 		 */
339 		n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
340 		assert(n >= 1 && n <= VTNET_MAXSEGS);
341 
342 		/*
343 		 * Get a pointer to the rx header, and use the
344 		 * data immediately following it for the packet buffer.
345 		 */
346 		vrx = iov[0].iov_base;
347 		riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
348 
349 		len = readv(sc->vsc_tapfd, riov, n);
350 
351 		if (len < 0 && errno == EWOULDBLOCK) {
352 			/*
353 			 * No more packets, but still some avail ring
354 			 * entries.  Interrupt if needed/appropriate.
355 			 */
356 			vq_retchain(vq);
357 			vq_endchains(vq, 0);
358 			return;
359 		}
360 
361 		/*
362 		 * The only valid field in the rx packet header is the
363 		 * number of buffers if merged rx bufs were negotiated.
364 		 */
365 		memset(vrx, 0, sc->rx_vhdrlen);
366 
367 		if (sc->rx_merge) {
368 			struct virtio_net_rxhdr *vrxh;
369 
370 			vrxh = vrx;
371 			vrxh->vrh_bufs = 1;
372 		}
373 
374 		/*
375 		 * Release this chain and handle more chains.
376 		 */
377 		vq_relchain(vq, idx, len + sc->rx_vhdrlen);
378 	} while (vq_has_descs(vq));
379 
380 	/* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
381 	vq_endchains(vq, 1);
382 }
383 
384 static int
385 pci_vtnet_netmap_writev(struct nm_desc *nmd, struct iovec *iov, int iovcnt)
386 {
387 	int r, i;
388 	int len = 0;
389 
390 	for (r = nmd->cur_tx_ring; ; ) {
391 		struct netmap_ring *ring = NETMAP_TXRING(nmd->nifp, r);
392 		uint32_t cur, idx;
393 		char *buf;
394 
395 		if (nm_ring_empty(ring)) {
396 			r++;
397 			if (r > nmd->last_tx_ring)
398 				r = nmd->first_tx_ring;
399 			if (r == nmd->cur_rx_ring)
400 				break;
401 			continue;
402 		}
403 		cur = ring->cur;
404 		idx = ring->slot[cur].buf_idx;
405 		buf = NETMAP_BUF(ring, idx);
406 
407 		for (i = 0; i < iovcnt; i++) {
408 			memcpy(&buf[len], iov[i].iov_base, iov[i].iov_len);
409 			len += iov[i].iov_len;
410 		}
411 		ring->slot[cur].len = len;
412 		ring->head = ring->cur = nm_ring_next(ring, cur);
413 		nmd->cur_tx_ring = r;
414 		ioctl(nmd->fd, NIOCTXSYNC, NULL);
415 		break;
416 	}
417 
418 	return (len);
419 }
420 
421 static inline int
422 pci_vtnet_netmap_readv(struct nm_desc *nmd, struct iovec *iov, int iovcnt)
423 {
424 	int len = 0;
425 	int i = 0;
426 	int r;
427 
428 	for (r = nmd->cur_rx_ring; ; ) {
429 		struct netmap_ring *ring = NETMAP_RXRING(nmd->nifp, r);
430 		uint32_t cur, idx;
431 		char *buf;
432 		size_t left;
433 
434 		if (nm_ring_empty(ring)) {
435 			r++;
436 			if (r > nmd->last_rx_ring)
437 				r = nmd->first_rx_ring;
438 			if (r == nmd->cur_rx_ring)
439 				break;
440 			continue;
441 		}
442 		cur = ring->cur;
443 		idx = ring->slot[cur].buf_idx;
444 		buf = NETMAP_BUF(ring, idx);
445 		left = ring->slot[cur].len;
446 
447 		for (i = 0; i < iovcnt && left > 0; i++) {
448 			if (iov[i].iov_len > left)
449 				iov[i].iov_len = left;
450 			memcpy(iov[i].iov_base, &buf[len], iov[i].iov_len);
451 			len += iov[i].iov_len;
452 			left -= iov[i].iov_len;
453 		}
454 		ring->head = ring->cur = nm_ring_next(ring, cur);
455 		nmd->cur_rx_ring = r;
456 		ioctl(nmd->fd, NIOCRXSYNC, NULL);
457 		break;
458 	}
459 	for (; i < iovcnt; i++)
460 		iov[i].iov_len = 0;
461 
462 	return (len);
463 }
464 
465 /*
466  * Called to send a buffer chain out to the vale port
467  */
468 static void
469 pci_vtnet_netmap_tx(struct pci_vtnet_softc *sc, struct iovec *iov, int iovcnt,
470 		    int len)
471 {
472 	static char pad[60]; /* all zero bytes */
473 
474 	if (sc->vsc_nmd == NULL)
475 		return;
476 
477 	/*
478 	 * If the length is < 60, pad out to that and add the
479 	 * extra zero'd segment to the iov. It is guaranteed that
480 	 * there is always an extra iov available by the caller.
481 	 */
482 	if (len < 60) {
483 		iov[iovcnt].iov_base = pad;
484 		iov[iovcnt].iov_len = 60 - len;
485 		iovcnt++;
486 	}
487 	(void) pci_vtnet_netmap_writev(sc->vsc_nmd, iov, iovcnt);
488 }
489 
490 static void
491 pci_vtnet_netmap_rx(struct pci_vtnet_softc *sc)
492 {
493 	struct iovec iov[VTNET_MAXSEGS], *riov;
494 	struct vqueue_info *vq;
495 	void *vrx;
496 	int len, n;
497 	uint16_t idx;
498 
499 	/*
500 	 * Should never be called without a valid netmap descriptor
501 	 */
502 	assert(sc->vsc_nmd != NULL);
503 
504 	/*
505 	 * But, will be called when the rx ring hasn't yet
506 	 * been set up or the guest is resetting the device.
507 	 */
508 	if (!sc->vsc_rx_ready || sc->resetting) {
509 		/*
510 		 * Drop the packet and try later.
511 		 */
512 		(void) nm_nextpkt(sc->vsc_nmd, (void *)dummybuf);
513 		return;
514 	}
515 
516 	/*
517 	 * Check for available rx buffers
518 	 */
519 	vq = &sc->vsc_queues[VTNET_RXQ];
520 	if (!vq_has_descs(vq)) {
521 		/*
522 		 * Drop the packet and try later.  Interrupt on
523 		 * empty, if that's negotiated.
524 		 */
525 		(void) nm_nextpkt(sc->vsc_nmd, (void *)dummybuf);
526 		vq_endchains(vq, 1);
527 		return;
528 	}
529 
530 	do {
531 		/*
532 		 * Get descriptor chain.
533 		 */
534 		n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
535 		assert(n >= 1 && n <= VTNET_MAXSEGS);
536 
537 		/*
538 		 * Get a pointer to the rx header, and use the
539 		 * data immediately following it for the packet buffer.
540 		 */
541 		vrx = iov[0].iov_base;
542 		riov = rx_iov_trim(iov, &n, sc->rx_vhdrlen);
543 
544 		len = pci_vtnet_netmap_readv(sc->vsc_nmd, riov, n);
545 
546 		if (len == 0) {
547 			/*
548 			 * No more packets, but still some avail ring
549 			 * entries.  Interrupt if needed/appropriate.
550 			 */
551 			vq_endchains(vq, 0);
552 			return;
553 		}
554 
555 		/*
556 		 * The only valid field in the rx packet header is the
557 		 * number of buffers if merged rx bufs were negotiated.
558 		 */
559 		memset(vrx, 0, sc->rx_vhdrlen);
560 
561 		if (sc->rx_merge) {
562 			struct virtio_net_rxhdr *vrxh;
563 
564 			vrxh = vrx;
565 			vrxh->vrh_bufs = 1;
566 		}
567 
568 		/*
569 		 * Release this chain and handle more chains.
570 		 */
571 		vq_relchain(vq, idx, len + sc->rx_vhdrlen);
572 	} while (vq_has_descs(vq));
573 
574 	/* Interrupt if needed, including for NOTIFY_ON_EMPTY. */
575 	vq_endchains(vq, 1);
576 }
577 
578 static void
579 pci_vtnet_rx_callback(int fd, enum ev_type type, void *param)
580 {
581 	struct pci_vtnet_softc *sc = param;
582 
583 	pthread_mutex_lock(&sc->rx_mtx);
584 	sc->rx_in_progress = 1;
585 	sc->pci_vtnet_rx(sc);
586 	sc->rx_in_progress = 0;
587 	pthread_mutex_unlock(&sc->rx_mtx);
588 
589 }
590 
591 static void
592 pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
593 {
594 	struct pci_vtnet_softc *sc = vsc;
595 
596 	/*
597 	 * A qnotify means that the rx process can now begin
598 	 */
599 	if (sc->vsc_rx_ready == 0) {
600 		sc->vsc_rx_ready = 1;
601 		vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
602 	}
603 }
604 
605 static void
606 pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
607 {
608 	struct iovec iov[VTNET_MAXSEGS + 1];
609 	int i, n;
610 	int plen, tlen;
611 	uint16_t idx;
612 
613 	/*
614 	 * Obtain chain of descriptors.  The first one is
615 	 * really the header descriptor, so we need to sum
616 	 * up two lengths: packet length and transfer length.
617 	 */
618 	n = vq_getchain(vq, &idx, iov, VTNET_MAXSEGS, NULL);
619 	assert(n >= 1 && n <= VTNET_MAXSEGS);
620 	plen = 0;
621 	tlen = iov[0].iov_len;
622 	for (i = 1; i < n; i++) {
623 		plen += iov[i].iov_len;
624 		tlen += iov[i].iov_len;
625 	}
626 
627 	DPRINTF(("virtio: packet send, %d bytes, %d segs\n\r", plen, n));
628 	sc->pci_vtnet_tx(sc, &iov[1], n - 1, plen);
629 
630 	/* chain is processed, release it and set tlen */
631 	vq_relchain(vq, idx, tlen);
632 }
633 
634 static void
635 pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
636 {
637 	struct pci_vtnet_softc *sc = vsc;
638 
639 	/*
640 	 * Any ring entries to process?
641 	 */
642 	if (!vq_has_descs(vq))
643 		return;
644 
645 	/* Signal the tx thread for processing */
646 	pthread_mutex_lock(&sc->tx_mtx);
647 	vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
648 	if (sc->tx_in_progress == 0)
649 		pthread_cond_signal(&sc->tx_cond);
650 	pthread_mutex_unlock(&sc->tx_mtx);
651 }
652 
653 /*
654  * Thread which will handle processing of TX desc
655  */
656 static void *
657 pci_vtnet_tx_thread(void *param)
658 {
659 	struct pci_vtnet_softc *sc = param;
660 	struct vqueue_info *vq;
661 	int error;
662 
663 	vq = &sc->vsc_queues[VTNET_TXQ];
664 
665 	/*
666 	 * Let us wait till the tx queue pointers get initialised &
667 	 * first tx signaled
668 	 */
669 	pthread_mutex_lock(&sc->tx_mtx);
670 	error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
671 	assert(error == 0);
672 
673 	for (;;) {
674 		/* note - tx mutex is locked here */
675 		while (sc->resetting || !vq_has_descs(vq)) {
676 			vq->vq_used->vu_flags &= ~VRING_USED_F_NO_NOTIFY;
677 			mb();
678 			if (!sc->resetting && vq_has_descs(vq))
679 				break;
680 
681 			sc->tx_in_progress = 0;
682 			error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
683 			assert(error == 0);
684 		}
685 		vq->vq_used->vu_flags |= VRING_USED_F_NO_NOTIFY;
686 		sc->tx_in_progress = 1;
687 		pthread_mutex_unlock(&sc->tx_mtx);
688 
689 		do {
690 			/*
691 			 * Run through entries, placing them into
692 			 * iovecs and sending when an end-of-packet
693 			 * is found
694 			 */
695 			pci_vtnet_proctx(sc, vq);
696 		} while (vq_has_descs(vq));
697 
698 		/*
699 		 * Generate an interrupt if needed.
700 		 */
701 		vq_endchains(vq, 1);
702 
703 		pthread_mutex_lock(&sc->tx_mtx);
704 	}
705 }
706 
707 #ifdef notyet
708 static void
709 pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
710 {
711 
712 	DPRINTF(("vtnet: control qnotify!\n\r"));
713 }
714 #endif
715 
716 static int
717 pci_vtnet_parsemac(char *mac_str, uint8_t *mac_addr)
718 {
719         struct ether_addr *ea;
720         char *tmpstr;
721         char zero_addr[ETHER_ADDR_LEN] = { 0, 0, 0, 0, 0, 0 };
722 
723         tmpstr = strsep(&mac_str,"=");
724 
725         if ((mac_str != NULL) && (!strcmp(tmpstr,"mac"))) {
726                 ea = ether_aton(mac_str);
727 
728                 if (ea == NULL || ETHER_IS_MULTICAST(ea->octet) ||
729                     memcmp(ea->octet, zero_addr, ETHER_ADDR_LEN) == 0) {
730 			fprintf(stderr, "Invalid MAC %s\n", mac_str);
731                         return (EINVAL);
732                 } else
733                         memcpy(mac_addr, ea->octet, ETHER_ADDR_LEN);
734         }
735 
736         return (0);
737 }
738 
739 static void
740 pci_vtnet_tap_setup(struct pci_vtnet_softc *sc, char *devname)
741 {
742 	char tbuf[80];
743 
744 	strcpy(tbuf, "/dev/");
745 	strlcat(tbuf, devname, sizeof(tbuf));
746 
747 	sc->pci_vtnet_rx = pci_vtnet_tap_rx;
748 	sc->pci_vtnet_tx = pci_vtnet_tap_tx;
749 
750 	sc->vsc_tapfd = open(tbuf, O_RDWR);
751 	if (sc->vsc_tapfd == -1) {
752 		WPRINTF(("open of tap device %s failed\n", tbuf));
753 		return;
754 	}
755 
756 	/*
757 	 * Set non-blocking and register for read
758 	 * notifications with the event loop
759 	 */
760 	int opt = 1;
761 	if (ioctl(sc->vsc_tapfd, FIONBIO, &opt) < 0) {
762 		WPRINTF(("tap device O_NONBLOCK failed\n"));
763 		close(sc->vsc_tapfd);
764 		sc->vsc_tapfd = -1;
765 	}
766 
767 	sc->vsc_mevp = mevent_add(sc->vsc_tapfd,
768 				  EVF_READ,
769 				  pci_vtnet_rx_callback,
770 				  sc);
771 	if (sc->vsc_mevp == NULL) {
772 		WPRINTF(("Could not register event\n"));
773 		close(sc->vsc_tapfd);
774 		sc->vsc_tapfd = -1;
775 	}
776 }
777 
778 static void
779 pci_vtnet_netmap_setup(struct pci_vtnet_softc *sc, char *ifname)
780 {
781 	sc->pci_vtnet_rx = pci_vtnet_netmap_rx;
782 	sc->pci_vtnet_tx = pci_vtnet_netmap_tx;
783 
784 	sc->vsc_nmd = nm_open(ifname, NULL, 0, 0);
785 	if (sc->vsc_nmd == NULL) {
786 		WPRINTF(("open of netmap device %s failed\n", ifname));
787 		return;
788 	}
789 
790 	sc->vsc_mevp = mevent_add(sc->vsc_nmd->fd,
791 				  EVF_READ,
792 				  pci_vtnet_rx_callback,
793 				  sc);
794 	if (sc->vsc_mevp == NULL) {
795 		WPRINTF(("Could not register event\n"));
796 		nm_close(sc->vsc_nmd);
797 		sc->vsc_nmd = NULL;
798 	}
799 }
800 
801 static int
802 pci_vtnet_init(struct vmctx *ctx, struct pci_devinst *pi, char *opts)
803 {
804 	MD5_CTX mdctx;
805 	unsigned char digest[16];
806 	char nstr[80];
807 	char tname[MAXCOMLEN + 1];
808 	struct pci_vtnet_softc *sc;
809 	char *devname;
810 	char *vtopts;
811 	int mac_provided;
812 
813 	sc = calloc(1, sizeof(struct pci_vtnet_softc));
814 
815 	pthread_mutex_init(&sc->vsc_mtx, NULL);
816 
817 	vi_softc_linkup(&sc->vsc_vs, &vtnet_vi_consts, sc, pi, sc->vsc_queues);
818 	sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
819 
820 	sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
821 	sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
822 	sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
823 	sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
824 #ifdef notyet
825 	sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
826         sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
827 #endif
828 
829 	/*
830 	 * Attempt to open the tap device and read the MAC address
831 	 * if specified
832 	 */
833 	mac_provided = 0;
834 	sc->vsc_tapfd = -1;
835 	sc->vsc_nmd = NULL;
836 	if (opts != NULL) {
837 		int err;
838 
839 		devname = vtopts = strdup(opts);
840 		(void) strsep(&vtopts, ",");
841 
842 		if (vtopts != NULL) {
843 			err = pci_vtnet_parsemac(vtopts, sc->vsc_config.mac);
844 			if (err != 0) {
845 				free(devname);
846 				return (err);
847 			}
848 			mac_provided = 1;
849 		}
850 
851 		if (strncmp(devname, "vale", 4) == 0)
852 			pci_vtnet_netmap_setup(sc, devname);
853 		if (strncmp(devname, "tap", 3) == 0 ||
854 		    strncmp(devname, "vmnet", 5) == 0)
855 			pci_vtnet_tap_setup(sc, devname);
856 
857 		free(devname);
858 	}
859 
860 	/*
861 	 * The default MAC address is the standard NetApp OUI of 00-a0-98,
862 	 * followed by an MD5 of the PCI slot/func number and dev name
863 	 */
864 	if (!mac_provided) {
865 		snprintf(nstr, sizeof(nstr), "%d-%d-%s", pi->pi_slot,
866 		    pi->pi_func, vmname);
867 
868 		MD5Init(&mdctx);
869 		MD5Update(&mdctx, nstr, strlen(nstr));
870 		MD5Final(digest, &mdctx);
871 
872 		sc->vsc_config.mac[0] = 0x00;
873 		sc->vsc_config.mac[1] = 0xa0;
874 		sc->vsc_config.mac[2] = 0x98;
875 		sc->vsc_config.mac[3] = digest[0];
876 		sc->vsc_config.mac[4] = digest[1];
877 		sc->vsc_config.mac[5] = digest[2];
878 	}
879 
880 	/* initialize config space */
881 	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
882 	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
883 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
884 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_TYPE_NET);
885 	pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
886 
887 	/* Link is up if we managed to open tap device. */
888 	sc->vsc_config.status = (opts == NULL || sc->vsc_tapfd >= 0);
889 
890 	/* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
891 	if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix()))
892 		return (1);
893 
894 	/* use BAR 0 to map config regs in IO space */
895 	vi_set_io_bar(&sc->vsc_vs, 0);
896 
897 	sc->resetting = 0;
898 
899 	sc->rx_merge = 1;
900 	sc->rx_vhdrlen = sizeof(struct virtio_net_rxhdr);
901 	sc->rx_in_progress = 0;
902 	pthread_mutex_init(&sc->rx_mtx, NULL);
903 
904 	/*
905 	 * Initialize tx semaphore & spawn TX processing thread.
906 	 * As of now, only one thread for TX desc processing is
907 	 * spawned.
908 	 */
909 	sc->tx_in_progress = 0;
910 	pthread_mutex_init(&sc->tx_mtx, NULL);
911 	pthread_cond_init(&sc->tx_cond, NULL);
912 	pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
913 	snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
914 	    pi->pi_func);
915         pthread_set_name_np(sc->tx_tid, tname);
916 
917 	return (0);
918 }
919 
920 static int
921 pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
922 {
923 	struct pci_vtnet_softc *sc = vsc;
924 	void *ptr;
925 
926 	if (offset < 6) {
927 		assert(offset + size <= 6);
928 		/*
929 		 * The driver is allowed to change the MAC address
930 		 */
931 		ptr = &sc->vsc_config.mac[offset];
932 		memcpy(ptr, &value, size);
933 	} else {
934 		/* silently ignore other writes */
935 		DPRINTF(("vtnet: write to readonly reg %d\n\r", offset));
936 	}
937 
938 	return (0);
939 }
940 
941 static int
942 pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
943 {
944 	struct pci_vtnet_softc *sc = vsc;
945 	void *ptr;
946 
947 	ptr = (uint8_t *)&sc->vsc_config + offset;
948 	memcpy(retval, ptr, size);
949 	return (0);
950 }
951 
952 static void
953 pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features)
954 {
955 	struct pci_vtnet_softc *sc = vsc;
956 
957 	sc->vsc_features = negotiated_features;
958 
959 	if (!(sc->vsc_features & VIRTIO_NET_F_MRG_RXBUF)) {
960 		sc->rx_merge = 0;
961 		/* non-merge rx header is 2 bytes shorter */
962 		sc->rx_vhdrlen -= 2;
963 	}
964 }
965 
966 struct pci_devemu pci_de_vnet = {
967 	.pe_emu = 	"virtio-net",
968 	.pe_init =	pci_vtnet_init,
969 	.pe_barwrite =	vi_pci_write,
970 	.pe_barread =	vi_pci_read
971 };
972 PCI_EMUL_SET(pci_de_vnet);
973