1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 NetApp, Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
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.
15  *
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
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 
31 #include <sys/param.h>
32 #include <sys/linker_set.h>
33 #include <sys/select.h>
34 #include <sys/uio.h>
35 #include <sys/ioctl.h>
36 #include <net/ethernet.h>
37 #include <net/if.h> /* IFNAMSIZ */
38 
39 #include <err.h>
40 #include <errno.h>
41 #include <fcntl.h>
42 #include <stdio.h>
43 #include <stdlib.h>
44 #include <stdint.h>
45 #include <string.h>
46 #include <strings.h>
47 #include <unistd.h>
48 #include <assert.h>
49 #include <pthread.h>
50 #include <pthread_np.h>
51 
52 #include "bhyverun.h"
53 #include "config.h"
54 #include "debug.h"
55 #include "pci_emul.h"
56 #include "mevent.h"
57 #include "virtio.h"
58 #include "net_utils.h"
59 #include "net_backends.h"
60 #include "iov.h"
61 
62 #define VTNET_RINGSZ	1024
63 
64 #define VTNET_MAXSEGS	256
65 
66 #define VTNET_MAX_PKT_LEN	(65536 + 64)
67 
68 #define VTNET_MIN_MTU	ETHERMIN
69 #define VTNET_MAX_MTU	65535
70 
71 #define VTNET_S_HOSTCAPS      \
72   ( VIRTIO_NET_F_MAC | VIRTIO_NET_F_STATUS | \
73     VIRTIO_F_NOTIFY_ON_EMPTY | VIRTIO_RING_F_INDIRECT_DESC)
74 
75 /*
76  * PCI config-space "registers"
77  */
78 struct virtio_net_config {
79 	uint8_t  mac[6];
80 	uint16_t status;
81 	uint16_t max_virtqueue_pairs;
82 	uint16_t mtu;
83 } __packed;
84 
85 /*
86  * Queue definitions.
87  */
88 #define VTNET_RXQ	0
89 #define VTNET_TXQ	1
90 #define VTNET_CTLQ	2	/* NB: not yet supported */
91 
92 #define VTNET_MAXQ	3
93 
94 /*
95  * Debug printf
96  */
97 static int pci_vtnet_debug;
98 #define DPRINTF(params) if (pci_vtnet_debug) PRINTLN params
99 #define WPRINTF(params) PRINTLN params
100 
101 /*
102  * Per-device softc
103  */
104 struct pci_vtnet_softc {
105 	struct virtio_softc vsc_vs;
106 	struct vqueue_info vsc_queues[VTNET_MAXQ - 1];
107 	pthread_mutex_t vsc_mtx;
108 
109 	net_backend_t	*vsc_be;
110 
111 	bool    features_negotiated;	/* protected by rx_mtx */
112 
113 	int		resetting;	/* protected by tx_mtx */
114 
115 	uint64_t	vsc_features;	/* negotiated features */
116 
117 	pthread_mutex_t	rx_mtx;
118 	int		rx_merge;	/* merged rx bufs in use */
119 
120 	pthread_t 	tx_tid;
121 	pthread_mutex_t	tx_mtx;
122 	pthread_cond_t	tx_cond;
123 	int		tx_in_progress;
124 
125 	size_t		vhdrlen;
126 	size_t		be_vhdrlen;
127 
128 	struct virtio_net_config vsc_config;
129 	struct virtio_consts vsc_consts;
130 };
131 
132 static void pci_vtnet_reset(void *);
133 /* static void pci_vtnet_notify(void *, struct vqueue_info *); */
134 static int pci_vtnet_cfgread(void *, int, int, uint32_t *);
135 static int pci_vtnet_cfgwrite(void *, int, int, uint32_t);
136 static void pci_vtnet_neg_features(void *, uint64_t);
137 
138 static struct virtio_consts vtnet_vi_consts = {
139 	.vc_name =	"vtnet",
140 	.vc_nvq =	VTNET_MAXQ - 1,
141 	.vc_cfgsize =	sizeof(struct virtio_net_config),
142 	.vc_reset =	pci_vtnet_reset,
143 	.vc_cfgread =	pci_vtnet_cfgread,
144 	.vc_cfgwrite =	pci_vtnet_cfgwrite,
145 	.vc_apply_features = pci_vtnet_neg_features,
146 	.vc_hv_caps =	VTNET_S_HOSTCAPS,
147 };
148 
149 static void
pci_vtnet_reset(void * vsc)150 pci_vtnet_reset(void *vsc)
151 {
152 	struct pci_vtnet_softc *sc = vsc;
153 
154 	DPRINTF(("vtnet: device reset requested !"));
155 
156 	/* Acquire the RX lock to block RX processing. */
157 	pthread_mutex_lock(&sc->rx_mtx);
158 
159 	/*
160 	 * Make sure receive operation is disabled at least until we
161 	 * re-negotiate the features, since receive operation depends
162 	 * on the value of sc->rx_merge and the header length, which
163 	 * are both set in pci_vtnet_neg_features().
164 	 * Receive operation will be enabled again once the guest adds
165 	 * the first receive buffers and kicks us.
166 	 */
167 	sc->features_negotiated = false;
168 	netbe_rx_disable(sc->vsc_be);
169 
170 	/* Set sc->resetting and give a chance to the TX thread to stop. */
171 	pthread_mutex_lock(&sc->tx_mtx);
172 	sc->resetting = 1;
173 	while (sc->tx_in_progress) {
174 		pthread_mutex_unlock(&sc->tx_mtx);
175 		usleep(10000);
176 		pthread_mutex_lock(&sc->tx_mtx);
177 	}
178 
179 	/*
180 	 * Now reset rings, MSI-X vectors, and negotiated capabilities.
181 	 * Do that with the TX lock held, since we need to reset
182 	 * sc->resetting.
183 	 */
184 	vi_reset_dev(&sc->vsc_vs);
185 
186 	sc->resetting = 0;
187 	pthread_mutex_unlock(&sc->tx_mtx);
188 	pthread_mutex_unlock(&sc->rx_mtx);
189 }
190 
191 static __inline struct iovec *
iov_trim_hdr(struct iovec * iov,int * iovcnt,unsigned int hlen)192 iov_trim_hdr(struct iovec *iov, int *iovcnt, unsigned int hlen)
193 {
194 	struct iovec *riov;
195 
196 	if (iov[0].iov_len < hlen) {
197 		/*
198 		 * Not enough header space in the first fragment.
199 		 * That's not ok for us.
200 		 */
201 		return NULL;
202 	}
203 
204 	iov[0].iov_len -= hlen;
205 	if (iov[0].iov_len == 0) {
206 		*iovcnt -= 1;
207 		if (*iovcnt == 0) {
208 			/*
209 			 * Only space for the header. That's not
210 			 * enough for us.
211 			 */
212 			return NULL;
213 		}
214 		riov = &iov[1];
215 	} else {
216 		iov[0].iov_base = (void *)((uintptr_t)iov[0].iov_base + hlen);
217 		riov = &iov[0];
218 	}
219 
220 	return (riov);
221 }
222 
223 struct virtio_mrg_rxbuf_info {
224 	uint16_t idx;
225 	uint16_t pad;
226 	uint32_t len;
227 };
228 
229 static void
pci_vtnet_rx(struct pci_vtnet_softc * sc)230 pci_vtnet_rx(struct pci_vtnet_softc *sc)
231 {
232 	int prepend_hdr_len = sc->vhdrlen - sc->be_vhdrlen;
233 	struct virtio_mrg_rxbuf_info info[VTNET_MAXSEGS];
234 	struct iovec iov[VTNET_MAXSEGS + 1];
235 	struct vqueue_info *vq;
236 	struct vi_req req;
237 
238 	vq = &sc->vsc_queues[VTNET_RXQ];
239 
240 	/* Features must be negotiated */
241 	if (!sc->features_negotiated) {
242 		return;
243 	}
244 
245 	for (;;) {
246 		struct virtio_net_rxhdr *hdr;
247 		uint32_t riov_bytes;
248 		struct iovec *riov;
249 		uint32_t ulen;
250 		int riov_len;
251 		int n_chains;
252 		ssize_t rlen;
253 		ssize_t plen;
254 
255 		plen = netbe_peek_recvlen(sc->vsc_be);
256 		if (plen <= 0) {
257 			/*
258 			 * No more packets (plen == 0), or backend errored
259 			 * (plen < 0). Interrupt if needed and stop.
260 			 */
261 			vq_endchains(vq, /*used_all_avail=*/0);
262 			return;
263 		}
264 		plen += prepend_hdr_len;
265 
266 		/*
267 		 * Get a descriptor chain to store the next ingress
268 		 * packet. In case of mergeable rx buffers, get as
269 		 * many chains as necessary in order to make room
270 		 * for plen bytes.
271 		 */
272 		riov_bytes = 0;
273 		riov_len = 0;
274 		riov = iov;
275 		n_chains = 0;
276 		do {
277 			int n = vq_getchain(vq, riov, VTNET_MAXSEGS - riov_len,
278 			    &req);
279 			info[n_chains].idx = req.idx;
280 
281 			if (n == 0) {
282 				/*
283 				 * No rx buffers. Enable RX kicks and double
284 				 * check.
285 				 */
286 				vq_kick_enable(vq);
287 				if (!vq_has_descs(vq)) {
288 					/*
289 					 * Still no buffers. Return the unused
290 					 * chains (if any), interrupt if needed
291 					 * (including for NOTIFY_ON_EMPTY), and
292 					 * disable the backend until the next
293 					 * kick.
294 					 */
295 					vq_retchains(vq, n_chains);
296 					vq_endchains(vq, /*used_all_avail=*/1);
297 					netbe_rx_disable(sc->vsc_be);
298 					return;
299 				}
300 
301 				/* More rx buffers found, so keep going. */
302 				vq_kick_disable(vq);
303 				continue;
304 			}
305 #ifndef __FreeBSD__
306 			if (n == -1) {
307 				/*
308 				 * An error from vq_getchain() means that
309 				 * an invalid descriptor was found.
310 				 */
311 				vq_retchains(vq, n_chains);
312 				vq_endchains(vq, /*used_all_avail=*/0);
313 				return;
314 			}
315 #endif
316 			assert(n >= 1 && riov_len + n <= VTNET_MAXSEGS);
317 			riov_len += n;
318 			if (!sc->rx_merge) {
319 				n_chains = 1;
320 				break;
321 			}
322 #ifndef __FreeBSD__
323 			size_t c = count_iov(riov, n);
324 			if (c > UINT32_MAX) {
325 				vq_retchains(vq, n_chains);
326 				vq_endchains(vq, /*used_all_avail=*/0);
327 				return;
328 			}
329 			info[n_chains].len = (uint32_t)c;
330 #else
331 			info[n_chains].len = (uint32_t)count_iov(riov, n);
332 #endif
333 			riov_bytes += info[n_chains].len;
334 			riov += n;
335 			n_chains++;
336 		} while (riov_bytes < plen && riov_len < VTNET_MAXSEGS);
337 
338 		riov = iov;
339 #ifdef __FreeBSD__
340 		hdr = riov[0].iov_base;
341 #else
342 		hdr = (struct virtio_net_rxhdr *)riov[0].iov_base;
343 #endif
344 		if (prepend_hdr_len > 0) {
345 			/*
346 			 * The frontend uses a virtio-net header, but the
347 			 * backend does not. We need to prepend a zeroed
348 			 * header.
349 			 */
350 			riov = iov_trim_hdr(riov, &riov_len, prepend_hdr_len);
351 			if (riov == NULL) {
352 				/*
353 				 * The first collected chain is nonsensical,
354 				 * as it is not even enough to store the
355 				 * virtio-net header. Just drop it.
356 				 */
357 				vq_relchain(vq, info[0].idx, 0);
358 				vq_retchains(vq, n_chains - 1);
359 				continue;
360 			}
361 			memset(hdr, 0, prepend_hdr_len);
362 		}
363 
364 		rlen = netbe_recv(sc->vsc_be, riov, riov_len);
365 		if (rlen != plen - prepend_hdr_len) {
366 			/*
367 			 * If this happens it means there is something
368 			 * wrong with the backend (e.g., some other
369 			 * process is stealing our packets).
370 			 */
371 			WPRINTF(("netbe_recv: expected %zd bytes, "
372 				"got %zd", plen - prepend_hdr_len, rlen));
373 			vq_retchains(vq, n_chains);
374 			continue;
375 		}
376 
377 		ulen = (uint32_t)plen;
378 
379 		/*
380 		 * Publish the used buffers to the guest, reporting the
381 		 * number of bytes that we wrote.
382 		 */
383 		if (!sc->rx_merge) {
384 			vq_relchain(vq, info[0].idx, ulen);
385 		} else {
386 			uint32_t iolen;
387 			int i = 0;
388 
389 			do {
390 				iolen = info[i].len;
391 				if (iolen > ulen) {
392 					iolen = ulen;
393 				}
394 				vq_relchain_prepare(vq, info[i].idx, iolen);
395 				ulen -= iolen;
396 				i++;
397 			} while (ulen > 0);
398 
399 			hdr->vrh_bufs = i;
400 			vq_relchain_publish(vq);
401 			assert(i == n_chains);
402 		}
403 	}
404 
405 }
406 
407 /*
408  * Called when there is read activity on the backend file descriptor.
409  * Each buffer posted by the guest is assumed to be able to contain
410  * an entire ethernet frame + rx header.
411  */
412 static void
pci_vtnet_rx_callback(int fd __unused,enum ev_type type __unused,void * param)413 pci_vtnet_rx_callback(int fd __unused, enum ev_type type __unused, void *param)
414 {
415 	struct pci_vtnet_softc *sc = param;
416 
417 	pthread_mutex_lock(&sc->rx_mtx);
418 	pci_vtnet_rx(sc);
419 	pthread_mutex_unlock(&sc->rx_mtx);
420 
421 }
422 
423 /* Called on RX kick. */
424 static void
pci_vtnet_ping_rxq(void * vsc,struct vqueue_info * vq)425 pci_vtnet_ping_rxq(void *vsc, struct vqueue_info *vq)
426 {
427 	struct pci_vtnet_softc *sc = vsc;
428 
429 	/*
430 	 * A qnotify means that the rx process can now begin.
431 	 * Enable RX only if features are negotiated.
432 	 */
433 	pthread_mutex_lock(&sc->rx_mtx);
434 	if (!sc->features_negotiated) {
435 		pthread_mutex_unlock(&sc->rx_mtx);
436 		return;
437 	}
438 
439 	vq_kick_disable(vq);
440 	netbe_rx_enable(sc->vsc_be);
441 	pthread_mutex_unlock(&sc->rx_mtx);
442 }
443 
444 /* TX virtqueue processing, called by the TX thread. */
445 static void
pci_vtnet_proctx(struct pci_vtnet_softc * sc,struct vqueue_info * vq)446 pci_vtnet_proctx(struct pci_vtnet_softc *sc, struct vqueue_info *vq)
447 {
448 	struct iovec iov[VTNET_MAXSEGS + 1];
449 	struct iovec *siov = iov;
450 	struct vi_req req;
451 	ssize_t len;
452 	int n;
453 
454 	/*
455 	 * Obtain chain of descriptors. The first descriptor also
456 	 * contains the virtio-net header.
457 	 */
458 	n = vq_getchain(vq, iov, VTNET_MAXSEGS, &req);
459 	assert(n >= 1 && n <= VTNET_MAXSEGS);
460 
461 	if (sc->vhdrlen != sc->be_vhdrlen) {
462 		/*
463 		 * The frontend uses a virtio-net header, but the backend
464 		 * does not. We simply strip the header and ignore it, as
465 		 * it should be zero-filled.
466 		 */
467 		siov = iov_trim_hdr(siov, &n, sc->vhdrlen);
468 	}
469 
470 	if (siov == NULL) {
471 		/* The chain is nonsensical. Just drop it. */
472 		len = 0;
473 	} else {
474 		len = netbe_send(sc->vsc_be, siov, n);
475 		if (len < 0) {
476 			/*
477 			 * If send failed, report that 0 bytes
478 			 * were read.
479 			 */
480 			len = 0;
481 		}
482 	}
483 
484 	/*
485 	 * Return the processed chain to the guest, reporting
486 	 * the number of bytes that we read.
487 	 */
488 	vq_relchain(vq, req.idx, len);
489 }
490 
491 /* Called on TX kick. */
492 static void
pci_vtnet_ping_txq(void * vsc,struct vqueue_info * vq)493 pci_vtnet_ping_txq(void *vsc, struct vqueue_info *vq)
494 {
495 	struct pci_vtnet_softc *sc = vsc;
496 
497 	/*
498 	 * Any ring entries to process?
499 	 */
500 	if (!vq_has_descs(vq))
501 		return;
502 
503 	/* Signal the tx thread for processing */
504 	pthread_mutex_lock(&sc->tx_mtx);
505 	vq_kick_disable(vq);
506 	if (sc->tx_in_progress == 0)
507 		pthread_cond_signal(&sc->tx_cond);
508 	pthread_mutex_unlock(&sc->tx_mtx);
509 }
510 
511 /*
512  * Thread which will handle processing of TX desc
513  */
514 static void *
pci_vtnet_tx_thread(void * param)515 pci_vtnet_tx_thread(void *param)
516 {
517 	struct pci_vtnet_softc *sc = param;
518 	struct vqueue_info *vq;
519 	int error;
520 
521 	vq = &sc->vsc_queues[VTNET_TXQ];
522 
523 	/*
524 	 * Let us wait till the tx queue pointers get initialised &
525 	 * first tx signaled
526 	 */
527 	pthread_mutex_lock(&sc->tx_mtx);
528 	error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
529 	assert(error == 0);
530 
531 	for (;;) {
532 		/* note - tx mutex is locked here */
533 		while (sc->resetting || !vq_has_descs(vq)) {
534 			vq_kick_enable(vq);
535 			if (!sc->resetting && vq_has_descs(vq))
536 				break;
537 
538 			sc->tx_in_progress = 0;
539 			error = pthread_cond_wait(&sc->tx_cond, &sc->tx_mtx);
540 			assert(error == 0);
541 		}
542 		vq_kick_disable(vq);
543 		sc->tx_in_progress = 1;
544 		pthread_mutex_unlock(&sc->tx_mtx);
545 
546 		do {
547 			/*
548 			 * Run through entries, placing them into
549 			 * iovecs and sending when an end-of-packet
550 			 * is found
551 			 */
552 			pci_vtnet_proctx(sc, vq);
553 		} while (vq_has_descs(vq));
554 
555 		/*
556 		 * Generate an interrupt if needed.
557 		 */
558 		vq_endchains(vq, /*used_all_avail=*/1);
559 
560 		pthread_mutex_lock(&sc->tx_mtx);
561 	}
562 #ifndef __FreeBSD__
563 	return (NULL);
564 #endif
565 }
566 
567 #ifdef notyet
568 static void
pci_vtnet_ping_ctlq(void * vsc,struct vqueue_info * vq)569 pci_vtnet_ping_ctlq(void *vsc, struct vqueue_info *vq)
570 {
571 
572 	DPRINTF(("vtnet: control qnotify!"));
573 }
574 #endif
575 
576 static int
pci_vtnet_init(struct pci_devinst * pi,nvlist_t * nvl)577 pci_vtnet_init(struct pci_devinst *pi, nvlist_t *nvl)
578 {
579 	struct pci_vtnet_softc *sc;
580 	const char *value;
581 	char tname[MAXCOMLEN + 1];
582 	unsigned long mtu = ETHERMTU;
583 	int err;
584 
585 	/*
586 	 * Allocate data structures for further virtio initializations.
587 	 * sc also contains a copy of vtnet_vi_consts, since capabilities
588 	 * change depending on the backend.
589 	 */
590 	sc = calloc(1, sizeof(struct pci_vtnet_softc));
591 
592 	sc->vsc_consts = vtnet_vi_consts;
593 	pthread_mutex_init(&sc->vsc_mtx, NULL);
594 
595 	sc->vsc_queues[VTNET_RXQ].vq_qsize = VTNET_RINGSZ;
596 	sc->vsc_queues[VTNET_RXQ].vq_notify = pci_vtnet_ping_rxq;
597 	sc->vsc_queues[VTNET_TXQ].vq_qsize = VTNET_RINGSZ;
598 	sc->vsc_queues[VTNET_TXQ].vq_notify = pci_vtnet_ping_txq;
599 #ifdef notyet
600 	sc->vsc_queues[VTNET_CTLQ].vq_qsize = VTNET_RINGSZ;
601         sc->vsc_queues[VTNET_CTLQ].vq_notify = pci_vtnet_ping_ctlq;
602 #endif
603 
604 	value = get_config_value_node(nvl, "mac");
605 	if (value != NULL) {
606 		err = net_parsemac(value, sc->vsc_config.mac);
607 		if (err) {
608 			free(sc);
609 			return (err);
610 		}
611 	} else
612 		net_genmac(pi, sc->vsc_config.mac);
613 
614 	value = get_config_value_node(nvl, "mtu");
615 	if (value != NULL) {
616 		err = net_parsemtu(value, &mtu);
617 		if (err) {
618 			free(sc);
619 			return (err);
620 		}
621 
622 		if (mtu < VTNET_MIN_MTU || mtu > VTNET_MAX_MTU) {
623 			err = EINVAL;
624 			errno = EINVAL;
625 			free(sc);
626 			return (err);
627 		}
628 		sc->vsc_consts.vc_hv_caps |= VIRTIO_NET_F_MTU;
629 	}
630 	sc->vsc_config.mtu = mtu;
631 
632 	/* Permit interfaces without a configured backend. */
633 	if (get_config_value_node(nvl, "backend") != NULL) {
634 		err = netbe_init(&sc->vsc_be, nvl, pci_vtnet_rx_callback, sc);
635 		if (err) {
636 			free(sc);
637 			return (err);
638 		}
639 #ifndef __FreeBSD__
640 		size_t buflen = sizeof (sc->vsc_config.mac);
641 
642 		err = netbe_get_mac(sc->vsc_be, sc->vsc_config.mac, &buflen);
643 		if (err != 0) {
644 		       free(sc);
645 		       return (err);
646 		}
647 #endif
648 	}
649 
650 	sc->vsc_consts.vc_hv_caps |= VIRTIO_NET_F_MRG_RXBUF |
651 	    netbe_get_cap(sc->vsc_be);
652 
653 	/*
654 	 * Since we do not actually support multiqueue,
655 	 * set the maximum virtqueue pairs to 1.
656 	 */
657 	sc->vsc_config.max_virtqueue_pairs = 1;
658 
659 	/* initialize config space */
660 	pci_set_cfgdata16(pi, PCIR_DEVICE, VIRTIO_DEV_NET);
661 	pci_set_cfgdata16(pi, PCIR_VENDOR, VIRTIO_VENDOR);
662 	pci_set_cfgdata8(pi, PCIR_CLASS, PCIC_NETWORK);
663 	pci_set_cfgdata16(pi, PCIR_SUBDEV_0, VIRTIO_ID_NETWORK);
664 	pci_set_cfgdata16(pi, PCIR_SUBVEND_0, VIRTIO_VENDOR);
665 
666 	/* Link is always up. */
667 	sc->vsc_config.status = 1;
668 
669 	vi_softc_linkup(&sc->vsc_vs, &sc->vsc_consts, sc, pi, sc->vsc_queues);
670 	sc->vsc_vs.vs_mtx = &sc->vsc_mtx;
671 
672 	/* use BAR 1 to map MSI-X table and PBA, if we're using MSI-X */
673 	if (vi_intr_init(&sc->vsc_vs, 1, fbsdrun_virtio_msix())) {
674 		free(sc);
675 		return (1);
676 	}
677 
678 	/* use BAR 0 to map config regs in IO space */
679 	vi_set_io_bar(&sc->vsc_vs, 0);
680 
681 	sc->resetting = 0;
682 
683 	sc->rx_merge = 0;
684 	sc->vhdrlen = sizeof(struct virtio_net_rxhdr) - 2;
685 	pthread_mutex_init(&sc->rx_mtx, NULL);
686 
687 	/*
688 	 * Initialize tx semaphore & spawn TX processing thread.
689 	 * As of now, only one thread for TX desc processing is
690 	 * spawned.
691 	 */
692 	sc->tx_in_progress = 0;
693 	pthread_mutex_init(&sc->tx_mtx, NULL);
694 	pthread_cond_init(&sc->tx_cond, NULL);
695 	pthread_create(&sc->tx_tid, NULL, pci_vtnet_tx_thread, (void *)sc);
696 	snprintf(tname, sizeof(tname), "vtnet-%d:%d tx", pi->pi_slot,
697 	    pi->pi_func);
698 	pthread_set_name_np(sc->tx_tid, tname);
699 
700 	return (0);
701 }
702 
703 static int
pci_vtnet_cfgwrite(void * vsc,int offset,int size,uint32_t value)704 pci_vtnet_cfgwrite(void *vsc, int offset, int size, uint32_t value)
705 {
706 	struct pci_vtnet_softc *sc = vsc;
707 	void *ptr;
708 
709 	if (offset < (int)sizeof(sc->vsc_config.mac)) {
710 		assert(offset + size <= (int)sizeof(sc->vsc_config.mac));
711 		/*
712 		 * The driver is allowed to change the MAC address
713 		 */
714 		ptr = &sc->vsc_config.mac[offset];
715 		memcpy(ptr, &value, size);
716 	} else {
717 		/* silently ignore other writes */
718 		DPRINTF(("vtnet: write to readonly reg %d", offset));
719 	}
720 
721 	return (0);
722 }
723 
724 static int
pci_vtnet_cfgread(void * vsc,int offset,int size,uint32_t * retval)725 pci_vtnet_cfgread(void *vsc, int offset, int size, uint32_t *retval)
726 {
727 	struct pci_vtnet_softc *sc = vsc;
728 	void *ptr;
729 
730 	ptr = (uint8_t *)&sc->vsc_config + offset;
731 	memcpy(retval, ptr, size);
732 	return (0);
733 }
734 
735 static void
pci_vtnet_neg_features(void * vsc,uint64_t negotiated_features)736 pci_vtnet_neg_features(void *vsc, uint64_t negotiated_features)
737 {
738 	struct pci_vtnet_softc *sc = vsc;
739 
740 	sc->vsc_features = negotiated_features;
741 
742 	if (negotiated_features & VIRTIO_NET_F_MRG_RXBUF) {
743 		sc->vhdrlen = sizeof(struct virtio_net_rxhdr);
744 		sc->rx_merge = 1;
745 	} else {
746 		/*
747 		 * Without mergeable rx buffers, virtio-net header is 2
748 		 * bytes shorter than sizeof(struct virtio_net_rxhdr).
749 		 */
750 		sc->vhdrlen = sizeof(struct virtio_net_rxhdr) - 2;
751 		sc->rx_merge = 0;
752 	}
753 
754 	/* Tell the backend to enable some capabilities it has advertised. */
755 	netbe_set_cap(sc->vsc_be, negotiated_features, sc->vhdrlen);
756 	sc->be_vhdrlen = netbe_get_vnet_hdr_len(sc->vsc_be);
757 	assert(sc->be_vhdrlen == 0 || sc->be_vhdrlen == sc->vhdrlen);
758 
759 	pthread_mutex_lock(&sc->rx_mtx);
760 	sc->features_negotiated = true;
761 	pthread_mutex_unlock(&sc->rx_mtx);
762 }
763 
764 static const struct pci_devemu pci_de_vnet = {
765 	.pe_emu = 	"virtio-net",
766 	.pe_init =	pci_vtnet_init,
767 	.pe_legacy_config = netbe_legacy_config,
768 	.pe_barwrite =	vi_pci_write,
769 	.pe_barread =	vi_pci_read,
770 };
771 PCI_EMUL_SET(pci_de_vnet);
772