xref: /freebsd/sys/dev/xen/netfront/netfront.c (revision b626f5a73a48f44a31a200291b141e1da408a2ff)
1 /*-
2  * Copyright (c) 2004-2006 Kip Macy
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 THE AUTHOR AND CONTRIBUTORS ``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 THE AUTHOR 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 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_inet.h"
31 #include "opt_inet6.h"
32 
33 #include <sys/param.h>
34 #include <sys/sockio.h>
35 #include <sys/limits.h>
36 #include <sys/mbuf.h>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/kernel.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 
43 #include <net/if.h>
44 #include <net/if_var.h>
45 #include <net/if_arp.h>
46 #include <net/ethernet.h>
47 #include <net/if_media.h>
48 
49 #include <net/bpf.h>
50 
51 #include <net/if_types.h>
52 
53 #include <netinet/in.h>
54 #include <netinet/ip.h>
55 #include <netinet/if_ether.h>
56 #include <netinet/tcp.h>
57 #include <netinet/tcp_lro.h>
58 
59 #include <vm/vm.h>
60 #include <vm/pmap.h>
61 
62 #include <sys/bus.h>
63 
64 #include <xen/xen-os.h>
65 #include <xen/hypervisor.h>
66 #include <xen/xen_intr.h>
67 #include <xen/gnttab.h>
68 #include <xen/interface/memory.h>
69 #include <xen/interface/io/netif.h>
70 #include <xen/xenbus/xenbusvar.h>
71 
72 #include "xenbus_if.h"
73 
74 /* Features supported by all backends.  TSO and LRO can be negotiated */
75 #define XN_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
76 
77 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
78 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
79 
80 /*
81  * Should the driver do LRO on the RX end
82  *  this can be toggled on the fly, but the
83  *  interface must be reset (down/up) for it
84  *  to take effect.
85  */
86 static int xn_enable_lro = 1;
87 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
88 
89 /**
90  * \brief The maximum allowed data fragments in a single transmit
91  *        request.
92  *
93  * This limit is imposed by the backend driver.  We assume here that
94  * we are dealing with a Linux driver domain and have set our limit
95  * to mirror the Linux MAX_SKB_FRAGS constant.
96  */
97 #define	MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2)
98 
99 #define RX_COPY_THRESHOLD 256
100 
101 #define net_ratelimit() 0
102 
103 struct netfront_info;
104 struct netfront_rx_info;
105 
106 static void xn_txeof(struct netfront_info *);
107 static void xn_rxeof(struct netfront_info *);
108 static void network_alloc_rx_buffers(struct netfront_info *);
109 
110 static void xn_tick_locked(struct netfront_info *);
111 static void xn_tick(void *);
112 
113 static void xn_intr(void *);
114 static inline int xn_count_frags(struct mbuf *m);
115 static int  xn_assemble_tx_request(struct netfront_info *sc,
116 				   struct mbuf *m_head);
117 static void xn_start_locked(struct ifnet *);
118 static void xn_start(struct ifnet *);
119 static int  xn_ioctl(struct ifnet *, u_long, caddr_t);
120 static void xn_ifinit_locked(struct netfront_info *);
121 static void xn_ifinit(void *);
122 static void xn_stop(struct netfront_info *);
123 static void xn_query_features(struct netfront_info *np);
124 static int  xn_configure_features(struct netfront_info *np);
125 #ifdef notyet
126 static void xn_watchdog(struct ifnet *);
127 #endif
128 
129 #ifdef notyet
130 static void netfront_closing(device_t dev);
131 #endif
132 static void netif_free(struct netfront_info *info);
133 static int netfront_detach(device_t dev);
134 
135 static int talk_to_backend(device_t dev, struct netfront_info *info);
136 static int create_netdev(device_t dev);
137 static void netif_disconnect_backend(struct netfront_info *info);
138 static int setup_device(device_t dev, struct netfront_info *info);
139 static void free_ring(int *ref, void *ring_ptr_ref);
140 
141 static int  xn_ifmedia_upd(struct ifnet *ifp);
142 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
143 
144 /* Xenolinux helper functions */
145 int network_connect(struct netfront_info *);
146 
147 static void xn_free_rx_ring(struct netfront_info *);
148 
149 static void xn_free_tx_ring(struct netfront_info *);
150 
151 static int xennet_get_responses(struct netfront_info *np,
152 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
153 	struct mbuf **list);
154 
155 #define virt_to_mfn(x) (vtophys(x) >> PAGE_SHIFT)
156 
157 #define INVALID_P2M_ENTRY (~0UL)
158 
159 /*
160  * Mbuf pointers. We need these to keep track of the virtual addresses
161  * of our mbuf chains since we can only convert from virtual to physical,
162  * not the other way around.  The size must track the free index arrays.
163  */
164 struct xn_chain_data {
165 	struct mbuf    *xn_tx_chain[NET_TX_RING_SIZE+1];
166 	int		xn_tx_chain_cnt;
167 	struct mbuf    *xn_rx_chain[NET_RX_RING_SIZE+1];
168 };
169 
170 struct netfront_stats
171 {
172 	u_long	rx_packets;		/* total packets received	*/
173 	u_long	tx_packets;		/* total packets transmitted	*/
174 	u_long	rx_bytes;		/* total bytes received 	*/
175 	u_long	tx_bytes;		/* total bytes transmitted	*/
176 	u_long	rx_errors;		/* bad packets received		*/
177 	u_long	tx_errors;		/* packet transmit problems	*/
178 };
179 
180 struct netfront_info {
181 	struct ifnet *xn_ifp;
182 	struct lro_ctrl xn_lro;
183 
184 	struct netfront_stats stats;
185 	u_int tx_full;
186 
187 	netif_tx_front_ring_t tx;
188 	netif_rx_front_ring_t rx;
189 
190 	struct mtx   tx_lock;
191 	struct mtx   rx_lock;
192 	struct mtx   sc_lock;
193 
194 	xen_intr_handle_t xen_intr_handle;
195 	u_int carrier;
196 	u_int maxfrags;
197 
198 	/* Receive-ring batched refills. */
199 #define RX_MIN_TARGET 32
200 #define RX_MAX_TARGET NET_RX_RING_SIZE
201 	int rx_min_target;
202 	int rx_max_target;
203 	int rx_target;
204 
205 	grant_ref_t gref_tx_head;
206 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
207 	grant_ref_t gref_rx_head;
208 	grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
209 
210 	device_t		xbdev;
211 	int			tx_ring_ref;
212 	int			rx_ring_ref;
213 	uint8_t			mac[ETHER_ADDR_LEN];
214 	struct xn_chain_data	xn_cdata;	/* mbufs */
215 	struct mbufq		xn_rx_batch;	/* batch queue */
216 
217 	int			xn_if_flags;
218 	struct callout	        xn_stat_ch;
219 
220 	xen_pfn_t		rx_pfn_array[NET_RX_RING_SIZE];
221 	struct ifmedia		sc_media;
222 
223 	bool			xn_resume;
224 };
225 
226 #define rx_mbufs xn_cdata.xn_rx_chain
227 #define tx_mbufs xn_cdata.xn_tx_chain
228 
229 #define XN_RX_LOCK(_sc)           mtx_lock(&(_sc)->rx_lock)
230 #define XN_RX_UNLOCK(_sc)         mtx_unlock(&(_sc)->rx_lock)
231 
232 #define XN_TX_LOCK(_sc)           mtx_lock(&(_sc)->tx_lock)
233 #define XN_TX_UNLOCK(_sc)         mtx_unlock(&(_sc)->tx_lock)
234 
235 #define XN_LOCK(_sc)           mtx_lock(&(_sc)->sc_lock);
236 #define XN_UNLOCK(_sc)         mtx_unlock(&(_sc)->sc_lock);
237 
238 #define XN_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->sc_lock, MA_OWNED);
239 #define XN_RX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->rx_lock, MA_OWNED);
240 #define XN_TX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->tx_lock, MA_OWNED);
241 
242 struct netfront_rx_info {
243 	struct netif_rx_response rx;
244 	struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
245 };
246 
247 #define netfront_carrier_on(netif)	((netif)->carrier = 1)
248 #define netfront_carrier_off(netif)	((netif)->carrier = 0)
249 #define netfront_carrier_ok(netif)	((netif)->carrier)
250 
251 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
252 
253 static inline void
254 add_id_to_freelist(struct mbuf **list, uintptr_t id)
255 {
256 	KASSERT(id != 0,
257 		("%s: the head item (0) must always be free.", __func__));
258 	list[id] = list[0];
259 	list[0]  = (struct mbuf *)id;
260 }
261 
262 static inline unsigned short
263 get_id_from_freelist(struct mbuf **list)
264 {
265 	uintptr_t id;
266 
267 	id = (uintptr_t)list[0];
268 	KASSERT(id != 0,
269 		("%s: the head item (0) must always remain free.", __func__));
270 	list[0] = list[id];
271 	return (id);
272 }
273 
274 static inline int
275 xennet_rxidx(RING_IDX idx)
276 {
277 	return idx & (NET_RX_RING_SIZE - 1);
278 }
279 
280 static inline struct mbuf *
281 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
282 {
283 	int i = xennet_rxidx(ri);
284 	struct mbuf *m;
285 
286 	m = np->rx_mbufs[i];
287 	np->rx_mbufs[i] = NULL;
288 	return (m);
289 }
290 
291 static inline grant_ref_t
292 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
293 {
294 	int i = xennet_rxidx(ri);
295 	grant_ref_t ref = np->grant_rx_ref[i];
296 	KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
297 	np->grant_rx_ref[i] = GRANT_REF_INVALID;
298 	return ref;
299 }
300 
301 #define IPRINTK(fmt, args...) \
302     printf("[XEN] " fmt, ##args)
303 #ifdef INVARIANTS
304 #define WPRINTK(fmt, args...) \
305     printf("[XEN] " fmt, ##args)
306 #else
307 #define WPRINTK(fmt, args...)
308 #endif
309 #ifdef DEBUG
310 #define DPRINTK(fmt, args...) \
311     printf("[XEN] %s: " fmt, __func__, ##args)
312 #else
313 #define DPRINTK(fmt, args...)
314 #endif
315 
316 /**
317  * Read the 'mac' node at the given device's node in the store, and parse that
318  * as colon-separated octets, placing result the given mac array.  mac must be
319  * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
320  * Return 0 on success, or errno on error.
321  */
322 static int
323 xen_net_read_mac(device_t dev, uint8_t mac[])
324 {
325 	int error, i;
326 	char *s, *e, *macstr;
327 	const char *path;
328 
329 	path = xenbus_get_node(dev);
330 	error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
331 	if (error == ENOENT) {
332 		/*
333 		 * Deal with missing mac XenStore nodes on devices with
334 		 * HVM emulation (the 'ioemu' configuration attribute)
335 		 * enabled.
336 		 *
337 		 * The HVM emulator may execute in a stub device model
338 		 * domain which lacks the permission, only given to Dom0,
339 		 * to update the guest's XenStore tree.  For this reason,
340 		 * the HVM emulator doesn't even attempt to write the
341 		 * front-side mac node, even when operating in Dom0.
342 		 * However, there should always be a mac listed in the
343 		 * backend tree.  Fallback to this version if our query
344 		 * of the front side XenStore location doesn't find
345 		 * anything.
346 		 */
347 		path = xenbus_get_otherend_path(dev);
348 		error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
349 	}
350 	if (error != 0) {
351 		xenbus_dev_fatal(dev, error, "parsing %s/mac", path);
352 		return (error);
353 	}
354 
355 	s = macstr;
356 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
357 		mac[i] = strtoul(s, &e, 16);
358 		if (s == e || (e[0] != ':' && e[0] != 0)) {
359 			free(macstr, M_XENBUS);
360 			return (ENOENT);
361 		}
362 		s = &e[1];
363 	}
364 	free(macstr, M_XENBUS);
365 	return (0);
366 }
367 
368 /**
369  * Entry point to this code when a new device is created.  Allocate the basic
370  * structures and the ring buffers for communication with the backend, and
371  * inform the backend of the appropriate details for those.  Switch to
372  * Connected state.
373  */
374 static int
375 netfront_probe(device_t dev)
376 {
377 
378 	if (xen_hvm_domain() && xen_disable_pv_nics != 0)
379 		return (ENXIO);
380 
381 	if (!strcmp(xenbus_get_type(dev), "vif")) {
382 		device_set_desc(dev, "Virtual Network Interface");
383 		return (0);
384 	}
385 
386 	return (ENXIO);
387 }
388 
389 static int
390 netfront_attach(device_t dev)
391 {
392 	int err;
393 
394 	err = create_netdev(dev);
395 	if (err) {
396 		xenbus_dev_fatal(dev, err, "creating netdev");
397 		return (err);
398 	}
399 
400 	SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
401 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
402 	    OID_AUTO, "enable_lro", CTLFLAG_RW,
403 	    &xn_enable_lro, 0, "Large Receive Offload");
404 
405 	return (0);
406 }
407 
408 static int
409 netfront_suspend(device_t dev)
410 {
411 	struct netfront_info *info = device_get_softc(dev);
412 
413 	XN_RX_LOCK(info);
414 	XN_TX_LOCK(info);
415 	netfront_carrier_off(info);
416 	XN_TX_UNLOCK(info);
417 	XN_RX_UNLOCK(info);
418 	return (0);
419 }
420 
421 /**
422  * We are reconnecting to the backend, due to a suspend/resume, or a backend
423  * driver restart.  We tear down our netif structure and recreate it, but
424  * leave the device-layer structures intact so that this is transparent to the
425  * rest of the kernel.
426  */
427 static int
428 netfront_resume(device_t dev)
429 {
430 	struct netfront_info *info = device_get_softc(dev);
431 
432 	info->xn_resume = true;
433 	netif_disconnect_backend(info);
434 	return (0);
435 }
436 
437 /* Common code used when first setting up, and when resuming. */
438 static int
439 talk_to_backend(device_t dev, struct netfront_info *info)
440 {
441 	const char *message;
442 	struct xs_transaction xst;
443 	const char *node = xenbus_get_node(dev);
444 	int err;
445 
446 	err = xen_net_read_mac(dev, info->mac);
447 	if (err) {
448 		xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
449 		goto out;
450 	}
451 
452 	/* Create shared ring, alloc event channel. */
453 	err = setup_device(dev, info);
454 	if (err)
455 		goto out;
456 
457  again:
458 	err = xs_transaction_start(&xst);
459 	if (err) {
460 		xenbus_dev_fatal(dev, err, "starting transaction");
461 		goto destroy_ring;
462 	}
463 	err = xs_printf(xst, node, "tx-ring-ref","%u",
464 			info->tx_ring_ref);
465 	if (err) {
466 		message = "writing tx ring-ref";
467 		goto abort_transaction;
468 	}
469 	err = xs_printf(xst, node, "rx-ring-ref","%u",
470 			info->rx_ring_ref);
471 	if (err) {
472 		message = "writing rx ring-ref";
473 		goto abort_transaction;
474 	}
475 	err = xs_printf(xst, node,
476 			"event-channel", "%u",
477 			xen_intr_port(info->xen_intr_handle));
478 	if (err) {
479 		message = "writing event-channel";
480 		goto abort_transaction;
481 	}
482 	err = xs_printf(xst, node, "request-rx-copy", "%u", 1);
483 	if (err) {
484 		message = "writing request-rx-copy";
485 		goto abort_transaction;
486 	}
487 	err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
488 	if (err) {
489 		message = "writing feature-rx-notify";
490 		goto abort_transaction;
491 	}
492 	err = xs_printf(xst, node, "feature-sg", "%d", 1);
493 	if (err) {
494 		message = "writing feature-sg";
495 		goto abort_transaction;
496 	}
497 	err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
498 	if (err) {
499 		message = "writing feature-gso-tcpv4";
500 		goto abort_transaction;
501 	}
502 
503 	err = xs_transaction_end(xst, 0);
504 	if (err) {
505 		if (err == EAGAIN)
506 			goto again;
507 		xenbus_dev_fatal(dev, err, "completing transaction");
508 		goto destroy_ring;
509 	}
510 
511 	return 0;
512 
513  abort_transaction:
514 	xs_transaction_end(xst, 1);
515 	xenbus_dev_fatal(dev, err, "%s", message);
516  destroy_ring:
517 	netif_free(info);
518  out:
519 	return err;
520 }
521 
522 static int
523 setup_device(device_t dev, struct netfront_info *info)
524 {
525 	netif_tx_sring_t *txs;
526 	netif_rx_sring_t *rxs;
527 	int error;
528 
529 	info->tx_ring_ref = GRANT_REF_INVALID;
530 	info->rx_ring_ref = GRANT_REF_INVALID;
531 	info->rx.sring = NULL;
532 	info->tx.sring = NULL;
533 
534 	txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
535 	if (!txs) {
536 		error = ENOMEM;
537 		xenbus_dev_fatal(dev, error, "allocating tx ring page");
538 		goto fail;
539 	}
540 	SHARED_RING_INIT(txs);
541 	FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
542 	error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
543 	if (error)
544 		goto fail;
545 
546 	rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
547 	if (!rxs) {
548 		error = ENOMEM;
549 		xenbus_dev_fatal(dev, error, "allocating rx ring page");
550 		goto fail;
551 	}
552 	SHARED_RING_INIT(rxs);
553 	FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
554 
555 	error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
556 	if (error)
557 		goto fail;
558 
559 	error = xen_intr_alloc_and_bind_local_port(dev,
560 	    xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info,
561 	    INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle);
562 
563 	if (error) {
564 		xenbus_dev_fatal(dev, error,
565 				 "xen_intr_alloc_and_bind_local_port failed");
566 		goto fail;
567 	}
568 
569 	return (0);
570 
571  fail:
572 	netif_free(info);
573 	return (error);
574 }
575 
576 #ifdef INET
577 /**
578  * If this interface has an ipv4 address, send an arp for it. This
579  * helps to get the network going again after migrating hosts.
580  */
581 static void
582 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
583 {
584 	struct ifnet *ifp;
585 	struct ifaddr *ifa;
586 
587 	ifp = info->xn_ifp;
588 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
589 		if (ifa->ifa_addr->sa_family == AF_INET) {
590 			arp_ifinit(ifp, ifa);
591 		}
592 	}
593 }
594 #endif
595 
596 /**
597  * Callback received when the backend's state changes.
598  */
599 static void
600 netfront_backend_changed(device_t dev, XenbusState newstate)
601 {
602 	struct netfront_info *sc = device_get_softc(dev);
603 
604 	DPRINTK("newstate=%d\n", newstate);
605 
606 	switch (newstate) {
607 	case XenbusStateInitialising:
608 	case XenbusStateInitialised:
609 	case XenbusStateUnknown:
610 	case XenbusStateClosed:
611 	case XenbusStateReconfigured:
612 	case XenbusStateReconfiguring:
613 		break;
614 	case XenbusStateInitWait:
615 		if (xenbus_get_state(dev) != XenbusStateInitialising)
616 			break;
617 		if (network_connect(sc) != 0)
618 			break;
619 		xenbus_set_state(dev, XenbusStateConnected);
620 		break;
621 	case XenbusStateClosing:
622 		xenbus_set_state(dev, XenbusStateClosed);
623 		break;
624 	case XenbusStateConnected:
625 #ifdef INET
626 		netfront_send_fake_arp(dev, sc);
627 #endif
628 		break;
629 	}
630 }
631 
632 static void
633 xn_free_rx_ring(struct netfront_info *sc)
634 {
635 #if 0
636 	int i;
637 
638 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
639 		if (sc->xn_cdata.rx_mbufs[i] != NULL) {
640 			m_freem(sc->rx_mbufs[i]);
641 			sc->rx_mbufs[i] = NULL;
642 		}
643 	}
644 
645 	sc->rx.rsp_cons = 0;
646 	sc->xn_rx_if->req_prod = 0;
647 	sc->xn_rx_if->event = sc->rx.rsp_cons ;
648 #endif
649 }
650 
651 static void
652 xn_free_tx_ring(struct netfront_info *sc)
653 {
654 #if 0
655 	int i;
656 
657 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
658 		if (sc->tx_mbufs[i] != NULL) {
659 			m_freem(sc->tx_mbufs[i]);
660 			sc->xn_cdata.xn_tx_chain[i] = NULL;
661 		}
662 	}
663 
664 	return;
665 #endif
666 }
667 
668 /**
669  * \brief Verify that there is sufficient space in the Tx ring
670  *        buffer for a maximally sized request to be enqueued.
671  *
672  * A transmit request requires a transmit descriptor for each packet
673  * fragment, plus up to 2 entries for "options" (e.g. TSO).
674  */
675 static inline int
676 xn_tx_slot_available(struct netfront_info *np)
677 {
678 	return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
679 }
680 
681 static void
682 netif_release_tx_bufs(struct netfront_info *np)
683 {
684 	int i;
685 
686 	for (i = 1; i <= NET_TX_RING_SIZE; i++) {
687 		struct mbuf *m;
688 
689 		m = np->tx_mbufs[i];
690 
691 		/*
692 		 * We assume that no kernel addresses are
693 		 * less than NET_TX_RING_SIZE.  Any entry
694 		 * in the table that is below this number
695 		 * must be an index from free-list tracking.
696 		 */
697 		if (((uintptr_t)m) <= NET_TX_RING_SIZE)
698 			continue;
699 		gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
700 		gnttab_release_grant_reference(&np->gref_tx_head,
701 		    np->grant_tx_ref[i]);
702 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
703 		add_id_to_freelist(np->tx_mbufs, i);
704 		np->xn_cdata.xn_tx_chain_cnt--;
705 		if (np->xn_cdata.xn_tx_chain_cnt < 0) {
706 			panic("%s: tx_chain_cnt must be >= 0", __func__);
707 		}
708 		m_free(m);
709 	}
710 }
711 
712 static void
713 network_alloc_rx_buffers(struct netfront_info *sc)
714 {
715 	int otherend_id = xenbus_get_otherend_id(sc->xbdev);
716 	unsigned short id;
717 	struct mbuf *m_new;
718 	int i, batch_target, notify;
719 	RING_IDX req_prod;
720 	grant_ref_t ref;
721 	netif_rx_request_t *req;
722 	vm_offset_t vaddr;
723 	u_long pfn;
724 
725 	req_prod = sc->rx.req_prod_pvt;
726 
727 	if (__predict_false(sc->carrier == 0))
728 		return;
729 
730 	/*
731 	 * Allocate mbufs greedily, even though we batch updates to the
732 	 * receive ring. This creates a less bursty demand on the memory
733 	 * allocator, and so should reduce the chance of failed allocation
734 	 * requests both for ourself and for other kernel subsystems.
735 	 *
736 	 * Here we attempt to maintain rx_target buffers in flight, counting
737 	 * buffers that we have yet to process in the receive ring.
738 	 */
739 	batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
740 	for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
741 		m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE);
742 		if (m_new == NULL) {
743 			if (i != 0)
744 				goto refill;
745 			/*
746 			 * XXX set timer
747 			 */
748 			break;
749 		}
750 		m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
751 
752 		/* queue the mbufs allocated */
753 		(void )mbufq_enqueue(&sc->xn_rx_batch, m_new);
754 	}
755 
756 	/*
757 	 * If we've allocated at least half of our target number of entries,
758 	 * submit them to the backend - we have enough to make the overhead
759 	 * of submission worthwhile.  Otherwise wait for more mbufs and
760 	 * request entries to become available.
761 	 */
762 	if (i < (sc->rx_target/2)) {
763 		if (req_prod >sc->rx.sring->req_prod)
764 			goto push;
765 		return;
766 	}
767 
768 	/*
769 	 * Double floating fill target if we risked having the backend
770 	 * run out of empty buffers for receive traffic.  We define "running
771 	 * low" as having less than a fourth of our target buffers free
772 	 * at the time we refilled the queue.
773 	 */
774 	if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
775 		sc->rx_target *= 2;
776 		if (sc->rx_target > sc->rx_max_target)
777 			sc->rx_target = sc->rx_max_target;
778 	}
779 
780 refill:
781 	for (i = 0; ; i++) {
782 		if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
783 			break;
784 
785 		m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
786 				vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
787 
788 		id = xennet_rxidx(req_prod + i);
789 
790 		KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
791 		sc->rx_mbufs[id] = m_new;
792 
793 		ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
794 		KASSERT(ref != GNTTAB_LIST_END,
795 			("reserved grant references exhuasted"));
796 		sc->grant_rx_ref[id] = ref;
797 
798 		vaddr = mtod(m_new, vm_offset_t);
799 		pfn = vtophys(vaddr) >> PAGE_SHIFT;
800 		req = RING_GET_REQUEST(&sc->rx, req_prod + i);
801 
802 		gnttab_grant_foreign_access_ref(ref, otherend_id, pfn, 0);
803 		req->id = id;
804 		req->gref = ref;
805 
806 		sc->rx_pfn_array[i] =
807 		    vtophys(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
808 	}
809 
810 	KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
811 	KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
812 	/*
813 	 * We may have allocated buffers which have entries outstanding
814 	 * in the page * update queue -- make sure we flush those first!
815 	 */
816 	wmb();
817 
818 	/* Above is a suitable barrier to ensure backend will see requests. */
819 	sc->rx.req_prod_pvt = req_prod + i;
820 push:
821 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
822 	if (notify)
823 		xen_intr_signal(sc->xen_intr_handle);
824 }
825 
826 static void
827 xn_rxeof(struct netfront_info *np)
828 {
829 	struct ifnet *ifp;
830 #if (defined(INET) || defined(INET6))
831 	struct lro_ctrl *lro = &np->xn_lro;
832 	struct lro_entry *queued;
833 #endif
834 	struct netfront_rx_info rinfo;
835 	struct netif_rx_response *rx = &rinfo.rx;
836 	struct netif_extra_info *extras = rinfo.extras;
837 	RING_IDX i, rp;
838 	struct mbuf *m;
839 	struct mbufq rxq, errq;
840 	int err, work_to_do;
841 
842 	do {
843 		XN_RX_LOCK_ASSERT(np);
844 		if (!netfront_carrier_ok(np))
845 			return;
846 
847 		/* XXX: there should be some sane limit. */
848 		mbufq_init(&errq, INT_MAX);
849 		mbufq_init(&rxq, INT_MAX);
850 
851 		ifp = np->xn_ifp;
852 
853 		rp = np->rx.sring->rsp_prod;
854 		rmb();	/* Ensure we see queued responses up to 'rp'. */
855 
856 		i = np->rx.rsp_cons;
857 		while ((i != rp)) {
858 			memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
859 			memset(extras, 0, sizeof(rinfo.extras));
860 
861 			m = NULL;
862 			err = xennet_get_responses(np, &rinfo, rp, &i, &m);
863 
864 			if (__predict_false(err)) {
865 				if (m)
866 					(void )mbufq_enqueue(&errq, m);
867 				np->stats.rx_errors++;
868 				continue;
869 			}
870 
871 			m->m_pkthdr.rcvif = ifp;
872 			if ( rx->flags & NETRXF_data_validated ) {
873 				/* Tell the stack the checksums are okay */
874 				/*
875 				 * XXX this isn't necessarily the case - need to add
876 				 * check
877 				 */
878 
879 				m->m_pkthdr.csum_flags |=
880 					(CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
881 					    | CSUM_PSEUDO_HDR);
882 				m->m_pkthdr.csum_data = 0xffff;
883 			}
884 
885 			np->stats.rx_packets++;
886 			np->stats.rx_bytes += m->m_pkthdr.len;
887 
888 			(void )mbufq_enqueue(&rxq, m);
889 			np->rx.rsp_cons = i;
890 		}
891 
892 		mbufq_drain(&errq);
893 
894 		/*
895 		 * Process all the mbufs after the remapping is complete.
896 		 * Break the mbuf chain first though.
897 		 */
898 		while ((m = mbufq_dequeue(&rxq)) != NULL) {
899 			if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
900 
901 			/*
902 			 * Do we really need to drop the rx lock?
903 			 */
904 			XN_RX_UNLOCK(np);
905 #if (defined(INET) || defined(INET6))
906 			/* Use LRO if possible */
907 			if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
908 			    lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
909 				/*
910 				 * If LRO fails, pass up to the stack
911 				 * directly.
912 				 */
913 				(*ifp->if_input)(ifp, m);
914 			}
915 #else
916 			(*ifp->if_input)(ifp, m);
917 #endif
918 			XN_RX_LOCK(np);
919 		}
920 
921 		np->rx.rsp_cons = i;
922 
923 #if (defined(INET) || defined(INET6))
924 		/*
925 		 * Flush any outstanding LRO work
926 		 */
927 		while (!SLIST_EMPTY(&lro->lro_active)) {
928 			queued = SLIST_FIRST(&lro->lro_active);
929 			SLIST_REMOVE_HEAD(&lro->lro_active, next);
930 			tcp_lro_flush(lro, queued);
931 		}
932 #endif
933 
934 #if 0
935 		/* If we get a callback with very few responses, reduce fill target. */
936 		/* NB. Note exponential increase, linear decrease. */
937 		if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
938 			((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
939 			np->rx_target = np->rx_min_target;
940 #endif
941 
942 		network_alloc_rx_buffers(np);
943 
944 		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
945 	} while (work_to_do);
946 }
947 
948 static void
949 xn_txeof(struct netfront_info *np)
950 {
951 	RING_IDX i, prod;
952 	unsigned short id;
953 	struct ifnet *ifp;
954 	netif_tx_response_t *txr;
955 	struct mbuf *m;
956 
957 	XN_TX_LOCK_ASSERT(np);
958 
959 	if (!netfront_carrier_ok(np))
960 		return;
961 
962 	ifp = np->xn_ifp;
963 
964 	do {
965 		prod = np->tx.sring->rsp_prod;
966 		rmb(); /* Ensure we see responses up to 'rp'. */
967 
968 		for (i = np->tx.rsp_cons; i != prod; i++) {
969 			txr = RING_GET_RESPONSE(&np->tx, i);
970 			if (txr->status == NETIF_RSP_NULL)
971 				continue;
972 
973 			if (txr->status != NETIF_RSP_OKAY) {
974 				printf("%s: WARNING: response is %d!\n",
975 				       __func__, txr->status);
976 			}
977 			id = txr->id;
978 			m = np->tx_mbufs[id];
979 			KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
980 			KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
981 				("mbuf already on the free list, but we're "
982 				"trying to free it again!"));
983 			M_ASSERTVALID(m);
984 
985 			/*
986 			 * Increment packet count if this is the last
987 			 * mbuf of the chain.
988 			 */
989 			if (!m->m_next)
990 				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
991 			if (__predict_false(gnttab_query_foreign_access(
992 			    np->grant_tx_ref[id]) != 0)) {
993 				panic("%s: grant id %u still in use by the "
994 				    "backend", __func__, id);
995 			}
996 			gnttab_end_foreign_access_ref(
997 				np->grant_tx_ref[id]);
998 			gnttab_release_grant_reference(
999 				&np->gref_tx_head, np->grant_tx_ref[id]);
1000 			np->grant_tx_ref[id] = GRANT_REF_INVALID;
1001 
1002 			np->tx_mbufs[id] = NULL;
1003 			add_id_to_freelist(np->tx_mbufs, id);
1004 			np->xn_cdata.xn_tx_chain_cnt--;
1005 			m_free(m);
1006 			/* Only mark the queue active if we've freed up at least one slot to try */
1007 			ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1008 		}
1009 		np->tx.rsp_cons = prod;
1010 
1011 		/*
1012 		 * Set a new event, then check for race with update of
1013 		 * tx_cons. Note that it is essential to schedule a
1014 		 * callback, no matter how few buffers are pending. Even if
1015 		 * there is space in the transmit ring, higher layers may
1016 		 * be blocked because too much data is outstanding: in such
1017 		 * cases notification from Xen is likely to be the only kick
1018 		 * that we'll get.
1019 		 */
1020 		np->tx.sring->rsp_event =
1021 		    prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1022 
1023 		mb();
1024 	} while (prod != np->tx.sring->rsp_prod);
1025 
1026 	if (np->tx_full &&
1027 	    ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1028 		np->tx_full = 0;
1029 #if 0
1030 		if (np->user_state == UST_OPEN)
1031 			netif_wake_queue(dev);
1032 #endif
1033 	}
1034 }
1035 
1036 static void
1037 xn_intr(void *xsc)
1038 {
1039 	struct netfront_info *np = xsc;
1040 	struct ifnet *ifp = np->xn_ifp;
1041 
1042 #if 0
1043 	if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1044 	    likely(netfront_carrier_ok(np)) &&
1045 	    ifp->if_drv_flags & IFF_DRV_RUNNING))
1046 		return;
1047 #endif
1048 	if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1049 		XN_TX_LOCK(np);
1050 		xn_txeof(np);
1051 		XN_TX_UNLOCK(np);
1052 	}
1053 
1054 	XN_RX_LOCK(np);
1055 	xn_rxeof(np);
1056 	XN_RX_UNLOCK(np);
1057 
1058 	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1059 	    !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1060 		xn_start(ifp);
1061 }
1062 
1063 static void
1064 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1065 	grant_ref_t ref)
1066 {
1067 	int new = xennet_rxidx(np->rx.req_prod_pvt);
1068 
1069 	KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1070 	np->rx_mbufs[new] = m;
1071 	np->grant_rx_ref[new] = ref;
1072 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1073 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1074 	np->rx.req_prod_pvt++;
1075 }
1076 
1077 static int
1078 xennet_get_extras(struct netfront_info *np,
1079     struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1080 {
1081 	struct netif_extra_info *extra;
1082 
1083 	int err = 0;
1084 
1085 	do {
1086 		struct mbuf *m;
1087 		grant_ref_t ref;
1088 
1089 		if (__predict_false(*cons + 1 == rp)) {
1090 #if 0
1091 			if (net_ratelimit())
1092 				WPRINTK("Missing extra info\n");
1093 #endif
1094 			err = EINVAL;
1095 			break;
1096 		}
1097 
1098 		extra = (struct netif_extra_info *)
1099 		RING_GET_RESPONSE(&np->rx, ++(*cons));
1100 
1101 		if (__predict_false(!extra->type ||
1102 			extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1103 #if 0
1104 			if (net_ratelimit())
1105 				WPRINTK("Invalid extra type: %d\n",
1106 					extra->type);
1107 #endif
1108 			err = EINVAL;
1109 		} else {
1110 			memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1111 		}
1112 
1113 		m = xennet_get_rx_mbuf(np, *cons);
1114 		ref = xennet_get_rx_ref(np, *cons);
1115 		xennet_move_rx_slot(np, m, ref);
1116 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1117 
1118 	return err;
1119 }
1120 
1121 static int
1122 xennet_get_responses(struct netfront_info *np,
1123 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1124 	struct mbuf  **list)
1125 {
1126 	struct netif_rx_response *rx = &rinfo->rx;
1127 	struct netif_extra_info *extras = rinfo->extras;
1128 	struct mbuf *m, *m0, *m_prev;
1129 	grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1130 	RING_IDX ref_cons = *cons;
1131 	int frags = 1;
1132 	int err = 0;
1133 	u_long ret;
1134 
1135 	m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1136 
1137 	if (rx->flags & NETRXF_extra_info) {
1138 		err = xennet_get_extras(np, extras, rp, cons);
1139 	}
1140 
1141 	if (m0 != NULL) {
1142 		m0->m_pkthdr.len = 0;
1143 		m0->m_next = NULL;
1144 	}
1145 
1146 	for (;;) {
1147 #if 0
1148 		DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1149 			rx->status, rx->offset, frags);
1150 #endif
1151 		if (__predict_false(rx->status < 0 ||
1152 			rx->offset + rx->status > PAGE_SIZE)) {
1153 
1154 #if 0
1155 			if (net_ratelimit())
1156 				WPRINTK("rx->offset: %x, size: %u\n",
1157 					rx->offset, rx->status);
1158 #endif
1159 			xennet_move_rx_slot(np, m, ref);
1160 			if (m0 == m)
1161 				m0 = NULL;
1162 			m = NULL;
1163 			err = EINVAL;
1164 			goto next_skip_queue;
1165 		}
1166 
1167 		/*
1168 		 * This definitely indicates a bug, either in this driver or in
1169 		 * the backend driver. In future this should flag the bad
1170 		 * situation to the system controller to reboot the backed.
1171 		 */
1172 		if (ref == GRANT_REF_INVALID) {
1173 
1174 #if 0
1175 			if (net_ratelimit())
1176 				WPRINTK("Bad rx response id %d.\n", rx->id);
1177 #endif
1178 			printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1179 			err = EINVAL;
1180 			goto next;
1181 		}
1182 
1183 		ret = gnttab_end_foreign_access_ref(ref);
1184 		KASSERT(ret, ("Unable to end access to grant references"));
1185 
1186 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1187 
1188 next:
1189 		if (m == NULL)
1190 			break;
1191 
1192 		m->m_len = rx->status;
1193 		m->m_data += rx->offset;
1194 		m0->m_pkthdr.len += rx->status;
1195 
1196 next_skip_queue:
1197 		if (!(rx->flags & NETRXF_more_data))
1198 			break;
1199 
1200 		if (*cons + frags == rp) {
1201 			if (net_ratelimit())
1202 				WPRINTK("Need more frags\n");
1203 			err = ENOENT;
1204 			printf("%s: cons %u frags %u rp %u, not enough frags\n",
1205 			       __func__, *cons, frags, rp);
1206 			break;
1207 		}
1208 		/*
1209 		 * Note that m can be NULL, if rx->status < 0 or if
1210 		 * rx->offset + rx->status > PAGE_SIZE above.
1211 		 */
1212 		m_prev = m;
1213 
1214 		rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1215 		m = xennet_get_rx_mbuf(np, *cons + frags);
1216 
1217 		/*
1218 		 * m_prev == NULL can happen if rx->status < 0 or if
1219 		 * rx->offset + * rx->status > PAGE_SIZE above.
1220 		 */
1221 		if (m_prev != NULL)
1222 			m_prev->m_next = m;
1223 
1224 		/*
1225 		 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1226 		 * rx->status > PAGE_SIZE above.
1227 		 */
1228 		if (m0 == NULL)
1229 			m0 = m;
1230 		m->m_next = NULL;
1231 		ref = xennet_get_rx_ref(np, *cons + frags);
1232 		ref_cons = *cons + frags;
1233 		frags++;
1234 	}
1235 	*list = m0;
1236 	*cons += frags;
1237 
1238 	return (err);
1239 }
1240 
1241 static void
1242 xn_tick_locked(struct netfront_info *sc)
1243 {
1244 	XN_RX_LOCK_ASSERT(sc);
1245 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1246 
1247 	/* XXX placeholder for printing debug information */
1248 }
1249 
1250 static void
1251 xn_tick(void *xsc)
1252 {
1253 	struct netfront_info *sc;
1254 
1255 	sc = xsc;
1256 	XN_RX_LOCK(sc);
1257 	xn_tick_locked(sc);
1258 	XN_RX_UNLOCK(sc);
1259 }
1260 
1261 /**
1262  * \brief Count the number of fragments in an mbuf chain.
1263  *
1264  * Surprisingly, there isn't an M* macro for this.
1265  */
1266 static inline int
1267 xn_count_frags(struct mbuf *m)
1268 {
1269 	int nfrags;
1270 
1271 	for (nfrags = 0; m != NULL; m = m->m_next)
1272 		nfrags++;
1273 
1274 	return (nfrags);
1275 }
1276 
1277 /**
1278  * Given an mbuf chain, make sure we have enough room and then push
1279  * it onto the transmit ring.
1280  */
1281 static int
1282 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1283 {
1284 	struct ifnet *ifp;
1285 	struct mbuf *m;
1286 	u_int nfrags;
1287 	int otherend_id;
1288 
1289 	ifp = sc->xn_ifp;
1290 
1291 	/**
1292 	 * Defragment the mbuf if necessary.
1293 	 */
1294 	nfrags = xn_count_frags(m_head);
1295 
1296 	/*
1297 	 * Check to see whether this request is longer than netback
1298 	 * can handle, and try to defrag it.
1299 	 */
1300 	/**
1301 	 * It is a bit lame, but the netback driver in Linux can't
1302 	 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1303 	 * the Linux network stack.
1304 	 */
1305 	if (nfrags > sc->maxfrags) {
1306 		m = m_defrag(m_head, M_NOWAIT);
1307 		if (!m) {
1308 			/*
1309 			 * Defrag failed, so free the mbuf and
1310 			 * therefore drop the packet.
1311 			 */
1312 			m_freem(m_head);
1313 			return (EMSGSIZE);
1314 		}
1315 		m_head = m;
1316 	}
1317 
1318 	/* Determine how many fragments now exist */
1319 	nfrags = xn_count_frags(m_head);
1320 
1321 	/*
1322 	 * Check to see whether the defragmented packet has too many
1323 	 * segments for the Linux netback driver.
1324 	 */
1325 	/**
1326 	 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1327 	 * of mbufs longer than Linux can handle.  Make sure we don't
1328 	 * pass a too-long chain over to the other side by dropping the
1329 	 * packet.  It doesn't look like there is currently a way to
1330 	 * tell the TCP stack to generate a shorter chain of packets.
1331 	 */
1332 	if (nfrags > MAX_TX_REQ_FRAGS) {
1333 #ifdef DEBUG
1334 		printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1335 		       "won't be able to handle it, dropping\n",
1336 		       __func__, nfrags, MAX_TX_REQ_FRAGS);
1337 #endif
1338 		m_freem(m_head);
1339 		return (EMSGSIZE);
1340 	}
1341 
1342 	/*
1343 	 * This check should be redundant.  We've already verified that we
1344 	 * have enough slots in the ring to handle a packet of maximum
1345 	 * size, and that our packet is less than the maximum size.  Keep
1346 	 * it in here as an assert for now just to make certain that
1347 	 * xn_tx_chain_cnt is accurate.
1348 	 */
1349 	KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1350 		("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1351 		 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1352                     (int) nfrags, (int) NET_TX_RING_SIZE));
1353 
1354 	/*
1355 	 * Start packing the mbufs in this chain into
1356 	 * the fragment pointers. Stop when we run out
1357 	 * of fragments or hit the end of the mbuf chain.
1358 	 */
1359 	m = m_head;
1360 	otherend_id = xenbus_get_otherend_id(sc->xbdev);
1361 	for (m = m_head; m; m = m->m_next) {
1362 		netif_tx_request_t *tx;
1363 		uintptr_t id;
1364 		grant_ref_t ref;
1365 		u_long mfn; /* XXX Wrong type? */
1366 
1367 		tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1368 		id = get_id_from_freelist(sc->tx_mbufs);
1369 		if (id == 0)
1370 			panic("%s: was allocated the freelist head!\n",
1371 			    __func__);
1372 		sc->xn_cdata.xn_tx_chain_cnt++;
1373 		if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1374 			panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1375 			    __func__);
1376 		sc->tx_mbufs[id] = m;
1377 		tx->id = id;
1378 		ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1379 		KASSERT((short)ref >= 0, ("Negative ref"));
1380 		mfn = virt_to_mfn(mtod(m, vm_offset_t));
1381 		gnttab_grant_foreign_access_ref(ref, otherend_id,
1382 		    mfn, GNTMAP_readonly);
1383 		tx->gref = sc->grant_tx_ref[id] = ref;
1384 		tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1385 		tx->flags = 0;
1386 		if (m == m_head) {
1387 			/*
1388 			 * The first fragment has the entire packet
1389 			 * size, subsequent fragments have just the
1390 			 * fragment size. The backend works out the
1391 			 * true size of the first fragment by
1392 			 * subtracting the sizes of the other
1393 			 * fragments.
1394 			 */
1395 			tx->size = m->m_pkthdr.len;
1396 
1397 			/*
1398 			 * The first fragment contains the checksum flags
1399 			 * and is optionally followed by extra data for
1400 			 * TSO etc.
1401 			 */
1402 			/**
1403 			 * CSUM_TSO requires checksum offloading.
1404 			 * Some versions of FreeBSD fail to
1405 			 * set CSUM_TCP in the CSUM_TSO case,
1406 			 * so we have to test for CSUM_TSO
1407 			 * explicitly.
1408 			 */
1409 			if (m->m_pkthdr.csum_flags
1410 			    & (CSUM_DELAY_DATA | CSUM_TSO)) {
1411 				tx->flags |= (NETTXF_csum_blank
1412 				    | NETTXF_data_validated);
1413 			}
1414 			if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1415 				struct netif_extra_info *gso =
1416 					(struct netif_extra_info *)
1417 					RING_GET_REQUEST(&sc->tx,
1418 							 ++sc->tx.req_prod_pvt);
1419 
1420 				tx->flags |= NETTXF_extra_info;
1421 
1422 				gso->u.gso.size = m->m_pkthdr.tso_segsz;
1423 				gso->u.gso.type =
1424 					XEN_NETIF_GSO_TYPE_TCPV4;
1425 				gso->u.gso.pad = 0;
1426 				gso->u.gso.features = 0;
1427 
1428 				gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1429 				gso->flags = 0;
1430 			}
1431 		} else {
1432 			tx->size = m->m_len;
1433 		}
1434 		if (m->m_next)
1435 			tx->flags |= NETTXF_more_data;
1436 
1437 		sc->tx.req_prod_pvt++;
1438 	}
1439 	BPF_MTAP(ifp, m_head);
1440 
1441 	sc->stats.tx_bytes += m_head->m_pkthdr.len;
1442 	sc->stats.tx_packets++;
1443 
1444 	return (0);
1445 }
1446 
1447 static void
1448 xn_start_locked(struct ifnet *ifp)
1449 {
1450 	struct netfront_info *sc;
1451 	struct mbuf *m_head;
1452 	int notify;
1453 
1454 	sc = ifp->if_softc;
1455 
1456 	if (!netfront_carrier_ok(sc))
1457 		return;
1458 
1459 	/*
1460 	 * While we have enough transmit slots available for at least one
1461 	 * maximum-sized packet, pull mbufs off the queue and put them on
1462 	 * the transmit ring.
1463 	 */
1464 	while (xn_tx_slot_available(sc)) {
1465 		IF_DEQUEUE(&ifp->if_snd, m_head);
1466 		if (m_head == NULL)
1467 			break;
1468 
1469 		if (xn_assemble_tx_request(sc, m_head) != 0)
1470 			break;
1471 	}
1472 
1473 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1474 	if (notify)
1475 		xen_intr_signal(sc->xen_intr_handle);
1476 
1477 	if (RING_FULL(&sc->tx)) {
1478 		sc->tx_full = 1;
1479 #if 0
1480 		netif_stop_queue(dev);
1481 #endif
1482 	}
1483 }
1484 
1485 static void
1486 xn_start(struct ifnet *ifp)
1487 {
1488 	struct netfront_info *sc;
1489 	sc = ifp->if_softc;
1490 	XN_TX_LOCK(sc);
1491 	xn_start_locked(ifp);
1492 	XN_TX_UNLOCK(sc);
1493 }
1494 
1495 /* equivalent of network_open() in Linux */
1496 static void
1497 xn_ifinit_locked(struct netfront_info *sc)
1498 {
1499 	struct ifnet *ifp;
1500 
1501 	XN_LOCK_ASSERT(sc);
1502 
1503 	ifp = sc->xn_ifp;
1504 
1505 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1506 		return;
1507 
1508 	xn_stop(sc);
1509 
1510 	network_alloc_rx_buffers(sc);
1511 	sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1512 
1513 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1514 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1515 	if_link_state_change(ifp, LINK_STATE_UP);
1516 
1517 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1518 }
1519 
1520 static void
1521 xn_ifinit(void *xsc)
1522 {
1523 	struct netfront_info *sc = xsc;
1524 
1525 	XN_LOCK(sc);
1526 	xn_ifinit_locked(sc);
1527 	XN_UNLOCK(sc);
1528 }
1529 
1530 static int
1531 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1532 {
1533 	struct netfront_info *sc = ifp->if_softc;
1534 	struct ifreq *ifr = (struct ifreq *) data;
1535 #ifdef INET
1536 	struct ifaddr *ifa = (struct ifaddr *)data;
1537 #endif
1538 
1539 	int mask, error = 0;
1540 	switch(cmd) {
1541 	case SIOCSIFADDR:
1542 #ifdef INET
1543 		XN_LOCK(sc);
1544 		if (ifa->ifa_addr->sa_family == AF_INET) {
1545 			ifp->if_flags |= IFF_UP;
1546 			if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1547 				xn_ifinit_locked(sc);
1548 			arp_ifinit(ifp, ifa);
1549 			XN_UNLOCK(sc);
1550 		} else {
1551 			XN_UNLOCK(sc);
1552 #endif
1553 			error = ether_ioctl(ifp, cmd, data);
1554 #ifdef INET
1555 		}
1556 #endif
1557 		break;
1558 	case SIOCSIFMTU:
1559 		/* XXX can we alter the MTU on a VN ?*/
1560 #ifdef notyet
1561 		if (ifr->ifr_mtu > XN_JUMBO_MTU)
1562 			error = EINVAL;
1563 		else
1564 #endif
1565 		{
1566 			ifp->if_mtu = ifr->ifr_mtu;
1567 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1568 			xn_ifinit(sc);
1569 		}
1570 		break;
1571 	case SIOCSIFFLAGS:
1572 		XN_LOCK(sc);
1573 		if (ifp->if_flags & IFF_UP) {
1574 			/*
1575 			 * If only the state of the PROMISC flag changed,
1576 			 * then just use the 'set promisc mode' command
1577 			 * instead of reinitializing the entire NIC. Doing
1578 			 * a full re-init means reloading the firmware and
1579 			 * waiting for it to start up, which may take a
1580 			 * second or two.
1581 			 */
1582 #ifdef notyet
1583 			/* No promiscuous mode with Xen */
1584 			if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1585 			    ifp->if_flags & IFF_PROMISC &&
1586 			    !(sc->xn_if_flags & IFF_PROMISC)) {
1587 				XN_SETBIT(sc, XN_RX_MODE,
1588 					  XN_RXMODE_RX_PROMISC);
1589 			} else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1590 				   !(ifp->if_flags & IFF_PROMISC) &&
1591 				   sc->xn_if_flags & IFF_PROMISC) {
1592 				XN_CLRBIT(sc, XN_RX_MODE,
1593 					  XN_RXMODE_RX_PROMISC);
1594 			} else
1595 #endif
1596 				xn_ifinit_locked(sc);
1597 		} else {
1598 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1599 				xn_stop(sc);
1600 			}
1601 		}
1602 		sc->xn_if_flags = ifp->if_flags;
1603 		XN_UNLOCK(sc);
1604 		error = 0;
1605 		break;
1606 	case SIOCSIFCAP:
1607 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1608 		if (mask & IFCAP_TXCSUM) {
1609 			if (IFCAP_TXCSUM & ifp->if_capenable) {
1610 				ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1611 				ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1612 				    | CSUM_IP | CSUM_TSO);
1613 			} else {
1614 				ifp->if_capenable |= IFCAP_TXCSUM;
1615 				ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1616 				    | CSUM_IP);
1617 			}
1618 		}
1619 		if (mask & IFCAP_RXCSUM) {
1620 			ifp->if_capenable ^= IFCAP_RXCSUM;
1621 		}
1622 		if (mask & IFCAP_TSO4) {
1623 			if (IFCAP_TSO4 & ifp->if_capenable) {
1624 				ifp->if_capenable &= ~IFCAP_TSO4;
1625 				ifp->if_hwassist &= ~CSUM_TSO;
1626 			} else if (IFCAP_TXCSUM & ifp->if_capenable) {
1627 				ifp->if_capenable |= IFCAP_TSO4;
1628 				ifp->if_hwassist |= CSUM_TSO;
1629 			} else {
1630 				IPRINTK("Xen requires tx checksum offload"
1631 				    " be enabled to use TSO\n");
1632 				error = EINVAL;
1633 			}
1634 		}
1635 		if (mask & IFCAP_LRO) {
1636 			ifp->if_capenable ^= IFCAP_LRO;
1637 
1638 		}
1639 		error = 0;
1640 		break;
1641 	case SIOCADDMULTI:
1642 	case SIOCDELMULTI:
1643 #ifdef notyet
1644 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1645 			XN_LOCK(sc);
1646 			xn_setmulti(sc);
1647 			XN_UNLOCK(sc);
1648 			error = 0;
1649 		}
1650 #endif
1651 		break;
1652 	case SIOCSIFMEDIA:
1653 	case SIOCGIFMEDIA:
1654 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1655 		break;
1656 	default:
1657 		error = ether_ioctl(ifp, cmd, data);
1658 	}
1659 
1660 	return (error);
1661 }
1662 
1663 static void
1664 xn_stop(struct netfront_info *sc)
1665 {
1666 	struct ifnet *ifp;
1667 
1668 	XN_LOCK_ASSERT(sc);
1669 
1670 	ifp = sc->xn_ifp;
1671 
1672 	callout_stop(&sc->xn_stat_ch);
1673 
1674 	xn_free_rx_ring(sc);
1675 	xn_free_tx_ring(sc);
1676 
1677 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1678 	if_link_state_change(ifp, LINK_STATE_DOWN);
1679 }
1680 
1681 /* START of Xenolinux helper functions adapted to FreeBSD */
1682 int
1683 network_connect(struct netfront_info *np)
1684 {
1685 	int i, requeue_idx, error;
1686 	grant_ref_t ref;
1687 	netif_rx_request_t *req;
1688 	u_int feature_rx_copy;
1689 
1690 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1691 	    "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1692 	if (error)
1693 		feature_rx_copy = 0;
1694 
1695 	/* We only support rx copy. */
1696 	if (!feature_rx_copy)
1697 		return (EPROTONOSUPPORT);
1698 
1699 	/* Recovery procedure: */
1700 	error = talk_to_backend(np->xbdev, np);
1701 	if (error)
1702 		return (error);
1703 
1704 	/* Step 1: Reinitialise variables. */
1705 	xn_query_features(np);
1706 	xn_configure_features(np);
1707 	netif_release_tx_bufs(np);
1708 
1709 	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1710 	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1711 		struct mbuf *m;
1712 		u_long pfn;
1713 
1714 		if (np->rx_mbufs[i] == NULL)
1715 			continue;
1716 
1717 		m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1718 		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1719 
1720 		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1721 		pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1722 
1723 		gnttab_grant_foreign_access_ref(ref,
1724 		    xenbus_get_otherend_id(np->xbdev),
1725 		    pfn, 0);
1726 
1727 		req->gref = ref;
1728 		req->id   = requeue_idx;
1729 
1730 		requeue_idx++;
1731 	}
1732 
1733 	np->rx.req_prod_pvt = requeue_idx;
1734 
1735 	/* Step 3: All public and private state should now be sane.  Get
1736 	 * ready to start sending and receiving packets and give the driver
1737 	 * domain a kick because we've probably just requeued some
1738 	 * packets.
1739 	 */
1740 	netfront_carrier_on(np);
1741 	xen_intr_signal(np->xen_intr_handle);
1742 	XN_TX_LOCK(np);
1743 	xn_txeof(np);
1744 	XN_TX_UNLOCK(np);
1745 	network_alloc_rx_buffers(np);
1746 
1747 	return (0);
1748 }
1749 
1750 static void
1751 xn_query_features(struct netfront_info *np)
1752 {
1753 	int val;
1754 
1755 	device_printf(np->xbdev, "backend features:");
1756 
1757 	if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1758 		"feature-sg", NULL, "%d", &val) < 0)
1759 		val = 0;
1760 
1761 	np->maxfrags = 1;
1762 	if (val) {
1763 		np->maxfrags = MAX_TX_REQ_FRAGS;
1764 		printf(" feature-sg");
1765 	}
1766 
1767 	if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1768 		"feature-gso-tcpv4", NULL, "%d", &val) < 0)
1769 		val = 0;
1770 
1771 	np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
1772 	if (val) {
1773 		np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
1774 		printf(" feature-gso-tcp4");
1775 	}
1776 
1777 	printf("\n");
1778 }
1779 
1780 static int
1781 xn_configure_features(struct netfront_info *np)
1782 {
1783 	int err, cap_enabled;
1784 
1785 	err = 0;
1786 
1787 	if (np->xn_resume &&
1788 	    ((np->xn_ifp->if_capenable & np->xn_ifp->if_capabilities)
1789 	    == np->xn_ifp->if_capenable)) {
1790 		/* Current options are available, no need to do anything. */
1791 		return (0);
1792 	}
1793 
1794 	/* Try to preserve as many options as possible. */
1795 	if (np->xn_resume)
1796 		cap_enabled = np->xn_ifp->if_capenable;
1797 	else
1798 		cap_enabled = UINT_MAX;
1799 
1800 #if (defined(INET) || defined(INET6))
1801 	if ((np->xn_ifp->if_capenable & IFCAP_LRO) == (cap_enabled & IFCAP_LRO))
1802 		tcp_lro_free(&np->xn_lro);
1803 #endif
1804     	np->xn_ifp->if_capenable =
1805 	    np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4) & cap_enabled;
1806 	np->xn_ifp->if_hwassist &= ~CSUM_TSO;
1807 #if (defined(INET) || defined(INET6))
1808 	if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) ==
1809 	    (cap_enabled & IFCAP_LRO)) {
1810 		err = tcp_lro_init(&np->xn_lro);
1811 		if (err) {
1812 			device_printf(np->xbdev, "LRO initialization failed\n");
1813 		} else {
1814 			np->xn_lro.ifp = np->xn_ifp;
1815 			np->xn_ifp->if_capenable |= IFCAP_LRO;
1816 		}
1817 	}
1818 	if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) ==
1819 	    (cap_enabled & IFCAP_TSO4)) {
1820 		np->xn_ifp->if_capenable |= IFCAP_TSO4;
1821 		np->xn_ifp->if_hwassist |= CSUM_TSO;
1822 	}
1823 #endif
1824 	return (err);
1825 }
1826 
1827 /**
1828  * Create a network device.
1829  * @param dev  Newbus device representing this virtual NIC.
1830  */
1831 int
1832 create_netdev(device_t dev)
1833 {
1834 	int i;
1835 	struct netfront_info *np;
1836 	int err;
1837 	struct ifnet *ifp;
1838 
1839 	np = device_get_softc(dev);
1840 
1841 	np->xbdev         = dev;
1842 
1843 	mtx_init(&np->tx_lock, "xntx", "netfront transmit lock", MTX_DEF);
1844 	mtx_init(&np->rx_lock, "xnrx", "netfront receive lock", MTX_DEF);
1845 	mtx_init(&np->sc_lock, "xnsc", "netfront softc lock", MTX_DEF);
1846 
1847 	ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
1848 	ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
1849 	ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
1850 
1851 	np->rx_target     = RX_MIN_TARGET;
1852 	np->rx_min_target = RX_MIN_TARGET;
1853 	np->rx_max_target = RX_MAX_TARGET;
1854 
1855 	/* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
1856 	for (i = 0; i <= NET_TX_RING_SIZE; i++) {
1857 		np->tx_mbufs[i] = (void *) ((u_long) i+1);
1858 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
1859 	}
1860 	np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
1861 
1862 	for (i = 0; i <= NET_RX_RING_SIZE; i++) {
1863 
1864 		np->rx_mbufs[i] = NULL;
1865 		np->grant_rx_ref[i] = GRANT_REF_INVALID;
1866 	}
1867 
1868 	mbufq_init(&np->xn_rx_batch, INT_MAX);
1869 
1870 	/* A grant for every tx ring slot */
1871 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1872 					  &np->gref_tx_head) != 0) {
1873 		IPRINTK("#### netfront can't alloc tx grant refs\n");
1874 		err = ENOMEM;
1875 		goto error;
1876 	}
1877 	/* A grant for every rx ring slot */
1878 	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1879 					  &np->gref_rx_head) != 0) {
1880 		WPRINTK("#### netfront can't alloc rx grant refs\n");
1881 		gnttab_free_grant_references(np->gref_tx_head);
1882 		err = ENOMEM;
1883 		goto error;
1884 	}
1885 
1886 	err = xen_net_read_mac(dev, np->mac);
1887 	if (err) {
1888 		gnttab_free_grant_references(np->gref_rx_head);
1889 		gnttab_free_grant_references(np->gref_tx_head);
1890 		goto error;
1891 	}
1892 
1893 	/* Set up ifnet structure */
1894 	ifp = np->xn_ifp = if_alloc(IFT_ETHER);
1895     	ifp->if_softc = np;
1896     	if_initname(ifp, "xn",  device_get_unit(dev));
1897     	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1898     	ifp->if_ioctl = xn_ioctl;
1899     	ifp->if_start = xn_start;
1900 #ifdef notyet
1901     	ifp->if_watchdog = xn_watchdog;
1902 #endif
1903     	ifp->if_init = xn_ifinit;
1904     	ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
1905 
1906     	ifp->if_hwassist = XN_CSUM_FEATURES;
1907     	ifp->if_capabilities = IFCAP_HWCSUM;
1908 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
1909 	ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS;
1910 	ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
1911 
1912     	ether_ifattach(ifp, np->mac);
1913     	callout_init(&np->xn_stat_ch, 1);
1914 	netfront_carrier_off(np);
1915 
1916 	return (0);
1917 
1918 error:
1919 	KASSERT(err != 0, ("Error path with no error code specified"));
1920 	return (err);
1921 }
1922 
1923 /**
1924  * Handle the change of state of the backend to Closing.  We must delete our
1925  * device-layer structures now, to ensure that writes are flushed through to
1926  * the backend.  Once is this done, we can switch to Closed in
1927  * acknowledgement.
1928  */
1929 #if 0
1930 static void
1931 netfront_closing(device_t dev)
1932 {
1933 #if 0
1934 	struct netfront_info *info = dev->dev_driver_data;
1935 
1936 	DPRINTK("netfront_closing: %s removed\n", dev->nodename);
1937 
1938 	close_netdev(info);
1939 #endif
1940 	xenbus_switch_state(dev, XenbusStateClosed);
1941 }
1942 #endif
1943 
1944 static int
1945 netfront_detach(device_t dev)
1946 {
1947 	struct netfront_info *info = device_get_softc(dev);
1948 
1949 	DPRINTK("%s\n", xenbus_get_node(dev));
1950 
1951 	netif_free(info);
1952 
1953 	return 0;
1954 }
1955 
1956 static void
1957 netif_free(struct netfront_info *info)
1958 {
1959 	XN_LOCK(info);
1960 	xn_stop(info);
1961 	XN_UNLOCK(info);
1962 	callout_drain(&info->xn_stat_ch);
1963 	netif_disconnect_backend(info);
1964 	if (info->xn_ifp != NULL) {
1965 		ether_ifdetach(info->xn_ifp);
1966 		if_free(info->xn_ifp);
1967 		info->xn_ifp = NULL;
1968 	}
1969 	ifmedia_removeall(&info->sc_media);
1970 }
1971 
1972 static void
1973 netif_disconnect_backend(struct netfront_info *info)
1974 {
1975 	XN_RX_LOCK(info);
1976 	XN_TX_LOCK(info);
1977 	netfront_carrier_off(info);
1978 	XN_TX_UNLOCK(info);
1979 	XN_RX_UNLOCK(info);
1980 
1981 	free_ring(&info->tx_ring_ref, &info->tx.sring);
1982 	free_ring(&info->rx_ring_ref, &info->rx.sring);
1983 
1984 	xen_intr_unbind(&info->xen_intr_handle);
1985 }
1986 
1987 static void
1988 free_ring(int *ref, void *ring_ptr_ref)
1989 {
1990 	void **ring_ptr_ptr = ring_ptr_ref;
1991 
1992 	if (*ref != GRANT_REF_INVALID) {
1993 		/* This API frees the associated storage. */
1994 		gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
1995 		*ref = GRANT_REF_INVALID;
1996 	}
1997 	*ring_ptr_ptr = NULL;
1998 }
1999 
2000 static int
2001 xn_ifmedia_upd(struct ifnet *ifp)
2002 {
2003 	return (0);
2004 }
2005 
2006 static void
2007 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2008 {
2009 	ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2010 	ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2011 }
2012 
2013 /* ** Driver registration ** */
2014 static device_method_t netfront_methods[] = {
2015 	/* Device interface */
2016 	DEVMETHOD(device_probe,         netfront_probe),
2017 	DEVMETHOD(device_attach,        netfront_attach),
2018 	DEVMETHOD(device_detach,        netfront_detach),
2019 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
2020 	DEVMETHOD(device_suspend,       netfront_suspend),
2021 	DEVMETHOD(device_resume,        netfront_resume),
2022 
2023 	/* Xenbus interface */
2024 	DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2025 
2026 	DEVMETHOD_END
2027 };
2028 
2029 static driver_t netfront_driver = {
2030 	"xn",
2031 	netfront_methods,
2032 	sizeof(struct netfront_info),
2033 };
2034 devclass_t netfront_devclass;
2035 
2036 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,
2037     NULL);
2038