xref: /freebsd/sys/dev/xen/netfront/netfront.c (revision 3cfe78242997eedd13ef4e5b3a5d45e391c5d3ca)
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/systm.h>
35 #include <sys/sockio.h>
36 #include <sys/limits.h>
37 #include <sys/mbuf.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/kernel.h>
41 #include <sys/socket.h>
42 #include <sys/sysctl.h>
43 #include <sys/queue.h>
44 #include <sys/lock.h>
45 #include <sys/sx.h>
46 
47 #include <net/if.h>
48 #include <net/if_var.h>
49 #include <net/if_arp.h>
50 #include <net/ethernet.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
53 
54 #include <net/bpf.h>
55 
56 #include <net/if_types.h>
57 
58 #include <netinet/in_systm.h>
59 #include <netinet/in.h>
60 #include <netinet/ip.h>
61 #include <netinet/if_ether.h>
62 #if __FreeBSD_version >= 700000
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_lro.h>
65 #endif
66 
67 #include <vm/vm.h>
68 #include <vm/pmap.h>
69 
70 #include <machine/clock.h>      /* for DELAY */
71 #include <machine/bus.h>
72 #include <machine/resource.h>
73 #include <machine/frame.h>
74 #include <machine/vmparam.h>
75 
76 #include <sys/bus.h>
77 #include <sys/rman.h>
78 
79 #include <machine/intr_machdep.h>
80 
81 #include <xen/xen-os.h>
82 #include <xen/hypervisor.h>
83 #include <xen/xen_intr.h>
84 #include <xen/gnttab.h>
85 #include <xen/interface/memory.h>
86 #include <xen/interface/io/netif.h>
87 #include <xen/xenbus/xenbusvar.h>
88 
89 #include "xenbus_if.h"
90 
91 /* Features supported by all backends.  TSO and LRO can be negotiated */
92 #define XN_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
93 
94 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
95 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
96 
97 #if __FreeBSD_version >= 700000
98 /*
99  * Should the driver do LRO on the RX end
100  *  this can be toggled on the fly, but the
101  *  interface must be reset (down/up) for it
102  *  to take effect.
103  */
104 static int xn_enable_lro = 1;
105 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
106 #else
107 
108 #define IFCAP_TSO4	0
109 #define CSUM_TSO	0
110 
111 #endif
112 
113 #ifdef CONFIG_XEN
114 static int MODPARM_rx_copy = 0;
115 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0);
116 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)");
117 static int MODPARM_rx_flip = 0;
118 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0);
119 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)");
120 #else
121 static const int MODPARM_rx_copy = 1;
122 static const int MODPARM_rx_flip = 0;
123 #endif
124 
125 /**
126  * \brief The maximum allowed data fragments in a single transmit
127  *        request.
128  *
129  * This limit is imposed by the backend driver.  We assume here that
130  * we are dealing with a Linux driver domain and have set our limit
131  * to mirror the Linux MAX_SKB_FRAGS constant.
132  */
133 #define	MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2)
134 
135 #define RX_COPY_THRESHOLD 256
136 
137 #define net_ratelimit() 0
138 
139 struct netfront_info;
140 struct netfront_rx_info;
141 
142 static void xn_txeof(struct netfront_info *);
143 static void xn_rxeof(struct netfront_info *);
144 static void network_alloc_rx_buffers(struct netfront_info *);
145 
146 static void xn_tick_locked(struct netfront_info *);
147 static void xn_tick(void *);
148 
149 static void xn_intr(void *);
150 static inline int xn_count_frags(struct mbuf *m);
151 static int  xn_assemble_tx_request(struct netfront_info *sc,
152 				   struct mbuf *m_head);
153 static void xn_start_locked(struct ifnet *);
154 static void xn_start(struct ifnet *);
155 static int  xn_ioctl(struct ifnet *, u_long, caddr_t);
156 static void xn_ifinit_locked(struct netfront_info *);
157 static void xn_ifinit(void *);
158 static void xn_stop(struct netfront_info *);
159 static void xn_query_features(struct netfront_info *np);
160 static int  xn_configure_features(struct netfront_info *np);
161 #ifdef notyet
162 static void xn_watchdog(struct ifnet *);
163 #endif
164 
165 #ifdef notyet
166 static void netfront_closing(device_t dev);
167 #endif
168 static void netif_free(struct netfront_info *info);
169 static int netfront_detach(device_t dev);
170 
171 static int talk_to_backend(device_t dev, struct netfront_info *info);
172 static int create_netdev(device_t dev);
173 static void netif_disconnect_backend(struct netfront_info *info);
174 static int setup_device(device_t dev, struct netfront_info *info);
175 static void free_ring(int *ref, void *ring_ptr_ref);
176 
177 static int  xn_ifmedia_upd(struct ifnet *ifp);
178 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
179 
180 /* Xenolinux helper functions */
181 int network_connect(struct netfront_info *);
182 
183 static void xn_free_rx_ring(struct netfront_info *);
184 
185 static void xn_free_tx_ring(struct netfront_info *);
186 
187 static int xennet_get_responses(struct netfront_info *np,
188 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
189 	struct mbuf **list, int *pages_flipped_p);
190 
191 #define virt_to_mfn(x) (vtophys(x) >> PAGE_SHIFT)
192 
193 #define INVALID_P2M_ENTRY (~0UL)
194 
195 /*
196  * Mbuf pointers. We need these to keep track of the virtual addresses
197  * of our mbuf chains since we can only convert from virtual to physical,
198  * not the other way around.  The size must track the free index arrays.
199  */
200 struct xn_chain_data {
201 	struct mbuf    *xn_tx_chain[NET_TX_RING_SIZE+1];
202 	int		xn_tx_chain_cnt;
203 	struct mbuf    *xn_rx_chain[NET_RX_RING_SIZE+1];
204 };
205 
206 struct net_device_stats
207 {
208 	u_long	rx_packets;		/* total packets received	*/
209 	u_long	tx_packets;		/* total packets transmitted	*/
210 	u_long	rx_bytes;		/* total bytes received 	*/
211 	u_long	tx_bytes;		/* total bytes transmitted	*/
212 	u_long	rx_errors;		/* bad packets received		*/
213 	u_long	tx_errors;		/* packet transmit problems	*/
214 	u_long	rx_dropped;		/* no space in linux buffers	*/
215 	u_long	tx_dropped;		/* no space available in linux	*/
216 	u_long	multicast;		/* multicast packets received	*/
217 	u_long	collisions;
218 
219 	/* detailed rx_errors: */
220 	u_long	rx_length_errors;
221 	u_long	rx_over_errors;		/* receiver ring buff overflow	*/
222 	u_long	rx_crc_errors;		/* recved pkt with crc error	*/
223 	u_long	rx_frame_errors;	/* recv'd frame alignment error */
224 	u_long	rx_fifo_errors;		/* recv'r fifo overrun		*/
225 	u_long	rx_missed_errors;	/* receiver missed packet	*/
226 
227 	/* detailed tx_errors */
228 	u_long	tx_aborted_errors;
229 	u_long	tx_carrier_errors;
230 	u_long	tx_fifo_errors;
231 	u_long	tx_heartbeat_errors;
232 	u_long	tx_window_errors;
233 
234 	/* for cslip etc */
235 	u_long	rx_compressed;
236 	u_long	tx_compressed;
237 };
238 
239 struct netfront_info {
240 	struct ifnet *xn_ifp;
241 #if __FreeBSD_version >= 700000
242 	struct lro_ctrl xn_lro;
243 #endif
244 
245 	struct net_device_stats stats;
246 	u_int tx_full;
247 
248 	netif_tx_front_ring_t tx;
249 	netif_rx_front_ring_t rx;
250 
251 	struct mtx   tx_lock;
252 	struct mtx   rx_lock;
253 	struct mtx   sc_lock;
254 
255 	xen_intr_handle_t xen_intr_handle;
256 	u_int copying_receiver;
257 	u_int carrier;
258 	u_int maxfrags;
259 
260 	/* Receive-ring batched refills. */
261 #define RX_MIN_TARGET 32
262 #define RX_MAX_TARGET NET_RX_RING_SIZE
263 	int rx_min_target;
264 	int rx_max_target;
265 	int rx_target;
266 
267 	grant_ref_t gref_tx_head;
268 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
269 	grant_ref_t gref_rx_head;
270 	grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
271 
272 	device_t		xbdev;
273 	int			tx_ring_ref;
274 	int			rx_ring_ref;
275 	uint8_t			mac[ETHER_ADDR_LEN];
276 	struct xn_chain_data	xn_cdata;	/* mbufs */
277 	struct mbufq		xn_rx_batch;	/* batch queue */
278 
279 	int			xn_if_flags;
280 	struct callout	        xn_stat_ch;
281 
282 	u_long			rx_pfn_array[NET_RX_RING_SIZE];
283 	struct ifmedia		sc_media;
284 
285 	bool			xn_resume;
286 };
287 
288 #define rx_mbufs xn_cdata.xn_rx_chain
289 #define tx_mbufs xn_cdata.xn_tx_chain
290 
291 #define XN_LOCK_INIT(_sc, _name) \
292         mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
293         mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF);  \
294         mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF)
295 
296 #define XN_RX_LOCK(_sc)           mtx_lock(&(_sc)->rx_lock)
297 #define XN_RX_UNLOCK(_sc)         mtx_unlock(&(_sc)->rx_lock)
298 
299 #define XN_TX_LOCK(_sc)           mtx_lock(&(_sc)->tx_lock)
300 #define XN_TX_UNLOCK(_sc)         mtx_unlock(&(_sc)->tx_lock)
301 
302 #define XN_LOCK(_sc)           mtx_lock(&(_sc)->sc_lock);
303 #define XN_UNLOCK(_sc)         mtx_unlock(&(_sc)->sc_lock);
304 
305 #define XN_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->sc_lock, MA_OWNED);
306 #define XN_RX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->rx_lock, MA_OWNED);
307 #define XN_TX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->tx_lock, MA_OWNED);
308 #define XN_LOCK_DESTROY(_sc)   mtx_destroy(&(_sc)->rx_lock); \
309                                mtx_destroy(&(_sc)->tx_lock); \
310                                mtx_destroy(&(_sc)->sc_lock);
311 
312 struct netfront_rx_info {
313 	struct netif_rx_response rx;
314 	struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
315 };
316 
317 #define netfront_carrier_on(netif)	((netif)->carrier = 1)
318 #define netfront_carrier_off(netif)	((netif)->carrier = 0)
319 #define netfront_carrier_ok(netif)	((netif)->carrier)
320 
321 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
322 
323 static inline void
324 add_id_to_freelist(struct mbuf **list, uintptr_t id)
325 {
326 	KASSERT(id != 0,
327 		("%s: the head item (0) must always be free.", __func__));
328 	list[id] = list[0];
329 	list[0]  = (struct mbuf *)id;
330 }
331 
332 static inline unsigned short
333 get_id_from_freelist(struct mbuf **list)
334 {
335 	uintptr_t id;
336 
337 	id = (uintptr_t)list[0];
338 	KASSERT(id != 0,
339 		("%s: the head item (0) must always remain free.", __func__));
340 	list[0] = list[id];
341 	return (id);
342 }
343 
344 static inline int
345 xennet_rxidx(RING_IDX idx)
346 {
347 	return idx & (NET_RX_RING_SIZE - 1);
348 }
349 
350 static inline struct mbuf *
351 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
352 {
353 	int i = xennet_rxidx(ri);
354 	struct mbuf *m;
355 
356 	m = np->rx_mbufs[i];
357 	np->rx_mbufs[i] = NULL;
358 	return (m);
359 }
360 
361 static inline grant_ref_t
362 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
363 {
364 	int i = xennet_rxidx(ri);
365 	grant_ref_t ref = np->grant_rx_ref[i];
366 	KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
367 	np->grant_rx_ref[i] = GRANT_REF_INVALID;
368 	return ref;
369 }
370 
371 #define IPRINTK(fmt, args...) \
372     printf("[XEN] " fmt, ##args)
373 #ifdef INVARIANTS
374 #define WPRINTK(fmt, args...) \
375     printf("[XEN] " fmt, ##args)
376 #else
377 #define WPRINTK(fmt, args...)
378 #endif
379 #ifdef DEBUG
380 #define DPRINTK(fmt, args...) \
381     printf("[XEN] %s: " fmt, __func__, ##args)
382 #else
383 #define DPRINTK(fmt, args...)
384 #endif
385 
386 /**
387  * Read the 'mac' node at the given device's node in the store, and parse that
388  * as colon-separated octets, placing result the given mac array.  mac must be
389  * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
390  * Return 0 on success, or errno on error.
391  */
392 static int
393 xen_net_read_mac(device_t dev, uint8_t mac[])
394 {
395 	int error, i;
396 	char *s, *e, *macstr;
397 	const char *path;
398 
399 	path = xenbus_get_node(dev);
400 	error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
401 	if (error == ENOENT) {
402 		/*
403 		 * Deal with missing mac XenStore nodes on devices with
404 		 * HVM emulation (the 'ioemu' configuration attribute)
405 		 * enabled.
406 		 *
407 		 * The HVM emulator may execute in a stub device model
408 		 * domain which lacks the permission, only given to Dom0,
409 		 * to update the guest's XenStore tree.  For this reason,
410 		 * the HVM emulator doesn't even attempt to write the
411 		 * front-side mac node, even when operating in Dom0.
412 		 * However, there should always be a mac listed in the
413 		 * backend tree.  Fallback to this version if our query
414 		 * of the front side XenStore location doesn't find
415 		 * anything.
416 		 */
417 		path = xenbus_get_otherend_path(dev);
418 		error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
419 	}
420 	if (error != 0) {
421 		xenbus_dev_fatal(dev, error, "parsing %s/mac", path);
422 		return (error);
423 	}
424 
425 	s = macstr;
426 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
427 		mac[i] = strtoul(s, &e, 16);
428 		if (s == e || (e[0] != ':' && e[0] != 0)) {
429 			free(macstr, M_XENBUS);
430 			return (ENOENT);
431 		}
432 		s = &e[1];
433 	}
434 	free(macstr, M_XENBUS);
435 	return (0);
436 }
437 
438 /**
439  * Entry point to this code when a new device is created.  Allocate the basic
440  * structures and the ring buffers for communication with the backend, and
441  * inform the backend of the appropriate details for those.  Switch to
442  * Connected state.
443  */
444 static int
445 netfront_probe(device_t dev)
446 {
447 
448 	if (xen_hvm_domain() && xen_disable_pv_nics != 0)
449 		return (ENXIO);
450 
451 	if (!strcmp(xenbus_get_type(dev), "vif")) {
452 		device_set_desc(dev, "Virtual Network Interface");
453 		return (0);
454 	}
455 
456 	return (ENXIO);
457 }
458 
459 static int
460 netfront_attach(device_t dev)
461 {
462 	int err;
463 
464 	err = create_netdev(dev);
465 	if (err) {
466 		xenbus_dev_fatal(dev, err, "creating netdev");
467 		return (err);
468 	}
469 
470 #if __FreeBSD_version >= 700000
471 	SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
472 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
473 	    OID_AUTO, "enable_lro", CTLFLAG_RW,
474 	    &xn_enable_lro, 0, "Large Receive Offload");
475 #endif
476 
477 	return (0);
478 }
479 
480 static int
481 netfront_suspend(device_t dev)
482 {
483 	struct netfront_info *info = device_get_softc(dev);
484 
485 	XN_RX_LOCK(info);
486 	XN_TX_LOCK(info);
487 	netfront_carrier_off(info);
488 	XN_TX_UNLOCK(info);
489 	XN_RX_UNLOCK(info);
490 	return (0);
491 }
492 
493 /**
494  * We are reconnecting to the backend, due to a suspend/resume, or a backend
495  * driver restart.  We tear down our netif structure and recreate it, but
496  * leave the device-layer structures intact so that this is transparent to the
497  * rest of the kernel.
498  */
499 static int
500 netfront_resume(device_t dev)
501 {
502 	struct netfront_info *info = device_get_softc(dev);
503 
504 	info->xn_resume = true;
505 	netif_disconnect_backend(info);
506 	return (0);
507 }
508 
509 /* Common code used when first setting up, and when resuming. */
510 static int
511 talk_to_backend(device_t dev, struct netfront_info *info)
512 {
513 	const char *message;
514 	struct xs_transaction xst;
515 	const char *node = xenbus_get_node(dev);
516 	int err;
517 
518 	err = xen_net_read_mac(dev, info->mac);
519 	if (err) {
520 		xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
521 		goto out;
522 	}
523 
524 	/* Create shared ring, alloc event channel. */
525 	err = setup_device(dev, info);
526 	if (err)
527 		goto out;
528 
529  again:
530 	err = xs_transaction_start(&xst);
531 	if (err) {
532 		xenbus_dev_fatal(dev, err, "starting transaction");
533 		goto destroy_ring;
534 	}
535 	err = xs_printf(xst, node, "tx-ring-ref","%u",
536 			info->tx_ring_ref);
537 	if (err) {
538 		message = "writing tx ring-ref";
539 		goto abort_transaction;
540 	}
541 	err = xs_printf(xst, node, "rx-ring-ref","%u",
542 			info->rx_ring_ref);
543 	if (err) {
544 		message = "writing rx ring-ref";
545 		goto abort_transaction;
546 	}
547 	err = xs_printf(xst, node,
548 			"event-channel", "%u",
549 			xen_intr_port(info->xen_intr_handle));
550 	if (err) {
551 		message = "writing event-channel";
552 		goto abort_transaction;
553 	}
554 	err = xs_printf(xst, node, "request-rx-copy", "%u",
555 			info->copying_receiver);
556 	if (err) {
557 		message = "writing request-rx-copy";
558 		goto abort_transaction;
559 	}
560 	err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
561 	if (err) {
562 		message = "writing feature-rx-notify";
563 		goto abort_transaction;
564 	}
565 	err = xs_printf(xst, node, "feature-sg", "%d", 1);
566 	if (err) {
567 		message = "writing feature-sg";
568 		goto abort_transaction;
569 	}
570 #if __FreeBSD_version >= 700000
571 	err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
572 	if (err) {
573 		message = "writing feature-gso-tcpv4";
574 		goto abort_transaction;
575 	}
576 #endif
577 
578 	err = xs_transaction_end(xst, 0);
579 	if (err) {
580 		if (err == EAGAIN)
581 			goto again;
582 		xenbus_dev_fatal(dev, err, "completing transaction");
583 		goto destroy_ring;
584 	}
585 
586 	return 0;
587 
588  abort_transaction:
589 	xs_transaction_end(xst, 1);
590 	xenbus_dev_fatal(dev, err, "%s", message);
591  destroy_ring:
592 	netif_free(info);
593  out:
594 	return err;
595 }
596 
597 static int
598 setup_device(device_t dev, struct netfront_info *info)
599 {
600 	netif_tx_sring_t *txs;
601 	netif_rx_sring_t *rxs;
602 	int error;
603 	struct ifnet *ifp;
604 
605 	ifp = info->xn_ifp;
606 
607 	info->tx_ring_ref = GRANT_REF_INVALID;
608 	info->rx_ring_ref = GRANT_REF_INVALID;
609 	info->rx.sring = NULL;
610 	info->tx.sring = NULL;
611 
612 	txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
613 	if (!txs) {
614 		error = ENOMEM;
615 		xenbus_dev_fatal(dev, error, "allocating tx ring page");
616 		goto fail;
617 	}
618 	SHARED_RING_INIT(txs);
619 	FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
620 	error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
621 	if (error)
622 		goto fail;
623 
624 	rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
625 	if (!rxs) {
626 		error = ENOMEM;
627 		xenbus_dev_fatal(dev, error, "allocating rx ring page");
628 		goto fail;
629 	}
630 	SHARED_RING_INIT(rxs);
631 	FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
632 
633 	error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
634 	if (error)
635 		goto fail;
636 
637 	error = xen_intr_alloc_and_bind_local_port(dev,
638 	    xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info,
639 	    INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle);
640 
641 	if (error) {
642 		xenbus_dev_fatal(dev, error,
643 				 "xen_intr_alloc_and_bind_local_port failed");
644 		goto fail;
645 	}
646 
647 	return (0);
648 
649  fail:
650 	netif_free(info);
651 	return (error);
652 }
653 
654 #ifdef INET
655 /**
656  * If this interface has an ipv4 address, send an arp for it. This
657  * helps to get the network going again after migrating hosts.
658  */
659 static void
660 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
661 {
662 	struct ifnet *ifp;
663 	struct ifaddr *ifa;
664 
665 	ifp = info->xn_ifp;
666 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
667 		if (ifa->ifa_addr->sa_family == AF_INET) {
668 			arp_ifinit(ifp, ifa);
669 		}
670 	}
671 }
672 #endif
673 
674 /**
675  * Callback received when the backend's state changes.
676  */
677 static void
678 netfront_backend_changed(device_t dev, XenbusState newstate)
679 {
680 	struct netfront_info *sc = device_get_softc(dev);
681 
682 	DPRINTK("newstate=%d\n", newstate);
683 
684 	switch (newstate) {
685 	case XenbusStateInitialising:
686 	case XenbusStateInitialised:
687 	case XenbusStateUnknown:
688 	case XenbusStateClosed:
689 	case XenbusStateReconfigured:
690 	case XenbusStateReconfiguring:
691 		break;
692 	case XenbusStateInitWait:
693 		if (xenbus_get_state(dev) != XenbusStateInitialising)
694 			break;
695 		if (network_connect(sc) != 0)
696 			break;
697 		xenbus_set_state(dev, XenbusStateConnected);
698 		break;
699 	case XenbusStateClosing:
700 		xenbus_set_state(dev, XenbusStateClosed);
701 		break;
702 	case XenbusStateConnected:
703 #ifdef INET
704 		netfront_send_fake_arp(dev, sc);
705 #endif
706 		break;
707 	}
708 }
709 
710 static void
711 xn_free_rx_ring(struct netfront_info *sc)
712 {
713 #if 0
714 	int i;
715 
716 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
717 		if (sc->xn_cdata.rx_mbufs[i] != NULL) {
718 			m_freem(sc->rx_mbufs[i]);
719 			sc->rx_mbufs[i] = NULL;
720 		}
721 	}
722 
723 	sc->rx.rsp_cons = 0;
724 	sc->xn_rx_if->req_prod = 0;
725 	sc->xn_rx_if->event = sc->rx.rsp_cons ;
726 #endif
727 }
728 
729 static void
730 xn_free_tx_ring(struct netfront_info *sc)
731 {
732 #if 0
733 	int i;
734 
735 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
736 		if (sc->tx_mbufs[i] != NULL) {
737 			m_freem(sc->tx_mbufs[i]);
738 			sc->xn_cdata.xn_tx_chain[i] = NULL;
739 		}
740 	}
741 
742 	return;
743 #endif
744 }
745 
746 /**
747  * \brief Verify that there is sufficient space in the Tx ring
748  *        buffer for a maximally sized request to be enqueued.
749  *
750  * A transmit request requires a transmit descriptor for each packet
751  * fragment, plus up to 2 entries for "options" (e.g. TSO).
752  */
753 static inline int
754 xn_tx_slot_available(struct netfront_info *np)
755 {
756 	return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
757 }
758 
759 static void
760 netif_release_tx_bufs(struct netfront_info *np)
761 {
762 	int i;
763 
764 	for (i = 1; i <= NET_TX_RING_SIZE; i++) {
765 		struct mbuf *m;
766 
767 		m = np->tx_mbufs[i];
768 
769 		/*
770 		 * We assume that no kernel addresses are
771 		 * less than NET_TX_RING_SIZE.  Any entry
772 		 * in the table that is below this number
773 		 * must be an index from free-list tracking.
774 		 */
775 		if (((uintptr_t)m) <= NET_TX_RING_SIZE)
776 			continue;
777 		gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
778 		gnttab_release_grant_reference(&np->gref_tx_head,
779 		    np->grant_tx_ref[i]);
780 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
781 		add_id_to_freelist(np->tx_mbufs, i);
782 		np->xn_cdata.xn_tx_chain_cnt--;
783 		if (np->xn_cdata.xn_tx_chain_cnt < 0) {
784 			panic("%s: tx_chain_cnt must be >= 0", __func__);
785 		}
786 		m_free(m);
787 	}
788 }
789 
790 static void
791 network_alloc_rx_buffers(struct netfront_info *sc)
792 {
793 	int otherend_id = xenbus_get_otherend_id(sc->xbdev);
794 	unsigned short id;
795 	struct mbuf *m_new;
796 	int i, batch_target, notify;
797 	RING_IDX req_prod;
798 	struct xen_memory_reservation reservation;
799 	grant_ref_t ref;
800 	int nr_flips;
801 	netif_rx_request_t *req;
802 	vm_offset_t vaddr;
803 	u_long pfn;
804 
805 	req_prod = sc->rx.req_prod_pvt;
806 
807 	if (__predict_false(sc->carrier == 0))
808 		return;
809 
810 	/*
811 	 * Allocate mbufs greedily, even though we batch updates to the
812 	 * receive ring. This creates a less bursty demand on the memory
813 	 * allocator, and so should reduce the chance of failed allocation
814 	 * requests both for ourself and for other kernel subsystems.
815 	 *
816 	 * Here we attempt to maintain rx_target buffers in flight, counting
817 	 * buffers that we have yet to process in the receive ring.
818 	 */
819 	batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
820 	for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
821 		m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE);
822 		if (m_new == NULL) {
823 			if (i != 0)
824 				goto refill;
825 			/*
826 			 * XXX set timer
827 			 */
828 			break;
829 		}
830 		m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
831 
832 		/* queue the mbufs allocated */
833 		(void )mbufq_enqueue(&sc->xn_rx_batch, m_new);
834 	}
835 
836 	/*
837 	 * If we've allocated at least half of our target number of entries,
838 	 * submit them to the backend - we have enough to make the overhead
839 	 * of submission worthwhile.  Otherwise wait for more mbufs and
840 	 * request entries to become available.
841 	 */
842 	if (i < (sc->rx_target/2)) {
843 		if (req_prod >sc->rx.sring->req_prod)
844 			goto push;
845 		return;
846 	}
847 
848 	/*
849 	 * Double floating fill target if we risked having the backend
850 	 * run out of empty buffers for receive traffic.  We define "running
851 	 * low" as having less than a fourth of our target buffers free
852 	 * at the time we refilled the queue.
853 	 */
854 	if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
855 		sc->rx_target *= 2;
856 		if (sc->rx_target > sc->rx_max_target)
857 			sc->rx_target = sc->rx_max_target;
858 	}
859 
860 refill:
861 	for (nr_flips = i = 0; ; i++) {
862 		if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
863 			break;
864 
865 		m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
866 				vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
867 
868 		id = xennet_rxidx(req_prod + i);
869 
870 		KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
871 		sc->rx_mbufs[id] = m_new;
872 
873 		ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
874 		KASSERT(ref != GNTTAB_LIST_END,
875 			("reserved grant references exhuasted"));
876 		sc->grant_rx_ref[id] = ref;
877 
878 		vaddr = mtod(m_new, vm_offset_t);
879 		pfn = vtophys(vaddr) >> PAGE_SHIFT;
880 		req = RING_GET_REQUEST(&sc->rx, req_prod + i);
881 
882 		if (sc->copying_receiver == 0) {
883 			gnttab_grant_foreign_transfer_ref(ref,
884 			    otherend_id, pfn);
885 			sc->rx_pfn_array[nr_flips] = pfn;
886 			nr_flips++;
887 		} else {
888 			gnttab_grant_foreign_access_ref(ref,
889 			    otherend_id,
890 			    pfn, 0);
891 		}
892 		req->id = id;
893 		req->gref = ref;
894 
895 		sc->rx_pfn_array[i] =
896 		    vtophys(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
897 	}
898 
899 	KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
900 	KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
901 	/*
902 	 * We may have allocated buffers which have entries outstanding
903 	 * in the page * update queue -- make sure we flush those first!
904 	 */
905 	if (nr_flips != 0) {
906 #ifdef notyet
907 		/* Tell the ballon driver what is going on. */
908 		balloon_update_driver_allowance(i);
909 #endif
910 		set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
911 		reservation.nr_extents   = i;
912 		reservation.extent_order = 0;
913 		reservation.address_bits = 0;
914 		reservation.domid        = DOMID_SELF;
915 	} else {
916 		wmb();
917 	}
918 
919 	/* Above is a suitable barrier to ensure backend will see requests. */
920 	sc->rx.req_prod_pvt = req_prod + i;
921 push:
922 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
923 	if (notify)
924 		xen_intr_signal(sc->xen_intr_handle);
925 }
926 
927 static void
928 xn_rxeof(struct netfront_info *np)
929 {
930 	struct ifnet *ifp;
931 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
932 	struct lro_ctrl *lro = &np->xn_lro;
933 	struct lro_entry *queued;
934 #endif
935 	struct netfront_rx_info rinfo;
936 	struct netif_rx_response *rx = &rinfo.rx;
937 	struct netif_extra_info *extras = rinfo.extras;
938 	RING_IDX i, rp;
939 	struct mbuf *m;
940 	struct mbufq rxq, errq;
941 	int err, pages_flipped = 0, work_to_do;
942 
943 	do {
944 		XN_RX_LOCK_ASSERT(np);
945 		if (!netfront_carrier_ok(np))
946 			return;
947 
948 		/* XXX: there should be some sane limit. */
949 		mbufq_init(&errq, INT_MAX);
950 		mbufq_init(&rxq, INT_MAX);
951 
952 		ifp = np->xn_ifp;
953 
954 		rp = np->rx.sring->rsp_prod;
955 		rmb();	/* Ensure we see queued responses up to 'rp'. */
956 
957 		i = np->rx.rsp_cons;
958 		while ((i != rp)) {
959 			memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
960 			memset(extras, 0, sizeof(rinfo.extras));
961 
962 			m = NULL;
963 			err = xennet_get_responses(np, &rinfo, rp, &i, &m,
964 			    &pages_flipped);
965 
966 			if (__predict_false(err)) {
967 				if (m)
968 					(void )mbufq_enqueue(&errq, m);
969 				np->stats.rx_errors++;
970 				continue;
971 			}
972 
973 			m->m_pkthdr.rcvif = ifp;
974 			if ( rx->flags & NETRXF_data_validated ) {
975 				/* Tell the stack the checksums are okay */
976 				/*
977 				 * XXX this isn't necessarily the case - need to add
978 				 * check
979 				 */
980 
981 				m->m_pkthdr.csum_flags |=
982 					(CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
983 					    | CSUM_PSEUDO_HDR);
984 				m->m_pkthdr.csum_data = 0xffff;
985 			}
986 
987 			np->stats.rx_packets++;
988 			np->stats.rx_bytes += m->m_pkthdr.len;
989 
990 			(void )mbufq_enqueue(&rxq, m);
991 			np->rx.rsp_cons = i;
992 		}
993 
994 		if (pages_flipped) {
995 			/* Some pages are no longer absent... */
996 #ifdef notyet
997 			balloon_update_driver_allowance(-pages_flipped);
998 #endif
999 		}
1000 
1001 		mbufq_drain(&errq);
1002 
1003 		/*
1004 		 * Process all the mbufs after the remapping is complete.
1005 		 * Break the mbuf chain first though.
1006 		 */
1007 		while ((m = mbufq_dequeue(&rxq)) != NULL) {
1008 			if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1009 
1010 			/*
1011 			 * Do we really need to drop the rx lock?
1012 			 */
1013 			XN_RX_UNLOCK(np);
1014 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1015 			/* Use LRO if possible */
1016 			if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1017 			    lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1018 				/*
1019 				 * If LRO fails, pass up to the stack
1020 				 * directly.
1021 				 */
1022 				(*ifp->if_input)(ifp, m);
1023 			}
1024 #else
1025 			(*ifp->if_input)(ifp, m);
1026 #endif
1027 			XN_RX_LOCK(np);
1028 		}
1029 
1030 		np->rx.rsp_cons = i;
1031 
1032 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1033 		/*
1034 		 * Flush any outstanding LRO work
1035 		 */
1036 		while (!SLIST_EMPTY(&lro->lro_active)) {
1037 			queued = SLIST_FIRST(&lro->lro_active);
1038 			SLIST_REMOVE_HEAD(&lro->lro_active, next);
1039 			tcp_lro_flush(lro, queued);
1040 		}
1041 #endif
1042 
1043 #if 0
1044 		/* If we get a callback with very few responses, reduce fill target. */
1045 		/* NB. Note exponential increase, linear decrease. */
1046 		if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1047 			((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1048 			np->rx_target = np->rx_min_target;
1049 #endif
1050 
1051 		network_alloc_rx_buffers(np);
1052 
1053 		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1054 	} while (work_to_do);
1055 }
1056 
1057 static void
1058 xn_txeof(struct netfront_info *np)
1059 {
1060 	RING_IDX i, prod;
1061 	unsigned short id;
1062 	struct ifnet *ifp;
1063 	netif_tx_response_t *txr;
1064 	struct mbuf *m;
1065 
1066 	XN_TX_LOCK_ASSERT(np);
1067 
1068 	if (!netfront_carrier_ok(np))
1069 		return;
1070 
1071 	ifp = np->xn_ifp;
1072 
1073 	do {
1074 		prod = np->tx.sring->rsp_prod;
1075 		rmb(); /* Ensure we see responses up to 'rp'. */
1076 
1077 		for (i = np->tx.rsp_cons; i != prod; i++) {
1078 			txr = RING_GET_RESPONSE(&np->tx, i);
1079 			if (txr->status == NETIF_RSP_NULL)
1080 				continue;
1081 
1082 			if (txr->status != NETIF_RSP_OKAY) {
1083 				printf("%s: WARNING: response is %d!\n",
1084 				       __func__, txr->status);
1085 			}
1086 			id = txr->id;
1087 			m = np->tx_mbufs[id];
1088 			KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1089 			KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1090 				("mbuf already on the free list, but we're "
1091 				"trying to free it again!"));
1092 			M_ASSERTVALID(m);
1093 
1094 			/*
1095 			 * Increment packet count if this is the last
1096 			 * mbuf of the chain.
1097 			 */
1098 			if (!m->m_next)
1099 				if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1100 			if (__predict_false(gnttab_query_foreign_access(
1101 			    np->grant_tx_ref[id]) != 0)) {
1102 				panic("%s: grant id %u still in use by the "
1103 				    "backend", __func__, id);
1104 			}
1105 			gnttab_end_foreign_access_ref(
1106 				np->grant_tx_ref[id]);
1107 			gnttab_release_grant_reference(
1108 				&np->gref_tx_head, np->grant_tx_ref[id]);
1109 			np->grant_tx_ref[id] = GRANT_REF_INVALID;
1110 
1111 			np->tx_mbufs[id] = NULL;
1112 			add_id_to_freelist(np->tx_mbufs, id);
1113 			np->xn_cdata.xn_tx_chain_cnt--;
1114 			m_free(m);
1115 			/* Only mark the queue active if we've freed up at least one slot to try */
1116 			ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1117 		}
1118 		np->tx.rsp_cons = prod;
1119 
1120 		/*
1121 		 * Set a new event, then check for race with update of
1122 		 * tx_cons. Note that it is essential to schedule a
1123 		 * callback, no matter how few buffers are pending. Even if
1124 		 * there is space in the transmit ring, higher layers may
1125 		 * be blocked because too much data is outstanding: in such
1126 		 * cases notification from Xen is likely to be the only kick
1127 		 * that we'll get.
1128 		 */
1129 		np->tx.sring->rsp_event =
1130 		    prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1131 
1132 		mb();
1133 	} while (prod != np->tx.sring->rsp_prod);
1134 
1135 	if (np->tx_full &&
1136 	    ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1137 		np->tx_full = 0;
1138 #if 0
1139 		if (np->user_state == UST_OPEN)
1140 			netif_wake_queue(dev);
1141 #endif
1142 	}
1143 }
1144 
1145 static void
1146 xn_intr(void *xsc)
1147 {
1148 	struct netfront_info *np = xsc;
1149 	struct ifnet *ifp = np->xn_ifp;
1150 
1151 #if 0
1152 	if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1153 	    likely(netfront_carrier_ok(np)) &&
1154 	    ifp->if_drv_flags & IFF_DRV_RUNNING))
1155 		return;
1156 #endif
1157 	if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1158 		XN_TX_LOCK(np);
1159 		xn_txeof(np);
1160 		XN_TX_UNLOCK(np);
1161 	}
1162 
1163 	XN_RX_LOCK(np);
1164 	xn_rxeof(np);
1165 	XN_RX_UNLOCK(np);
1166 
1167 	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1168 	    !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1169 		xn_start(ifp);
1170 }
1171 
1172 static void
1173 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1174 	grant_ref_t ref)
1175 {
1176 	int new = xennet_rxidx(np->rx.req_prod_pvt);
1177 
1178 	KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1179 	np->rx_mbufs[new] = m;
1180 	np->grant_rx_ref[new] = ref;
1181 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1182 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1183 	np->rx.req_prod_pvt++;
1184 }
1185 
1186 static int
1187 xennet_get_extras(struct netfront_info *np,
1188     struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1189 {
1190 	struct netif_extra_info *extra;
1191 
1192 	int err = 0;
1193 
1194 	do {
1195 		struct mbuf *m;
1196 		grant_ref_t ref;
1197 
1198 		if (__predict_false(*cons + 1 == rp)) {
1199 #if 0
1200 			if (net_ratelimit())
1201 				WPRINTK("Missing extra info\n");
1202 #endif
1203 			err = EINVAL;
1204 			break;
1205 		}
1206 
1207 		extra = (struct netif_extra_info *)
1208 		RING_GET_RESPONSE(&np->rx, ++(*cons));
1209 
1210 		if (__predict_false(!extra->type ||
1211 			extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1212 #if 0
1213 			if (net_ratelimit())
1214 				WPRINTK("Invalid extra type: %d\n",
1215 					extra->type);
1216 #endif
1217 			err = EINVAL;
1218 		} else {
1219 			memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1220 		}
1221 
1222 		m = xennet_get_rx_mbuf(np, *cons);
1223 		ref = xennet_get_rx_ref(np, *cons);
1224 		xennet_move_rx_slot(np, m, ref);
1225 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1226 
1227 	return err;
1228 }
1229 
1230 static int
1231 xennet_get_responses(struct netfront_info *np,
1232 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1233 	struct mbuf  **list,
1234 	int *pages_flipped_p)
1235 {
1236 	int pages_flipped = *pages_flipped_p;
1237 	struct netif_rx_response *rx = &rinfo->rx;
1238 	struct netif_extra_info *extras = rinfo->extras;
1239 	struct mbuf *m, *m0, *m_prev;
1240 	grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1241 	RING_IDX ref_cons = *cons;
1242 	int frags = 1;
1243 	int err = 0;
1244 	u_long ret;
1245 
1246 	m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1247 
1248 	if (rx->flags & NETRXF_extra_info) {
1249 		err = xennet_get_extras(np, extras, rp, cons);
1250 	}
1251 
1252 	if (m0 != NULL) {
1253 		m0->m_pkthdr.len = 0;
1254 		m0->m_next = NULL;
1255 	}
1256 
1257 	for (;;) {
1258 		u_long mfn;
1259 
1260 #if 0
1261 		DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1262 			rx->status, rx->offset, frags);
1263 #endif
1264 		if (__predict_false(rx->status < 0 ||
1265 			rx->offset + rx->status > PAGE_SIZE)) {
1266 
1267 #if 0
1268 			if (net_ratelimit())
1269 				WPRINTK("rx->offset: %x, size: %u\n",
1270 					rx->offset, rx->status);
1271 #endif
1272 			xennet_move_rx_slot(np, m, ref);
1273 			if (m0 == m)
1274 				m0 = NULL;
1275 			m = NULL;
1276 			err = EINVAL;
1277 			goto next_skip_queue;
1278 		}
1279 
1280 		/*
1281 		 * This definitely indicates a bug, either in this driver or in
1282 		 * the backend driver. In future this should flag the bad
1283 		 * situation to the system controller to reboot the backed.
1284 		 */
1285 		if (ref == GRANT_REF_INVALID) {
1286 
1287 #if 0
1288 			if (net_ratelimit())
1289 				WPRINTK("Bad rx response id %d.\n", rx->id);
1290 #endif
1291 			printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1292 			err = EINVAL;
1293 			goto next;
1294 		}
1295 
1296 		if (!np->copying_receiver) {
1297 			/* Memory pressure, insufficient buffer
1298 			 * headroom, ...
1299 			 */
1300 			if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1301 				WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1302 					rx->id, rx->status);
1303 				xennet_move_rx_slot(np, m, ref);
1304 				err = ENOMEM;
1305 				goto next;
1306 			}
1307 
1308 			pages_flipped++;
1309 		} else {
1310 			ret = gnttab_end_foreign_access_ref(ref);
1311 			KASSERT(ret, ("ret != 0"));
1312 		}
1313 
1314 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1315 
1316 next:
1317 		if (m == NULL)
1318 			break;
1319 
1320 		m->m_len = rx->status;
1321 		m->m_data += rx->offset;
1322 		m0->m_pkthdr.len += rx->status;
1323 
1324 next_skip_queue:
1325 		if (!(rx->flags & NETRXF_more_data))
1326 			break;
1327 
1328 		if (*cons + frags == rp) {
1329 			if (net_ratelimit())
1330 				WPRINTK("Need more frags\n");
1331 			err = ENOENT;
1332 			printf("%s: cons %u frags %u rp %u, not enough frags\n",
1333 			       __func__, *cons, frags, rp);
1334 			break;
1335 		}
1336 		/*
1337 		 * Note that m can be NULL, if rx->status < 0 or if
1338 		 * rx->offset + rx->status > PAGE_SIZE above.
1339 		 */
1340 		m_prev = m;
1341 
1342 		rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1343 		m = xennet_get_rx_mbuf(np, *cons + frags);
1344 
1345 		/*
1346 		 * m_prev == NULL can happen if rx->status < 0 or if
1347 		 * rx->offset + * rx->status > PAGE_SIZE above.
1348 		 */
1349 		if (m_prev != NULL)
1350 			m_prev->m_next = m;
1351 
1352 		/*
1353 		 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1354 		 * rx->status > PAGE_SIZE above.
1355 		 */
1356 		if (m0 == NULL)
1357 			m0 = m;
1358 		m->m_next = NULL;
1359 		ref = xennet_get_rx_ref(np, *cons + frags);
1360 		ref_cons = *cons + frags;
1361 		frags++;
1362 	}
1363 	*list = m0;
1364 	*cons += frags;
1365 	*pages_flipped_p = pages_flipped;
1366 
1367 	return (err);
1368 }
1369 
1370 static void
1371 xn_tick_locked(struct netfront_info *sc)
1372 {
1373 	XN_RX_LOCK_ASSERT(sc);
1374 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1375 
1376 	/* XXX placeholder for printing debug information */
1377 }
1378 
1379 static void
1380 xn_tick(void *xsc)
1381 {
1382 	struct netfront_info *sc;
1383 
1384 	sc = xsc;
1385 	XN_RX_LOCK(sc);
1386 	xn_tick_locked(sc);
1387 	XN_RX_UNLOCK(sc);
1388 }
1389 
1390 /**
1391  * \brief Count the number of fragments in an mbuf chain.
1392  *
1393  * Surprisingly, there isn't an M* macro for this.
1394  */
1395 static inline int
1396 xn_count_frags(struct mbuf *m)
1397 {
1398 	int nfrags;
1399 
1400 	for (nfrags = 0; m != NULL; m = m->m_next)
1401 		nfrags++;
1402 
1403 	return (nfrags);
1404 }
1405 
1406 /**
1407  * Given an mbuf chain, make sure we have enough room and then push
1408  * it onto the transmit ring.
1409  */
1410 static int
1411 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1412 {
1413 	struct ifnet *ifp;
1414 	struct mbuf *m;
1415 	u_int nfrags;
1416 	int otherend_id;
1417 
1418 	ifp = sc->xn_ifp;
1419 
1420 	/**
1421 	 * Defragment the mbuf if necessary.
1422 	 */
1423 	nfrags = xn_count_frags(m_head);
1424 
1425 	/*
1426 	 * Check to see whether this request is longer than netback
1427 	 * can handle, and try to defrag it.
1428 	 */
1429 	/**
1430 	 * It is a bit lame, but the netback driver in Linux can't
1431 	 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1432 	 * the Linux network stack.
1433 	 */
1434 	if (nfrags > sc->maxfrags) {
1435 		m = m_defrag(m_head, M_NOWAIT);
1436 		if (!m) {
1437 			/*
1438 			 * Defrag failed, so free the mbuf and
1439 			 * therefore drop the packet.
1440 			 */
1441 			m_freem(m_head);
1442 			return (EMSGSIZE);
1443 		}
1444 		m_head = m;
1445 	}
1446 
1447 	/* Determine how many fragments now exist */
1448 	nfrags = xn_count_frags(m_head);
1449 
1450 	/*
1451 	 * Check to see whether the defragmented packet has too many
1452 	 * segments for the Linux netback driver.
1453 	 */
1454 	/**
1455 	 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1456 	 * of mbufs longer than Linux can handle.  Make sure we don't
1457 	 * pass a too-long chain over to the other side by dropping the
1458 	 * packet.  It doesn't look like there is currently a way to
1459 	 * tell the TCP stack to generate a shorter chain of packets.
1460 	 */
1461 	if (nfrags > MAX_TX_REQ_FRAGS) {
1462 #ifdef DEBUG
1463 		printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1464 		       "won't be able to handle it, dropping\n",
1465 		       __func__, nfrags, MAX_TX_REQ_FRAGS);
1466 #endif
1467 		m_freem(m_head);
1468 		return (EMSGSIZE);
1469 	}
1470 
1471 	/*
1472 	 * This check should be redundant.  We've already verified that we
1473 	 * have enough slots in the ring to handle a packet of maximum
1474 	 * size, and that our packet is less than the maximum size.  Keep
1475 	 * it in here as an assert for now just to make certain that
1476 	 * xn_tx_chain_cnt is accurate.
1477 	 */
1478 	KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1479 		("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1480 		 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1481                     (int) nfrags, (int) NET_TX_RING_SIZE));
1482 
1483 	/*
1484 	 * Start packing the mbufs in this chain into
1485 	 * the fragment pointers. Stop when we run out
1486 	 * of fragments or hit the end of the mbuf chain.
1487 	 */
1488 	m = m_head;
1489 	otherend_id = xenbus_get_otherend_id(sc->xbdev);
1490 	for (m = m_head; m; m = m->m_next) {
1491 		netif_tx_request_t *tx;
1492 		uintptr_t id;
1493 		grant_ref_t ref;
1494 		u_long mfn; /* XXX Wrong type? */
1495 
1496 		tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1497 		id = get_id_from_freelist(sc->tx_mbufs);
1498 		if (id == 0)
1499 			panic("%s: was allocated the freelist head!\n",
1500 			    __func__);
1501 		sc->xn_cdata.xn_tx_chain_cnt++;
1502 		if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1503 			panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1504 			    __func__);
1505 		sc->tx_mbufs[id] = m;
1506 		tx->id = id;
1507 		ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1508 		KASSERT((short)ref >= 0, ("Negative ref"));
1509 		mfn = virt_to_mfn(mtod(m, vm_offset_t));
1510 		gnttab_grant_foreign_access_ref(ref, otherend_id,
1511 		    mfn, GNTMAP_readonly);
1512 		tx->gref = sc->grant_tx_ref[id] = ref;
1513 		tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1514 		tx->flags = 0;
1515 		if (m == m_head) {
1516 			/*
1517 			 * The first fragment has the entire packet
1518 			 * size, subsequent fragments have just the
1519 			 * fragment size. The backend works out the
1520 			 * true size of the first fragment by
1521 			 * subtracting the sizes of the other
1522 			 * fragments.
1523 			 */
1524 			tx->size = m->m_pkthdr.len;
1525 
1526 			/*
1527 			 * The first fragment contains the checksum flags
1528 			 * and is optionally followed by extra data for
1529 			 * TSO etc.
1530 			 */
1531 			/**
1532 			 * CSUM_TSO requires checksum offloading.
1533 			 * Some versions of FreeBSD fail to
1534 			 * set CSUM_TCP in the CSUM_TSO case,
1535 			 * so we have to test for CSUM_TSO
1536 			 * explicitly.
1537 			 */
1538 			if (m->m_pkthdr.csum_flags
1539 			    & (CSUM_DELAY_DATA | CSUM_TSO)) {
1540 				tx->flags |= (NETTXF_csum_blank
1541 				    | NETTXF_data_validated);
1542 			}
1543 #if __FreeBSD_version >= 700000
1544 			if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1545 				struct netif_extra_info *gso =
1546 					(struct netif_extra_info *)
1547 					RING_GET_REQUEST(&sc->tx,
1548 							 ++sc->tx.req_prod_pvt);
1549 
1550 				tx->flags |= NETTXF_extra_info;
1551 
1552 				gso->u.gso.size = m->m_pkthdr.tso_segsz;
1553 				gso->u.gso.type =
1554 					XEN_NETIF_GSO_TYPE_TCPV4;
1555 				gso->u.gso.pad = 0;
1556 				gso->u.gso.features = 0;
1557 
1558 				gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1559 				gso->flags = 0;
1560 			}
1561 #endif
1562 		} else {
1563 			tx->size = m->m_len;
1564 		}
1565 		if (m->m_next)
1566 			tx->flags |= NETTXF_more_data;
1567 
1568 		sc->tx.req_prod_pvt++;
1569 	}
1570 	BPF_MTAP(ifp, m_head);
1571 
1572 	sc->stats.tx_bytes += m_head->m_pkthdr.len;
1573 	sc->stats.tx_packets++;
1574 
1575 	return (0);
1576 }
1577 
1578 static void
1579 xn_start_locked(struct ifnet *ifp)
1580 {
1581 	struct netfront_info *sc;
1582 	struct mbuf *m_head;
1583 	int notify;
1584 
1585 	sc = ifp->if_softc;
1586 
1587 	if (!netfront_carrier_ok(sc))
1588 		return;
1589 
1590 	/*
1591 	 * While we have enough transmit slots available for at least one
1592 	 * maximum-sized packet, pull mbufs off the queue and put them on
1593 	 * the transmit ring.
1594 	 */
1595 	while (xn_tx_slot_available(sc)) {
1596 		IF_DEQUEUE(&ifp->if_snd, m_head);
1597 		if (m_head == NULL)
1598 			break;
1599 
1600 		if (xn_assemble_tx_request(sc, m_head) != 0)
1601 			break;
1602 	}
1603 
1604 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1605 	if (notify)
1606 		xen_intr_signal(sc->xen_intr_handle);
1607 
1608 	if (RING_FULL(&sc->tx)) {
1609 		sc->tx_full = 1;
1610 #if 0
1611 		netif_stop_queue(dev);
1612 #endif
1613 	}
1614 }
1615 
1616 static void
1617 xn_start(struct ifnet *ifp)
1618 {
1619 	struct netfront_info *sc;
1620 	sc = ifp->if_softc;
1621 	XN_TX_LOCK(sc);
1622 	xn_start_locked(ifp);
1623 	XN_TX_UNLOCK(sc);
1624 }
1625 
1626 /* equivalent of network_open() in Linux */
1627 static void
1628 xn_ifinit_locked(struct netfront_info *sc)
1629 {
1630 	struct ifnet *ifp;
1631 
1632 	XN_LOCK_ASSERT(sc);
1633 
1634 	ifp = sc->xn_ifp;
1635 
1636 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1637 		return;
1638 
1639 	xn_stop(sc);
1640 
1641 	network_alloc_rx_buffers(sc);
1642 	sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1643 
1644 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1645 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1646 	if_link_state_change(ifp, LINK_STATE_UP);
1647 
1648 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1649 }
1650 
1651 static void
1652 xn_ifinit(void *xsc)
1653 {
1654 	struct netfront_info *sc = xsc;
1655 
1656 	XN_LOCK(sc);
1657 	xn_ifinit_locked(sc);
1658 	XN_UNLOCK(sc);
1659 }
1660 
1661 static int
1662 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1663 {
1664 	struct netfront_info *sc = ifp->if_softc;
1665 	struct ifreq *ifr = (struct ifreq *) data;
1666 #ifdef INET
1667 	struct ifaddr *ifa = (struct ifaddr *)data;
1668 #endif
1669 
1670 	int mask, error = 0;
1671 	switch(cmd) {
1672 	case SIOCSIFADDR:
1673 #ifdef INET
1674 		XN_LOCK(sc);
1675 		if (ifa->ifa_addr->sa_family == AF_INET) {
1676 			ifp->if_flags |= IFF_UP;
1677 			if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1678 				xn_ifinit_locked(sc);
1679 			arp_ifinit(ifp, ifa);
1680 			XN_UNLOCK(sc);
1681 		} else {
1682 			XN_UNLOCK(sc);
1683 #endif
1684 			error = ether_ioctl(ifp, cmd, data);
1685 #ifdef INET
1686 		}
1687 #endif
1688 		break;
1689 	case SIOCSIFMTU:
1690 		/* XXX can we alter the MTU on a VN ?*/
1691 #ifdef notyet
1692 		if (ifr->ifr_mtu > XN_JUMBO_MTU)
1693 			error = EINVAL;
1694 		else
1695 #endif
1696 		{
1697 			ifp->if_mtu = ifr->ifr_mtu;
1698 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1699 			xn_ifinit(sc);
1700 		}
1701 		break;
1702 	case SIOCSIFFLAGS:
1703 		XN_LOCK(sc);
1704 		if (ifp->if_flags & IFF_UP) {
1705 			/*
1706 			 * If only the state of the PROMISC flag changed,
1707 			 * then just use the 'set promisc mode' command
1708 			 * instead of reinitializing the entire NIC. Doing
1709 			 * a full re-init means reloading the firmware and
1710 			 * waiting for it to start up, which may take a
1711 			 * second or two.
1712 			 */
1713 #ifdef notyet
1714 			/* No promiscuous mode with Xen */
1715 			if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1716 			    ifp->if_flags & IFF_PROMISC &&
1717 			    !(sc->xn_if_flags & IFF_PROMISC)) {
1718 				XN_SETBIT(sc, XN_RX_MODE,
1719 					  XN_RXMODE_RX_PROMISC);
1720 			} else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1721 				   !(ifp->if_flags & IFF_PROMISC) &&
1722 				   sc->xn_if_flags & IFF_PROMISC) {
1723 				XN_CLRBIT(sc, XN_RX_MODE,
1724 					  XN_RXMODE_RX_PROMISC);
1725 			} else
1726 #endif
1727 				xn_ifinit_locked(sc);
1728 		} else {
1729 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1730 				xn_stop(sc);
1731 			}
1732 		}
1733 		sc->xn_if_flags = ifp->if_flags;
1734 		XN_UNLOCK(sc);
1735 		error = 0;
1736 		break;
1737 	case SIOCSIFCAP:
1738 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1739 		if (mask & IFCAP_TXCSUM) {
1740 			if (IFCAP_TXCSUM & ifp->if_capenable) {
1741 				ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1742 				ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1743 				    | CSUM_IP | CSUM_TSO);
1744 			} else {
1745 				ifp->if_capenable |= IFCAP_TXCSUM;
1746 				ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1747 				    | CSUM_IP);
1748 			}
1749 		}
1750 		if (mask & IFCAP_RXCSUM) {
1751 			ifp->if_capenable ^= IFCAP_RXCSUM;
1752 		}
1753 #if __FreeBSD_version >= 700000
1754 		if (mask & IFCAP_TSO4) {
1755 			if (IFCAP_TSO4 & ifp->if_capenable) {
1756 				ifp->if_capenable &= ~IFCAP_TSO4;
1757 				ifp->if_hwassist &= ~CSUM_TSO;
1758 			} else if (IFCAP_TXCSUM & ifp->if_capenable) {
1759 				ifp->if_capenable |= IFCAP_TSO4;
1760 				ifp->if_hwassist |= CSUM_TSO;
1761 			} else {
1762 				IPRINTK("Xen requires tx checksum offload"
1763 				    " be enabled to use TSO\n");
1764 				error = EINVAL;
1765 			}
1766 		}
1767 		if (mask & IFCAP_LRO) {
1768 			ifp->if_capenable ^= IFCAP_LRO;
1769 
1770 		}
1771 #endif
1772 		error = 0;
1773 		break;
1774 	case SIOCADDMULTI:
1775 	case SIOCDELMULTI:
1776 #ifdef notyet
1777 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1778 			XN_LOCK(sc);
1779 			xn_setmulti(sc);
1780 			XN_UNLOCK(sc);
1781 			error = 0;
1782 		}
1783 #endif
1784 		/* FALLTHROUGH */
1785 	case SIOCSIFMEDIA:
1786 	case SIOCGIFMEDIA:
1787 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1788 		break;
1789 	default:
1790 		error = ether_ioctl(ifp, cmd, data);
1791 	}
1792 
1793 	return (error);
1794 }
1795 
1796 static void
1797 xn_stop(struct netfront_info *sc)
1798 {
1799 	struct ifnet *ifp;
1800 
1801 	XN_LOCK_ASSERT(sc);
1802 
1803 	ifp = sc->xn_ifp;
1804 
1805 	callout_stop(&sc->xn_stat_ch);
1806 
1807 	xn_free_rx_ring(sc);
1808 	xn_free_tx_ring(sc);
1809 
1810 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1811 	if_link_state_change(ifp, LINK_STATE_DOWN);
1812 }
1813 
1814 /* START of Xenolinux helper functions adapted to FreeBSD */
1815 int
1816 network_connect(struct netfront_info *np)
1817 {
1818 	int i, requeue_idx, error;
1819 	grant_ref_t ref;
1820 	netif_rx_request_t *req;
1821 	u_int feature_rx_copy, feature_rx_flip;
1822 
1823 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1824 	    "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1825 	if (error)
1826 		feature_rx_copy = 0;
1827 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1828 	    "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1829 	if (error)
1830 		feature_rx_flip = 1;
1831 
1832 	/*
1833 	 * Copy packets on receive path if:
1834 	 *  (a) This was requested by user, and the backend supports it; or
1835 	 *  (b) Flipping was requested, but this is unsupported by the backend.
1836 	 */
1837 	np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1838 				(MODPARM_rx_flip && !feature_rx_flip));
1839 
1840 	/* Recovery procedure: */
1841 	error = talk_to_backend(np->xbdev, np);
1842 	if (error)
1843 		return (error);
1844 
1845 	/* Step 1: Reinitialise variables. */
1846 	xn_query_features(np);
1847 	xn_configure_features(np);
1848 	netif_release_tx_bufs(np);
1849 
1850 	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1851 	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1852 		struct mbuf *m;
1853 		u_long pfn;
1854 
1855 		if (np->rx_mbufs[i] == NULL)
1856 			continue;
1857 
1858 		m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1859 		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1860 
1861 		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1862 		pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1863 
1864 		if (!np->copying_receiver) {
1865 			gnttab_grant_foreign_transfer_ref(ref,
1866 			    xenbus_get_otherend_id(np->xbdev),
1867 			    pfn);
1868 		} else {
1869 			gnttab_grant_foreign_access_ref(ref,
1870 			    xenbus_get_otherend_id(np->xbdev),
1871 			    pfn, 0);
1872 		}
1873 		req->gref = ref;
1874 		req->id   = requeue_idx;
1875 
1876 		requeue_idx++;
1877 	}
1878 
1879 	np->rx.req_prod_pvt = requeue_idx;
1880 
1881 	/* Step 3: All public and private state should now be sane.  Get
1882 	 * ready to start sending and receiving packets and give the driver
1883 	 * domain a kick because we've probably just requeued some
1884 	 * packets.
1885 	 */
1886 	netfront_carrier_on(np);
1887 	xen_intr_signal(np->xen_intr_handle);
1888 	XN_TX_LOCK(np);
1889 	xn_txeof(np);
1890 	XN_TX_UNLOCK(np);
1891 	network_alloc_rx_buffers(np);
1892 
1893 	return (0);
1894 }
1895 
1896 static void
1897 xn_query_features(struct netfront_info *np)
1898 {
1899 	int val;
1900 
1901 	device_printf(np->xbdev, "backend features:");
1902 
1903 	if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1904 		"feature-sg", NULL, "%d", &val) < 0)
1905 		val = 0;
1906 
1907 	np->maxfrags = 1;
1908 	if (val) {
1909 		np->maxfrags = MAX_TX_REQ_FRAGS;
1910 		printf(" feature-sg");
1911 	}
1912 
1913 	if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1914 		"feature-gso-tcpv4", NULL, "%d", &val) < 0)
1915 		val = 0;
1916 
1917 	np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
1918 	if (val) {
1919 		np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
1920 		printf(" feature-gso-tcp4");
1921 	}
1922 
1923 	printf("\n");
1924 }
1925 
1926 static int
1927 xn_configure_features(struct netfront_info *np)
1928 {
1929 	int err, cap_enabled;
1930 
1931 	err = 0;
1932 
1933 	if (np->xn_resume &&
1934 	    ((np->xn_ifp->if_capenable & np->xn_ifp->if_capabilities)
1935 	    == np->xn_ifp->if_capenable)) {
1936 		/* Current options are available, no need to do anything. */
1937 		return (0);
1938 	}
1939 
1940 	/* Try to preserve as many options as possible. */
1941 	if (np->xn_resume)
1942 		cap_enabled = np->xn_ifp->if_capenable;
1943 	else
1944 		cap_enabled = UINT_MAX;
1945 
1946 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1947 	if ((np->xn_ifp->if_capenable & IFCAP_LRO) == (cap_enabled & IFCAP_LRO))
1948 		tcp_lro_free(&np->xn_lro);
1949 #endif
1950     	np->xn_ifp->if_capenable =
1951 	    np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4) & cap_enabled;
1952 	np->xn_ifp->if_hwassist &= ~CSUM_TSO;
1953 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1954 	if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) ==
1955 	    (cap_enabled & IFCAP_LRO)) {
1956 		err = tcp_lro_init(&np->xn_lro);
1957 		if (err) {
1958 			device_printf(np->xbdev, "LRO initialization failed\n");
1959 		} else {
1960 			np->xn_lro.ifp = np->xn_ifp;
1961 			np->xn_ifp->if_capenable |= IFCAP_LRO;
1962 		}
1963 	}
1964 	if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) ==
1965 	    (cap_enabled & IFCAP_TSO4)) {
1966 		np->xn_ifp->if_capenable |= IFCAP_TSO4;
1967 		np->xn_ifp->if_hwassist |= CSUM_TSO;
1968 	}
1969 #endif
1970 	return (err);
1971 }
1972 
1973 /**
1974  * Create a network device.
1975  * @param dev  Newbus device representing this virtual NIC.
1976  */
1977 int
1978 create_netdev(device_t dev)
1979 {
1980 	int i;
1981 	struct netfront_info *np;
1982 	int err;
1983 	struct ifnet *ifp;
1984 
1985 	np = device_get_softc(dev);
1986 
1987 	np->xbdev         = dev;
1988 
1989 	XN_LOCK_INIT(np, xennetif);
1990 
1991 	ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
1992 	ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
1993 	ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
1994 
1995 	np->rx_target     = RX_MIN_TARGET;
1996 	np->rx_min_target = RX_MIN_TARGET;
1997 	np->rx_max_target = RX_MAX_TARGET;
1998 
1999 	/* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2000 	for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2001 		np->tx_mbufs[i] = (void *) ((u_long) i+1);
2002 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
2003 	}
2004 	np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2005 
2006 	for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2007 
2008 		np->rx_mbufs[i] = NULL;
2009 		np->grant_rx_ref[i] = GRANT_REF_INVALID;
2010 	}
2011 
2012 	mbufq_init(&np->xn_rx_batch, INT_MAX);
2013 
2014 	/* A grant for every tx ring slot */
2015 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2016 					  &np->gref_tx_head) != 0) {
2017 		IPRINTK("#### netfront can't alloc tx grant refs\n");
2018 		err = ENOMEM;
2019 		goto exit;
2020 	}
2021 	/* A grant for every rx ring slot */
2022 	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2023 					  &np->gref_rx_head) != 0) {
2024 		WPRINTK("#### netfront can't alloc rx grant refs\n");
2025 		gnttab_free_grant_references(np->gref_tx_head);
2026 		err = ENOMEM;
2027 		goto exit;
2028 	}
2029 
2030 	err = xen_net_read_mac(dev, np->mac);
2031 	if (err)
2032 		goto out;
2033 
2034 	/* Set up ifnet structure */
2035 	ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2036     	ifp->if_softc = np;
2037     	if_initname(ifp, "xn",  device_get_unit(dev));
2038     	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2039     	ifp->if_ioctl = xn_ioctl;
2040     	ifp->if_output = ether_output;
2041     	ifp->if_start = xn_start;
2042 #ifdef notyet
2043     	ifp->if_watchdog = xn_watchdog;
2044 #endif
2045     	ifp->if_init = xn_ifinit;
2046     	ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2047 
2048     	ifp->if_hwassist = XN_CSUM_FEATURES;
2049     	ifp->if_capabilities = IFCAP_HWCSUM;
2050 	ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2051 	ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS;
2052 	ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
2053 
2054     	ether_ifattach(ifp, np->mac);
2055     	callout_init(&np->xn_stat_ch, 1);
2056 	netfront_carrier_off(np);
2057 
2058 	return (0);
2059 
2060 exit:
2061 	gnttab_free_grant_references(np->gref_tx_head);
2062 out:
2063 	return (err);
2064 }
2065 
2066 /**
2067  * Handle the change of state of the backend to Closing.  We must delete our
2068  * device-layer structures now, to ensure that writes are flushed through to
2069  * the backend.  Once is this done, we can switch to Closed in
2070  * acknowledgement.
2071  */
2072 #if 0
2073 static void
2074 netfront_closing(device_t dev)
2075 {
2076 #if 0
2077 	struct netfront_info *info = dev->dev_driver_data;
2078 
2079 	DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2080 
2081 	close_netdev(info);
2082 #endif
2083 	xenbus_switch_state(dev, XenbusStateClosed);
2084 }
2085 #endif
2086 
2087 static int
2088 netfront_detach(device_t dev)
2089 {
2090 	struct netfront_info *info = device_get_softc(dev);
2091 
2092 	DPRINTK("%s\n", xenbus_get_node(dev));
2093 
2094 	netif_free(info);
2095 
2096 	return 0;
2097 }
2098 
2099 static void
2100 netif_free(struct netfront_info *info)
2101 {
2102 	XN_LOCK(info);
2103 	xn_stop(info);
2104 	XN_UNLOCK(info);
2105 	callout_drain(&info->xn_stat_ch);
2106 	netif_disconnect_backend(info);
2107 	if (info->xn_ifp != NULL) {
2108 		ether_ifdetach(info->xn_ifp);
2109 		if_free(info->xn_ifp);
2110 		info->xn_ifp = NULL;
2111 	}
2112 	ifmedia_removeall(&info->sc_media);
2113 }
2114 
2115 static void
2116 netif_disconnect_backend(struct netfront_info *info)
2117 {
2118 	XN_RX_LOCK(info);
2119 	XN_TX_LOCK(info);
2120 	netfront_carrier_off(info);
2121 	XN_TX_UNLOCK(info);
2122 	XN_RX_UNLOCK(info);
2123 
2124 	free_ring(&info->tx_ring_ref, &info->tx.sring);
2125 	free_ring(&info->rx_ring_ref, &info->rx.sring);
2126 
2127 	xen_intr_unbind(&info->xen_intr_handle);
2128 }
2129 
2130 static void
2131 free_ring(int *ref, void *ring_ptr_ref)
2132 {
2133 	void **ring_ptr_ptr = ring_ptr_ref;
2134 
2135 	if (*ref != GRANT_REF_INVALID) {
2136 		/* This API frees the associated storage. */
2137 		gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
2138 		*ref = GRANT_REF_INVALID;
2139 	}
2140 	*ring_ptr_ptr = NULL;
2141 }
2142 
2143 static int
2144 xn_ifmedia_upd(struct ifnet *ifp)
2145 {
2146 	return (0);
2147 }
2148 
2149 static void
2150 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2151 {
2152 	ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2153 	ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2154 }
2155 
2156 /* ** Driver registration ** */
2157 static device_method_t netfront_methods[] = {
2158 	/* Device interface */
2159 	DEVMETHOD(device_probe,         netfront_probe),
2160 	DEVMETHOD(device_attach,        netfront_attach),
2161 	DEVMETHOD(device_detach,        netfront_detach),
2162 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
2163 	DEVMETHOD(device_suspend,       netfront_suspend),
2164 	DEVMETHOD(device_resume,        netfront_resume),
2165 
2166 	/* Xenbus interface */
2167 	DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2168 
2169 	DEVMETHOD_END
2170 };
2171 
2172 static driver_t netfront_driver = {
2173 	"xn",
2174 	netfront_methods,
2175 	sizeof(struct netfront_info),
2176 };
2177 devclass_t netfront_devclass;
2178 
2179 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,
2180     NULL);
2181