xref: /freebsd/sys/dev/xen/netfront/netfront.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/sockio.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 #include <sys/queue.h>
43 #include <sys/lock.h>
44 #include <sys/sx.h>
45 
46 #include <net/if.h>
47 #include <net/if_arp.h>
48 #include <net/ethernet.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
51 
52 #include <net/bpf.h>
53 
54 #include <net/if_types.h>
55 #include <net/if.h>
56 
57 #include <netinet/in_systm.h>
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #include <netinet/if_ether.h>
61 #if __FreeBSD_version >= 700000
62 #include <netinet/tcp.h>
63 #include <netinet/tcp_lro.h>
64 #endif
65 
66 #include <vm/vm.h>
67 #include <vm/pmap.h>
68 
69 #include <machine/clock.h>      /* for DELAY */
70 #include <machine/bus.h>
71 #include <machine/resource.h>
72 #include <machine/frame.h>
73 #include <machine/vmparam.h>
74 
75 #include <sys/bus.h>
76 #include <sys/rman.h>
77 
78 #include <machine/intr_machdep.h>
79 
80 #include <machine/xen/xen-os.h>
81 #include <machine/xen/xenfunc.h>
82 #include <machine/xen/xenvar.h>
83 #include <xen/hypervisor.h>
84 #include <xen/xen_intr.h>
85 #include <xen/evtchn.h>
86 #include <xen/gnttab.h>
87 #include <xen/interface/memory.h>
88 #include <xen/interface/io/netif.h>
89 #include <xen/xenbus/xenbusvar.h>
90 
91 #include <dev/xen/netfront/mbufq.h>
92 
93 #include "xenbus_if.h"
94 
95 #define XN_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP | CSUM_TSO)
96 
97 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
98 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
99 
100 #if __FreeBSD_version >= 700000
101 /*
102  * Should the driver do LRO on the RX end
103  *  this can be toggled on the fly, but the
104  *  interface must be reset (down/up) for it
105  *  to take effect.
106  */
107 static int xn_enable_lro = 1;
108 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
109 #else
110 
111 #define IFCAP_TSO4	0
112 #define CSUM_TSO	0
113 
114 #endif
115 
116 #ifdef CONFIG_XEN
117 static int MODPARM_rx_copy = 0;
118 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0);
119 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)");
120 static int MODPARM_rx_flip = 0;
121 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0);
122 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)");
123 #else
124 static const int MODPARM_rx_copy = 1;
125 static const int MODPARM_rx_flip = 0;
126 #endif
127 
128 /**
129  * \brief The maximum allowed data fragments in a single transmit
130  *        request.
131  *
132  * This limit is imposed by the backend driver.  We assume here that
133  * we are dealing with a Linux driver domain and have set our limit
134  * to mirror the Linux MAX_SKB_FRAGS constant.
135  */
136 #define	MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2)
137 
138 #define RX_COPY_THRESHOLD 256
139 
140 #define net_ratelimit() 0
141 
142 struct netfront_info;
143 struct netfront_rx_info;
144 
145 static void xn_txeof(struct netfront_info *);
146 static void xn_rxeof(struct netfront_info *);
147 static void network_alloc_rx_buffers(struct netfront_info *);
148 
149 static void xn_tick_locked(struct netfront_info *);
150 static void xn_tick(void *);
151 
152 static void xn_intr(void *);
153 static inline int xn_count_frags(struct mbuf *m);
154 static int  xn_assemble_tx_request(struct netfront_info *sc,
155 				   struct mbuf *m_head);
156 static void xn_start_locked(struct ifnet *);
157 static void xn_start(struct ifnet *);
158 static int  xn_ioctl(struct ifnet *, u_long, caddr_t);
159 static void xn_ifinit_locked(struct netfront_info *);
160 static void xn_ifinit(void *);
161 static void xn_stop(struct netfront_info *);
162 #ifdef notyet
163 static void xn_watchdog(struct ifnet *);
164 #endif
165 
166 static void show_device(struct netfront_info *sc);
167 #ifdef notyet
168 static void netfront_closing(device_t dev);
169 #endif
170 static void netif_free(struct netfront_info *info);
171 static int netfront_detach(device_t dev);
172 
173 static int talk_to_backend(device_t dev, struct netfront_info *info);
174 static int create_netdev(device_t dev);
175 static void netif_disconnect_backend(struct netfront_info *info);
176 static int setup_device(device_t dev, struct netfront_info *info);
177 static void end_access(int ref, void *page);
178 
179 static int  xn_ifmedia_upd(struct ifnet *ifp);
180 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
181 
182 /* Xenolinux helper functions */
183 int network_connect(struct netfront_info *);
184 
185 static void xn_free_rx_ring(struct netfront_info *);
186 
187 static void xn_free_tx_ring(struct netfront_info *);
188 
189 static int xennet_get_responses(struct netfront_info *np,
190 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
191 	struct mbuf **list, int *pages_flipped_p);
192 
193 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
194 
195 #define INVALID_P2M_ENTRY (~0UL)
196 
197 /*
198  * Mbuf pointers. We need these to keep track of the virtual addresses
199  * of our mbuf chains since we can only convert from virtual to physical,
200  * not the other way around.  The size must track the free index arrays.
201  */
202 struct xn_chain_data {
203 	struct mbuf    *xn_tx_chain[NET_TX_RING_SIZE+1];
204 	int		xn_tx_chain_cnt;
205 	struct mbuf    *xn_rx_chain[NET_RX_RING_SIZE+1];
206 };
207 
208 #define NUM_ELEMENTS(x) (sizeof(x)/sizeof(*x))
209 
210 struct net_device_stats
211 {
212 	u_long	rx_packets;		/* total packets received	*/
213 	u_long	tx_packets;		/* total packets transmitted	*/
214 	u_long	rx_bytes;		/* total bytes received 	*/
215 	u_long	tx_bytes;		/* total bytes transmitted	*/
216 	u_long	rx_errors;		/* bad packets received		*/
217 	u_long	tx_errors;		/* packet transmit problems	*/
218 	u_long	rx_dropped;		/* no space in linux buffers	*/
219 	u_long	tx_dropped;		/* no space available in linux	*/
220 	u_long	multicast;		/* multicast packets received	*/
221 	u_long	collisions;
222 
223 	/* detailed rx_errors: */
224 	u_long	rx_length_errors;
225 	u_long	rx_over_errors;		/* receiver ring buff overflow	*/
226 	u_long	rx_crc_errors;		/* recved pkt with crc error	*/
227 	u_long	rx_frame_errors;	/* recv'd frame alignment error */
228 	u_long	rx_fifo_errors;		/* recv'r fifo overrun		*/
229 	u_long	rx_missed_errors;	/* receiver missed packet	*/
230 
231 	/* detailed tx_errors */
232 	u_long	tx_aborted_errors;
233 	u_long	tx_carrier_errors;
234 	u_long	tx_fifo_errors;
235 	u_long	tx_heartbeat_errors;
236 	u_long	tx_window_errors;
237 
238 	/* for cslip etc */
239 	u_long	rx_compressed;
240 	u_long	tx_compressed;
241 };
242 
243 struct netfront_info {
244 
245 	struct ifnet *xn_ifp;
246 #if __FreeBSD_version >= 700000
247 	struct lro_ctrl xn_lro;
248 #endif
249 
250 	struct net_device_stats stats;
251 	u_int tx_full;
252 
253 	netif_tx_front_ring_t tx;
254 	netif_rx_front_ring_t rx;
255 
256 	struct mtx   tx_lock;
257 	struct mtx   rx_lock;
258 	struct mtx   sc_lock;
259 
260 	u_int handle;
261 	u_int irq;
262 	u_int copying_receiver;
263 	u_int carrier;
264 
265 	/* Receive-ring batched refills. */
266 #define RX_MIN_TARGET 32
267 #define RX_MAX_TARGET NET_RX_RING_SIZE
268 	int rx_min_target;
269 	int rx_max_target;
270 	int rx_target;
271 
272 	grant_ref_t gref_tx_head;
273 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
274 	grant_ref_t gref_rx_head;
275 	grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
276 
277 	device_t		xbdev;
278 	int			tx_ring_ref;
279 	int			rx_ring_ref;
280 	uint8_t			mac[ETHER_ADDR_LEN];
281 	struct xn_chain_data	xn_cdata;	/* mbufs */
282 	struct mbuf_head	xn_rx_batch;	/* head of the batch queue */
283 
284 	int			xn_if_flags;
285 	struct callout	        xn_stat_ch;
286 
287 	u_long			rx_pfn_array[NET_RX_RING_SIZE];
288 	multicall_entry_t	rx_mcl[NET_RX_RING_SIZE+1];
289 	mmu_update_t		rx_mmu[NET_RX_RING_SIZE];
290 	struct ifmedia		sc_media;
291 };
292 
293 #define rx_mbufs xn_cdata.xn_rx_chain
294 #define tx_mbufs xn_cdata.xn_tx_chain
295 
296 #define XN_LOCK_INIT(_sc, _name) \
297         mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
298         mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF);  \
299         mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF)
300 
301 #define XN_RX_LOCK(_sc)           mtx_lock(&(_sc)->rx_lock)
302 #define XN_RX_UNLOCK(_sc)         mtx_unlock(&(_sc)->rx_lock)
303 
304 #define XN_TX_LOCK(_sc)           mtx_lock(&(_sc)->tx_lock)
305 #define XN_TX_UNLOCK(_sc)         mtx_unlock(&(_sc)->tx_lock)
306 
307 #define XN_LOCK(_sc)           mtx_lock(&(_sc)->sc_lock);
308 #define XN_UNLOCK(_sc)         mtx_unlock(&(_sc)->sc_lock);
309 
310 #define XN_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->sc_lock, MA_OWNED);
311 #define XN_RX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->rx_lock, MA_OWNED);
312 #define XN_TX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->tx_lock, MA_OWNED);
313 #define XN_LOCK_DESTROY(_sc)   mtx_destroy(&(_sc)->rx_lock); \
314                                mtx_destroy(&(_sc)->tx_lock); \
315                                mtx_destroy(&(_sc)->sc_lock);
316 
317 struct netfront_rx_info {
318 	struct netif_rx_response rx;
319 	struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
320 };
321 
322 #define netfront_carrier_on(netif)	((netif)->carrier = 1)
323 #define netfront_carrier_off(netif)	((netif)->carrier = 0)
324 #define netfront_carrier_ok(netif)	((netif)->carrier)
325 
326 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
327 
328 
329 
330 /*
331  * Access macros for acquiring freeing slots in tx_skbs[].
332  */
333 
334 static inline void
335 add_id_to_freelist(struct mbuf **list, uintptr_t id)
336 {
337 	KASSERT(id != 0,
338 		("%s: the head item (0) must always be free.", __func__));
339 	list[id] = list[0];
340 	list[0]  = (struct mbuf *)id;
341 }
342 
343 static inline unsigned short
344 get_id_from_freelist(struct mbuf **list)
345 {
346 	uintptr_t id;
347 
348 	id = (uintptr_t)list[0];
349 	KASSERT(id != 0,
350 		("%s: the head item (0) must always remain free.", __func__));
351 	list[0] = list[id];
352 	return (id);
353 }
354 
355 static inline int
356 xennet_rxidx(RING_IDX idx)
357 {
358 	return idx & (NET_RX_RING_SIZE - 1);
359 }
360 
361 static inline struct mbuf *
362 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
363 {
364 	int i = xennet_rxidx(ri);
365 	struct mbuf *m;
366 
367 	m = np->rx_mbufs[i];
368 	np->rx_mbufs[i] = NULL;
369 	return (m);
370 }
371 
372 static inline grant_ref_t
373 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
374 {
375 	int i = xennet_rxidx(ri);
376 	grant_ref_t ref = np->grant_rx_ref[i];
377 	KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
378 	np->grant_rx_ref[i] = GRANT_REF_INVALID;
379 	return ref;
380 }
381 
382 #define IPRINTK(fmt, args...) \
383     printf("[XEN] " fmt, ##args)
384 #ifdef INVARIANTS
385 #define WPRINTK(fmt, args...) \
386     printf("[XEN] " fmt, ##args)
387 #else
388 #define WPRINTK(fmt, args...)
389 #endif
390 #ifdef DEBUG
391 #define DPRINTK(fmt, args...) \
392     printf("[XEN] %s: " fmt, __func__, ##args)
393 #else
394 #define DPRINTK(fmt, args...)
395 #endif
396 
397 /**
398  * Read the 'mac' node at the given device's node in the store, and parse that
399  * as colon-separated octets, placing result the given mac array.  mac must be
400  * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
401  * Return 0 on success, or errno on error.
402  */
403 static int
404 xen_net_read_mac(device_t dev, uint8_t mac[])
405 {
406 	int error, i;
407 	char *s, *e, *macstr;
408 
409 	error = xs_read(XST_NIL, xenbus_get_node(dev), "mac", NULL,
410 	    (void **) &macstr);
411 	if (error)
412 		return (error);
413 
414 	s = macstr;
415 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
416 		mac[i] = strtoul(s, &e, 16);
417 		if (s == e || (e[0] != ':' && e[0] != 0)) {
418 			free(macstr, M_XENBUS);
419 			return (ENOENT);
420 		}
421 		s = &e[1];
422 	}
423 	free(macstr, M_XENBUS);
424 	return (0);
425 }
426 
427 /**
428  * Entry point to this code when a new device is created.  Allocate the basic
429  * structures and the ring buffers for communication with the backend, and
430  * inform the backend of the appropriate details for those.  Switch to
431  * Connected state.
432  */
433 static int
434 netfront_probe(device_t dev)
435 {
436 
437 	if (!strcmp(xenbus_get_type(dev), "vif")) {
438 		device_set_desc(dev, "Virtual Network Interface");
439 		return (0);
440 	}
441 
442 	return (ENXIO);
443 }
444 
445 static int
446 netfront_attach(device_t dev)
447 {
448 	int err;
449 
450 	err = create_netdev(dev);
451 	if (err) {
452 		xenbus_dev_fatal(dev, err, "creating netdev");
453 		return err;
454 	}
455 
456 #if __FreeBSD_version >= 700000
457 	SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
458 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
459 	    OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
460 	    &xn_enable_lro, 0, "Large Receive Offload");
461 #endif
462 
463 	return 0;
464 }
465 
466 
467 /**
468  * We are reconnecting to the backend, due to a suspend/resume, or a backend
469  * driver restart.  We tear down our netif structure and recreate it, but
470  * leave the device-layer structures intact so that this is transparent to the
471  * rest of the kernel.
472  */
473 static int
474 netfront_resume(device_t dev)
475 {
476 	struct netfront_info *info = device_get_softc(dev);
477 
478 	netif_disconnect_backend(info);
479 	return (0);
480 }
481 
482 
483 /* Common code used when first setting up, and when resuming. */
484 static int
485 talk_to_backend(device_t dev, struct netfront_info *info)
486 {
487 	const char *message;
488 	struct xs_transaction xst;
489 	const char *node = xenbus_get_node(dev);
490 	int err;
491 
492 	err = xen_net_read_mac(dev, info->mac);
493 	if (err) {
494 		xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
495 		goto out;
496 	}
497 
498 	/* Create shared ring, alloc event channel. */
499 	err = setup_device(dev, info);
500 	if (err)
501 		goto out;
502 
503  again:
504 	err = xs_transaction_start(&xst);
505 	if (err) {
506 		xenbus_dev_fatal(dev, err, "starting transaction");
507 		goto destroy_ring;
508 	}
509 	err = xs_printf(xst, node, "tx-ring-ref","%u",
510 			info->tx_ring_ref);
511 	if (err) {
512 		message = "writing tx ring-ref";
513 		goto abort_transaction;
514 	}
515 	err = xs_printf(xst, node, "rx-ring-ref","%u",
516 			info->rx_ring_ref);
517 	if (err) {
518 		message = "writing rx ring-ref";
519 		goto abort_transaction;
520 	}
521 	err = xs_printf(xst, node,
522 			"event-channel", "%u", irq_to_evtchn_port(info->irq));
523 	if (err) {
524 		message = "writing event-channel";
525 		goto abort_transaction;
526 	}
527 	err = xs_printf(xst, node, "request-rx-copy", "%u",
528 			info->copying_receiver);
529 	if (err) {
530 		message = "writing request-rx-copy";
531 		goto abort_transaction;
532 	}
533 	err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
534 	if (err) {
535 		message = "writing feature-rx-notify";
536 		goto abort_transaction;
537 	}
538 	err = xs_printf(xst, node, "feature-sg", "%d", 1);
539 	if (err) {
540 		message = "writing feature-sg";
541 		goto abort_transaction;
542 	}
543 #if __FreeBSD_version >= 700000
544 	err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
545 	if (err) {
546 		message = "writing feature-gso-tcpv4";
547 		goto abort_transaction;
548 	}
549 #endif
550 
551 	err = xs_transaction_end(xst, 0);
552 	if (err) {
553 		if (err == EAGAIN)
554 			goto again;
555 		xenbus_dev_fatal(dev, err, "completing transaction");
556 		goto destroy_ring;
557 	}
558 
559 	return 0;
560 
561  abort_transaction:
562 	xs_transaction_end(xst, 1);
563 	xenbus_dev_fatal(dev, err, "%s", message);
564  destroy_ring:
565 	netif_free(info);
566  out:
567 	return err;
568 }
569 
570 
571 static int
572 setup_device(device_t dev, struct netfront_info *info)
573 {
574 	netif_tx_sring_t *txs;
575 	netif_rx_sring_t *rxs;
576 	int error;
577 	struct ifnet *ifp;
578 
579 	ifp = info->xn_ifp;
580 
581 	info->tx_ring_ref = GRANT_REF_INVALID;
582 	info->rx_ring_ref = GRANT_REF_INVALID;
583 	info->rx.sring = NULL;
584 	info->tx.sring = NULL;
585 	info->irq = 0;
586 
587 	txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
588 	if (!txs) {
589 		error = ENOMEM;
590 		xenbus_dev_fatal(dev, error, "allocating tx ring page");
591 		goto fail;
592 	}
593 	SHARED_RING_INIT(txs);
594 	FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
595 	error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
596 	if (error)
597 		goto fail;
598 
599 	rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
600 	if (!rxs) {
601 		error = ENOMEM;
602 		xenbus_dev_fatal(dev, error, "allocating rx ring page");
603 		goto fail;
604 	}
605 	SHARED_RING_INIT(rxs);
606 	FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
607 
608 	error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
609 	if (error)
610 		goto fail;
611 
612 	error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
613 	    "xn", xn_intr, info, INTR_TYPE_NET | INTR_MPSAFE, &info->irq);
614 
615 	if (error) {
616 		xenbus_dev_fatal(dev, error,
617 				 "bind_evtchn_to_irqhandler failed");
618 		goto fail;
619 	}
620 
621 	show_device(info);
622 
623 	return (0);
624 
625  fail:
626 	netif_free(info);
627 	return (error);
628 }
629 
630 #ifdef INET
631 /**
632  * If this interface has an ipv4 address, send an arp for it. This
633  * helps to get the network going again after migrating hosts.
634  */
635 static void
636 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
637 {
638 	struct ifnet *ifp;
639 	struct ifaddr *ifa;
640 
641 	ifp = info->xn_ifp;
642 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
643 		if (ifa->ifa_addr->sa_family == AF_INET) {
644 			arp_ifinit(ifp, ifa);
645 		}
646 	}
647 }
648 #endif
649 
650 /**
651  * Callback received when the backend's state changes.
652  */
653 static int
654 netfront_backend_changed(device_t dev, XenbusState newstate)
655 {
656 	struct netfront_info *sc = device_get_softc(dev);
657 
658 	DPRINTK("newstate=%d\n", newstate);
659 
660 	switch (newstate) {
661 	case XenbusStateInitialising:
662 	case XenbusStateInitialised:
663 	case XenbusStateConnected:
664 	case XenbusStateUnknown:
665 	case XenbusStateClosed:
666 	case XenbusStateReconfigured:
667 	case XenbusStateReconfiguring:
668 		break;
669 	case XenbusStateInitWait:
670 		if (xenbus_get_state(dev) != XenbusStateInitialising)
671 			break;
672 		if (network_connect(sc) != 0)
673 			break;
674 		xenbus_set_state(dev, XenbusStateConnected);
675 #ifdef INET
676 		netfront_send_fake_arp(dev, sc);
677 #endif
678 		break;
679 	case XenbusStateClosing:
680 		xenbus_set_state(dev, XenbusStateClosed);
681 		break;
682 	}
683 	return (0);
684 }
685 
686 static void
687 xn_free_rx_ring(struct netfront_info *sc)
688 {
689 #if 0
690 	int i;
691 
692 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
693 		if (sc->xn_cdata.rx_mbufs[i] != NULL) {
694 			m_freem(sc->rx_mbufs[i]);
695 			sc->rx_mbufs[i] = NULL;
696 		}
697 	}
698 
699 	sc->rx.rsp_cons = 0;
700 	sc->xn_rx_if->req_prod = 0;
701 	sc->xn_rx_if->event = sc->rx.rsp_cons ;
702 #endif
703 }
704 
705 static void
706 xn_free_tx_ring(struct netfront_info *sc)
707 {
708 #if 0
709 	int i;
710 
711 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
712 		if (sc->tx_mbufs[i] != NULL) {
713 			m_freem(sc->tx_mbufs[i]);
714 			sc->xn_cdata.xn_tx_chain[i] = NULL;
715 		}
716 	}
717 
718 	return;
719 #endif
720 }
721 
722 /**
723  * \brief Verify that there is sufficient space in the Tx ring
724  *        buffer for a maximally sized request to be enqueued.
725  *
726  * A transmit request requires a transmit descriptor for each packet
727  * fragment, plus up to 2 entries for "options" (e.g. TSO).
728  */
729 static inline int
730 xn_tx_slot_available(struct netfront_info *np)
731 {
732 	return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
733 }
734 
735 static void
736 netif_release_tx_bufs(struct netfront_info *np)
737 {
738 	int i;
739 
740 	for (i = 1; i <= NET_TX_RING_SIZE; i++) {
741 		struct mbuf *m;
742 
743 		m = np->tx_mbufs[i];
744 
745 		/*
746 		 * We assume that no kernel addresses are
747 		 * less than NET_TX_RING_SIZE.  Any entry
748 		 * in the table that is below this number
749 		 * must be an index from free-list tracking.
750 		 */
751 		if (((uintptr_t)m) <= NET_TX_RING_SIZE)
752 			continue;
753 		gnttab_grant_foreign_access_ref(np->grant_tx_ref[i],
754 		    xenbus_get_otherend_id(np->xbdev),
755 		    virt_to_mfn(mtod(m, vm_offset_t)),
756 		    GNTMAP_readonly);
757 		gnttab_release_grant_reference(&np->gref_tx_head,
758 		    np->grant_tx_ref[i]);
759 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
760 		add_id_to_freelist(np->tx_mbufs, i);
761 		np->xn_cdata.xn_tx_chain_cnt--;
762 		if (np->xn_cdata.xn_tx_chain_cnt < 0) {
763 			panic("netif_release_tx_bufs: tx_chain_cnt must be >= 0");
764 		}
765 		m_freem(m);
766 	}
767 }
768 
769 static void
770 network_alloc_rx_buffers(struct netfront_info *sc)
771 {
772 	int otherend_id = xenbus_get_otherend_id(sc->xbdev);
773 	unsigned short id;
774 	struct mbuf *m_new;
775 	int i, batch_target, notify;
776 	RING_IDX req_prod;
777 	struct xen_memory_reservation reservation;
778 	grant_ref_t ref;
779 	int nr_flips;
780 	netif_rx_request_t *req;
781 	vm_offset_t vaddr;
782 	u_long pfn;
783 
784 	req_prod = sc->rx.req_prod_pvt;
785 
786 	if (unlikely(sc->carrier == 0))
787 		return;
788 
789 	/*
790 	 * Allocate mbufs greedily, even though we batch updates to the
791 	 * receive ring. This creates a less bursty demand on the memory
792 	 * allocator, and so should reduce the chance of failed allocation
793 	 * requests both for ourself and for other kernel subsystems.
794 	 *
795 	 * Here we attempt to maintain rx_target buffers in flight, counting
796 	 * buffers that we have yet to process in the receive ring.
797 	 */
798 	batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
799 	for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
800 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
801 		if (m_new == NULL) {
802 			printf("%s: MGETHDR failed\n", __func__);
803 			goto no_mbuf;
804 		}
805 
806 		m_cljget(m_new, M_DONTWAIT, MJUMPAGESIZE);
807 		if ((m_new->m_flags & M_EXT) == 0) {
808 			printf("%s: m_cljget failed\n", __func__);
809 			m_freem(m_new);
810 
811 no_mbuf:
812 			if (i != 0)
813 				goto refill;
814 			/*
815 			 * XXX set timer
816 			 */
817 			break;
818 		}
819 		m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
820 
821 		/* queue the mbufs allocated */
822 		mbufq_tail(&sc->xn_rx_batch, m_new);
823 	}
824 
825 	/*
826 	 * If we've allocated at least half of our target number of entries,
827 	 * submit them to the backend - we have enough to make the overhead
828 	 * of submission worthwhile.  Otherwise wait for more mbufs and
829 	 * request entries to become available.
830 	 */
831 	if (i < (sc->rx_target/2)) {
832 		if (req_prod >sc->rx.sring->req_prod)
833 			goto push;
834 		return;
835 	}
836 
837 	/*
838 	 * Double floating fill target if we risked having the backend
839 	 * run out of empty buffers for receive traffic.  We define "running
840 	 * low" as having less than a fourth of our target buffers free
841 	 * at the time we refilled the queue.
842 	 */
843 	if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
844 		sc->rx_target *= 2;
845 		if (sc->rx_target > sc->rx_max_target)
846 			sc->rx_target = sc->rx_max_target;
847 	}
848 
849 refill:
850 	for (nr_flips = i = 0; ; i++) {
851 		if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
852 			break;
853 
854 		m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
855 				vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
856 
857 		id = xennet_rxidx(req_prod + i);
858 
859 		KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
860 		sc->rx_mbufs[id] = m_new;
861 
862 		ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
863 		KASSERT(ref != GNTTAB_LIST_END,
864 			("reserved grant references exhuasted"));
865 		sc->grant_rx_ref[id] = ref;
866 
867 		vaddr = mtod(m_new, vm_offset_t);
868 		pfn = vtophys(vaddr) >> PAGE_SHIFT;
869 		req = RING_GET_REQUEST(&sc->rx, req_prod + i);
870 
871 		if (sc->copying_receiver == 0) {
872 			gnttab_grant_foreign_transfer_ref(ref,
873 			    otherend_id, pfn);
874 			sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
875 			if (!xen_feature(XENFEAT_auto_translated_physmap)) {
876 				/* Remove this page before passing
877 				 * back to Xen.
878 				 */
879 				set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
880 				MULTI_update_va_mapping(&sc->rx_mcl[i],
881 				    vaddr, 0, 0);
882 			}
883 			nr_flips++;
884 		} else {
885 			gnttab_grant_foreign_access_ref(ref,
886 			    otherend_id,
887 			    PFNTOMFN(pfn), 0);
888 		}
889 		req->id = id;
890 		req->gref = ref;
891 
892 		sc->rx_pfn_array[i] =
893 		    vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
894 	}
895 
896 	KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
897 	KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
898 	/*
899 	 * We may have allocated buffers which have entries outstanding
900 	 * in the page * update queue -- make sure we flush those first!
901 	 */
902 	PT_UPDATES_FLUSH();
903 	if (nr_flips != 0) {
904 #ifdef notyet
905 		/* Tell the ballon driver what is going on. */
906 		balloon_update_driver_allowance(i);
907 #endif
908 		set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
909 		reservation.nr_extents   = i;
910 		reservation.extent_order = 0;
911 		reservation.address_bits = 0;
912 		reservation.domid        = DOMID_SELF;
913 
914 		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
915 
916 			/* After all PTEs have been zapped, flush the TLB. */
917 			sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
918 			    UVMF_TLB_FLUSH|UVMF_ALL;
919 
920 			/* Give away a batch of pages. */
921 			sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
922 			sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
923 			sc->rx_mcl[i].args[1] =  (u_long)&reservation;
924 			/* Zap PTEs and give away pages in one big multicall. */
925 			(void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
926 
927 			/* Check return status of HYPERVISOR_dom_mem_op(). */
928 			if (unlikely(sc->rx_mcl[i].result != i))
929 				panic("Unable to reduce memory reservation\n");
930 			} else {
931 				if (HYPERVISOR_memory_op(
932 				    XENMEM_decrease_reservation, &reservation)
933 				    != i)
934 					panic("Unable to reduce memory "
935 					    "reservation\n");
936 		}
937 	} else {
938 		wmb();
939 	}
940 
941 	/* Above is a suitable barrier to ensure backend will see requests. */
942 	sc->rx.req_prod_pvt = req_prod + i;
943 push:
944 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
945 	if (notify)
946 		notify_remote_via_irq(sc->irq);
947 }
948 
949 static void
950 xn_rxeof(struct netfront_info *np)
951 {
952 	struct ifnet *ifp;
953 #if __FreeBSD_version >= 700000
954 	struct lro_ctrl *lro = &np->xn_lro;
955 	struct lro_entry *queued;
956 #endif
957 	struct netfront_rx_info rinfo;
958 	struct netif_rx_response *rx = &rinfo.rx;
959 	struct netif_extra_info *extras = rinfo.extras;
960 	RING_IDX i, rp;
961 	multicall_entry_t *mcl;
962 	struct mbuf *m;
963 	struct mbuf_head rxq, errq;
964 	int err, pages_flipped = 0, work_to_do;
965 
966 	do {
967 		XN_RX_LOCK_ASSERT(np);
968 		if (!netfront_carrier_ok(np))
969 			return;
970 
971 		mbufq_init(&errq);
972 		mbufq_init(&rxq);
973 
974 		ifp = np->xn_ifp;
975 
976 		rp = np->rx.sring->rsp_prod;
977 		rmb();	/* Ensure we see queued responses up to 'rp'. */
978 
979 		i = np->rx.rsp_cons;
980 		while ((i != rp)) {
981 			memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
982 			memset(extras, 0, sizeof(rinfo.extras));
983 
984 			m = NULL;
985 			err = xennet_get_responses(np, &rinfo, rp, &i, &m,
986 			    &pages_flipped);
987 
988 			if (unlikely(err)) {
989 				if (m)
990 					mbufq_tail(&errq, m);
991 				np->stats.rx_errors++;
992 				continue;
993 			}
994 
995 			m->m_pkthdr.rcvif = ifp;
996 			if ( rx->flags & NETRXF_data_validated ) {
997 				/* Tell the stack the checksums are okay */
998 				/*
999 				 * XXX this isn't necessarily the case - need to add
1000 				 * check
1001 				 */
1002 
1003 				m->m_pkthdr.csum_flags |=
1004 					(CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
1005 					    | CSUM_PSEUDO_HDR);
1006 				m->m_pkthdr.csum_data = 0xffff;
1007 			}
1008 
1009 			np->stats.rx_packets++;
1010 			np->stats.rx_bytes += m->m_pkthdr.len;
1011 
1012 			mbufq_tail(&rxq, m);
1013 			np->rx.rsp_cons = i;
1014 		}
1015 
1016 		if (pages_flipped) {
1017 			/* Some pages are no longer absent... */
1018 #ifdef notyet
1019 			balloon_update_driver_allowance(-pages_flipped);
1020 #endif
1021 			/* Do all the remapping work, and M->P updates, in one big
1022 			 * hypercall.
1023 			 */
1024 			if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1025 				mcl = np->rx_mcl + pages_flipped;
1026 				mcl->op = __HYPERVISOR_mmu_update;
1027 				mcl->args[0] = (u_long)np->rx_mmu;
1028 				mcl->args[1] = pages_flipped;
1029 				mcl->args[2] = 0;
1030 				mcl->args[3] = DOMID_SELF;
1031 				(void)HYPERVISOR_multicall(np->rx_mcl,
1032 				    pages_flipped + 1);
1033 			}
1034 		}
1035 
1036 		while ((m = mbufq_dequeue(&errq)))
1037 			m_freem(m);
1038 
1039 		/*
1040 		 * Process all the mbufs after the remapping is complete.
1041 		 * Break the mbuf chain first though.
1042 		 */
1043 		while ((m = mbufq_dequeue(&rxq)) != NULL) {
1044 			ifp->if_ipackets++;
1045 
1046 			/*
1047 			 * Do we really need to drop the rx lock?
1048 			 */
1049 			XN_RX_UNLOCK(np);
1050 #if __FreeBSD_version >= 700000
1051 			/* Use LRO if possible */
1052 			if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1053 			    lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1054 				/*
1055 				 * If LRO fails, pass up to the stack
1056 				 * directly.
1057 				 */
1058 				(*ifp->if_input)(ifp, m);
1059 			}
1060 #else
1061 			(*ifp->if_input)(ifp, m);
1062 #endif
1063 			XN_RX_LOCK(np);
1064 		}
1065 
1066 		np->rx.rsp_cons = i;
1067 
1068 #if __FreeBSD_version >= 700000
1069 		/*
1070 		 * Flush any outstanding LRO work
1071 		 */
1072 		while (!SLIST_EMPTY(&lro->lro_active)) {
1073 			queued = SLIST_FIRST(&lro->lro_active);
1074 			SLIST_REMOVE_HEAD(&lro->lro_active, next);
1075 			tcp_lro_flush(lro, queued);
1076 		}
1077 #endif
1078 
1079 #if 0
1080 		/* If we get a callback with very few responses, reduce fill target. */
1081 		/* NB. Note exponential increase, linear decrease. */
1082 		if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1083 			((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1084 			np->rx_target = np->rx_min_target;
1085 #endif
1086 
1087 		network_alloc_rx_buffers(np);
1088 
1089 		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1090 	} while (work_to_do);
1091 }
1092 
1093 static void
1094 xn_txeof(struct netfront_info *np)
1095 {
1096 	RING_IDX i, prod;
1097 	unsigned short id;
1098 	struct ifnet *ifp;
1099 	netif_tx_response_t *txr;
1100 	struct mbuf *m;
1101 
1102 	XN_TX_LOCK_ASSERT(np);
1103 
1104 	if (!netfront_carrier_ok(np))
1105 		return;
1106 
1107 	ifp = np->xn_ifp;
1108 
1109 	do {
1110 		prod = np->tx.sring->rsp_prod;
1111 		rmb(); /* Ensure we see responses up to 'rp'. */
1112 
1113 		for (i = np->tx.rsp_cons; i != prod; i++) {
1114 			txr = RING_GET_RESPONSE(&np->tx, i);
1115 			if (txr->status == NETIF_RSP_NULL)
1116 				continue;
1117 
1118 			if (txr->status != NETIF_RSP_OKAY) {
1119 				printf("%s: WARNING: response is %d!\n",
1120 				       __func__, txr->status);
1121 			}
1122 			id = txr->id;
1123 			m = np->tx_mbufs[id];
1124 			KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1125 			KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1126 				("mbuf already on the free list, but we're "
1127 				"trying to free it again!"));
1128 			M_ASSERTVALID(m);
1129 
1130 			/*
1131 			 * Increment packet count if this is the last
1132 			 * mbuf of the chain.
1133 			 */
1134 			if (!m->m_next)
1135 				ifp->if_opackets++;
1136 			if (unlikely(gnttab_query_foreign_access(
1137 			    np->grant_tx_ref[id]) != 0)) {
1138 				panic("grant id %u still in use by the backend",
1139 				      id);
1140 			}
1141 			gnttab_end_foreign_access_ref(
1142 				np->grant_tx_ref[id]);
1143 			gnttab_release_grant_reference(
1144 				&np->gref_tx_head, np->grant_tx_ref[id]);
1145 			np->grant_tx_ref[id] = GRANT_REF_INVALID;
1146 
1147 			np->tx_mbufs[id] = NULL;
1148 			add_id_to_freelist(np->tx_mbufs, id);
1149 			np->xn_cdata.xn_tx_chain_cnt--;
1150 			m_free(m);
1151 			/* Only mark the queue active if we've freed up at least one slot to try */
1152 			ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1153 		}
1154 		np->tx.rsp_cons = prod;
1155 
1156 		/*
1157 		 * Set a new event, then check for race with update of
1158 		 * tx_cons. Note that it is essential to schedule a
1159 		 * callback, no matter how few buffers are pending. Even if
1160 		 * there is space in the transmit ring, higher layers may
1161 		 * be blocked because too much data is outstanding: in such
1162 		 * cases notification from Xen is likely to be the only kick
1163 		 * that we'll get.
1164 		 */
1165 		np->tx.sring->rsp_event =
1166 		    prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1167 
1168 		mb();
1169 	} while (prod != np->tx.sring->rsp_prod);
1170 
1171 	if (np->tx_full &&
1172 	    ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1173 		np->tx_full = 0;
1174 #if 0
1175 		if (np->user_state == UST_OPEN)
1176 			netif_wake_queue(dev);
1177 #endif
1178 	}
1179 
1180 }
1181 
1182 static void
1183 xn_intr(void *xsc)
1184 {
1185 	struct netfront_info *np = xsc;
1186 	struct ifnet *ifp = np->xn_ifp;
1187 
1188 #if 0
1189 	if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1190 	    likely(netfront_carrier_ok(np)) &&
1191 	    ifp->if_drv_flags & IFF_DRV_RUNNING))
1192 		return;
1193 #endif
1194 	if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1195 		XN_TX_LOCK(np);
1196 		xn_txeof(np);
1197 		XN_TX_UNLOCK(np);
1198 	}
1199 
1200 	XN_RX_LOCK(np);
1201 	xn_rxeof(np);
1202 	XN_RX_UNLOCK(np);
1203 
1204 	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1205 	    !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1206 		xn_start(ifp);
1207 }
1208 
1209 
1210 static void
1211 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1212 	grant_ref_t ref)
1213 {
1214 	int new = xennet_rxidx(np->rx.req_prod_pvt);
1215 
1216 	KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1217 	np->rx_mbufs[new] = m;
1218 	np->grant_rx_ref[new] = ref;
1219 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1220 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1221 	np->rx.req_prod_pvt++;
1222 }
1223 
1224 static int
1225 xennet_get_extras(struct netfront_info *np,
1226     struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1227 {
1228 	struct netif_extra_info *extra;
1229 
1230 	int err = 0;
1231 
1232 	do {
1233 		struct mbuf *m;
1234 		grant_ref_t ref;
1235 
1236 		if (unlikely(*cons + 1 == rp)) {
1237 #if 0
1238 			if (net_ratelimit())
1239 				WPRINTK("Missing extra info\n");
1240 #endif
1241 			err = EINVAL;
1242 			break;
1243 		}
1244 
1245 		extra = (struct netif_extra_info *)
1246 		RING_GET_RESPONSE(&np->rx, ++(*cons));
1247 
1248 		if (unlikely(!extra->type ||
1249 			extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1250 #if 0
1251 			if (net_ratelimit())
1252 				WPRINTK("Invalid extra type: %d\n",
1253 					extra->type);
1254 #endif
1255 			err = EINVAL;
1256 		} else {
1257 			memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1258 		}
1259 
1260 		m = xennet_get_rx_mbuf(np, *cons);
1261 		ref = xennet_get_rx_ref(np, *cons);
1262 		xennet_move_rx_slot(np, m, ref);
1263 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1264 
1265 	return err;
1266 }
1267 
1268 static int
1269 xennet_get_responses(struct netfront_info *np,
1270 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1271 	struct mbuf  **list,
1272 	int *pages_flipped_p)
1273 {
1274 	int pages_flipped = *pages_flipped_p;
1275 	struct mmu_update *mmu;
1276 	struct multicall_entry *mcl;
1277 	struct netif_rx_response *rx = &rinfo->rx;
1278 	struct netif_extra_info *extras = rinfo->extras;
1279 	struct mbuf *m, *m0, *m_prev;
1280 	grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1281 	RING_IDX ref_cons = *cons;
1282 	int frags = 1;
1283 	int err = 0;
1284 	u_long ret;
1285 
1286 	m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1287 
1288 
1289 	if (rx->flags & NETRXF_extra_info) {
1290 		err = xennet_get_extras(np, extras, rp, cons);
1291 	}
1292 
1293 
1294 	if (m0 != NULL) {
1295 		m0->m_pkthdr.len = 0;
1296 		m0->m_next = NULL;
1297 	}
1298 
1299 	for (;;) {
1300 		u_long mfn;
1301 
1302 #if 0
1303 		DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1304 			rx->status, rx->offset, frags);
1305 #endif
1306 		if (unlikely(rx->status < 0 ||
1307 			rx->offset + rx->status > PAGE_SIZE)) {
1308 
1309 #if 0
1310 			if (net_ratelimit())
1311 				WPRINTK("rx->offset: %x, size: %u\n",
1312 					rx->offset, rx->status);
1313 #endif
1314 			xennet_move_rx_slot(np, m, ref);
1315 			if (m0 == m)
1316 				m0 = NULL;
1317 			m = NULL;
1318 			err = EINVAL;
1319 			goto next_skip_queue;
1320 		}
1321 
1322 		/*
1323 		 * This definitely indicates a bug, either in this driver or in
1324 		 * the backend driver. In future this should flag the bad
1325 		 * situation to the system controller to reboot the backed.
1326 		 */
1327 		if (ref == GRANT_REF_INVALID) {
1328 
1329 #if 0
1330 			if (net_ratelimit())
1331 				WPRINTK("Bad rx response id %d.\n", rx->id);
1332 #endif
1333 			printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1334 			err = EINVAL;
1335 			goto next;
1336 		}
1337 
1338 		if (!np->copying_receiver) {
1339 			/* Memory pressure, insufficient buffer
1340 			 * headroom, ...
1341 			 */
1342 			if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1343 				WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1344 					rx->id, rx->status);
1345 				xennet_move_rx_slot(np, m, ref);
1346 				err = ENOMEM;
1347 				goto next;
1348 			}
1349 
1350 			if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1351 				/* Remap the page. */
1352 				void *vaddr = mtod(m, void *);
1353 				uint32_t pfn;
1354 
1355 				mcl = np->rx_mcl + pages_flipped;
1356 				mmu = np->rx_mmu + pages_flipped;
1357 
1358 				MULTI_update_va_mapping(mcl, (u_long)vaddr,
1359 				    (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1360 				    PG_V | PG_M | PG_A, 0);
1361 				pfn = (uintptr_t)m->m_ext.ext_arg1;
1362 				mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1363 				    MMU_MACHPHYS_UPDATE;
1364 				mmu->val = pfn;
1365 
1366 				set_phys_to_machine(pfn, mfn);
1367 			}
1368 			pages_flipped++;
1369 		} else {
1370 			ret = gnttab_end_foreign_access_ref(ref);
1371 			KASSERT(ret, ("ret != 0"));
1372 		}
1373 
1374 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1375 
1376 next:
1377 		if (m == NULL)
1378 			break;
1379 
1380 		m->m_len = rx->status;
1381 		m->m_data += rx->offset;
1382 		m0->m_pkthdr.len += rx->status;
1383 
1384 next_skip_queue:
1385 		if (!(rx->flags & NETRXF_more_data))
1386 			break;
1387 
1388 		if (*cons + frags == rp) {
1389 			if (net_ratelimit())
1390 				WPRINTK("Need more frags\n");
1391 			err = ENOENT;
1392 			printf("%s: cons %u frags %u rp %u, not enough frags\n",
1393 			       __func__, *cons, frags, rp);
1394 			break;
1395 		}
1396 		/*
1397 		 * Note that m can be NULL, if rx->status < 0 or if
1398 		 * rx->offset + rx->status > PAGE_SIZE above.
1399 		 */
1400 		m_prev = m;
1401 
1402 		rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1403 		m = xennet_get_rx_mbuf(np, *cons + frags);
1404 
1405 		/*
1406 		 * m_prev == NULL can happen if rx->status < 0 or if
1407 		 * rx->offset + * rx->status > PAGE_SIZE above.
1408 		 */
1409 		if (m_prev != NULL)
1410 			m_prev->m_next = m;
1411 
1412 		/*
1413 		 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1414 		 * rx->status > PAGE_SIZE above.
1415 		 */
1416 		if (m0 == NULL)
1417 			m0 = m;
1418 		m->m_next = NULL;
1419 		ref = xennet_get_rx_ref(np, *cons + frags);
1420 		ref_cons = *cons + frags;
1421 		frags++;
1422 	}
1423 	*list = m0;
1424 	*cons += frags;
1425 	*pages_flipped_p = pages_flipped;
1426 
1427 	return (err);
1428 }
1429 
1430 static void
1431 xn_tick_locked(struct netfront_info *sc)
1432 {
1433 	XN_RX_LOCK_ASSERT(sc);
1434 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1435 
1436 	/* XXX placeholder for printing debug information */
1437 
1438 }
1439 
1440 
1441 static void
1442 xn_tick(void *xsc)
1443 {
1444 	struct netfront_info *sc;
1445 
1446 	sc = xsc;
1447 	XN_RX_LOCK(sc);
1448 	xn_tick_locked(sc);
1449 	XN_RX_UNLOCK(sc);
1450 
1451 }
1452 
1453 /**
1454  * \brief Count the number of fragments in an mbuf chain.
1455  *
1456  * Surprisingly, there isn't an M* macro for this.
1457  */
1458 static inline int
1459 xn_count_frags(struct mbuf *m)
1460 {
1461 	int nfrags;
1462 
1463 	for (nfrags = 0; m != NULL; m = m->m_next)
1464 		nfrags++;
1465 
1466 	return (nfrags);
1467 }
1468 
1469 /**
1470  * Given an mbuf chain, make sure we have enough room and then push
1471  * it onto the transmit ring.
1472  */
1473 static int
1474 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1475 {
1476 	struct ifnet *ifp;
1477 	struct mbuf *m;
1478 	u_int nfrags;
1479 	netif_extra_info_t *extra;
1480 	int otherend_id;
1481 
1482 	ifp = sc->xn_ifp;
1483 
1484 	/**
1485 	 * Defragment the mbuf if necessary.
1486 	 */
1487 	nfrags = xn_count_frags(m_head);
1488 
1489 	/*
1490 	 * Check to see whether this request is longer than netback
1491 	 * can handle, and try to defrag it.
1492 	 */
1493 	/**
1494 	 * It is a bit lame, but the netback driver in Linux can't
1495 	 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1496 	 * the Linux network stack.
1497 	 */
1498 	if (nfrags > MAX_TX_REQ_FRAGS) {
1499 		m = m_defrag(m_head, M_DONTWAIT);
1500 		if (!m) {
1501 			/*
1502 			 * Defrag failed, so free the mbuf and
1503 			 * therefore drop the packet.
1504 			 */
1505 			m_freem(m_head);
1506 			return (EMSGSIZE);
1507 		}
1508 		m_head = m;
1509 	}
1510 
1511 	/* Determine how many fragments now exist */
1512 	nfrags = xn_count_frags(m_head);
1513 
1514 	/*
1515 	 * Check to see whether the defragmented packet has too many
1516 	 * segments for the Linux netback driver.
1517 	 */
1518 	/**
1519 	 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1520 	 * of mbufs longer than Linux can handle.  Make sure we don't
1521 	 * pass a too-long chain over to the other side by dropping the
1522 	 * packet.  It doesn't look like there is currently a way to
1523 	 * tell the TCP stack to generate a shorter chain of packets.
1524 	 */
1525 	if (nfrags > MAX_TX_REQ_FRAGS) {
1526 #ifdef DEBUG
1527 		printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1528 		       "won't be able to handle it, dropping\n",
1529 		       __func__, nfrags, MAX_TX_REQ_FRAGS);
1530 #endif
1531 		m_freem(m_head);
1532 		return (EMSGSIZE);
1533 	}
1534 
1535 	/*
1536 	 * This check should be redundant.  We've already verified that we
1537 	 * have enough slots in the ring to handle a packet of maximum
1538 	 * size, and that our packet is less than the maximum size.  Keep
1539 	 * it in here as an assert for now just to make certain that
1540 	 * xn_tx_chain_cnt is accurate.
1541 	 */
1542 	KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1543 		("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1544 		 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1545                     (int) nfrags, (int) NET_TX_RING_SIZE));
1546 
1547 	/*
1548 	 * Start packing the mbufs in this chain into
1549 	 * the fragment pointers. Stop when we run out
1550 	 * of fragments or hit the end of the mbuf chain.
1551 	 */
1552 	m = m_head;
1553 	extra = NULL;
1554 	otherend_id = xenbus_get_otherend_id(sc->xbdev);
1555 	for (m = m_head; m; m = m->m_next) {
1556 		netif_tx_request_t *tx;
1557 		uintptr_t id;
1558 		grant_ref_t ref;
1559 		u_long mfn; /* XXX Wrong type? */
1560 
1561 		tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1562 		id = get_id_from_freelist(sc->tx_mbufs);
1563 		if (id == 0)
1564 			panic("xn_start_locked: was allocated the freelist head!\n");
1565 		sc->xn_cdata.xn_tx_chain_cnt++;
1566 		if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1567 			panic("xn_start_locked: tx_chain_cnt must be <= NET_TX_RING_SIZE\n");
1568 		sc->tx_mbufs[id] = m;
1569 		tx->id = id;
1570 		ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1571 		KASSERT((short)ref >= 0, ("Negative ref"));
1572 		mfn = virt_to_mfn(mtod(m, vm_offset_t));
1573 		gnttab_grant_foreign_access_ref(ref, otherend_id,
1574 		    mfn, GNTMAP_readonly);
1575 		tx->gref = sc->grant_tx_ref[id] = ref;
1576 		tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1577 		tx->flags = 0;
1578 		if (m == m_head) {
1579 			/*
1580 			 * The first fragment has the entire packet
1581 			 * size, subsequent fragments have just the
1582 			 * fragment size. The backend works out the
1583 			 * true size of the first fragment by
1584 			 * subtracting the sizes of the other
1585 			 * fragments.
1586 			 */
1587 			tx->size = m->m_pkthdr.len;
1588 
1589 			/*
1590 			 * The first fragment contains the checksum flags
1591 			 * and is optionally followed by extra data for
1592 			 * TSO etc.
1593 			 */
1594 			/**
1595 			 * CSUM_TSO requires checksum offloading.
1596 			 * Some versions of FreeBSD fail to
1597 			 * set CSUM_TCP in the CSUM_TSO case,
1598 			 * so we have to test for CSUM_TSO
1599 			 * explicitly.
1600 			 */
1601 			if (m->m_pkthdr.csum_flags
1602 			    & (CSUM_DELAY_DATA | CSUM_TSO)) {
1603 				tx->flags |= (NETTXF_csum_blank
1604 				    | NETTXF_data_validated);
1605 			}
1606 #if __FreeBSD_version >= 700000
1607 			if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1608 				struct netif_extra_info *gso =
1609 					(struct netif_extra_info *)
1610 					RING_GET_REQUEST(&sc->tx,
1611 							 ++sc->tx.req_prod_pvt);
1612 
1613 				tx->flags |= NETTXF_extra_info;
1614 
1615 				gso->u.gso.size = m->m_pkthdr.tso_segsz;
1616 				gso->u.gso.type =
1617 					XEN_NETIF_GSO_TYPE_TCPV4;
1618 				gso->u.gso.pad = 0;
1619 				gso->u.gso.features = 0;
1620 
1621 				gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1622 				gso->flags = 0;
1623 			}
1624 #endif
1625 		} else {
1626 			tx->size = m->m_len;
1627 		}
1628 		if (m->m_next)
1629 			tx->flags |= NETTXF_more_data;
1630 
1631 		sc->tx.req_prod_pvt++;
1632 	}
1633 	BPF_MTAP(ifp, m_head);
1634 
1635 	sc->stats.tx_bytes += m_head->m_pkthdr.len;
1636 	sc->stats.tx_packets++;
1637 
1638 	return (0);
1639 }
1640 
1641 static void
1642 xn_start_locked(struct ifnet *ifp)
1643 {
1644 	struct netfront_info *sc;
1645 	struct mbuf *m_head;
1646 	int notify;
1647 
1648 	sc = ifp->if_softc;
1649 
1650 	if (!netfront_carrier_ok(sc))
1651 		return;
1652 
1653 	/*
1654 	 * While we have enough transmit slots available for at least one
1655 	 * maximum-sized packet, pull mbufs off the queue and put them on
1656 	 * the transmit ring.
1657 	 */
1658 	while (xn_tx_slot_available(sc)) {
1659 		IF_DEQUEUE(&ifp->if_snd, m_head);
1660 		if (m_head == NULL)
1661 			break;
1662 
1663 		if (xn_assemble_tx_request(sc, m_head) != 0)
1664 			break;
1665 	}
1666 
1667 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1668 	if (notify)
1669 		notify_remote_via_irq(sc->irq);
1670 
1671 	if (RING_FULL(&sc->tx)) {
1672 		sc->tx_full = 1;
1673 #if 0
1674 		netif_stop_queue(dev);
1675 #endif
1676 	}
1677 }
1678 
1679 
1680 static void
1681 xn_start(struct ifnet *ifp)
1682 {
1683 	struct netfront_info *sc;
1684 	sc = ifp->if_softc;
1685 	XN_TX_LOCK(sc);
1686 	xn_start_locked(ifp);
1687 	XN_TX_UNLOCK(sc);
1688 }
1689 
1690 /* equivalent of network_open() in Linux */
1691 static void
1692 xn_ifinit_locked(struct netfront_info *sc)
1693 {
1694 	struct ifnet *ifp;
1695 
1696 	XN_LOCK_ASSERT(sc);
1697 
1698 	ifp = sc->xn_ifp;
1699 
1700 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1701 		return;
1702 
1703 	xn_stop(sc);
1704 
1705 	network_alloc_rx_buffers(sc);
1706 	sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1707 
1708 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1709 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1710 	if_link_state_change(ifp, LINK_STATE_UP);
1711 
1712 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1713 
1714 }
1715 
1716 
1717 static void
1718 xn_ifinit(void *xsc)
1719 {
1720 	struct netfront_info *sc = xsc;
1721 
1722 	XN_LOCK(sc);
1723 	xn_ifinit_locked(sc);
1724 	XN_UNLOCK(sc);
1725 
1726 }
1727 
1728 
1729 static int
1730 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1731 {
1732 	struct netfront_info *sc = ifp->if_softc;
1733 	struct ifreq *ifr = (struct ifreq *) data;
1734 #ifdef INET
1735 	struct ifaddr *ifa = (struct ifaddr *)data;
1736 #endif
1737 
1738 	int mask, error = 0;
1739 	switch(cmd) {
1740 	case SIOCSIFADDR:
1741 	case SIOCGIFADDR:
1742 #ifdef INET
1743 		XN_LOCK(sc);
1744 		if (ifa->ifa_addr->sa_family == AF_INET) {
1745 			ifp->if_flags |= IFF_UP;
1746 			if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1747 				xn_ifinit_locked(sc);
1748 			arp_ifinit(ifp, ifa);
1749 			XN_UNLOCK(sc);
1750 		} else {
1751 			XN_UNLOCK(sc);
1752 #endif
1753 			error = ether_ioctl(ifp, cmd, data);
1754 #ifdef INET
1755 		}
1756 #endif
1757 		break;
1758 	case SIOCSIFMTU:
1759 		/* XXX can we alter the MTU on a VN ?*/
1760 #ifdef notyet
1761 		if (ifr->ifr_mtu > XN_JUMBO_MTU)
1762 			error = EINVAL;
1763 		else
1764 #endif
1765 		{
1766 			ifp->if_mtu = ifr->ifr_mtu;
1767 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1768 			xn_ifinit(sc);
1769 		}
1770 		break;
1771 	case SIOCSIFFLAGS:
1772 		XN_LOCK(sc);
1773 		if (ifp->if_flags & IFF_UP) {
1774 			/*
1775 			 * If only the state of the PROMISC flag changed,
1776 			 * then just use the 'set promisc mode' command
1777 			 * instead of reinitializing the entire NIC. Doing
1778 			 * a full re-init means reloading the firmware and
1779 			 * waiting for it to start up, which may take a
1780 			 * second or two.
1781 			 */
1782 #ifdef notyet
1783 			/* No promiscuous mode with Xen */
1784 			if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1785 			    ifp->if_flags & IFF_PROMISC &&
1786 			    !(sc->xn_if_flags & IFF_PROMISC)) {
1787 				XN_SETBIT(sc, XN_RX_MODE,
1788 					  XN_RXMODE_RX_PROMISC);
1789 			} else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1790 				   !(ifp->if_flags & IFF_PROMISC) &&
1791 				   sc->xn_if_flags & IFF_PROMISC) {
1792 				XN_CLRBIT(sc, XN_RX_MODE,
1793 					  XN_RXMODE_RX_PROMISC);
1794 			} else
1795 #endif
1796 				xn_ifinit_locked(sc);
1797 		} else {
1798 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1799 				xn_stop(sc);
1800 			}
1801 		}
1802 		sc->xn_if_flags = ifp->if_flags;
1803 		XN_UNLOCK(sc);
1804 		error = 0;
1805 		break;
1806 	case SIOCSIFCAP:
1807 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1808 		if (mask & IFCAP_TXCSUM) {
1809 			if (IFCAP_TXCSUM & ifp->if_capenable) {
1810 				ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1811 				ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1812 				    | CSUM_IP | CSUM_TSO);
1813 			} else {
1814 				ifp->if_capenable |= IFCAP_TXCSUM;
1815 				ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1816 				    | CSUM_IP);
1817 			}
1818 		}
1819 		if (mask & IFCAP_RXCSUM) {
1820 			ifp->if_capenable ^= IFCAP_RXCSUM;
1821 		}
1822 #if __FreeBSD_version >= 700000
1823 		if (mask & IFCAP_TSO4) {
1824 			if (IFCAP_TSO4 & ifp->if_capenable) {
1825 				ifp->if_capenable &= ~IFCAP_TSO4;
1826 				ifp->if_hwassist &= ~CSUM_TSO;
1827 			} else if (IFCAP_TXCSUM & ifp->if_capenable) {
1828 				ifp->if_capenable |= IFCAP_TSO4;
1829 				ifp->if_hwassist |= CSUM_TSO;
1830 			} else {
1831 				IPRINTK("Xen requires tx checksum offload"
1832 				    " be enabled to use TSO\n");
1833 				error = EINVAL;
1834 			}
1835 		}
1836 		if (mask & IFCAP_LRO) {
1837 			ifp->if_capenable ^= IFCAP_LRO;
1838 
1839 		}
1840 #endif
1841 		error = 0;
1842 		break;
1843 	case SIOCADDMULTI:
1844 	case SIOCDELMULTI:
1845 #ifdef notyet
1846 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1847 			XN_LOCK(sc);
1848 			xn_setmulti(sc);
1849 			XN_UNLOCK(sc);
1850 			error = 0;
1851 		}
1852 #endif
1853 		/* FALLTHROUGH */
1854 	case SIOCSIFMEDIA:
1855 	case SIOCGIFMEDIA:
1856 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1857 		break;
1858 	default:
1859 		error = ether_ioctl(ifp, cmd, data);
1860 	}
1861 
1862 	return (error);
1863 }
1864 
1865 static void
1866 xn_stop(struct netfront_info *sc)
1867 {
1868 	struct ifnet *ifp;
1869 
1870 	XN_LOCK_ASSERT(sc);
1871 
1872 	ifp = sc->xn_ifp;
1873 
1874 	callout_stop(&sc->xn_stat_ch);
1875 
1876 	xn_free_rx_ring(sc);
1877 	xn_free_tx_ring(sc);
1878 
1879 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1880 	if_link_state_change(ifp, LINK_STATE_DOWN);
1881 }
1882 
1883 /* START of Xenolinux helper functions adapted to FreeBSD */
1884 int
1885 network_connect(struct netfront_info *np)
1886 {
1887 	int i, requeue_idx, error;
1888 	grant_ref_t ref;
1889 	netif_rx_request_t *req;
1890 	u_int feature_rx_copy, feature_rx_flip;
1891 
1892 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1893 	    "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1894 	if (error)
1895 		feature_rx_copy = 0;
1896 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1897 	    "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1898 	if (error)
1899 		feature_rx_flip = 1;
1900 
1901 	/*
1902 	 * Copy packets on receive path if:
1903 	 *  (a) This was requested by user, and the backend supports it; or
1904 	 *  (b) Flipping was requested, but this is unsupported by the backend.
1905 	 */
1906 	np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1907 				(MODPARM_rx_flip && !feature_rx_flip));
1908 
1909 	/* Recovery procedure: */
1910 	error = talk_to_backend(np->xbdev, np);
1911 	if (error)
1912 		return (error);
1913 
1914 	/* Step 1: Reinitialise variables. */
1915 	netif_release_tx_bufs(np);
1916 
1917 	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1918 	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1919 		struct mbuf *m;
1920 		u_long pfn;
1921 
1922 		if (np->rx_mbufs[i] == NULL)
1923 			continue;
1924 
1925 		m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1926 		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1927 
1928 		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1929 		pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1930 
1931 		if (!np->copying_receiver) {
1932 			gnttab_grant_foreign_transfer_ref(ref,
1933 			    xenbus_get_otherend_id(np->xbdev),
1934 			    pfn);
1935 		} else {
1936 			gnttab_grant_foreign_access_ref(ref,
1937 			    xenbus_get_otherend_id(np->xbdev),
1938 			    PFNTOMFN(pfn), 0);
1939 		}
1940 		req->gref = ref;
1941 		req->id   = requeue_idx;
1942 
1943 		requeue_idx++;
1944 	}
1945 
1946 	np->rx.req_prod_pvt = requeue_idx;
1947 
1948 	/* Step 3: All public and private state should now be sane.  Get
1949 	 * ready to start sending and receiving packets and give the driver
1950 	 * domain a kick because we've probably just requeued some
1951 	 * packets.
1952 	 */
1953 	netfront_carrier_on(np);
1954 	notify_remote_via_irq(np->irq);
1955 	XN_TX_LOCK(np);
1956 	xn_txeof(np);
1957 	XN_TX_UNLOCK(np);
1958 	network_alloc_rx_buffers(np);
1959 
1960 	return (0);
1961 }
1962 
1963 static void
1964 show_device(struct netfront_info *sc)
1965 {
1966 #ifdef DEBUG
1967 	if (sc) {
1968 		IPRINTK("<vif handle=%u %s(%s) evtchn=%u irq=%u tx=%p rx=%p>\n",
1969 			sc->xn_ifno,
1970 			be_state_name[sc->xn_backend_state],
1971 			sc->xn_user_state ? "open" : "closed",
1972 			sc->xn_evtchn,
1973 			sc->xn_irq,
1974 			sc->xn_tx_if,
1975 			sc->xn_rx_if);
1976 	} else {
1977 		IPRINTK("<vif NULL>\n");
1978 	}
1979 #endif
1980 }
1981 
1982 /** Create a network device.
1983  * @param handle device handle
1984  */
1985 int
1986 create_netdev(device_t dev)
1987 {
1988 	int i;
1989 	struct netfront_info *np;
1990 	int err;
1991 	struct ifnet *ifp;
1992 
1993 	np = device_get_softc(dev);
1994 
1995 	np->xbdev         = dev;
1996 
1997 	XN_LOCK_INIT(np, xennetif);
1998 
1999 	ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
2000 	ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
2001 	ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
2002 
2003 	np->rx_target     = RX_MIN_TARGET;
2004 	np->rx_min_target = RX_MIN_TARGET;
2005 	np->rx_max_target = RX_MAX_TARGET;
2006 
2007 	/* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2008 	for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2009 		np->tx_mbufs[i] = (void *) ((u_long) i+1);
2010 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
2011 	}
2012 	np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2013 
2014 	for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2015 
2016 		np->rx_mbufs[i] = NULL;
2017 		np->grant_rx_ref[i] = GRANT_REF_INVALID;
2018 	}
2019 	/* A grant for every tx ring slot */
2020 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2021 					  &np->gref_tx_head) != 0) {
2022 		IPRINTK("#### netfront can't alloc tx grant refs\n");
2023 		err = ENOMEM;
2024 		goto exit;
2025 	}
2026 	/* A grant for every rx ring slot */
2027 	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2028 					  &np->gref_rx_head) != 0) {
2029 		WPRINTK("#### netfront can't alloc rx grant refs\n");
2030 		gnttab_free_grant_references(np->gref_tx_head);
2031 		err = ENOMEM;
2032 		goto exit;
2033 	}
2034 
2035 	err = xen_net_read_mac(dev, np->mac);
2036 	if (err) {
2037 		xenbus_dev_fatal(dev, err, "parsing %s/mac",
2038 		    xenbus_get_node(dev));
2039 		goto out;
2040 	}
2041 
2042 	/* Set up ifnet structure */
2043 	ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2044     	ifp->if_softc = np;
2045     	if_initname(ifp, "xn",  device_get_unit(dev));
2046     	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2047     	ifp->if_ioctl = xn_ioctl;
2048     	ifp->if_output = ether_output;
2049     	ifp->if_start = xn_start;
2050 #ifdef notyet
2051     	ifp->if_watchdog = xn_watchdog;
2052 #endif
2053     	ifp->if_init = xn_ifinit;
2054     	ifp->if_mtu = ETHERMTU;
2055     	ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2056 
2057     	ifp->if_hwassist = XN_CSUM_FEATURES;
2058     	ifp->if_capabilities = IFCAP_HWCSUM;
2059 #if __FreeBSD_version >= 700000
2060 	ifp->if_capabilities |= IFCAP_TSO4;
2061 	if (xn_enable_lro) {
2062 		int err = tcp_lro_init(&np->xn_lro);
2063 		if (err) {
2064 			device_printf(dev, "LRO initialization failed\n");
2065 			goto exit;
2066 		}
2067 		np->xn_lro.ifp = ifp;
2068 		ifp->if_capabilities |= IFCAP_LRO;
2069 	}
2070 #endif
2071     	ifp->if_capenable = ifp->if_capabilities;
2072 
2073     	ether_ifattach(ifp, np->mac);
2074     	callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2075 	netfront_carrier_off(np);
2076 
2077 	return (0);
2078 
2079 exit:
2080 	gnttab_free_grant_references(np->gref_tx_head);
2081 out:
2082 	panic("do something smart");
2083 
2084 }
2085 
2086 /**
2087  * Handle the change of state of the backend to Closing.  We must delete our
2088  * device-layer structures now, to ensure that writes are flushed through to
2089  * the backend.  Once is this done, we can switch to Closed in
2090  * acknowledgement.
2091  */
2092 #if 0
2093 static void
2094 netfront_closing(device_t dev)
2095 {
2096 #if 0
2097 	struct netfront_info *info = dev->dev_driver_data;
2098 
2099 	DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2100 
2101 	close_netdev(info);
2102 #endif
2103 	xenbus_switch_state(dev, XenbusStateClosed);
2104 }
2105 #endif
2106 
2107 static int
2108 netfront_detach(device_t dev)
2109 {
2110 	struct netfront_info *info = device_get_softc(dev);
2111 
2112 	DPRINTK("%s\n", xenbus_get_node(dev));
2113 
2114 	netif_free(info);
2115 
2116 	return 0;
2117 }
2118 
2119 static void
2120 netif_free(struct netfront_info *info)
2121 {
2122 	netif_disconnect_backend(info);
2123 #if 0
2124 	close_netdev(info);
2125 #endif
2126 }
2127 
2128 static void
2129 netif_disconnect_backend(struct netfront_info *info)
2130 {
2131 	XN_RX_LOCK(info);
2132 	XN_TX_LOCK(info);
2133 	netfront_carrier_off(info);
2134 	XN_TX_UNLOCK(info);
2135 	XN_RX_UNLOCK(info);
2136 
2137 	end_access(info->tx_ring_ref, info->tx.sring);
2138 	end_access(info->rx_ring_ref, info->rx.sring);
2139 	info->tx_ring_ref = GRANT_REF_INVALID;
2140 	info->rx_ring_ref = GRANT_REF_INVALID;
2141 	info->tx.sring = NULL;
2142 	info->rx.sring = NULL;
2143 
2144 	if (info->irq)
2145 		unbind_from_irqhandler(info->irq);
2146 
2147 	info->irq = 0;
2148 }
2149 
2150 
2151 static void
2152 end_access(int ref, void *page)
2153 {
2154 	if (ref != GRANT_REF_INVALID)
2155 		gnttab_end_foreign_access(ref, page);
2156 }
2157 
2158 static int
2159 xn_ifmedia_upd(struct ifnet *ifp)
2160 {
2161 	return (0);
2162 }
2163 
2164 static void
2165 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2166 {
2167 	ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2168 	ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2169 }
2170 
2171 /* ** Driver registration ** */
2172 static device_method_t netfront_methods[] = {
2173 	/* Device interface */
2174 	DEVMETHOD(device_probe,         netfront_probe),
2175 	DEVMETHOD(device_attach,        netfront_attach),
2176 	DEVMETHOD(device_detach,        netfront_detach),
2177 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
2178 	DEVMETHOD(device_suspend,       bus_generic_suspend),
2179 	DEVMETHOD(device_resume,        netfront_resume),
2180 
2181 	/* Xenbus interface */
2182 	DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2183 
2184 	{ 0, 0 }
2185 };
2186 
2187 static driver_t netfront_driver = {
2188 	"xn",
2189 	netfront_methods,
2190 	sizeof(struct netfront_info),
2191 };
2192 devclass_t netfront_devclass;
2193 
2194 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, 0, 0);
2195