xref: /freebsd/sys/dev/xen/netfront/netfront.c (revision 3b8f08459569bf0faa21473e5cec2491e95c9349)
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/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_var.h>
48 #include <net/if_arp.h>
49 #include <net/ethernet.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
52 
53 #include <net/bpf.h>
54 
55 #include <net/if_types.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 <xen/xen-os.h>
81 #include <xen/hypervisor.h>
82 #include <xen/xen_intr.h>
83 #include <xen/gnttab.h>
84 #include <xen/interface/memory.h>
85 #include <xen/interface/io/netif.h>
86 #include <xen/xenbus/xenbusvar.h>
87 
88 #include <machine/xen/xenvar.h>
89 
90 #include <dev/xen/netfront/mbufq.h>
91 
92 #include "xenbus_if.h"
93 
94 /* Features supported by all backends.  TSO and LRO can be negotiated */
95 #define XN_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
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 #define	NF_TSO_MAXBURST ((IP_MAXPACKET / PAGE_SIZE) * MCLBYTES)
138 
139 #define RX_COPY_THRESHOLD 256
140 
141 #define net_ratelimit() 0
142 
143 struct netfront_info;
144 struct netfront_rx_info;
145 
146 static void xn_txeof(struct netfront_info *);
147 static void xn_rxeof(struct netfront_info *);
148 static void network_alloc_rx_buffers(struct netfront_info *);
149 
150 static void xn_tick_locked(struct netfront_info *);
151 static void xn_tick(void *);
152 
153 static void xn_intr(void *);
154 static inline int xn_count_frags(struct mbuf *m);
155 static int  xn_assemble_tx_request(struct netfront_info *sc,
156 				   struct mbuf *m_head);
157 static void xn_start_locked(struct ifnet *);
158 static void xn_start(struct ifnet *);
159 static int  xn_ioctl(struct ifnet *, u_long, caddr_t);
160 static void xn_ifinit_locked(struct netfront_info *);
161 static void xn_ifinit(void *);
162 static void xn_stop(struct netfront_info *);
163 static void xn_query_features(struct netfront_info *np);
164 static int  xn_configure_features(struct netfront_info *np);
165 #ifdef notyet
166 static void xn_watchdog(struct ifnet *);
167 #endif
168 
169 #ifdef notyet
170 static void netfront_closing(device_t dev);
171 #endif
172 static void netif_free(struct netfront_info *info);
173 static int netfront_detach(device_t dev);
174 
175 static int talk_to_backend(device_t dev, struct netfront_info *info);
176 static int create_netdev(device_t dev);
177 static void netif_disconnect_backend(struct netfront_info *info);
178 static int setup_device(device_t dev, struct netfront_info *info);
179 static void free_ring(int *ref, void *ring_ptr_ref);
180 
181 static int  xn_ifmedia_upd(struct ifnet *ifp);
182 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
183 
184 /* Xenolinux helper functions */
185 int network_connect(struct netfront_info *);
186 
187 static void xn_free_rx_ring(struct netfront_info *);
188 
189 static void xn_free_tx_ring(struct netfront_info *);
190 
191 static int xennet_get_responses(struct netfront_info *np,
192 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
193 	struct mbuf **list, int *pages_flipped_p);
194 
195 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
196 
197 #define INVALID_P2M_ENTRY (~0UL)
198 
199 /*
200  * Mbuf pointers. We need these to keep track of the virtual addresses
201  * of our mbuf chains since we can only convert from virtual to physical,
202  * not the other way around.  The size must track the free index arrays.
203  */
204 struct xn_chain_data {
205 	struct mbuf    *xn_tx_chain[NET_TX_RING_SIZE+1];
206 	int		xn_tx_chain_cnt;
207 	struct mbuf    *xn_rx_chain[NET_RX_RING_SIZE+1];
208 };
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 	struct ifnet *xn_ifp;
245 #if __FreeBSD_version >= 700000
246 	struct lro_ctrl xn_lro;
247 #endif
248 
249 	struct net_device_stats stats;
250 	u_int tx_full;
251 
252 	netif_tx_front_ring_t tx;
253 	netif_rx_front_ring_t rx;
254 
255 	struct mtx   tx_lock;
256 	struct mtx   rx_lock;
257 	struct mtx   sc_lock;
258 
259 	xen_intr_handle_t xen_intr_handle;
260 	u_int copying_receiver;
261 	u_int carrier;
262 	u_int maxfrags;
263 
264 	/* Receive-ring batched refills. */
265 #define RX_MIN_TARGET 32
266 #define RX_MAX_TARGET NET_RX_RING_SIZE
267 	int rx_min_target;
268 	int rx_max_target;
269 	int rx_target;
270 
271 	grant_ref_t gref_tx_head;
272 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
273 	grant_ref_t gref_rx_head;
274 	grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
275 
276 	device_t		xbdev;
277 	int			tx_ring_ref;
278 	int			rx_ring_ref;
279 	uint8_t			mac[ETHER_ADDR_LEN];
280 	struct xn_chain_data	xn_cdata;	/* mbufs */
281 	struct mbuf_head	xn_rx_batch;	/* head of the batch queue */
282 
283 	int			xn_if_flags;
284 	struct callout	        xn_stat_ch;
285 
286 	u_long			rx_pfn_array[NET_RX_RING_SIZE];
287 	multicall_entry_t	rx_mcl[NET_RX_RING_SIZE+1];
288 	mmu_update_t		rx_mmu[NET_RX_RING_SIZE];
289 	struct ifmedia		sc_media;
290 };
291 
292 #define rx_mbufs xn_cdata.xn_rx_chain
293 #define tx_mbufs xn_cdata.xn_tx_chain
294 
295 #define XN_LOCK_INIT(_sc, _name) \
296         mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
297         mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF);  \
298         mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF)
299 
300 #define XN_RX_LOCK(_sc)           mtx_lock(&(_sc)->rx_lock)
301 #define XN_RX_UNLOCK(_sc)         mtx_unlock(&(_sc)->rx_lock)
302 
303 #define XN_TX_LOCK(_sc)           mtx_lock(&(_sc)->tx_lock)
304 #define XN_TX_UNLOCK(_sc)         mtx_unlock(&(_sc)->tx_lock)
305 
306 #define XN_LOCK(_sc)           mtx_lock(&(_sc)->sc_lock);
307 #define XN_UNLOCK(_sc)         mtx_unlock(&(_sc)->sc_lock);
308 
309 #define XN_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->sc_lock, MA_OWNED);
310 #define XN_RX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->rx_lock, MA_OWNED);
311 #define XN_TX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->tx_lock, MA_OWNED);
312 #define XN_LOCK_DESTROY(_sc)   mtx_destroy(&(_sc)->rx_lock); \
313                                mtx_destroy(&(_sc)->tx_lock); \
314                                mtx_destroy(&(_sc)->sc_lock);
315 
316 struct netfront_rx_info {
317 	struct netif_rx_response rx;
318 	struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
319 };
320 
321 #define netfront_carrier_on(netif)	((netif)->carrier = 1)
322 #define netfront_carrier_off(netif)	((netif)->carrier = 0)
323 #define netfront_carrier_ok(netif)	((netif)->carrier)
324 
325 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
326 
327 static inline void
328 add_id_to_freelist(struct mbuf **list, uintptr_t id)
329 {
330 	KASSERT(id != 0,
331 		("%s: the head item (0) must always be free.", __func__));
332 	list[id] = list[0];
333 	list[0]  = (struct mbuf *)id;
334 }
335 
336 static inline unsigned short
337 get_id_from_freelist(struct mbuf **list)
338 {
339 	uintptr_t id;
340 
341 	id = (uintptr_t)list[0];
342 	KASSERT(id != 0,
343 		("%s: the head item (0) must always remain free.", __func__));
344 	list[0] = list[id];
345 	return (id);
346 }
347 
348 static inline int
349 xennet_rxidx(RING_IDX idx)
350 {
351 	return idx & (NET_RX_RING_SIZE - 1);
352 }
353 
354 static inline struct mbuf *
355 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
356 {
357 	int i = xennet_rxidx(ri);
358 	struct mbuf *m;
359 
360 	m = np->rx_mbufs[i];
361 	np->rx_mbufs[i] = NULL;
362 	return (m);
363 }
364 
365 static inline grant_ref_t
366 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
367 {
368 	int i = xennet_rxidx(ri);
369 	grant_ref_t ref = np->grant_rx_ref[i];
370 	KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
371 	np->grant_rx_ref[i] = GRANT_REF_INVALID;
372 	return ref;
373 }
374 
375 #define IPRINTK(fmt, args...) \
376     printf("[XEN] " fmt, ##args)
377 #ifdef INVARIANTS
378 #define WPRINTK(fmt, args...) \
379     printf("[XEN] " fmt, ##args)
380 #else
381 #define WPRINTK(fmt, args...)
382 #endif
383 #ifdef DEBUG
384 #define DPRINTK(fmt, args...) \
385     printf("[XEN] %s: " fmt, __func__, ##args)
386 #else
387 #define DPRINTK(fmt, args...)
388 #endif
389 
390 /**
391  * Read the 'mac' node at the given device's node in the store, and parse that
392  * as colon-separated octets, placing result the given mac array.  mac must be
393  * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
394  * Return 0 on success, or errno on error.
395  */
396 static int
397 xen_net_read_mac(device_t dev, uint8_t mac[])
398 {
399 	int error, i;
400 	char *s, *e, *macstr;
401 	const char *path;
402 
403 	path = xenbus_get_node(dev);
404 	error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
405 	if (error == ENOENT) {
406 		/*
407 		 * Deal with missing mac XenStore nodes on devices with
408 		 * HVM emulation (the 'ioemu' configuration attribute)
409 		 * enabled.
410 		 *
411 		 * The HVM emulator may execute in a stub device model
412 		 * domain which lacks the permission, only given to Dom0,
413 		 * to update the guest's XenStore tree.  For this reason,
414 		 * the HVM emulator doesn't even attempt to write the
415 		 * front-side mac node, even when operating in Dom0.
416 		 * However, there should always be a mac listed in the
417 		 * backend tree.  Fallback to this version if our query
418 		 * of the front side XenStore location doesn't find
419 		 * anything.
420 		 */
421 		path = xenbus_get_otherend_path(dev);
422 		error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
423 	}
424 	if (error != 0) {
425 		xenbus_dev_fatal(dev, error, "parsing %s/mac", path);
426 		return (error);
427 	}
428 
429 	s = macstr;
430 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
431 		mac[i] = strtoul(s, &e, 16);
432 		if (s == e || (e[0] != ':' && e[0] != 0)) {
433 			free(macstr, M_XENBUS);
434 			return (ENOENT);
435 		}
436 		s = &e[1];
437 	}
438 	free(macstr, M_XENBUS);
439 	return (0);
440 }
441 
442 /**
443  * Entry point to this code when a new device is created.  Allocate the basic
444  * structures and the ring buffers for communication with the backend, and
445  * inform the backend of the appropriate details for those.  Switch to
446  * Connected state.
447  */
448 static int
449 netfront_probe(device_t dev)
450 {
451 
452 	if (!strcmp(xenbus_get_type(dev), "vif")) {
453 		device_set_desc(dev, "Virtual Network Interface");
454 		return (0);
455 	}
456 
457 	return (ENXIO);
458 }
459 
460 static int
461 netfront_attach(device_t dev)
462 {
463 	int err;
464 
465 	err = create_netdev(dev);
466 	if (err) {
467 		xenbus_dev_fatal(dev, err, "creating netdev");
468 		return (err);
469 	}
470 
471 #if __FreeBSD_version >= 700000
472 	SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
473 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
474 	    OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
475 	    &xn_enable_lro, 0, "Large Receive Offload");
476 #endif
477 
478 	return (0);
479 }
480 
481 static int
482 netfront_suspend(device_t dev)
483 {
484 	struct netfront_info *info = device_get_softc(dev);
485 
486 	XN_RX_LOCK(info);
487 	XN_TX_LOCK(info);
488 	netfront_carrier_off(info);
489 	XN_TX_UNLOCK(info);
490 	XN_RX_UNLOCK(info);
491 	return (0);
492 }
493 
494 /**
495  * We are reconnecting to the backend, due to a suspend/resume, or a backend
496  * driver restart.  We tear down our netif structure and recreate it, but
497  * leave the device-layer structures intact so that this is transparent to the
498  * rest of the kernel.
499  */
500 static int
501 netfront_resume(device_t dev)
502 {
503 	struct netfront_info *info = device_get_softc(dev);
504 
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 XenbusStateConnected:
688 	case XenbusStateUnknown:
689 	case XenbusStateClosed:
690 	case XenbusStateReconfigured:
691 	case XenbusStateReconfiguring:
692 		break;
693 	case XenbusStateInitWait:
694 		if (xenbus_get_state(dev) != XenbusStateInitialising)
695 			break;
696 		if (network_connect(sc) != 0)
697 			break;
698 		xenbus_set_state(dev, XenbusStateConnected);
699 #ifdef INET
700 		netfront_send_fake_arp(dev, sc);
701 #endif
702 		break;
703 	case XenbusStateClosing:
704 		xenbus_set_state(dev, XenbusStateClosed);
705 		break;
706 	}
707 }
708 
709 static void
710 xn_free_rx_ring(struct netfront_info *sc)
711 {
712 #if 0
713 	int i;
714 
715 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
716 		if (sc->xn_cdata.rx_mbufs[i] != NULL) {
717 			m_freem(sc->rx_mbufs[i]);
718 			sc->rx_mbufs[i] = NULL;
719 		}
720 	}
721 
722 	sc->rx.rsp_cons = 0;
723 	sc->xn_rx_if->req_prod = 0;
724 	sc->xn_rx_if->event = sc->rx.rsp_cons ;
725 #endif
726 }
727 
728 static void
729 xn_free_tx_ring(struct netfront_info *sc)
730 {
731 #if 0
732 	int i;
733 
734 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
735 		if (sc->tx_mbufs[i] != NULL) {
736 			m_freem(sc->tx_mbufs[i]);
737 			sc->xn_cdata.xn_tx_chain[i] = NULL;
738 		}
739 	}
740 
741 	return;
742 #endif
743 }
744 
745 /**
746  * \brief Verify that there is sufficient space in the Tx ring
747  *        buffer for a maximally sized request to be enqueued.
748  *
749  * A transmit request requires a transmit descriptor for each packet
750  * fragment, plus up to 2 entries for "options" (e.g. TSO).
751  */
752 static inline int
753 xn_tx_slot_available(struct netfront_info *np)
754 {
755 	return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
756 }
757 
758 static void
759 netif_release_tx_bufs(struct netfront_info *np)
760 {
761 	int i;
762 
763 	for (i = 1; i <= NET_TX_RING_SIZE; i++) {
764 		struct mbuf *m;
765 
766 		m = np->tx_mbufs[i];
767 
768 		/*
769 		 * We assume that no kernel addresses are
770 		 * less than NET_TX_RING_SIZE.  Any entry
771 		 * in the table that is below this number
772 		 * must be an index from free-list tracking.
773 		 */
774 		if (((uintptr_t)m) <= NET_TX_RING_SIZE)
775 			continue;
776 		gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
777 		gnttab_release_grant_reference(&np->gref_tx_head,
778 		    np->grant_tx_ref[i]);
779 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
780 		add_id_to_freelist(np->tx_mbufs, i);
781 		np->xn_cdata.xn_tx_chain_cnt--;
782 		if (np->xn_cdata.xn_tx_chain_cnt < 0) {
783 			panic("%s: tx_chain_cnt must be >= 0", __func__);
784 		}
785 		m_free(m);
786 	}
787 }
788 
789 static void
790 network_alloc_rx_buffers(struct netfront_info *sc)
791 {
792 	int otherend_id = xenbus_get_otherend_id(sc->xbdev);
793 	unsigned short id;
794 	struct mbuf *m_new;
795 	int i, batch_target, notify;
796 	RING_IDX req_prod;
797 	struct xen_memory_reservation reservation;
798 	grant_ref_t ref;
799 	int nr_flips;
800 	netif_rx_request_t *req;
801 	vm_offset_t vaddr;
802 	u_long pfn;
803 
804 	req_prod = sc->rx.req_prod_pvt;
805 
806 	if (__predict_false(sc->carrier == 0))
807 		return;
808 
809 	/*
810 	 * Allocate mbufs greedily, even though we batch updates to the
811 	 * receive ring. This creates a less bursty demand on the memory
812 	 * allocator, and so should reduce the chance of failed allocation
813 	 * requests both for ourself and for other kernel subsystems.
814 	 *
815 	 * Here we attempt to maintain rx_target buffers in flight, counting
816 	 * buffers that we have yet to process in the receive ring.
817 	 */
818 	batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
819 	for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
820 		MGETHDR(m_new, M_NOWAIT, MT_DATA);
821 		if (m_new == NULL) {
822 			printf("%s: MGETHDR failed\n", __func__);
823 			goto no_mbuf;
824 		}
825 
826 		m_cljget(m_new, M_NOWAIT, MJUMPAGESIZE);
827 		if ((m_new->m_flags & M_EXT) == 0) {
828 			printf("%s: m_cljget failed\n", __func__);
829 			m_freem(m_new);
830 
831 no_mbuf:
832 			if (i != 0)
833 				goto refill;
834 			/*
835 			 * XXX set timer
836 			 */
837 			break;
838 		}
839 		m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
840 
841 		/* queue the mbufs allocated */
842 		mbufq_tail(&sc->xn_rx_batch, m_new);
843 	}
844 
845 	/*
846 	 * If we've allocated at least half of our target number of entries,
847 	 * submit them to the backend - we have enough to make the overhead
848 	 * of submission worthwhile.  Otherwise wait for more mbufs and
849 	 * request entries to become available.
850 	 */
851 	if (i < (sc->rx_target/2)) {
852 		if (req_prod >sc->rx.sring->req_prod)
853 			goto push;
854 		return;
855 	}
856 
857 	/*
858 	 * Double floating fill target if we risked having the backend
859 	 * run out of empty buffers for receive traffic.  We define "running
860 	 * low" as having less than a fourth of our target buffers free
861 	 * at the time we refilled the queue.
862 	 */
863 	if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
864 		sc->rx_target *= 2;
865 		if (sc->rx_target > sc->rx_max_target)
866 			sc->rx_target = sc->rx_max_target;
867 	}
868 
869 refill:
870 	for (nr_flips = i = 0; ; i++) {
871 		if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
872 			break;
873 
874 		m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
875 				vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
876 
877 		id = xennet_rxidx(req_prod + i);
878 
879 		KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
880 		sc->rx_mbufs[id] = m_new;
881 
882 		ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
883 		KASSERT(ref != GNTTAB_LIST_END,
884 			("reserved grant references exhuasted"));
885 		sc->grant_rx_ref[id] = ref;
886 
887 		vaddr = mtod(m_new, vm_offset_t);
888 		pfn = vtophys(vaddr) >> PAGE_SHIFT;
889 		req = RING_GET_REQUEST(&sc->rx, req_prod + i);
890 
891 		if (sc->copying_receiver == 0) {
892 			gnttab_grant_foreign_transfer_ref(ref,
893 			    otherend_id, pfn);
894 			sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
895 			if (!xen_feature(XENFEAT_auto_translated_physmap)) {
896 				/* Remove this page before passing
897 				 * back to Xen.
898 				 */
899 				set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
900 				MULTI_update_va_mapping(&sc->rx_mcl[i],
901 				    vaddr, 0, 0);
902 			}
903 			nr_flips++;
904 		} else {
905 			gnttab_grant_foreign_access_ref(ref,
906 			    otherend_id,
907 			    PFNTOMFN(pfn), 0);
908 		}
909 		req->id = id;
910 		req->gref = ref;
911 
912 		sc->rx_pfn_array[i] =
913 		    vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
914 	}
915 
916 	KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
917 	KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
918 	/*
919 	 * We may have allocated buffers which have entries outstanding
920 	 * in the page * update queue -- make sure we flush those first!
921 	 */
922 	PT_UPDATES_FLUSH();
923 	if (nr_flips != 0) {
924 #ifdef notyet
925 		/* Tell the ballon driver what is going on. */
926 		balloon_update_driver_allowance(i);
927 #endif
928 		set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
929 		reservation.nr_extents   = i;
930 		reservation.extent_order = 0;
931 		reservation.address_bits = 0;
932 		reservation.domid        = DOMID_SELF;
933 
934 		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
935 			/* After all PTEs have been zapped, flush the TLB. */
936 			sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
937 			    UVMF_TLB_FLUSH|UVMF_ALL;
938 
939 			/* Give away a batch of pages. */
940 			sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
941 			sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
942 			sc->rx_mcl[i].args[1] =  (u_long)&reservation;
943 			/* Zap PTEs and give away pages in one big multicall. */
944 			(void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
945 
946 			if (__predict_false(sc->rx_mcl[i].result != i ||
947 			    HYPERVISOR_memory_op(XENMEM_decrease_reservation,
948 			    &reservation) != i))
949 				panic("%s: unable to reduce memory "
950 				    "reservation\n", __func__);
951 		}
952 	} else {
953 		wmb();
954 	}
955 
956 	/* Above is a suitable barrier to ensure backend will see requests. */
957 	sc->rx.req_prod_pvt = req_prod + i;
958 push:
959 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
960 	if (notify)
961 		xen_intr_signal(sc->xen_intr_handle);
962 }
963 
964 static void
965 xn_rxeof(struct netfront_info *np)
966 {
967 	struct ifnet *ifp;
968 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
969 	struct lro_ctrl *lro = &np->xn_lro;
970 	struct lro_entry *queued;
971 #endif
972 	struct netfront_rx_info rinfo;
973 	struct netif_rx_response *rx = &rinfo.rx;
974 	struct netif_extra_info *extras = rinfo.extras;
975 	RING_IDX i, rp;
976 	multicall_entry_t *mcl;
977 	struct mbuf *m;
978 	struct mbuf_head rxq, errq;
979 	int err, pages_flipped = 0, work_to_do;
980 
981 	do {
982 		XN_RX_LOCK_ASSERT(np);
983 		if (!netfront_carrier_ok(np))
984 			return;
985 
986 		mbufq_init(&errq);
987 		mbufq_init(&rxq);
988 
989 		ifp = np->xn_ifp;
990 
991 		rp = np->rx.sring->rsp_prod;
992 		rmb();	/* Ensure we see queued responses up to 'rp'. */
993 
994 		i = np->rx.rsp_cons;
995 		while ((i != rp)) {
996 			memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
997 			memset(extras, 0, sizeof(rinfo.extras));
998 
999 			m = NULL;
1000 			err = xennet_get_responses(np, &rinfo, rp, &i, &m,
1001 			    &pages_flipped);
1002 
1003 			if (__predict_false(err)) {
1004 				if (m)
1005 					mbufq_tail(&errq, m);
1006 				np->stats.rx_errors++;
1007 				continue;
1008 			}
1009 
1010 			m->m_pkthdr.rcvif = ifp;
1011 			if ( rx->flags & NETRXF_data_validated ) {
1012 				/* Tell the stack the checksums are okay */
1013 				/*
1014 				 * XXX this isn't necessarily the case - need to add
1015 				 * check
1016 				 */
1017 
1018 				m->m_pkthdr.csum_flags |=
1019 					(CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
1020 					    | CSUM_PSEUDO_HDR);
1021 				m->m_pkthdr.csum_data = 0xffff;
1022 			}
1023 
1024 			np->stats.rx_packets++;
1025 			np->stats.rx_bytes += m->m_pkthdr.len;
1026 
1027 			mbufq_tail(&rxq, m);
1028 			np->rx.rsp_cons = i;
1029 		}
1030 
1031 		if (pages_flipped) {
1032 			/* Some pages are no longer absent... */
1033 #ifdef notyet
1034 			balloon_update_driver_allowance(-pages_flipped);
1035 #endif
1036 			/* Do all the remapping work, and M->P updates, in one big
1037 			 * hypercall.
1038 			 */
1039 			if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1040 				mcl = np->rx_mcl + pages_flipped;
1041 				mcl->op = __HYPERVISOR_mmu_update;
1042 				mcl->args[0] = (u_long)np->rx_mmu;
1043 				mcl->args[1] = pages_flipped;
1044 				mcl->args[2] = 0;
1045 				mcl->args[3] = DOMID_SELF;
1046 				(void)HYPERVISOR_multicall(np->rx_mcl,
1047 				    pages_flipped + 1);
1048 			}
1049 		}
1050 
1051 		while ((m = mbufq_dequeue(&errq)))
1052 			m_freem(m);
1053 
1054 		/*
1055 		 * Process all the mbufs after the remapping is complete.
1056 		 * Break the mbuf chain first though.
1057 		 */
1058 		while ((m = mbufq_dequeue(&rxq)) != NULL) {
1059 			ifp->if_ipackets++;
1060 
1061 			/*
1062 			 * Do we really need to drop the rx lock?
1063 			 */
1064 			XN_RX_UNLOCK(np);
1065 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1066 			/* Use LRO if possible */
1067 			if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1068 			    lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1069 				/*
1070 				 * If LRO fails, pass up to the stack
1071 				 * directly.
1072 				 */
1073 				(*ifp->if_input)(ifp, m);
1074 			}
1075 #else
1076 			(*ifp->if_input)(ifp, m);
1077 #endif
1078 			XN_RX_LOCK(np);
1079 		}
1080 
1081 		np->rx.rsp_cons = i;
1082 
1083 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1084 		/*
1085 		 * Flush any outstanding LRO work
1086 		 */
1087 		while (!SLIST_EMPTY(&lro->lro_active)) {
1088 			queued = SLIST_FIRST(&lro->lro_active);
1089 			SLIST_REMOVE_HEAD(&lro->lro_active, next);
1090 			tcp_lro_flush(lro, queued);
1091 		}
1092 #endif
1093 
1094 #if 0
1095 		/* If we get a callback with very few responses, reduce fill target. */
1096 		/* NB. Note exponential increase, linear decrease. */
1097 		if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1098 			((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1099 			np->rx_target = np->rx_min_target;
1100 #endif
1101 
1102 		network_alloc_rx_buffers(np);
1103 
1104 		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1105 	} while (work_to_do);
1106 }
1107 
1108 static void
1109 xn_txeof(struct netfront_info *np)
1110 {
1111 	RING_IDX i, prod;
1112 	unsigned short id;
1113 	struct ifnet *ifp;
1114 	netif_tx_response_t *txr;
1115 	struct mbuf *m;
1116 
1117 	XN_TX_LOCK_ASSERT(np);
1118 
1119 	if (!netfront_carrier_ok(np))
1120 		return;
1121 
1122 	ifp = np->xn_ifp;
1123 
1124 	do {
1125 		prod = np->tx.sring->rsp_prod;
1126 		rmb(); /* Ensure we see responses up to 'rp'. */
1127 
1128 		for (i = np->tx.rsp_cons; i != prod; i++) {
1129 			txr = RING_GET_RESPONSE(&np->tx, i);
1130 			if (txr->status == NETIF_RSP_NULL)
1131 				continue;
1132 
1133 			if (txr->status != NETIF_RSP_OKAY) {
1134 				printf("%s: WARNING: response is %d!\n",
1135 				       __func__, txr->status);
1136 			}
1137 			id = txr->id;
1138 			m = np->tx_mbufs[id];
1139 			KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1140 			KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1141 				("mbuf already on the free list, but we're "
1142 				"trying to free it again!"));
1143 			M_ASSERTVALID(m);
1144 
1145 			/*
1146 			 * Increment packet count if this is the last
1147 			 * mbuf of the chain.
1148 			 */
1149 			if (!m->m_next)
1150 				ifp->if_opackets++;
1151 			if (__predict_false(gnttab_query_foreign_access(
1152 			    np->grant_tx_ref[id]) != 0)) {
1153 				panic("%s: grant id %u still in use by the "
1154 				    "backend", __func__, id);
1155 			}
1156 			gnttab_end_foreign_access_ref(
1157 				np->grant_tx_ref[id]);
1158 			gnttab_release_grant_reference(
1159 				&np->gref_tx_head, np->grant_tx_ref[id]);
1160 			np->grant_tx_ref[id] = GRANT_REF_INVALID;
1161 
1162 			np->tx_mbufs[id] = NULL;
1163 			add_id_to_freelist(np->tx_mbufs, id);
1164 			np->xn_cdata.xn_tx_chain_cnt--;
1165 			m_free(m);
1166 			/* Only mark the queue active if we've freed up at least one slot to try */
1167 			ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1168 		}
1169 		np->tx.rsp_cons = prod;
1170 
1171 		/*
1172 		 * Set a new event, then check for race with update of
1173 		 * tx_cons. Note that it is essential to schedule a
1174 		 * callback, no matter how few buffers are pending. Even if
1175 		 * there is space in the transmit ring, higher layers may
1176 		 * be blocked because too much data is outstanding: in such
1177 		 * cases notification from Xen is likely to be the only kick
1178 		 * that we'll get.
1179 		 */
1180 		np->tx.sring->rsp_event =
1181 		    prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1182 
1183 		mb();
1184 	} while (prod != np->tx.sring->rsp_prod);
1185 
1186 	if (np->tx_full &&
1187 	    ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1188 		np->tx_full = 0;
1189 #if 0
1190 		if (np->user_state == UST_OPEN)
1191 			netif_wake_queue(dev);
1192 #endif
1193 	}
1194 }
1195 
1196 static void
1197 xn_intr(void *xsc)
1198 {
1199 	struct netfront_info *np = xsc;
1200 	struct ifnet *ifp = np->xn_ifp;
1201 
1202 #if 0
1203 	if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1204 	    likely(netfront_carrier_ok(np)) &&
1205 	    ifp->if_drv_flags & IFF_DRV_RUNNING))
1206 		return;
1207 #endif
1208 	if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1209 		XN_TX_LOCK(np);
1210 		xn_txeof(np);
1211 		XN_TX_UNLOCK(np);
1212 	}
1213 
1214 	XN_RX_LOCK(np);
1215 	xn_rxeof(np);
1216 	XN_RX_UNLOCK(np);
1217 
1218 	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1219 	    !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1220 		xn_start(ifp);
1221 }
1222 
1223 static void
1224 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1225 	grant_ref_t ref)
1226 {
1227 	int new = xennet_rxidx(np->rx.req_prod_pvt);
1228 
1229 	KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1230 	np->rx_mbufs[new] = m;
1231 	np->grant_rx_ref[new] = ref;
1232 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1233 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1234 	np->rx.req_prod_pvt++;
1235 }
1236 
1237 static int
1238 xennet_get_extras(struct netfront_info *np,
1239     struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1240 {
1241 	struct netif_extra_info *extra;
1242 
1243 	int err = 0;
1244 
1245 	do {
1246 		struct mbuf *m;
1247 		grant_ref_t ref;
1248 
1249 		if (__predict_false(*cons + 1 == rp)) {
1250 #if 0
1251 			if (net_ratelimit())
1252 				WPRINTK("Missing extra info\n");
1253 #endif
1254 			err = EINVAL;
1255 			break;
1256 		}
1257 
1258 		extra = (struct netif_extra_info *)
1259 		RING_GET_RESPONSE(&np->rx, ++(*cons));
1260 
1261 		if (__predict_false(!extra->type ||
1262 			extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1263 #if 0
1264 			if (net_ratelimit())
1265 				WPRINTK("Invalid extra type: %d\n",
1266 					extra->type);
1267 #endif
1268 			err = EINVAL;
1269 		} else {
1270 			memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1271 		}
1272 
1273 		m = xennet_get_rx_mbuf(np, *cons);
1274 		ref = xennet_get_rx_ref(np, *cons);
1275 		xennet_move_rx_slot(np, m, ref);
1276 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1277 
1278 	return err;
1279 }
1280 
1281 static int
1282 xennet_get_responses(struct netfront_info *np,
1283 	struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1284 	struct mbuf  **list,
1285 	int *pages_flipped_p)
1286 {
1287 	int pages_flipped = *pages_flipped_p;
1288 	struct mmu_update *mmu;
1289 	struct multicall_entry *mcl;
1290 	struct netif_rx_response *rx = &rinfo->rx;
1291 	struct netif_extra_info *extras = rinfo->extras;
1292 	struct mbuf *m, *m0, *m_prev;
1293 	grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1294 	RING_IDX ref_cons = *cons;
1295 	int frags = 1;
1296 	int err = 0;
1297 	u_long ret;
1298 
1299 	m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1300 
1301 	if (rx->flags & NETRXF_extra_info) {
1302 		err = xennet_get_extras(np, extras, rp, cons);
1303 	}
1304 
1305 	if (m0 != NULL) {
1306 		m0->m_pkthdr.len = 0;
1307 		m0->m_next = NULL;
1308 	}
1309 
1310 	for (;;) {
1311 		u_long mfn;
1312 
1313 #if 0
1314 		DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1315 			rx->status, rx->offset, frags);
1316 #endif
1317 		if (__predict_false(rx->status < 0 ||
1318 			rx->offset + rx->status > PAGE_SIZE)) {
1319 
1320 #if 0
1321 			if (net_ratelimit())
1322 				WPRINTK("rx->offset: %x, size: %u\n",
1323 					rx->offset, rx->status);
1324 #endif
1325 			xennet_move_rx_slot(np, m, ref);
1326 			if (m0 == m)
1327 				m0 = NULL;
1328 			m = NULL;
1329 			err = EINVAL;
1330 			goto next_skip_queue;
1331 		}
1332 
1333 		/*
1334 		 * This definitely indicates a bug, either in this driver or in
1335 		 * the backend driver. In future this should flag the bad
1336 		 * situation to the system controller to reboot the backed.
1337 		 */
1338 		if (ref == GRANT_REF_INVALID) {
1339 
1340 #if 0
1341 			if (net_ratelimit())
1342 				WPRINTK("Bad rx response id %d.\n", rx->id);
1343 #endif
1344 			printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1345 			err = EINVAL;
1346 			goto next;
1347 		}
1348 
1349 		if (!np->copying_receiver) {
1350 			/* Memory pressure, insufficient buffer
1351 			 * headroom, ...
1352 			 */
1353 			if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1354 				WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1355 					rx->id, rx->status);
1356 				xennet_move_rx_slot(np, m, ref);
1357 				err = ENOMEM;
1358 				goto next;
1359 			}
1360 
1361 			if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1362 				/* Remap the page. */
1363 				void *vaddr = mtod(m, void *);
1364 				uint32_t pfn;
1365 
1366 				mcl = np->rx_mcl + pages_flipped;
1367 				mmu = np->rx_mmu + pages_flipped;
1368 
1369 				MULTI_update_va_mapping(mcl, (u_long)vaddr,
1370 				    (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1371 				    PG_V | PG_M | PG_A, 0);
1372 				pfn = (uintptr_t)m->m_ext.ext_arg1;
1373 				mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1374 				    MMU_MACHPHYS_UPDATE;
1375 				mmu->val = pfn;
1376 
1377 				set_phys_to_machine(pfn, mfn);
1378 			}
1379 			pages_flipped++;
1380 		} else {
1381 			ret = gnttab_end_foreign_access_ref(ref);
1382 			KASSERT(ret, ("ret != 0"));
1383 		}
1384 
1385 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1386 
1387 next:
1388 		if (m == NULL)
1389 			break;
1390 
1391 		m->m_len = rx->status;
1392 		m->m_data += rx->offset;
1393 		m0->m_pkthdr.len += rx->status;
1394 
1395 next_skip_queue:
1396 		if (!(rx->flags & NETRXF_more_data))
1397 			break;
1398 
1399 		if (*cons + frags == rp) {
1400 			if (net_ratelimit())
1401 				WPRINTK("Need more frags\n");
1402 			err = ENOENT;
1403 			printf("%s: cons %u frags %u rp %u, not enough frags\n",
1404 			       __func__, *cons, frags, rp);
1405 			break;
1406 		}
1407 		/*
1408 		 * Note that m can be NULL, if rx->status < 0 or if
1409 		 * rx->offset + rx->status > PAGE_SIZE above.
1410 		 */
1411 		m_prev = m;
1412 
1413 		rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1414 		m = xennet_get_rx_mbuf(np, *cons + frags);
1415 
1416 		/*
1417 		 * m_prev == NULL can happen if rx->status < 0 or if
1418 		 * rx->offset + * rx->status > PAGE_SIZE above.
1419 		 */
1420 		if (m_prev != NULL)
1421 			m_prev->m_next = m;
1422 
1423 		/*
1424 		 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1425 		 * rx->status > PAGE_SIZE above.
1426 		 */
1427 		if (m0 == NULL)
1428 			m0 = m;
1429 		m->m_next = NULL;
1430 		ref = xennet_get_rx_ref(np, *cons + frags);
1431 		ref_cons = *cons + frags;
1432 		frags++;
1433 	}
1434 	*list = m0;
1435 	*cons += frags;
1436 	*pages_flipped_p = pages_flipped;
1437 
1438 	return (err);
1439 }
1440 
1441 static void
1442 xn_tick_locked(struct netfront_info *sc)
1443 {
1444 	XN_RX_LOCK_ASSERT(sc);
1445 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1446 
1447 	/* XXX placeholder for printing debug information */
1448 }
1449 
1450 static void
1451 xn_tick(void *xsc)
1452 {
1453 	struct netfront_info *sc;
1454 
1455 	sc = xsc;
1456 	XN_RX_LOCK(sc);
1457 	xn_tick_locked(sc);
1458 	XN_RX_UNLOCK(sc);
1459 }
1460 
1461 /**
1462  * \brief Count the number of fragments in an mbuf chain.
1463  *
1464  * Surprisingly, there isn't an M* macro for this.
1465  */
1466 static inline int
1467 xn_count_frags(struct mbuf *m)
1468 {
1469 	int nfrags;
1470 
1471 	for (nfrags = 0; m != NULL; m = m->m_next)
1472 		nfrags++;
1473 
1474 	return (nfrags);
1475 }
1476 
1477 /**
1478  * Given an mbuf chain, make sure we have enough room and then push
1479  * it onto the transmit ring.
1480  */
1481 static int
1482 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1483 {
1484 	struct ifnet *ifp;
1485 	struct mbuf *m;
1486 	u_int nfrags;
1487 	netif_extra_info_t *extra;
1488 	int otherend_id;
1489 
1490 	ifp = sc->xn_ifp;
1491 
1492 	/**
1493 	 * Defragment the mbuf if necessary.
1494 	 */
1495 	nfrags = xn_count_frags(m_head);
1496 
1497 	/*
1498 	 * Check to see whether this request is longer than netback
1499 	 * can handle, and try to defrag it.
1500 	 */
1501 	/**
1502 	 * It is a bit lame, but the netback driver in Linux can't
1503 	 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1504 	 * the Linux network stack.
1505 	 */
1506 	if (nfrags > sc->maxfrags) {
1507 		m = m_defrag(m_head, M_NOWAIT);
1508 		if (!m) {
1509 			/*
1510 			 * Defrag failed, so free the mbuf and
1511 			 * therefore drop the packet.
1512 			 */
1513 			m_freem(m_head);
1514 			return (EMSGSIZE);
1515 		}
1516 		m_head = m;
1517 	}
1518 
1519 	/* Determine how many fragments now exist */
1520 	nfrags = xn_count_frags(m_head);
1521 
1522 	/*
1523 	 * Check to see whether the defragmented packet has too many
1524 	 * segments for the Linux netback driver.
1525 	 */
1526 	/**
1527 	 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1528 	 * of mbufs longer than Linux can handle.  Make sure we don't
1529 	 * pass a too-long chain over to the other side by dropping the
1530 	 * packet.  It doesn't look like there is currently a way to
1531 	 * tell the TCP stack to generate a shorter chain of packets.
1532 	 */
1533 	if (nfrags > MAX_TX_REQ_FRAGS) {
1534 #ifdef DEBUG
1535 		printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1536 		       "won't be able to handle it, dropping\n",
1537 		       __func__, nfrags, MAX_TX_REQ_FRAGS);
1538 #endif
1539 		m_freem(m_head);
1540 		return (EMSGSIZE);
1541 	}
1542 
1543 	/*
1544 	 * This check should be redundant.  We've already verified that we
1545 	 * have enough slots in the ring to handle a packet of maximum
1546 	 * size, and that our packet is less than the maximum size.  Keep
1547 	 * it in here as an assert for now just to make certain that
1548 	 * xn_tx_chain_cnt is accurate.
1549 	 */
1550 	KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1551 		("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1552 		 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1553                     (int) nfrags, (int) NET_TX_RING_SIZE));
1554 
1555 	/*
1556 	 * Start packing the mbufs in this chain into
1557 	 * the fragment pointers. Stop when we run out
1558 	 * of fragments or hit the end of the mbuf chain.
1559 	 */
1560 	m = m_head;
1561 	extra = NULL;
1562 	otherend_id = xenbus_get_otherend_id(sc->xbdev);
1563 	for (m = m_head; m; m = m->m_next) {
1564 		netif_tx_request_t *tx;
1565 		uintptr_t id;
1566 		grant_ref_t ref;
1567 		u_long mfn; /* XXX Wrong type? */
1568 
1569 		tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1570 		id = get_id_from_freelist(sc->tx_mbufs);
1571 		if (id == 0)
1572 			panic("%s: was allocated the freelist head!\n",
1573 			    __func__);
1574 		sc->xn_cdata.xn_tx_chain_cnt++;
1575 		if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1576 			panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1577 			    __func__);
1578 		sc->tx_mbufs[id] = m;
1579 		tx->id = id;
1580 		ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1581 		KASSERT((short)ref >= 0, ("Negative ref"));
1582 		mfn = virt_to_mfn(mtod(m, vm_offset_t));
1583 		gnttab_grant_foreign_access_ref(ref, otherend_id,
1584 		    mfn, GNTMAP_readonly);
1585 		tx->gref = sc->grant_tx_ref[id] = ref;
1586 		tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1587 		tx->flags = 0;
1588 		if (m == m_head) {
1589 			/*
1590 			 * The first fragment has the entire packet
1591 			 * size, subsequent fragments have just the
1592 			 * fragment size. The backend works out the
1593 			 * true size of the first fragment by
1594 			 * subtracting the sizes of the other
1595 			 * fragments.
1596 			 */
1597 			tx->size = m->m_pkthdr.len;
1598 
1599 			/*
1600 			 * The first fragment contains the checksum flags
1601 			 * and is optionally followed by extra data for
1602 			 * TSO etc.
1603 			 */
1604 			/**
1605 			 * CSUM_TSO requires checksum offloading.
1606 			 * Some versions of FreeBSD fail to
1607 			 * set CSUM_TCP in the CSUM_TSO case,
1608 			 * so we have to test for CSUM_TSO
1609 			 * explicitly.
1610 			 */
1611 			if (m->m_pkthdr.csum_flags
1612 			    & (CSUM_DELAY_DATA | CSUM_TSO)) {
1613 				tx->flags |= (NETTXF_csum_blank
1614 				    | NETTXF_data_validated);
1615 			}
1616 #if __FreeBSD_version >= 700000
1617 			if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1618 				struct netif_extra_info *gso =
1619 					(struct netif_extra_info *)
1620 					RING_GET_REQUEST(&sc->tx,
1621 							 ++sc->tx.req_prod_pvt);
1622 
1623 				tx->flags |= NETTXF_extra_info;
1624 
1625 				gso->u.gso.size = m->m_pkthdr.tso_segsz;
1626 				gso->u.gso.type =
1627 					XEN_NETIF_GSO_TYPE_TCPV4;
1628 				gso->u.gso.pad = 0;
1629 				gso->u.gso.features = 0;
1630 
1631 				gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1632 				gso->flags = 0;
1633 			}
1634 #endif
1635 		} else {
1636 			tx->size = m->m_len;
1637 		}
1638 		if (m->m_next)
1639 			tx->flags |= NETTXF_more_data;
1640 
1641 		sc->tx.req_prod_pvt++;
1642 	}
1643 	BPF_MTAP(ifp, m_head);
1644 
1645 	sc->stats.tx_bytes += m_head->m_pkthdr.len;
1646 	sc->stats.tx_packets++;
1647 
1648 	return (0);
1649 }
1650 
1651 static void
1652 xn_start_locked(struct ifnet *ifp)
1653 {
1654 	struct netfront_info *sc;
1655 	struct mbuf *m_head;
1656 	int notify;
1657 
1658 	sc = ifp->if_softc;
1659 
1660 	if (!netfront_carrier_ok(sc))
1661 		return;
1662 
1663 	/*
1664 	 * While we have enough transmit slots available for at least one
1665 	 * maximum-sized packet, pull mbufs off the queue and put them on
1666 	 * the transmit ring.
1667 	 */
1668 	while (xn_tx_slot_available(sc)) {
1669 		IF_DEQUEUE(&ifp->if_snd, m_head);
1670 		if (m_head == NULL)
1671 			break;
1672 
1673 		if (xn_assemble_tx_request(sc, m_head) != 0)
1674 			break;
1675 	}
1676 
1677 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1678 	if (notify)
1679 		xen_intr_signal(sc->xen_intr_handle);
1680 
1681 	if (RING_FULL(&sc->tx)) {
1682 		sc->tx_full = 1;
1683 #if 0
1684 		netif_stop_queue(dev);
1685 #endif
1686 	}
1687 }
1688 
1689 static void
1690 xn_start(struct ifnet *ifp)
1691 {
1692 	struct netfront_info *sc;
1693 	sc = ifp->if_softc;
1694 	XN_TX_LOCK(sc);
1695 	xn_start_locked(ifp);
1696 	XN_TX_UNLOCK(sc);
1697 }
1698 
1699 /* equivalent of network_open() in Linux */
1700 static void
1701 xn_ifinit_locked(struct netfront_info *sc)
1702 {
1703 	struct ifnet *ifp;
1704 
1705 	XN_LOCK_ASSERT(sc);
1706 
1707 	ifp = sc->xn_ifp;
1708 
1709 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1710 		return;
1711 
1712 	xn_stop(sc);
1713 
1714 	network_alloc_rx_buffers(sc);
1715 	sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1716 
1717 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1718 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1719 	if_link_state_change(ifp, LINK_STATE_UP);
1720 
1721 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1722 }
1723 
1724 static void
1725 xn_ifinit(void *xsc)
1726 {
1727 	struct netfront_info *sc = xsc;
1728 
1729 	XN_LOCK(sc);
1730 	xn_ifinit_locked(sc);
1731 	XN_UNLOCK(sc);
1732 }
1733 
1734 static int
1735 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1736 {
1737 	struct netfront_info *sc = ifp->if_softc;
1738 	struct ifreq *ifr = (struct ifreq *) data;
1739 #ifdef INET
1740 	struct ifaddr *ifa = (struct ifaddr *)data;
1741 #endif
1742 
1743 	int mask, error = 0;
1744 	switch(cmd) {
1745 	case SIOCSIFADDR:
1746 	case SIOCGIFADDR:
1747 #ifdef INET
1748 		XN_LOCK(sc);
1749 		if (ifa->ifa_addr->sa_family == AF_INET) {
1750 			ifp->if_flags |= IFF_UP;
1751 			if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1752 				xn_ifinit_locked(sc);
1753 			arp_ifinit(ifp, ifa);
1754 			XN_UNLOCK(sc);
1755 		} else {
1756 			XN_UNLOCK(sc);
1757 #endif
1758 			error = ether_ioctl(ifp, cmd, data);
1759 #ifdef INET
1760 		}
1761 #endif
1762 		break;
1763 	case SIOCSIFMTU:
1764 		/* XXX can we alter the MTU on a VN ?*/
1765 #ifdef notyet
1766 		if (ifr->ifr_mtu > XN_JUMBO_MTU)
1767 			error = EINVAL;
1768 		else
1769 #endif
1770 		{
1771 			ifp->if_mtu = ifr->ifr_mtu;
1772 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1773 			xn_ifinit(sc);
1774 		}
1775 		break;
1776 	case SIOCSIFFLAGS:
1777 		XN_LOCK(sc);
1778 		if (ifp->if_flags & IFF_UP) {
1779 			/*
1780 			 * If only the state of the PROMISC flag changed,
1781 			 * then just use the 'set promisc mode' command
1782 			 * instead of reinitializing the entire NIC. Doing
1783 			 * a full re-init means reloading the firmware and
1784 			 * waiting for it to start up, which may take a
1785 			 * second or two.
1786 			 */
1787 #ifdef notyet
1788 			/* No promiscuous mode with Xen */
1789 			if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1790 			    ifp->if_flags & IFF_PROMISC &&
1791 			    !(sc->xn_if_flags & IFF_PROMISC)) {
1792 				XN_SETBIT(sc, XN_RX_MODE,
1793 					  XN_RXMODE_RX_PROMISC);
1794 			} else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1795 				   !(ifp->if_flags & IFF_PROMISC) &&
1796 				   sc->xn_if_flags & IFF_PROMISC) {
1797 				XN_CLRBIT(sc, XN_RX_MODE,
1798 					  XN_RXMODE_RX_PROMISC);
1799 			} else
1800 #endif
1801 				xn_ifinit_locked(sc);
1802 		} else {
1803 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1804 				xn_stop(sc);
1805 			}
1806 		}
1807 		sc->xn_if_flags = ifp->if_flags;
1808 		XN_UNLOCK(sc);
1809 		error = 0;
1810 		break;
1811 	case SIOCSIFCAP:
1812 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1813 		if (mask & IFCAP_TXCSUM) {
1814 			if (IFCAP_TXCSUM & ifp->if_capenable) {
1815 				ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1816 				ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1817 				    | CSUM_IP | CSUM_TSO);
1818 			} else {
1819 				ifp->if_capenable |= IFCAP_TXCSUM;
1820 				ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1821 				    | CSUM_IP);
1822 			}
1823 		}
1824 		if (mask & IFCAP_RXCSUM) {
1825 			ifp->if_capenable ^= IFCAP_RXCSUM;
1826 		}
1827 #if __FreeBSD_version >= 700000
1828 		if (mask & IFCAP_TSO4) {
1829 			if (IFCAP_TSO4 & ifp->if_capenable) {
1830 				ifp->if_capenable &= ~IFCAP_TSO4;
1831 				ifp->if_hwassist &= ~CSUM_TSO;
1832 			} else if (IFCAP_TXCSUM & ifp->if_capenable) {
1833 				ifp->if_capenable |= IFCAP_TSO4;
1834 				ifp->if_hwassist |= CSUM_TSO;
1835 			} else {
1836 				IPRINTK("Xen requires tx checksum offload"
1837 				    " be enabled to use TSO\n");
1838 				error = EINVAL;
1839 			}
1840 		}
1841 		if (mask & IFCAP_LRO) {
1842 			ifp->if_capenable ^= IFCAP_LRO;
1843 
1844 		}
1845 #endif
1846 		error = 0;
1847 		break;
1848 	case SIOCADDMULTI:
1849 	case SIOCDELMULTI:
1850 #ifdef notyet
1851 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1852 			XN_LOCK(sc);
1853 			xn_setmulti(sc);
1854 			XN_UNLOCK(sc);
1855 			error = 0;
1856 		}
1857 #endif
1858 		/* FALLTHROUGH */
1859 	case SIOCSIFMEDIA:
1860 	case SIOCGIFMEDIA:
1861 		error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1862 		break;
1863 	default:
1864 		error = ether_ioctl(ifp, cmd, data);
1865 	}
1866 
1867 	return (error);
1868 }
1869 
1870 static void
1871 xn_stop(struct netfront_info *sc)
1872 {
1873 	struct ifnet *ifp;
1874 
1875 	XN_LOCK_ASSERT(sc);
1876 
1877 	ifp = sc->xn_ifp;
1878 
1879 	callout_stop(&sc->xn_stat_ch);
1880 
1881 	xn_free_rx_ring(sc);
1882 	xn_free_tx_ring(sc);
1883 
1884 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1885 	if_link_state_change(ifp, LINK_STATE_DOWN);
1886 }
1887 
1888 /* START of Xenolinux helper functions adapted to FreeBSD */
1889 int
1890 network_connect(struct netfront_info *np)
1891 {
1892 	int i, requeue_idx, error;
1893 	grant_ref_t ref;
1894 	netif_rx_request_t *req;
1895 	u_int feature_rx_copy, feature_rx_flip;
1896 
1897 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1898 	    "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1899 	if (error)
1900 		feature_rx_copy = 0;
1901 	error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1902 	    "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1903 	if (error)
1904 		feature_rx_flip = 1;
1905 
1906 	/*
1907 	 * Copy packets on receive path if:
1908 	 *  (a) This was requested by user, and the backend supports it; or
1909 	 *  (b) Flipping was requested, but this is unsupported by the backend.
1910 	 */
1911 	np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1912 				(MODPARM_rx_flip && !feature_rx_flip));
1913 
1914 	/* Recovery procedure: */
1915 	error = talk_to_backend(np->xbdev, np);
1916 	if (error)
1917 		return (error);
1918 
1919 	/* Step 1: Reinitialise variables. */
1920 	xn_query_features(np);
1921 	xn_configure_features(np);
1922 	netif_release_tx_bufs(np);
1923 
1924 	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1925 	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1926 		struct mbuf *m;
1927 		u_long pfn;
1928 
1929 		if (np->rx_mbufs[i] == NULL)
1930 			continue;
1931 
1932 		m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1933 		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1934 
1935 		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1936 		pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1937 
1938 		if (!np->copying_receiver) {
1939 			gnttab_grant_foreign_transfer_ref(ref,
1940 			    xenbus_get_otherend_id(np->xbdev),
1941 			    pfn);
1942 		} else {
1943 			gnttab_grant_foreign_access_ref(ref,
1944 			    xenbus_get_otherend_id(np->xbdev),
1945 			    PFNTOMFN(pfn), 0);
1946 		}
1947 		req->gref = ref;
1948 		req->id   = requeue_idx;
1949 
1950 		requeue_idx++;
1951 	}
1952 
1953 	np->rx.req_prod_pvt = requeue_idx;
1954 
1955 	/* Step 3: All public and private state should now be sane.  Get
1956 	 * ready to start sending and receiving packets and give the driver
1957 	 * domain a kick because we've probably just requeued some
1958 	 * packets.
1959 	 */
1960 	netfront_carrier_on(np);
1961 	xen_intr_signal(np->xen_intr_handle);
1962 	XN_TX_LOCK(np);
1963 	xn_txeof(np);
1964 	XN_TX_UNLOCK(np);
1965 	network_alloc_rx_buffers(np);
1966 
1967 	return (0);
1968 }
1969 
1970 static void
1971 xn_query_features(struct netfront_info *np)
1972 {
1973 	int val;
1974 
1975 	device_printf(np->xbdev, "backend features:");
1976 
1977 	if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1978 		"feature-sg", NULL, "%d", &val) < 0)
1979 		val = 0;
1980 
1981 	np->maxfrags = 1;
1982 	if (val) {
1983 		np->maxfrags = MAX_TX_REQ_FRAGS;
1984 		printf(" feature-sg");
1985 	}
1986 
1987 	if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1988 		"feature-gso-tcpv4", NULL, "%d", &val) < 0)
1989 		val = 0;
1990 
1991 	np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
1992 	if (val) {
1993 		np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
1994 		printf(" feature-gso-tcp4");
1995 	}
1996 
1997 	printf("\n");
1998 }
1999 
2000 static int
2001 xn_configure_features(struct netfront_info *np)
2002 {
2003 	int err;
2004 
2005 	err = 0;
2006 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
2007 	if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0)
2008 		tcp_lro_free(&np->xn_lro);
2009 #endif
2010     	np->xn_ifp->if_capenable =
2011 	    np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4);
2012 	np->xn_ifp->if_hwassist &= ~CSUM_TSO;
2013 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
2014 	if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) {
2015 		err = tcp_lro_init(&np->xn_lro);
2016 		if (err) {
2017 			device_printf(np->xbdev, "LRO initialization failed\n");
2018 		} else {
2019 			np->xn_lro.ifp = np->xn_ifp;
2020 			np->xn_ifp->if_capenable |= IFCAP_LRO;
2021 		}
2022 	}
2023 	if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) {
2024 		np->xn_ifp->if_capenable |= IFCAP_TSO4;
2025 		np->xn_ifp->if_hwassist |= CSUM_TSO;
2026 	}
2027 #endif
2028 	return (err);
2029 }
2030 
2031 /**
2032  * Create a network device.
2033  * @param dev  Newbus device representing this virtual NIC.
2034  */
2035 int
2036 create_netdev(device_t dev)
2037 {
2038 	int i;
2039 	struct netfront_info *np;
2040 	int err;
2041 	struct ifnet *ifp;
2042 
2043 	np = device_get_softc(dev);
2044 
2045 	np->xbdev         = dev;
2046 
2047 	XN_LOCK_INIT(np, xennetif);
2048 
2049 	ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
2050 	ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
2051 	ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
2052 
2053 	np->rx_target     = RX_MIN_TARGET;
2054 	np->rx_min_target = RX_MIN_TARGET;
2055 	np->rx_max_target = RX_MAX_TARGET;
2056 
2057 	/* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2058 	for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2059 		np->tx_mbufs[i] = (void *) ((u_long) i+1);
2060 		np->grant_tx_ref[i] = GRANT_REF_INVALID;
2061 	}
2062 	np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2063 
2064 	for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2065 
2066 		np->rx_mbufs[i] = NULL;
2067 		np->grant_rx_ref[i] = GRANT_REF_INVALID;
2068 	}
2069 	/* A grant for every tx ring slot */
2070 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2071 					  &np->gref_tx_head) != 0) {
2072 		IPRINTK("#### netfront can't alloc tx grant refs\n");
2073 		err = ENOMEM;
2074 		goto exit;
2075 	}
2076 	/* A grant for every rx ring slot */
2077 	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2078 					  &np->gref_rx_head) != 0) {
2079 		WPRINTK("#### netfront can't alloc rx grant refs\n");
2080 		gnttab_free_grant_references(np->gref_tx_head);
2081 		err = ENOMEM;
2082 		goto exit;
2083 	}
2084 
2085 	err = xen_net_read_mac(dev, np->mac);
2086 	if (err)
2087 		goto out;
2088 
2089 	/* Set up ifnet structure */
2090 	ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2091     	ifp->if_softc = np;
2092     	if_initname(ifp, "xn",  device_get_unit(dev));
2093     	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2094     	ifp->if_ioctl = xn_ioctl;
2095     	ifp->if_output = ether_output;
2096     	ifp->if_start = xn_start;
2097 #ifdef notyet
2098     	ifp->if_watchdog = xn_watchdog;
2099 #endif
2100     	ifp->if_init = xn_ifinit;
2101     	ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2102 
2103     	ifp->if_hwassist = XN_CSUM_FEATURES;
2104     	ifp->if_capabilities = IFCAP_HWCSUM;
2105 	ifp->if_hw_tsomax = NF_TSO_MAXBURST;
2106 
2107     	ether_ifattach(ifp, np->mac);
2108     	callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2109 	netfront_carrier_off(np);
2110 
2111 	return (0);
2112 
2113 exit:
2114 	gnttab_free_grant_references(np->gref_tx_head);
2115 out:
2116 	return (err);
2117 }
2118 
2119 /**
2120  * Handle the change of state of the backend to Closing.  We must delete our
2121  * device-layer structures now, to ensure that writes are flushed through to
2122  * the backend.  Once is this done, we can switch to Closed in
2123  * acknowledgement.
2124  */
2125 #if 0
2126 static void
2127 netfront_closing(device_t dev)
2128 {
2129 #if 0
2130 	struct netfront_info *info = dev->dev_driver_data;
2131 
2132 	DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2133 
2134 	close_netdev(info);
2135 #endif
2136 	xenbus_switch_state(dev, XenbusStateClosed);
2137 }
2138 #endif
2139 
2140 static int
2141 netfront_detach(device_t dev)
2142 {
2143 	struct netfront_info *info = device_get_softc(dev);
2144 
2145 	DPRINTK("%s\n", xenbus_get_node(dev));
2146 
2147 	netif_free(info);
2148 
2149 	return 0;
2150 }
2151 
2152 static void
2153 netif_free(struct netfront_info *info)
2154 {
2155 	XN_LOCK(info);
2156 	xn_stop(info);
2157 	XN_UNLOCK(info);
2158 	callout_drain(&info->xn_stat_ch);
2159 	netif_disconnect_backend(info);
2160 	if (info->xn_ifp != NULL) {
2161 		ether_ifdetach(info->xn_ifp);
2162 		if_free(info->xn_ifp);
2163 		info->xn_ifp = NULL;
2164 	}
2165 	ifmedia_removeall(&info->sc_media);
2166 }
2167 
2168 static void
2169 netif_disconnect_backend(struct netfront_info *info)
2170 {
2171 	XN_RX_LOCK(info);
2172 	XN_TX_LOCK(info);
2173 	netfront_carrier_off(info);
2174 	XN_TX_UNLOCK(info);
2175 	XN_RX_UNLOCK(info);
2176 
2177 	free_ring(&info->tx_ring_ref, &info->tx.sring);
2178 	free_ring(&info->rx_ring_ref, &info->rx.sring);
2179 
2180 	xen_intr_unbind(&info->xen_intr_handle);
2181 }
2182 
2183 static void
2184 free_ring(int *ref, void *ring_ptr_ref)
2185 {
2186 	void **ring_ptr_ptr = ring_ptr_ref;
2187 
2188 	if (*ref != GRANT_REF_INVALID) {
2189 		/* This API frees the associated storage. */
2190 		gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
2191 		*ref = GRANT_REF_INVALID;
2192 	}
2193 	*ring_ptr_ptr = NULL;
2194 }
2195 
2196 static int
2197 xn_ifmedia_upd(struct ifnet *ifp)
2198 {
2199 	return (0);
2200 }
2201 
2202 static void
2203 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2204 {
2205 	ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2206 	ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2207 }
2208 
2209 /* ** Driver registration ** */
2210 static device_method_t netfront_methods[] = {
2211 	/* Device interface */
2212 	DEVMETHOD(device_probe,         netfront_probe),
2213 	DEVMETHOD(device_attach,        netfront_attach),
2214 	DEVMETHOD(device_detach,        netfront_detach),
2215 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
2216 	DEVMETHOD(device_suspend,       netfront_suspend),
2217 	DEVMETHOD(device_resume,        netfront_resume),
2218 
2219 	/* Xenbus interface */
2220 	DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2221 
2222 	DEVMETHOD_END
2223 };
2224 
2225 static driver_t netfront_driver = {
2226 	"xn",
2227 	netfront_methods,
2228 	sizeof(struct netfront_info),
2229 };
2230 devclass_t netfront_devclass;
2231 
2232 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,
2233     NULL);
2234