xref: /freebsd/sys/dev/xen/netfront/netfront.c (revision 2f98b86b92c842f67e08a7f2e6cf43f39b8ad3ce)
1 /*
2  *
3  * Copyright (c) 2004-2006 Kip Macy
4  * All rights reserved.
5  *
6  *
7  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
8  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
9  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
10  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
11  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
12  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
13  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
14  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
15  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
16  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
17  */
18 
19 
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
22 
23 #include <sys/param.h>
24 #include <sys/systm.h>
25 #include <sys/sockio.h>
26 #include <sys/mbuf.h>
27 #include <sys/malloc.h>
28 #include <sys/module.h>
29 #include <sys/kernel.h>
30 #include <sys/socket.h>
31 #include <sys/sysctl.h>
32 #include <sys/queue.h>
33 #include <sys/sx.h>
34 
35 #include <net/if.h>
36 #include <net/if_arp.h>
37 #include <net/ethernet.h>
38 #include <net/if_dl.h>
39 #include <net/if_media.h>
40 
41 #include <net/bpf.h>
42 
43 #include <net/if_types.h>
44 #include <net/if.h>
45 
46 #include <netinet/in_systm.h>
47 #include <netinet/in.h>
48 #include <netinet/ip.h>
49 #include <netinet/if_ether.h>
50 #if __FreeBSD_version >= 700000
51 #include <netinet/tcp.h>
52 #include <netinet/tcp_lro.h>
53 #endif
54 
55 #include <vm/vm.h>
56 #include <vm/pmap.h>
57 
58 #include <machine/clock.h>      /* for DELAY */
59 #include <machine/bus.h>
60 #include <machine/resource.h>
61 #include <machine/frame.h>
62 #include <machine/vmparam.h>
63 
64 #include <sys/bus.h>
65 #include <sys/rman.h>
66 
67 #include <machine/intr_machdep.h>
68 
69 #include <machine/xen/xen-os.h>
70 #include <machine/xen/xenfunc.h>
71 #include <xen/hypervisor.h>
72 #include <xen/xen_intr.h>
73 #include <xen/evtchn.h>
74 #include <xen/gnttab.h>
75 #include <xen/interface/memory.h>
76 #include <xen/interface/io/netif.h>
77 #include <xen/xenbus/xenbusvar.h>
78 
79 #include <dev/xen/netfront/mbufq.h>
80 
81 #include "xenbus_if.h"
82 
83 #define XN_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP | CSUM_TSO)
84 
85 #define GRANT_INVALID_REF	0
86 
87 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
88 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
89 
90 #if __FreeBSD_version >= 700000
91 /*
92  * Should the driver do LRO on the RX end
93  *  this can be toggled on the fly, but the
94  *  interface must be reset (down/up) for it
95  *  to take effect.
96  */
97 static int xn_enable_lro = 1;
98 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
99 #else
100 
101 #define IFCAP_TSO4	0
102 #define CSUM_TSO	0
103 
104 #endif
105 
106 #ifdef CONFIG_XEN
107 static int MODPARM_rx_copy = 0;
108 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0);
109 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)");
110 static int MODPARM_rx_flip = 0;
111 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0);
112 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)");
113 #else
114 static const int MODPARM_rx_copy = 1;
115 static const int MODPARM_rx_flip = 0;
116 #endif
117 
118 #define MAX_SKB_FRAGS	(65536/PAGE_SIZE + 2)
119 #define RX_COPY_THRESHOLD 256
120 
121 #define net_ratelimit() 0
122 
123 struct netfront_info;
124 struct netfront_rx_info;
125 
126 static void xn_txeof(struct netfront_info *);
127 static void xn_rxeof(struct netfront_info *);
128 static void network_alloc_rx_buffers(struct netfront_info *);
129 
130 static void xn_tick_locked(struct netfront_info *);
131 static void xn_tick(void *);
132 
133 static void xn_intr(void *);
134 static void xn_start_locked(struct ifnet *);
135 static void xn_start(struct ifnet *);
136 static int  xn_ioctl(struct ifnet *, u_long, caddr_t);
137 static void xn_ifinit_locked(struct netfront_info *);
138 static void xn_ifinit(void *);
139 static void xn_stop(struct netfront_info *);
140 #ifdef notyet
141 static void xn_watchdog(struct ifnet *);
142 #endif
143 
144 static void show_device(struct netfront_info *sc);
145 #ifdef notyet
146 static void netfront_closing(device_t dev);
147 #endif
148 static void netif_free(struct netfront_info *info);
149 static int netfront_detach(device_t dev);
150 
151 static int talk_to_backend(device_t dev, struct netfront_info *info);
152 static int create_netdev(device_t dev);
153 static void netif_disconnect_backend(struct netfront_info *info);
154 static int setup_device(device_t dev, struct netfront_info *info);
155 static void end_access(int ref, void *page);
156 
157 /* Xenolinux helper functions */
158 int network_connect(struct netfront_info *);
159 
160 static void xn_free_rx_ring(struct netfront_info *);
161 
162 static void xn_free_tx_ring(struct netfront_info *);
163 
164 static int xennet_get_responses(struct netfront_info *np,
165 	struct netfront_rx_info *rinfo, RING_IDX rp, struct mbuf **list,
166 	int *pages_flipped_p);
167 
168 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
169 
170 #define INVALID_P2M_ENTRY (~0UL)
171 
172 /*
173  * Mbuf pointers. We need these to keep track of the virtual addresses
174  * of our mbuf chains since we can only convert from virtual to physical,
175  * not the other way around.  The size must track the free index arrays.
176  */
177 struct xn_chain_data {
178 		struct mbuf		*xn_tx_chain[NET_TX_RING_SIZE+1];
179 		int			xn_tx_chain_cnt;
180 		struct mbuf		*xn_rx_chain[NET_RX_RING_SIZE+1];
181 };
182 
183 
184 struct net_device_stats
185 {
186 	u_long	rx_packets;		/* total packets received	*/
187 	u_long	tx_packets;		/* total packets transmitted	*/
188 	u_long	rx_bytes;		/* total bytes received 	*/
189 	u_long	tx_bytes;		/* total bytes transmitted	*/
190 	u_long	rx_errors;		/* bad packets received		*/
191 	u_long	tx_errors;		/* packet transmit problems	*/
192 	u_long	rx_dropped;		/* no space in linux buffers	*/
193 	u_long	tx_dropped;		/* no space available in linux	*/
194 	u_long	multicast;		/* multicast packets received	*/
195 	u_long	collisions;
196 
197 	/* detailed rx_errors: */
198 	u_long	rx_length_errors;
199 	u_long	rx_over_errors;		/* receiver ring buff overflow	*/
200 	u_long	rx_crc_errors;		/* recved pkt with crc error	*/
201 	u_long	rx_frame_errors;	/* recv'd frame alignment error */
202 	u_long	rx_fifo_errors;		/* recv'r fifo overrun		*/
203 	u_long	rx_missed_errors;	/* receiver missed packet	*/
204 
205 	/* detailed tx_errors */
206 	u_long	tx_aborted_errors;
207 	u_long	tx_carrier_errors;
208 	u_long	tx_fifo_errors;
209 	u_long	tx_heartbeat_errors;
210 	u_long	tx_window_errors;
211 
212 	/* for cslip etc */
213 	u_long	rx_compressed;
214 	u_long	tx_compressed;
215 };
216 
217 struct netfront_info {
218 
219 	struct ifnet *xn_ifp;
220 #if __FreeBSD_version >= 700000
221 	struct lro_ctrl xn_lro;
222 #endif
223 
224 	struct net_device_stats stats;
225 	u_int tx_full;
226 
227 	netif_tx_front_ring_t tx;
228 	netif_rx_front_ring_t rx;
229 
230 	struct mtx   tx_lock;
231 	struct mtx   rx_lock;
232 	struct sx    sc_lock;
233 
234 	u_int handle;
235 	u_int irq;
236 	u_int copying_receiver;
237 	u_int carrier;
238 
239 	/* Receive-ring batched refills. */
240 #define RX_MIN_TARGET 32
241 #define RX_MAX_TARGET NET_RX_RING_SIZE
242 	int rx_min_target, rx_max_target, rx_target;
243 
244 	/*
245 	 * {tx,rx}_skbs store outstanding skbuffs. The first entry in each
246 	 * array is an index into a chain of free entries.
247 	 */
248 
249 	grant_ref_t gref_tx_head;
250 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
251 	grant_ref_t gref_rx_head;
252 	grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
253 
254 #define TX_MAX_TARGET min(NET_RX_RING_SIZE, 256)
255 	device_t xbdev;
256 	int tx_ring_ref;
257 	int rx_ring_ref;
258 	uint8_t mac[ETHER_ADDR_LEN];
259 	struct xn_chain_data	xn_cdata;	/* mbufs */
260 	struct mbuf_head xn_rx_batch;	/* head of the batch queue */
261 
262 	int			xn_if_flags;
263 	struct callout	        xn_stat_ch;
264 
265 	u_long rx_pfn_array[NET_RX_RING_SIZE];
266 	multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1];
267 	mmu_update_t rx_mmu[NET_RX_RING_SIZE];
268 };
269 
270 #define rx_mbufs xn_cdata.xn_rx_chain
271 #define tx_mbufs xn_cdata.xn_tx_chain
272 
273 #define XN_LOCK_INIT(_sc, _name) \
274         mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
275         mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF);  \
276         sx_init(&(_sc)->sc_lock, #_name"_rx")
277 
278 #define XN_RX_LOCK(_sc)           mtx_lock(&(_sc)->rx_lock)
279 #define XN_RX_UNLOCK(_sc)         mtx_unlock(&(_sc)->rx_lock)
280 
281 #define XN_TX_LOCK(_sc)           mtx_lock(&(_sc)->tx_lock)
282 #define XN_TX_UNLOCK(_sc)         mtx_unlock(&(_sc)->tx_lock)
283 
284 #define XN_LOCK(_sc)           sx_xlock(&(_sc)->sc_lock);
285 #define XN_UNLOCK(_sc)         sx_xunlock(&(_sc)->sc_lock);
286 
287 #define XN_LOCK_ASSERT(_sc)    sx_assert(&(_sc)->sc_lock, SX_LOCKED);
288 #define XN_RX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->rx_lock, MA_OWNED);
289 #define XN_TX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->tx_lock, MA_OWNED);
290 #define XN_LOCK_DESTROY(_sc)   mtx_destroy(&(_sc)->rx_lock); \
291                                mtx_destroy(&(_sc)->tx_lock); \
292                                sx_destroy(&(_sc)->sc_lock);
293 
294 struct netfront_rx_info {
295 	struct netif_rx_response rx;
296 	struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
297 };
298 
299 #define netfront_carrier_on(netif)	((netif)->carrier = 1)
300 #define netfront_carrier_off(netif)	((netif)->carrier = 0)
301 #define netfront_carrier_ok(netif)	((netif)->carrier)
302 
303 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
304 
305 
306 
307 /*
308  * Access macros for acquiring freeing slots in tx_skbs[].
309  */
310 
311 static inline void
312 add_id_to_freelist(struct mbuf **list, unsigned short id)
313 {
314 	KASSERT(id != 0, ("add_id_to_freelist: the head item (0) must always be free."));
315 	list[id] = list[0];
316 	list[0]  = (void *)(u_long)id;
317 }
318 
319 static inline unsigned short
320 get_id_from_freelist(struct mbuf **list)
321 {
322 	u_int id = (u_int)(u_long)list[0];
323 	KASSERT(id != 0, ("get_id_from_freelist: the head item (0) must always remain free."));
324 	list[0] = list[id];
325 	return (id);
326 }
327 
328 static inline int
329 xennet_rxidx(RING_IDX idx)
330 {
331 	return idx & (NET_RX_RING_SIZE - 1);
332 }
333 
334 static inline struct mbuf *
335 xennet_get_rx_mbuf(struct netfront_info *np,
336 						RING_IDX ri)
337 {
338 	int i = xennet_rxidx(ri);
339 	struct mbuf *m;
340 
341 	m = np->rx_mbufs[i];
342 	np->rx_mbufs[i] = NULL;
343 	return (m);
344 }
345 
346 static inline grant_ref_t
347 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
348 {
349 	int i = xennet_rxidx(ri);
350 	grant_ref_t ref = np->grant_rx_ref[i];
351 	np->grant_rx_ref[i] = GRANT_INVALID_REF;
352 	return ref;
353 }
354 
355 #ifdef DEBUG
356 
357 #endif
358 #define IPRINTK(fmt, args...) \
359     printf("[XEN] " fmt, ##args)
360 #define WPRINTK(fmt, args...) \
361     printf("[XEN] " fmt, ##args)
362 #if 0
363 #define DPRINTK(fmt, args...) \
364     printf("[XEN] %s: " fmt, __func__, ##args)
365 #else
366 #define DPRINTK(fmt, args...)
367 #endif
368 
369 /**
370  * Read the 'mac' node at the given device's node in the store, and parse that
371  * as colon-separated octets, placing result the given mac array.  mac must be
372  * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
373  * Return 0 on success, or errno on error.
374  */
375 static int
376 xen_net_read_mac(device_t dev, uint8_t mac[])
377 {
378 	int error, i;
379 	char *s, *e, *macstr;
380 
381 	error = xenbus_read(XBT_NIL, xenbus_get_node(dev), "mac", NULL,
382 	    (void **) &macstr);
383 	if (error)
384 		return (error);
385 
386 	s = macstr;
387 	for (i = 0; i < ETHER_ADDR_LEN; i++) {
388 		mac[i] = strtoul(s, &e, 16);
389 		if (s == e || (e[0] != ':' && e[0] != 0)) {
390 			free(macstr, M_DEVBUF);
391 			return (ENOENT);
392 		}
393 		s = &e[1];
394 	}
395 	free(macstr, M_DEVBUF);
396 	return (0);
397 }
398 
399 /**
400  * Entry point to this code when a new device is created.  Allocate the basic
401  * structures and the ring buffers for communication with the backend, and
402  * inform the backend of the appropriate details for those.  Switch to
403  * Connected state.
404  */
405 static int
406 netfront_probe(device_t dev)
407 {
408 
409 	if (!strcmp(xenbus_get_type(dev), "vif")) {
410 		device_set_desc(dev, "Virtual Network Interface");
411 		return (0);
412 	}
413 
414 	return (ENXIO);
415 }
416 
417 static int
418 netfront_attach(device_t dev)
419 {
420 	int err;
421 
422 	err = create_netdev(dev);
423 	if (err) {
424 		xenbus_dev_fatal(dev, err, "creating netdev");
425 		return err;
426 	}
427 
428 #if __FreeBSD_version >= 700000
429 	SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
430 	    SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
431 	    OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
432 	    &xn_enable_lro, 0, "Large Receive Offload");
433 #endif
434 
435 	return 0;
436 }
437 
438 
439 /**
440  * We are reconnecting to the backend, due to a suspend/resume, or a backend
441  * driver restart.  We tear down our netif structure and recreate it, but
442  * leave the device-layer structures intact so that this is transparent to the
443  * rest of the kernel.
444  */
445 static int
446 netfront_resume(device_t dev)
447 {
448 	struct netfront_info *info = device_get_softc(dev);
449 
450 	netif_disconnect_backend(info);
451 	return (0);
452 }
453 
454 
455 /* Common code used when first setting up, and when resuming. */
456 static int
457 talk_to_backend(device_t dev, struct netfront_info *info)
458 {
459 	const char *message;
460 	struct xenbus_transaction xbt;
461 	const char *node = xenbus_get_node(dev);
462 	int err;
463 
464 	err = xen_net_read_mac(dev, info->mac);
465 	if (err) {
466 		xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
467 		goto out;
468 	}
469 
470 	/* Create shared ring, alloc event channel. */
471 	err = setup_device(dev, info);
472 	if (err)
473 		goto out;
474 
475  again:
476 	err = xenbus_transaction_start(&xbt);
477 	if (err) {
478 		xenbus_dev_fatal(dev, err, "starting transaction");
479 		goto destroy_ring;
480 	}
481 	err = xenbus_printf(xbt, node, "tx-ring-ref","%u",
482 			    info->tx_ring_ref);
483 	if (err) {
484 		message = "writing tx ring-ref";
485 		goto abort_transaction;
486 	}
487 	err = xenbus_printf(xbt, node, "rx-ring-ref","%u",
488 			    info->rx_ring_ref);
489 	if (err) {
490 		message = "writing rx ring-ref";
491 		goto abort_transaction;
492 	}
493 	err = xenbus_printf(xbt, node,
494 		"event-channel", "%u", irq_to_evtchn_port(info->irq));
495 	if (err) {
496 		message = "writing event-channel";
497 		goto abort_transaction;
498 	}
499 	err = xenbus_printf(xbt, node, "request-rx-copy", "%u",
500 			    info->copying_receiver);
501 	if (err) {
502 		message = "writing request-rx-copy";
503 		goto abort_transaction;
504 	}
505 	err = xenbus_printf(xbt, node, "feature-rx-notify", "%d", 1);
506 	if (err) {
507 		message = "writing feature-rx-notify";
508 		goto abort_transaction;
509 	}
510 	err = xenbus_printf(xbt, node, "feature-sg", "%d", 1);
511 	if (err) {
512 		message = "writing feature-sg";
513 		goto abort_transaction;
514 	}
515 #if __FreeBSD_version >= 700000
516 	err = xenbus_printf(xbt, node, "feature-gso-tcpv4", "%d", 1);
517 	if (err) {
518 		message = "writing feature-gso-tcpv4";
519 		goto abort_transaction;
520 	}
521 #endif
522 
523 	err = xenbus_transaction_end(xbt, 0);
524 	if (err) {
525 		if (err == EAGAIN)
526 			goto again;
527 		xenbus_dev_fatal(dev, err, "completing transaction");
528 		goto destroy_ring;
529 	}
530 
531 	return 0;
532 
533  abort_transaction:
534 	xenbus_transaction_end(xbt, 1);
535 	xenbus_dev_fatal(dev, err, "%s", message);
536  destroy_ring:
537 	netif_free(info);
538  out:
539 	return err;
540 }
541 
542 
543 static int
544 setup_device(device_t dev, struct netfront_info *info)
545 {
546 	netif_tx_sring_t *txs;
547 	netif_rx_sring_t *rxs;
548 	int error;
549 	struct ifnet *ifp;
550 
551 	ifp = info->xn_ifp;
552 
553 	info->tx_ring_ref = GRANT_INVALID_REF;
554 	info->rx_ring_ref = GRANT_INVALID_REF;
555 	info->rx.sring = NULL;
556 	info->tx.sring = NULL;
557 	info->irq = 0;
558 
559 	txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
560 	if (!txs) {
561 		error = ENOMEM;
562 		xenbus_dev_fatal(dev, error, "allocating tx ring page");
563 		goto fail;
564 	}
565 	SHARED_RING_INIT(txs);
566 	FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
567 	error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
568 	if (error)
569 		goto fail;
570 
571 	rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
572 	if (!rxs) {
573 		error = ENOMEM;
574 		xenbus_dev_fatal(dev, error, "allocating rx ring page");
575 		goto fail;
576 	}
577 	SHARED_RING_INIT(rxs);
578 	FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
579 
580 	error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
581 	if (error)
582 		goto fail;
583 
584 	error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
585 	    "xn", xn_intr, info, INTR_TYPE_NET | INTR_MPSAFE, &info->irq);
586 
587 	if (error) {
588 		xenbus_dev_fatal(dev, error,
589 				 "bind_evtchn_to_irqhandler failed");
590 		goto fail;
591 	}
592 
593 	show_device(info);
594 
595 	return (0);
596 
597  fail:
598 	netif_free(info);
599 	return (error);
600 }
601 
602 /**
603  * If this interface has an ipv4 address, send an arp for it. This
604  * helps to get the network going again after migrating hosts.
605  */
606 static void
607 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
608 {
609 	struct ifnet *ifp;
610 	struct ifaddr *ifa;
611 
612 	ifp = info->xn_ifp;
613 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
614 		if (ifa->ifa_addr->sa_family == AF_INET) {
615 			arp_ifinit(ifp, ifa);
616 		}
617 	}
618 }
619 
620 /**
621  * Callback received when the backend's state changes.
622  */
623 static void
624 netfront_backend_changed(device_t dev, XenbusState newstate)
625 {
626 	struct netfront_info *sc = device_get_softc(dev);
627 
628 	DPRINTK("newstate=%d\n", newstate);
629 
630 	switch (newstate) {
631 	case XenbusStateInitialising:
632 	case XenbusStateInitialised:
633 	case XenbusStateConnected:
634 	case XenbusStateUnknown:
635 	case XenbusStateClosed:
636 	case XenbusStateReconfigured:
637 	case XenbusStateReconfiguring:
638 		break;
639 	case XenbusStateInitWait:
640 		if (xenbus_get_state(dev) != XenbusStateInitialising)
641 			break;
642 		if (network_connect(sc) != 0)
643 			break;
644 		xenbus_set_state(dev, XenbusStateConnected);
645 		netfront_send_fake_arp(dev, sc);
646 		break;
647 	case XenbusStateClosing:
648 		xenbus_set_state(dev, XenbusStateClosed);
649 		break;
650 	}
651 }
652 
653 static void
654 xn_free_rx_ring(struct netfront_info *sc)
655 {
656 #if 0
657 	int i;
658 
659 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
660 		if (sc->xn_cdata.xn_rx_chain[i] != NULL) {
661 			m_freem(sc->xn_cdata.xn_rx_chain[i]);
662 			sc->xn_cdata.xn_rx_chain[i] = NULL;
663 		}
664 	}
665 
666 	sc->rx.rsp_cons = 0;
667 	sc->xn_rx_if->req_prod = 0;
668 	sc->xn_rx_if->event = sc->rx.rsp_cons ;
669 #endif
670 }
671 
672 static void
673 xn_free_tx_ring(struct netfront_info *sc)
674 {
675 #if 0
676 	int i;
677 
678 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
679 		if (sc->xn_cdata.xn_tx_chain[i] != NULL) {
680 			m_freem(sc->xn_cdata.xn_tx_chain[i]);
681 			sc->xn_cdata.xn_tx_chain[i] = NULL;
682 		}
683 	}
684 
685 	return;
686 #endif
687 }
688 
689 /*
690  * Do some brief math on the number of descriptors available to
691  * determine how many slots are available.
692  *
693  * Firstly - wouldn't something with RING_FREE_REQUESTS() be more applicable?
694  * Secondly - MAX_SKB_FRAGS is a Linux construct which may not apply here.
695  * Thirdly - it isn't used here anyway; the magic constant '24' is possibly
696  *   wrong?
697  * The "2" is presumably to ensure there are also enough slots available for
698  * the ring entries used for "options" (eg, the TSO entry before a packet
699  * is queued); I'm not sure why its 2 and not 1. Perhaps to make sure there's
700  * a "free" node in the tx mbuf list (node 0) to represent the freelist?
701  *
702  * This only figures out whether any xenbus ring descriptors are available;
703  * it doesn't at all reflect how many tx mbuf ring descriptors are also
704  * available.
705  */
706 static inline int
707 netfront_tx_slot_available(struct netfront_info *np)
708 {
709 	return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
710 		(TX_MAX_TARGET - /* MAX_SKB_FRAGS */ 24 - 2));
711 }
712 static void
713 netif_release_tx_bufs(struct netfront_info *np)
714 {
715 	struct mbuf *m;
716 	int i;
717 
718 	for (i = 1; i <= NET_TX_RING_SIZE; i++) {
719 		m = np->xn_cdata.xn_tx_chain[i];
720 
721 		if (((u_long)m) < KERNBASE)
722 			continue;
723 		gnttab_grant_foreign_access_ref(np->grant_tx_ref[i],
724 		    xenbus_get_otherend_id(np->xbdev),
725 		    virt_to_mfn(mtod(m, vm_offset_t)),
726 		    GNTMAP_readonly);
727 		gnttab_release_grant_reference(&np->gref_tx_head,
728 		    np->grant_tx_ref[i]);
729 		np->grant_tx_ref[i] = GRANT_INVALID_REF;
730 		add_id_to_freelist(np->tx_mbufs, i);
731 		np->xn_cdata.xn_tx_chain_cnt--;
732 		if (np->xn_cdata.xn_tx_chain_cnt < 0) {
733 			panic("netif_release_tx_bufs: tx_chain_cnt must be >= 0");
734 		}
735 		m_freem(m);
736 	}
737 }
738 
739 static void
740 network_alloc_rx_buffers(struct netfront_info *sc)
741 {
742 	int otherend_id = xenbus_get_otherend_id(sc->xbdev);
743 	unsigned short id;
744 	struct mbuf *m_new;
745 	int i, batch_target, notify;
746 	RING_IDX req_prod;
747 	struct xen_memory_reservation reservation;
748 	grant_ref_t ref;
749 	int nr_flips;
750 	netif_rx_request_t *req;
751 	vm_offset_t vaddr;
752 	u_long pfn;
753 
754 	req_prod = sc->rx.req_prod_pvt;
755 
756 	if (unlikely(sc->carrier == 0))
757 		return;
758 
759 	/*
760 	 * Allocate skbuffs greedily, even though we batch updates to the
761 	 * receive ring. This creates a less bursty demand on the memory
762 	 * allocator, so should reduce the chance of failed allocation
763 	 * requests both for ourself and for other kernel subsystems.
764 	 */
765 	batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
766 	for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
767 		MGETHDR(m_new, M_DONTWAIT, MT_DATA);
768 		if (m_new == NULL)
769 			goto no_mbuf;
770 
771 		m_cljget(m_new, M_DONTWAIT, MJUMPAGESIZE);
772 		if ((m_new->m_flags & M_EXT) == 0) {
773 			m_freem(m_new);
774 
775 no_mbuf:
776 			if (i != 0)
777 				goto refill;
778 			/*
779 			 * XXX set timer
780 			 */
781 			break;
782 		}
783 		m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
784 
785 		/* queue the mbufs allocated */
786 		mbufq_tail(&sc->xn_rx_batch, m_new);
787 	}
788 
789 	/* Is the batch large enough to be worthwhile? */
790 	if (i < (sc->rx_target/2)) {
791 		if (req_prod >sc->rx.sring->req_prod)
792 			goto push;
793 		return;
794 	}
795 	/* Adjust floating fill target if we risked running out of buffers. */
796 	if ( ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) &&
797 	     ((sc->rx_target *= 2) > sc->rx_max_target) )
798 		sc->rx_target = sc->rx_max_target;
799 
800 refill:
801 	for (nr_flips = i = 0; ; i++) {
802 		if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
803 			break;
804 
805 		m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
806 				vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
807 
808 		id = xennet_rxidx(req_prod + i);
809 
810 		KASSERT(sc->xn_cdata.xn_rx_chain[id] == NULL,
811 		    ("non-NULL xm_rx_chain"));
812 		sc->xn_cdata.xn_rx_chain[id] = m_new;
813 
814 		ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
815 		KASSERT((short)ref >= 0, ("negative ref"));
816 		sc->grant_rx_ref[id] = ref;
817 
818 		vaddr = mtod(m_new, vm_offset_t);
819 		pfn = vtophys(vaddr) >> PAGE_SHIFT;
820 		req = RING_GET_REQUEST(&sc->rx, req_prod + i);
821 
822 		if (sc->copying_receiver == 0) {
823 			gnttab_grant_foreign_transfer_ref(ref,
824 			    otherend_id, pfn);
825 			sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
826 			if (!xen_feature(XENFEAT_auto_translated_physmap)) {
827 				/* Remove this page before passing
828 				 * back to Xen.
829 				 */
830 				set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
831 				MULTI_update_va_mapping(&sc->rx_mcl[i],
832 				    vaddr, 0, 0);
833 			}
834 			nr_flips++;
835 		} else {
836 			gnttab_grant_foreign_access_ref(ref,
837 			    otherend_id,
838 			    PFNTOMFN(pfn), 0);
839 		}
840 		req->id = id;
841 		req->gref = ref;
842 
843 		sc->rx_pfn_array[i] =
844 		    vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
845 	}
846 
847 	KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
848 	KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
849 	/*
850 	 * We may have allocated buffers which have entries outstanding
851 	 * in the page * update queue -- make sure we flush those first!
852 	 */
853 	PT_UPDATES_FLUSH();
854 	if (nr_flips != 0) {
855 #ifdef notyet
856 		/* Tell the ballon driver what is going on. */
857 		balloon_update_driver_allowance(i);
858 #endif
859 		set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
860 		reservation.nr_extents   = i;
861 		reservation.extent_order = 0;
862 		reservation.address_bits = 0;
863 		reservation.domid        = DOMID_SELF;
864 
865 		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
866 
867 			/* After all PTEs have been zapped, flush the TLB. */
868 			sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
869 			    UVMF_TLB_FLUSH|UVMF_ALL;
870 
871 			/* Give away a batch of pages. */
872 			sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
873 			sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
874 			sc->rx_mcl[i].args[1] =  (u_long)&reservation;
875 			/* Zap PTEs and give away pages in one big multicall. */
876 			(void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
877 
878 			/* Check return status of HYPERVISOR_dom_mem_op(). */
879 			if (unlikely(sc->rx_mcl[i].result != i))
880 				panic("Unable to reduce memory reservation\n");
881 			} else {
882 				if (HYPERVISOR_memory_op(
883 				    XENMEM_decrease_reservation, &reservation)
884 				    != i)
885 					panic("Unable to reduce memory "
886 					    "reservation\n");
887 		}
888 	} else {
889 		wmb();
890 	}
891 
892 	/* Above is a suitable barrier to ensure backend will see requests. */
893 	sc->rx.req_prod_pvt = req_prod + i;
894 push:
895 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
896 	if (notify)
897 		notify_remote_via_irq(sc->irq);
898 }
899 
900 static void
901 xn_rxeof(struct netfront_info *np)
902 {
903 	struct ifnet *ifp;
904 #if __FreeBSD_version >= 700000
905 	struct lro_ctrl *lro = &np->xn_lro;
906 	struct lro_entry *queued;
907 #endif
908 	struct netfront_rx_info rinfo;
909 	struct netif_rx_response *rx = &rinfo.rx;
910 	struct netif_extra_info *extras = rinfo.extras;
911 	RING_IDX i, rp;
912 	multicall_entry_t *mcl;
913 	struct mbuf *m;
914 	struct mbuf_head rxq, errq;
915 	int err, pages_flipped = 0, work_to_do;
916 
917 	do {
918 		XN_RX_LOCK_ASSERT(np);
919 		if (!netfront_carrier_ok(np))
920 			return;
921 
922 		mbufq_init(&errq);
923 		mbufq_init(&rxq);
924 
925 		ifp = np->xn_ifp;
926 
927 		rp = np->rx.sring->rsp_prod;
928 		rmb();	/* Ensure we see queued responses up to 'rp'. */
929 
930 		i = np->rx.rsp_cons;
931 		while ((i != rp)) {
932 			memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
933 			memset(extras, 0, sizeof(rinfo.extras));
934 
935 			m = NULL;
936 			err = xennet_get_responses(np, &rinfo, rp, &m,
937 			    &pages_flipped);
938 
939 			if (unlikely(err)) {
940 				if (m)
941 					mbufq_tail(&errq, m);
942 				np->stats.rx_errors++;
943 				i = np->rx.rsp_cons;
944 				continue;
945 			}
946 
947 			m->m_pkthdr.rcvif = ifp;
948 			if ( rx->flags & NETRXF_data_validated ) {
949 				/* Tell the stack the checksums are okay */
950 				/*
951 				 * XXX this isn't necessarily the case - need to add
952 				 * check
953 				 */
954 
955 				m->m_pkthdr.csum_flags |=
956 					(CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
957 					    | CSUM_PSEUDO_HDR);
958 				m->m_pkthdr.csum_data = 0xffff;
959 			}
960 
961 			np->stats.rx_packets++;
962 			np->stats.rx_bytes += m->m_pkthdr.len;
963 
964 			mbufq_tail(&rxq, m);
965 			np->rx.rsp_cons = ++i;
966 		}
967 
968 		if (pages_flipped) {
969 			/* Some pages are no longer absent... */
970 #ifdef notyet
971 			balloon_update_driver_allowance(-pages_flipped);
972 #endif
973 			/* Do all the remapping work, and M->P updates, in one big
974 			 * hypercall.
975 			 */
976 			if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
977 				mcl = np->rx_mcl + pages_flipped;
978 				mcl->op = __HYPERVISOR_mmu_update;
979 				mcl->args[0] = (u_long)np->rx_mmu;
980 				mcl->args[1] = pages_flipped;
981 				mcl->args[2] = 0;
982 				mcl->args[3] = DOMID_SELF;
983 				(void)HYPERVISOR_multicall(np->rx_mcl,
984 				    pages_flipped + 1);
985 			}
986 		}
987 
988 		while ((m = mbufq_dequeue(&errq)))
989 			m_freem(m);
990 
991 		/*
992 		 * Process all the mbufs after the remapping is complete.
993 		 * Break the mbuf chain first though.
994 		 */
995 		while ((m = mbufq_dequeue(&rxq)) != NULL) {
996 			ifp->if_ipackets++;
997 
998 			/*
999 			 * Do we really need to drop the rx lock?
1000 			 */
1001 			XN_RX_UNLOCK(np);
1002 #if __FreeBSD_version >= 700000
1003 			/* Use LRO if possible */
1004 			if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1005 			    lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1006 				/*
1007 				 * If LRO fails, pass up to the stack
1008 				 * directly.
1009 				 */
1010 				(*ifp->if_input)(ifp, m);
1011 			}
1012 #else
1013 			(*ifp->if_input)(ifp, m);
1014 #endif
1015 			XN_RX_LOCK(np);
1016 		}
1017 
1018 		np->rx.rsp_cons = i;
1019 
1020 #if __FreeBSD_version >= 700000
1021 		/*
1022 		 * Flush any outstanding LRO work
1023 		 */
1024 		while (!SLIST_EMPTY(&lro->lro_active)) {
1025 			queued = SLIST_FIRST(&lro->lro_active);
1026 			SLIST_REMOVE_HEAD(&lro->lro_active, next);
1027 			tcp_lro_flush(lro, queued);
1028 		}
1029 #endif
1030 
1031 #if 0
1032 		/* If we get a callback with very few responses, reduce fill target. */
1033 		/* NB. Note exponential increase, linear decrease. */
1034 		if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1035 			((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1036 			np->rx_target = np->rx_min_target;
1037 #endif
1038 
1039 		network_alloc_rx_buffers(np);
1040 
1041 		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1042 	} while (work_to_do);
1043 }
1044 
1045 static void
1046 xn_txeof(struct netfront_info *np)
1047 {
1048 	RING_IDX i, prod;
1049 	unsigned short id;
1050 	struct ifnet *ifp;
1051 	netif_tx_response_t *txr;
1052 	struct mbuf *m;
1053 
1054 	XN_TX_LOCK_ASSERT(np);
1055 
1056 	if (!netfront_carrier_ok(np))
1057 		return;
1058 
1059 	ifp = np->xn_ifp;
1060 	ifp->if_timer = 0;
1061 
1062 	do {
1063 		prod = np->tx.sring->rsp_prod;
1064 		rmb(); /* Ensure we see responses up to 'rp'. */
1065 
1066 		for (i = np->tx.rsp_cons; i != prod; i++) {
1067 			txr = RING_GET_RESPONSE(&np->tx, i);
1068 			if (txr->status == NETIF_RSP_NULL)
1069 				continue;
1070 
1071 			id = txr->id;
1072 			m = np->xn_cdata.xn_tx_chain[id];
1073 			KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1074 			M_ASSERTVALID(m);
1075 
1076 			/*
1077 			 * Increment packet count if this is the last
1078 			 * mbuf of the chain.
1079 			 */
1080 			if (!m->m_next)
1081 				ifp->if_opackets++;
1082 			if (unlikely(gnttab_query_foreign_access(
1083 			    np->grant_tx_ref[id]) != 0)) {
1084 				printf("network_tx_buf_gc: warning "
1085 				    "-- grant still in use by backend "
1086 				    "domain.\n");
1087 				goto out;
1088 			}
1089 			gnttab_end_foreign_access_ref(
1090 				np->grant_tx_ref[id]);
1091 			gnttab_release_grant_reference(
1092 				&np->gref_tx_head, np->grant_tx_ref[id]);
1093 			np->grant_tx_ref[id] = GRANT_INVALID_REF;
1094 
1095 			np->xn_cdata.xn_tx_chain[id] = NULL;
1096 			add_id_to_freelist(np->xn_cdata.xn_tx_chain, id);
1097 			np->xn_cdata.xn_tx_chain_cnt--;
1098 			if (np->xn_cdata.xn_tx_chain_cnt < 0) {
1099 				panic("netif_release_tx_bufs: tx_chain_cnt must be >= 0");
1100 			}
1101 			m_free(m);
1102 			/* Only mark the queue active if we've freed up at least one slot to try */
1103 			ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1104 		}
1105 		np->tx.rsp_cons = prod;
1106 
1107 		/*
1108 		 * Set a new event, then check for race with update of
1109 		 * tx_cons. Note that it is essential to schedule a
1110 		 * callback, no matter how few buffers are pending. Even if
1111 		 * there is space in the transmit ring, higher layers may
1112 		 * be blocked because too much data is outstanding: in such
1113 		 * cases notification from Xen is likely to be the only kick
1114 		 * that we'll get.
1115 		 */
1116 		np->tx.sring->rsp_event =
1117 		    prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1118 
1119 		mb();
1120 	} while (prod != np->tx.sring->rsp_prod);
1121 
1122  out:
1123 	if (np->tx_full &&
1124 	    ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1125 		np->tx_full = 0;
1126 #if 0
1127 		if (np->user_state == UST_OPEN)
1128 			netif_wake_queue(dev);
1129 #endif
1130 	}
1131 
1132 }
1133 
1134 static void
1135 xn_intr(void *xsc)
1136 {
1137 	struct netfront_info *np = xsc;
1138 	struct ifnet *ifp = np->xn_ifp;
1139 
1140 #if 0
1141 	if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1142 	    likely(netfront_carrier_ok(np)) &&
1143 	    ifp->if_drv_flags & IFF_DRV_RUNNING))
1144 		return;
1145 #endif
1146 	if (np->tx.rsp_cons != np->tx.sring->rsp_prod) {
1147 		XN_TX_LOCK(np);
1148 		xn_txeof(np);
1149 		XN_TX_UNLOCK(np);
1150 	}
1151 
1152 	XN_RX_LOCK(np);
1153 	xn_rxeof(np);
1154 	XN_RX_UNLOCK(np);
1155 
1156 	if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1157 	    !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1158 		xn_start(ifp);
1159 }
1160 
1161 
1162 static void
1163 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1164 	grant_ref_t ref)
1165 {
1166 	int new = xennet_rxidx(np->rx.req_prod_pvt);
1167 
1168 	KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1169 	np->rx_mbufs[new] = m;
1170 	np->grant_rx_ref[new] = ref;
1171 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1172 	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1173 	np->rx.req_prod_pvt++;
1174 }
1175 
1176 static int
1177 xennet_get_extras(struct netfront_info *np,
1178     struct netif_extra_info *extras, RING_IDX rp)
1179 {
1180 	struct netif_extra_info *extra;
1181 	RING_IDX cons = np->rx.rsp_cons;
1182 
1183 	int err = 0;
1184 
1185 	do {
1186 		struct mbuf *m;
1187 		grant_ref_t ref;
1188 
1189 		if (unlikely(cons + 1 == rp)) {
1190 #if 0
1191 			if (net_ratelimit())
1192 				WPRINTK("Missing extra info\n");
1193 #endif
1194 			err = -EINVAL;
1195 			break;
1196 		}
1197 
1198 		extra = (struct netif_extra_info *)
1199 		RING_GET_RESPONSE(&np->rx, ++cons);
1200 
1201 		if (unlikely(!extra->type ||
1202 			extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1203 #if 0
1204 			if (net_ratelimit())
1205 				WPRINTK("Invalid extra type: %d\n",
1206 					extra->type);
1207 #endif
1208 			err = -EINVAL;
1209 		} else {
1210 			memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1211 		}
1212 
1213 		m = xennet_get_rx_mbuf(np, cons);
1214 		ref = xennet_get_rx_ref(np, cons);
1215 		xennet_move_rx_slot(np, m, ref);
1216 	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1217 
1218 	np->rx.rsp_cons = cons;
1219 	return err;
1220 }
1221 
1222 static int
1223 xennet_get_responses(struct netfront_info *np,
1224 	struct netfront_rx_info *rinfo, RING_IDX rp,
1225 	struct mbuf  **list,
1226 	int *pages_flipped_p)
1227 {
1228 	int pages_flipped = *pages_flipped_p;
1229 	struct mmu_update *mmu;
1230 	struct multicall_entry *mcl;
1231 	struct netif_rx_response *rx = &rinfo->rx;
1232 	struct netif_extra_info *extras = rinfo->extras;
1233 	RING_IDX cons = np->rx.rsp_cons;
1234 	struct mbuf *m, *m0, *m_prev;
1235 	grant_ref_t ref = xennet_get_rx_ref(np, cons);
1236 	int max = 5 /* MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD) */;
1237 	int frags = 1;
1238 	int err = 0;
1239 	u_long ret;
1240 
1241 	m0 = m = m_prev = xennet_get_rx_mbuf(np, cons);
1242 
1243 
1244 	if (rx->flags & NETRXF_extra_info) {
1245 		err = xennet_get_extras(np, extras, rp);
1246 		cons = np->rx.rsp_cons;
1247 	}
1248 
1249 
1250 	if (m0 != NULL) {
1251 			m0->m_pkthdr.len = 0;
1252 			m0->m_next = NULL;
1253 	}
1254 
1255 	for (;;) {
1256 		u_long mfn;
1257 
1258 #if 0
1259 		printf("rx->status=%hd rx->offset=%hu frags=%u\n",
1260 			rx->status, rx->offset, frags);
1261 #endif
1262 		if (unlikely(rx->status < 0 ||
1263 			rx->offset + rx->status > PAGE_SIZE)) {
1264 #if 0
1265 			if (net_ratelimit())
1266 				WPRINTK("rx->offset: %x, size: %u\n",
1267 					rx->offset, rx->status);
1268 #endif
1269 			xennet_move_rx_slot(np, m, ref);
1270 			err = -EINVAL;
1271 			goto next;
1272 		}
1273 
1274 		/*
1275 		 * This definitely indicates a bug, either in this driver or in
1276 		 * the backend driver. In future this should flag the bad
1277 		 * situation to the system controller to reboot the backed.
1278 		 */
1279 		if (ref == GRANT_INVALID_REF) {
1280 #if 0
1281 			if (net_ratelimit())
1282 				WPRINTK("Bad rx response id %d.\n", rx->id);
1283 #endif
1284 			err = -EINVAL;
1285 			goto next;
1286 		}
1287 
1288 		if (!np->copying_receiver) {
1289 			/* Memory pressure, insufficient buffer
1290 			 * headroom, ...
1291 			 */
1292 			if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1293 				if (net_ratelimit())
1294 					WPRINTK("Unfulfilled rx req "
1295 						"(id=%d, st=%d).\n",
1296 						rx->id, rx->status);
1297 				xennet_move_rx_slot(np, m, ref);
1298 				err = -ENOMEM;
1299 				goto next;
1300 			}
1301 
1302 			if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1303 				/* Remap the page. */
1304 				void *vaddr = mtod(m, void *);
1305 				uint32_t pfn;
1306 
1307 				mcl = np->rx_mcl + pages_flipped;
1308 				mmu = np->rx_mmu + pages_flipped;
1309 
1310 				MULTI_update_va_mapping(mcl, (u_long)vaddr,
1311 				    (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1312 				    PG_V | PG_M | PG_A, 0);
1313 				pfn = (uintptr_t)m->m_ext.ext_arg1;
1314 				mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1315 				    MMU_MACHPHYS_UPDATE;
1316 				mmu->val = pfn;
1317 
1318 				set_phys_to_machine(pfn, mfn);
1319 			}
1320 			pages_flipped++;
1321 		} else {
1322 			ret = gnttab_end_foreign_access_ref(ref);
1323 			KASSERT(ret, ("ret != 0"));
1324 		}
1325 
1326 		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1327 
1328 next:
1329 		if (m == NULL)
1330 			break;
1331 
1332 		m->m_len = rx->status;
1333 		m->m_data += rx->offset;
1334 		m0->m_pkthdr.len += rx->status;
1335 
1336 		if (!(rx->flags & NETRXF_more_data))
1337 			break;
1338 
1339 		if (cons + frags == rp) {
1340 			if (net_ratelimit())
1341 				WPRINTK("Need more frags\n");
1342 			err = -ENOENT;
1343 				break;
1344 		}
1345 		m_prev = m;
1346 
1347 		rx = RING_GET_RESPONSE(&np->rx, cons + frags);
1348 		m = xennet_get_rx_mbuf(np, cons + frags);
1349 
1350 		m_prev->m_next = m;
1351 		m->m_next = NULL;
1352 		ref = xennet_get_rx_ref(np, cons + frags);
1353 		frags++;
1354 	}
1355 	*list = m0;
1356 
1357 	if (unlikely(frags > max)) {
1358 		if (net_ratelimit())
1359 			WPRINTK("Too many frags\n");
1360 		err = -E2BIG;
1361 	}
1362 
1363 	if (unlikely(err))
1364 		np->rx.rsp_cons = cons + frags;
1365 
1366 	*pages_flipped_p = pages_flipped;
1367 
1368 	return err;
1369 }
1370 
1371 static void
1372 xn_tick_locked(struct netfront_info *sc)
1373 {
1374 	XN_RX_LOCK_ASSERT(sc);
1375 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1376 
1377 	/* XXX placeholder for printing debug information */
1378 
1379 }
1380 
1381 
1382 static void
1383 xn_tick(void *xsc)
1384 {
1385 	struct netfront_info *sc;
1386 
1387 	sc = xsc;
1388 	XN_RX_LOCK(sc);
1389 	xn_tick_locked(sc);
1390 	XN_RX_UNLOCK(sc);
1391 
1392 }
1393 static void
1394 xn_start_locked(struct ifnet *ifp)
1395 {
1396 	int otherend_id;
1397 	unsigned short id;
1398 	struct mbuf *m_head, *m;
1399 	struct netfront_info *sc;
1400 	netif_tx_request_t *tx;
1401 	netif_extra_info_t *extra;
1402 	RING_IDX i;
1403 	grant_ref_t ref;
1404 	u_long mfn, tx_bytes;
1405 	int notify, nfrags;
1406 
1407 	sc = ifp->if_softc;
1408 	otherend_id = xenbus_get_otherend_id(sc->xbdev);
1409 	tx_bytes = 0;
1410 
1411 	if (!netfront_carrier_ok(sc))
1412 		return;
1413 
1414 	for (i = sc->tx.req_prod_pvt; TRUE; i++) {
1415 		IF_DEQUEUE(&ifp->if_snd, m_head);
1416 		if (m_head == NULL)
1417 			break;
1418 
1419 		/*
1420 		 * netfront_tx_slot_available() tries to do some math to
1421 		 * ensure that there'll be enough xenbus ring slots available
1422 		 * for the maximum number of packet fragments (and a couple more
1423 		 * for what I guess are TSO and other ring entry items.)
1424 		 */
1425 		if (!netfront_tx_slot_available(sc)) {
1426 			IF_PREPEND(&ifp->if_snd, m_head);
1427 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1428 			break;
1429 		}
1430 
1431 		/*
1432 		 * Defragment the mbuf if necessary.
1433 		 */
1434 		for (m = m_head, nfrags = 0; m; m = m->m_next)
1435 			nfrags++;
1436 		if (nfrags > MAX_SKB_FRAGS) {
1437 			m = m_defrag(m_head, M_DONTWAIT);
1438 			if (!m) {
1439 				m_freem(m_head);
1440 				break;
1441 			}
1442 			m_head = m;
1443 		}
1444 
1445 		/* Determine how many fragments now exist */
1446 		for (m = m_head, nfrags = 0; m; m = m->m_next)
1447 			nfrags++;
1448 
1449 		/*
1450 		 * Don't attempt to queue this packet if there aren't
1451 		 * enough free entries in the chain.
1452 		 *
1453 		 * There isn't a 1:1 correspondance between the mbuf TX ring
1454 		 * and the xenbus TX ring.
1455 		 * xn_txeof() may need to be called to free up some slots.
1456 		 *
1457 		 * It is quite possible that this can be later eliminated if
1458 		 * it turns out that partial * packets can be pushed into
1459 		 * the ringbuffer, with fragments pushed in when further slots
1460 		 * free up.
1461 		 *
1462 		 * It is also quite possible that the driver will lock up
1463 		 * if the TX queue fills up with no RX traffic, and
1464 		 * the mbuf ring is exhausted. The queue may need
1465 		 * a swift kick to continue.
1466 		 */
1467 
1468 		/*
1469 		 * It is not +1 like the allocation because we need to keep
1470 		 * slot [0] free for the freelist head
1471 		 */
1472 		if (sc->xn_cdata.xn_tx_chain_cnt + nfrags >= NET_TX_RING_SIZE) {
1473 			printf("xn_start_locked: xn_tx_chain_cnt (%d) + nfrags %d >= NET_TX_RING_SIZE (%d); must be full!\n",
1474 			    (int) sc->xn_cdata.xn_tx_chain_cnt,
1475 			    (int) nfrags, (int) NET_TX_RING_SIZE);
1476 			IF_PREPEND(&ifp->if_snd, m_head);
1477 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1478 			break;
1479 		}
1480 
1481 		/*
1482 		 * Make sure there's actually space available in the
1483 		 * Xen TX ring for this. Overcompensate for the possibility
1484 		 * of having a TCP offload fragment just in case for now
1485 		 * (the +1) rather than adding logic to accurately calculate
1486 		 * the required size.
1487 		 */
1488 		if (RING_FREE_REQUESTS(&sc->tx) < (nfrags + 1)) {
1489 			printf("xn_start_locked: free ring slots (%d) < (nfrags + 1) (%d); must be full!\n",
1490 			    (int) RING_FREE_REQUESTS(&sc->tx),
1491 			    (int) (nfrags + 1));
1492 			IF_PREPEND(&ifp->if_snd, m_head);
1493 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1494 			break;
1495 		}
1496 
1497 		/*
1498 		 * Start packing the mbufs in this chain into
1499 		 * the fragment pointers. Stop when we run out
1500 		 * of fragments or hit the end of the mbuf chain.
1501 		 */
1502 		m = m_head;
1503 		extra = NULL;
1504 		for (m = m_head; m; m = m->m_next) {
1505 			tx = RING_GET_REQUEST(&sc->tx, i);
1506 			id = get_id_from_freelist(sc->xn_cdata.xn_tx_chain);
1507 			if (id == 0)
1508 				panic("xn_start_locked: was allocated the freelist head!\n");
1509 			sc->xn_cdata.xn_tx_chain_cnt++;
1510 			if (sc->xn_cdata.xn_tx_chain_cnt >= NET_TX_RING_SIZE+1)
1511 				panic("xn_start_locked: tx_chain_cnt must be < NET_TX_RING_SIZE+1\n");
1512 			sc->xn_cdata.xn_tx_chain[id] = m;
1513 			tx->id = id;
1514 			ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1515 			KASSERT((short)ref >= 0, ("Negative ref"));
1516 			mfn = virt_to_mfn(mtod(m, vm_offset_t));
1517 			gnttab_grant_foreign_access_ref(ref, otherend_id,
1518 			    mfn, GNTMAP_readonly);
1519 			tx->gref = sc->grant_tx_ref[id] = ref;
1520 			tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1521 			tx->flags = 0;
1522 			if (m == m_head) {
1523 				/*
1524 				 * The first fragment has the entire packet
1525 				 * size, subsequent fragments have just the
1526 				 * fragment size. The backend works out the
1527 				 * true size of the first fragment by
1528 				 * subtracting the sizes of the other
1529 				 * fragments.
1530 				 */
1531 				tx->size = m->m_pkthdr.len;
1532 
1533 				/*
1534 				 * The first fragment contains the
1535 				 * checksum flags and is optionally
1536 				 * followed by extra data for TSO etc.
1537 				 */
1538 				if (m->m_pkthdr.csum_flags
1539 				    & CSUM_DELAY_DATA) {
1540 					tx->flags |= (NETTXF_csum_blank
1541 					    | NETTXF_data_validated);
1542 				}
1543 #if __FreeBSD_version >= 700000
1544 				if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1545 					struct netif_extra_info *gso =
1546 						(struct netif_extra_info *)
1547 						RING_GET_REQUEST(&sc->tx, ++i);
1548 
1549 					if (extra)
1550 						extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
1551 					else
1552 						tx->flags |= NETTXF_extra_info;
1553 
1554 					gso->u.gso.size = m->m_pkthdr.tso_segsz;
1555 					gso->u.gso.type =
1556 						XEN_NETIF_GSO_TYPE_TCPV4;
1557 					gso->u.gso.pad = 0;
1558 					gso->u.gso.features = 0;
1559 
1560 					gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1561 					gso->flags = 0;
1562 					extra = gso;
1563 				}
1564 #endif
1565 			} else {
1566 				tx->size = m->m_len;
1567 			}
1568 			if (m->m_next) {
1569 				tx->flags |= NETTXF_more_data;
1570 				i++;
1571 			}
1572 		}
1573 
1574 		BPF_MTAP(ifp, m_head);
1575 
1576 		sc->stats.tx_bytes += m_head->m_pkthdr.len;
1577 		sc->stats.tx_packets++;
1578 	}
1579 
1580 	sc->tx.req_prod_pvt = i;
1581 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1582 	if (notify)
1583 		notify_remote_via_irq(sc->irq);
1584 
1585 	xn_txeof(sc);
1586 
1587 	if (RING_FULL(&sc->tx)) {
1588 		sc->tx_full = 1;
1589 #if 0
1590 		netif_stop_queue(dev);
1591 #endif
1592 	}
1593 
1594 	return;
1595 }
1596 
1597 static void
1598 xn_start(struct ifnet *ifp)
1599 {
1600 	struct netfront_info *sc;
1601 	sc = ifp->if_softc;
1602 	XN_TX_LOCK(sc);
1603 	xn_start_locked(ifp);
1604 	XN_TX_UNLOCK(sc);
1605 }
1606 
1607 /* equivalent of network_open() in Linux */
1608 static void
1609 xn_ifinit_locked(struct netfront_info *sc)
1610 {
1611 	struct ifnet *ifp;
1612 
1613 	XN_LOCK_ASSERT(sc);
1614 
1615 	ifp = sc->xn_ifp;
1616 
1617 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1618 		return;
1619 
1620 	xn_stop(sc);
1621 
1622 	network_alloc_rx_buffers(sc);
1623 	sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1624 
1625 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1626 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1627 
1628 	callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1629 
1630 }
1631 
1632 
1633 static void
1634 xn_ifinit(void *xsc)
1635 {
1636 	struct netfront_info *sc = xsc;
1637 
1638 	XN_LOCK(sc);
1639 	xn_ifinit_locked(sc);
1640 	XN_UNLOCK(sc);
1641 
1642 }
1643 
1644 
1645 static int
1646 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1647 {
1648 	struct netfront_info *sc = ifp->if_softc;
1649 	struct ifreq *ifr = (struct ifreq *) data;
1650 	struct ifaddr *ifa = (struct ifaddr *)data;
1651 
1652 	int mask, error = 0;
1653 	switch(cmd) {
1654 	case SIOCSIFADDR:
1655 	case SIOCGIFADDR:
1656 		XN_LOCK(sc);
1657 		if (ifa->ifa_addr->sa_family == AF_INET) {
1658 			ifp->if_flags |= IFF_UP;
1659 			if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1660 				xn_ifinit_locked(sc);
1661 			arp_ifinit(ifp, ifa);
1662 			XN_UNLOCK(sc);
1663 		} else {
1664 			XN_UNLOCK(sc);
1665 			error = ether_ioctl(ifp, cmd, data);
1666 		}
1667 		break;
1668 	case SIOCSIFMTU:
1669 		/* XXX can we alter the MTU on a VN ?*/
1670 #ifdef notyet
1671 		if (ifr->ifr_mtu > XN_JUMBO_MTU)
1672 			error = EINVAL;
1673 		else
1674 #endif
1675 		{
1676 			ifp->if_mtu = ifr->ifr_mtu;
1677 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1678 			xn_ifinit(sc);
1679 		}
1680 		break;
1681 	case SIOCSIFFLAGS:
1682 		XN_LOCK(sc);
1683 		if (ifp->if_flags & IFF_UP) {
1684 			/*
1685 			 * If only the state of the PROMISC flag changed,
1686 			 * then just use the 'set promisc mode' command
1687 			 * instead of reinitializing the entire NIC. Doing
1688 			 * a full re-init means reloading the firmware and
1689 			 * waiting for it to start up, which may take a
1690 			 * second or two.
1691 			 */
1692 #ifdef notyet
1693 			/* No promiscuous mode with Xen */
1694 			if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1695 			    ifp->if_flags & IFF_PROMISC &&
1696 			    !(sc->xn_if_flags & IFF_PROMISC)) {
1697 				XN_SETBIT(sc, XN_RX_MODE,
1698 					  XN_RXMODE_RX_PROMISC);
1699 			} else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1700 				   !(ifp->if_flags & IFF_PROMISC) &&
1701 				   sc->xn_if_flags & IFF_PROMISC) {
1702 				XN_CLRBIT(sc, XN_RX_MODE,
1703 					  XN_RXMODE_RX_PROMISC);
1704 			} else
1705 #endif
1706 				xn_ifinit_locked(sc);
1707 		} else {
1708 			if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1709 				xn_stop(sc);
1710 			}
1711 		}
1712 		sc->xn_if_flags = ifp->if_flags;
1713 		XN_UNLOCK(sc);
1714 		error = 0;
1715 		break;
1716 	case SIOCSIFCAP:
1717 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1718 		if (mask & IFCAP_TXCSUM) {
1719 			if (IFCAP_TXCSUM & ifp->if_capenable) {
1720 				ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1721 				ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1722 				    | CSUM_IP | CSUM_TSO);
1723 			} else {
1724 				ifp->if_capenable |= IFCAP_TXCSUM;
1725 				ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1726 				    | CSUM_IP);
1727 			}
1728 		}
1729 		if (mask & IFCAP_RXCSUM) {
1730 			ifp->if_capenable ^= IFCAP_RXCSUM;
1731 		}
1732 #if __FreeBSD_version >= 700000
1733 		if (mask & IFCAP_TSO4) {
1734 			if (IFCAP_TSO4 & ifp->if_capenable) {
1735 				ifp->if_capenable &= ~IFCAP_TSO4;
1736 				ifp->if_hwassist &= ~CSUM_TSO;
1737 			} else if (IFCAP_TXCSUM & ifp->if_capenable) {
1738 				ifp->if_capenable |= IFCAP_TSO4;
1739 				ifp->if_hwassist |= CSUM_TSO;
1740 			} else {
1741 				DPRINTK("Xen requires tx checksum offload"
1742 				    " be enabled to use TSO\n");
1743 				error = EINVAL;
1744 			}
1745 		}
1746 		if (mask & IFCAP_LRO) {
1747 			ifp->if_capenable ^= IFCAP_LRO;
1748 
1749 		}
1750 #endif
1751 		error = 0;
1752 		break;
1753 	case SIOCADDMULTI:
1754 	case SIOCDELMULTI:
1755 #ifdef notyet
1756 		if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1757 			XN_LOCK(sc);
1758 			xn_setmulti(sc);
1759 			XN_UNLOCK(sc);
1760 			error = 0;
1761 		}
1762 #endif
1763 		/* FALLTHROUGH */
1764 	case SIOCSIFMEDIA:
1765 	case SIOCGIFMEDIA:
1766 		error = EINVAL;
1767 		break;
1768 	default:
1769 		error = ether_ioctl(ifp, cmd, data);
1770 	}
1771 
1772 	return (error);
1773 }
1774 
1775 static void
1776 xn_stop(struct netfront_info *sc)
1777 {
1778 	struct ifnet *ifp;
1779 
1780 	XN_LOCK_ASSERT(sc);
1781 
1782 	ifp = sc->xn_ifp;
1783 
1784 	callout_stop(&sc->xn_stat_ch);
1785 
1786 	xn_free_rx_ring(sc);
1787 	xn_free_tx_ring(sc);
1788 
1789 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1790 }
1791 
1792 /* START of Xenolinux helper functions adapted to FreeBSD */
1793 int
1794 network_connect(struct netfront_info *np)
1795 {
1796 	int i, requeue_idx, error;
1797 	grant_ref_t ref;
1798 	netif_rx_request_t *req;
1799 	u_int feature_rx_copy, feature_rx_flip;
1800 
1801 	error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1802 	    "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1803 	if (error)
1804 		feature_rx_copy = 0;
1805 	error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1806 	    "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1807 	if (error)
1808 		feature_rx_flip = 1;
1809 
1810 	/*
1811 	 * Copy packets on receive path if:
1812 	 *  (a) This was requested by user, and the backend supports it; or
1813 	 *  (b) Flipping was requested, but this is unsupported by the backend.
1814 	 */
1815 	np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1816 				(MODPARM_rx_flip && !feature_rx_flip));
1817 
1818 	XN_LOCK(np);
1819 	/* Recovery procedure: */
1820 	error = talk_to_backend(np->xbdev, np);
1821 	if (error)
1822 		return (error);
1823 
1824 	/* Step 1: Reinitialise variables. */
1825 	netif_release_tx_bufs(np);
1826 
1827 	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1828 	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1829 		struct mbuf *m;
1830 		u_long pfn;
1831 
1832 		if (np->rx_mbufs[i] == NULL)
1833 			continue;
1834 
1835 		m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1836 		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1837 		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1838 		pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1839 
1840 		if (!np->copying_receiver) {
1841 			gnttab_grant_foreign_transfer_ref(ref,
1842 			    xenbus_get_otherend_id(np->xbdev),
1843 			    pfn);
1844 		} else {
1845 			gnttab_grant_foreign_access_ref(ref,
1846 			    xenbus_get_otherend_id(np->xbdev),
1847 			    PFNTOMFN(pfn), 0);
1848 		}
1849 		req->gref = ref;
1850 		req->id   = requeue_idx;
1851 
1852 		requeue_idx++;
1853 	}
1854 
1855 	np->rx.req_prod_pvt = requeue_idx;
1856 
1857 	/* Step 3: All public and private state should now be sane.  Get
1858 	 * ready to start sending and receiving packets and give the driver
1859 	 * domain a kick because we've probably just requeued some
1860 	 * packets.
1861 	 */
1862 	netfront_carrier_on(np);
1863 	notify_remote_via_irq(np->irq);
1864 	XN_TX_LOCK(np);
1865 	xn_txeof(np);
1866 	XN_TX_UNLOCK(np);
1867 	network_alloc_rx_buffers(np);
1868 	XN_UNLOCK(np);
1869 
1870 	return (0);
1871 }
1872 
1873 static void
1874 show_device(struct netfront_info *sc)
1875 {
1876 #ifdef DEBUG
1877 	if (sc) {
1878 		IPRINTK("<vif handle=%u %s(%s) evtchn=%u irq=%u tx=%p rx=%p>\n",
1879 			sc->xn_ifno,
1880 			be_state_name[sc->xn_backend_state],
1881 			sc->xn_user_state ? "open" : "closed",
1882 			sc->xn_evtchn,
1883 			sc->xn_irq,
1884 			sc->xn_tx_if,
1885 			sc->xn_rx_if);
1886 	} else {
1887 		IPRINTK("<vif NULL>\n");
1888 	}
1889 #endif
1890 }
1891 
1892 /** Create a network device.
1893  * @param handle device handle
1894  */
1895 int
1896 create_netdev(device_t dev)
1897 {
1898 	int i;
1899 	struct netfront_info *np;
1900 	int err;
1901 	struct ifnet *ifp;
1902 
1903 	np = device_get_softc(dev);
1904 
1905 	np->xbdev         = dev;
1906 
1907 	XN_LOCK_INIT(np, xennetif);
1908 	np->rx_target     = RX_MIN_TARGET;
1909 	np->rx_min_target = RX_MIN_TARGET;
1910 	np->rx_max_target = RX_MAX_TARGET;
1911 
1912 	/* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
1913 	for (i = 0; i <= NET_TX_RING_SIZE; i++) {
1914 		np->tx_mbufs[i] = (void *) ((u_long) i+1);
1915 		np->grant_tx_ref[i] = GRANT_INVALID_REF;
1916 	}
1917 	for (i = 0; i <= NET_RX_RING_SIZE; i++) {
1918 		np->rx_mbufs[i] = NULL;
1919 		np->grant_rx_ref[i] = GRANT_INVALID_REF;
1920 	}
1921 	/* A grant for every tx ring slot */
1922 	if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1923 					  &np->gref_tx_head) < 0) {
1924 		printf("#### netfront can't alloc tx grant refs\n");
1925 		err = ENOMEM;
1926 		goto exit;
1927 	}
1928 	/* A grant for every rx ring slot */
1929 	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1930 					  &np->gref_rx_head) < 0) {
1931 		printf("#### netfront can't alloc rx grant refs\n");
1932 		gnttab_free_grant_references(np->gref_tx_head);
1933 		err = ENOMEM;
1934 		goto exit;
1935 	}
1936 
1937 	err = xen_net_read_mac(dev, np->mac);
1938 	if (err) {
1939 		xenbus_dev_fatal(dev, err, "parsing %s/mac",
1940 		    xenbus_get_node(dev));
1941 		goto out;
1942 	}
1943 
1944 	/* Set up ifnet structure */
1945 	ifp = np->xn_ifp = if_alloc(IFT_ETHER);
1946     	ifp->if_softc = np;
1947     	if_initname(ifp, "xn",  device_get_unit(dev));
1948     	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1949     	ifp->if_ioctl = xn_ioctl;
1950     	ifp->if_output = ether_output;
1951     	ifp->if_start = xn_start;
1952 #ifdef notyet
1953     	ifp->if_watchdog = xn_watchdog;
1954 #endif
1955     	ifp->if_init = xn_ifinit;
1956     	ifp->if_mtu = ETHERMTU;
1957     	ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
1958 
1959     	ifp->if_hwassist = XN_CSUM_FEATURES;
1960     	ifp->if_capabilities = IFCAP_HWCSUM;
1961 #if __FreeBSD_version >= 700000
1962 	ifp->if_capabilities |= IFCAP_TSO4;
1963 	if (xn_enable_lro) {
1964 		int err = tcp_lro_init(&np->xn_lro);
1965 		if (err) {
1966 			device_printf(dev, "LRO initialization failed\n");
1967 			goto exit;
1968 		}
1969 		np->xn_lro.ifp = ifp;
1970 		ifp->if_capabilities |= IFCAP_LRO;
1971 	}
1972 #endif
1973     	ifp->if_capenable = ifp->if_capabilities;
1974 
1975     	ether_ifattach(ifp, np->mac);
1976     	callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
1977 	netfront_carrier_off(np);
1978 
1979 	return (0);
1980 
1981 exit:
1982 	gnttab_free_grant_references(np->gref_tx_head);
1983 out:
1984 	panic("do something smart");
1985 
1986 }
1987 
1988 /**
1989  * Handle the change of state of the backend to Closing.  We must delete our
1990  * device-layer structures now, to ensure that writes are flushed through to
1991  * the backend.  Once is this done, we can switch to Closed in
1992  * acknowledgement.
1993  */
1994 #if 0
1995 static void netfront_closing(device_t dev)
1996 {
1997 #if 0
1998 	struct netfront_info *info = dev->dev_driver_data;
1999 
2000 	DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2001 
2002 	close_netdev(info);
2003 #endif
2004 	xenbus_switch_state(dev, XenbusStateClosed);
2005 }
2006 #endif
2007 
2008 static int netfront_detach(device_t dev)
2009 {
2010 	struct netfront_info *info = device_get_softc(dev);
2011 
2012 	DPRINTK("%s\n", xenbus_get_node(dev));
2013 
2014 	netif_free(info);
2015 
2016 	return 0;
2017 }
2018 
2019 
2020 static void netif_free(struct netfront_info *info)
2021 {
2022 	netif_disconnect_backend(info);
2023 #if 0
2024 	close_netdev(info);
2025 #endif
2026 }
2027 
2028 static void netif_disconnect_backend(struct netfront_info *info)
2029 {
2030 	XN_RX_LOCK(info);
2031 	XN_TX_LOCK(info);
2032 	netfront_carrier_off(info);
2033 	XN_TX_UNLOCK(info);
2034 	XN_RX_UNLOCK(info);
2035 
2036 	end_access(info->tx_ring_ref, info->tx.sring);
2037 	end_access(info->rx_ring_ref, info->rx.sring);
2038 	info->tx_ring_ref = GRANT_INVALID_REF;
2039 	info->rx_ring_ref = GRANT_INVALID_REF;
2040 	info->tx.sring = NULL;
2041 	info->rx.sring = NULL;
2042 
2043 	if (info->irq)
2044 		unbind_from_irqhandler(info->irq);
2045 
2046 	info->irq = 0;
2047 }
2048 
2049 
2050 static void end_access(int ref, void *page)
2051 {
2052 	if (ref != GRANT_INVALID_REF)
2053 		gnttab_end_foreign_access(ref, page);
2054 }
2055 
2056 /* ** Driver registration ** */
2057 static device_method_t netfront_methods[] = {
2058 	/* Device interface */
2059 	DEVMETHOD(device_probe,         netfront_probe),
2060 	DEVMETHOD(device_attach,        netfront_attach),
2061 	DEVMETHOD(device_detach,        netfront_detach),
2062 	DEVMETHOD(device_shutdown,      bus_generic_shutdown),
2063 	DEVMETHOD(device_suspend,       bus_generic_suspend),
2064 	DEVMETHOD(device_resume,        netfront_resume),
2065 
2066 	/* Xenbus interface */
2067 	DEVMETHOD(xenbus_backend_changed, netfront_backend_changed),
2068 
2069 	{ 0, 0 }
2070 };
2071 
2072 static driver_t netfront_driver = {
2073 	"xn",
2074 	netfront_methods,
2075 	sizeof(struct netfront_info),
2076 };
2077 devclass_t netfront_devclass;
2078 
2079 DRIVER_MODULE(xe, xenbus, netfront_driver, netfront_devclass, 0, 0);
2080