3 * Copyright (c) 2004-2006 Kip Macy
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.
20 #include <sys/cdefs.h>
21 __FBSDID("$FreeBSD$");
23 #include <sys/param.h>
24 #include <sys/systm.h>
25 #include <sys/sockio.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>
36 #include <net/if_arp.h>
37 #include <net/ethernet.h>
38 #include <net/if_dl.h>
39 #include <net/if_media.h>
43 #include <net/if_types.h>
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>
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>
67 #include <machine/intr_machdep.h>
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>
79 #include <dev/xen/netfront/mbufq.h>
81 #include "xenbus_if.h"
83 #define XN_CSUM_FEATURES (CSUM_TCP | CSUM_UDP | CSUM_TSO)
85 #define GRANT_INVALID_REF 0
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)
90 #if __FreeBSD_version >= 700000
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
97 static int xn_enable_lro = 1;
98 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
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)");
114 static const int MODPARM_rx_copy = 1;
115 static const int MODPARM_rx_flip = 0;
118 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
119 #define RX_COPY_THRESHOLD 256
121 #define net_ratelimit() 0
123 struct netfront_info;
124 struct netfront_rx_info;
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 *);
130 static void xn_tick_locked(struct netfront_info *);
131 static void xn_tick(void *);
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 *);
141 static void xn_watchdog(struct ifnet *);
144 static void show_device(struct netfront_info *sc);
146 static void netfront_closing(device_t dev);
148 static void netif_free(struct netfront_info *info);
149 static int netfront_detach(device_t dev);
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);
157 /* Xenolinux helper functions */
158 int network_connect(struct netfront_info *);
160 static void xn_free_rx_ring(struct netfront_info *);
162 static void xn_free_tx_ring(struct netfront_info *);
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);
168 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
170 #define INVALID_P2M_ENTRY (~0UL)
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.
177 struct xn_chain_data {
178 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1];
179 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1];
183 struct net_device_stats
185 u_long rx_packets; /* total packets received */
186 u_long tx_packets; /* total packets transmitted */
187 u_long rx_bytes; /* total bytes received */
188 u_long tx_bytes; /* total bytes transmitted */
189 u_long rx_errors; /* bad packets received */
190 u_long tx_errors; /* packet transmit problems */
191 u_long rx_dropped; /* no space in linux buffers */
192 u_long tx_dropped; /* no space available in linux */
193 u_long multicast; /* multicast packets received */
196 /* detailed rx_errors: */
197 u_long rx_length_errors;
198 u_long rx_over_errors; /* receiver ring buff overflow */
199 u_long rx_crc_errors; /* recved pkt with crc error */
200 u_long rx_frame_errors; /* recv'd frame alignment error */
201 u_long rx_fifo_errors; /* recv'r fifo overrun */
202 u_long rx_missed_errors; /* receiver missed packet */
204 /* detailed tx_errors */
205 u_long tx_aborted_errors;
206 u_long tx_carrier_errors;
207 u_long tx_fifo_errors;
208 u_long tx_heartbeat_errors;
209 u_long tx_window_errors;
212 u_long rx_compressed;
213 u_long tx_compressed;
216 struct netfront_info {
218 struct ifnet *xn_ifp;
219 #if __FreeBSD_version >= 700000
220 struct lro_ctrl xn_lro;
223 struct net_device_stats stats;
226 netif_tx_front_ring_t tx;
227 netif_rx_front_ring_t rx;
235 u_int copying_receiver;
238 /* Receive-ring batched refills. */
239 #define RX_MIN_TARGET 32
240 #define RX_MAX_TARGET NET_RX_RING_SIZE
241 int rx_min_target, rx_max_target, rx_target;
244 * {tx,rx}_skbs store outstanding skbuffs. The first entry in each
245 * array is an index into a chain of free entries.
248 grant_ref_t gref_tx_head;
249 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
250 grant_ref_t gref_rx_head;
251 grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
253 #define TX_MAX_TARGET min(NET_RX_RING_SIZE, 256)
257 uint8_t mac[ETHER_ADDR_LEN];
258 struct xn_chain_data xn_cdata; /* mbufs */
259 struct mbuf_head xn_rx_batch; /* head of the batch queue */
262 struct callout xn_stat_ch;
264 u_long rx_pfn_array[NET_RX_RING_SIZE];
265 multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1];
266 mmu_update_t rx_mmu[NET_RX_RING_SIZE];
269 #define rx_mbufs xn_cdata.xn_rx_chain
270 #define tx_mbufs xn_cdata.xn_tx_chain
272 #define XN_LOCK_INIT(_sc, _name) \
273 mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
274 mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF); \
275 sx_init(&(_sc)->sc_lock, #_name"_rx")
277 #define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock)
278 #define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock)
280 #define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock)
281 #define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock)
283 #define XN_LOCK(_sc) sx_xlock(&(_sc)->sc_lock);
284 #define XN_UNLOCK(_sc) sx_xunlock(&(_sc)->sc_lock);
286 #define XN_LOCK_ASSERT(_sc) sx_assert(&(_sc)->sc_lock, SX_LOCKED);
287 #define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED);
288 #define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED);
289 #define XN_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->rx_lock); \
290 mtx_destroy(&(_sc)->tx_lock); \
291 sx_destroy(&(_sc)->sc_lock);
293 struct netfront_rx_info {
294 struct netif_rx_response rx;
295 struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
298 #define netfront_carrier_on(netif) ((netif)->carrier = 1)
299 #define netfront_carrier_off(netif) ((netif)->carrier = 0)
300 #define netfront_carrier_ok(netif) ((netif)->carrier)
302 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
307 * Access macros for acquiring freeing slots in tx_skbs[].
311 add_id_to_freelist(struct mbuf **list, unsigned short id)
314 list[0] = (void *)(u_long)id;
317 static inline unsigned short
318 get_id_from_freelist(struct mbuf **list)
320 u_int id = (u_int)(u_long)list[0];
326 xennet_rxidx(RING_IDX idx)
328 return idx & (NET_RX_RING_SIZE - 1);
331 static inline struct mbuf *
332 xennet_get_rx_mbuf(struct netfront_info *np,
335 int i = xennet_rxidx(ri);
339 np->rx_mbufs[i] = NULL;
343 static inline grant_ref_t
344 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
346 int i = xennet_rxidx(ri);
347 grant_ref_t ref = np->grant_rx_ref[i];
348 np->grant_rx_ref[i] = GRANT_INVALID_REF;
355 #define IPRINTK(fmt, args...) \
356 printf("[XEN] " fmt, ##args)
357 #define WPRINTK(fmt, args...) \
358 printf("[XEN] " fmt, ##args)
360 #define DPRINTK(fmt, args...) \
361 printf("[XEN] %s: " fmt, __func__, ##args)
363 #define DPRINTK(fmt, args...)
367 * Read the 'mac' node at the given device's node in the store, and parse that
368 * as colon-separated octets, placing result the given mac array. mac must be
369 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
370 * Return 0 on success, or errno on error.
373 xen_net_read_mac(device_t dev, uint8_t mac[])
376 char *s, *e, *macstr;
378 error = xenbus_read(XBT_NIL, xenbus_get_node(dev), "mac", NULL,
384 for (i = 0; i < ETHER_ADDR_LEN; i++) {
385 mac[i] = strtoul(s, &e, 16);
386 if (s == e || (e[0] != ':' && e[0] != 0)) {
387 free(macstr, M_DEVBUF);
392 free(macstr, M_DEVBUF);
397 * Entry point to this code when a new device is created. Allocate the basic
398 * structures and the ring buffers for communication with the backend, and
399 * inform the backend of the appropriate details for those. Switch to
403 netfront_probe(device_t dev)
406 if (!strcmp(xenbus_get_type(dev), "vif")) {
407 device_set_desc(dev, "Virtual Network Interface");
415 netfront_attach(device_t dev)
419 err = create_netdev(dev);
421 xenbus_dev_fatal(dev, err, "creating netdev");
425 #if __FreeBSD_version >= 700000
426 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
427 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
428 OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
429 &xn_enable_lro, 0, "Large Receive Offload");
437 * We are reconnecting to the backend, due to a suspend/resume, or a backend
438 * driver restart. We tear down our netif structure and recreate it, but
439 * leave the device-layer structures intact so that this is transparent to the
440 * rest of the kernel.
443 netfront_resume(device_t dev)
445 struct netfront_info *info = device_get_softc(dev);
447 netif_disconnect_backend(info);
452 /* Common code used when first setting up, and when resuming. */
454 talk_to_backend(device_t dev, struct netfront_info *info)
457 struct xenbus_transaction xbt;
458 const char *node = xenbus_get_node(dev);
461 err = xen_net_read_mac(dev, info->mac);
463 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
467 /* Create shared ring, alloc event channel. */
468 err = setup_device(dev, info);
473 err = xenbus_transaction_start(&xbt);
475 xenbus_dev_fatal(dev, err, "starting transaction");
478 err = xenbus_printf(xbt, node, "tx-ring-ref","%u",
481 message = "writing tx ring-ref";
482 goto abort_transaction;
484 err = xenbus_printf(xbt, node, "rx-ring-ref","%u",
487 message = "writing rx ring-ref";
488 goto abort_transaction;
490 err = xenbus_printf(xbt, node,
491 "event-channel", "%u", irq_to_evtchn_port(info->irq));
493 message = "writing event-channel";
494 goto abort_transaction;
496 err = xenbus_printf(xbt, node, "request-rx-copy", "%u",
497 info->copying_receiver);
499 message = "writing request-rx-copy";
500 goto abort_transaction;
502 err = xenbus_printf(xbt, node, "feature-rx-notify", "%d", 1);
504 message = "writing feature-rx-notify";
505 goto abort_transaction;
507 err = xenbus_printf(xbt, node, "feature-sg", "%d", 1);
509 message = "writing feature-sg";
510 goto abort_transaction;
512 #if __FreeBSD_version >= 700000
513 err = xenbus_printf(xbt, node, "feature-gso-tcpv4", "%d", 1);
515 message = "writing feature-gso-tcpv4";
516 goto abort_transaction;
520 err = xenbus_transaction_end(xbt, 0);
524 xenbus_dev_fatal(dev, err, "completing transaction");
531 xenbus_transaction_end(xbt, 1);
532 xenbus_dev_fatal(dev, err, "%s", message);
541 setup_device(device_t dev, struct netfront_info *info)
543 netif_tx_sring_t *txs;
544 netif_rx_sring_t *rxs;
550 info->tx_ring_ref = GRANT_INVALID_REF;
551 info->rx_ring_ref = GRANT_INVALID_REF;
552 info->rx.sring = NULL;
553 info->tx.sring = NULL;
556 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
559 xenbus_dev_fatal(dev, error, "allocating tx ring page");
562 SHARED_RING_INIT(txs);
563 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
564 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
568 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
571 xenbus_dev_fatal(dev, error, "allocating rx ring page");
574 SHARED_RING_INIT(rxs);
575 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
577 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
581 error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
582 "xn", xn_intr, info, INTR_TYPE_NET | INTR_MPSAFE, &info->irq);
585 xenbus_dev_fatal(dev, error,
586 "bind_evtchn_to_irqhandler failed");
600 * If this interface has an ipv4 address, send an arp for it. This
601 * helps to get the network going again after migrating hosts.
604 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
610 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
611 if (ifa->ifa_addr->sa_family == AF_INET) {
612 arp_ifinit(ifp, ifa);
618 * Callback received when the backend's state changes.
621 netfront_backend_changed(device_t dev, XenbusState newstate)
623 struct netfront_info *sc = device_get_softc(dev);
625 DPRINTK("newstate=%d\n", newstate);
628 case XenbusStateInitialising:
629 case XenbusStateInitialised:
630 case XenbusStateConnected:
631 case XenbusStateUnknown:
632 case XenbusStateClosed:
633 case XenbusStateReconfigured:
634 case XenbusStateReconfiguring:
636 case XenbusStateInitWait:
637 if (xenbus_get_state(dev) != XenbusStateInitialising)
639 if (network_connect(sc) != 0)
641 xenbus_set_state(dev, XenbusStateConnected);
642 netfront_send_fake_arp(dev, sc);
644 case XenbusStateClosing:
645 xenbus_set_state(dev, XenbusStateClosed);
651 xn_free_rx_ring(struct netfront_info *sc)
656 for (i = 0; i < NET_RX_RING_SIZE; i++) {
657 if (sc->xn_cdata.xn_rx_chain[i] != NULL) {
658 m_freem(sc->xn_cdata.xn_rx_chain[i]);
659 sc->xn_cdata.xn_rx_chain[i] = NULL;
664 sc->xn_rx_if->req_prod = 0;
665 sc->xn_rx_if->event = sc->rx.rsp_cons ;
670 xn_free_tx_ring(struct netfront_info *sc)
675 for (i = 0; i < NET_TX_RING_SIZE; i++) {
676 if (sc->xn_cdata.xn_tx_chain[i] != NULL) {
677 m_freem(sc->xn_cdata.xn_tx_chain[i]);
678 sc->xn_cdata.xn_tx_chain[i] = NULL;
687 netfront_tx_slot_available(struct netfront_info *np)
689 return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
690 (TX_MAX_TARGET - /* MAX_SKB_FRAGS */ 24 - 2));
693 netif_release_tx_bufs(struct netfront_info *np)
698 for (i = 1; i <= NET_TX_RING_SIZE; i++) {
699 m = np->xn_cdata.xn_tx_chain[i];
701 if (((u_long)m) < KERNBASE)
703 gnttab_grant_foreign_access_ref(np->grant_tx_ref[i],
704 xenbus_get_otherend_id(np->xbdev),
705 virt_to_mfn(mtod(m, vm_offset_t)),
707 gnttab_release_grant_reference(&np->gref_tx_head,
708 np->grant_tx_ref[i]);
709 np->grant_tx_ref[i] = GRANT_INVALID_REF;
710 add_id_to_freelist(np->tx_mbufs, i);
716 network_alloc_rx_buffers(struct netfront_info *sc)
718 int otherend_id = xenbus_get_otherend_id(sc->xbdev);
721 int i, batch_target, notify;
723 struct xen_memory_reservation reservation;
726 netif_rx_request_t *req;
730 req_prod = sc->rx.req_prod_pvt;
732 if (unlikely(sc->carrier == 0))
736 * Allocate skbuffs greedily, even though we batch updates to the
737 * receive ring. This creates a less bursty demand on the memory
738 * allocator, so should reduce the chance of failed allocation
739 * requests both for ourself and for other kernel subsystems.
741 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
742 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
743 MGETHDR(m_new, M_DONTWAIT, MT_DATA);
747 m_cljget(m_new, M_DONTWAIT, MJUMPAGESIZE);
748 if ((m_new->m_flags & M_EXT) == 0) {
759 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
761 /* queue the mbufs allocated */
762 mbufq_tail(&sc->xn_rx_batch, m_new);
765 /* Is the batch large enough to be worthwhile? */
766 if (i < (sc->rx_target/2)) {
767 if (req_prod >sc->rx.sring->req_prod)
771 /* Adjust floating fill target if we risked running out of buffers. */
772 if ( ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) &&
773 ((sc->rx_target *= 2) > sc->rx_max_target) )
774 sc->rx_target = sc->rx_max_target;
777 for (nr_flips = i = 0; ; i++) {
778 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
781 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
782 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
784 id = xennet_rxidx(req_prod + i);
786 KASSERT(sc->xn_cdata.xn_rx_chain[id] == NULL,
787 ("non-NULL xm_rx_chain"));
788 sc->xn_cdata.xn_rx_chain[id] = m_new;
790 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
791 KASSERT((short)ref >= 0, ("negative ref"));
792 sc->grant_rx_ref[id] = ref;
794 vaddr = mtod(m_new, vm_offset_t);
795 pfn = vtophys(vaddr) >> PAGE_SHIFT;
796 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
798 if (sc->copying_receiver == 0) {
799 gnttab_grant_foreign_transfer_ref(ref,
801 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
802 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
803 /* Remove this page before passing
806 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
807 MULTI_update_va_mapping(&sc->rx_mcl[i],
812 gnttab_grant_foreign_access_ref(ref,
819 sc->rx_pfn_array[i] =
820 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
823 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
824 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
826 * We may have allocated buffers which have entries outstanding
827 * in the page * update queue -- make sure we flush those first!
832 /* Tell the ballon driver what is going on. */
833 balloon_update_driver_allowance(i);
835 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
836 reservation.nr_extents = i;
837 reservation.extent_order = 0;
838 reservation.address_bits = 0;
839 reservation.domid = DOMID_SELF;
841 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
843 /* After all PTEs have been zapped, flush the TLB. */
844 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
845 UVMF_TLB_FLUSH|UVMF_ALL;
847 /* Give away a batch of pages. */
848 sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
849 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
850 sc->rx_mcl[i].args[1] = (u_long)&reservation;
851 /* Zap PTEs and give away pages in one big multicall. */
852 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
854 /* Check return status of HYPERVISOR_dom_mem_op(). */
855 if (unlikely(sc->rx_mcl[i].result != i))
856 panic("Unable to reduce memory reservation\n");
858 if (HYPERVISOR_memory_op(
859 XENMEM_decrease_reservation, &reservation)
861 panic("Unable to reduce memory "
868 /* Above is a suitable barrier to ensure backend will see requests. */
869 sc->rx.req_prod_pvt = req_prod + i;
871 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
873 notify_remote_via_irq(sc->irq);
877 xn_rxeof(struct netfront_info *np)
880 #if __FreeBSD_version >= 700000
881 struct lro_ctrl *lro = &np->xn_lro;
882 struct lro_entry *queued;
884 struct netfront_rx_info rinfo;
885 struct netif_rx_response *rx = &rinfo.rx;
886 struct netif_extra_info *extras = rinfo.extras;
888 multicall_entry_t *mcl;
890 struct mbuf_head rxq, errq;
891 int err, pages_flipped = 0, work_to_do;
894 XN_RX_LOCK_ASSERT(np);
895 if (!netfront_carrier_ok(np))
903 rp = np->rx.sring->rsp_prod;
904 rmb(); /* Ensure we see queued responses up to 'rp'. */
908 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
909 memset(extras, 0, sizeof(rinfo.extras));
912 err = xennet_get_responses(np, &rinfo, rp, &m,
917 mbufq_tail(&errq, m);
918 np->stats.rx_errors++;
923 m->m_pkthdr.rcvif = ifp;
924 if ( rx->flags & NETRXF_data_validated ) {
925 /* Tell the stack the checksums are okay */
927 * XXX this isn't necessarily the case - need to add
931 m->m_pkthdr.csum_flags |=
932 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
934 m->m_pkthdr.csum_data = 0xffff;
937 np->stats.rx_packets++;
938 np->stats.rx_bytes += m->m_pkthdr.len;
941 np->rx.rsp_cons = ++i;
945 /* Some pages are no longer absent... */
947 balloon_update_driver_allowance(-pages_flipped);
949 /* Do all the remapping work, and M->P updates, in one big
952 if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
953 mcl = np->rx_mcl + pages_flipped;
954 mcl->op = __HYPERVISOR_mmu_update;
955 mcl->args[0] = (u_long)np->rx_mmu;
956 mcl->args[1] = pages_flipped;
958 mcl->args[3] = DOMID_SELF;
959 (void)HYPERVISOR_multicall(np->rx_mcl,
964 while ((m = mbufq_dequeue(&errq)))
968 * Process all the mbufs after the remapping is complete.
969 * Break the mbuf chain first though.
971 while ((m = mbufq_dequeue(&rxq)) != NULL) {
975 * Do we really need to drop the rx lock?
978 #if __FreeBSD_version >= 700000
979 /* Use LRO if possible */
980 if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
981 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
983 * If LRO fails, pass up to the stack
986 (*ifp->if_input)(ifp, m);
989 (*ifp->if_input)(ifp, m);
996 #if __FreeBSD_version >= 700000
998 * Flush any outstanding LRO work
1000 while (!SLIST_EMPTY(&lro->lro_active)) {
1001 queued = SLIST_FIRST(&lro->lro_active);
1002 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1003 tcp_lro_flush(lro, queued);
1008 /* If we get a callback with very few responses, reduce fill target. */
1009 /* NB. Note exponential increase, linear decrease. */
1010 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1011 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1012 np->rx_target = np->rx_min_target;
1015 network_alloc_rx_buffers(np);
1017 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1018 } while (work_to_do);
1022 xn_txeof(struct netfront_info *np)
1027 netif_tx_response_t *txr;
1030 XN_TX_LOCK_ASSERT(np);
1032 if (!netfront_carrier_ok(np))
1039 prod = np->tx.sring->rsp_prod;
1040 rmb(); /* Ensure we see responses up to 'rp'. */
1042 for (i = np->tx.rsp_cons; i != prod; i++) {
1043 txr = RING_GET_RESPONSE(&np->tx, i);
1044 if (txr->status == NETIF_RSP_NULL)
1048 m = np->xn_cdata.xn_tx_chain[id];
1051 * Increment packet count if this is the last
1052 * mbuf of the chain.
1056 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1058 if (unlikely(gnttab_query_foreign_access(
1059 np->grant_tx_ref[id]) != 0)) {
1060 printf("network_tx_buf_gc: warning "
1061 "-- grant still in use by backend "
1065 gnttab_end_foreign_access_ref(
1066 np->grant_tx_ref[id]);
1067 gnttab_release_grant_reference(
1068 &np->gref_tx_head, np->grant_tx_ref[id]);
1069 np->grant_tx_ref[id] = GRANT_INVALID_REF;
1071 np->xn_cdata.xn_tx_chain[id] = NULL;
1072 add_id_to_freelist(np->xn_cdata.xn_tx_chain, id);
1075 np->tx.rsp_cons = prod;
1078 * Set a new event, then check for race with update of
1079 * tx_cons. Note that it is essential to schedule a
1080 * callback, no matter how few buffers are pending. Even if
1081 * there is space in the transmit ring, higher layers may
1082 * be blocked because too much data is outstanding: in such
1083 * cases notification from Xen is likely to be the only kick
1086 np->tx.sring->rsp_event =
1087 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1091 } while (prod != np->tx.sring->rsp_prod);
1095 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1098 if (np->user_state == UST_OPEN)
1099 netif_wake_queue(dev);
1108 struct netfront_info *np = xsc;
1109 struct ifnet *ifp = np->xn_ifp;
1112 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1113 likely(netfront_carrier_ok(np)) &&
1114 ifp->if_drv_flags & IFF_DRV_RUNNING))
1117 if (np->tx.rsp_cons != np->tx.sring->rsp_prod) {
1127 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1128 !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1134 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1137 int new = xennet_rxidx(np->rx.req_prod_pvt);
1139 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1140 np->rx_mbufs[new] = m;
1141 np->grant_rx_ref[new] = ref;
1142 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1143 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1144 np->rx.req_prod_pvt++;
1148 xennet_get_extras(struct netfront_info *np,
1149 struct netif_extra_info *extras, RING_IDX rp)
1151 struct netif_extra_info *extra;
1152 RING_IDX cons = np->rx.rsp_cons;
1160 if (unlikely(cons + 1 == rp)) {
1162 if (net_ratelimit())
1163 WPRINTK("Missing extra info\n");
1169 extra = (struct netif_extra_info *)
1170 RING_GET_RESPONSE(&np->rx, ++cons);
1172 if (unlikely(!extra->type ||
1173 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1175 if (net_ratelimit())
1176 WPRINTK("Invalid extra type: %d\n",
1181 memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1184 m = xennet_get_rx_mbuf(np, cons);
1185 ref = xennet_get_rx_ref(np, cons);
1186 xennet_move_rx_slot(np, m, ref);
1187 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1189 np->rx.rsp_cons = cons;
1194 xennet_get_responses(struct netfront_info *np,
1195 struct netfront_rx_info *rinfo, RING_IDX rp,
1197 int *pages_flipped_p)
1199 int pages_flipped = *pages_flipped_p;
1200 struct mmu_update *mmu;
1201 struct multicall_entry *mcl;
1202 struct netif_rx_response *rx = &rinfo->rx;
1203 struct netif_extra_info *extras = rinfo->extras;
1204 RING_IDX cons = np->rx.rsp_cons;
1205 struct mbuf *m, *m0, *m_prev;
1206 grant_ref_t ref = xennet_get_rx_ref(np, cons);
1207 int max = 5 /* MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD) */;
1212 m0 = m = m_prev = xennet_get_rx_mbuf(np, cons);
1215 if (rx->flags & NETRXF_extra_info) {
1216 err = xennet_get_extras(np, extras, rp);
1217 cons = np->rx.rsp_cons;
1222 m0->m_pkthdr.len = 0;
1230 printf("rx->status=%hd rx->offset=%hu frags=%u\n",
1231 rx->status, rx->offset, frags);
1233 if (unlikely(rx->status < 0 ||
1234 rx->offset + rx->status > PAGE_SIZE)) {
1236 if (net_ratelimit())
1237 WPRINTK("rx->offset: %x, size: %u\n",
1238 rx->offset, rx->status);
1240 xennet_move_rx_slot(np, m, ref);
1246 * This definitely indicates a bug, either in this driver or in
1247 * the backend driver. In future this should flag the bad
1248 * situation to the system controller to reboot the backed.
1250 if (ref == GRANT_INVALID_REF) {
1252 if (net_ratelimit())
1253 WPRINTK("Bad rx response id %d.\n", rx->id);
1259 if (!np->copying_receiver) {
1260 /* Memory pressure, insufficient buffer
1263 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1264 if (net_ratelimit())
1265 WPRINTK("Unfulfilled rx req "
1266 "(id=%d, st=%d).\n",
1267 rx->id, rx->status);
1268 xennet_move_rx_slot(np, m, ref);
1273 if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1274 /* Remap the page. */
1275 void *vaddr = mtod(m, void *);
1278 mcl = np->rx_mcl + pages_flipped;
1279 mmu = np->rx_mmu + pages_flipped;
1281 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1282 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1283 PG_V | PG_M | PG_A, 0);
1284 pfn = (uintptr_t)m->m_ext.ext_arg1;
1285 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1286 MMU_MACHPHYS_UPDATE;
1289 set_phys_to_machine(pfn, mfn);
1293 ret = gnttab_end_foreign_access_ref(ref);
1294 KASSERT(ret, ("ret != 0"));
1297 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1301 m->m_len = rx->status;
1302 m->m_data += rx->offset;
1303 m0->m_pkthdr.len += rx->status;
1306 if (!(rx->flags & NETRXF_more_data))
1309 if (cons + frags == rp) {
1310 if (net_ratelimit())
1311 WPRINTK("Need more frags\n");
1317 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
1318 m = xennet_get_rx_mbuf(np, cons + frags);
1322 ref = xennet_get_rx_ref(np, cons + frags);
1327 if (unlikely(frags > max)) {
1328 if (net_ratelimit())
1329 WPRINTK("Too many frags\n");
1334 np->rx.rsp_cons = cons + frags;
1336 *pages_flipped_p = pages_flipped;
1342 xn_tick_locked(struct netfront_info *sc)
1344 XN_RX_LOCK_ASSERT(sc);
1345 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1347 /* XXX placeholder for printing debug information */
1355 struct netfront_info *sc;
1364 xn_start_locked(struct ifnet *ifp)
1368 struct mbuf *m_head, *m;
1369 struct netfront_info *sc;
1370 netif_tx_request_t *tx;
1371 netif_extra_info_t *extra;
1374 u_long mfn, tx_bytes;
1378 otherend_id = xenbus_get_otherend_id(sc->xbdev);
1381 if (!netfront_carrier_ok(sc))
1384 for (i = sc->tx.req_prod_pvt; TRUE; i++) {
1385 IF_DEQUEUE(&ifp->if_snd, m_head);
1389 if (!netfront_tx_slot_available(sc)) {
1390 IF_PREPEND(&ifp->if_snd, m_head);
1391 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1397 * Defragment the mbuf if necessary.
1399 for (m = m_head, nfrags = 0; m; m = m->m_next)
1401 if (nfrags > MAX_SKB_FRAGS) {
1402 m = m_defrag(m_head, M_DONTWAIT);
1411 * Start packing the mbufs in this chain into
1412 * the fragment pointers. Stop when we run out
1413 * of fragments or hit the end of the mbuf chain.
1417 for (m = m_head; m; m = m->m_next) {
1418 tx = RING_GET_REQUEST(&sc->tx, i);
1419 id = get_id_from_freelist(sc->xn_cdata.xn_tx_chain);
1420 sc->xn_cdata.xn_tx_chain[id] = m;
1422 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1423 KASSERT((short)ref >= 0, ("Negative ref"));
1424 mfn = virt_to_mfn(mtod(m, vm_offset_t));
1425 gnttab_grant_foreign_access_ref(ref, otherend_id,
1426 mfn, GNTMAP_readonly);
1427 tx->gref = sc->grant_tx_ref[id] = ref;
1428 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1432 * The first fragment has the entire packet
1433 * size, subsequent fragments have just the
1434 * fragment size. The backend works out the
1435 * true size of the first fragment by
1436 * subtracting the sizes of the other
1439 tx->size = m->m_pkthdr.len;
1442 * The first fragment contains the
1443 * checksum flags and is optionally
1444 * followed by extra data for TSO etc.
1446 if (m->m_pkthdr.csum_flags
1447 & CSUM_DELAY_DATA) {
1448 tx->flags |= (NETTXF_csum_blank
1449 | NETTXF_data_validated);
1451 #if __FreeBSD_version >= 700000
1452 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1453 struct netif_extra_info *gso =
1454 (struct netif_extra_info *)
1455 RING_GET_REQUEST(&sc->tx, ++i);
1458 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
1460 tx->flags |= NETTXF_extra_info;
1462 gso->u.gso.size = m->m_pkthdr.tso_segsz;
1464 XEN_NETIF_GSO_TYPE_TCPV4;
1466 gso->u.gso.features = 0;
1468 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1474 tx->size = m->m_len;
1477 tx->flags |= NETTXF_more_data;
1482 BPF_MTAP(ifp, m_head);
1484 sc->stats.tx_bytes += m_head->m_pkthdr.len;
1485 sc->stats.tx_packets++;
1488 sc->tx.req_prod_pvt = i;
1489 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1491 notify_remote_via_irq(sc->irq);
1495 if (RING_FULL(&sc->tx)) {
1498 netif_stop_queue(dev);
1506 xn_start(struct ifnet *ifp)
1508 struct netfront_info *sc;
1511 xn_start_locked(ifp);
1515 /* equivalent of network_open() in Linux */
1517 xn_ifinit_locked(struct netfront_info *sc)
1525 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1530 network_alloc_rx_buffers(sc);
1531 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1533 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1534 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1536 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1542 xn_ifinit(void *xsc)
1544 struct netfront_info *sc = xsc;
1547 xn_ifinit_locked(sc);
1554 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1556 struct netfront_info *sc = ifp->if_softc;
1557 struct ifreq *ifr = (struct ifreq *) data;
1558 struct ifaddr *ifa = (struct ifaddr *)data;
1560 int mask, error = 0;
1565 if (ifa->ifa_addr->sa_family == AF_INET) {
1566 ifp->if_flags |= IFF_UP;
1567 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1568 xn_ifinit_locked(sc);
1569 arp_ifinit(ifp, ifa);
1573 error = ether_ioctl(ifp, cmd, data);
1577 /* XXX can we alter the MTU on a VN ?*/
1579 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1584 ifp->if_mtu = ifr->ifr_mtu;
1585 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1591 if (ifp->if_flags & IFF_UP) {
1593 * If only the state of the PROMISC flag changed,
1594 * then just use the 'set promisc mode' command
1595 * instead of reinitializing the entire NIC. Doing
1596 * a full re-init means reloading the firmware and
1597 * waiting for it to start up, which may take a
1601 /* No promiscuous mode with Xen */
1602 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1603 ifp->if_flags & IFF_PROMISC &&
1604 !(sc->xn_if_flags & IFF_PROMISC)) {
1605 XN_SETBIT(sc, XN_RX_MODE,
1606 XN_RXMODE_RX_PROMISC);
1607 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1608 !(ifp->if_flags & IFF_PROMISC) &&
1609 sc->xn_if_flags & IFF_PROMISC) {
1610 XN_CLRBIT(sc, XN_RX_MODE,
1611 XN_RXMODE_RX_PROMISC);
1614 xn_ifinit_locked(sc);
1616 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1620 sc->xn_if_flags = ifp->if_flags;
1625 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1626 if (mask & IFCAP_TXCSUM) {
1627 if (IFCAP_TXCSUM & ifp->if_capenable) {
1628 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1629 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1630 | CSUM_IP | CSUM_TSO);
1632 ifp->if_capenable |= IFCAP_TXCSUM;
1633 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1637 if (mask & IFCAP_RXCSUM) {
1638 ifp->if_capenable ^= IFCAP_RXCSUM;
1640 #if __FreeBSD_version >= 700000
1641 if (mask & IFCAP_TSO4) {
1642 if (IFCAP_TSO4 & ifp->if_capenable) {
1643 ifp->if_capenable &= ~IFCAP_TSO4;
1644 ifp->if_hwassist &= ~CSUM_TSO;
1645 } else if (IFCAP_TXCSUM & ifp->if_capenable) {
1646 ifp->if_capenable |= IFCAP_TSO4;
1647 ifp->if_hwassist |= CSUM_TSO;
1649 DPRINTK("Xen requires tx checksum offload"
1650 " be enabled to use TSO\n");
1654 if (mask & IFCAP_LRO) {
1655 ifp->if_capenable ^= IFCAP_LRO;
1664 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1677 error = ether_ioctl(ifp, cmd, data);
1684 xn_stop(struct netfront_info *sc)
1692 callout_stop(&sc->xn_stat_ch);
1694 xn_free_rx_ring(sc);
1695 xn_free_tx_ring(sc);
1697 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1700 /* START of Xenolinux helper functions adapted to FreeBSD */
1702 network_connect(struct netfront_info *np)
1704 int i, requeue_idx, error;
1706 netif_rx_request_t *req;
1707 u_int feature_rx_copy, feature_rx_flip;
1709 error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1710 "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1712 feature_rx_copy = 0;
1713 error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1714 "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1716 feature_rx_flip = 1;
1719 * Copy packets on receive path if:
1720 * (a) This was requested by user, and the backend supports it; or
1721 * (b) Flipping was requested, but this is unsupported by the backend.
1723 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1724 (MODPARM_rx_flip && !feature_rx_flip));
1727 /* Recovery procedure: */
1728 error = talk_to_backend(np->xbdev, np);
1732 /* Step 1: Reinitialise variables. */
1733 netif_release_tx_bufs(np);
1735 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1736 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1740 if (np->rx_mbufs[i] == NULL)
1743 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1744 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1745 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1746 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1748 if (!np->copying_receiver) {
1749 gnttab_grant_foreign_transfer_ref(ref,
1750 xenbus_get_otherend_id(np->xbdev),
1753 gnttab_grant_foreign_access_ref(ref,
1754 xenbus_get_otherend_id(np->xbdev),
1758 req->id = requeue_idx;
1763 np->rx.req_prod_pvt = requeue_idx;
1765 /* Step 3: All public and private state should now be sane. Get
1766 * ready to start sending and receiving packets and give the driver
1767 * domain a kick because we've probably just requeued some
1770 netfront_carrier_on(np);
1771 notify_remote_via_irq(np->irq);
1775 network_alloc_rx_buffers(np);
1782 show_device(struct netfront_info *sc)
1786 IPRINTK("<vif handle=%u %s(%s) evtchn=%u irq=%u tx=%p rx=%p>\n",
1788 be_state_name[sc->xn_backend_state],
1789 sc->xn_user_state ? "open" : "closed",
1795 IPRINTK("<vif NULL>\n");
1800 /** Create a network device.
1801 * @param handle device handle
1804 create_netdev(device_t dev)
1807 struct netfront_info *np;
1811 np = device_get_softc(dev);
1815 XN_LOCK_INIT(np, xennetif);
1816 np->rx_target = RX_MIN_TARGET;
1817 np->rx_min_target = RX_MIN_TARGET;
1818 np->rx_max_target = RX_MAX_TARGET;
1820 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
1821 for (i = 0; i <= NET_TX_RING_SIZE; i++) {
1822 np->tx_mbufs[i] = (void *) ((u_long) i+1);
1823 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1825 for (i = 0; i <= NET_RX_RING_SIZE; i++) {
1826 np->rx_mbufs[i] = NULL;
1827 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1829 /* A grant for every tx ring slot */
1830 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1831 &np->gref_tx_head) < 0) {
1832 printf("#### netfront can't alloc tx grant refs\n");
1836 /* A grant for every rx ring slot */
1837 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1838 &np->gref_rx_head) < 0) {
1839 printf("#### netfront can't alloc rx grant refs\n");
1840 gnttab_free_grant_references(np->gref_tx_head);
1845 err = xen_net_read_mac(dev, np->mac);
1847 xenbus_dev_fatal(dev, err, "parsing %s/mac",
1848 xenbus_get_node(dev));
1852 /* Set up ifnet structure */
1853 ifp = np->xn_ifp = if_alloc(IFT_ETHER);
1855 if_initname(ifp, "xn", device_get_unit(dev));
1856 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1857 ifp->if_ioctl = xn_ioctl;
1858 ifp->if_output = ether_output;
1859 ifp->if_start = xn_start;
1861 ifp->if_watchdog = xn_watchdog;
1863 ifp->if_init = xn_ifinit;
1864 ifp->if_mtu = ETHERMTU;
1865 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
1867 ifp->if_hwassist = XN_CSUM_FEATURES;
1868 ifp->if_capabilities = IFCAP_HWCSUM;
1869 #if __FreeBSD_version >= 700000
1870 ifp->if_capabilities |= IFCAP_TSO4;
1871 if (xn_enable_lro) {
1872 int err = tcp_lro_init(&np->xn_lro);
1874 device_printf(dev, "LRO initialization failed\n");
1877 np->xn_lro.ifp = ifp;
1878 ifp->if_capabilities |= IFCAP_LRO;
1881 ifp->if_capenable = ifp->if_capabilities;
1883 ether_ifattach(ifp, np->mac);
1884 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
1885 netfront_carrier_off(np);
1890 gnttab_free_grant_references(np->gref_tx_head);
1892 panic("do something smart");
1897 * Handle the change of state of the backend to Closing. We must delete our
1898 * device-layer structures now, to ensure that writes are flushed through to
1899 * the backend. Once is this done, we can switch to Closed in
1903 static void netfront_closing(device_t dev)
1906 struct netfront_info *info = dev->dev_driver_data;
1908 DPRINTK("netfront_closing: %s removed\n", dev->nodename);
1912 xenbus_switch_state(dev, XenbusStateClosed);
1916 static int netfront_detach(device_t dev)
1918 struct netfront_info *info = device_get_softc(dev);
1920 DPRINTK("%s\n", xenbus_get_node(dev));
1928 static void netif_free(struct netfront_info *info)
1930 netif_disconnect_backend(info);
1936 static void netif_disconnect_backend(struct netfront_info *info)
1940 netfront_carrier_off(info);
1944 end_access(info->tx_ring_ref, info->tx.sring);
1945 end_access(info->rx_ring_ref, info->rx.sring);
1946 info->tx_ring_ref = GRANT_INVALID_REF;
1947 info->rx_ring_ref = GRANT_INVALID_REF;
1948 info->tx.sring = NULL;
1949 info->rx.sring = NULL;
1952 unbind_from_irqhandler(info->irq);
1958 static void end_access(int ref, void *page)
1960 if (ref != GRANT_INVALID_REF)
1961 gnttab_end_foreign_access(ref, page);
1964 /* ** Driver registration ** */
1965 static device_method_t netfront_methods[] = {
1966 /* Device interface */
1967 DEVMETHOD(device_probe, netfront_probe),
1968 DEVMETHOD(device_attach, netfront_attach),
1969 DEVMETHOD(device_detach, netfront_detach),
1970 DEVMETHOD(device_shutdown, bus_generic_shutdown),
1971 DEVMETHOD(device_suspend, bus_generic_suspend),
1972 DEVMETHOD(device_resume, netfront_resume),
1974 /* Xenbus interface */
1975 DEVMETHOD(xenbus_backend_changed, netfront_backend_changed),
1980 static driver_t netfront_driver = {
1983 sizeof(struct netfront_info),
1985 devclass_t netfront_devclass;
1987 DRIVER_MODULE(xe, xenbus, netfront_driver, netfront_devclass, 0, 0);