2 * Copyright (c) 2004-2006 Kip Macy
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/sockio.h>
35 #include <sys/malloc.h>
36 #include <sys/module.h>
37 #include <sys/kernel.h>
38 #include <sys/socket.h>
39 #include <sys/sysctl.h>
40 #include <sys/queue.h>
45 #include <net/if_arp.h>
46 #include <net/ethernet.h>
47 #include <net/if_dl.h>
48 #include <net/if_media.h>
52 #include <net/if_types.h>
55 #include <netinet/in_systm.h>
56 #include <netinet/in.h>
57 #include <netinet/ip.h>
58 #include <netinet/if_ether.h>
59 #if __FreeBSD_version >= 700000
60 #include <netinet/tcp.h>
61 #include <netinet/tcp_lro.h>
67 #include <machine/clock.h> /* for DELAY */
68 #include <machine/bus.h>
69 #include <machine/resource.h>
70 #include <machine/frame.h>
71 #include <machine/vmparam.h>
76 #include <machine/intr_machdep.h>
78 #include <machine/xen/xen-os.h>
79 #include <machine/xen/xenfunc.h>
80 #include <xen/hypervisor.h>
81 #include <xen/xen_intr.h>
82 #include <xen/evtchn.h>
83 #include <xen/gnttab.h>
84 #include <xen/interface/memory.h>
85 #include <xen/interface/io/netif.h>
86 #include <xen/xenbus/xenbusvar.h>
88 #include <dev/xen/netfront/mbufq.h>
90 #include "xenbus_if.h"
92 #define XN_CSUM_FEATURES (CSUM_TCP | CSUM_UDP | CSUM_TSO)
94 #define GRANT_INVALID_REF 0
96 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
97 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
99 #if __FreeBSD_version >= 700000
101 * Should the driver do LRO on the RX end
102 * this can be toggled on the fly, but the
103 * interface must be reset (down/up) for it
106 static int xn_enable_lro = 1;
107 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
116 static int MODPARM_rx_copy = 0;
117 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0);
118 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)");
119 static int MODPARM_rx_flip = 0;
120 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0);
121 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)");
123 static const int MODPARM_rx_copy = 1;
124 static const int MODPARM_rx_flip = 0;
128 * \brief The maximum allowed data fragments in a single transmit
131 * This limit is imposed by the backend driver. We assume here that
132 * we are dealing with a Linux driver domain and have set our limit
133 * to mirror the Linux MAX_SKB_FRAGS constant.
135 #define MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2)
137 #define RX_COPY_THRESHOLD 256
139 #define net_ratelimit() 0
141 struct netfront_info;
142 struct netfront_rx_info;
144 static void xn_txeof(struct netfront_info *);
145 static void xn_rxeof(struct netfront_info *);
146 static void network_alloc_rx_buffers(struct netfront_info *);
148 static void xn_tick_locked(struct netfront_info *);
149 static void xn_tick(void *);
151 static void xn_intr(void *);
152 static inline int xn_count_frags(struct mbuf *m);
153 static int xn_assemble_tx_request(struct netfront_info *sc,
154 struct mbuf *m_head);
155 static void xn_start_locked(struct ifnet *);
156 static void xn_start(struct ifnet *);
157 static int xn_ioctl(struct ifnet *, u_long, caddr_t);
158 static void xn_ifinit_locked(struct netfront_info *);
159 static void xn_ifinit(void *);
160 static void xn_stop(struct netfront_info *);
162 static void xn_watchdog(struct ifnet *);
165 static void show_device(struct netfront_info *sc);
167 static void netfront_closing(device_t dev);
169 static void netif_free(struct netfront_info *info);
170 static int netfront_detach(device_t dev);
172 static int talk_to_backend(device_t dev, struct netfront_info *info);
173 static int create_netdev(device_t dev);
174 static void netif_disconnect_backend(struct netfront_info *info);
175 static int setup_device(device_t dev, struct netfront_info *info);
176 static void end_access(int ref, void *page);
178 static int xn_ifmedia_upd(struct ifnet *ifp);
179 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
181 /* Xenolinux helper functions */
182 int network_connect(struct netfront_info *);
184 static void xn_free_rx_ring(struct netfront_info *);
186 static void xn_free_tx_ring(struct netfront_info *);
188 static int xennet_get_responses(struct netfront_info *np,
189 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
190 struct mbuf **list, int *pages_flipped_p);
192 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
194 #define INVALID_P2M_ENTRY (~0UL)
197 * Mbuf pointers. We need these to keep track of the virtual addresses
198 * of our mbuf chains since we can only convert from virtual to physical,
199 * not the other way around. The size must track the free index arrays.
201 struct xn_chain_data {
202 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1];
204 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1];
207 #define NUM_ELEMENTS(x) (sizeof(x)/sizeof(*x))
209 struct net_device_stats
211 u_long rx_packets; /* total packets received */
212 u_long tx_packets; /* total packets transmitted */
213 u_long rx_bytes; /* total bytes received */
214 u_long tx_bytes; /* total bytes transmitted */
215 u_long rx_errors; /* bad packets received */
216 u_long tx_errors; /* packet transmit problems */
217 u_long rx_dropped; /* no space in linux buffers */
218 u_long tx_dropped; /* no space available in linux */
219 u_long multicast; /* multicast packets received */
222 /* detailed rx_errors: */
223 u_long rx_length_errors;
224 u_long rx_over_errors; /* receiver ring buff overflow */
225 u_long rx_crc_errors; /* recved pkt with crc error */
226 u_long rx_frame_errors; /* recv'd frame alignment error */
227 u_long rx_fifo_errors; /* recv'r fifo overrun */
228 u_long rx_missed_errors; /* receiver missed packet */
230 /* detailed tx_errors */
231 u_long tx_aborted_errors;
232 u_long tx_carrier_errors;
233 u_long tx_fifo_errors;
234 u_long tx_heartbeat_errors;
235 u_long tx_window_errors;
238 u_long rx_compressed;
239 u_long tx_compressed;
242 struct netfront_info {
244 struct ifnet *xn_ifp;
245 #if __FreeBSD_version >= 700000
246 struct lro_ctrl xn_lro;
249 struct net_device_stats stats;
252 netif_tx_front_ring_t tx;
253 netif_rx_front_ring_t rx;
261 u_int copying_receiver;
264 /* Receive-ring batched refills. */
265 #define RX_MIN_TARGET 32
266 #define RX_MAX_TARGET NET_RX_RING_SIZE
271 grant_ref_t gref_tx_head;
272 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
273 grant_ref_t gref_rx_head;
274 grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
279 uint8_t mac[ETHER_ADDR_LEN];
280 struct xn_chain_data xn_cdata; /* mbufs */
281 struct mbuf_head xn_rx_batch; /* head of the batch queue */
284 struct callout xn_stat_ch;
286 u_long rx_pfn_array[NET_RX_RING_SIZE];
287 multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1];
288 mmu_update_t rx_mmu[NET_RX_RING_SIZE];
289 struct ifmedia sc_media;
292 #define rx_mbufs xn_cdata.xn_rx_chain
293 #define tx_mbufs xn_cdata.xn_tx_chain
295 #define XN_LOCK_INIT(_sc, _name) \
296 mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
297 mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF); \
298 mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF)
300 #define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock)
301 #define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock)
303 #define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock)
304 #define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock)
306 #define XN_LOCK(_sc) mtx_lock(&(_sc)->sc_lock);
307 #define XN_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_lock);
309 #define XN_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->sc_lock, MA_OWNED);
310 #define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED);
311 #define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED);
312 #define XN_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->rx_lock); \
313 mtx_destroy(&(_sc)->tx_lock); \
314 mtx_destroy(&(_sc)->sc_lock);
316 struct netfront_rx_info {
317 struct netif_rx_response rx;
318 struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
321 #define netfront_carrier_on(netif) ((netif)->carrier = 1)
322 #define netfront_carrier_off(netif) ((netif)->carrier = 0)
323 #define netfront_carrier_ok(netif) ((netif)->carrier)
325 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
330 * Access macros for acquiring freeing slots in tx_skbs[].
334 add_id_to_freelist(struct mbuf **list, uintptr_t id)
337 ("%s: the head item (0) must always be free.", __func__));
339 list[0] = (struct mbuf *)id;
342 static inline unsigned short
343 get_id_from_freelist(struct mbuf **list)
347 id = (uintptr_t)list[0];
349 ("%s: the head item (0) must always remain free.", __func__));
355 xennet_rxidx(RING_IDX idx)
357 return idx & (NET_RX_RING_SIZE - 1);
360 static inline struct mbuf *
361 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
363 int i = xennet_rxidx(ri);
367 np->rx_mbufs[i] = NULL;
371 static inline grant_ref_t
372 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
374 int i = xennet_rxidx(ri);
375 grant_ref_t ref = np->grant_rx_ref[i];
376 np->grant_rx_ref[i] = GRANT_INVALID_REF;
380 #define IPRINTK(fmt, args...) \
381 printf("[XEN] " fmt, ##args)
383 #define WPRINTK(fmt, args...) \
384 printf("[XEN] " fmt, ##args)
386 #define WPRINTK(fmt, args...)
389 #define DPRINTK(fmt, args...) \
390 printf("[XEN] %s: " fmt, __func__, ##args)
392 #define DPRINTK(fmt, args...)
396 * Read the 'mac' node at the given device's node in the store, and parse that
397 * as colon-separated octets, placing result the given mac array. mac must be
398 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
399 * Return 0 on success, or errno on error.
402 xen_net_read_mac(device_t dev, uint8_t mac[])
405 char *s, *e, *macstr;
407 error = xenbus_read(XBT_NIL, xenbus_get_node(dev), "mac", NULL,
413 for (i = 0; i < ETHER_ADDR_LEN; i++) {
414 mac[i] = strtoul(s, &e, 16);
415 if (s == e || (e[0] != ':' && e[0] != 0)) {
416 free(macstr, M_DEVBUF);
421 free(macstr, M_DEVBUF);
426 * Entry point to this code when a new device is created. Allocate the basic
427 * structures and the ring buffers for communication with the backend, and
428 * inform the backend of the appropriate details for those. Switch to
432 netfront_probe(device_t dev)
435 if (!strcmp(xenbus_get_type(dev), "vif")) {
436 device_set_desc(dev, "Virtual Network Interface");
444 netfront_attach(device_t dev)
448 err = create_netdev(dev);
450 xenbus_dev_fatal(dev, err, "creating netdev");
454 #if __FreeBSD_version >= 700000
455 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
456 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
457 OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
458 &xn_enable_lro, 0, "Large Receive Offload");
466 * We are reconnecting to the backend, due to a suspend/resume, or a backend
467 * driver restart. We tear down our netif structure and recreate it, but
468 * leave the device-layer structures intact so that this is transparent to the
469 * rest of the kernel.
472 netfront_resume(device_t dev)
474 struct netfront_info *info = device_get_softc(dev);
476 netif_disconnect_backend(info);
481 /* Common code used when first setting up, and when resuming. */
483 talk_to_backend(device_t dev, struct netfront_info *info)
486 struct xenbus_transaction xbt;
487 const char *node = xenbus_get_node(dev);
490 err = xen_net_read_mac(dev, info->mac);
492 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
496 /* Create shared ring, alloc event channel. */
497 err = setup_device(dev, info);
502 err = xenbus_transaction_start(&xbt);
504 xenbus_dev_fatal(dev, err, "starting transaction");
507 err = xenbus_printf(xbt, node, "tx-ring-ref","%u",
510 message = "writing tx ring-ref";
511 goto abort_transaction;
513 err = xenbus_printf(xbt, node, "rx-ring-ref","%u",
516 message = "writing rx ring-ref";
517 goto abort_transaction;
519 err = xenbus_printf(xbt, node,
520 "event-channel", "%u", irq_to_evtchn_port(info->irq));
522 message = "writing event-channel";
523 goto abort_transaction;
525 err = xenbus_printf(xbt, node, "request-rx-copy", "%u",
526 info->copying_receiver);
528 message = "writing request-rx-copy";
529 goto abort_transaction;
531 err = xenbus_printf(xbt, node, "feature-rx-notify", "%d", 1);
533 message = "writing feature-rx-notify";
534 goto abort_transaction;
536 err = xenbus_printf(xbt, node, "feature-sg", "%d", 1);
538 message = "writing feature-sg";
539 goto abort_transaction;
541 #if __FreeBSD_version >= 700000
542 err = xenbus_printf(xbt, node, "feature-gso-tcpv4", "%d", 1);
544 message = "writing feature-gso-tcpv4";
545 goto abort_transaction;
549 err = xenbus_transaction_end(xbt, 0);
553 xenbus_dev_fatal(dev, err, "completing transaction");
560 xenbus_transaction_end(xbt, 1);
561 xenbus_dev_fatal(dev, err, "%s", message);
570 setup_device(device_t dev, struct netfront_info *info)
572 netif_tx_sring_t *txs;
573 netif_rx_sring_t *rxs;
579 info->tx_ring_ref = GRANT_INVALID_REF;
580 info->rx_ring_ref = GRANT_INVALID_REF;
581 info->rx.sring = NULL;
582 info->tx.sring = NULL;
585 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
588 xenbus_dev_fatal(dev, error, "allocating tx ring page");
591 SHARED_RING_INIT(txs);
592 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
593 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
597 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
600 xenbus_dev_fatal(dev, error, "allocating rx ring page");
603 SHARED_RING_INIT(rxs);
604 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
606 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
610 error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
611 "xn", xn_intr, info, INTR_TYPE_NET | INTR_MPSAFE, &info->irq);
614 xenbus_dev_fatal(dev, error,
615 "bind_evtchn_to_irqhandler failed");
629 * If this interface has an ipv4 address, send an arp for it. This
630 * helps to get the network going again after migrating hosts.
633 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
639 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
640 if (ifa->ifa_addr->sa_family == AF_INET) {
641 arp_ifinit(ifp, ifa);
647 * Callback received when the backend's state changes.
650 netfront_backend_changed(device_t dev, XenbusState newstate)
652 struct netfront_info *sc = device_get_softc(dev);
654 DPRINTK("newstate=%d\n", newstate);
657 case XenbusStateInitialising:
658 case XenbusStateInitialised:
659 case XenbusStateConnected:
660 case XenbusStateUnknown:
661 case XenbusStateClosed:
662 case XenbusStateReconfigured:
663 case XenbusStateReconfiguring:
665 case XenbusStateInitWait:
666 if (xenbus_get_state(dev) != XenbusStateInitialising)
668 if (network_connect(sc) != 0)
670 xenbus_set_state(dev, XenbusStateConnected);
671 netfront_send_fake_arp(dev, sc);
673 case XenbusStateClosing:
674 xenbus_set_state(dev, XenbusStateClosed);
681 xn_free_rx_ring(struct netfront_info *sc)
686 for (i = 0; i < NET_RX_RING_SIZE; i++) {
687 if (sc->xn_cdata.rx_mbufs[i] != NULL) {
688 m_freem(sc->rx_mbufs[i]);
689 sc->rx_mbufs[i] = NULL;
694 sc->xn_rx_if->req_prod = 0;
695 sc->xn_rx_if->event = sc->rx.rsp_cons ;
700 xn_free_tx_ring(struct netfront_info *sc)
705 for (i = 0; i < NET_TX_RING_SIZE; i++) {
706 if (sc->tx_mbufs[i] != NULL) {
707 m_freem(sc->tx_mbufs[i]);
708 sc->xn_cdata.xn_tx_chain[i] = NULL;
717 * \brief Verify that there is sufficient space in the Tx ring
718 * buffer for a maximally sized request to be enqueued.
720 * A transmit request requires a transmit descriptor for each packet
721 * fragment, plus up to 2 entries for "options" (e.g. TSO).
724 xn_tx_slot_available(struct netfront_info *np)
726 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
730 netif_release_tx_bufs(struct netfront_info *np)
734 for (i = 1; i <= NET_TX_RING_SIZE; i++) {
740 * We assume that no kernel addresses are
741 * less than NET_TX_RING_SIZE. Any entry
742 * in the table that is below this number
743 * must be an index from free-list tracking.
745 if (((uintptr_t)m) <= NET_TX_RING_SIZE)
747 gnttab_grant_foreign_access_ref(np->grant_tx_ref[i],
748 xenbus_get_otherend_id(np->xbdev),
749 virt_to_mfn(mtod(m, vm_offset_t)),
751 gnttab_release_grant_reference(&np->gref_tx_head,
752 np->grant_tx_ref[i]);
753 np->grant_tx_ref[i] = GRANT_INVALID_REF;
754 add_id_to_freelist(np->tx_mbufs, i);
755 np->xn_cdata.xn_tx_chain_cnt--;
756 if (np->xn_cdata.xn_tx_chain_cnt < 0) {
757 panic("netif_release_tx_bufs: tx_chain_cnt must be >= 0");
764 network_alloc_rx_buffers(struct netfront_info *sc)
766 int otherend_id = xenbus_get_otherend_id(sc->xbdev);
769 int i, batch_target, notify;
771 struct xen_memory_reservation reservation;
774 netif_rx_request_t *req;
778 req_prod = sc->rx.req_prod_pvt;
780 if (unlikely(sc->carrier == 0))
784 * Allocate mbufs greedily, even though we batch updates to the
785 * receive ring. This creates a less bursty demand on the memory
786 * allocator, and so should reduce the chance of failed allocation
787 * requests both for ourself and for other kernel subsystems.
789 * Here we attempt to maintain rx_target buffers in flight, counting
790 * buffers that we have yet to process in the receive ring.
792 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
793 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
794 MGETHDR(m_new, M_DONTWAIT, MT_DATA);
796 printf("%s: MGETHDR failed\n", __func__);
800 m_cljget(m_new, M_DONTWAIT, MJUMPAGESIZE);
801 if ((m_new->m_flags & M_EXT) == 0) {
802 printf("%s: m_cljget failed\n", __func__);
813 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
815 /* queue the mbufs allocated */
816 mbufq_tail(&sc->xn_rx_batch, m_new);
820 * If we've allocated at least half of our target number of entries,
821 * submit them to the backend - we have enough to make the overhead
822 * of submission worthwhile. Otherwise wait for more mbufs and
823 * request entries to become available.
825 if (i < (sc->rx_target/2)) {
826 if (req_prod >sc->rx.sring->req_prod)
832 * Double floating fill target if we risked having the backend
833 * run out of empty buffers for receive traffic. We define "running
834 * low" as having less than a fourth of our target buffers free
835 * at the time we refilled the queue.
837 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
839 if (sc->rx_target > sc->rx_max_target)
840 sc->rx_target = sc->rx_max_target;
844 for (nr_flips = i = 0; ; i++) {
845 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
848 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
849 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
851 id = xennet_rxidx(req_prod + i);
853 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
854 sc->rx_mbufs[id] = m_new;
856 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
857 KASSERT((short)ref >= 0, ("negative ref"));
858 sc->grant_rx_ref[id] = ref;
860 vaddr = mtod(m_new, vm_offset_t);
861 pfn = vtophys(vaddr) >> PAGE_SHIFT;
862 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
864 if (sc->copying_receiver == 0) {
865 gnttab_grant_foreign_transfer_ref(ref,
867 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
868 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
869 /* Remove this page before passing
872 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
873 MULTI_update_va_mapping(&sc->rx_mcl[i],
878 gnttab_grant_foreign_access_ref(ref,
885 sc->rx_pfn_array[i] =
886 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
889 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
890 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
892 * We may have allocated buffers which have entries outstanding
893 * in the page * update queue -- make sure we flush those first!
898 /* Tell the ballon driver what is going on. */
899 balloon_update_driver_allowance(i);
901 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
902 reservation.nr_extents = i;
903 reservation.extent_order = 0;
904 reservation.address_bits = 0;
905 reservation.domid = DOMID_SELF;
907 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
909 /* After all PTEs have been zapped, flush the TLB. */
910 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
911 UVMF_TLB_FLUSH|UVMF_ALL;
913 /* Give away a batch of pages. */
914 sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
915 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
916 sc->rx_mcl[i].args[1] = (u_long)&reservation;
917 /* Zap PTEs and give away pages in one big multicall. */
918 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
920 /* Check return status of HYPERVISOR_dom_mem_op(). */
921 if (unlikely(sc->rx_mcl[i].result != i))
922 panic("Unable to reduce memory reservation\n");
924 if (HYPERVISOR_memory_op(
925 XENMEM_decrease_reservation, &reservation)
927 panic("Unable to reduce memory "
934 /* Above is a suitable barrier to ensure backend will see requests. */
935 sc->rx.req_prod_pvt = req_prod + i;
937 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
939 notify_remote_via_irq(sc->irq);
943 xn_rxeof(struct netfront_info *np)
946 #if __FreeBSD_version >= 700000
947 struct lro_ctrl *lro = &np->xn_lro;
948 struct lro_entry *queued;
950 struct netfront_rx_info rinfo;
951 struct netif_rx_response *rx = &rinfo.rx;
952 struct netif_extra_info *extras = rinfo.extras;
954 multicall_entry_t *mcl;
956 struct mbuf_head rxq, errq;
957 int err, pages_flipped = 0, work_to_do;
960 XN_RX_LOCK_ASSERT(np);
961 if (!netfront_carrier_ok(np))
969 rp = np->rx.sring->rsp_prod;
970 rmb(); /* Ensure we see queued responses up to 'rp'. */
974 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
975 memset(extras, 0, sizeof(rinfo.extras));
978 err = xennet_get_responses(np, &rinfo, rp, &i, &m,
983 mbufq_tail(&errq, m);
984 np->stats.rx_errors++;
988 m->m_pkthdr.rcvif = ifp;
989 if ( rx->flags & NETRXF_data_validated ) {
990 /* Tell the stack the checksums are okay */
992 * XXX this isn't necessarily the case - need to add
996 m->m_pkthdr.csum_flags |=
997 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
999 m->m_pkthdr.csum_data = 0xffff;
1002 np->stats.rx_packets++;
1003 np->stats.rx_bytes += m->m_pkthdr.len;
1005 mbufq_tail(&rxq, m);
1006 np->rx.rsp_cons = i;
1009 if (pages_flipped) {
1010 /* Some pages are no longer absent... */
1012 balloon_update_driver_allowance(-pages_flipped);
1014 /* Do all the remapping work, and M->P updates, in one big
1017 if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1018 mcl = np->rx_mcl + pages_flipped;
1019 mcl->op = __HYPERVISOR_mmu_update;
1020 mcl->args[0] = (u_long)np->rx_mmu;
1021 mcl->args[1] = pages_flipped;
1023 mcl->args[3] = DOMID_SELF;
1024 (void)HYPERVISOR_multicall(np->rx_mcl,
1029 while ((m = mbufq_dequeue(&errq)))
1033 * Process all the mbufs after the remapping is complete.
1034 * Break the mbuf chain first though.
1036 while ((m = mbufq_dequeue(&rxq)) != NULL) {
1040 * Do we really need to drop the rx lock?
1043 #if __FreeBSD_version >= 700000
1044 /* Use LRO if possible */
1045 if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1046 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1048 * If LRO fails, pass up to the stack
1051 (*ifp->if_input)(ifp, m);
1054 (*ifp->if_input)(ifp, m);
1059 np->rx.rsp_cons = i;
1061 #if __FreeBSD_version >= 700000
1063 * Flush any outstanding LRO work
1065 while (!SLIST_EMPTY(&lro->lro_active)) {
1066 queued = SLIST_FIRST(&lro->lro_active);
1067 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1068 tcp_lro_flush(lro, queued);
1073 /* If we get a callback with very few responses, reduce fill target. */
1074 /* NB. Note exponential increase, linear decrease. */
1075 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1076 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1077 np->rx_target = np->rx_min_target;
1080 network_alloc_rx_buffers(np);
1082 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1083 } while (work_to_do);
1087 xn_txeof(struct netfront_info *np)
1092 netif_tx_response_t *txr;
1095 XN_TX_LOCK_ASSERT(np);
1097 if (!netfront_carrier_ok(np))
1103 prod = np->tx.sring->rsp_prod;
1104 rmb(); /* Ensure we see responses up to 'rp'. */
1106 for (i = np->tx.rsp_cons; i != prod; i++) {
1107 txr = RING_GET_RESPONSE(&np->tx, i);
1108 if (txr->status == NETIF_RSP_NULL)
1111 if (txr->status != NETIF_RSP_OKAY) {
1112 printf("%s: WARNING: response is %d!\n",
1113 __func__, txr->status);
1116 m = np->tx_mbufs[id];
1117 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1118 KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1119 ("mbuf already on the free list, but we're "
1120 "trying to free it again!"));
1124 * Increment packet count if this is the last
1125 * mbuf of the chain.
1129 if (unlikely(gnttab_query_foreign_access(
1130 np->grant_tx_ref[id]) != 0)) {
1131 panic("grant id %u still in use by the backend",
1134 gnttab_end_foreign_access_ref(
1135 np->grant_tx_ref[id]);
1136 gnttab_release_grant_reference(
1137 &np->gref_tx_head, np->grant_tx_ref[id]);
1138 np->grant_tx_ref[id] = GRANT_INVALID_REF;
1140 np->tx_mbufs[id] = NULL;
1141 add_id_to_freelist(np->tx_mbufs, id);
1142 np->xn_cdata.xn_tx_chain_cnt--;
1144 /* Only mark the queue active if we've freed up at least one slot to try */
1145 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1147 np->tx.rsp_cons = prod;
1150 * Set a new event, then check for race with update of
1151 * tx_cons. Note that it is essential to schedule a
1152 * callback, no matter how few buffers are pending. Even if
1153 * there is space in the transmit ring, higher layers may
1154 * be blocked because too much data is outstanding: in such
1155 * cases notification from Xen is likely to be the only kick
1158 np->tx.sring->rsp_event =
1159 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1162 } while (prod != np->tx.sring->rsp_prod);
1165 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1168 if (np->user_state == UST_OPEN)
1169 netif_wake_queue(dev);
1178 struct netfront_info *np = xsc;
1179 struct ifnet *ifp = np->xn_ifp;
1182 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1183 likely(netfront_carrier_ok(np)) &&
1184 ifp->if_drv_flags & IFF_DRV_RUNNING))
1187 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1197 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1198 !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1204 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1207 int new = xennet_rxidx(np->rx.req_prod_pvt);
1209 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1210 np->rx_mbufs[new] = m;
1211 np->grant_rx_ref[new] = ref;
1212 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1213 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1214 np->rx.req_prod_pvt++;
1218 xennet_get_extras(struct netfront_info *np,
1219 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1221 struct netif_extra_info *extra;
1229 if (unlikely(*cons + 1 == rp)) {
1231 if (net_ratelimit())
1232 WPRINTK("Missing extra info\n");
1238 extra = (struct netif_extra_info *)
1239 RING_GET_RESPONSE(&np->rx, ++(*cons));
1241 if (unlikely(!extra->type ||
1242 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1244 if (net_ratelimit())
1245 WPRINTK("Invalid extra type: %d\n",
1250 memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1253 m = xennet_get_rx_mbuf(np, *cons);
1254 ref = xennet_get_rx_ref(np, *cons);
1255 xennet_move_rx_slot(np, m, ref);
1256 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1262 xennet_get_responses(struct netfront_info *np,
1263 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1265 int *pages_flipped_p)
1267 int pages_flipped = *pages_flipped_p;
1268 struct mmu_update *mmu;
1269 struct multicall_entry *mcl;
1270 struct netif_rx_response *rx = &rinfo->rx;
1271 struct netif_extra_info *extras = rinfo->extras;
1272 struct mbuf *m, *m0, *m_prev;
1273 grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1274 RING_IDX ref_cons = *cons;
1275 int max = 5 /* MAX_TX_REQ_FRAGS + (rx->status <= RX_COPY_THRESHOLD) */;
1280 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1283 if (rx->flags & NETRXF_extra_info) {
1284 err = xennet_get_extras(np, extras, rp, cons);
1289 m0->m_pkthdr.len = 0;
1297 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1298 rx->status, rx->offset, frags);
1300 if (unlikely(rx->status < 0 ||
1301 rx->offset + rx->status > PAGE_SIZE)) {
1304 if (net_ratelimit())
1305 WPRINTK("rx->offset: %x, size: %u\n",
1306 rx->offset, rx->status);
1308 xennet_move_rx_slot(np, m, ref);
1313 goto next_skip_queue;
1317 * This definitely indicates a bug, either in this driver or in
1318 * the backend driver. In future this should flag the bad
1319 * situation to the system controller to reboot the backed.
1321 if (ref == GRANT_INVALID_REF) {
1324 if (net_ratelimit())
1325 WPRINTK("Bad rx response id %d.\n", rx->id);
1331 if (!np->copying_receiver) {
1332 /* Memory pressure, insufficient buffer
1335 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1336 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1337 rx->id, rx->status);
1338 xennet_move_rx_slot(np, m, ref);
1343 if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1344 /* Remap the page. */
1345 void *vaddr = mtod(m, void *);
1348 mcl = np->rx_mcl + pages_flipped;
1349 mmu = np->rx_mmu + pages_flipped;
1351 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1352 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1353 PG_V | PG_M | PG_A, 0);
1354 pfn = (uintptr_t)m->m_ext.ext_arg1;
1355 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1356 MMU_MACHPHYS_UPDATE;
1359 set_phys_to_machine(pfn, mfn);
1363 ret = gnttab_end_foreign_access_ref(ref);
1364 KASSERT(ret, ("ret != 0"));
1367 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1373 m->m_len = rx->status;
1374 m->m_data += rx->offset;
1375 m0->m_pkthdr.len += rx->status;
1378 if (!(rx->flags & NETRXF_more_data))
1381 if (*cons + frags == rp) {
1382 if (net_ratelimit())
1383 WPRINTK("Need more frags\n");
1385 printf("%s: cons %u frags %u rp %u, not enough frags\n",
1386 __func__, *cons, frags, rp);
1390 * Note that m can be NULL, if rx->status < 0 or if
1391 * rx->offset + rx->status > PAGE_SIZE above.
1395 rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1396 m = xennet_get_rx_mbuf(np, *cons + frags);
1399 * m_prev == NULL can happen if rx->status < 0 or if
1400 * rx->offset + * rx->status > PAGE_SIZE above.
1406 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1407 * rx->status > PAGE_SIZE above.
1412 ref = xennet_get_rx_ref(np, *cons + frags);
1413 ref_cons = *cons + frags;
1418 if (unlikely(frags > max)) {
1419 if (net_ratelimit())
1420 WPRINTK("Too many frags\n");
1421 printf("%s: too many frags %d > max %d\n", __func__, frags,
1428 *pages_flipped_p = pages_flipped;
1434 xn_tick_locked(struct netfront_info *sc)
1436 XN_RX_LOCK_ASSERT(sc);
1437 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1439 /* XXX placeholder for printing debug information */
1447 struct netfront_info *sc;
1457 * \brief Count the number of fragments in an mbuf chain.
1459 * Surprisingly, there isn't an M* macro for this.
1462 xn_count_frags(struct mbuf *m)
1466 for (nfrags = 0; m != NULL; m = m->m_next)
1473 * Given an mbuf chain, make sure we have enough room and then push
1474 * it onto the transmit ring.
1477 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1482 netif_extra_info_t *extra;
1488 * Defragment the mbuf if necessary.
1490 nfrags = xn_count_frags(m_head);
1493 * Check to see whether this request is longer than netback
1494 * can handle, and try to defrag it.
1497 * It is a bit lame, but the netback driver in Linux can't
1498 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1499 * the Linux network stack.
1501 if (nfrags > MAX_TX_REQ_FRAGS) {
1502 m = m_defrag(m_head, M_DONTWAIT);
1505 * Defrag failed, so free the mbuf and
1506 * therefore drop the packet.
1514 /* Determine how many fragments now exist */
1515 nfrags = xn_count_frags(m_head);
1518 * Check to see whether the defragmented packet has too many
1519 * segments for the Linux netback driver.
1522 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1523 * of mbufs longer than Linux can handle. Make sure we don't
1524 * pass a too-long chain over to the other side by dropping the
1525 * packet. It doesn't look like there is currently a way to
1526 * tell the TCP stack to generate a shorter chain of packets.
1528 if (nfrags > MAX_TX_REQ_FRAGS) {
1534 * This check should be redundant. We've already verified that we
1535 * have enough slots in the ring to handle a packet of maximum
1536 * size, and that our packet is less than the maximum size. Keep
1537 * it in here as an assert for now just to make certain that
1538 * xn_tx_chain_cnt is accurate.
1540 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1541 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1542 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1543 (int) nfrags, (int) NET_TX_RING_SIZE));
1546 * Start packing the mbufs in this chain into
1547 * the fragment pointers. Stop when we run out
1548 * of fragments or hit the end of the mbuf chain.
1552 otherend_id = xenbus_get_otherend_id(sc->xbdev);
1553 for (m = m_head; m; m = m->m_next) {
1554 netif_tx_request_t *tx;
1557 u_long mfn; /* XXX Wrong type? */
1559 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1560 id = get_id_from_freelist(sc->tx_mbufs);
1562 panic("xn_start_locked: was allocated the freelist head!\n");
1563 sc->xn_cdata.xn_tx_chain_cnt++;
1564 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1565 panic("xn_start_locked: tx_chain_cnt must be <= NET_TX_RING_SIZE\n");
1566 sc->tx_mbufs[id] = m;
1568 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1569 KASSERT((short)ref >= 0, ("Negative ref"));
1570 mfn = virt_to_mfn(mtod(m, vm_offset_t));
1571 gnttab_grant_foreign_access_ref(ref, otherend_id,
1572 mfn, GNTMAP_readonly);
1573 tx->gref = sc->grant_tx_ref[id] = ref;
1574 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1578 * The first fragment has the entire packet
1579 * size, subsequent fragments have just the
1580 * fragment size. The backend works out the
1581 * true size of the first fragment by
1582 * subtracting the sizes of the other
1585 tx->size = m->m_pkthdr.len;
1588 * The first fragment contains the checksum flags
1589 * and is optionally followed by extra data for
1593 * CSUM_TSO requires checksum offloading.
1594 * Some versions of FreeBSD fail to
1595 * set CSUM_TCP in the CSUM_TSO case,
1596 * so we have to test for CSUM_TSO
1599 if (m->m_pkthdr.csum_flags
1600 & (CSUM_DELAY_DATA | CSUM_TSO)) {
1601 tx->flags |= (NETTXF_csum_blank
1602 | NETTXF_data_validated);
1604 #if __FreeBSD_version >= 700000
1605 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1606 struct netif_extra_info *gso =
1607 (struct netif_extra_info *)
1608 RING_GET_REQUEST(&sc->tx,
1609 ++sc->tx.req_prod_pvt);
1611 tx->flags |= NETTXF_extra_info;
1613 gso->u.gso.size = m->m_pkthdr.tso_segsz;
1615 XEN_NETIF_GSO_TYPE_TCPV4;
1617 gso->u.gso.features = 0;
1619 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1624 tx->size = m->m_len;
1627 tx->flags |= NETTXF_more_data;
1629 sc->tx.req_prod_pvt++;
1631 BPF_MTAP(ifp, m_head);
1633 sc->stats.tx_bytes += m_head->m_pkthdr.len;
1634 sc->stats.tx_packets++;
1640 xn_start_locked(struct ifnet *ifp)
1642 struct netfront_info *sc;
1643 struct mbuf *m_head;
1648 if (!netfront_carrier_ok(sc))
1652 * While we have enough transmit slots available for at least one
1653 * maximum-sized packet, pull mbufs off the queue and put them on
1654 * the transmit ring.
1656 while (xn_tx_slot_available(sc)) {
1657 IF_DEQUEUE(&ifp->if_snd, m_head);
1661 if (xn_assemble_tx_request(sc, m_head) != 0)
1665 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1667 notify_remote_via_irq(sc->irq);
1669 if (RING_FULL(&sc->tx)) {
1672 netif_stop_queue(dev);
1679 xn_start(struct ifnet *ifp)
1681 struct netfront_info *sc;
1684 xn_start_locked(ifp);
1688 /* equivalent of network_open() in Linux */
1690 xn_ifinit_locked(struct netfront_info *sc)
1698 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1703 network_alloc_rx_buffers(sc);
1704 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1706 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1707 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1708 if_link_state_change(ifp, LINK_STATE_UP);
1710 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1716 xn_ifinit(void *xsc)
1718 struct netfront_info *sc = xsc;
1721 xn_ifinit_locked(sc);
1728 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1730 struct netfront_info *sc = ifp->if_softc;
1731 struct ifreq *ifr = (struct ifreq *) data;
1732 struct ifaddr *ifa = (struct ifaddr *)data;
1734 int mask, error = 0;
1739 if (ifa->ifa_addr->sa_family == AF_INET) {
1740 ifp->if_flags |= IFF_UP;
1741 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1742 xn_ifinit_locked(sc);
1743 arp_ifinit(ifp, ifa);
1747 error = ether_ioctl(ifp, cmd, data);
1751 /* XXX can we alter the MTU on a VN ?*/
1753 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1758 ifp->if_mtu = ifr->ifr_mtu;
1759 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1765 if (ifp->if_flags & IFF_UP) {
1767 * If only the state of the PROMISC flag changed,
1768 * then just use the 'set promisc mode' command
1769 * instead of reinitializing the entire NIC. Doing
1770 * a full re-init means reloading the firmware and
1771 * waiting for it to start up, which may take a
1775 /* No promiscuous mode with Xen */
1776 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1777 ifp->if_flags & IFF_PROMISC &&
1778 !(sc->xn_if_flags & IFF_PROMISC)) {
1779 XN_SETBIT(sc, XN_RX_MODE,
1780 XN_RXMODE_RX_PROMISC);
1781 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1782 !(ifp->if_flags & IFF_PROMISC) &&
1783 sc->xn_if_flags & IFF_PROMISC) {
1784 XN_CLRBIT(sc, XN_RX_MODE,
1785 XN_RXMODE_RX_PROMISC);
1788 xn_ifinit_locked(sc);
1790 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1794 sc->xn_if_flags = ifp->if_flags;
1799 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1800 if (mask & IFCAP_TXCSUM) {
1801 if (IFCAP_TXCSUM & ifp->if_capenable) {
1802 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1803 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1804 | CSUM_IP | CSUM_TSO);
1806 ifp->if_capenable |= IFCAP_TXCSUM;
1807 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1811 if (mask & IFCAP_RXCSUM) {
1812 ifp->if_capenable ^= IFCAP_RXCSUM;
1814 #if __FreeBSD_version >= 700000
1815 if (mask & IFCAP_TSO4) {
1816 if (IFCAP_TSO4 & ifp->if_capenable) {
1817 ifp->if_capenable &= ~IFCAP_TSO4;
1818 ifp->if_hwassist &= ~CSUM_TSO;
1819 } else if (IFCAP_TXCSUM & ifp->if_capenable) {
1820 ifp->if_capenable |= IFCAP_TSO4;
1821 ifp->if_hwassist |= CSUM_TSO;
1823 IPRINTK("Xen requires tx checksum offload"
1824 " be enabled to use TSO\n");
1828 if (mask & IFCAP_LRO) {
1829 ifp->if_capenable ^= IFCAP_LRO;
1838 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1848 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1851 error = ether_ioctl(ifp, cmd, data);
1858 xn_stop(struct netfront_info *sc)
1866 callout_stop(&sc->xn_stat_ch);
1868 xn_free_rx_ring(sc);
1869 xn_free_tx_ring(sc);
1871 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1872 if_link_state_change(ifp, LINK_STATE_DOWN);
1875 /* START of Xenolinux helper functions adapted to FreeBSD */
1877 network_connect(struct netfront_info *np)
1879 int i, requeue_idx, error;
1881 netif_rx_request_t *req;
1882 u_int feature_rx_copy, feature_rx_flip;
1884 error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1885 "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1887 feature_rx_copy = 0;
1888 error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1889 "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1891 feature_rx_flip = 1;
1894 * Copy packets on receive path if:
1895 * (a) This was requested by user, and the backend supports it; or
1896 * (b) Flipping was requested, but this is unsupported by the backend.
1898 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1899 (MODPARM_rx_flip && !feature_rx_flip));
1901 /* Recovery procedure: */
1902 error = talk_to_backend(np->xbdev, np);
1906 /* Step 1: Reinitialise variables. */
1907 netif_release_tx_bufs(np);
1909 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1910 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1914 if (np->rx_mbufs[i] == NULL)
1917 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1918 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1920 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1921 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1923 if (!np->copying_receiver) {
1924 gnttab_grant_foreign_transfer_ref(ref,
1925 xenbus_get_otherend_id(np->xbdev),
1928 gnttab_grant_foreign_access_ref(ref,
1929 xenbus_get_otherend_id(np->xbdev),
1933 req->id = requeue_idx;
1938 np->rx.req_prod_pvt = requeue_idx;
1940 /* Step 3: All public and private state should now be sane. Get
1941 * ready to start sending and receiving packets and give the driver
1942 * domain a kick because we've probably just requeued some
1945 netfront_carrier_on(np);
1946 notify_remote_via_irq(np->irq);
1950 network_alloc_rx_buffers(np);
1956 show_device(struct netfront_info *sc)
1960 IPRINTK("<vif handle=%u %s(%s) evtchn=%u irq=%u tx=%p rx=%p>\n",
1962 be_state_name[sc->xn_backend_state],
1963 sc->xn_user_state ? "open" : "closed",
1969 IPRINTK("<vif NULL>\n");
1974 /** Create a network device.
1975 * @param handle device handle
1978 create_netdev(device_t dev)
1981 struct netfront_info *np;
1985 np = device_get_softc(dev);
1989 XN_LOCK_INIT(np, xennetif);
1991 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
1992 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
1993 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
1995 np->rx_target = RX_MIN_TARGET;
1996 np->rx_min_target = RX_MIN_TARGET;
1997 np->rx_max_target = RX_MAX_TARGET;
1999 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2000 for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2001 np->tx_mbufs[i] = (void *) ((u_long) i+1);
2002 np->grant_tx_ref[i] = GRANT_INVALID_REF;
2004 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2006 for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2008 np->rx_mbufs[i] = NULL;
2009 np->grant_rx_ref[i] = GRANT_INVALID_REF;
2011 /* A grant for every tx ring slot */
2012 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2013 &np->gref_tx_head) != 0) {
2014 IPRINTK("#### netfront can't alloc tx grant refs\n");
2018 /* A grant for every rx ring slot */
2019 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2020 &np->gref_rx_head) != 0) {
2021 WPRINTK("#### netfront can't alloc rx grant refs\n");
2022 gnttab_free_grant_references(np->gref_tx_head);
2027 err = xen_net_read_mac(dev, np->mac);
2029 xenbus_dev_fatal(dev, err, "parsing %s/mac",
2030 xenbus_get_node(dev));
2034 /* Set up ifnet structure */
2035 ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2037 if_initname(ifp, "xn", device_get_unit(dev));
2038 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2039 ifp->if_ioctl = xn_ioctl;
2040 ifp->if_output = ether_output;
2041 ifp->if_start = xn_start;
2043 ifp->if_watchdog = xn_watchdog;
2045 ifp->if_init = xn_ifinit;
2046 ifp->if_mtu = ETHERMTU;
2047 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2049 ifp->if_hwassist = XN_CSUM_FEATURES;
2050 ifp->if_capabilities = IFCAP_HWCSUM;
2051 #if __FreeBSD_version >= 700000
2052 ifp->if_capabilities |= IFCAP_TSO4;
2053 if (xn_enable_lro) {
2054 int err = tcp_lro_init(&np->xn_lro);
2056 device_printf(dev, "LRO initialization failed\n");
2059 np->xn_lro.ifp = ifp;
2060 ifp->if_capabilities |= IFCAP_LRO;
2063 ifp->if_capenable = ifp->if_capabilities;
2065 ether_ifattach(ifp, np->mac);
2066 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2067 netfront_carrier_off(np);
2072 gnttab_free_grant_references(np->gref_tx_head);
2074 panic("do something smart");
2079 * Handle the change of state of the backend to Closing. We must delete our
2080 * device-layer structures now, to ensure that writes are flushed through to
2081 * the backend. Once is this done, we can switch to Closed in
2086 netfront_closing(device_t dev)
2089 struct netfront_info *info = dev->dev_driver_data;
2091 DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2095 xenbus_switch_state(dev, XenbusStateClosed);
2100 netfront_detach(device_t dev)
2102 struct netfront_info *info = device_get_softc(dev);
2104 DPRINTK("%s\n", xenbus_get_node(dev));
2112 netif_free(struct netfront_info *info)
2114 netif_disconnect_backend(info);
2121 netif_disconnect_backend(struct netfront_info *info)
2125 netfront_carrier_off(info);
2129 end_access(info->tx_ring_ref, info->tx.sring);
2130 end_access(info->rx_ring_ref, info->rx.sring);
2131 info->tx_ring_ref = GRANT_INVALID_REF;
2132 info->rx_ring_ref = GRANT_INVALID_REF;
2133 info->tx.sring = NULL;
2134 info->rx.sring = NULL;
2137 unbind_from_irqhandler(info->irq);
2144 end_access(int ref, void *page)
2146 if (ref != GRANT_INVALID_REF)
2147 gnttab_end_foreign_access(ref, page);
2151 xn_ifmedia_upd(struct ifnet *ifp)
2157 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2159 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2160 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2163 /* ** Driver registration ** */
2164 static device_method_t netfront_methods[] = {
2165 /* Device interface */
2166 DEVMETHOD(device_probe, netfront_probe),
2167 DEVMETHOD(device_attach, netfront_attach),
2168 DEVMETHOD(device_detach, netfront_detach),
2169 DEVMETHOD(device_shutdown, bus_generic_shutdown),
2170 DEVMETHOD(device_suspend, bus_generic_suspend),
2171 DEVMETHOD(device_resume, netfront_resume),
2173 /* Xenbus interface */
2174 DEVMETHOD(xenbus_backend_changed, netfront_backend_changed),
2179 static driver_t netfront_driver = {
2182 sizeof(struct netfront_info),
2184 devclass_t netfront_devclass;
2186 DRIVER_MODULE(xe, xenbus, netfront_driver, netfront_devclass, 0, 0);