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
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
31 #include "opt_inet6.h"
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/sockio.h>
36 #include <sys/limits.h>
38 #include <sys/malloc.h>
39 #include <sys/module.h>
40 #include <sys/kernel.h>
41 #include <sys/socket.h>
42 #include <sys/sysctl.h>
43 #include <sys/queue.h>
48 #include <net/if_var.h>
49 #include <net/if_arp.h>
50 #include <net/ethernet.h>
51 #include <net/if_dl.h>
52 #include <net/if_media.h>
56 #include <net/if_types.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in.h>
60 #include <netinet/ip.h>
61 #include <netinet/if_ether.h>
62 #if __FreeBSD_version >= 700000
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_lro.h>
70 #include <machine/clock.h> /* for DELAY */
71 #include <machine/bus.h>
72 #include <machine/resource.h>
73 #include <machine/frame.h>
74 #include <machine/vmparam.h>
79 #include <machine/intr_machdep.h>
81 #include <xen/xen-os.h>
82 #include <xen/hypervisor.h>
83 #include <xen/xen_intr.h>
84 #include <xen/gnttab.h>
85 #include <xen/interface/memory.h>
86 #include <xen/interface/io/netif.h>
87 #include <xen/xenbus/xenbusvar.h>
89 #include <machine/xen/xenvar.h>
91 #include "xenbus_if.h"
93 /* Features supported by all backends. TSO and LRO can be negotiated */
94 #define XN_CSUM_FEATURES (CSUM_TCP | CSUM_UDP)
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 *);
161 static void xn_query_features(struct netfront_info *np);
162 static int xn_configure_features(struct netfront_info *np);
164 static void xn_watchdog(struct ifnet *);
168 static void netfront_closing(device_t dev);
170 static void netif_free(struct netfront_info *info);
171 static int netfront_detach(device_t dev);
173 static int talk_to_backend(device_t dev, struct netfront_info *info);
174 static int create_netdev(device_t dev);
175 static void netif_disconnect_backend(struct netfront_info *info);
176 static int setup_device(device_t dev, struct netfront_info *info);
177 static void free_ring(int *ref, void *ring_ptr_ref);
179 static int xn_ifmedia_upd(struct ifnet *ifp);
180 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
182 /* Xenolinux helper functions */
183 int network_connect(struct netfront_info *);
185 static void xn_free_rx_ring(struct netfront_info *);
187 static void xn_free_tx_ring(struct netfront_info *);
189 static int xennet_get_responses(struct netfront_info *np,
190 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
191 struct mbuf **list, int *pages_flipped_p);
193 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
195 #define INVALID_P2M_ENTRY (~0UL)
198 * Mbuf pointers. We need these to keep track of the virtual addresses
199 * of our mbuf chains since we can only convert from virtual to physical,
200 * not the other way around. The size must track the free index arrays.
202 struct xn_chain_data {
203 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1];
205 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1];
208 struct net_device_stats
210 u_long rx_packets; /* total packets received */
211 u_long tx_packets; /* total packets transmitted */
212 u_long rx_bytes; /* total bytes received */
213 u_long tx_bytes; /* total bytes transmitted */
214 u_long rx_errors; /* bad packets received */
215 u_long tx_errors; /* packet transmit problems */
216 u_long rx_dropped; /* no space in linux buffers */
217 u_long tx_dropped; /* no space available in linux */
218 u_long multicast; /* multicast packets received */
221 /* detailed rx_errors: */
222 u_long rx_length_errors;
223 u_long rx_over_errors; /* receiver ring buff overflow */
224 u_long rx_crc_errors; /* recved pkt with crc error */
225 u_long rx_frame_errors; /* recv'd frame alignment error */
226 u_long rx_fifo_errors; /* recv'r fifo overrun */
227 u_long rx_missed_errors; /* receiver missed packet */
229 /* detailed tx_errors */
230 u_long tx_aborted_errors;
231 u_long tx_carrier_errors;
232 u_long tx_fifo_errors;
233 u_long tx_heartbeat_errors;
234 u_long tx_window_errors;
237 u_long rx_compressed;
238 u_long tx_compressed;
241 struct netfront_info {
242 struct ifnet *xn_ifp;
243 #if __FreeBSD_version >= 700000
244 struct lro_ctrl xn_lro;
247 struct net_device_stats stats;
250 netif_tx_front_ring_t tx;
251 netif_rx_front_ring_t rx;
257 xen_intr_handle_t xen_intr_handle;
258 u_int copying_receiver;
262 /* Receive-ring batched refills. */
263 #define RX_MIN_TARGET 32
264 #define RX_MAX_TARGET NET_RX_RING_SIZE
269 grant_ref_t gref_tx_head;
270 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
271 grant_ref_t gref_rx_head;
272 grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
277 uint8_t mac[ETHER_ADDR_LEN];
278 struct xn_chain_data xn_cdata; /* mbufs */
279 struct mbufq xn_rx_batch; /* batch queue */
282 struct callout xn_stat_ch;
284 u_long rx_pfn_array[NET_RX_RING_SIZE];
285 multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1];
286 mmu_update_t rx_mmu[NET_RX_RING_SIZE];
287 struct ifmedia sc_media;
290 #define rx_mbufs xn_cdata.xn_rx_chain
291 #define tx_mbufs xn_cdata.xn_tx_chain
293 #define XN_LOCK_INIT(_sc, _name) \
294 mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
295 mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF); \
296 mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF)
298 #define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock)
299 #define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock)
301 #define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock)
302 #define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock)
304 #define XN_LOCK(_sc) mtx_lock(&(_sc)->sc_lock);
305 #define XN_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_lock);
307 #define XN_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->sc_lock, MA_OWNED);
308 #define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED);
309 #define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED);
310 #define XN_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->rx_lock); \
311 mtx_destroy(&(_sc)->tx_lock); \
312 mtx_destroy(&(_sc)->sc_lock);
314 struct netfront_rx_info {
315 struct netif_rx_response rx;
316 struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
319 #define netfront_carrier_on(netif) ((netif)->carrier = 1)
320 #define netfront_carrier_off(netif) ((netif)->carrier = 0)
321 #define netfront_carrier_ok(netif) ((netif)->carrier)
323 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
326 add_id_to_freelist(struct mbuf **list, uintptr_t id)
329 ("%s: the head item (0) must always be free.", __func__));
331 list[0] = (struct mbuf *)id;
334 static inline unsigned short
335 get_id_from_freelist(struct mbuf **list)
339 id = (uintptr_t)list[0];
341 ("%s: the head item (0) must always remain free.", __func__));
347 xennet_rxidx(RING_IDX idx)
349 return idx & (NET_RX_RING_SIZE - 1);
352 static inline struct mbuf *
353 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
355 int i = xennet_rxidx(ri);
359 np->rx_mbufs[i] = NULL;
363 static inline grant_ref_t
364 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
366 int i = xennet_rxidx(ri);
367 grant_ref_t ref = np->grant_rx_ref[i];
368 KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
369 np->grant_rx_ref[i] = GRANT_REF_INVALID;
373 #define IPRINTK(fmt, args...) \
374 printf("[XEN] " fmt, ##args)
376 #define WPRINTK(fmt, args...) \
377 printf("[XEN] " fmt, ##args)
379 #define WPRINTK(fmt, args...)
382 #define DPRINTK(fmt, args...) \
383 printf("[XEN] %s: " fmt, __func__, ##args)
385 #define DPRINTK(fmt, args...)
389 * Read the 'mac' node at the given device's node in the store, and parse that
390 * as colon-separated octets, placing result the given mac array. mac must be
391 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
392 * Return 0 on success, or errno on error.
395 xen_net_read_mac(device_t dev, uint8_t mac[])
398 char *s, *e, *macstr;
401 path = xenbus_get_node(dev);
402 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
403 if (error == ENOENT) {
405 * Deal with missing mac XenStore nodes on devices with
406 * HVM emulation (the 'ioemu' configuration attribute)
409 * The HVM emulator may execute in a stub device model
410 * domain which lacks the permission, only given to Dom0,
411 * to update the guest's XenStore tree. For this reason,
412 * the HVM emulator doesn't even attempt to write the
413 * front-side mac node, even when operating in Dom0.
414 * However, there should always be a mac listed in the
415 * backend tree. Fallback to this version if our query
416 * of the front side XenStore location doesn't find
419 path = xenbus_get_otherend_path(dev);
420 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
423 xenbus_dev_fatal(dev, error, "parsing %s/mac", path);
428 for (i = 0; i < ETHER_ADDR_LEN; i++) {
429 mac[i] = strtoul(s, &e, 16);
430 if (s == e || (e[0] != ':' && e[0] != 0)) {
431 free(macstr, M_XENBUS);
436 free(macstr, M_XENBUS);
441 * Entry point to this code when a new device is created. Allocate the basic
442 * structures and the ring buffers for communication with the backend, and
443 * inform the backend of the appropriate details for those. Switch to
447 netfront_probe(device_t dev)
450 if (!strcmp(xenbus_get_type(dev), "vif")) {
451 device_set_desc(dev, "Virtual Network Interface");
459 netfront_attach(device_t dev)
463 err = create_netdev(dev);
465 xenbus_dev_fatal(dev, err, "creating netdev");
469 #if __FreeBSD_version >= 700000
470 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
471 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
472 OID_AUTO, "enable_lro", CTLFLAG_RW,
473 &xn_enable_lro, 0, "Large Receive Offload");
480 netfront_suspend(device_t dev)
482 struct netfront_info *info = device_get_softc(dev);
486 netfront_carrier_off(info);
493 * We are reconnecting to the backend, due to a suspend/resume, or a backend
494 * driver restart. We tear down our netif structure and recreate it, but
495 * leave the device-layer structures intact so that this is transparent to the
496 * rest of the kernel.
499 netfront_resume(device_t dev)
501 struct netfront_info *info = device_get_softc(dev);
503 netif_disconnect_backend(info);
507 /* Common code used when first setting up, and when resuming. */
509 talk_to_backend(device_t dev, struct netfront_info *info)
512 struct xs_transaction xst;
513 const char *node = xenbus_get_node(dev);
516 err = xen_net_read_mac(dev, info->mac);
518 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
522 /* Create shared ring, alloc event channel. */
523 err = setup_device(dev, info);
528 err = xs_transaction_start(&xst);
530 xenbus_dev_fatal(dev, err, "starting transaction");
533 err = xs_printf(xst, node, "tx-ring-ref","%u",
536 message = "writing tx ring-ref";
537 goto abort_transaction;
539 err = xs_printf(xst, node, "rx-ring-ref","%u",
542 message = "writing rx ring-ref";
543 goto abort_transaction;
545 err = xs_printf(xst, node,
546 "event-channel", "%u",
547 xen_intr_port(info->xen_intr_handle));
549 message = "writing event-channel";
550 goto abort_transaction;
552 err = xs_printf(xst, node, "request-rx-copy", "%u",
553 info->copying_receiver);
555 message = "writing request-rx-copy";
556 goto abort_transaction;
558 err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
560 message = "writing feature-rx-notify";
561 goto abort_transaction;
563 err = xs_printf(xst, node, "feature-sg", "%d", 1);
565 message = "writing feature-sg";
566 goto abort_transaction;
568 #if __FreeBSD_version >= 700000
569 err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
571 message = "writing feature-gso-tcpv4";
572 goto abort_transaction;
576 err = xs_transaction_end(xst, 0);
580 xenbus_dev_fatal(dev, err, "completing transaction");
587 xs_transaction_end(xst, 1);
588 xenbus_dev_fatal(dev, err, "%s", message);
596 setup_device(device_t dev, struct netfront_info *info)
598 netif_tx_sring_t *txs;
599 netif_rx_sring_t *rxs;
605 info->tx_ring_ref = GRANT_REF_INVALID;
606 info->rx_ring_ref = GRANT_REF_INVALID;
607 info->rx.sring = NULL;
608 info->tx.sring = NULL;
610 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
613 xenbus_dev_fatal(dev, error, "allocating tx ring page");
616 SHARED_RING_INIT(txs);
617 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
618 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
622 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
625 xenbus_dev_fatal(dev, error, "allocating rx ring page");
628 SHARED_RING_INIT(rxs);
629 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
631 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
635 error = xen_intr_alloc_and_bind_local_port(dev,
636 xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info,
637 INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle);
640 xenbus_dev_fatal(dev, error,
641 "xen_intr_alloc_and_bind_local_port failed");
654 * If this interface has an ipv4 address, send an arp for it. This
655 * helps to get the network going again after migrating hosts.
658 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
664 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
665 if (ifa->ifa_addr->sa_family == AF_INET) {
666 arp_ifinit(ifp, ifa);
673 * Callback received when the backend's state changes.
676 netfront_backend_changed(device_t dev, XenbusState newstate)
678 struct netfront_info *sc = device_get_softc(dev);
680 DPRINTK("newstate=%d\n", newstate);
683 case XenbusStateInitialising:
684 case XenbusStateInitialised:
685 case XenbusStateUnknown:
686 case XenbusStateClosed:
687 case XenbusStateReconfigured:
688 case XenbusStateReconfiguring:
690 case XenbusStateInitWait:
691 if (xenbus_get_state(dev) != XenbusStateInitialising)
693 if (network_connect(sc) != 0)
695 xenbus_set_state(dev, XenbusStateConnected);
697 case XenbusStateClosing:
698 xenbus_set_state(dev, XenbusStateClosed);
700 case XenbusStateConnected:
702 netfront_send_fake_arp(dev, sc);
709 xn_free_rx_ring(struct netfront_info *sc)
714 for (i = 0; i < NET_RX_RING_SIZE; i++) {
715 if (sc->xn_cdata.rx_mbufs[i] != NULL) {
716 m_freem(sc->rx_mbufs[i]);
717 sc->rx_mbufs[i] = NULL;
722 sc->xn_rx_if->req_prod = 0;
723 sc->xn_rx_if->event = sc->rx.rsp_cons ;
728 xn_free_tx_ring(struct netfront_info *sc)
733 for (i = 0; i < NET_TX_RING_SIZE; i++) {
734 if (sc->tx_mbufs[i] != NULL) {
735 m_freem(sc->tx_mbufs[i]);
736 sc->xn_cdata.xn_tx_chain[i] = NULL;
745 * \brief Verify that there is sufficient space in the Tx ring
746 * buffer for a maximally sized request to be enqueued.
748 * A transmit request requires a transmit descriptor for each packet
749 * fragment, plus up to 2 entries for "options" (e.g. TSO).
752 xn_tx_slot_available(struct netfront_info *np)
754 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
758 netif_release_tx_bufs(struct netfront_info *np)
762 for (i = 1; i <= NET_TX_RING_SIZE; i++) {
768 * We assume that no kernel addresses are
769 * less than NET_TX_RING_SIZE. Any entry
770 * in the table that is below this number
771 * must be an index from free-list tracking.
773 if (((uintptr_t)m) <= NET_TX_RING_SIZE)
775 gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
776 gnttab_release_grant_reference(&np->gref_tx_head,
777 np->grant_tx_ref[i]);
778 np->grant_tx_ref[i] = GRANT_REF_INVALID;
779 add_id_to_freelist(np->tx_mbufs, i);
780 np->xn_cdata.xn_tx_chain_cnt--;
781 if (np->xn_cdata.xn_tx_chain_cnt < 0) {
782 panic("%s: tx_chain_cnt must be >= 0", __func__);
789 network_alloc_rx_buffers(struct netfront_info *sc)
791 int otherend_id = xenbus_get_otherend_id(sc->xbdev);
794 int i, batch_target, notify;
796 struct xen_memory_reservation reservation;
799 netif_rx_request_t *req;
803 req_prod = sc->rx.req_prod_pvt;
805 if (__predict_false(sc->carrier == 0))
809 * Allocate mbufs greedily, even though we batch updates to the
810 * receive ring. This creates a less bursty demand on the memory
811 * allocator, and so should reduce the chance of failed allocation
812 * requests both for ourself and for other kernel subsystems.
814 * Here we attempt to maintain rx_target buffers in flight, counting
815 * buffers that we have yet to process in the receive ring.
817 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
818 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
819 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE);
828 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
830 /* queue the mbufs allocated */
831 (void )mbufq_enqueue(&sc->xn_rx_batch, m_new);
835 * If we've allocated at least half of our target number of entries,
836 * submit them to the backend - we have enough to make the overhead
837 * of submission worthwhile. Otherwise wait for more mbufs and
838 * request entries to become available.
840 if (i < (sc->rx_target/2)) {
841 if (req_prod >sc->rx.sring->req_prod)
847 * Double floating fill target if we risked having the backend
848 * run out of empty buffers for receive traffic. We define "running
849 * low" as having less than a fourth of our target buffers free
850 * at the time we refilled the queue.
852 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
854 if (sc->rx_target > sc->rx_max_target)
855 sc->rx_target = sc->rx_max_target;
859 for (nr_flips = i = 0; ; i++) {
860 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
863 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
864 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
866 id = xennet_rxidx(req_prod + i);
868 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
869 sc->rx_mbufs[id] = m_new;
871 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
872 KASSERT(ref != GNTTAB_LIST_END,
873 ("reserved grant references exhuasted"));
874 sc->grant_rx_ref[id] = ref;
876 vaddr = mtod(m_new, vm_offset_t);
877 pfn = vtophys(vaddr) >> PAGE_SHIFT;
878 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
880 if (sc->copying_receiver == 0) {
881 gnttab_grant_foreign_transfer_ref(ref,
883 sc->rx_pfn_array[nr_flips] = pfn;
884 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
885 /* Remove this page before passing
888 MULTI_update_va_mapping(&sc->rx_mcl[i],
893 gnttab_grant_foreign_access_ref(ref,
900 sc->rx_pfn_array[i] =
901 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
904 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
905 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
907 * We may have allocated buffers which have entries outstanding
908 * in the page * update queue -- make sure we flush those first!
912 /* Tell the ballon driver what is going on. */
913 balloon_update_driver_allowance(i);
915 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
916 reservation.nr_extents = i;
917 reservation.extent_order = 0;
918 reservation.address_bits = 0;
919 reservation.domid = DOMID_SELF;
921 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
922 /* After all PTEs have been zapped, flush the TLB. */
923 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
924 UVMF_TLB_FLUSH|UVMF_ALL;
926 /* Give away a batch of pages. */
927 sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
928 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
929 sc->rx_mcl[i].args[1] = (u_long)&reservation;
930 /* Zap PTEs and give away pages in one big multicall. */
931 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
933 if (__predict_false(sc->rx_mcl[i].result != i ||
934 HYPERVISOR_memory_op(XENMEM_decrease_reservation,
936 panic("%s: unable to reduce memory "
937 "reservation\n", __func__);
943 /* Above is a suitable barrier to ensure backend will see requests. */
944 sc->rx.req_prod_pvt = req_prod + i;
946 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
948 xen_intr_signal(sc->xen_intr_handle);
952 xn_rxeof(struct netfront_info *np)
955 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
956 struct lro_ctrl *lro = &np->xn_lro;
957 struct lro_entry *queued;
959 struct netfront_rx_info rinfo;
960 struct netif_rx_response *rx = &rinfo.rx;
961 struct netif_extra_info *extras = rinfo.extras;
963 multicall_entry_t *mcl;
965 struct mbufq rxq, errq;
966 int err, pages_flipped = 0, work_to_do;
969 XN_RX_LOCK_ASSERT(np);
970 if (!netfront_carrier_ok(np))
973 /* XXX: there should be some sane limit. */
974 mbufq_init(&errq, INT_MAX);
975 mbufq_init(&rxq, INT_MAX);
979 rp = np->rx.sring->rsp_prod;
980 rmb(); /* Ensure we see queued responses up to 'rp'. */
984 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
985 memset(extras, 0, sizeof(rinfo.extras));
988 err = xennet_get_responses(np, &rinfo, rp, &i, &m,
991 if (__predict_false(err)) {
993 (void )mbufq_enqueue(&errq, m);
994 np->stats.rx_errors++;
998 m->m_pkthdr.rcvif = ifp;
999 if ( rx->flags & NETRXF_data_validated ) {
1000 /* Tell the stack the checksums are okay */
1002 * XXX this isn't necessarily the case - need to add
1006 m->m_pkthdr.csum_flags |=
1007 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
1009 m->m_pkthdr.csum_data = 0xffff;
1012 np->stats.rx_packets++;
1013 np->stats.rx_bytes += m->m_pkthdr.len;
1015 (void )mbufq_enqueue(&rxq, m);
1016 np->rx.rsp_cons = i;
1019 if (pages_flipped) {
1020 /* Some pages are no longer absent... */
1022 balloon_update_driver_allowance(-pages_flipped);
1024 /* Do all the remapping work, and M->P updates, in one big
1027 if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1028 mcl = np->rx_mcl + pages_flipped;
1029 mcl->op = __HYPERVISOR_mmu_update;
1030 mcl->args[0] = (u_long)np->rx_mmu;
1031 mcl->args[1] = pages_flipped;
1033 mcl->args[3] = DOMID_SELF;
1034 (void)HYPERVISOR_multicall(np->rx_mcl,
1042 * Process all the mbufs after the remapping is complete.
1043 * Break the mbuf chain first though.
1045 while ((m = mbufq_dequeue(&rxq)) != NULL) {
1046 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1049 * Do we really need to drop the rx lock?
1052 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1053 /* Use LRO if possible */
1054 if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1055 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1057 * If LRO fails, pass up to the stack
1060 (*ifp->if_input)(ifp, m);
1063 (*ifp->if_input)(ifp, m);
1068 np->rx.rsp_cons = i;
1070 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1072 * Flush any outstanding LRO work
1074 while (!SLIST_EMPTY(&lro->lro_active)) {
1075 queued = SLIST_FIRST(&lro->lro_active);
1076 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1077 tcp_lro_flush(lro, queued);
1082 /* If we get a callback with very few responses, reduce fill target. */
1083 /* NB. Note exponential increase, linear decrease. */
1084 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1085 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1086 np->rx_target = np->rx_min_target;
1089 network_alloc_rx_buffers(np);
1091 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1092 } while (work_to_do);
1096 xn_txeof(struct netfront_info *np)
1101 netif_tx_response_t *txr;
1104 XN_TX_LOCK_ASSERT(np);
1106 if (!netfront_carrier_ok(np))
1112 prod = np->tx.sring->rsp_prod;
1113 rmb(); /* Ensure we see responses up to 'rp'. */
1115 for (i = np->tx.rsp_cons; i != prod; i++) {
1116 txr = RING_GET_RESPONSE(&np->tx, i);
1117 if (txr->status == NETIF_RSP_NULL)
1120 if (txr->status != NETIF_RSP_OKAY) {
1121 printf("%s: WARNING: response is %d!\n",
1122 __func__, txr->status);
1125 m = np->tx_mbufs[id];
1126 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1127 KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1128 ("mbuf already on the free list, but we're "
1129 "trying to free it again!"));
1133 * Increment packet count if this is the last
1134 * mbuf of the chain.
1137 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1138 if (__predict_false(gnttab_query_foreign_access(
1139 np->grant_tx_ref[id]) != 0)) {
1140 panic("%s: grant id %u still in use by the "
1141 "backend", __func__, id);
1143 gnttab_end_foreign_access_ref(
1144 np->grant_tx_ref[id]);
1145 gnttab_release_grant_reference(
1146 &np->gref_tx_head, np->grant_tx_ref[id]);
1147 np->grant_tx_ref[id] = GRANT_REF_INVALID;
1149 np->tx_mbufs[id] = NULL;
1150 add_id_to_freelist(np->tx_mbufs, id);
1151 np->xn_cdata.xn_tx_chain_cnt--;
1153 /* Only mark the queue active if we've freed up at least one slot to try */
1154 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1156 np->tx.rsp_cons = prod;
1159 * Set a new event, then check for race with update of
1160 * tx_cons. Note that it is essential to schedule a
1161 * callback, no matter how few buffers are pending. Even if
1162 * there is space in the transmit ring, higher layers may
1163 * be blocked because too much data is outstanding: in such
1164 * cases notification from Xen is likely to be the only kick
1167 np->tx.sring->rsp_event =
1168 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1171 } while (prod != np->tx.sring->rsp_prod);
1174 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1177 if (np->user_state == UST_OPEN)
1178 netif_wake_queue(dev);
1186 struct netfront_info *np = xsc;
1187 struct ifnet *ifp = np->xn_ifp;
1190 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1191 likely(netfront_carrier_ok(np)) &&
1192 ifp->if_drv_flags & IFF_DRV_RUNNING))
1195 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1205 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1206 !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1211 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1214 int new = xennet_rxidx(np->rx.req_prod_pvt);
1216 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1217 np->rx_mbufs[new] = m;
1218 np->grant_rx_ref[new] = ref;
1219 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1220 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1221 np->rx.req_prod_pvt++;
1225 xennet_get_extras(struct netfront_info *np,
1226 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1228 struct netif_extra_info *extra;
1236 if (__predict_false(*cons + 1 == rp)) {
1238 if (net_ratelimit())
1239 WPRINTK("Missing extra info\n");
1245 extra = (struct netif_extra_info *)
1246 RING_GET_RESPONSE(&np->rx, ++(*cons));
1248 if (__predict_false(!extra->type ||
1249 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1251 if (net_ratelimit())
1252 WPRINTK("Invalid extra type: %d\n",
1257 memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1260 m = xennet_get_rx_mbuf(np, *cons);
1261 ref = xennet_get_rx_ref(np, *cons);
1262 xennet_move_rx_slot(np, m, ref);
1263 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1269 xennet_get_responses(struct netfront_info *np,
1270 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1272 int *pages_flipped_p)
1274 int pages_flipped = *pages_flipped_p;
1275 struct mmu_update *mmu;
1276 struct multicall_entry *mcl;
1277 struct netif_rx_response *rx = &rinfo->rx;
1278 struct netif_extra_info *extras = rinfo->extras;
1279 struct mbuf *m, *m0, *m_prev;
1280 grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1281 RING_IDX ref_cons = *cons;
1286 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1288 if (rx->flags & NETRXF_extra_info) {
1289 err = xennet_get_extras(np, extras, rp, cons);
1293 m0->m_pkthdr.len = 0;
1301 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1302 rx->status, rx->offset, frags);
1304 if (__predict_false(rx->status < 0 ||
1305 rx->offset + rx->status > PAGE_SIZE)) {
1308 if (net_ratelimit())
1309 WPRINTK("rx->offset: %x, size: %u\n",
1310 rx->offset, rx->status);
1312 xennet_move_rx_slot(np, m, ref);
1317 goto next_skip_queue;
1321 * This definitely indicates a bug, either in this driver or in
1322 * the backend driver. In future this should flag the bad
1323 * situation to the system controller to reboot the backed.
1325 if (ref == GRANT_REF_INVALID) {
1328 if (net_ratelimit())
1329 WPRINTK("Bad rx response id %d.\n", rx->id);
1331 printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1336 if (!np->copying_receiver) {
1337 /* Memory pressure, insufficient buffer
1340 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1341 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1342 rx->id, rx->status);
1343 xennet_move_rx_slot(np, m, ref);
1348 if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1349 /* Remap the page. */
1350 void *vaddr = mtod(m, void *);
1353 mcl = np->rx_mcl + pages_flipped;
1354 mmu = np->rx_mmu + pages_flipped;
1356 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1357 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1358 PG_V | PG_M | PG_A, 0);
1359 pfn = (uintptr_t)m->m_ext.ext_arg1;
1360 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1361 MMU_MACHPHYS_UPDATE;
1366 ret = gnttab_end_foreign_access_ref(ref);
1367 KASSERT(ret, ("ret != 0"));
1370 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1376 m->m_len = rx->status;
1377 m->m_data += rx->offset;
1378 m0->m_pkthdr.len += rx->status;
1381 if (!(rx->flags & NETRXF_more_data))
1384 if (*cons + frags == rp) {
1385 if (net_ratelimit())
1386 WPRINTK("Need more frags\n");
1388 printf("%s: cons %u frags %u rp %u, not enough frags\n",
1389 __func__, *cons, frags, rp);
1393 * Note that m can be NULL, if rx->status < 0 or if
1394 * rx->offset + rx->status > PAGE_SIZE above.
1398 rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1399 m = xennet_get_rx_mbuf(np, *cons + frags);
1402 * m_prev == NULL can happen if rx->status < 0 or if
1403 * rx->offset + * rx->status > PAGE_SIZE above.
1409 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1410 * rx->status > PAGE_SIZE above.
1415 ref = xennet_get_rx_ref(np, *cons + frags);
1416 ref_cons = *cons + frags;
1421 *pages_flipped_p = pages_flipped;
1427 xn_tick_locked(struct netfront_info *sc)
1429 XN_RX_LOCK_ASSERT(sc);
1430 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1432 /* XXX placeholder for printing debug information */
1438 struct netfront_info *sc;
1447 * \brief Count the number of fragments in an mbuf chain.
1449 * Surprisingly, there isn't an M* macro for this.
1452 xn_count_frags(struct mbuf *m)
1456 for (nfrags = 0; m != NULL; m = m->m_next)
1463 * Given an mbuf chain, make sure we have enough room and then push
1464 * it onto the transmit ring.
1467 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1477 * Defragment the mbuf if necessary.
1479 nfrags = xn_count_frags(m_head);
1482 * Check to see whether this request is longer than netback
1483 * can handle, and try to defrag it.
1486 * It is a bit lame, but the netback driver in Linux can't
1487 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1488 * the Linux network stack.
1490 if (nfrags > sc->maxfrags) {
1491 m = m_defrag(m_head, M_NOWAIT);
1494 * Defrag failed, so free the mbuf and
1495 * therefore drop the packet.
1503 /* Determine how many fragments now exist */
1504 nfrags = xn_count_frags(m_head);
1507 * Check to see whether the defragmented packet has too many
1508 * segments for the Linux netback driver.
1511 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1512 * of mbufs longer than Linux can handle. Make sure we don't
1513 * pass a too-long chain over to the other side by dropping the
1514 * packet. It doesn't look like there is currently a way to
1515 * tell the TCP stack to generate a shorter chain of packets.
1517 if (nfrags > MAX_TX_REQ_FRAGS) {
1519 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1520 "won't be able to handle it, dropping\n",
1521 __func__, nfrags, MAX_TX_REQ_FRAGS);
1528 * This check should be redundant. We've already verified that we
1529 * have enough slots in the ring to handle a packet of maximum
1530 * size, and that our packet is less than the maximum size. Keep
1531 * it in here as an assert for now just to make certain that
1532 * xn_tx_chain_cnt is accurate.
1534 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1535 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1536 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1537 (int) nfrags, (int) NET_TX_RING_SIZE));
1540 * Start packing the mbufs in this chain into
1541 * the fragment pointers. Stop when we run out
1542 * of fragments or hit the end of the mbuf chain.
1545 otherend_id = xenbus_get_otherend_id(sc->xbdev);
1546 for (m = m_head; m; m = m->m_next) {
1547 netif_tx_request_t *tx;
1550 u_long mfn; /* XXX Wrong type? */
1552 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1553 id = get_id_from_freelist(sc->tx_mbufs);
1555 panic("%s: was allocated the freelist head!\n",
1557 sc->xn_cdata.xn_tx_chain_cnt++;
1558 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1559 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1561 sc->tx_mbufs[id] = m;
1563 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1564 KASSERT((short)ref >= 0, ("Negative ref"));
1565 mfn = virt_to_mfn(mtod(m, vm_offset_t));
1566 gnttab_grant_foreign_access_ref(ref, otherend_id,
1567 mfn, GNTMAP_readonly);
1568 tx->gref = sc->grant_tx_ref[id] = ref;
1569 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1573 * The first fragment has the entire packet
1574 * size, subsequent fragments have just the
1575 * fragment size. The backend works out the
1576 * true size of the first fragment by
1577 * subtracting the sizes of the other
1580 tx->size = m->m_pkthdr.len;
1583 * The first fragment contains the checksum flags
1584 * and is optionally followed by extra data for
1588 * CSUM_TSO requires checksum offloading.
1589 * Some versions of FreeBSD fail to
1590 * set CSUM_TCP in the CSUM_TSO case,
1591 * so we have to test for CSUM_TSO
1594 if (m->m_pkthdr.csum_flags
1595 & (CSUM_DELAY_DATA | CSUM_TSO)) {
1596 tx->flags |= (NETTXF_csum_blank
1597 | NETTXF_data_validated);
1599 #if __FreeBSD_version >= 700000
1600 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1601 struct netif_extra_info *gso =
1602 (struct netif_extra_info *)
1603 RING_GET_REQUEST(&sc->tx,
1604 ++sc->tx.req_prod_pvt);
1606 tx->flags |= NETTXF_extra_info;
1608 gso->u.gso.size = m->m_pkthdr.tso_segsz;
1610 XEN_NETIF_GSO_TYPE_TCPV4;
1612 gso->u.gso.features = 0;
1614 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1619 tx->size = m->m_len;
1622 tx->flags |= NETTXF_more_data;
1624 sc->tx.req_prod_pvt++;
1626 BPF_MTAP(ifp, m_head);
1628 sc->stats.tx_bytes += m_head->m_pkthdr.len;
1629 sc->stats.tx_packets++;
1635 xn_start_locked(struct ifnet *ifp)
1637 struct netfront_info *sc;
1638 struct mbuf *m_head;
1643 if (!netfront_carrier_ok(sc))
1647 * While we have enough transmit slots available for at least one
1648 * maximum-sized packet, pull mbufs off the queue and put them on
1649 * the transmit ring.
1651 while (xn_tx_slot_available(sc)) {
1652 IF_DEQUEUE(&ifp->if_snd, m_head);
1656 if (xn_assemble_tx_request(sc, m_head) != 0)
1660 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1662 xen_intr_signal(sc->xen_intr_handle);
1664 if (RING_FULL(&sc->tx)) {
1667 netif_stop_queue(dev);
1673 xn_start(struct ifnet *ifp)
1675 struct netfront_info *sc;
1678 xn_start_locked(ifp);
1682 /* equivalent of network_open() in Linux */
1684 xn_ifinit_locked(struct netfront_info *sc)
1692 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1697 network_alloc_rx_buffers(sc);
1698 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1700 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1701 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1702 if_link_state_change(ifp, LINK_STATE_UP);
1704 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1708 xn_ifinit(void *xsc)
1710 struct netfront_info *sc = xsc;
1713 xn_ifinit_locked(sc);
1718 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1720 struct netfront_info *sc = ifp->if_softc;
1721 struct ifreq *ifr = (struct ifreq *) data;
1723 struct ifaddr *ifa = (struct ifaddr *)data;
1726 int mask, error = 0;
1731 if (ifa->ifa_addr->sa_family == AF_INET) {
1732 ifp->if_flags |= IFF_UP;
1733 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1734 xn_ifinit_locked(sc);
1735 arp_ifinit(ifp, ifa);
1740 error = ether_ioctl(ifp, cmd, data);
1746 /* XXX can we alter the MTU on a VN ?*/
1748 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1753 ifp->if_mtu = ifr->ifr_mtu;
1754 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1760 if (ifp->if_flags & IFF_UP) {
1762 * If only the state of the PROMISC flag changed,
1763 * then just use the 'set promisc mode' command
1764 * instead of reinitializing the entire NIC. Doing
1765 * a full re-init means reloading the firmware and
1766 * waiting for it to start up, which may take a
1770 /* No promiscuous mode with Xen */
1771 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1772 ifp->if_flags & IFF_PROMISC &&
1773 !(sc->xn_if_flags & IFF_PROMISC)) {
1774 XN_SETBIT(sc, XN_RX_MODE,
1775 XN_RXMODE_RX_PROMISC);
1776 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1777 !(ifp->if_flags & IFF_PROMISC) &&
1778 sc->xn_if_flags & IFF_PROMISC) {
1779 XN_CLRBIT(sc, XN_RX_MODE,
1780 XN_RXMODE_RX_PROMISC);
1783 xn_ifinit_locked(sc);
1785 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1789 sc->xn_if_flags = ifp->if_flags;
1794 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1795 if (mask & IFCAP_TXCSUM) {
1796 if (IFCAP_TXCSUM & ifp->if_capenable) {
1797 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1798 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1799 | CSUM_IP | CSUM_TSO);
1801 ifp->if_capenable |= IFCAP_TXCSUM;
1802 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1806 if (mask & IFCAP_RXCSUM) {
1807 ifp->if_capenable ^= IFCAP_RXCSUM;
1809 #if __FreeBSD_version >= 700000
1810 if (mask & IFCAP_TSO4) {
1811 if (IFCAP_TSO4 & ifp->if_capenable) {
1812 ifp->if_capenable &= ~IFCAP_TSO4;
1813 ifp->if_hwassist &= ~CSUM_TSO;
1814 } else if (IFCAP_TXCSUM & ifp->if_capenable) {
1815 ifp->if_capenable |= IFCAP_TSO4;
1816 ifp->if_hwassist |= CSUM_TSO;
1818 IPRINTK("Xen requires tx checksum offload"
1819 " be enabled to use TSO\n");
1823 if (mask & IFCAP_LRO) {
1824 ifp->if_capenable ^= IFCAP_LRO;
1833 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1843 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1846 error = ether_ioctl(ifp, cmd, data);
1853 xn_stop(struct netfront_info *sc)
1861 callout_stop(&sc->xn_stat_ch);
1863 xn_free_rx_ring(sc);
1864 xn_free_tx_ring(sc);
1866 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1867 if_link_state_change(ifp, LINK_STATE_DOWN);
1870 /* START of Xenolinux helper functions adapted to FreeBSD */
1872 network_connect(struct netfront_info *np)
1874 int i, requeue_idx, error;
1876 netif_rx_request_t *req;
1877 u_int feature_rx_copy, feature_rx_flip;
1879 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1880 "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1882 feature_rx_copy = 0;
1883 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1884 "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1886 feature_rx_flip = 1;
1889 * Copy packets on receive path if:
1890 * (a) This was requested by user, and the backend supports it; or
1891 * (b) Flipping was requested, but this is unsupported by the backend.
1893 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1894 (MODPARM_rx_flip && !feature_rx_flip));
1896 /* Recovery procedure: */
1897 error = talk_to_backend(np->xbdev, np);
1901 /* Step 1: Reinitialise variables. */
1902 xn_query_features(np);
1903 xn_configure_features(np);
1904 netif_release_tx_bufs(np);
1906 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1907 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1911 if (np->rx_mbufs[i] == NULL)
1914 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1915 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1917 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1918 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1920 if (!np->copying_receiver) {
1921 gnttab_grant_foreign_transfer_ref(ref,
1922 xenbus_get_otherend_id(np->xbdev),
1925 gnttab_grant_foreign_access_ref(ref,
1926 xenbus_get_otherend_id(np->xbdev),
1930 req->id = requeue_idx;
1935 np->rx.req_prod_pvt = requeue_idx;
1937 /* Step 3: All public and private state should now be sane. Get
1938 * ready to start sending and receiving packets and give the driver
1939 * domain a kick because we've probably just requeued some
1942 netfront_carrier_on(np);
1943 xen_intr_signal(np->xen_intr_handle);
1947 network_alloc_rx_buffers(np);
1953 xn_query_features(struct netfront_info *np)
1957 device_printf(np->xbdev, "backend features:");
1959 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1960 "feature-sg", NULL, "%d", &val) < 0)
1965 np->maxfrags = MAX_TX_REQ_FRAGS;
1966 printf(" feature-sg");
1969 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1970 "feature-gso-tcpv4", NULL, "%d", &val) < 0)
1973 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
1975 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
1976 printf(" feature-gso-tcp4");
1983 xn_configure_features(struct netfront_info *np)
1988 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1989 if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0)
1990 tcp_lro_free(&np->xn_lro);
1992 np->xn_ifp->if_capenable =
1993 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4);
1994 np->xn_ifp->if_hwassist &= ~CSUM_TSO;
1995 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1996 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) {
1997 err = tcp_lro_init(&np->xn_lro);
1999 device_printf(np->xbdev, "LRO initialization failed\n");
2001 np->xn_lro.ifp = np->xn_ifp;
2002 np->xn_ifp->if_capenable |= IFCAP_LRO;
2005 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) {
2006 np->xn_ifp->if_capenable |= IFCAP_TSO4;
2007 np->xn_ifp->if_hwassist |= CSUM_TSO;
2014 * Create a network device.
2015 * @param dev Newbus device representing this virtual NIC.
2018 create_netdev(device_t dev)
2021 struct netfront_info *np;
2025 np = device_get_softc(dev);
2029 XN_LOCK_INIT(np, xennetif);
2031 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
2032 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
2033 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
2035 np->rx_target = RX_MIN_TARGET;
2036 np->rx_min_target = RX_MIN_TARGET;
2037 np->rx_max_target = RX_MAX_TARGET;
2039 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2040 for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2041 np->tx_mbufs[i] = (void *) ((u_long) i+1);
2042 np->grant_tx_ref[i] = GRANT_REF_INVALID;
2044 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2046 for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2048 np->rx_mbufs[i] = NULL;
2049 np->grant_rx_ref[i] = GRANT_REF_INVALID;
2052 mbufq_init(&np->xn_rx_batch, INT_MAX);
2054 /* A grant for every tx ring slot */
2055 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2056 &np->gref_tx_head) != 0) {
2057 IPRINTK("#### netfront can't alloc tx grant refs\n");
2061 /* A grant for every rx ring slot */
2062 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2063 &np->gref_rx_head) != 0) {
2064 WPRINTK("#### netfront can't alloc rx grant refs\n");
2065 gnttab_free_grant_references(np->gref_tx_head);
2070 err = xen_net_read_mac(dev, np->mac);
2074 /* Set up ifnet structure */
2075 ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2077 if_initname(ifp, "xn", device_get_unit(dev));
2078 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2079 ifp->if_ioctl = xn_ioctl;
2080 ifp->if_output = ether_output;
2081 ifp->if_start = xn_start;
2083 ifp->if_watchdog = xn_watchdog;
2085 ifp->if_init = xn_ifinit;
2086 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2088 ifp->if_hwassist = XN_CSUM_FEATURES;
2089 ifp->if_capabilities = IFCAP_HWCSUM;
2090 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2091 ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS;
2092 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
2094 ether_ifattach(ifp, np->mac);
2095 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2096 netfront_carrier_off(np);
2101 gnttab_free_grant_references(np->gref_tx_head);
2107 * Handle the change of state of the backend to Closing. We must delete our
2108 * device-layer structures now, to ensure that writes are flushed through to
2109 * the backend. Once is this done, we can switch to Closed in
2114 netfront_closing(device_t dev)
2117 struct netfront_info *info = dev->dev_driver_data;
2119 DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2123 xenbus_switch_state(dev, XenbusStateClosed);
2128 netfront_detach(device_t dev)
2130 struct netfront_info *info = device_get_softc(dev);
2132 DPRINTK("%s\n", xenbus_get_node(dev));
2140 netif_free(struct netfront_info *info)
2145 callout_drain(&info->xn_stat_ch);
2146 netif_disconnect_backend(info);
2147 if (info->xn_ifp != NULL) {
2148 ether_ifdetach(info->xn_ifp);
2149 if_free(info->xn_ifp);
2150 info->xn_ifp = NULL;
2152 ifmedia_removeall(&info->sc_media);
2156 netif_disconnect_backend(struct netfront_info *info)
2160 netfront_carrier_off(info);
2164 free_ring(&info->tx_ring_ref, &info->tx.sring);
2165 free_ring(&info->rx_ring_ref, &info->rx.sring);
2167 xen_intr_unbind(&info->xen_intr_handle);
2171 free_ring(int *ref, void *ring_ptr_ref)
2173 void **ring_ptr_ptr = ring_ptr_ref;
2175 if (*ref != GRANT_REF_INVALID) {
2176 /* This API frees the associated storage. */
2177 gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
2178 *ref = GRANT_REF_INVALID;
2180 *ring_ptr_ptr = NULL;
2184 xn_ifmedia_upd(struct ifnet *ifp)
2190 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2192 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2193 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2196 /* ** Driver registration ** */
2197 static device_method_t netfront_methods[] = {
2198 /* Device interface */
2199 DEVMETHOD(device_probe, netfront_probe),
2200 DEVMETHOD(device_attach, netfront_attach),
2201 DEVMETHOD(device_detach, netfront_detach),
2202 DEVMETHOD(device_shutdown, bus_generic_shutdown),
2203 DEVMETHOD(device_suspend, netfront_suspend),
2204 DEVMETHOD(device_resume, netfront_resume),
2206 /* Xenbus interface */
2207 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2212 static driver_t netfront_driver = {
2215 sizeof(struct netfront_info),
2217 devclass_t netfront_devclass;
2219 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,