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("$FreeBSD$");
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sockio.h>
36 #include <sys/malloc.h>
37 #include <sys/module.h>
38 #include <sys/kernel.h>
39 #include <sys/socket.h>
40 #include <sys/sysctl.h>
41 #include <sys/queue.h>
46 #include <net/if_arp.h>
47 #include <net/ethernet.h>
48 #include <net/if_dl.h>
49 #include <net/if_media.h>
53 #include <net/if_types.h>
56 #include <netinet/in_systm.h>
57 #include <netinet/in.h>
58 #include <netinet/ip.h>
59 #include <netinet/if_ether.h>
60 #if __FreeBSD_version >= 700000
61 #include <netinet/tcp.h>
62 #include <netinet/tcp_lro.h>
68 #include <machine/clock.h> /* for DELAY */
69 #include <machine/bus.h>
70 #include <machine/resource.h>
71 #include <machine/frame.h>
72 #include <machine/vmparam.h>
77 #include <machine/intr_machdep.h>
79 #include <xen/xen-os.h>
80 #include <xen/hypervisor.h>
81 #include <xen/xen_intr.h>
82 #include <xen/gnttab.h>
83 #include <xen/interface/memory.h>
84 #include <xen/interface/io/netif.h>
85 #include <xen/xenbus/xenbusvar.h>
87 #include <machine/xen/xenvar.h>
89 #include <dev/xen/netfront/mbufq.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)
136 #define NF_TSO_MAXBURST ((IP_MAXPACKET / PAGE_SIZE) * MCLBYTES)
138 #define RX_COPY_THRESHOLD 256
140 #define net_ratelimit() 0
142 struct netfront_info;
143 struct netfront_rx_info;
145 static void xn_txeof(struct netfront_info *);
146 static void xn_rxeof(struct netfront_info *);
147 static void network_alloc_rx_buffers(struct netfront_info *);
149 static void xn_tick_locked(struct netfront_info *);
150 static void xn_tick(void *);
152 static void xn_intr(void *);
153 static inline int xn_count_frags(struct mbuf *m);
154 static int xn_assemble_tx_request(struct netfront_info *sc,
155 struct mbuf *m_head);
156 static void xn_start_locked(struct ifnet *);
157 static void xn_start(struct ifnet *);
158 static int xn_ioctl(struct ifnet *, u_long, caddr_t);
159 static void xn_ifinit_locked(struct netfront_info *);
160 static void xn_ifinit(void *);
161 static void xn_stop(struct netfront_info *);
162 static void xn_query_features(struct netfront_info *np);
163 static int xn_configure_features(struct netfront_info *np);
165 static void xn_watchdog(struct ifnet *);
168 static void show_device(struct netfront_info *sc);
170 static void netfront_closing(device_t dev);
172 static void netif_free(struct netfront_info *info);
173 static int netfront_detach(device_t dev);
175 static int talk_to_backend(device_t dev, struct netfront_info *info);
176 static int create_netdev(device_t dev);
177 static void netif_disconnect_backend(struct netfront_info *info);
178 static int setup_device(device_t dev, struct netfront_info *info);
179 static void free_ring(int *ref, void *ring_ptr_ref);
181 static int xn_ifmedia_upd(struct ifnet *ifp);
182 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
184 /* Xenolinux helper functions */
185 int network_connect(struct netfront_info *);
187 static void xn_free_rx_ring(struct netfront_info *);
189 static void xn_free_tx_ring(struct netfront_info *);
191 static int xennet_get_responses(struct netfront_info *np,
192 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
193 struct mbuf **list, int *pages_flipped_p);
195 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
197 #define INVALID_P2M_ENTRY (~0UL)
200 * Mbuf pointers. We need these to keep track of the virtual addresses
201 * of our mbuf chains since we can only convert from virtual to physical,
202 * not the other way around. The size must track the free index arrays.
204 struct xn_chain_data {
205 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1];
207 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1];
210 struct net_device_stats
212 u_long rx_packets; /* total packets received */
213 u_long tx_packets; /* total packets transmitted */
214 u_long rx_bytes; /* total bytes received */
215 u_long tx_bytes; /* total bytes transmitted */
216 u_long rx_errors; /* bad packets received */
217 u_long tx_errors; /* packet transmit problems */
218 u_long rx_dropped; /* no space in linux buffers */
219 u_long tx_dropped; /* no space available in linux */
220 u_long multicast; /* multicast packets received */
223 /* detailed rx_errors: */
224 u_long rx_length_errors;
225 u_long rx_over_errors; /* receiver ring buff overflow */
226 u_long rx_crc_errors; /* recved pkt with crc error */
227 u_long rx_frame_errors; /* recv'd frame alignment error */
228 u_long rx_fifo_errors; /* recv'r fifo overrun */
229 u_long rx_missed_errors; /* receiver missed packet */
231 /* detailed tx_errors */
232 u_long tx_aborted_errors;
233 u_long tx_carrier_errors;
234 u_long tx_fifo_errors;
235 u_long tx_heartbeat_errors;
236 u_long tx_window_errors;
239 u_long rx_compressed;
240 u_long tx_compressed;
243 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;
259 xen_intr_handle_t xen_intr_handle;
260 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[]. */
328 add_id_to_freelist(struct mbuf **list, uintptr_t id)
331 ("%s: the head item (0) must always be free.", __func__));
333 list[0] = (struct mbuf *)id;
336 static inline unsigned short
337 get_id_from_freelist(struct mbuf **list)
341 id = (uintptr_t)list[0];
343 ("%s: the head item (0) must always remain free.", __func__));
349 xennet_rxidx(RING_IDX idx)
351 return idx & (NET_RX_RING_SIZE - 1);
354 static inline struct mbuf *
355 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
357 int i = xennet_rxidx(ri);
361 np->rx_mbufs[i] = NULL;
365 static inline grant_ref_t
366 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
368 int i = xennet_rxidx(ri);
369 grant_ref_t ref = np->grant_rx_ref[i];
370 KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
371 np->grant_rx_ref[i] = GRANT_REF_INVALID;
375 #define IPRINTK(fmt, args...) \
376 printf("[XEN] " fmt, ##args)
378 #define WPRINTK(fmt, args...) \
379 printf("[XEN] " fmt, ##args)
381 #define WPRINTK(fmt, args...)
384 #define DPRINTK(fmt, args...) \
385 printf("[XEN] %s: " fmt, __func__, ##args)
387 #define DPRINTK(fmt, args...)
391 * Read the 'mac' node at the given device's node in the store, and parse that
392 * as colon-separated octets, placing result the given mac array. mac must be
393 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
394 * Return 0 on success, or errno on error.
397 xen_net_read_mac(device_t dev, uint8_t mac[])
400 char *s, *e, *macstr;
403 path = xenbus_get_node(dev);
404 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
405 if (error == ENOENT) {
407 * Deal with missing mac XenStore nodes on devices with
408 * HVM emulation (the 'ioemu' configuration attribute)
411 * The HVM emulator may execute in a stub device model
412 * domain which lacks the permission, only given to Dom0,
413 * to update the guest's XenStore tree. For this reason,
414 * the HVM emulator doesn't even attempt to write the
415 * front-side mac node, even when operating in Dom0.
416 * However, there should always be a mac listed in the
417 * backend tree. Fallback to this version if our query
418 * of the front side XenStore location doesn't find
421 path = xenbus_get_otherend_path(dev);
422 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
425 xenbus_dev_fatal(dev, error, "parsing %s/mac", path);
430 for (i = 0; i < ETHER_ADDR_LEN; i++) {
431 mac[i] = strtoul(s, &e, 16);
432 if (s == e || (e[0] != ':' && e[0] != 0)) {
433 free(macstr, M_XENBUS);
438 free(macstr, M_XENBUS);
443 * Entry point to this code when a new device is created. Allocate the basic
444 * structures and the ring buffers for communication with the backend, and
445 * inform the backend of the appropriate details for those. Switch to
449 netfront_probe(device_t dev)
452 if (!strcmp(xenbus_get_type(dev), "vif")) {
453 device_set_desc(dev, "Virtual Network Interface");
461 netfront_attach(device_t dev)
465 err = create_netdev(dev);
467 xenbus_dev_fatal(dev, err, "creating netdev");
471 #if __FreeBSD_version >= 700000
472 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
473 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
474 OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
475 &xn_enable_lro, 0, "Large Receive Offload");
482 netfront_suspend(device_t dev)
484 struct netfront_info *info = device_get_softc(dev);
488 netfront_carrier_off(info);
495 * We are reconnecting to the backend, due to a suspend/resume, or a backend
496 * driver restart. We tear down our netif structure and recreate it, but
497 * leave the device-layer structures intact so that this is transparent to the
498 * rest of the kernel.
501 netfront_resume(device_t dev)
503 struct netfront_info *info = device_get_softc(dev);
505 netif_disconnect_backend(info);
509 /* Common code used when first setting up, and when resuming. */
511 talk_to_backend(device_t dev, struct netfront_info *info)
514 struct xs_transaction xst;
515 const char *node = xenbus_get_node(dev);
518 err = xen_net_read_mac(dev, info->mac);
520 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
524 /* Create shared ring, alloc event channel. */
525 err = setup_device(dev, info);
530 err = xs_transaction_start(&xst);
532 xenbus_dev_fatal(dev, err, "starting transaction");
535 err = xs_printf(xst, node, "tx-ring-ref","%u",
538 message = "writing tx ring-ref";
539 goto abort_transaction;
541 err = xs_printf(xst, node, "rx-ring-ref","%u",
544 message = "writing rx ring-ref";
545 goto abort_transaction;
547 err = xs_printf(xst, node,
548 "event-channel", "%u",
549 xen_intr_port(info->xen_intr_handle));
551 message = "writing event-channel";
552 goto abort_transaction;
554 err = xs_printf(xst, node, "request-rx-copy", "%u",
555 info->copying_receiver);
557 message = "writing request-rx-copy";
558 goto abort_transaction;
560 err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
562 message = "writing feature-rx-notify";
563 goto abort_transaction;
565 err = xs_printf(xst, node, "feature-sg", "%d", 1);
567 message = "writing feature-sg";
568 goto abort_transaction;
570 #if __FreeBSD_version >= 700000
571 err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
573 message = "writing feature-gso-tcpv4";
574 goto abort_transaction;
578 err = xs_transaction_end(xst, 0);
582 xenbus_dev_fatal(dev, err, "completing transaction");
589 xs_transaction_end(xst, 1);
590 xenbus_dev_fatal(dev, err, "%s", message);
598 setup_device(device_t dev, struct netfront_info *info)
600 netif_tx_sring_t *txs;
601 netif_rx_sring_t *rxs;
607 info->tx_ring_ref = GRANT_REF_INVALID;
608 info->rx_ring_ref = GRANT_REF_INVALID;
609 info->rx.sring = NULL;
610 info->tx.sring = NULL;
612 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
615 xenbus_dev_fatal(dev, error, "allocating tx ring page");
618 SHARED_RING_INIT(txs);
619 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
620 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
624 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
627 xenbus_dev_fatal(dev, error, "allocating rx ring page");
630 SHARED_RING_INIT(rxs);
631 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
633 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
637 error = xen_intr_alloc_and_bind_local_port(dev,
638 xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info,
639 INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle);
642 xenbus_dev_fatal(dev, error,
643 "xen_intr_alloc_and_bind_local_port failed");
658 * If this interface has an ipv4 address, send an arp for it. This
659 * helps to get the network going again after migrating hosts.
662 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
668 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
669 if (ifa->ifa_addr->sa_family == AF_INET) {
670 arp_ifinit(ifp, ifa);
677 * Callback received when the backend's state changes.
680 netfront_backend_changed(device_t dev, XenbusState newstate)
682 struct netfront_info *sc = device_get_softc(dev);
684 DPRINTK("newstate=%d\n", newstate);
687 case XenbusStateInitialising:
688 case XenbusStateInitialised:
689 case XenbusStateConnected:
690 case XenbusStateUnknown:
691 case XenbusStateClosed:
692 case XenbusStateReconfigured:
693 case XenbusStateReconfiguring:
695 case XenbusStateInitWait:
696 if (xenbus_get_state(dev) != XenbusStateInitialising)
698 if (network_connect(sc) != 0)
700 xenbus_set_state(dev, XenbusStateConnected);
702 netfront_send_fake_arp(dev, sc);
705 case XenbusStateClosing:
706 xenbus_set_state(dev, XenbusStateClosed);
712 xn_free_rx_ring(struct netfront_info *sc)
717 for (i = 0; i < NET_RX_RING_SIZE; i++) {
718 if (sc->xn_cdata.rx_mbufs[i] != NULL) {
719 m_freem(sc->rx_mbufs[i]);
720 sc->rx_mbufs[i] = NULL;
725 sc->xn_rx_if->req_prod = 0;
726 sc->xn_rx_if->event = sc->rx.rsp_cons ;
731 xn_free_tx_ring(struct netfront_info *sc)
736 for (i = 0; i < NET_TX_RING_SIZE; i++) {
737 if (sc->tx_mbufs[i] != NULL) {
738 m_freem(sc->tx_mbufs[i]);
739 sc->xn_cdata.xn_tx_chain[i] = NULL;
748 * \brief Verify that there is sufficient space in the Tx ring
749 * buffer for a maximally sized request to be enqueued.
751 * A transmit request requires a transmit descriptor for each packet
752 * fragment, plus up to 2 entries for "options" (e.g. TSO).
755 xn_tx_slot_available(struct netfront_info *np)
757 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
761 netif_release_tx_bufs(struct netfront_info *np)
765 for (i = 1; i <= NET_TX_RING_SIZE; i++) {
771 * We assume that no kernel addresses are
772 * less than NET_TX_RING_SIZE. Any entry
773 * in the table that is below this number
774 * must be an index from free-list tracking.
776 if (((uintptr_t)m) <= NET_TX_RING_SIZE)
778 gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
779 gnttab_release_grant_reference(&np->gref_tx_head,
780 np->grant_tx_ref[i]);
781 np->grant_tx_ref[i] = GRANT_REF_INVALID;
782 add_id_to_freelist(np->tx_mbufs, i);
783 np->xn_cdata.xn_tx_chain_cnt--;
784 if (np->xn_cdata.xn_tx_chain_cnt < 0) {
785 panic("%s: tx_chain_cnt must be >= 0", __func__);
792 network_alloc_rx_buffers(struct netfront_info *sc)
794 int otherend_id = xenbus_get_otherend_id(sc->xbdev);
797 int i, batch_target, notify;
799 struct xen_memory_reservation reservation;
802 netif_rx_request_t *req;
806 req_prod = sc->rx.req_prod_pvt;
808 if (__predict_false(sc->carrier == 0))
812 * Allocate mbufs greedily, even though we batch updates to the
813 * receive ring. This creates a less bursty demand on the memory
814 * allocator, and so should reduce the chance of failed allocation
815 * requests both for ourself and for other kernel subsystems.
817 * Here we attempt to maintain rx_target buffers in flight, counting
818 * buffers that we have yet to process in the receive ring.
820 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
821 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
822 MGETHDR(m_new, M_NOWAIT, MT_DATA);
824 printf("%s: MGETHDR failed\n", __func__);
828 m_cljget(m_new, M_NOWAIT, MJUMPAGESIZE);
829 if ((m_new->m_flags & M_EXT) == 0) {
830 printf("%s: m_cljget failed\n", __func__);
841 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
843 /* queue the mbufs allocated */
844 mbufq_tail(&sc->xn_rx_batch, m_new);
848 * If we've allocated at least half of our target number of entries,
849 * submit them to the backend - we have enough to make the overhead
850 * of submission worthwhile. Otherwise wait for more mbufs and
851 * request entries to become available.
853 if (i < (sc->rx_target/2)) {
854 if (req_prod >sc->rx.sring->req_prod)
860 * Double floating fill target if we risked having the backend
861 * run out of empty buffers for receive traffic. We define "running
862 * low" as having less than a fourth of our target buffers free
863 * at the time we refilled the queue.
865 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
867 if (sc->rx_target > sc->rx_max_target)
868 sc->rx_target = sc->rx_max_target;
872 for (nr_flips = i = 0; ; i++) {
873 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
876 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
877 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
879 id = xennet_rxidx(req_prod + i);
881 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
882 sc->rx_mbufs[id] = m_new;
884 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
885 KASSERT(ref != GNTTAB_LIST_END,
886 ("reserved grant references exhuasted"));
887 sc->grant_rx_ref[id] = ref;
889 vaddr = mtod(m_new, vm_offset_t);
890 pfn = vtophys(vaddr) >> PAGE_SHIFT;
891 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
893 if (sc->copying_receiver == 0) {
894 gnttab_grant_foreign_transfer_ref(ref,
896 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
897 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
898 /* Remove this page before passing
901 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
902 MULTI_update_va_mapping(&sc->rx_mcl[i],
907 gnttab_grant_foreign_access_ref(ref,
914 sc->rx_pfn_array[i] =
915 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
918 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
919 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
921 * We may have allocated buffers which have entries outstanding
922 * in the page * update queue -- make sure we flush those first!
927 /* Tell the ballon driver what is going on. */
928 balloon_update_driver_allowance(i);
930 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
931 reservation.nr_extents = i;
932 reservation.extent_order = 0;
933 reservation.address_bits = 0;
934 reservation.domid = DOMID_SELF;
936 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
937 /* After all PTEs have been zapped, flush the TLB. */
938 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
939 UVMF_TLB_FLUSH|UVMF_ALL;
941 /* Give away a batch of pages. */
942 sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
943 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
944 sc->rx_mcl[i].args[1] = (u_long)&reservation;
945 /* Zap PTEs and give away pages in one big multicall. */
946 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
948 if (__predict_false(sc->rx_mcl[i].result != i ||
949 HYPERVISOR_memory_op(XENMEM_decrease_reservation,
951 panic("%s: unable to reduce memory "
952 "reservation\n", __func__);
958 /* Above is a suitable barrier to ensure backend will see requests. */
959 sc->rx.req_prod_pvt = req_prod + i;
961 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
963 xen_intr_signal(sc->xen_intr_handle);
967 xn_rxeof(struct netfront_info *np)
970 #if __FreeBSD_version >= 700000
971 struct lro_ctrl *lro = &np->xn_lro;
972 struct lro_entry *queued;
974 struct netfront_rx_info rinfo;
975 struct netif_rx_response *rx = &rinfo.rx;
976 struct netif_extra_info *extras = rinfo.extras;
978 multicall_entry_t *mcl;
980 struct mbuf_head rxq, errq;
981 int err, pages_flipped = 0, work_to_do;
984 XN_RX_LOCK_ASSERT(np);
985 if (!netfront_carrier_ok(np))
993 rp = np->rx.sring->rsp_prod;
994 rmb(); /* Ensure we see queued responses up to 'rp'. */
998 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
999 memset(extras, 0, sizeof(rinfo.extras));
1002 err = xennet_get_responses(np, &rinfo, rp, &i, &m,
1005 if (__predict_false(err)) {
1007 mbufq_tail(&errq, m);
1008 np->stats.rx_errors++;
1012 m->m_pkthdr.rcvif = ifp;
1013 if ( rx->flags & NETRXF_data_validated ) {
1014 /* Tell the stack the checksums are okay */
1016 * XXX this isn't necessarily the case - need to add
1020 m->m_pkthdr.csum_flags |=
1021 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
1023 m->m_pkthdr.csum_data = 0xffff;
1026 np->stats.rx_packets++;
1027 np->stats.rx_bytes += m->m_pkthdr.len;
1029 mbufq_tail(&rxq, m);
1030 np->rx.rsp_cons = i;
1033 if (pages_flipped) {
1034 /* Some pages are no longer absent... */
1036 balloon_update_driver_allowance(-pages_flipped);
1038 /* Do all the remapping work, and M->P updates, in one big
1041 if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1042 mcl = np->rx_mcl + pages_flipped;
1043 mcl->op = __HYPERVISOR_mmu_update;
1044 mcl->args[0] = (u_long)np->rx_mmu;
1045 mcl->args[1] = pages_flipped;
1047 mcl->args[3] = DOMID_SELF;
1048 (void)HYPERVISOR_multicall(np->rx_mcl,
1053 while ((m = mbufq_dequeue(&errq)))
1057 * Process all the mbufs after the remapping is complete.
1058 * Break the mbuf chain first though.
1060 while ((m = mbufq_dequeue(&rxq)) != NULL) {
1064 * Do we really need to drop the rx lock?
1067 #if __FreeBSD_version >= 700000
1068 /* Use LRO if possible */
1069 if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1070 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1072 * If LRO fails, pass up to the stack
1075 (*ifp->if_input)(ifp, m);
1078 (*ifp->if_input)(ifp, m);
1083 np->rx.rsp_cons = i;
1085 #if __FreeBSD_version >= 700000
1087 * Flush any outstanding LRO work
1089 while (!SLIST_EMPTY(&lro->lro_active)) {
1090 queued = SLIST_FIRST(&lro->lro_active);
1091 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1092 tcp_lro_flush(lro, queued);
1097 /* If we get a callback with very few responses, reduce fill target. */
1098 /* NB. Note exponential increase, linear decrease. */
1099 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1100 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1101 np->rx_target = np->rx_min_target;
1104 network_alloc_rx_buffers(np);
1106 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1107 } while (work_to_do);
1111 xn_txeof(struct netfront_info *np)
1116 netif_tx_response_t *txr;
1119 XN_TX_LOCK_ASSERT(np);
1121 if (!netfront_carrier_ok(np))
1127 prod = np->tx.sring->rsp_prod;
1128 rmb(); /* Ensure we see responses up to 'rp'. */
1130 for (i = np->tx.rsp_cons; i != prod; i++) {
1131 txr = RING_GET_RESPONSE(&np->tx, i);
1132 if (txr->status == NETIF_RSP_NULL)
1135 if (txr->status != NETIF_RSP_OKAY) {
1136 printf("%s: WARNING: response is %d!\n",
1137 __func__, txr->status);
1140 m = np->tx_mbufs[id];
1141 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1142 KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1143 ("mbuf already on the free list, but we're "
1144 "trying to free it again!"));
1148 * Increment packet count if this is the last
1149 * mbuf of the chain.
1153 if (__predict_false(gnttab_query_foreign_access(
1154 np->grant_tx_ref[id]) != 0)) {
1155 panic("%s: grant id %u still in use by the "
1156 "backend", __func__, id);
1158 gnttab_end_foreign_access_ref(
1159 np->grant_tx_ref[id]);
1160 gnttab_release_grant_reference(
1161 &np->gref_tx_head, np->grant_tx_ref[id]);
1162 np->grant_tx_ref[id] = GRANT_REF_INVALID;
1164 np->tx_mbufs[id] = NULL;
1165 add_id_to_freelist(np->tx_mbufs, id);
1166 np->xn_cdata.xn_tx_chain_cnt--;
1168 /* Only mark the queue active if we've freed up at least one slot to try */
1169 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1171 np->tx.rsp_cons = prod;
1174 * Set a new event, then check for race with update of
1175 * tx_cons. Note that it is essential to schedule a
1176 * callback, no matter how few buffers are pending. Even if
1177 * there is space in the transmit ring, higher layers may
1178 * be blocked because too much data is outstanding: in such
1179 * cases notification from Xen is likely to be the only kick
1182 np->tx.sring->rsp_event =
1183 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1186 } while (prod != np->tx.sring->rsp_prod);
1189 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1192 if (np->user_state == UST_OPEN)
1193 netif_wake_queue(dev);
1201 struct netfront_info *np = xsc;
1202 struct ifnet *ifp = np->xn_ifp;
1205 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1206 likely(netfront_carrier_ok(np)) &&
1207 ifp->if_drv_flags & IFF_DRV_RUNNING))
1210 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1220 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1221 !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1226 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1229 int new = xennet_rxidx(np->rx.req_prod_pvt);
1231 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1232 np->rx_mbufs[new] = m;
1233 np->grant_rx_ref[new] = ref;
1234 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1235 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1236 np->rx.req_prod_pvt++;
1240 xennet_get_extras(struct netfront_info *np,
1241 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1243 struct netif_extra_info *extra;
1251 if (__predict_false(*cons + 1 == rp)) {
1253 if (net_ratelimit())
1254 WPRINTK("Missing extra info\n");
1260 extra = (struct netif_extra_info *)
1261 RING_GET_RESPONSE(&np->rx, ++(*cons));
1263 if (__predict_false(!extra->type ||
1264 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1266 if (net_ratelimit())
1267 WPRINTK("Invalid extra type: %d\n",
1272 memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1275 m = xennet_get_rx_mbuf(np, *cons);
1276 ref = xennet_get_rx_ref(np, *cons);
1277 xennet_move_rx_slot(np, m, ref);
1278 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1284 xennet_get_responses(struct netfront_info *np,
1285 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1287 int *pages_flipped_p)
1289 int pages_flipped = *pages_flipped_p;
1290 struct mmu_update *mmu;
1291 struct multicall_entry *mcl;
1292 struct netif_rx_response *rx = &rinfo->rx;
1293 struct netif_extra_info *extras = rinfo->extras;
1294 struct mbuf *m, *m0, *m_prev;
1295 grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1296 RING_IDX ref_cons = *cons;
1301 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1303 if (rx->flags & NETRXF_extra_info) {
1304 err = xennet_get_extras(np, extras, rp, cons);
1308 m0->m_pkthdr.len = 0;
1316 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1317 rx->status, rx->offset, frags);
1319 if (__predict_false(rx->status < 0 ||
1320 rx->offset + rx->status > PAGE_SIZE)) {
1323 if (net_ratelimit())
1324 WPRINTK("rx->offset: %x, size: %u\n",
1325 rx->offset, rx->status);
1327 xennet_move_rx_slot(np, m, ref);
1332 goto next_skip_queue;
1336 * This definitely indicates a bug, either in this driver or in
1337 * the backend driver. In future this should flag the bad
1338 * situation to the system controller to reboot the backed.
1340 if (ref == GRANT_REF_INVALID) {
1343 if (net_ratelimit())
1344 WPRINTK("Bad rx response id %d.\n", rx->id);
1346 printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1351 if (!np->copying_receiver) {
1352 /* Memory pressure, insufficient buffer
1355 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1356 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1357 rx->id, rx->status);
1358 xennet_move_rx_slot(np, m, ref);
1363 if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1364 /* Remap the page. */
1365 void *vaddr = mtod(m, void *);
1368 mcl = np->rx_mcl + pages_flipped;
1369 mmu = np->rx_mmu + pages_flipped;
1371 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1372 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1373 PG_V | PG_M | PG_A, 0);
1374 pfn = (uintptr_t)m->m_ext.ext_arg1;
1375 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1376 MMU_MACHPHYS_UPDATE;
1379 set_phys_to_machine(pfn, mfn);
1383 ret = gnttab_end_foreign_access_ref(ref);
1384 KASSERT(ret, ("ret != 0"));
1387 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1393 m->m_len = rx->status;
1394 m->m_data += rx->offset;
1395 m0->m_pkthdr.len += rx->status;
1398 if (!(rx->flags & NETRXF_more_data))
1401 if (*cons + frags == rp) {
1402 if (net_ratelimit())
1403 WPRINTK("Need more frags\n");
1405 printf("%s: cons %u frags %u rp %u, not enough frags\n",
1406 __func__, *cons, frags, rp);
1410 * Note that m can be NULL, if rx->status < 0 or if
1411 * rx->offset + rx->status > PAGE_SIZE above.
1415 rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1416 m = xennet_get_rx_mbuf(np, *cons + frags);
1419 * m_prev == NULL can happen if rx->status < 0 or if
1420 * rx->offset + * rx->status > PAGE_SIZE above.
1426 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1427 * rx->status > PAGE_SIZE above.
1432 ref = xennet_get_rx_ref(np, *cons + frags);
1433 ref_cons = *cons + frags;
1438 *pages_flipped_p = pages_flipped;
1444 xn_tick_locked(struct netfront_info *sc)
1446 XN_RX_LOCK_ASSERT(sc);
1447 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1449 /* XXX placeholder for printing debug information */
1455 struct netfront_info *sc;
1464 * \brief Count the number of fragments in an mbuf chain.
1466 * Surprisingly, there isn't an M* macro for this.
1469 xn_count_frags(struct mbuf *m)
1473 for (nfrags = 0; m != NULL; m = m->m_next)
1480 * Given an mbuf chain, make sure we have enough room and then push
1481 * it onto the transmit ring.
1484 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1489 netif_extra_info_t *extra;
1495 * Defragment the mbuf if necessary.
1497 nfrags = xn_count_frags(m_head);
1500 * Check to see whether this request is longer than netback
1501 * can handle, and try to defrag it.
1504 * It is a bit lame, but the netback driver in Linux can't
1505 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1506 * the Linux network stack.
1508 if (nfrags > sc->maxfrags) {
1509 m = m_defrag(m_head, M_NOWAIT);
1512 * Defrag failed, so free the mbuf and
1513 * therefore drop the packet.
1521 /* Determine how many fragments now exist */
1522 nfrags = xn_count_frags(m_head);
1525 * Check to see whether the defragmented packet has too many
1526 * segments for the Linux netback driver.
1529 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1530 * of mbufs longer than Linux can handle. Make sure we don't
1531 * pass a too-long chain over to the other side by dropping the
1532 * packet. It doesn't look like there is currently a way to
1533 * tell the TCP stack to generate a shorter chain of packets.
1535 if (nfrags > MAX_TX_REQ_FRAGS) {
1537 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1538 "won't be able to handle it, dropping\n",
1539 __func__, nfrags, MAX_TX_REQ_FRAGS);
1546 * This check should be redundant. We've already verified that we
1547 * have enough slots in the ring to handle a packet of maximum
1548 * size, and that our packet is less than the maximum size. Keep
1549 * it in here as an assert for now just to make certain that
1550 * xn_tx_chain_cnt is accurate.
1552 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1553 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1554 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1555 (int) nfrags, (int) NET_TX_RING_SIZE));
1558 * Start packing the mbufs in this chain into
1559 * the fragment pointers. Stop when we run out
1560 * of fragments or hit the end of the mbuf chain.
1564 otherend_id = xenbus_get_otherend_id(sc->xbdev);
1565 for (m = m_head; m; m = m->m_next) {
1566 netif_tx_request_t *tx;
1569 u_long mfn; /* XXX Wrong type? */
1571 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1572 id = get_id_from_freelist(sc->tx_mbufs);
1574 panic("%s: was allocated the freelist head!\n",
1576 sc->xn_cdata.xn_tx_chain_cnt++;
1577 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1578 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1580 sc->tx_mbufs[id] = m;
1582 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1583 KASSERT((short)ref >= 0, ("Negative ref"));
1584 mfn = virt_to_mfn(mtod(m, vm_offset_t));
1585 gnttab_grant_foreign_access_ref(ref, otherend_id,
1586 mfn, GNTMAP_readonly);
1587 tx->gref = sc->grant_tx_ref[id] = ref;
1588 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1592 * The first fragment has the entire packet
1593 * size, subsequent fragments have just the
1594 * fragment size. The backend works out the
1595 * true size of the first fragment by
1596 * subtracting the sizes of the other
1599 tx->size = m->m_pkthdr.len;
1602 * The first fragment contains the checksum flags
1603 * and is optionally followed by extra data for
1607 * CSUM_TSO requires checksum offloading.
1608 * Some versions of FreeBSD fail to
1609 * set CSUM_TCP in the CSUM_TSO case,
1610 * so we have to test for CSUM_TSO
1613 if (m->m_pkthdr.csum_flags
1614 & (CSUM_DELAY_DATA | CSUM_TSO)) {
1615 tx->flags |= (NETTXF_csum_blank
1616 | NETTXF_data_validated);
1618 #if __FreeBSD_version >= 700000
1619 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1620 struct netif_extra_info *gso =
1621 (struct netif_extra_info *)
1622 RING_GET_REQUEST(&sc->tx,
1623 ++sc->tx.req_prod_pvt);
1625 tx->flags |= NETTXF_extra_info;
1627 gso->u.gso.size = m->m_pkthdr.tso_segsz;
1629 XEN_NETIF_GSO_TYPE_TCPV4;
1631 gso->u.gso.features = 0;
1633 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1638 tx->size = m->m_len;
1641 tx->flags |= NETTXF_more_data;
1643 sc->tx.req_prod_pvt++;
1645 BPF_MTAP(ifp, m_head);
1647 sc->stats.tx_bytes += m_head->m_pkthdr.len;
1648 sc->stats.tx_packets++;
1654 xn_start_locked(struct ifnet *ifp)
1656 struct netfront_info *sc;
1657 struct mbuf *m_head;
1662 if (!netfront_carrier_ok(sc))
1666 * While we have enough transmit slots available for at least one
1667 * maximum-sized packet, pull mbufs off the queue and put them on
1668 * the transmit ring.
1670 while (xn_tx_slot_available(sc)) {
1671 IF_DEQUEUE(&ifp->if_snd, m_head);
1675 if (xn_assemble_tx_request(sc, m_head) != 0)
1679 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1681 xen_intr_signal(sc->xen_intr_handle);
1683 if (RING_FULL(&sc->tx)) {
1686 netif_stop_queue(dev);
1692 xn_start(struct ifnet *ifp)
1694 struct netfront_info *sc;
1697 xn_start_locked(ifp);
1701 /* equivalent of network_open() in Linux */
1703 xn_ifinit_locked(struct netfront_info *sc)
1711 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1716 network_alloc_rx_buffers(sc);
1717 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1719 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1720 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1721 if_link_state_change(ifp, LINK_STATE_UP);
1723 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1727 xn_ifinit(void *xsc)
1729 struct netfront_info *sc = xsc;
1732 xn_ifinit_locked(sc);
1737 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1739 struct netfront_info *sc = ifp->if_softc;
1740 struct ifreq *ifr = (struct ifreq *) data;
1742 struct ifaddr *ifa = (struct ifaddr *)data;
1745 int mask, error = 0;
1751 if (ifa->ifa_addr->sa_family == AF_INET) {
1752 ifp->if_flags |= IFF_UP;
1753 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1754 xn_ifinit_locked(sc);
1755 arp_ifinit(ifp, ifa);
1760 error = ether_ioctl(ifp, cmd, data);
1766 /* XXX can we alter the MTU on a VN ?*/
1768 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1773 ifp->if_mtu = ifr->ifr_mtu;
1774 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1780 if (ifp->if_flags & IFF_UP) {
1782 * If only the state of the PROMISC flag changed,
1783 * then just use the 'set promisc mode' command
1784 * instead of reinitializing the entire NIC. Doing
1785 * a full re-init means reloading the firmware and
1786 * waiting for it to start up, which may take a
1790 /* No promiscuous mode with Xen */
1791 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1792 ifp->if_flags & IFF_PROMISC &&
1793 !(sc->xn_if_flags & IFF_PROMISC)) {
1794 XN_SETBIT(sc, XN_RX_MODE,
1795 XN_RXMODE_RX_PROMISC);
1796 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1797 !(ifp->if_flags & IFF_PROMISC) &&
1798 sc->xn_if_flags & IFF_PROMISC) {
1799 XN_CLRBIT(sc, XN_RX_MODE,
1800 XN_RXMODE_RX_PROMISC);
1803 xn_ifinit_locked(sc);
1805 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1809 sc->xn_if_flags = ifp->if_flags;
1814 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1815 if (mask & IFCAP_TXCSUM) {
1816 if (IFCAP_TXCSUM & ifp->if_capenable) {
1817 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1818 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1819 | CSUM_IP | CSUM_TSO);
1821 ifp->if_capenable |= IFCAP_TXCSUM;
1822 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1826 if (mask & IFCAP_RXCSUM) {
1827 ifp->if_capenable ^= IFCAP_RXCSUM;
1829 #if __FreeBSD_version >= 700000
1830 if (mask & IFCAP_TSO4) {
1831 if (IFCAP_TSO4 & ifp->if_capenable) {
1832 ifp->if_capenable &= ~IFCAP_TSO4;
1833 ifp->if_hwassist &= ~CSUM_TSO;
1834 } else if (IFCAP_TXCSUM & ifp->if_capenable) {
1835 ifp->if_capenable |= IFCAP_TSO4;
1836 ifp->if_hwassist |= CSUM_TSO;
1838 IPRINTK("Xen requires tx checksum offload"
1839 " be enabled to use TSO\n");
1843 if (mask & IFCAP_LRO) {
1844 ifp->if_capenable ^= IFCAP_LRO;
1853 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1863 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1866 error = ether_ioctl(ifp, cmd, data);
1873 xn_stop(struct netfront_info *sc)
1881 callout_stop(&sc->xn_stat_ch);
1883 xn_free_rx_ring(sc);
1884 xn_free_tx_ring(sc);
1886 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1887 if_link_state_change(ifp, LINK_STATE_DOWN);
1890 /* START of Xenolinux helper functions adapted to FreeBSD */
1892 network_connect(struct netfront_info *np)
1894 int i, requeue_idx, error;
1896 netif_rx_request_t *req;
1897 u_int feature_rx_copy, feature_rx_flip;
1899 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1900 "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1902 feature_rx_copy = 0;
1903 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1904 "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1906 feature_rx_flip = 1;
1909 * Copy packets on receive path if:
1910 * (a) This was requested by user, and the backend supports it; or
1911 * (b) Flipping was requested, but this is unsupported by the backend.
1913 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1914 (MODPARM_rx_flip && !feature_rx_flip));
1916 /* Recovery procedure: */
1917 error = talk_to_backend(np->xbdev, np);
1921 /* Step 1: Reinitialise variables. */
1922 xn_query_features(np);
1923 xn_configure_features(np);
1924 netif_release_tx_bufs(np);
1926 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1927 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1931 if (np->rx_mbufs[i] == NULL)
1934 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1935 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1937 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1938 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1940 if (!np->copying_receiver) {
1941 gnttab_grant_foreign_transfer_ref(ref,
1942 xenbus_get_otherend_id(np->xbdev),
1945 gnttab_grant_foreign_access_ref(ref,
1946 xenbus_get_otherend_id(np->xbdev),
1950 req->id = requeue_idx;
1955 np->rx.req_prod_pvt = requeue_idx;
1957 /* Step 3: All public and private state should now be sane. Get
1958 * ready to start sending and receiving packets and give the driver
1959 * domain a kick because we've probably just requeued some
1962 netfront_carrier_on(np);
1963 xen_intr_signal(np->xen_intr_handle);
1967 network_alloc_rx_buffers(np);
1973 show_device(struct netfront_info *sc)
1977 IPRINTK("<vif handle=%u %s(%s) evtchn=%u irq=%u tx=%p rx=%p>\n",
1979 be_state_name[sc->xn_backend_state],
1980 sc->xn_user_state ? "open" : "closed",
1986 IPRINTK("<vif NULL>\n");
1992 xn_query_features(struct netfront_info *np)
1996 device_printf(np->xbdev, "backend features:");
1998 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1999 "feature-sg", NULL, "%d", &val) < 0)
2004 np->maxfrags = MAX_TX_REQ_FRAGS;
2005 printf(" feature-sg");
2008 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
2009 "feature-gso-tcpv4", NULL, "%d", &val) < 0)
2012 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
2014 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
2015 printf(" feature-gso-tcp4");
2022 xn_configure_features(struct netfront_info *np)
2027 #if __FreeBSD_version >= 700000
2028 if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0)
2029 tcp_lro_free(&np->xn_lro);
2031 np->xn_ifp->if_capenable =
2032 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4);
2033 np->xn_ifp->if_hwassist &= ~CSUM_TSO;
2034 #if __FreeBSD_version >= 700000
2035 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) {
2036 err = tcp_lro_init(&np->xn_lro);
2038 device_printf(np->xbdev, "LRO initialization failed\n");
2040 np->xn_lro.ifp = np->xn_ifp;
2041 np->xn_ifp->if_capenable |= IFCAP_LRO;
2044 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) {
2045 np->xn_ifp->if_capenable |= IFCAP_TSO4;
2046 np->xn_ifp->if_hwassist |= CSUM_TSO;
2053 * Create a network device.
2054 * @param dev Newbus device representing this virtual NIC.
2057 create_netdev(device_t dev)
2060 struct netfront_info *np;
2064 np = device_get_softc(dev);
2068 XN_LOCK_INIT(np, xennetif);
2070 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
2071 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
2072 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
2074 np->rx_target = RX_MIN_TARGET;
2075 np->rx_min_target = RX_MIN_TARGET;
2076 np->rx_max_target = RX_MAX_TARGET;
2078 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2079 for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2080 np->tx_mbufs[i] = (void *) ((u_long) i+1);
2081 np->grant_tx_ref[i] = GRANT_REF_INVALID;
2083 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2085 for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2087 np->rx_mbufs[i] = NULL;
2088 np->grant_rx_ref[i] = GRANT_REF_INVALID;
2090 /* A grant for every tx ring slot */
2091 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2092 &np->gref_tx_head) != 0) {
2093 IPRINTK("#### netfront can't alloc tx grant refs\n");
2097 /* A grant for every rx ring slot */
2098 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2099 &np->gref_rx_head) != 0) {
2100 WPRINTK("#### netfront can't alloc rx grant refs\n");
2101 gnttab_free_grant_references(np->gref_tx_head);
2106 err = xen_net_read_mac(dev, np->mac);
2110 /* Set up ifnet structure */
2111 ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2113 if_initname(ifp, "xn", device_get_unit(dev));
2114 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2115 ifp->if_ioctl = xn_ioctl;
2116 ifp->if_output = ether_output;
2117 ifp->if_start = xn_start;
2119 ifp->if_watchdog = xn_watchdog;
2121 ifp->if_init = xn_ifinit;
2122 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2124 ifp->if_hwassist = XN_CSUM_FEATURES;
2125 ifp->if_capabilities = IFCAP_HWCSUM;
2126 ifp->if_hw_tsomax = NF_TSO_MAXBURST;
2128 ether_ifattach(ifp, np->mac);
2129 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2130 netfront_carrier_off(np);
2135 gnttab_free_grant_references(np->gref_tx_head);
2141 * Handle the change of state of the backend to Closing. We must delete our
2142 * device-layer structures now, to ensure that writes are flushed through to
2143 * the backend. Once is this done, we can switch to Closed in
2148 netfront_closing(device_t dev)
2151 struct netfront_info *info = dev->dev_driver_data;
2153 DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2157 xenbus_switch_state(dev, XenbusStateClosed);
2162 netfront_detach(device_t dev)
2164 struct netfront_info *info = device_get_softc(dev);
2166 DPRINTK("%s\n", xenbus_get_node(dev));
2174 netif_free(struct netfront_info *info)
2179 callout_drain(&info->xn_stat_ch);
2180 netif_disconnect_backend(info);
2181 if (info->xn_ifp != NULL) {
2182 ether_ifdetach(info->xn_ifp);
2183 if_free(info->xn_ifp);
2184 info->xn_ifp = NULL;
2186 ifmedia_removeall(&info->sc_media);
2190 netif_disconnect_backend(struct netfront_info *info)
2194 netfront_carrier_off(info);
2198 free_ring(&info->tx_ring_ref, &info->tx.sring);
2199 free_ring(&info->rx_ring_ref, &info->rx.sring);
2201 xen_intr_unbind(&info->xen_intr_handle);
2205 free_ring(int *ref, void *ring_ptr_ref)
2207 void **ring_ptr_ptr = ring_ptr_ref;
2209 if (*ref != GRANT_REF_INVALID) {
2210 /* This API frees the associated storage. */
2211 gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
2212 *ref = GRANT_REF_INVALID;
2214 *ring_ptr_ptr = NULL;
2218 xn_ifmedia_upd(struct ifnet *ifp)
2224 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2226 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2227 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2230 /* ** Driver registration ** */
2231 static device_method_t netfront_methods[] = {
2232 /* Device interface */
2233 DEVMETHOD(device_probe, netfront_probe),
2234 DEVMETHOD(device_attach, netfront_attach),
2235 DEVMETHOD(device_detach, netfront_detach),
2236 DEVMETHOD(device_shutdown, bus_generic_shutdown),
2237 DEVMETHOD(device_suspend, netfront_suspend),
2238 DEVMETHOD(device_resume, netfront_resume),
2240 /* Xenbus interface */
2241 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2246 static driver_t netfront_driver = {
2249 sizeof(struct netfront_info),
2251 devclass_t netfront_devclass;
2253 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,