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>
37 #include <sys/malloc.h>
38 #include <sys/module.h>
39 #include <sys/kernel.h>
40 #include <sys/socket.h>
41 #include <sys/sysctl.h>
42 #include <sys/queue.h>
47 #include <net/if_arp.h>
48 #include <net/ethernet.h>
49 #include <net/if_dl.h>
50 #include <net/if_media.h>
54 #include <net/if_types.h>
57 #include <netinet/in_systm.h>
58 #include <netinet/in.h>
59 #include <netinet/ip.h>
60 #include <netinet/if_ether.h>
61 #if __FreeBSD_version >= 700000
62 #include <netinet/tcp.h>
63 #include <netinet/tcp_lro.h>
69 #include <machine/clock.h> /* for DELAY */
70 #include <machine/bus.h>
71 #include <machine/resource.h>
72 #include <machine/frame.h>
73 #include <machine/vmparam.h>
78 #include <machine/intr_machdep.h>
80 #include <machine/xen/xen-os.h>
81 #include <machine/xen/xenfunc.h>
82 #include <machine/xen/xenvar.h>
83 #include <xen/hypervisor.h>
84 #include <xen/xen_intr.h>
85 #include <xen/evtchn.h>
86 #include <xen/gnttab.h>
87 #include <xen/interface/memory.h>
88 #include <xen/interface/io/netif.h>
89 #include <xen/xenbus/xenbusvar.h>
91 #include <dev/xen/netfront/mbufq.h>
93 #include "xenbus_if.h"
95 /* Features supported by all backends. TSO and LRO can be negotiated */
96 #define XN_CSUM_FEATURES (CSUM_TCP | CSUM_UDP)
98 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
99 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
101 #if __FreeBSD_version >= 700000
103 * Should the driver do LRO on the RX end
104 * this can be toggled on the fly, but the
105 * interface must be reset (down/up) for it
108 static int xn_enable_lro = 1;
109 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro);
118 static int MODPARM_rx_copy = 0;
119 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0);
120 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)");
121 static int MODPARM_rx_flip = 0;
122 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0);
123 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)");
125 static const int MODPARM_rx_copy = 1;
126 static const int MODPARM_rx_flip = 0;
130 * \brief The maximum allowed data fragments in a single transmit
133 * This limit is imposed by the backend driver. We assume here that
134 * we are dealing with a Linux driver domain and have set our limit
135 * to mirror the Linux MAX_SKB_FRAGS constant.
137 #define MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2)
138 #define NF_TSO_MAXBURST ((IP_MAXPACKET / PAGE_SIZE) * MCLBYTES)
140 #define RX_COPY_THRESHOLD 256
142 #define net_ratelimit() 0
144 struct netfront_info;
145 struct netfront_rx_info;
147 static void xn_txeof(struct netfront_info *);
148 static void xn_rxeof(struct netfront_info *);
149 static void network_alloc_rx_buffers(struct netfront_info *);
151 static void xn_tick_locked(struct netfront_info *);
152 static void xn_tick(void *);
154 static void xn_intr(void *);
155 static inline int xn_count_frags(struct mbuf *m);
156 static int xn_assemble_tx_request(struct netfront_info *sc,
157 struct mbuf *m_head);
158 static void xn_start_locked(struct ifnet *);
159 static void xn_start(struct ifnet *);
160 static int xn_ioctl(struct ifnet *, u_long, caddr_t);
161 static void xn_ifinit_locked(struct netfront_info *);
162 static void xn_ifinit(void *);
163 static void xn_stop(struct netfront_info *);
164 static void xn_query_features(struct netfront_info *np);
165 static int xn_configure_features(struct netfront_info *np);
167 static void xn_watchdog(struct ifnet *);
171 static void netfront_closing(device_t dev);
173 static void netif_free(struct netfront_info *info);
174 static int netfront_detach(device_t dev);
176 static int talk_to_backend(device_t dev, struct netfront_info *info);
177 static int create_netdev(device_t dev);
178 static void netif_disconnect_backend(struct netfront_info *info);
179 static int setup_device(device_t dev, struct netfront_info *info);
180 static void free_ring(int *ref, void *ring_ptr_ref);
182 static int xn_ifmedia_upd(struct ifnet *ifp);
183 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
185 /* Xenolinux helper functions */
186 int network_connect(struct netfront_info *);
188 static void xn_free_rx_ring(struct netfront_info *);
190 static void xn_free_tx_ring(struct netfront_info *);
192 static int xennet_get_responses(struct netfront_info *np,
193 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
194 struct mbuf **list, int *pages_flipped_p);
196 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
198 #define INVALID_P2M_ENTRY (~0UL)
201 * Mbuf pointers. We need these to keep track of the virtual addresses
202 * of our mbuf chains since we can only convert from virtual to physical,
203 * not the other way around. The size must track the free index arrays.
205 struct xn_chain_data {
206 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1];
208 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1];
211 struct net_device_stats
213 u_long rx_packets; /* total packets received */
214 u_long tx_packets; /* total packets transmitted */
215 u_long rx_bytes; /* total bytes received */
216 u_long tx_bytes; /* total bytes transmitted */
217 u_long rx_errors; /* bad packets received */
218 u_long tx_errors; /* packet transmit problems */
219 u_long rx_dropped; /* no space in linux buffers */
220 u_long tx_dropped; /* no space available in linux */
221 u_long multicast; /* multicast packets received */
224 /* detailed rx_errors: */
225 u_long rx_length_errors;
226 u_long rx_over_errors; /* receiver ring buff overflow */
227 u_long rx_crc_errors; /* recved pkt with crc error */
228 u_long rx_frame_errors; /* recv'd frame alignment error */
229 u_long rx_fifo_errors; /* recv'r fifo overrun */
230 u_long rx_missed_errors; /* receiver missed packet */
232 /* detailed tx_errors */
233 u_long tx_aborted_errors;
234 u_long tx_carrier_errors;
235 u_long tx_fifo_errors;
236 u_long tx_heartbeat_errors;
237 u_long tx_window_errors;
240 u_long rx_compressed;
241 u_long tx_compressed;
244 struct netfront_info {
245 struct ifnet *xn_ifp;
246 #if __FreeBSD_version >= 700000
247 struct lro_ctrl xn_lro;
250 struct net_device_stats stats;
253 netif_tx_front_ring_t tx;
254 netif_rx_front_ring_t rx;
262 u_int copying_receiver;
266 /* Receive-ring batched refills. */
267 #define RX_MIN_TARGET 32
268 #define RX_MAX_TARGET NET_RX_RING_SIZE
273 grant_ref_t gref_tx_head;
274 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
275 grant_ref_t gref_rx_head;
276 grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
281 uint8_t mac[ETHER_ADDR_LEN];
282 struct xn_chain_data xn_cdata; /* mbufs */
283 struct mbuf_head xn_rx_batch; /* head of the batch queue */
286 struct callout xn_stat_ch;
288 u_long rx_pfn_array[NET_RX_RING_SIZE];
289 multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1];
290 mmu_update_t rx_mmu[NET_RX_RING_SIZE];
291 struct ifmedia sc_media;
294 #define rx_mbufs xn_cdata.xn_rx_chain
295 #define tx_mbufs xn_cdata.xn_tx_chain
297 #define XN_LOCK_INIT(_sc, _name) \
298 mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
299 mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF); \
300 mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF)
302 #define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock)
303 #define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock)
305 #define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock)
306 #define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock)
308 #define XN_LOCK(_sc) mtx_lock(&(_sc)->sc_lock);
309 #define XN_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_lock);
311 #define XN_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->sc_lock, MA_OWNED);
312 #define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED);
313 #define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED);
314 #define XN_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->rx_lock); \
315 mtx_destroy(&(_sc)->tx_lock); \
316 mtx_destroy(&(_sc)->sc_lock);
318 struct netfront_rx_info {
319 struct netif_rx_response rx;
320 struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
323 #define netfront_carrier_on(netif) ((netif)->carrier = 1)
324 #define netfront_carrier_off(netif) ((netif)->carrier = 0)
325 #define netfront_carrier_ok(netif) ((netif)->carrier)
327 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
330 add_id_to_freelist(struct mbuf **list, uintptr_t id)
333 ("%s: the head item (0) must always be free.", __func__));
335 list[0] = (struct mbuf *)id;
338 static inline unsigned short
339 get_id_from_freelist(struct mbuf **list)
343 id = (uintptr_t)list[0];
345 ("%s: the head item (0) must always remain free.", __func__));
351 xennet_rxidx(RING_IDX idx)
353 return idx & (NET_RX_RING_SIZE - 1);
356 static inline struct mbuf *
357 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri)
359 int i = xennet_rxidx(ri);
363 np->rx_mbufs[i] = NULL;
367 static inline grant_ref_t
368 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
370 int i = xennet_rxidx(ri);
371 grant_ref_t ref = np->grant_rx_ref[i];
372 KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n"));
373 np->grant_rx_ref[i] = GRANT_REF_INVALID;
377 #define IPRINTK(fmt, args...) \
378 printf("[XEN] " fmt, ##args)
380 #define WPRINTK(fmt, args...) \
381 printf("[XEN] " fmt, ##args)
383 #define WPRINTK(fmt, args...)
386 #define DPRINTK(fmt, args...) \
387 printf("[XEN] %s: " fmt, __func__, ##args)
389 #define DPRINTK(fmt, args...)
393 * Read the 'mac' node at the given device's node in the store, and parse that
394 * as colon-separated octets, placing result the given mac array. mac must be
395 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
396 * Return 0 on success, or errno on error.
399 xen_net_read_mac(device_t dev, uint8_t mac[])
402 char *s, *e, *macstr;
405 path = xenbus_get_node(dev);
406 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
407 if (error == ENOENT) {
409 * Deal with missing mac XenStore nodes on devices with
410 * HVM emulation (the 'ioemu' configuration attribute)
413 * The HVM emulator may execute in a stub device model
414 * domain which lacks the permission, only given to Dom0,
415 * to update the guest's XenStore tree. For this reason,
416 * the HVM emulator doesn't even attempt to write the
417 * front-side mac node, even when operating in Dom0.
418 * However, there should always be a mac listed in the
419 * backend tree. Fallback to this version if our query
420 * of the front side XenStore location doesn't find
423 path = xenbus_get_otherend_path(dev);
424 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr);
427 xenbus_dev_fatal(dev, error, "parsing %s/mac", path);
432 for (i = 0; i < ETHER_ADDR_LEN; i++) {
433 mac[i] = strtoul(s, &e, 16);
434 if (s == e || (e[0] != ':' && e[0] != 0)) {
435 free(macstr, M_XENBUS);
440 free(macstr, M_XENBUS);
445 * Entry point to this code when a new device is created. Allocate the basic
446 * structures and the ring buffers for communication with the backend, and
447 * inform the backend of the appropriate details for those. Switch to
451 netfront_probe(device_t dev)
454 if (!strcmp(xenbus_get_type(dev), "vif")) {
455 device_set_desc(dev, "Virtual Network Interface");
463 netfront_attach(device_t dev)
467 err = create_netdev(dev);
469 xenbus_dev_fatal(dev, err, "creating netdev");
473 #if __FreeBSD_version >= 700000
474 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
475 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
476 OID_AUTO, "enable_lro", CTLTYPE_INT|CTLFLAG_RW,
477 &xn_enable_lro, 0, "Large Receive Offload");
484 netfront_suspend(device_t dev)
486 struct netfront_info *info = device_get_softc(dev);
490 netfront_carrier_off(info);
497 * We are reconnecting to the backend, due to a suspend/resume, or a backend
498 * driver restart. We tear down our netif structure and recreate it, but
499 * leave the device-layer structures intact so that this is transparent to the
500 * rest of the kernel.
503 netfront_resume(device_t dev)
505 struct netfront_info *info = device_get_softc(dev);
507 netif_disconnect_backend(info);
511 /* Common code used when first setting up, and when resuming. */
513 talk_to_backend(device_t dev, struct netfront_info *info)
516 struct xs_transaction xst;
517 const char *node = xenbus_get_node(dev);
520 err = xen_net_read_mac(dev, info->mac);
522 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
526 /* Create shared ring, alloc event channel. */
527 err = setup_device(dev, info);
532 err = xs_transaction_start(&xst);
534 xenbus_dev_fatal(dev, err, "starting transaction");
537 err = xs_printf(xst, node, "tx-ring-ref","%u",
540 message = "writing tx ring-ref";
541 goto abort_transaction;
543 err = xs_printf(xst, node, "rx-ring-ref","%u",
546 message = "writing rx ring-ref";
547 goto abort_transaction;
549 err = xs_printf(xst, node,
550 "event-channel", "%u", irq_to_evtchn_port(info->irq));
552 message = "writing event-channel";
553 goto abort_transaction;
555 err = xs_printf(xst, node, "request-rx-copy", "%u",
556 info->copying_receiver);
558 message = "writing request-rx-copy";
559 goto abort_transaction;
561 err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
563 message = "writing feature-rx-notify";
564 goto abort_transaction;
566 err = xs_printf(xst, node, "feature-sg", "%d", 1);
568 message = "writing feature-sg";
569 goto abort_transaction;
571 #if __FreeBSD_version >= 700000
572 err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
574 message = "writing feature-gso-tcpv4";
575 goto abort_transaction;
579 err = xs_transaction_end(xst, 0);
583 xenbus_dev_fatal(dev, err, "completing transaction");
590 xs_transaction_end(xst, 1);
591 xenbus_dev_fatal(dev, err, "%s", message);
599 setup_device(device_t dev, struct netfront_info *info)
601 netif_tx_sring_t *txs;
602 netif_rx_sring_t *rxs;
608 info->tx_ring_ref = GRANT_REF_INVALID;
609 info->rx_ring_ref = GRANT_REF_INVALID;
610 info->rx.sring = NULL;
611 info->tx.sring = NULL;
614 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
617 xenbus_dev_fatal(dev, error, "allocating tx ring page");
620 SHARED_RING_INIT(txs);
621 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
622 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
626 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
629 xenbus_dev_fatal(dev, error, "allocating rx ring page");
632 SHARED_RING_INIT(rxs);
633 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
635 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
639 error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
640 "xn", xn_intr, info, INTR_TYPE_NET | INTR_MPSAFE, &info->irq);
643 xenbus_dev_fatal(dev, error,
644 "bind_evtchn_to_irqhandler failed");
657 * If this interface has an ipv4 address, send an arp for it. This
658 * helps to get the network going again after migrating hosts.
661 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
667 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
668 if (ifa->ifa_addr->sa_family == AF_INET) {
669 arp_ifinit(ifp, ifa);
676 * Callback received when the backend's state changes.
679 netfront_backend_changed(device_t dev, XenbusState newstate)
681 struct netfront_info *sc = device_get_softc(dev);
683 DPRINTK("newstate=%d\n", newstate);
686 case XenbusStateInitialising:
687 case XenbusStateInitialised:
688 case XenbusStateConnected:
689 case XenbusStateUnknown:
690 case XenbusStateClosed:
691 case XenbusStateReconfigured:
692 case XenbusStateReconfiguring:
694 case XenbusStateInitWait:
695 if (xenbus_get_state(dev) != XenbusStateInitialising)
697 if (network_connect(sc) != 0)
699 xenbus_set_state(dev, XenbusStateConnected);
701 netfront_send_fake_arp(dev, sc);
704 case XenbusStateClosing:
705 xenbus_set_state(dev, XenbusStateClosed);
711 xn_free_rx_ring(struct netfront_info *sc)
716 for (i = 0; i < NET_RX_RING_SIZE; i++) {
717 if (sc->xn_cdata.rx_mbufs[i] != NULL) {
718 m_freem(sc->rx_mbufs[i]);
719 sc->rx_mbufs[i] = NULL;
724 sc->xn_rx_if->req_prod = 0;
725 sc->xn_rx_if->event = sc->rx.rsp_cons ;
730 xn_free_tx_ring(struct netfront_info *sc)
735 for (i = 0; i < NET_TX_RING_SIZE; i++) {
736 if (sc->tx_mbufs[i] != NULL) {
737 m_freem(sc->tx_mbufs[i]);
738 sc->xn_cdata.xn_tx_chain[i] = NULL;
747 * \brief Verify that there is sufficient space in the Tx ring
748 * buffer for a maximally sized request to be enqueued.
750 * A transmit request requires a transmit descriptor for each packet
751 * fragment, plus up to 2 entries for "options" (e.g. TSO).
754 xn_tx_slot_available(struct netfront_info *np)
756 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
760 netif_release_tx_bufs(struct netfront_info *np)
764 for (i = 1; i <= NET_TX_RING_SIZE; i++) {
770 * We assume that no kernel addresses are
771 * less than NET_TX_RING_SIZE. Any entry
772 * in the table that is below this number
773 * must be an index from free-list tracking.
775 if (((uintptr_t)m) <= NET_TX_RING_SIZE)
777 gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
778 gnttab_release_grant_reference(&np->gref_tx_head,
779 np->grant_tx_ref[i]);
780 np->grant_tx_ref[i] = GRANT_REF_INVALID;
781 add_id_to_freelist(np->tx_mbufs, i);
782 np->xn_cdata.xn_tx_chain_cnt--;
783 if (np->xn_cdata.xn_tx_chain_cnt < 0) {
784 panic("%s: tx_chain_cnt must be >= 0", __func__);
791 network_alloc_rx_buffers(struct netfront_info *sc)
793 int otherend_id = xenbus_get_otherend_id(sc->xbdev);
796 int i, batch_target, notify;
798 struct xen_memory_reservation reservation;
801 netif_rx_request_t *req;
805 req_prod = sc->rx.req_prod_pvt;
807 if (unlikely(sc->carrier == 0))
811 * Allocate mbufs greedily, even though we batch updates to the
812 * receive ring. This creates a less bursty demand on the memory
813 * allocator, and so should reduce the chance of failed allocation
814 * requests both for ourself and for other kernel subsystems.
816 * Here we attempt to maintain rx_target buffers in flight, counting
817 * buffers that we have yet to process in the receive ring.
819 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
820 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
821 MGETHDR(m_new, M_NOWAIT, MT_DATA);
823 printf("%s: MGETHDR failed\n", __func__);
827 m_cljget(m_new, M_NOWAIT, MJUMPAGESIZE);
828 if ((m_new->m_flags & M_EXT) == 0) {
829 printf("%s: m_cljget failed\n", __func__);
840 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
842 /* queue the mbufs allocated */
843 mbufq_tail(&sc->xn_rx_batch, m_new);
847 * If we've allocated at least half of our target number of entries,
848 * submit them to the backend - we have enough to make the overhead
849 * of submission worthwhile. Otherwise wait for more mbufs and
850 * request entries to become available.
852 if (i < (sc->rx_target/2)) {
853 if (req_prod >sc->rx.sring->req_prod)
859 * Double floating fill target if we risked having the backend
860 * run out of empty buffers for receive traffic. We define "running
861 * low" as having less than a fourth of our target buffers free
862 * at the time we refilled the queue.
864 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
866 if (sc->rx_target > sc->rx_max_target)
867 sc->rx_target = sc->rx_max_target;
871 for (nr_flips = i = 0; ; i++) {
872 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
875 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
876 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
878 id = xennet_rxidx(req_prod + i);
880 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
881 sc->rx_mbufs[id] = m_new;
883 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
884 KASSERT(ref != GNTTAB_LIST_END,
885 ("reserved grant references exhuasted"));
886 sc->grant_rx_ref[id] = ref;
888 vaddr = mtod(m_new, vm_offset_t);
889 pfn = vtophys(vaddr) >> PAGE_SHIFT;
890 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
892 if (sc->copying_receiver == 0) {
893 gnttab_grant_foreign_transfer_ref(ref,
895 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
896 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
897 /* Remove this page before passing
900 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
901 MULTI_update_va_mapping(&sc->rx_mcl[i],
906 gnttab_grant_foreign_access_ref(ref,
913 sc->rx_pfn_array[i] =
914 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
917 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
918 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
920 * We may have allocated buffers which have entries outstanding
921 * in the page * update queue -- make sure we flush those first!
926 /* Tell the ballon driver what is going on. */
927 balloon_update_driver_allowance(i);
929 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
930 reservation.nr_extents = i;
931 reservation.extent_order = 0;
932 reservation.address_bits = 0;
933 reservation.domid = DOMID_SELF;
935 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
936 /* After all PTEs have been zapped, flush the TLB. */
937 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
938 UVMF_TLB_FLUSH|UVMF_ALL;
940 /* Give away a batch of pages. */
941 sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
942 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
943 sc->rx_mcl[i].args[1] = (u_long)&reservation;
944 /* Zap PTEs and give away pages in one big multicall. */
945 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
947 if (unlikely(sc->rx_mcl[i].result != i ||
948 HYPERVISOR_memory_op(XENMEM_decrease_reservation,
950 panic("%s: unable to reduce memory "
951 "reservation\n", __func__);
957 /* Above is a suitable barrier to ensure backend will see requests. */
958 sc->rx.req_prod_pvt = req_prod + i;
960 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
962 notify_remote_via_irq(sc->irq);
966 xn_rxeof(struct netfront_info *np)
969 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
970 struct lro_ctrl *lro = &np->xn_lro;
971 struct lro_entry *queued;
973 struct netfront_rx_info rinfo;
974 struct netif_rx_response *rx = &rinfo.rx;
975 struct netif_extra_info *extras = rinfo.extras;
977 multicall_entry_t *mcl;
979 struct mbuf_head rxq, errq;
980 int err, pages_flipped = 0, work_to_do;
983 XN_RX_LOCK_ASSERT(np);
984 if (!netfront_carrier_ok(np))
992 rp = np->rx.sring->rsp_prod;
993 rmb(); /* Ensure we see queued responses up to 'rp'. */
997 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
998 memset(extras, 0, sizeof(rinfo.extras));
1001 err = xennet_get_responses(np, &rinfo, rp, &i, &m,
1004 if (unlikely(err)) {
1006 mbufq_tail(&errq, m);
1007 np->stats.rx_errors++;
1011 m->m_pkthdr.rcvif = ifp;
1012 if ( rx->flags & NETRXF_data_validated ) {
1013 /* Tell the stack the checksums are okay */
1015 * XXX this isn't necessarily the case - need to add
1019 m->m_pkthdr.csum_flags |=
1020 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
1022 m->m_pkthdr.csum_data = 0xffff;
1025 np->stats.rx_packets++;
1026 np->stats.rx_bytes += m->m_pkthdr.len;
1028 mbufq_tail(&rxq, m);
1029 np->rx.rsp_cons = i;
1032 if (pages_flipped) {
1033 /* Some pages are no longer absent... */
1035 balloon_update_driver_allowance(-pages_flipped);
1037 /* Do all the remapping work, and M->P updates, in one big
1040 if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1041 mcl = np->rx_mcl + pages_flipped;
1042 mcl->op = __HYPERVISOR_mmu_update;
1043 mcl->args[0] = (u_long)np->rx_mmu;
1044 mcl->args[1] = pages_flipped;
1046 mcl->args[3] = DOMID_SELF;
1047 (void)HYPERVISOR_multicall(np->rx_mcl,
1052 while ((m = mbufq_dequeue(&errq)))
1056 * Process all the mbufs after the remapping is complete.
1057 * Break the mbuf chain first though.
1059 while ((m = mbufq_dequeue(&rxq)) != NULL) {
1063 * Do we really need to drop the rx lock?
1066 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1067 /* Use LRO if possible */
1068 if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1069 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1071 * If LRO fails, pass up to the stack
1074 (*ifp->if_input)(ifp, m);
1077 (*ifp->if_input)(ifp, m);
1082 np->rx.rsp_cons = i;
1084 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1086 * Flush any outstanding LRO work
1088 while (!SLIST_EMPTY(&lro->lro_active)) {
1089 queued = SLIST_FIRST(&lro->lro_active);
1090 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1091 tcp_lro_flush(lro, queued);
1096 /* If we get a callback with very few responses, reduce fill target. */
1097 /* NB. Note exponential increase, linear decrease. */
1098 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1099 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1100 np->rx_target = np->rx_min_target;
1103 network_alloc_rx_buffers(np);
1105 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1106 } while (work_to_do);
1110 xn_txeof(struct netfront_info *np)
1115 netif_tx_response_t *txr;
1118 XN_TX_LOCK_ASSERT(np);
1120 if (!netfront_carrier_ok(np))
1126 prod = np->tx.sring->rsp_prod;
1127 rmb(); /* Ensure we see responses up to 'rp'. */
1129 for (i = np->tx.rsp_cons; i != prod; i++) {
1130 txr = RING_GET_RESPONSE(&np->tx, i);
1131 if (txr->status == NETIF_RSP_NULL)
1134 if (txr->status != NETIF_RSP_OKAY) {
1135 printf("%s: WARNING: response is %d!\n",
1136 __func__, txr->status);
1139 m = np->tx_mbufs[id];
1140 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1141 KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1142 ("mbuf already on the free list, but we're "
1143 "trying to free it again!"));
1147 * Increment packet count if this is the last
1148 * mbuf of the chain.
1152 if (unlikely(gnttab_query_foreign_access(
1153 np->grant_tx_ref[id]) != 0)) {
1154 panic("%s: grant id %u still in use by the "
1155 "backend", __func__, id);
1157 gnttab_end_foreign_access_ref(
1158 np->grant_tx_ref[id]);
1159 gnttab_release_grant_reference(
1160 &np->gref_tx_head, np->grant_tx_ref[id]);
1161 np->grant_tx_ref[id] = GRANT_REF_INVALID;
1163 np->tx_mbufs[id] = NULL;
1164 add_id_to_freelist(np->tx_mbufs, id);
1165 np->xn_cdata.xn_tx_chain_cnt--;
1167 /* Only mark the queue active if we've freed up at least one slot to try */
1168 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1170 np->tx.rsp_cons = prod;
1173 * Set a new event, then check for race with update of
1174 * tx_cons. Note that it is essential to schedule a
1175 * callback, no matter how few buffers are pending. Even if
1176 * there is space in the transmit ring, higher layers may
1177 * be blocked because too much data is outstanding: in such
1178 * cases notification from Xen is likely to be the only kick
1181 np->tx.sring->rsp_event =
1182 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1185 } while (prod != np->tx.sring->rsp_prod);
1188 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1191 if (np->user_state == UST_OPEN)
1192 netif_wake_queue(dev);
1200 struct netfront_info *np = xsc;
1201 struct ifnet *ifp = np->xn_ifp;
1204 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1205 likely(netfront_carrier_ok(np)) &&
1206 ifp->if_drv_flags & IFF_DRV_RUNNING))
1209 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1219 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1220 !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1225 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1228 int new = xennet_rxidx(np->rx.req_prod_pvt);
1230 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1231 np->rx_mbufs[new] = m;
1232 np->grant_rx_ref[new] = ref;
1233 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1234 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1235 np->rx.req_prod_pvt++;
1239 xennet_get_extras(struct netfront_info *np,
1240 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1242 struct netif_extra_info *extra;
1250 if (unlikely(*cons + 1 == rp)) {
1252 if (net_ratelimit())
1253 WPRINTK("Missing extra info\n");
1259 extra = (struct netif_extra_info *)
1260 RING_GET_RESPONSE(&np->rx, ++(*cons));
1262 if (unlikely(!extra->type ||
1263 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1265 if (net_ratelimit())
1266 WPRINTK("Invalid extra type: %d\n",
1271 memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1274 m = xennet_get_rx_mbuf(np, *cons);
1275 ref = xennet_get_rx_ref(np, *cons);
1276 xennet_move_rx_slot(np, m, ref);
1277 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1283 xennet_get_responses(struct netfront_info *np,
1284 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1286 int *pages_flipped_p)
1288 int pages_flipped = *pages_flipped_p;
1289 struct mmu_update *mmu;
1290 struct multicall_entry *mcl;
1291 struct netif_rx_response *rx = &rinfo->rx;
1292 struct netif_extra_info *extras = rinfo->extras;
1293 struct mbuf *m, *m0, *m_prev;
1294 grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1295 RING_IDX ref_cons = *cons;
1300 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1302 if (rx->flags & NETRXF_extra_info) {
1303 err = xennet_get_extras(np, extras, rp, cons);
1307 m0->m_pkthdr.len = 0;
1315 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1316 rx->status, rx->offset, frags);
1318 if (unlikely(rx->status < 0 ||
1319 rx->offset + rx->status > PAGE_SIZE)) {
1322 if (net_ratelimit())
1323 WPRINTK("rx->offset: %x, size: %u\n",
1324 rx->offset, rx->status);
1326 xennet_move_rx_slot(np, m, ref);
1331 goto next_skip_queue;
1335 * This definitely indicates a bug, either in this driver or in
1336 * the backend driver. In future this should flag the bad
1337 * situation to the system controller to reboot the backed.
1339 if (ref == GRANT_REF_INVALID) {
1342 if (net_ratelimit())
1343 WPRINTK("Bad rx response id %d.\n", rx->id);
1345 printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1350 if (!np->copying_receiver) {
1351 /* Memory pressure, insufficient buffer
1354 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1355 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1356 rx->id, rx->status);
1357 xennet_move_rx_slot(np, m, ref);
1362 if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1363 /* Remap the page. */
1364 void *vaddr = mtod(m, void *);
1367 mcl = np->rx_mcl + pages_flipped;
1368 mmu = np->rx_mmu + pages_flipped;
1370 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1371 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1372 PG_V | PG_M | PG_A, 0);
1373 pfn = (uintptr_t)m->m_ext.ext_arg1;
1374 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1375 MMU_MACHPHYS_UPDATE;
1378 set_phys_to_machine(pfn, mfn);
1382 ret = gnttab_end_foreign_access_ref(ref);
1383 KASSERT(ret, ("ret != 0"));
1386 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1392 m->m_len = rx->status;
1393 m->m_data += rx->offset;
1394 m0->m_pkthdr.len += rx->status;
1397 if (!(rx->flags & NETRXF_more_data))
1400 if (*cons + frags == rp) {
1401 if (net_ratelimit())
1402 WPRINTK("Need more frags\n");
1404 printf("%s: cons %u frags %u rp %u, not enough frags\n",
1405 __func__, *cons, frags, rp);
1409 * Note that m can be NULL, if rx->status < 0 or if
1410 * rx->offset + rx->status > PAGE_SIZE above.
1414 rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1415 m = xennet_get_rx_mbuf(np, *cons + frags);
1418 * m_prev == NULL can happen if rx->status < 0 or if
1419 * rx->offset + * rx->status > PAGE_SIZE above.
1425 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1426 * rx->status > PAGE_SIZE above.
1431 ref = xennet_get_rx_ref(np, *cons + frags);
1432 ref_cons = *cons + frags;
1437 *pages_flipped_p = pages_flipped;
1443 xn_tick_locked(struct netfront_info *sc)
1445 XN_RX_LOCK_ASSERT(sc);
1446 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1448 /* XXX placeholder for printing debug information */
1454 struct netfront_info *sc;
1463 * \brief Count the number of fragments in an mbuf chain.
1465 * Surprisingly, there isn't an M* macro for this.
1468 xn_count_frags(struct mbuf *m)
1472 for (nfrags = 0; m != NULL; m = m->m_next)
1479 * Given an mbuf chain, make sure we have enough room and then push
1480 * it onto the transmit ring.
1483 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1488 netif_extra_info_t *extra;
1494 * Defragment the mbuf if necessary.
1496 nfrags = xn_count_frags(m_head);
1499 * Check to see whether this request is longer than netback
1500 * can handle, and try to defrag it.
1503 * It is a bit lame, but the netback driver in Linux can't
1504 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1505 * the Linux network stack.
1507 if (nfrags > sc->maxfrags) {
1508 m = m_defrag(m_head, M_NOWAIT);
1511 * Defrag failed, so free the mbuf and
1512 * therefore drop the packet.
1520 /* Determine how many fragments now exist */
1521 nfrags = xn_count_frags(m_head);
1524 * Check to see whether the defragmented packet has too many
1525 * segments for the Linux netback driver.
1528 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1529 * of mbufs longer than Linux can handle. Make sure we don't
1530 * pass a too-long chain over to the other side by dropping the
1531 * packet. It doesn't look like there is currently a way to
1532 * tell the TCP stack to generate a shorter chain of packets.
1534 if (nfrags > MAX_TX_REQ_FRAGS) {
1536 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1537 "won't be able to handle it, dropping\n",
1538 __func__, nfrags, MAX_TX_REQ_FRAGS);
1545 * This check should be redundant. We've already verified that we
1546 * have enough slots in the ring to handle a packet of maximum
1547 * size, and that our packet is less than the maximum size. Keep
1548 * it in here as an assert for now just to make certain that
1549 * xn_tx_chain_cnt is accurate.
1551 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1552 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1553 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1554 (int) nfrags, (int) NET_TX_RING_SIZE));
1557 * Start packing the mbufs in this chain into
1558 * the fragment pointers. Stop when we run out
1559 * of fragments or hit the end of the mbuf chain.
1563 otherend_id = xenbus_get_otherend_id(sc->xbdev);
1564 for (m = m_head; m; m = m->m_next) {
1565 netif_tx_request_t *tx;
1568 u_long mfn; /* XXX Wrong type? */
1570 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1571 id = get_id_from_freelist(sc->tx_mbufs);
1573 panic("%s: was allocated the freelist head!\n",
1575 sc->xn_cdata.xn_tx_chain_cnt++;
1576 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1577 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1579 sc->tx_mbufs[id] = m;
1581 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1582 KASSERT((short)ref >= 0, ("Negative ref"));
1583 mfn = virt_to_mfn(mtod(m, vm_offset_t));
1584 gnttab_grant_foreign_access_ref(ref, otherend_id,
1585 mfn, GNTMAP_readonly);
1586 tx->gref = sc->grant_tx_ref[id] = ref;
1587 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1591 * The first fragment has the entire packet
1592 * size, subsequent fragments have just the
1593 * fragment size. The backend works out the
1594 * true size of the first fragment by
1595 * subtracting the sizes of the other
1598 tx->size = m->m_pkthdr.len;
1601 * The first fragment contains the checksum flags
1602 * and is optionally followed by extra data for
1606 * CSUM_TSO requires checksum offloading.
1607 * Some versions of FreeBSD fail to
1608 * set CSUM_TCP in the CSUM_TSO case,
1609 * so we have to test for CSUM_TSO
1612 if (m->m_pkthdr.csum_flags
1613 & (CSUM_DELAY_DATA | CSUM_TSO)) {
1614 tx->flags |= (NETTXF_csum_blank
1615 | NETTXF_data_validated);
1617 #if __FreeBSD_version >= 700000
1618 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1619 struct netif_extra_info *gso =
1620 (struct netif_extra_info *)
1621 RING_GET_REQUEST(&sc->tx,
1622 ++sc->tx.req_prod_pvt);
1624 tx->flags |= NETTXF_extra_info;
1626 gso->u.gso.size = m->m_pkthdr.tso_segsz;
1628 XEN_NETIF_GSO_TYPE_TCPV4;
1630 gso->u.gso.features = 0;
1632 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1637 tx->size = m->m_len;
1640 tx->flags |= NETTXF_more_data;
1642 sc->tx.req_prod_pvt++;
1644 BPF_MTAP(ifp, m_head);
1646 sc->stats.tx_bytes += m_head->m_pkthdr.len;
1647 sc->stats.tx_packets++;
1653 xn_start_locked(struct ifnet *ifp)
1655 struct netfront_info *sc;
1656 struct mbuf *m_head;
1661 if (!netfront_carrier_ok(sc))
1665 * While we have enough transmit slots available for at least one
1666 * maximum-sized packet, pull mbufs off the queue and put them on
1667 * the transmit ring.
1669 while (xn_tx_slot_available(sc)) {
1670 IF_DEQUEUE(&ifp->if_snd, m_head);
1674 if (xn_assemble_tx_request(sc, m_head) != 0)
1678 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1680 notify_remote_via_irq(sc->irq);
1682 if (RING_FULL(&sc->tx)) {
1685 netif_stop_queue(dev);
1691 xn_start(struct ifnet *ifp)
1693 struct netfront_info *sc;
1696 xn_start_locked(ifp);
1700 /* equivalent of network_open() in Linux */
1702 xn_ifinit_locked(struct netfront_info *sc)
1710 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1715 network_alloc_rx_buffers(sc);
1716 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1718 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1719 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1720 if_link_state_change(ifp, LINK_STATE_UP);
1722 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1726 xn_ifinit(void *xsc)
1728 struct netfront_info *sc = xsc;
1731 xn_ifinit_locked(sc);
1736 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1738 struct netfront_info *sc = ifp->if_softc;
1739 struct ifreq *ifr = (struct ifreq *) data;
1741 struct ifaddr *ifa = (struct ifaddr *)data;
1744 int mask, error = 0;
1750 if (ifa->ifa_addr->sa_family == AF_INET) {
1751 ifp->if_flags |= IFF_UP;
1752 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1753 xn_ifinit_locked(sc);
1754 arp_ifinit(ifp, ifa);
1759 error = ether_ioctl(ifp, cmd, data);
1765 /* XXX can we alter the MTU on a VN ?*/
1767 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1772 ifp->if_mtu = ifr->ifr_mtu;
1773 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1779 if (ifp->if_flags & IFF_UP) {
1781 * If only the state of the PROMISC flag changed,
1782 * then just use the 'set promisc mode' command
1783 * instead of reinitializing the entire NIC. Doing
1784 * a full re-init means reloading the firmware and
1785 * waiting for it to start up, which may take a
1789 /* No promiscuous mode with Xen */
1790 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1791 ifp->if_flags & IFF_PROMISC &&
1792 !(sc->xn_if_flags & IFF_PROMISC)) {
1793 XN_SETBIT(sc, XN_RX_MODE,
1794 XN_RXMODE_RX_PROMISC);
1795 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1796 !(ifp->if_flags & IFF_PROMISC) &&
1797 sc->xn_if_flags & IFF_PROMISC) {
1798 XN_CLRBIT(sc, XN_RX_MODE,
1799 XN_RXMODE_RX_PROMISC);
1802 xn_ifinit_locked(sc);
1804 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1808 sc->xn_if_flags = ifp->if_flags;
1813 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1814 if (mask & IFCAP_TXCSUM) {
1815 if (IFCAP_TXCSUM & ifp->if_capenable) {
1816 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1817 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1818 | CSUM_IP | CSUM_TSO);
1820 ifp->if_capenable |= IFCAP_TXCSUM;
1821 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1825 if (mask & IFCAP_RXCSUM) {
1826 ifp->if_capenable ^= IFCAP_RXCSUM;
1828 #if __FreeBSD_version >= 700000
1829 if (mask & IFCAP_TSO4) {
1830 if (IFCAP_TSO4 & ifp->if_capenable) {
1831 ifp->if_capenable &= ~IFCAP_TSO4;
1832 ifp->if_hwassist &= ~CSUM_TSO;
1833 } else if (IFCAP_TXCSUM & ifp->if_capenable) {
1834 ifp->if_capenable |= IFCAP_TSO4;
1835 ifp->if_hwassist |= CSUM_TSO;
1837 IPRINTK("Xen requires tx checksum offload"
1838 " be enabled to use TSO\n");
1842 if (mask & IFCAP_LRO) {
1843 ifp->if_capenable ^= IFCAP_LRO;
1852 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1862 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1865 error = ether_ioctl(ifp, cmd, data);
1872 xn_stop(struct netfront_info *sc)
1880 callout_stop(&sc->xn_stat_ch);
1882 xn_free_rx_ring(sc);
1883 xn_free_tx_ring(sc);
1885 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1886 if_link_state_change(ifp, LINK_STATE_DOWN);
1889 /* START of Xenolinux helper functions adapted to FreeBSD */
1891 network_connect(struct netfront_info *np)
1893 int i, requeue_idx, error;
1895 netif_rx_request_t *req;
1896 u_int feature_rx_copy, feature_rx_flip;
1898 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1899 "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1901 feature_rx_copy = 0;
1902 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1903 "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1905 feature_rx_flip = 1;
1908 * Copy packets on receive path if:
1909 * (a) This was requested by user, and the backend supports it; or
1910 * (b) Flipping was requested, but this is unsupported by the backend.
1912 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1913 (MODPARM_rx_flip && !feature_rx_flip));
1915 /* Recovery procedure: */
1916 error = talk_to_backend(np->xbdev, np);
1920 /* Step 1: Reinitialise variables. */
1921 xn_query_features(np);
1922 xn_configure_features(np);
1923 netif_release_tx_bufs(np);
1925 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1926 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1930 if (np->rx_mbufs[i] == NULL)
1933 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1934 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1936 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1937 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1939 if (!np->copying_receiver) {
1940 gnttab_grant_foreign_transfer_ref(ref,
1941 xenbus_get_otherend_id(np->xbdev),
1944 gnttab_grant_foreign_access_ref(ref,
1945 xenbus_get_otherend_id(np->xbdev),
1949 req->id = requeue_idx;
1954 np->rx.req_prod_pvt = requeue_idx;
1956 /* Step 3: All public and private state should now be sane. Get
1957 * ready to start sending and receiving packets and give the driver
1958 * domain a kick because we've probably just requeued some
1961 netfront_carrier_on(np);
1962 notify_remote_via_irq(np->irq);
1966 network_alloc_rx_buffers(np);
1972 xn_query_features(struct netfront_info *np)
1976 device_printf(np->xbdev, "backend features:");
1978 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1979 "feature-sg", NULL, "%d", &val) < 0)
1984 np->maxfrags = MAX_TX_REQ_FRAGS;
1985 printf(" feature-sg");
1988 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1989 "feature-gso-tcpv4", NULL, "%d", &val) < 0)
1992 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
1994 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
1995 printf(" feature-gso-tcp4");
2002 xn_configure_features(struct netfront_info *np)
2007 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
2008 if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0)
2009 tcp_lro_free(&np->xn_lro);
2011 np->xn_ifp->if_capenable =
2012 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4);
2013 np->xn_ifp->if_hwassist &= ~CSUM_TSO;
2014 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
2015 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) {
2016 err = tcp_lro_init(&np->xn_lro);
2018 device_printf(np->xbdev, "LRO initialization failed\n");
2020 np->xn_lro.ifp = np->xn_ifp;
2021 np->xn_ifp->if_capenable |= IFCAP_LRO;
2024 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) {
2025 np->xn_ifp->if_capenable |= IFCAP_TSO4;
2026 np->xn_ifp->if_hwassist |= CSUM_TSO;
2032 /** Create a network device.
2033 * @param handle device handle
2036 create_netdev(device_t dev)
2039 struct netfront_info *np;
2043 np = device_get_softc(dev);
2047 XN_LOCK_INIT(np, xennetif);
2049 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
2050 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
2051 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
2053 np->rx_target = RX_MIN_TARGET;
2054 np->rx_min_target = RX_MIN_TARGET;
2055 np->rx_max_target = RX_MAX_TARGET;
2057 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2058 for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2059 np->tx_mbufs[i] = (void *) ((u_long) i+1);
2060 np->grant_tx_ref[i] = GRANT_REF_INVALID;
2062 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2064 for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2066 np->rx_mbufs[i] = NULL;
2067 np->grant_rx_ref[i] = GRANT_REF_INVALID;
2069 /* A grant for every tx ring slot */
2070 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2071 &np->gref_tx_head) != 0) {
2072 IPRINTK("#### netfront can't alloc tx grant refs\n");
2076 /* A grant for every rx ring slot */
2077 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2078 &np->gref_rx_head) != 0) {
2079 WPRINTK("#### netfront can't alloc rx grant refs\n");
2080 gnttab_free_grant_references(np->gref_tx_head);
2085 err = xen_net_read_mac(dev, np->mac);
2089 /* Set up ifnet structure */
2090 ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2092 if_initname(ifp, "xn", device_get_unit(dev));
2093 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2094 ifp->if_ioctl = xn_ioctl;
2095 ifp->if_output = ether_output;
2096 ifp->if_start = xn_start;
2098 ifp->if_watchdog = xn_watchdog;
2100 ifp->if_init = xn_ifinit;
2101 ifp->if_mtu = ETHERMTU;
2102 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2104 ifp->if_hwassist = XN_CSUM_FEATURES;
2105 ifp->if_capabilities = IFCAP_HWCSUM;
2106 ifp->if_hw_tsomax = NF_TSO_MAXBURST;
2108 ether_ifattach(ifp, np->mac);
2109 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2110 netfront_carrier_off(np);
2115 gnttab_free_grant_references(np->gref_tx_head);
2121 * Handle the change of state of the backend to Closing. We must delete our
2122 * device-layer structures now, to ensure that writes are flushed through to
2123 * the backend. Once is this done, we can switch to Closed in
2128 netfront_closing(device_t dev)
2131 struct netfront_info *info = dev->dev_driver_data;
2133 DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2137 xenbus_switch_state(dev, XenbusStateClosed);
2142 netfront_detach(device_t dev)
2144 struct netfront_info *info = device_get_softc(dev);
2146 DPRINTK("%s\n", xenbus_get_node(dev));
2154 netif_free(struct netfront_info *info)
2159 callout_drain(&info->xn_stat_ch);
2160 netif_disconnect_backend(info);
2161 if (info->xn_ifp != NULL) {
2162 ether_ifdetach(info->xn_ifp);
2163 if_free(info->xn_ifp);
2164 info->xn_ifp = NULL;
2166 ifmedia_removeall(&info->sc_media);
2170 netif_disconnect_backend(struct netfront_info *info)
2174 netfront_carrier_off(info);
2178 free_ring(&info->tx_ring_ref, &info->tx.sring);
2179 free_ring(&info->rx_ring_ref, &info->rx.sring);
2182 unbind_from_irqhandler(info->irq);
2188 free_ring(int *ref, void *ring_ptr_ref)
2190 void **ring_ptr_ptr = ring_ptr_ref;
2192 if (*ref != GRANT_REF_INVALID) {
2193 /* This API frees the associated storage. */
2194 gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
2195 *ref = GRANT_REF_INVALID;
2197 *ring_ptr_ptr = NULL;
2201 xn_ifmedia_upd(struct ifnet *ifp)
2207 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2209 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2210 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2213 /* ** Driver registration ** */
2214 static device_method_t netfront_methods[] = {
2215 /* Device interface */
2216 DEVMETHOD(device_probe, netfront_probe),
2217 DEVMETHOD(device_attach, netfront_attach),
2218 DEVMETHOD(device_detach, netfront_detach),
2219 DEVMETHOD(device_shutdown, bus_generic_shutdown),
2220 DEVMETHOD(device_suspend, netfront_suspend),
2221 DEVMETHOD(device_resume, netfront_resume),
2223 /* Xenbus interface */
2224 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2229 static driver_t netfront_driver = {
2232 sizeof(struct netfront_info),
2234 devclass_t netfront_devclass;
2236 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,