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>
45 #include <sys/limits.h>
48 #include <net/if_arp.h>
49 #include <net/ethernet.h>
50 #include <net/if_dl.h>
51 #include <net/if_media.h>
55 #include <net/if_types.h>
58 #include <netinet/in_systm.h>
59 #include <netinet/in.h>
60 #include <netinet/ip.h>
61 #include <netinet/if_ether.h>
62 #if __FreeBSD_version >= 700000
63 #include <netinet/tcp.h>
64 #include <netinet/tcp_lro.h>
70 #include <machine/clock.h> /* for DELAY */
71 #include <machine/bus.h>
72 #include <machine/resource.h>
73 #include <machine/frame.h>
74 #include <machine/vmparam.h>
79 #include <machine/intr_machdep.h>
81 #include <xen/xen-os.h>
82 #include <xen/hypervisor.h>
83 #include <xen/xen_intr.h>
84 #include <xen/gnttab.h>
85 #include <xen/interface/memory.h>
86 #include <xen/interface/io/netif.h>
87 #include <xen/xenbus/xenbusvar.h>
89 #include <machine/xen/xenvar.h>
91 #include <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)
139 #define RX_COPY_THRESHOLD 256
141 #define net_ratelimit() 0
143 struct netfront_info;
144 struct netfront_rx_info;
146 static void xn_txeof(struct netfront_info *);
147 static void xn_rxeof(struct netfront_info *);
148 static void network_alloc_rx_buffers(struct netfront_info *);
150 static void xn_tick_locked(struct netfront_info *);
151 static void xn_tick(void *);
153 static void xn_intr(void *);
154 static inline int xn_count_frags(struct mbuf *m);
155 static int xn_assemble_tx_request(struct netfront_info *sc,
156 struct mbuf *m_head);
157 static void xn_start_locked(struct ifnet *);
158 static void xn_start(struct ifnet *);
159 static int xn_ioctl(struct ifnet *, u_long, caddr_t);
160 static void xn_ifinit_locked(struct netfront_info *);
161 static void xn_ifinit(void *);
162 static void xn_stop(struct netfront_info *);
163 static void xn_query_features(struct netfront_info *np);
164 static int xn_configure_features(struct netfront_info *np);
166 static void xn_watchdog(struct ifnet *);
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;
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)
455 if (xen_disable_pv_nics != 0)
459 if (!strcmp(xenbus_get_type(dev), "vif")) {
460 device_set_desc(dev, "Virtual Network Interface");
468 netfront_attach(device_t dev)
472 err = create_netdev(dev);
474 xenbus_dev_fatal(dev, err, "creating netdev");
478 #if __FreeBSD_version >= 700000
479 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev),
480 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)),
481 OID_AUTO, "enable_lro", CTLFLAG_RW,
482 &xn_enable_lro, 0, "Large Receive Offload");
489 netfront_suspend(device_t dev)
491 struct netfront_info *info = device_get_softc(dev);
495 netfront_carrier_off(info);
502 * We are reconnecting to the backend, due to a suspend/resume, or a backend
503 * driver restart. We tear down our netif structure and recreate it, but
504 * leave the device-layer structures intact so that this is transparent to the
505 * rest of the kernel.
508 netfront_resume(device_t dev)
510 struct netfront_info *info = device_get_softc(dev);
512 info->xn_resume = true;
513 netif_disconnect_backend(info);
517 /* Common code used when first setting up, and when resuming. */
519 talk_to_backend(device_t dev, struct netfront_info *info)
522 struct xs_transaction xst;
523 const char *node = xenbus_get_node(dev);
526 err = xen_net_read_mac(dev, info->mac);
528 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
532 /* Create shared ring, alloc event channel. */
533 err = setup_device(dev, info);
538 err = xs_transaction_start(&xst);
540 xenbus_dev_fatal(dev, err, "starting transaction");
543 err = xs_printf(xst, node, "tx-ring-ref","%u",
546 message = "writing tx ring-ref";
547 goto abort_transaction;
549 err = xs_printf(xst, node, "rx-ring-ref","%u",
552 message = "writing rx ring-ref";
553 goto abort_transaction;
555 err = xs_printf(xst, node,
556 "event-channel", "%u",
557 xen_intr_port(info->xen_intr_handle));
559 message = "writing event-channel";
560 goto abort_transaction;
562 err = xs_printf(xst, node, "request-rx-copy", "%u",
563 info->copying_receiver);
565 message = "writing request-rx-copy";
566 goto abort_transaction;
568 err = xs_printf(xst, node, "feature-rx-notify", "%d", 1);
570 message = "writing feature-rx-notify";
571 goto abort_transaction;
573 err = xs_printf(xst, node, "feature-sg", "%d", 1);
575 message = "writing feature-sg";
576 goto abort_transaction;
578 #if __FreeBSD_version >= 700000
579 err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1);
581 message = "writing feature-gso-tcpv4";
582 goto abort_transaction;
586 err = xs_transaction_end(xst, 0);
590 xenbus_dev_fatal(dev, err, "completing transaction");
597 xs_transaction_end(xst, 1);
598 xenbus_dev_fatal(dev, err, "%s", message);
606 setup_device(device_t dev, struct netfront_info *info)
608 netif_tx_sring_t *txs;
609 netif_rx_sring_t *rxs;
615 info->tx_ring_ref = GRANT_REF_INVALID;
616 info->rx_ring_ref = GRANT_REF_INVALID;
617 info->rx.sring = NULL;
618 info->tx.sring = NULL;
620 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
623 xenbus_dev_fatal(dev, error, "allocating tx ring page");
626 SHARED_RING_INIT(txs);
627 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
628 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
632 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
635 xenbus_dev_fatal(dev, error, "allocating rx ring page");
638 SHARED_RING_INIT(rxs);
639 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
641 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
645 error = xen_intr_alloc_and_bind_local_port(dev,
646 xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info,
647 INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle);
650 xenbus_dev_fatal(dev, error,
651 "xen_intr_alloc_and_bind_local_port failed");
664 * If this interface has an ipv4 address, send an arp for it. This
665 * helps to get the network going again after migrating hosts.
668 netfront_send_fake_arp(device_t dev, struct netfront_info *info)
674 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
675 if (ifa->ifa_addr->sa_family == AF_INET) {
676 arp_ifinit(ifp, ifa);
683 * Callback received when the backend's state changes.
686 netfront_backend_changed(device_t dev, XenbusState newstate)
688 struct netfront_info *sc = device_get_softc(dev);
690 DPRINTK("newstate=%d\n", newstate);
693 case XenbusStateInitialising:
694 case XenbusStateInitialised:
695 case XenbusStateUnknown:
696 case XenbusStateClosed:
697 case XenbusStateReconfigured:
698 case XenbusStateReconfiguring:
700 case XenbusStateInitWait:
701 if (xenbus_get_state(dev) != XenbusStateInitialising)
703 if (network_connect(sc) != 0)
705 xenbus_set_state(dev, XenbusStateConnected);
707 case XenbusStateClosing:
708 xenbus_set_state(dev, XenbusStateClosed);
710 case XenbusStateConnected:
712 netfront_send_fake_arp(dev, sc);
719 xn_free_rx_ring(struct netfront_info *sc)
724 for (i = 0; i < NET_RX_RING_SIZE; i++) {
725 if (sc->xn_cdata.rx_mbufs[i] != NULL) {
726 m_freem(sc->rx_mbufs[i]);
727 sc->rx_mbufs[i] = NULL;
732 sc->xn_rx_if->req_prod = 0;
733 sc->xn_rx_if->event = sc->rx.rsp_cons ;
738 xn_free_tx_ring(struct netfront_info *sc)
743 for (i = 0; i < NET_TX_RING_SIZE; i++) {
744 if (sc->tx_mbufs[i] != NULL) {
745 m_freem(sc->tx_mbufs[i]);
746 sc->xn_cdata.xn_tx_chain[i] = NULL;
755 * \brief Verify that there is sufficient space in the Tx ring
756 * buffer for a maximally sized request to be enqueued.
758 * A transmit request requires a transmit descriptor for each packet
759 * fragment, plus up to 2 entries for "options" (e.g. TSO).
762 xn_tx_slot_available(struct netfront_info *np)
764 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2));
768 netif_release_tx_bufs(struct netfront_info *np)
772 for (i = 1; i <= NET_TX_RING_SIZE; i++) {
778 * We assume that no kernel addresses are
779 * less than NET_TX_RING_SIZE. Any entry
780 * in the table that is below this number
781 * must be an index from free-list tracking.
783 if (((uintptr_t)m) <= NET_TX_RING_SIZE)
785 gnttab_end_foreign_access_ref(np->grant_tx_ref[i]);
786 gnttab_release_grant_reference(&np->gref_tx_head,
787 np->grant_tx_ref[i]);
788 np->grant_tx_ref[i] = GRANT_REF_INVALID;
789 add_id_to_freelist(np->tx_mbufs, i);
790 np->xn_cdata.xn_tx_chain_cnt--;
791 if (np->xn_cdata.xn_tx_chain_cnt < 0) {
792 panic("%s: tx_chain_cnt must be >= 0", __func__);
799 network_alloc_rx_buffers(struct netfront_info *sc)
801 int otherend_id = xenbus_get_otherend_id(sc->xbdev);
804 int i, batch_target, notify;
806 struct xen_memory_reservation reservation;
809 netif_rx_request_t *req;
813 req_prod = sc->rx.req_prod_pvt;
815 if (__predict_false(sc->carrier == 0))
819 * Allocate mbufs greedily, even though we batch updates to the
820 * receive ring. This creates a less bursty demand on the memory
821 * allocator, and so should reduce the chance of failed allocation
822 * requests both for ourself and for other kernel subsystems.
824 * Here we attempt to maintain rx_target buffers in flight, counting
825 * buffers that we have yet to process in the receive ring.
827 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
828 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
829 MGETHDR(m_new, M_NOWAIT, MT_DATA);
831 printf("%s: MGETHDR failed\n", __func__);
835 m_cljget(m_new, M_NOWAIT, MJUMPAGESIZE);
836 if ((m_new->m_flags & M_EXT) == 0) {
837 printf("%s: m_cljget failed\n", __func__);
848 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
850 /* queue the mbufs allocated */
851 mbufq_tail(&sc->xn_rx_batch, m_new);
855 * If we've allocated at least half of our target number of entries,
856 * submit them to the backend - we have enough to make the overhead
857 * of submission worthwhile. Otherwise wait for more mbufs and
858 * request entries to become available.
860 if (i < (sc->rx_target/2)) {
861 if (req_prod >sc->rx.sring->req_prod)
867 * Double floating fill target if we risked having the backend
868 * run out of empty buffers for receive traffic. We define "running
869 * low" as having less than a fourth of our target buffers free
870 * at the time we refilled the queue.
872 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) {
874 if (sc->rx_target > sc->rx_max_target)
875 sc->rx_target = sc->rx_max_target;
879 for (nr_flips = i = 0; ; i++) {
880 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
883 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
884 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
886 id = xennet_rxidx(req_prod + i);
888 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain"));
889 sc->rx_mbufs[id] = m_new;
891 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
892 KASSERT(ref != GNTTAB_LIST_END,
893 ("reserved grant references exhuasted"));
894 sc->grant_rx_ref[id] = ref;
896 vaddr = mtod(m_new, vm_offset_t);
897 pfn = vtophys(vaddr) >> PAGE_SHIFT;
898 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
900 if (sc->copying_receiver == 0) {
901 gnttab_grant_foreign_transfer_ref(ref,
903 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
904 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
905 /* Remove this page before passing
908 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
909 MULTI_update_va_mapping(&sc->rx_mcl[i],
914 gnttab_grant_foreign_access_ref(ref,
921 sc->rx_pfn_array[i] =
922 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
925 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
926 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
928 * We may have allocated buffers which have entries outstanding
929 * in the page * update queue -- make sure we flush those first!
934 /* Tell the ballon driver what is going on. */
935 balloon_update_driver_allowance(i);
937 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
938 reservation.nr_extents = i;
939 reservation.extent_order = 0;
940 reservation.address_bits = 0;
941 reservation.domid = DOMID_SELF;
943 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
944 /* After all PTEs have been zapped, flush the TLB. */
945 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
946 UVMF_TLB_FLUSH|UVMF_ALL;
948 /* Give away a batch of pages. */
949 sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
950 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
951 sc->rx_mcl[i].args[1] = (u_long)&reservation;
952 /* Zap PTEs and give away pages in one big multicall. */
953 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
955 if (__predict_false(sc->rx_mcl[i].result != i ||
956 HYPERVISOR_memory_op(XENMEM_decrease_reservation,
958 panic("%s: unable to reduce memory "
959 "reservation\n", __func__);
965 /* Above is a suitable barrier to ensure backend will see requests. */
966 sc->rx.req_prod_pvt = req_prod + i;
968 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
970 xen_intr_signal(sc->xen_intr_handle);
974 xn_rxeof(struct netfront_info *np)
977 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
978 struct lro_ctrl *lro = &np->xn_lro;
979 struct lro_entry *queued;
981 struct netfront_rx_info rinfo;
982 struct netif_rx_response *rx = &rinfo.rx;
983 struct netif_extra_info *extras = rinfo.extras;
985 multicall_entry_t *mcl;
987 struct mbuf_head rxq, errq;
988 int err, pages_flipped = 0, work_to_do;
991 XN_RX_LOCK_ASSERT(np);
992 if (!netfront_carrier_ok(np))
1000 rp = np->rx.sring->rsp_prod;
1001 rmb(); /* Ensure we see queued responses up to 'rp'. */
1003 i = np->rx.rsp_cons;
1005 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
1006 memset(extras, 0, sizeof(rinfo.extras));
1009 err = xennet_get_responses(np, &rinfo, rp, &i, &m,
1012 if (__predict_false(err)) {
1014 mbufq_tail(&errq, m);
1015 np->stats.rx_errors++;
1019 m->m_pkthdr.rcvif = ifp;
1020 if ( rx->flags & NETRXF_data_validated ) {
1021 /* Tell the stack the checksums are okay */
1023 * XXX this isn't necessarily the case - need to add
1027 m->m_pkthdr.csum_flags |=
1028 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
1030 m->m_pkthdr.csum_data = 0xffff;
1033 np->stats.rx_packets++;
1034 np->stats.rx_bytes += m->m_pkthdr.len;
1036 mbufq_tail(&rxq, m);
1037 np->rx.rsp_cons = i;
1040 if (pages_flipped) {
1041 /* Some pages are no longer absent... */
1043 balloon_update_driver_allowance(-pages_flipped);
1045 /* Do all the remapping work, and M->P updates, in one big
1048 if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
1049 mcl = np->rx_mcl + pages_flipped;
1050 mcl->op = __HYPERVISOR_mmu_update;
1051 mcl->args[0] = (u_long)np->rx_mmu;
1052 mcl->args[1] = pages_flipped;
1054 mcl->args[3] = DOMID_SELF;
1055 (void)HYPERVISOR_multicall(np->rx_mcl,
1060 while ((m = mbufq_dequeue(&errq)))
1064 * Process all the mbufs after the remapping is complete.
1065 * Break the mbuf chain first though.
1067 while ((m = mbufq_dequeue(&rxq)) != NULL) {
1071 * Do we really need to drop the rx lock?
1074 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1075 /* Use LRO if possible */
1076 if ((ifp->if_capenable & IFCAP_LRO) == 0 ||
1077 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) {
1079 * If LRO fails, pass up to the stack
1082 (*ifp->if_input)(ifp, m);
1085 (*ifp->if_input)(ifp, m);
1090 np->rx.rsp_cons = i;
1092 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
1094 * Flush any outstanding LRO work
1096 while (!SLIST_EMPTY(&lro->lro_active)) {
1097 queued = SLIST_FIRST(&lro->lro_active);
1098 SLIST_REMOVE_HEAD(&lro->lro_active, next);
1099 tcp_lro_flush(lro, queued);
1104 /* If we get a callback with very few responses, reduce fill target. */
1105 /* NB. Note exponential increase, linear decrease. */
1106 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1107 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
1108 np->rx_target = np->rx_min_target;
1111 network_alloc_rx_buffers(np);
1113 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
1114 } while (work_to_do);
1118 xn_txeof(struct netfront_info *np)
1123 netif_tx_response_t *txr;
1126 XN_TX_LOCK_ASSERT(np);
1128 if (!netfront_carrier_ok(np))
1134 prod = np->tx.sring->rsp_prod;
1135 rmb(); /* Ensure we see responses up to 'rp'. */
1137 for (i = np->tx.rsp_cons; i != prod; i++) {
1138 txr = RING_GET_RESPONSE(&np->tx, i);
1139 if (txr->status == NETIF_RSP_NULL)
1142 if (txr->status != NETIF_RSP_OKAY) {
1143 printf("%s: WARNING: response is %d!\n",
1144 __func__, txr->status);
1147 m = np->tx_mbufs[id];
1148 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
1149 KASSERT((uintptr_t)m > NET_TX_RING_SIZE,
1150 ("mbuf already on the free list, but we're "
1151 "trying to free it again!"));
1155 * Increment packet count if this is the last
1156 * mbuf of the chain.
1160 if (__predict_false(gnttab_query_foreign_access(
1161 np->grant_tx_ref[id]) != 0)) {
1162 panic("%s: grant id %u still in use by the "
1163 "backend", __func__, id);
1165 gnttab_end_foreign_access_ref(
1166 np->grant_tx_ref[id]);
1167 gnttab_release_grant_reference(
1168 &np->gref_tx_head, np->grant_tx_ref[id]);
1169 np->grant_tx_ref[id] = GRANT_REF_INVALID;
1171 np->tx_mbufs[id] = NULL;
1172 add_id_to_freelist(np->tx_mbufs, id);
1173 np->xn_cdata.xn_tx_chain_cnt--;
1175 /* Only mark the queue active if we've freed up at least one slot to try */
1176 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1178 np->tx.rsp_cons = prod;
1181 * Set a new event, then check for race with update of
1182 * tx_cons. Note that it is essential to schedule a
1183 * callback, no matter how few buffers are pending. Even if
1184 * there is space in the transmit ring, higher layers may
1185 * be blocked because too much data is outstanding: in such
1186 * cases notification from Xen is likely to be the only kick
1189 np->tx.sring->rsp_event =
1190 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1193 } while (prod != np->tx.sring->rsp_prod);
1196 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1199 if (np->user_state == UST_OPEN)
1200 netif_wake_queue(dev);
1208 struct netfront_info *np = xsc;
1209 struct ifnet *ifp = np->xn_ifp;
1212 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1213 likely(netfront_carrier_ok(np)) &&
1214 ifp->if_drv_flags & IFF_DRV_RUNNING))
1217 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) {
1227 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1228 !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1233 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1236 int new = xennet_rxidx(np->rx.req_prod_pvt);
1238 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1239 np->rx_mbufs[new] = m;
1240 np->grant_rx_ref[new] = ref;
1241 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1242 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1243 np->rx.req_prod_pvt++;
1247 xennet_get_extras(struct netfront_info *np,
1248 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons)
1250 struct netif_extra_info *extra;
1258 if (__predict_false(*cons + 1 == rp)) {
1260 if (net_ratelimit())
1261 WPRINTK("Missing extra info\n");
1267 extra = (struct netif_extra_info *)
1268 RING_GET_RESPONSE(&np->rx, ++(*cons));
1270 if (__predict_false(!extra->type ||
1271 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1273 if (net_ratelimit())
1274 WPRINTK("Invalid extra type: %d\n",
1279 memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1282 m = xennet_get_rx_mbuf(np, *cons);
1283 ref = xennet_get_rx_ref(np, *cons);
1284 xennet_move_rx_slot(np, m, ref);
1285 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1291 xennet_get_responses(struct netfront_info *np,
1292 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons,
1294 int *pages_flipped_p)
1296 int pages_flipped = *pages_flipped_p;
1297 struct mmu_update *mmu;
1298 struct multicall_entry *mcl;
1299 struct netif_rx_response *rx = &rinfo->rx;
1300 struct netif_extra_info *extras = rinfo->extras;
1301 struct mbuf *m, *m0, *m_prev;
1302 grant_ref_t ref = xennet_get_rx_ref(np, *cons);
1303 RING_IDX ref_cons = *cons;
1308 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons);
1310 if (rx->flags & NETRXF_extra_info) {
1311 err = xennet_get_extras(np, extras, rp, cons);
1315 m0->m_pkthdr.len = 0;
1323 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n",
1324 rx->status, rx->offset, frags);
1326 if (__predict_false(rx->status < 0 ||
1327 rx->offset + rx->status > PAGE_SIZE)) {
1330 if (net_ratelimit())
1331 WPRINTK("rx->offset: %x, size: %u\n",
1332 rx->offset, rx->status);
1334 xennet_move_rx_slot(np, m, ref);
1339 goto next_skip_queue;
1343 * This definitely indicates a bug, either in this driver or in
1344 * the backend driver. In future this should flag the bad
1345 * situation to the system controller to reboot the backed.
1347 if (ref == GRANT_REF_INVALID) {
1350 if (net_ratelimit())
1351 WPRINTK("Bad rx response id %d.\n", rx->id);
1353 printf("%s: Bad rx response id %d.\n", __func__,rx->id);
1358 if (!np->copying_receiver) {
1359 /* Memory pressure, insufficient buffer
1362 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1363 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n",
1364 rx->id, rx->status);
1365 xennet_move_rx_slot(np, m, ref);
1370 if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1371 /* Remap the page. */
1372 void *vaddr = mtod(m, void *);
1375 mcl = np->rx_mcl + pages_flipped;
1376 mmu = np->rx_mmu + pages_flipped;
1378 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1379 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1380 PG_V | PG_M | PG_A, 0);
1381 pfn = (uintptr_t)m->m_ext.ext_arg1;
1382 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1383 MMU_MACHPHYS_UPDATE;
1386 set_phys_to_machine(pfn, mfn);
1390 ret = gnttab_end_foreign_access_ref(ref);
1391 KASSERT(ret, ("ret != 0"));
1394 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1400 m->m_len = rx->status;
1401 m->m_data += rx->offset;
1402 m0->m_pkthdr.len += rx->status;
1405 if (!(rx->flags & NETRXF_more_data))
1408 if (*cons + frags == rp) {
1409 if (net_ratelimit())
1410 WPRINTK("Need more frags\n");
1412 printf("%s: cons %u frags %u rp %u, not enough frags\n",
1413 __func__, *cons, frags, rp);
1417 * Note that m can be NULL, if rx->status < 0 or if
1418 * rx->offset + rx->status > PAGE_SIZE above.
1422 rx = RING_GET_RESPONSE(&np->rx, *cons + frags);
1423 m = xennet_get_rx_mbuf(np, *cons + frags);
1426 * m_prev == NULL can happen if rx->status < 0 or if
1427 * rx->offset + * rx->status > PAGE_SIZE above.
1433 * m0 can be NULL if rx->status < 0 or if * rx->offset +
1434 * rx->status > PAGE_SIZE above.
1439 ref = xennet_get_rx_ref(np, *cons + frags);
1440 ref_cons = *cons + frags;
1445 *pages_flipped_p = pages_flipped;
1451 xn_tick_locked(struct netfront_info *sc)
1453 XN_RX_LOCK_ASSERT(sc);
1454 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1456 /* XXX placeholder for printing debug information */
1462 struct netfront_info *sc;
1471 * \brief Count the number of fragments in an mbuf chain.
1473 * Surprisingly, there isn't an M* macro for this.
1476 xn_count_frags(struct mbuf *m)
1480 for (nfrags = 0; m != NULL; m = m->m_next)
1487 * Given an mbuf chain, make sure we have enough room and then push
1488 * it onto the transmit ring.
1491 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head)
1496 netif_extra_info_t *extra;
1502 * Defragment the mbuf if necessary.
1504 nfrags = xn_count_frags(m_head);
1507 * Check to see whether this request is longer than netback
1508 * can handle, and try to defrag it.
1511 * It is a bit lame, but the netback driver in Linux can't
1512 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of
1513 * the Linux network stack.
1515 if (nfrags > sc->maxfrags) {
1516 m = m_defrag(m_head, M_NOWAIT);
1519 * Defrag failed, so free the mbuf and
1520 * therefore drop the packet.
1528 /* Determine how many fragments now exist */
1529 nfrags = xn_count_frags(m_head);
1532 * Check to see whether the defragmented packet has too many
1533 * segments for the Linux netback driver.
1536 * The FreeBSD TCP stack, with TSO enabled, can produce a chain
1537 * of mbufs longer than Linux can handle. Make sure we don't
1538 * pass a too-long chain over to the other side by dropping the
1539 * packet. It doesn't look like there is currently a way to
1540 * tell the TCP stack to generate a shorter chain of packets.
1542 if (nfrags > MAX_TX_REQ_FRAGS) {
1544 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback "
1545 "won't be able to handle it, dropping\n",
1546 __func__, nfrags, MAX_TX_REQ_FRAGS);
1553 * This check should be redundant. We've already verified that we
1554 * have enough slots in the ring to handle a packet of maximum
1555 * size, and that our packet is less than the maximum size. Keep
1556 * it in here as an assert for now just to make certain that
1557 * xn_tx_chain_cnt is accurate.
1559 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE,
1560 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE "
1561 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt,
1562 (int) nfrags, (int) NET_TX_RING_SIZE));
1565 * Start packing the mbufs in this chain into
1566 * the fragment pointers. Stop when we run out
1567 * of fragments or hit the end of the mbuf chain.
1571 otherend_id = xenbus_get_otherend_id(sc->xbdev);
1572 for (m = m_head; m; m = m->m_next) {
1573 netif_tx_request_t *tx;
1576 u_long mfn; /* XXX Wrong type? */
1578 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt);
1579 id = get_id_from_freelist(sc->tx_mbufs);
1581 panic("%s: was allocated the freelist head!\n",
1583 sc->xn_cdata.xn_tx_chain_cnt++;
1584 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE)
1585 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n",
1587 sc->tx_mbufs[id] = m;
1589 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1590 KASSERT((short)ref >= 0, ("Negative ref"));
1591 mfn = virt_to_mfn(mtod(m, vm_offset_t));
1592 gnttab_grant_foreign_access_ref(ref, otherend_id,
1593 mfn, GNTMAP_readonly);
1594 tx->gref = sc->grant_tx_ref[id] = ref;
1595 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1);
1599 * The first fragment has the entire packet
1600 * size, subsequent fragments have just the
1601 * fragment size. The backend works out the
1602 * true size of the first fragment by
1603 * subtracting the sizes of the other
1606 tx->size = m->m_pkthdr.len;
1609 * The first fragment contains the checksum flags
1610 * and is optionally followed by extra data for
1614 * CSUM_TSO requires checksum offloading.
1615 * Some versions of FreeBSD fail to
1616 * set CSUM_TCP in the CSUM_TSO case,
1617 * so we have to test for CSUM_TSO
1620 if (m->m_pkthdr.csum_flags
1621 & (CSUM_DELAY_DATA | CSUM_TSO)) {
1622 tx->flags |= (NETTXF_csum_blank
1623 | NETTXF_data_validated);
1625 #if __FreeBSD_version >= 700000
1626 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1627 struct netif_extra_info *gso =
1628 (struct netif_extra_info *)
1629 RING_GET_REQUEST(&sc->tx,
1630 ++sc->tx.req_prod_pvt);
1632 tx->flags |= NETTXF_extra_info;
1634 gso->u.gso.size = m->m_pkthdr.tso_segsz;
1636 XEN_NETIF_GSO_TYPE_TCPV4;
1638 gso->u.gso.features = 0;
1640 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
1645 tx->size = m->m_len;
1648 tx->flags |= NETTXF_more_data;
1650 sc->tx.req_prod_pvt++;
1652 BPF_MTAP(ifp, m_head);
1654 sc->stats.tx_bytes += m_head->m_pkthdr.len;
1655 sc->stats.tx_packets++;
1661 xn_start_locked(struct ifnet *ifp)
1663 struct netfront_info *sc;
1664 struct mbuf *m_head;
1669 if (!netfront_carrier_ok(sc))
1673 * While we have enough transmit slots available for at least one
1674 * maximum-sized packet, pull mbufs off the queue and put them on
1675 * the transmit ring.
1677 while (xn_tx_slot_available(sc)) {
1678 IF_DEQUEUE(&ifp->if_snd, m_head);
1682 if (xn_assemble_tx_request(sc, m_head) != 0)
1686 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1688 xen_intr_signal(sc->xen_intr_handle);
1690 if (RING_FULL(&sc->tx)) {
1693 netif_stop_queue(dev);
1699 xn_start(struct ifnet *ifp)
1701 struct netfront_info *sc;
1704 xn_start_locked(ifp);
1708 /* equivalent of network_open() in Linux */
1710 xn_ifinit_locked(struct netfront_info *sc)
1718 if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1723 network_alloc_rx_buffers(sc);
1724 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1726 ifp->if_drv_flags |= IFF_DRV_RUNNING;
1727 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1728 if_link_state_change(ifp, LINK_STATE_UP);
1730 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1734 xn_ifinit(void *xsc)
1736 struct netfront_info *sc = xsc;
1739 xn_ifinit_locked(sc);
1744 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1746 struct netfront_info *sc = ifp->if_softc;
1747 struct ifreq *ifr = (struct ifreq *) data;
1749 struct ifaddr *ifa = (struct ifaddr *)data;
1752 int mask, error = 0;
1758 if (ifa->ifa_addr->sa_family == AF_INET) {
1759 ifp->if_flags |= IFF_UP;
1760 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1761 xn_ifinit_locked(sc);
1762 arp_ifinit(ifp, ifa);
1767 error = ether_ioctl(ifp, cmd, data);
1773 /* XXX can we alter the MTU on a VN ?*/
1775 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1780 ifp->if_mtu = ifr->ifr_mtu;
1781 ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1787 if (ifp->if_flags & IFF_UP) {
1789 * If only the state of the PROMISC flag changed,
1790 * then just use the 'set promisc mode' command
1791 * instead of reinitializing the entire NIC. Doing
1792 * a full re-init means reloading the firmware and
1793 * waiting for it to start up, which may take a
1797 /* No promiscuous mode with Xen */
1798 if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1799 ifp->if_flags & IFF_PROMISC &&
1800 !(sc->xn_if_flags & IFF_PROMISC)) {
1801 XN_SETBIT(sc, XN_RX_MODE,
1802 XN_RXMODE_RX_PROMISC);
1803 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1804 !(ifp->if_flags & IFF_PROMISC) &&
1805 sc->xn_if_flags & IFF_PROMISC) {
1806 XN_CLRBIT(sc, XN_RX_MODE,
1807 XN_RXMODE_RX_PROMISC);
1810 xn_ifinit_locked(sc);
1812 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1816 sc->xn_if_flags = ifp->if_flags;
1821 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1822 if (mask & IFCAP_TXCSUM) {
1823 if (IFCAP_TXCSUM & ifp->if_capenable) {
1824 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4);
1825 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP
1826 | CSUM_IP | CSUM_TSO);
1828 ifp->if_capenable |= IFCAP_TXCSUM;
1829 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP
1833 if (mask & IFCAP_RXCSUM) {
1834 ifp->if_capenable ^= IFCAP_RXCSUM;
1836 #if __FreeBSD_version >= 700000
1837 if (mask & IFCAP_TSO4) {
1838 if (IFCAP_TSO4 & ifp->if_capenable) {
1839 ifp->if_capenable &= ~IFCAP_TSO4;
1840 ifp->if_hwassist &= ~CSUM_TSO;
1841 } else if (IFCAP_TXCSUM & ifp->if_capenable) {
1842 ifp->if_capenable |= IFCAP_TSO4;
1843 ifp->if_hwassist |= CSUM_TSO;
1845 IPRINTK("Xen requires tx checksum offload"
1846 " be enabled to use TSO\n");
1850 if (mask & IFCAP_LRO) {
1851 ifp->if_capenable ^= IFCAP_LRO;
1860 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1870 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
1873 error = ether_ioctl(ifp, cmd, data);
1880 xn_stop(struct netfront_info *sc)
1888 callout_stop(&sc->xn_stat_ch);
1890 xn_free_rx_ring(sc);
1891 xn_free_tx_ring(sc);
1893 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1894 if_link_state_change(ifp, LINK_STATE_DOWN);
1897 /* START of Xenolinux helper functions adapted to FreeBSD */
1899 network_connect(struct netfront_info *np)
1901 int i, requeue_idx, error;
1903 netif_rx_request_t *req;
1904 u_int feature_rx_copy, feature_rx_flip;
1906 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1907 "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1909 feature_rx_copy = 0;
1910 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1911 "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1913 feature_rx_flip = 1;
1916 * Copy packets on receive path if:
1917 * (a) This was requested by user, and the backend supports it; or
1918 * (b) Flipping was requested, but this is unsupported by the backend.
1920 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1921 (MODPARM_rx_flip && !feature_rx_flip));
1923 /* Recovery procedure: */
1924 error = talk_to_backend(np->xbdev, np);
1928 /* Step 1: Reinitialise variables. */
1929 xn_query_features(np);
1930 xn_configure_features(np);
1931 netif_release_tx_bufs(np);
1933 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1934 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1938 if (np->rx_mbufs[i] == NULL)
1941 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1942 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1944 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1945 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1947 if (!np->copying_receiver) {
1948 gnttab_grant_foreign_transfer_ref(ref,
1949 xenbus_get_otherend_id(np->xbdev),
1952 gnttab_grant_foreign_access_ref(ref,
1953 xenbus_get_otherend_id(np->xbdev),
1957 req->id = requeue_idx;
1962 np->rx.req_prod_pvt = requeue_idx;
1964 /* Step 3: All public and private state should now be sane. Get
1965 * ready to start sending and receiving packets and give the driver
1966 * domain a kick because we've probably just requeued some
1969 netfront_carrier_on(np);
1970 xen_intr_signal(np->xen_intr_handle);
1974 network_alloc_rx_buffers(np);
1980 xn_query_features(struct netfront_info *np)
1984 device_printf(np->xbdev, "backend features:");
1986 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1987 "feature-sg", NULL, "%d", &val) < 0)
1992 np->maxfrags = MAX_TX_REQ_FRAGS;
1993 printf(" feature-sg");
1996 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev),
1997 "feature-gso-tcpv4", NULL, "%d", &val) < 0)
2000 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO);
2002 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO;
2003 printf(" feature-gso-tcp4");
2010 xn_configure_features(struct netfront_info *np)
2012 int err, cap_enabled;
2016 if (np->xn_resume &&
2017 ((np->xn_ifp->if_capenable & np->xn_ifp->if_capabilities)
2018 == np->xn_ifp->if_capenable)) {
2019 /* Current options are available, no need to do anything. */
2023 /* Try to preserve as many options as possible. */
2025 cap_enabled = np->xn_ifp->if_capenable;
2027 cap_enabled = UINT_MAX;
2029 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
2030 if ((np->xn_ifp->if_capenable & IFCAP_LRO) == (cap_enabled & IFCAP_LRO))
2031 tcp_lro_free(&np->xn_lro);
2033 np->xn_ifp->if_capenable =
2034 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4) & cap_enabled;
2035 np->xn_ifp->if_hwassist &= ~CSUM_TSO;
2036 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6))
2037 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) ==
2038 (cap_enabled & IFCAP_LRO)) {
2039 err = tcp_lro_init(&np->xn_lro);
2041 device_printf(np->xbdev, "LRO initialization failed\n");
2043 np->xn_lro.ifp = np->xn_ifp;
2044 np->xn_ifp->if_capenable |= IFCAP_LRO;
2047 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) ==
2048 (cap_enabled & IFCAP_TSO4)) {
2049 np->xn_ifp->if_capenable |= IFCAP_TSO4;
2050 np->xn_ifp->if_hwassist |= CSUM_TSO;
2057 * Create a network device.
2058 * @param dev Newbus device representing this virtual NIC.
2061 create_netdev(device_t dev)
2064 struct netfront_info *np;
2068 np = device_get_softc(dev);
2072 XN_LOCK_INIT(np, xennetif);
2074 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts);
2075 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL);
2076 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL);
2078 np->rx_target = RX_MIN_TARGET;
2079 np->rx_min_target = RX_MIN_TARGET;
2080 np->rx_max_target = RX_MAX_TARGET;
2082 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
2083 for (i = 0; i <= NET_TX_RING_SIZE; i++) {
2084 np->tx_mbufs[i] = (void *) ((u_long) i+1);
2085 np->grant_tx_ref[i] = GRANT_REF_INVALID;
2087 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0;
2089 for (i = 0; i <= NET_RX_RING_SIZE; i++) {
2091 np->rx_mbufs[i] = NULL;
2092 np->grant_rx_ref[i] = GRANT_REF_INVALID;
2094 /* A grant for every tx ring slot */
2095 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2096 &np->gref_tx_head) != 0) {
2097 IPRINTK("#### netfront can't alloc tx grant refs\n");
2101 /* A grant for every rx ring slot */
2102 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
2103 &np->gref_rx_head) != 0) {
2104 WPRINTK("#### netfront can't alloc rx grant refs\n");
2105 gnttab_free_grant_references(np->gref_tx_head);
2110 err = xen_net_read_mac(dev, np->mac);
2114 /* Set up ifnet structure */
2115 ifp = np->xn_ifp = if_alloc(IFT_ETHER);
2117 if_initname(ifp, "xn", device_get_unit(dev));
2118 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2119 ifp->if_ioctl = xn_ioctl;
2120 ifp->if_output = ether_output;
2121 ifp->if_start = xn_start;
2123 ifp->if_watchdog = xn_watchdog;
2125 ifp->if_init = xn_ifinit;
2126 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
2128 ifp->if_hwassist = XN_CSUM_FEATURES;
2129 ifp->if_capabilities = IFCAP_HWCSUM;
2130 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2131 ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS;
2132 ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
2134 ether_ifattach(ifp, np->mac);
2135 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
2136 netfront_carrier_off(np);
2141 gnttab_free_grant_references(np->gref_tx_head);
2147 * Handle the change of state of the backend to Closing. We must delete our
2148 * device-layer structures now, to ensure that writes are flushed through to
2149 * the backend. Once is this done, we can switch to Closed in
2154 netfront_closing(device_t dev)
2157 struct netfront_info *info = dev->dev_driver_data;
2159 DPRINTK("netfront_closing: %s removed\n", dev->nodename);
2163 xenbus_switch_state(dev, XenbusStateClosed);
2168 netfront_detach(device_t dev)
2170 struct netfront_info *info = device_get_softc(dev);
2172 DPRINTK("%s\n", xenbus_get_node(dev));
2180 netif_free(struct netfront_info *info)
2185 callout_drain(&info->xn_stat_ch);
2186 netif_disconnect_backend(info);
2187 if (info->xn_ifp != NULL) {
2188 ether_ifdetach(info->xn_ifp);
2189 if_free(info->xn_ifp);
2190 info->xn_ifp = NULL;
2192 ifmedia_removeall(&info->sc_media);
2196 netif_disconnect_backend(struct netfront_info *info)
2200 netfront_carrier_off(info);
2204 free_ring(&info->tx_ring_ref, &info->tx.sring);
2205 free_ring(&info->rx_ring_ref, &info->rx.sring);
2207 xen_intr_unbind(&info->xen_intr_handle);
2211 free_ring(int *ref, void *ring_ptr_ref)
2213 void **ring_ptr_ptr = ring_ptr_ref;
2215 if (*ref != GRANT_REF_INVALID) {
2216 /* This API frees the associated storage. */
2217 gnttab_end_foreign_access(*ref, *ring_ptr_ptr);
2218 *ref = GRANT_REF_INVALID;
2220 *ring_ptr_ptr = NULL;
2224 xn_ifmedia_upd(struct ifnet *ifp)
2230 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2232 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE;
2233 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL;
2236 /* ** Driver registration ** */
2237 static device_method_t netfront_methods[] = {
2238 /* Device interface */
2239 DEVMETHOD(device_probe, netfront_probe),
2240 DEVMETHOD(device_attach, netfront_attach),
2241 DEVMETHOD(device_detach, netfront_detach),
2242 DEVMETHOD(device_shutdown, bus_generic_shutdown),
2243 DEVMETHOD(device_suspend, netfront_suspend),
2244 DEVMETHOD(device_resume, netfront_resume),
2246 /* Xenbus interface */
2247 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed),
2252 static driver_t netfront_driver = {
2255 sizeof(struct netfront_info),
2257 devclass_t netfront_devclass;
2259 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL,