sys/dev/xen/netfront/netfront.c

   1 /*
   2  *
   3  * Copyright (c) 2004-2006 Kip Macy
   4  * All rights reserved.
   5  *
   6  *
   7  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
   8  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
   9  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  10  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  11  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  12  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  13  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  14  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  15  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  16  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  17  */
  18
  19
  20 #include <sys/cdefs.h>
  21 __FBSDID("$FreeBSD$");
  22
  23 #include <sys/param.h>
  24 #include <sys/systm.h>
  25 #include <sys/sockio.h>
  26 #include <sys/mbuf.h>
  27 #include <sys/lock.h>
  28 #include <sys/malloc.h>
  29 #include <sys/module.h>
  30 #include <sys/kernel.h>
  31 #include <sys/socket.h>
  32 #include <sys/queue.h>
  33 #include <sys/sx.h>
  34
  35 #include <net/if.h>
  36 #include <net/if_arp.h>
  37 #include <net/ethernet.h>
  38 #include <net/if_dl.h>
  39 #include <net/if_media.h>
  40
  41 #include <net/bpf.h>
  42
  43 #include <net/if_types.h>
  44 #include <net/if.h>
  45
  46 #include <netinet/in_systm.h>
  47 #include <netinet/in.h>
  48 #include <netinet/ip.h>
  49 #include <netinet/if_ether.h>
  50
  51 #include <vm/vm.h>
  52 #include <vm/pmap.h>
  53
  54 #include <machine/clock.h>      /* for DELAY */
  55 #include <machine/bus.h>
  56 #include <machine/resource.h>
  57 #include <machine/frame.h>
  58 #include <machine/vmparam.h>
  59
  60 #include <sys/bus.h>
  61 #include <sys/rman.h>
  62
  63 #include <machine/intr_machdep.h>
  64
  65 #include <machine/xen/xen-os.h>
  66 #include <xen/hypervisor.h>
  67 #include <xen/xen_intr.h>
  68 #include <xen/evtchn.h>
  69 #include <xen/gnttab.h>
  70 #include <xen/interface/memory.h>
  71 #include <dev/xen/netfront/mbufq.h>
  72 #include <machine/xen/features.h>
  73 #include <xen/interface/io/netif.h>
  74 #include <xen/xenbus/xenbusvar.h>
  75
  76 #include "xenbus_if.h"
  77
  78 #define GRANT_INVALID_REF       0
  79
  80 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE)
  81 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE)
  82
  83 #ifdef CONFIG_XEN
  84 static int MODPARM_rx_copy = 0;
  85 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0);
  86 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)");
  87 static int MODPARM_rx_flip = 0;
  88 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0);
  89 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)");
  90 #else
  91 static const int MODPARM_rx_copy = 1;
  92 static const int MODPARM_rx_flip = 0;
  93 #endif
  94
  95 #define RX_COPY_THRESHOLD 256
  96
  97 #define net_ratelimit() 0
  98
  99 struct netfront_info;
 100 struct netfront_rx_info;
 101
 102 static void xn_txeof(struct netfront_info *);
 103 static void xn_rxeof(struct netfront_info *);
 104 static void network_alloc_rx_buffers(struct netfront_info *);
 105
 106 static void xn_tick_locked(struct netfront_info *);
 107 static void xn_tick(void *);
 108
 109 static void xn_intr(void *);
 110 static void xn_start_locked(struct ifnet *);
 111 static void xn_start(struct ifnet *);
 112 static int  xn_ioctl(struct ifnet *, u_long, caddr_t);
 113 static void xn_ifinit_locked(struct netfront_info *);
 114 static void xn_ifinit(void *);
 115 static void xn_stop(struct netfront_info *);
 116 #ifdef notyet
 117 static void xn_watchdog(struct ifnet *);
 118 #endif
 119
 120 static void show_device(struct netfront_info *sc);
 121 #ifdef notyet
 122 static void netfront_closing(device_t dev);
 123 #endif
 124 static void netif_free(struct netfront_info *info);
 125 static int netfront_detach(device_t dev);
 126
 127 static int talk_to_backend(device_t dev, struct netfront_info *info);
 128 static int create_netdev(device_t dev);
 129 static void netif_disconnect_backend(struct netfront_info *info);
 130 static int setup_device(device_t dev, struct netfront_info *info);
 131 static void end_access(int ref, void *page);
 132
 133 /* Xenolinux helper functions */
 134 int network_connect(struct netfront_info *);
 135
 136 static void xn_free_rx_ring(struct netfront_info *);
 137
 138 static void xn_free_tx_ring(struct netfront_info *);
 139
 140 static int xennet_get_responses(struct netfront_info *np,
 141         struct netfront_rx_info *rinfo, RING_IDX rp, struct mbuf **list,
 142         int *pages_flipped_p);
 143
 144 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT)
 145
 146 #define INVALID_P2M_ENTRY (~0UL)
 147
 148 /*
 149  * Mbuf pointers. We need these to keep track of the virtual addresses
 150  * of our mbuf chains since we can only convert from virtual to physical,
 151  * not the other way around.  The size must track the free index arrays.
 152  */
 153 struct xn_chain_data {
 154                 struct mbuf             *xn_tx_chain[NET_TX_RING_SIZE+1];
 155                 struct mbuf             *xn_rx_chain[NET_RX_RING_SIZE+1];
 156 };
 157
 158
 159 struct net_device_stats
 160 {
 161         u_long  rx_packets;             /* total packets received       */
 162         u_long  tx_packets;             /* total packets transmitted    */
 163         u_long  rx_bytes;               /* total bytes received         */
 164         u_long  tx_bytes;               /* total bytes transmitted      */
 165         u_long  rx_errors;              /* bad packets received         */
 166         u_long  tx_errors;              /* packet transmit problems     */
 167         u_long  rx_dropped;             /* no space in linux buffers    */
 168         u_long  tx_dropped;             /* no space available in linux  */
 169         u_long  multicast;              /* multicast packets received   */
 170         u_long  collisions;
 171
 172         /* detailed rx_errors: */
 173         u_long  rx_length_errors;
 174         u_long  rx_over_errors;         /* receiver ring buff overflow  */
 175         u_long  rx_crc_errors;          /* recved pkt with crc error    */
 176         u_long  rx_frame_errors;        /* recv'd frame alignment error */
 177         u_long  rx_fifo_errors;         /* recv'r fifo overrun          */
 178         u_long  rx_missed_errors;       /* receiver missed packet       */
 179
 180         /* detailed tx_errors */
 181         u_long  tx_aborted_errors;
 182         u_long  tx_carrier_errors;
 183         u_long  tx_fifo_errors;
 184         u_long  tx_heartbeat_errors;
 185         u_long  tx_window_errors;
 186
 187         /* for cslip etc */
 188         u_long  rx_compressed;
 189         u_long  tx_compressed;
 190 };
 191
 192 struct netfront_info {
 193
 194         struct ifnet *xn_ifp;
 195
 196         struct net_device_stats stats;
 197         u_int tx_full;
 198
 199         netif_tx_front_ring_t tx;
 200         netif_rx_front_ring_t rx;
 201
 202         struct mtx   tx_lock;
 203         struct mtx   rx_lock;
 204         struct sx    sc_lock;
 205
 206         u_int handle;
 207         u_int irq;
 208         u_int copying_receiver;
 209         u_int carrier;
 210
 211         /* Receive-ring batched refills. */
 212 #define RX_MIN_TARGET 32
 213 #define RX_MAX_TARGET NET_RX_RING_SIZE
 214         int rx_min_target, rx_max_target, rx_target;
 215
 216         /*
 217          * {tx,rx}_skbs store outstanding skbuffs. The first entry in each
 218          * array is an index into a chain of free entries.
 219          */
 220
 221         grant_ref_t gref_tx_head;
 222         grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1];
 223         grant_ref_t gref_rx_head;
 224         grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1];
 225
 226 #define TX_MAX_TARGET min(NET_RX_RING_SIZE, 256)
 227         device_t xbdev;
 228         int tx_ring_ref;
 229         int rx_ring_ref;
 230         uint8_t mac[ETHER_ADDR_LEN];
 231         struct xn_chain_data    xn_cdata;       /* mbufs */
 232         struct mbuf_head xn_rx_batch;   /* head of the batch queue */
 233
 234         int                     xn_if_flags;
 235         struct callout          xn_stat_ch;
 236
 237         u_long rx_pfn_array[NET_RX_RING_SIZE];
 238         multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1];
 239         mmu_update_t rx_mmu[NET_RX_RING_SIZE];
 240 };
 241
 242 #define rx_mbufs xn_cdata.xn_rx_chain
 243 #define tx_mbufs xn_cdata.xn_tx_chain
 244
 245 #define XN_LOCK_INIT(_sc, _name) \
 246         mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \
 247         mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF);  \
 248         sx_init(&(_sc)->sc_lock, #_name"_rx")
 249
 250 #define XN_RX_LOCK(_sc)           mtx_lock(&(_sc)->rx_lock)
 251 #define XN_RX_UNLOCK(_sc)         mtx_unlock(&(_sc)->rx_lock)
 252
 253 #define XN_TX_LOCK(_sc)           mtx_lock(&(_sc)->tx_lock)
 254 #define XN_TX_UNLOCK(_sc)         mtx_unlock(&(_sc)->tx_lock)
 255
 256 #define XN_LOCK(_sc)           sx_xlock(&(_sc)->sc_lock);
 257 #define XN_UNLOCK(_sc)         sx_xunlock(&(_sc)->sc_lock);
 258
 259 #define XN_LOCK_ASSERT(_sc)    sx_assert(&(_sc)->sc_lock, SX_LOCKED);
 260 #define XN_RX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->rx_lock, MA_OWNED);
 261 #define XN_TX_LOCK_ASSERT(_sc)    mtx_assert(&(_sc)->tx_lock, MA_OWNED);
 262 #define XN_LOCK_DESTROY(_sc)   mtx_destroy(&(_sc)->rx_lock); \
 263                                mtx_destroy(&(_sc)->tx_lock); \
 264                                sx_destroy(&(_sc)->sc_lock);
 265
 266 struct netfront_rx_info {
 267         struct netif_rx_response rx;
 268         struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
 269 };
 270
 271 #define netfront_carrier_on(netif)      ((netif)->carrier = 1)
 272 #define netfront_carrier_off(netif)     ((netif)->carrier = 0)
 273 #define netfront_carrier_ok(netif)      ((netif)->carrier)
 274
 275 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */
 276
 277
 278
 279 /*
 280  * Access macros for acquiring freeing slots in tx_skbs[].
 281  */
 282
 283 static inline void
 284 add_id_to_freelist(struct mbuf **list, unsigned short id)
 285 {
 286         list[id] = list[0];
 287         list[0]  = (void *)(u_long)id;
 288 }
 289
 290 static inline unsigned short
 291 get_id_from_freelist(struct mbuf **list)
 292 {
 293         u_int id = (u_int)(u_long)list[0];
 294         list[0] = list[id];
 295         return (id);
 296 }
 297
 298 static inline int
 299 xennet_rxidx(RING_IDX idx)
 300 {
 301         return idx & (NET_RX_RING_SIZE - 1);
 302 }
 303
 304 static inline struct mbuf *
 305 xennet_get_rx_mbuf(struct netfront_info *np,
 306                                                 RING_IDX ri)
 307 {
 308         int i = xennet_rxidx(ri);
 309         struct mbuf *m;
 310
 311         m = np->rx_mbufs[i];
 312         np->rx_mbufs[i] = NULL;
 313         return (m);
 314 }
 315
 316 static inline grant_ref_t
 317 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri)
 318 {
 319         int i = xennet_rxidx(ri);
 320         grant_ref_t ref = np->grant_rx_ref[i];
 321         np->grant_rx_ref[i] = GRANT_INVALID_REF;
 322         return ref;
 323 }
 324
 325 #ifdef DEBUG
 326
 327 #endif
 328 #define IPRINTK(fmt, args...) \
 329     printf("[XEN] " fmt, ##args)
 330 #define WPRINTK(fmt, args...) \
 331     printf("[XEN] " fmt, ##args)
 332 #define DPRINTK(fmt, args...) \
 333     printf("[XEN] %s: " fmt, __func__, ##args)
 334
 335
 336 static __inline struct mbuf*
 337 makembuf (struct mbuf *buf)
 338 {
 339         struct mbuf *m = NULL;
 340
 341         MGETHDR (m, M_DONTWAIT, MT_DATA);
 342
 343         if (! m)
 344                 return 0;
 345
 346         M_MOVE_PKTHDR(m, buf);
 347
 348         m_cljget(m, M_DONTWAIT, MJUMPAGESIZE);
 349         m->m_pkthdr.len = buf->m_pkthdr.len;
 350         m->m_len = buf->m_len;
 351         m_copydata(buf, 0, buf->m_pkthdr.len, mtod(m,caddr_t) );
 352
 353                 m->m_ext.ext_arg1 = (caddr_t *)(uintptr_t)(vtophys(mtod(m,caddr_t)) >> PAGE_SHIFT);
 354
 355         return m;
 356 }
 357
 358 /**
 359  * Read the 'mac' node at the given device's node in the store, and parse that
 360  * as colon-separated octets, placing result the given mac array.  mac must be
 361  * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h).
 362  * Return 0 on success, or errno on error.
 363  */
 364 static int
 365 xen_net_read_mac(device_t dev, uint8_t mac[])
 366 {
 367         int error, i;
 368         char *s, *e, *macstr;
 369
 370         error = xenbus_read(XBT_NIL, xenbus_get_node(dev), "mac", NULL,
 371             (void **) &macstr);
 372         if (error)
 373                 return (error);
 374
 375         s = macstr;
 376         for (i = 0; i < ETHER_ADDR_LEN; i++) {
 377                 mac[i] = strtoul(s, &e, 16);
 378                 if (s == e || (e[0] != ':' && e[0] != 0)) {
 379                         free(macstr, M_DEVBUF);
 380                         return (ENOENT);
 381                 }
 382                 s = &e[1];
 383         }
 384         free(macstr, M_DEVBUF);
 385         return (0);
 386 }
 387
 388 /**
 389  * Entry point to this code when a new device is created.  Allocate the basic
 390  * structures and the ring buffers for communication with the backend, and
 391  * inform the backend of the appropriate details for those.  Switch to
 392  * Connected state.
 393  */
 394 static int
 395 netfront_probe(device_t dev)
 396 {
 397
 398         if (!strcmp(xenbus_get_type(dev), "vif")) {
 399                 device_set_desc(dev, "Virtual Network Interface");
 400                 return (0);
 401         }
 402
 403         return (ENXIO);
 404 }
 405
 406 static int
 407 netfront_attach(device_t dev)
 408 {
 409         int err;
 410
 411         err = create_netdev(dev);
 412         if (err) {
 413                 xenbus_dev_fatal(dev, err, "creating netdev");
 414                 return err;
 415         }
 416
 417         return 0;
 418 }
 419
 420
 421 /**
 422  * We are reconnecting to the backend, due to a suspend/resume, or a backend
 423  * driver restart.  We tear down our netif structure and recreate it, but
 424  * leave the device-layer structures intact so that this is transparent to the
 425  * rest of the kernel.
 426  */
 427 static int
 428 netfront_resume(device_t dev)
 429 {
 430         struct netfront_info *info = device_get_softc(dev);
 431
 432         netif_disconnect_backend(info);
 433         return (0);
 434 }
 435
 436
 437 /* Common code used when first setting up, and when resuming. */
 438 static int
 439 talk_to_backend(device_t dev, struct netfront_info *info)
 440 {
 441         const char *message;
 442         struct xenbus_transaction xbt;
 443         const char *node = xenbus_get_node(dev);
 444         int err;
 445
 446         err = xen_net_read_mac(dev, info->mac);
 447         if (err) {
 448                 xenbus_dev_fatal(dev, err, "parsing %s/mac", node);
 449                 goto out;
 450         }
 451
 452         /* Create shared ring, alloc event channel. */
 453         err = setup_device(dev, info);
 454         if (err)
 455                 goto out;
 456
 457  again:
 458         err = xenbus_transaction_start(&xbt);
 459         if (err) {
 460                 xenbus_dev_fatal(dev, err, "starting transaction");
 461                 goto destroy_ring;
 462         }
 463         err = xenbus_printf(xbt, node, "tx-ring-ref","%u",
 464                             info->tx_ring_ref);
 465         if (err) {
 466                 message = "writing tx ring-ref";
 467                 goto abort_transaction;
 468         }
 469         err = xenbus_printf(xbt, node, "rx-ring-ref","%u",
 470                             info->rx_ring_ref);
 471         if (err) {
 472                 message = "writing rx ring-ref";
 473                 goto abort_transaction;
 474         }
 475         err = xenbus_printf(xbt, node,
 476                 "event-channel", "%u", irq_to_evtchn_port(info->irq));
 477         if (err) {
 478                 message = "writing event-channel";
 479                 goto abort_transaction;
 480         }
 481         err = xenbus_printf(xbt, node, "request-rx-copy", "%u",
 482                             info->copying_receiver);
 483         if (err) {
 484                 message = "writing request-rx-copy";
 485                 goto abort_transaction;
 486         }
 487         err = xenbus_printf(xbt, node, "feature-rx-notify", "%d", 1);
 488         if (err) {
 489                 message = "writing feature-rx-notify";
 490                 goto abort_transaction;
 491         }
 492         err = xenbus_printf(xbt, node, "feature-no-csum-offload", "%d", 1);
 493         if (err) {
 494                 message = "writing feature-no-csum-offload";
 495                 goto abort_transaction;
 496         }
 497         err = xenbus_printf(xbt, node, "feature-sg", "%d", 1);
 498         if (err) {
 499                 message = "writing feature-sg";
 500                 goto abort_transaction;
 501         }
 502 #ifdef HAVE_TSO
 503         err = xenbus_printf(xbt, node, "feature-gso-tcpv4", "%d", 1);
 504         if (err) {
 505                 message = "writing feature-gso-tcpv4";
 506                 goto abort_transaction;
 507         }
 508 #endif
 509
 510         err = xenbus_transaction_end(xbt, 0);
 511         if (err) {
 512                 if (err == EAGAIN)
 513                         goto again;
 514                 xenbus_dev_fatal(dev, err, "completing transaction");
 515                 goto destroy_ring;
 516         }
 517
 518         return 0;
 519
 520  abort_transaction:
 521         xenbus_transaction_end(xbt, 1);
 522         xenbus_dev_fatal(dev, err, "%s", message);
 523  destroy_ring:
 524         netif_free(info);
 525  out:
 526         return err;
 527 }
 528
 529
 530 static int
 531 setup_device(device_t dev, struct netfront_info *info)
 532 {
 533         netif_tx_sring_t *txs;
 534         netif_rx_sring_t *rxs;
 535         int error;
 536         struct ifnet *ifp;
 537
 538         ifp = info->xn_ifp;
 539
 540         info->tx_ring_ref = GRANT_INVALID_REF;
 541         info->rx_ring_ref = GRANT_INVALID_REF;
 542         info->rx.sring = NULL;
 543         info->tx.sring = NULL;
 544         info->irq = 0;
 545
 546         txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
 547         if (!txs) {
 548                 error = ENOMEM;
 549                 xenbus_dev_fatal(dev, error, "allocating tx ring page");
 550                 goto fail;
 551         }
 552         SHARED_RING_INIT(txs);
 553         FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
 554         error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref);
 555         if (error)
 556                 goto fail;
 557
 558         rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO);
 559         if (!rxs) {
 560                 error = ENOMEM;
 561                 xenbus_dev_fatal(dev, error, "allocating rx ring page");
 562                 goto fail;
 563         }
 564         SHARED_RING_INIT(rxs);
 565         FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
 566
 567         error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref);
 568         if (error)
 569                 goto fail;
 570
 571         error = bind_listening_port_to_irqhandler(xenbus_get_otherend_id(dev),
 572                  "xn", xn_intr, info, INTR_TYPE_NET | INTR_MPSAFE, &info->irq);
 573
 574         if (error) {
 575                 xenbus_dev_fatal(dev, error,
 576                                  "bind_evtchn_to_irqhandler failed");
 577                 goto fail;
 578         }
 579
 580         show_device(info);
 581
 582         return (0);
 583
 584  fail:
 585         netif_free(info);
 586         return (error);
 587 }
 588
 589 /**
 590  * Callback received when the backend's state changes.
 591  */
 592 static void
 593 netfront_backend_changed(device_t dev, XenbusState newstate)
 594 {
 595         struct netfront_info *sc = device_get_softc(dev);
 596
 597         DPRINTK("newstate=%d\n", newstate);
 598
 599         switch (newstate) {
 600         case XenbusStateInitialising:
 601         case XenbusStateInitialised:
 602         case XenbusStateConnected:
 603         case XenbusStateUnknown:
 604         case XenbusStateClosed:
 605         case XenbusStateReconfigured:
 606         case XenbusStateReconfiguring:
 607                 break;
 608         case XenbusStateInitWait:
 609                 if (xenbus_get_state(dev) != XenbusStateInitialising)
 610                         break;
 611                 if (network_connect(sc) != 0)
 612                         break;
 613                 xenbus_set_state(dev, XenbusStateConnected);
 614 #ifdef notyet
 615                 (void)send_fake_arp(netdev);
 616 #endif
 617                 break;
 618         case XenbusStateClosing:
 619                 xenbus_set_state(dev, XenbusStateClosed);
 620                 break;
 621         }
 622 }
 623
 624 static void
 625 xn_free_rx_ring(struct netfront_info *sc)
 626 {
 627 #if 0
 628         int i;
 629
 630         for (i = 0; i < NET_RX_RING_SIZE; i++) {
 631                 if (sc->xn_cdata.xn_rx_chain[i] != NULL) {
 632                         m_freem(sc->xn_cdata.xn_rx_chain[i]);
 633                         sc->xn_cdata.xn_rx_chain[i] = NULL;
 634                 }
 635         }
 636
 637         sc->rx.rsp_cons = 0;
 638         sc->xn_rx_if->req_prod = 0;
 639         sc->xn_rx_if->event = sc->rx.rsp_cons ;
 640 #endif
 641 }
 642
 643 static void
 644 xn_free_tx_ring(struct netfront_info *sc)
 645 {
 646 #if 0
 647         int i;
 648
 649         for (i = 0; i < NET_TX_RING_SIZE; i++) {
 650                 if (sc->xn_cdata.xn_tx_chain[i] != NULL) {
 651                         m_freem(sc->xn_cdata.xn_tx_chain[i]);
 652                         sc->xn_cdata.xn_tx_chain[i] = NULL;
 653                 }
 654         }
 655
 656         return;
 657 #endif
 658 }
 659
 660 static inline int
 661 netfront_tx_slot_available(struct netfront_info *np)
 662 {
 663         return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
 664                 (TX_MAX_TARGET - /* MAX_SKB_FRAGS */ 24 - 2));
 665 }
 666 static void
 667 netif_release_tx_bufs(struct netfront_info *np)
 668 {
 669         struct mbuf *m;
 670         int i;
 671
 672         for (i = 1; i <= NET_TX_RING_SIZE; i++) {
 673                 m = np->xn_cdata.xn_tx_chain[i];
 674
 675                 if (((u_long)m) < KERNBASE)
 676                         continue;
 677                 gnttab_grant_foreign_access_ref(np->grant_tx_ref[i],
 678                     xenbus_get_otherend_id(np->xbdev),
 679                     virt_to_mfn(mtod(m, vm_offset_t)),
 680                     GNTMAP_readonly);
 681                 gnttab_release_grant_reference(&np->gref_tx_head,
 682                     np->grant_tx_ref[i]);
 683                 np->grant_tx_ref[i] = GRANT_INVALID_REF;
 684                 add_id_to_freelist(np->tx_mbufs, i);
 685                 m_freem(m);
 686         }
 687 }
 688
 689 static void
 690 network_alloc_rx_buffers(struct netfront_info *sc)
 691 {
 692         int otherend_id = xenbus_get_otherend_id(sc->xbdev);
 693         unsigned short id;
 694         struct mbuf *m_new;
 695         int i, batch_target, notify;
 696         RING_IDX req_prod;
 697         struct xen_memory_reservation reservation;
 698         grant_ref_t ref;
 699         int nr_flips;
 700         netif_rx_request_t *req;
 701         vm_offset_t vaddr;
 702         u_long pfn;
 703
 704         req_prod = sc->rx.req_prod_pvt;
 705
 706         if (unlikely(sc->carrier == 0))
 707                 return;
 708
 709         /*
 710          * Allocate skbuffs greedily, even though we batch updates to the
 711          * receive ring. This creates a less bursty demand on the memory
 712          * allocator, so should reduce the chance of failed allocation
 713          * requests both for ourself and for other kernel subsystems.
 714          */
 715         batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons);
 716         for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) {
 717                 MGETHDR(m_new, M_DONTWAIT, MT_DATA);
 718                 if (m_new == NULL)
 719                         goto no_mbuf;
 720
 721                 m_cljget(m_new, M_DONTWAIT, MJUMPAGESIZE);
 722                 if ((m_new->m_flags & M_EXT) == 0) {
 723                         m_freem(m_new);
 724
 725 no_mbuf:
 726                         if (i != 0)
 727                                 goto refill;
 728                         /*
 729                          * XXX set timer
 730                          */
 731                         break;
 732                 }
 733                 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE;
 734
 735                 /* queue the mbufs allocated */
 736                 mbufq_tail(&sc->xn_rx_batch, m_new);
 737         }
 738
 739         /* Is the batch large enough to be worthwhile? */
 740         if (i < (sc->rx_target/2)) {
 741                 if (req_prod >sc->rx.sring->req_prod)
 742                         goto push;
 743                 return;
 744         }
 745         /* Adjust floating fill target if we risked running out of buffers. */
 746         if ( ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) &&
 747              ((sc->rx_target *= 2) > sc->rx_max_target) )
 748                 sc->rx_target = sc->rx_max_target;
 749
 750 refill:
 751         for (nr_flips = i = 0; ; i++) {
 752                 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL)
 753                         break;
 754
 755                 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)(
 756                                 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT);
 757
 758                 id = xennet_rxidx(req_prod + i);
 759
 760                 KASSERT(sc->xn_cdata.xn_rx_chain[id] == NULL,
 761                     ("non-NULL xm_rx_chain"));
 762                 sc->xn_cdata.xn_rx_chain[id] = m_new;
 763
 764                 ref = gnttab_claim_grant_reference(&sc->gref_rx_head);
 765                 KASSERT((short)ref >= 0, ("negative ref"));
 766                 sc->grant_rx_ref[id] = ref;
 767
 768                 vaddr = mtod(m_new, vm_offset_t);
 769                 pfn = vtophys(vaddr) >> PAGE_SHIFT;
 770                 req = RING_GET_REQUEST(&sc->rx, req_prod + i);
 771
 772                 if (sc->copying_receiver == 0) {
 773                         gnttab_grant_foreign_transfer_ref(ref,
 774                             otherend_id, pfn);
 775                         sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn);
 776                         if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 777                                 /* Remove this page before passing
 778                                  * back to Xen.
 779                                  */
 780                                 set_phys_to_machine(pfn, INVALID_P2M_ENTRY);
 781                                 MULTI_update_va_mapping(&sc->rx_mcl[i],
 782                                     vaddr, 0, 0);
 783                         }
 784                         nr_flips++;
 785                 } else {
 786                         gnttab_grant_foreign_access_ref(ref,
 787                             otherend_id,
 788                             PFNTOMFN(pfn), 0);
 789                 }
 790                 req->id = id;
 791                 req->gref = ref;
 792
 793                 sc->rx_pfn_array[i] =
 794                     vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT;
 795         }
 796
 797         KASSERT(i, ("no mbufs processed")); /* should have returned earlier */
 798         KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed"));
 799         /*
 800          * We may have allocated buffers which have entries outstanding
 801          * in the page * update queue -- make sure we flush those first!
 802          */
 803         PT_UPDATES_FLUSH();
 804         if (nr_flips != 0) {
 805 #ifdef notyet
 806                 /* Tell the ballon driver what is going on. */
 807                 balloon_update_driver_allowance(i);
 808 #endif
 809                 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array);
 810                 reservation.nr_extents   = i;
 811                 reservation.extent_order = 0;
 812                 reservation.address_bits = 0;
 813                 reservation.domid        = DOMID_SELF;
 814
 815                 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
 816
 817                         /* After all PTEs have been zapped, flush the TLB. */
 818                         sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
 819                             UVMF_TLB_FLUSH|UVMF_ALL;
 820
 821                         /* Give away a batch of pages. */
 822                         sc->rx_mcl[i].op = __HYPERVISOR_memory_op;
 823                         sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
 824                         sc->rx_mcl[i].args[1] =  (u_long)&reservation;
 825                         /* Zap PTEs and give away pages in one big multicall. */
 826                         (void)HYPERVISOR_multicall(sc->rx_mcl, i+1);
 827
 828                         /* Check return status of HYPERVISOR_dom_mem_op(). */
 829                         if (unlikely(sc->rx_mcl[i].result != i))
 830                                 panic("Unable to reduce memory reservation\n");
 831                         } else {
 832                                 if (HYPERVISOR_memory_op(
 833                                     XENMEM_decrease_reservation, &reservation)
 834                                     != i)
 835                                         panic("Unable to reduce memory "
 836                                             "reservation\n");
 837                 }
 838         } else {
 839                 wmb();
 840         }
 841
 842         /* Above is a suitable barrier to ensure backend will see requests. */
 843         sc->rx.req_prod_pvt = req_prod + i;
 844 push:
 845         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify);
 846         if (notify)
 847                 notify_remote_via_irq(sc->irq);
 848 }
 849
 850 static void
 851 xn_rxeof(struct netfront_info *np)
 852 {
 853         struct ifnet *ifp;
 854         struct netfront_rx_info rinfo;
 855         struct netif_rx_response *rx = &rinfo.rx;
 856         struct netif_extra_info *extras = rinfo.extras;
 857         RING_IDX i, rp;
 858         multicall_entry_t *mcl;
 859         struct mbuf *m;
 860         struct mbuf_head rxq, errq;
 861         int err, pages_flipped = 0, work_to_do;
 862
 863         do {
 864                 XN_RX_LOCK_ASSERT(np);
 865                 if (!netfront_carrier_ok(np))
 866                         return;
 867
 868                 mbufq_init(&errq);
 869                 mbufq_init(&rxq);
 870
 871                 ifp = np->xn_ifp;
 872
 873                 rp = np->rx.sring->rsp_prod;
 874                 rmb();  /* Ensure we see queued responses up to 'rp'. */
 875
 876                 i = np->rx.rsp_cons;
 877                 while ((i != rp)) {
 878                         memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
 879                         memset(extras, 0, sizeof(rinfo.extras));
 880
 881                         m = NULL;
 882                         err = xennet_get_responses(np, &rinfo, rp, &m,
 883                             &pages_flipped);
 884
 885                         if (unlikely(err)) {
 886                                 if (m)
 887                                         mbufq_tail(&errq, m);
 888                                 np->stats.rx_errors++;
 889                                 i = np->rx.rsp_cons;
 890                                 continue;
 891                         }
 892
 893                         m->m_pkthdr.rcvif = ifp;
 894                         if ( rx->flags & NETRXF_data_validated ) {
 895                                 /* Tell the stack the checksums are okay */
 896                                 /*
 897                                  * XXX this isn't necessarily the case - need to add
 898                                  * check
 899                                  */
 900
 901                                 m->m_pkthdr.csum_flags |=
 902                                         (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID
 903                                             | CSUM_PSEUDO_HDR);
 904                                 m->m_pkthdr.csum_data = 0xffff;
 905                         }
 906
 907                         np->stats.rx_packets++;
 908                         np->stats.rx_bytes += m->m_pkthdr.len;
 909
 910                         mbufq_tail(&rxq, m);
 911                         np->rx.rsp_cons = ++i;
 912                 }
 913
 914                 if (pages_flipped) {
 915                         /* Some pages are no longer absent... */
 916 #ifdef notyet
 917                         balloon_update_driver_allowance(-pages_flipped);
 918 #endif
 919                         /* Do all the remapping work, and M->P updates, in one big
 920                          * hypercall.
 921                          */
 922                         if (!!xen_feature(XENFEAT_auto_translated_physmap)) {
 923                                 mcl = np->rx_mcl + pages_flipped;
 924                                 mcl->op = __HYPERVISOR_mmu_update;
 925                                 mcl->args[0] = (u_long)np->rx_mmu;
 926                                 mcl->args[1] = pages_flipped;
 927                                 mcl->args[2] = 0;
 928                                 mcl->args[3] = DOMID_SELF;
 929                                 (void)HYPERVISOR_multicall(np->rx_mcl,
 930                                     pages_flipped + 1);
 931                         }
 932                 }
 933
 934                 while ((m = mbufq_dequeue(&errq)))
 935                         m_freem(m);
 936
 937                 /*
 938                  * Process all the mbufs after the remapping is complete.
 939                  * Break the mbuf chain first though.
 940                  */
 941                 while ((m = mbufq_dequeue(&rxq)) != NULL) {
 942                         ifp->if_ipackets++;
 943
 944                         /*
 945                          * Do we really need to drop the rx lock?
 946                          */
 947                         XN_RX_UNLOCK(np);
 948                         /* Pass it up. */
 949                         (*ifp->if_input)(ifp, m);
 950                         XN_RX_LOCK(np);
 951                 }
 952
 953                 np->rx.rsp_cons = i;
 954
 955 #if 0
 956                 /* If we get a callback with very few responses, reduce fill target. */
 957                 /* NB. Note exponential increase, linear decrease. */
 958                 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
 959                         ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target))
 960                         np->rx_target = np->rx_min_target;
 961 #endif
 962
 963                 network_alloc_rx_buffers(np);
 964
 965                 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do);
 966         } while (work_to_do);
 967 }
 968
 969 static void
 970 xn_txeof(struct netfront_info *np)
 971 {
 972         RING_IDX i, prod;
 973         unsigned short id;
 974         struct ifnet *ifp;
 975         struct mbuf *m;
 976
 977         XN_TX_LOCK_ASSERT(np);
 978
 979         if (!netfront_carrier_ok(np))
 980                 return;
 981
 982         ifp = np->xn_ifp;
 983         ifp->if_timer = 0;
 984
 985         do {
 986                 prod = np->tx.sring->rsp_prod;
 987                 rmb(); /* Ensure we see responses up to 'rp'. */
 988
 989                 for (i = np->tx.rsp_cons; i != prod; i++) {
 990                         id = RING_GET_RESPONSE(&np->tx, i)->id;
 991                         m = np->xn_cdata.xn_tx_chain[id];
 992
 993                         ifp->if_opackets++;
 994                         KASSERT(m != NULL, ("mbuf not found in xn_tx_chain"));
 995                         M_ASSERTVALID(m);
 996                         if (unlikely(gnttab_query_foreign_access(
 997                             np->grant_tx_ref[id]) != 0)) {
 998                                 printf("network_tx_buf_gc: warning "
 999                                     "-- grant still in use by backend "
1000                                     "domain.\n");
1001                                 goto out;
1002                         }
1003                         gnttab_end_foreign_access_ref(
1004                                 np->grant_tx_ref[id]);
1005                         gnttab_release_grant_reference(
1006                                 &np->gref_tx_head, np->grant_tx_ref[id]);
1007                         np->grant_tx_ref[id] = GRANT_INVALID_REF;
1008
1009                         np->xn_cdata.xn_tx_chain[id] = NULL;
1010                         add_id_to_freelist(np->xn_cdata.xn_tx_chain, id);
1011                         m_freem(m);
1012                 }
1013                 np->tx.rsp_cons = prod;
1014
1015                 /*
1016                  * Set a new event, then check for race with update of
1017                  * tx_cons. Note that it is essential to schedule a
1018                  * callback, no matter how few buffers are pending. Even if
1019                  * there is space in the transmit ring, higher layers may
1020                  * be blocked because too much data is outstanding: in such
1021                  * cases notification from Xen is likely to be the only kick
1022                  * that we'll get.
1023                  */
1024                 np->tx.sring->rsp_event =
1025                     prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
1026
1027                 mb();
1028
1029         } while (prod != np->tx.sring->rsp_prod);
1030
1031  out:
1032         if (np->tx_full &&
1033             ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) {
1034                 np->tx_full = 0;
1035 #if 0
1036                 if (np->user_state == UST_OPEN)
1037                         netif_wake_queue(dev);
1038 #endif
1039         }
1040
1041 }
1042
1043 static void
1044 xn_intr(void *xsc)
1045 {
1046         struct netfront_info *np = xsc;
1047         struct ifnet *ifp = np->xn_ifp;
1048
1049 #if 0
1050         if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod &&
1051             likely(netfront_carrier_ok(np)) &&
1052             ifp->if_drv_flags & IFF_DRV_RUNNING))
1053                 return;
1054 #endif
1055         if (np->tx.rsp_cons != np->tx.sring->rsp_prod) {
1056                 XN_TX_LOCK(np);
1057                 xn_txeof(np);
1058                 XN_TX_UNLOCK(np);
1059         }
1060
1061         XN_RX_LOCK(np);
1062         xn_rxeof(np);
1063         XN_RX_UNLOCK(np);
1064
1065         if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1066             !IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1067                 xn_start(ifp);
1068 }
1069
1070
1071 static void
1072 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m,
1073         grant_ref_t ref)
1074 {
1075         int new = xennet_rxidx(np->rx.req_prod_pvt);
1076
1077         KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL"));
1078         np->rx_mbufs[new] = m;
1079         np->grant_rx_ref[new] = ref;
1080         RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
1081         RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
1082         np->rx.req_prod_pvt++;
1083 }
1084
1085 static int
1086 xennet_get_extras(struct netfront_info *np,
1087     struct netif_extra_info *extras, RING_IDX rp)
1088 {
1089         struct netif_extra_info *extra;
1090         RING_IDX cons = np->rx.rsp_cons;
1091
1092         int err = 0;
1093
1094         do {
1095                 struct mbuf *m;
1096                 grant_ref_t ref;
1097
1098                 if (unlikely(cons + 1 == rp)) {
1099 #if 0
1100                         if (net_ratelimit())
1101                                 WPRINTK("Missing extra info\n");
1102 #endif
1103                         err = -EINVAL;
1104                         break;
1105                 }
1106
1107                 extra = (struct netif_extra_info *)
1108                 RING_GET_RESPONSE(&np->rx, ++cons);
1109
1110                 if (unlikely(!extra->type ||
1111                         extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1112 #if 0
1113                         if (net_ratelimit())
1114                                 WPRINTK("Invalid extra type: %d\n",
1115                                         extra->type);
1116 #endif
1117                         err = -EINVAL;
1118                 } else {
1119                         memcpy(&extras[extra->type - 1], extra, sizeof(*extra));
1120                 }
1121
1122                 m = xennet_get_rx_mbuf(np, cons);
1123                 ref = xennet_get_rx_ref(np, cons);
1124                 xennet_move_rx_slot(np, m, ref);
1125         } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
1126
1127         np->rx.rsp_cons = cons;
1128         return err;
1129 }
1130
1131 static int
1132 xennet_get_responses(struct netfront_info *np,
1133         struct netfront_rx_info *rinfo, RING_IDX rp,
1134         struct mbuf  **list,
1135         int *pages_flipped_p)
1136 {
1137         int pages_flipped = *pages_flipped_p;
1138         struct mmu_update *mmu;
1139         struct multicall_entry *mcl;
1140         struct netif_rx_response *rx = &rinfo->rx;
1141         struct netif_extra_info *extras = rinfo->extras;
1142         RING_IDX cons = np->rx.rsp_cons;
1143         struct mbuf *m, *m0, *m_prev;
1144         grant_ref_t ref = xennet_get_rx_ref(np, cons);
1145         int max = 5 /* MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD) */;
1146         int frags = 1;
1147         int err = 0;
1148         u_long ret;
1149
1150         m0 = m = m_prev = xennet_get_rx_mbuf(np, cons);
1151
1152
1153         if (rx->flags & NETRXF_extra_info) {
1154                 err = xennet_get_extras(np, extras, rp);
1155                 cons = np->rx.rsp_cons;
1156         }
1157
1158
1159         if (m0 != NULL) {
1160                         m0->m_pkthdr.len = 0;
1161                         m0->m_next = NULL;
1162         }
1163
1164         for (;;) {
1165                 u_long mfn;
1166
1167 #if 0
1168                 printf("rx->status=%hd rx->offset=%hu frags=%u\n",
1169                         rx->status, rx->offset, frags);
1170 #endif
1171                 if (unlikely(rx->status < 0 ||
1172                         rx->offset + rx->status > PAGE_SIZE)) {
1173 #if 0
1174                         if (net_ratelimit())
1175                                 WPRINTK("rx->offset: %x, size: %u\n",
1176                                         rx->offset, rx->status);
1177 #endif
1178                         xennet_move_rx_slot(np, m, ref);
1179                         err = -EINVAL;
1180                         goto next;
1181                 }
1182
1183                 /*
1184                  * This definitely indicates a bug, either in this driver or in
1185                  * the backend driver. In future this should flag the bad
1186                  * situation to the system controller to reboot the backed.
1187                  */
1188                 if (ref == GRANT_INVALID_REF) {
1189 #if 0
1190                         if (net_ratelimit())
1191                                 WPRINTK("Bad rx response id %d.\n", rx->id);
1192 #endif
1193                         err = -EINVAL;
1194                         goto next;
1195                 }
1196
1197                 if (!np->copying_receiver) {
1198                         /* Memory pressure, insufficient buffer
1199                          * headroom, ...
1200                          */
1201                         if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) {
1202                                 if (net_ratelimit())
1203                                         WPRINTK("Unfulfilled rx req "
1204                                                 "(id=%d, st=%d).\n",
1205                                                 rx->id, rx->status);
1206                                 xennet_move_rx_slot(np, m, ref);
1207                                 err = -ENOMEM;
1208                                 goto next;
1209                         }
1210
1211                         if (!xen_feature( XENFEAT_auto_translated_physmap)) {
1212                                 /* Remap the page. */
1213                                 void *vaddr = mtod(m, void *);
1214                                 uint32_t pfn;
1215
1216                                 mcl = np->rx_mcl + pages_flipped;
1217                                 mmu = np->rx_mmu + pages_flipped;
1218
1219                                 MULTI_update_va_mapping(mcl, (u_long)vaddr,
1220                                     (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW |
1221                                     PG_V | PG_M | PG_A, 0);
1222                                 pfn = (uintptr_t)m->m_ext.ext_arg1;
1223                                 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) |
1224                                     MMU_MACHPHYS_UPDATE;
1225                                 mmu->val = pfn;
1226
1227                                 set_phys_to_machine(pfn, mfn);
1228                         }
1229                         pages_flipped++;
1230                 } else {
1231                         ret = gnttab_end_foreign_access_ref(ref);
1232                         KASSERT(ret, ("ret != 0"));
1233                 }
1234
1235                 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1236
1237 next:
1238                 if (m == NULL)
1239                         break;
1240
1241                 m->m_len = rx->status;
1242                 m->m_data += rx->offset;
1243                 m0->m_pkthdr.len += rx->status;
1244
1245                 if (!(rx->flags & NETRXF_more_data))
1246                         break;
1247
1248                 if (cons + frags == rp) {
1249                         if (net_ratelimit())
1250                                 WPRINTK("Need more frags\n");
1251                         err = -ENOENT;
1252                                 break;
1253                 }
1254                 m_prev = m;
1255
1256                 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
1257                 m = xennet_get_rx_mbuf(np, cons + frags);
1258
1259                 m_prev->m_next = m;
1260                 m->m_next = NULL;
1261                 ref = xennet_get_rx_ref(np, cons + frags);
1262                 frags++;
1263         }
1264         *list = m0;
1265
1266         if (unlikely(frags > max)) {
1267                 if (net_ratelimit())
1268                         WPRINTK("Too many frags\n");
1269                 err = -E2BIG;
1270         }
1271
1272         if (unlikely(err))
1273                 np->rx.rsp_cons = cons + frags;
1274
1275         *pages_flipped_p = pages_flipped;
1276
1277         return err;
1278 }
1279
1280 static void
1281 xn_tick_locked(struct netfront_info *sc)
1282 {
1283         XN_RX_LOCK_ASSERT(sc);
1284         callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1285
1286         /* XXX placeholder for printing debug information */
1287
1288 }
1289
1290
1291 static void
1292 xn_tick(void *xsc)
1293 {
1294         struct netfront_info *sc;
1295
1296         sc = xsc;
1297         XN_RX_LOCK(sc);
1298         xn_tick_locked(sc);
1299         XN_RX_UNLOCK(sc);
1300
1301 }
1302 static void
1303 xn_start_locked(struct ifnet *ifp)
1304 {
1305         int otherend_id;
1306         unsigned short id;
1307         struct mbuf *m_head, *new_m;
1308         struct netfront_info *sc;
1309         netif_tx_request_t *tx;
1310         RING_IDX i;
1311         grant_ref_t ref;
1312         u_long mfn, tx_bytes;
1313         int notify;
1314
1315         sc = ifp->if_softc;
1316         otherend_id = xenbus_get_otherend_id(sc->xbdev);
1317         tx_bytes = 0;
1318
1319         if (!netfront_carrier_ok(sc))
1320                 return;
1321
1322         for (i = sc->tx.req_prod_pvt; TRUE; i++) {
1323                 IF_DEQUEUE(&ifp->if_snd, m_head);
1324                 if (m_head == NULL)
1325                         break;
1326
1327                 if (!netfront_tx_slot_available(sc)) {
1328                         IF_PREPEND(&ifp->if_snd, m_head);
1329                         ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1330                         break;
1331                 }
1332
1333                 id = get_id_from_freelist(sc->xn_cdata.xn_tx_chain);
1334
1335                 /*
1336                  * Start packing the mbufs in this chain into
1337                  * the fragment pointers. Stop when we run out
1338                  * of fragments or hit the end of the mbuf chain.
1339                  */
1340                 new_m = makembuf(m_head);
1341                 tx = RING_GET_REQUEST(&sc->tx, i);
1342                 tx->id = id;
1343                 ref = gnttab_claim_grant_reference(&sc->gref_tx_head);
1344                 KASSERT((short)ref >= 0, ("Negative ref"));
1345                 mfn = virt_to_mfn(mtod(new_m, vm_offset_t));
1346                 gnttab_grant_foreign_access_ref(ref, otherend_id,
1347                     mfn, GNTMAP_readonly);
1348                 tx->gref = sc->grant_tx_ref[id] = ref;
1349                 tx->size = new_m->m_pkthdr.len;
1350 #if 0
1351                 tx->flags = (skb->ip_summed == CHECKSUM_HW) ? NETTXF_csum_blank : 0;
1352 #endif
1353                 tx->flags = 0;
1354                 new_m->m_next = NULL;
1355                 new_m->m_nextpkt = NULL;
1356
1357                 m_freem(m_head);
1358
1359                 sc->xn_cdata.xn_tx_chain[id] = new_m;
1360                 BPF_MTAP(ifp, new_m);
1361
1362                 sc->stats.tx_bytes += new_m->m_pkthdr.len;
1363                 sc->stats.tx_packets++;
1364         }
1365
1366         sc->tx.req_prod_pvt = i;
1367         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify);
1368         if (notify)
1369                 notify_remote_via_irq(sc->irq);
1370
1371         xn_txeof(sc);
1372
1373         if (RING_FULL(&sc->tx)) {
1374                 sc->tx_full = 1;
1375 #if 0
1376                 netif_stop_queue(dev);
1377 #endif
1378         }
1379
1380         return;
1381 }
1382
1383 static void
1384 xn_start(struct ifnet *ifp)
1385 {
1386         struct netfront_info *sc;
1387         sc = ifp->if_softc;
1388         XN_TX_LOCK(sc);
1389         xn_start_locked(ifp);
1390         XN_TX_UNLOCK(sc);
1391 }
1392
1393 /* equivalent of network_open() in Linux */
1394 static void
1395 xn_ifinit_locked(struct netfront_info *sc)
1396 {
1397         struct ifnet *ifp;
1398
1399         XN_LOCK_ASSERT(sc);
1400
1401         ifp = sc->xn_ifp;
1402
1403         if (ifp->if_drv_flags & IFF_DRV_RUNNING)
1404                 return;
1405
1406         xn_stop(sc);
1407
1408         network_alloc_rx_buffers(sc);
1409         sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1;
1410
1411         ifp->if_drv_flags |= IFF_DRV_RUNNING;
1412         ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1413
1414         callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc);
1415
1416 }
1417
1418
1419 static void
1420 xn_ifinit(void *xsc)
1421 {
1422         struct netfront_info *sc = xsc;
1423
1424         XN_LOCK(sc);
1425         xn_ifinit_locked(sc);
1426         XN_UNLOCK(sc);
1427
1428 }
1429
1430
1431 static int
1432 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1433 {
1434         struct netfront_info *sc = ifp->if_softc;
1435         struct ifreq *ifr = (struct ifreq *) data;
1436         struct ifaddr *ifa = (struct ifaddr *)data;
1437
1438         int mask, error = 0;
1439         switch(cmd) {
1440         case SIOCSIFADDR:
1441         case SIOCGIFADDR:
1442                 XN_LOCK(sc);
1443                 if (ifa->ifa_addr->sa_family == AF_INET) {
1444                         ifp->if_flags |= IFF_UP;
1445                         if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
1446                                 xn_ifinit_locked(sc);
1447                         arp_ifinit(ifp, ifa);
1448                         XN_UNLOCK(sc);
1449                 } else {
1450                         XN_UNLOCK(sc);
1451                         error = ether_ioctl(ifp, cmd, data);
1452                 }
1453                 break;
1454         case SIOCSIFMTU:
1455                 /* XXX can we alter the MTU on a VN ?*/
1456 #ifdef notyet
1457                 if (ifr->ifr_mtu > XN_JUMBO_MTU)
1458                         error = EINVAL;
1459                 else
1460 #endif
1461                 {
1462                         ifp->if_mtu = ifr->ifr_mtu;
1463                         ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1464                         xn_ifinit(sc);
1465                 }
1466                 break;
1467         case SIOCSIFFLAGS:
1468                 XN_LOCK(sc);
1469                 if (ifp->if_flags & IFF_UP) {
1470                         /*
1471                          * If only the state of the PROMISC flag changed,
1472                          * then just use the 'set promisc mode' command
1473                          * instead of reinitializing the entire NIC. Doing
1474                          * a full re-init means reloading the firmware and
1475                          * waiting for it to start up, which may take a
1476                          * second or two.
1477                          */
1478 #ifdef notyet
1479                         /* No promiscuous mode with Xen */
1480                         if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1481                             ifp->if_flags & IFF_PROMISC &&
1482                             !(sc->xn_if_flags & IFF_PROMISC)) {
1483                                 XN_SETBIT(sc, XN_RX_MODE,
1484                                           XN_RXMODE_RX_PROMISC);
1485                         } else if (ifp->if_drv_flags & IFF_DRV_RUNNING &&
1486                                    !(ifp->if_flags & IFF_PROMISC) &&
1487                                    sc->xn_if_flags & IFF_PROMISC) {
1488                                 XN_CLRBIT(sc, XN_RX_MODE,
1489                                           XN_RXMODE_RX_PROMISC);
1490                         } else
1491 #endif
1492                                 xn_ifinit_locked(sc);
1493                 } else {
1494                         if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1495                                 xn_stop(sc);
1496                         }
1497                 }
1498                 sc->xn_if_flags = ifp->if_flags;
1499                 XN_UNLOCK(sc);
1500                 error = 0;
1501                 break;
1502         case SIOCSIFCAP:
1503                 mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1504                 if (mask & IFCAP_HWCSUM) {
1505                         if (IFCAP_HWCSUM & ifp->if_capenable)
1506                                 ifp->if_capenable &= ~IFCAP_HWCSUM;
1507                         else
1508                                 ifp->if_capenable |= IFCAP_HWCSUM;
1509                 }
1510                 error = 0;
1511                 break;
1512         case SIOCADDMULTI:
1513         case SIOCDELMULTI:
1514 #ifdef notyet
1515                 if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1516                         XN_LOCK(sc);
1517                         xn_setmulti(sc);
1518                         XN_UNLOCK(sc);
1519                         error = 0;
1520                 }
1521 #endif
1522                 /* FALLTHROUGH */
1523         case SIOCSIFMEDIA:
1524         case SIOCGIFMEDIA:
1525                 error = EINVAL;
1526                 break;
1527         default:
1528                 error = ether_ioctl(ifp, cmd, data);
1529         }
1530
1531         return (error);
1532 }
1533
1534 static void
1535 xn_stop(struct netfront_info *sc)
1536 {
1537         struct ifnet *ifp;
1538
1539         XN_LOCK_ASSERT(sc);
1540
1541         ifp = sc->xn_ifp;
1542
1543         callout_stop(&sc->xn_stat_ch);
1544
1545         xn_free_rx_ring(sc);
1546         xn_free_tx_ring(sc);
1547
1548         ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
1549 }
1550
1551 /* START of Xenolinux helper functions adapted to FreeBSD */
1552 int
1553 network_connect(struct netfront_info *np)
1554 {
1555         int i, requeue_idx, error;
1556         grant_ref_t ref;
1557         netif_rx_request_t *req;
1558         u_int feature_rx_copy, feature_rx_flip;
1559
1560         error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1561             "feature-rx-copy", NULL, "%u", &feature_rx_copy);
1562         if (error)
1563                 feature_rx_copy = 0;
1564         error = xenbus_scanf(XBT_NIL, xenbus_get_otherend_path(np->xbdev),
1565             "feature-rx-flip", NULL, "%u", &feature_rx_flip);
1566         if (error)
1567                 feature_rx_flip = 1;
1568
1569         /*
1570          * Copy packets on receive path if:
1571          *  (a) This was requested by user, and the backend supports it; or
1572          *  (b) Flipping was requested, but this is unsupported by the backend.
1573          */
1574         np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) ||
1575                                 (MODPARM_rx_flip && !feature_rx_flip));
1576
1577         XN_LOCK(np);
1578         /* Recovery procedure: */
1579         error = talk_to_backend(np->xbdev, np);
1580         if (error)
1581                 return (error);
1582
1583         /* Step 1: Reinitialise variables. */
1584         netif_release_tx_bufs(np);
1585
1586         /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1587         for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1588                 struct mbuf *m;
1589                 u_long pfn;
1590
1591                 if (np->rx_mbufs[i] == NULL)
1592                         continue;
1593
1594                 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i);
1595                 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1596                 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1597                 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT;
1598
1599                 if (!np->copying_receiver) {
1600                         gnttab_grant_foreign_transfer_ref(ref,
1601                             xenbus_get_otherend_id(np->xbdev),
1602                             pfn);
1603                 } else {
1604                         gnttab_grant_foreign_access_ref(ref,
1605                             xenbus_get_otherend_id(np->xbdev),
1606                             PFNTOMFN(pfn), 0);
1607                 }
1608                 req->gref = ref;
1609                 req->id   = requeue_idx;
1610
1611                 requeue_idx++;
1612         }
1613
1614         np->rx.req_prod_pvt = requeue_idx;
1615
1616         /* Step 3: All public and private state should now be sane.  Get
1617          * ready to start sending and receiving packets and give the driver
1618          * domain a kick because we've probably just requeued some
1619          * packets.
1620          */
1621         netfront_carrier_on(np);
1622         notify_remote_via_irq(np->irq);
1623         XN_TX_LOCK(np);
1624         xn_txeof(np);
1625         XN_TX_UNLOCK(np);
1626         network_alloc_rx_buffers(np);
1627         XN_UNLOCK(np);
1628
1629         return (0);
1630 }
1631
1632 static void
1633 show_device(struct netfront_info *sc)
1634 {
1635 #ifdef DEBUG
1636         if (sc) {
1637                 IPRINTK("<vif handle=%u %s(%s) evtchn=%u irq=%u tx=%p rx=%p>\n",
1638                         sc->xn_ifno,
1639                         be_state_name[sc->xn_backend_state],
1640                         sc->xn_user_state ? "open" : "closed",
1641                         sc->xn_evtchn,
1642                         sc->xn_irq,
1643                         sc->xn_tx_if,
1644                         sc->xn_rx_if);
1645         } else {
1646                 IPRINTK("<vif NULL>\n");
1647         }
1648 #endif
1649 }
1650
1651 /** Create a network device.
1652  * @param handle device handle
1653  */
1654 int
1655 create_netdev(device_t dev)
1656 {
1657         int i;
1658         struct netfront_info *np;
1659         int err;
1660         struct ifnet *ifp;
1661
1662         np = device_get_softc(dev);
1663
1664         np->xbdev         = dev;
1665
1666         XN_LOCK_INIT(np, xennetif);
1667         np->rx_target     = RX_MIN_TARGET;
1668         np->rx_min_target = RX_MIN_TARGET;
1669         np->rx_max_target = RX_MAX_TARGET;
1670
1671         /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */
1672         for (i = 0; i <= NET_TX_RING_SIZE; i++) {
1673                 np->tx_mbufs[i] = (void *) ((u_long) i+1);
1674                 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1675         }
1676         for (i = 0; i <= NET_RX_RING_SIZE; i++) {
1677                 np->rx_mbufs[i] = NULL;
1678                 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1679         }
1680         /* A grant for every tx ring slot */
1681         if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1682                                           &np->gref_tx_head) < 0) {
1683                 printf("#### netfront can't alloc tx grant refs\n");
1684                 err = ENOMEM;
1685                 goto exit;
1686         }
1687         /* A grant for every rx ring slot */
1688         if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1689                                           &np->gref_rx_head) < 0) {
1690                 printf("#### netfront can't alloc rx grant refs\n");
1691                 gnttab_free_grant_references(np->gref_tx_head);
1692                 err = ENOMEM;
1693                 goto exit;
1694         }
1695
1696         err = xen_net_read_mac(dev, np->mac);
1697         if (err) {
1698                 xenbus_dev_fatal(dev, err, "parsing %s/mac",
1699                     xenbus_get_node(dev));
1700                 goto out;
1701         }
1702
1703         /* Set up ifnet structure */
1704         ifp = np->xn_ifp = if_alloc(IFT_ETHER);
1705         ifp->if_softc = np;
1706         if_initname(ifp, "xn",  device_get_unit(dev));
1707         ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1708         ifp->if_ioctl = xn_ioctl;
1709         ifp->if_output = ether_output;
1710         ifp->if_start = xn_start;
1711 #ifdef notyet
1712         ifp->if_watchdog = xn_watchdog;
1713 #endif
1714         ifp->if_init = xn_ifinit;
1715         ifp->if_mtu = ETHERMTU;
1716         ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1;
1717
1718 #ifdef notyet
1719         ifp->if_hwassist = XN_CSUM_FEATURES;
1720         ifp->if_capabilities = IFCAP_HWCSUM;
1721         ifp->if_capenable = ifp->if_capabilities;
1722 #endif
1723
1724         ether_ifattach(ifp, np->mac);
1725         callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE);
1726         netfront_carrier_off(np);
1727
1728         return (0);
1729
1730 exit:
1731         gnttab_free_grant_references(np->gref_tx_head);
1732 out:
1733         panic("do something smart");
1734
1735 }
1736
1737 /**
1738  * Handle the change of state of the backend to Closing.  We must delete our
1739  * device-layer structures now, to ensure that writes are flushed through to
1740  * the backend.  Once is this done, we can switch to Closed in
1741  * acknowledgement.
1742  */
1743 #if 0
1744 static void netfront_closing(device_t dev)
1745 {
1746 #if 0
1747         struct netfront_info *info = dev->dev_driver_data;
1748
1749         DPRINTK("netfront_closing: %s removed\n", dev->nodename);
1750
1751         close_netdev(info);
1752 #endif
1753         xenbus_switch_state(dev, XenbusStateClosed);
1754 }
1755 #endif
1756
1757 static int netfront_detach(device_t dev)
1758 {
1759         struct netfront_info *info = device_get_softc(dev);
1760
1761         DPRINTK("%s\n", xenbus_get_node(dev));
1762
1763         netif_free(info);
1764
1765         return 0;
1766 }
1767
1768
1769 static void netif_free(struct netfront_info *info)
1770 {
1771         netif_disconnect_backend(info);
1772 #if 0
1773         close_netdev(info);
1774 #endif
1775 }
1776
1777 static void netif_disconnect_backend(struct netfront_info *info)
1778 {
1779         XN_RX_LOCK(info);
1780         XN_TX_LOCK(info);
1781         netfront_carrier_off(info);
1782         XN_TX_UNLOCK(info);
1783         XN_RX_UNLOCK(info);
1784
1785         end_access(info->tx_ring_ref, info->tx.sring);
1786         end_access(info->rx_ring_ref, info->rx.sring);
1787         info->tx_ring_ref = GRANT_INVALID_REF;
1788         info->rx_ring_ref = GRANT_INVALID_REF;
1789         info->tx.sring = NULL;
1790         info->rx.sring = NULL;
1791
1792         if (info->irq)
1793                 unbind_from_irqhandler(info->irq);
1794
1795         info->irq = 0;
1796 }
1797
1798
1799 static void end_access(int ref, void *page)
1800 {
1801         if (ref != GRANT_INVALID_REF)
1802                 gnttab_end_foreign_access(ref, page);
1803 }
1804
1805 /* ** Driver registration ** */
1806 static device_method_t netfront_methods[] = {
1807         /* Device interface */
1808         DEVMETHOD(device_probe,         netfront_probe),
1809         DEVMETHOD(device_attach,        netfront_attach),
1810         DEVMETHOD(device_detach,        netfront_detach),
1811         DEVMETHOD(device_shutdown,      bus_generic_shutdown),
1812         DEVMETHOD(device_suspend,       bus_generic_suspend),
1813         DEVMETHOD(device_resume,        netfront_resume),
1814
1815         /* Xenbus interface */
1816         DEVMETHOD(xenbus_backend_changed, netfront_backend_changed),
1817
1818         { 0, 0 }
1819 };
1820
1821 static driver_t netfront_driver = {
1822         "xn",
1823         netfront_methods,
1824         sizeof(struct netfront_info),
1825 };
1826 devclass_t netfront_devclass;
1827
1828 DRIVER_MODULE(xe, xenbus, netfront_driver, netfront_devclass, 0, 0);