1 /*- 2 * Copyright (c) 2004-2006 Kip Macy 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 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. 13 * 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 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include "opt_inet.h" 31 #include "opt_inet6.h" 32 33 #include <sys/param.h> 34 #include <sys/systm.h> 35 #include <sys/sockio.h> 36 #include <sys/limits.h> 37 #include <sys/mbuf.h> 38 #include <sys/malloc.h> 39 #include <sys/module.h> 40 #include <sys/kernel.h> 41 #include <sys/socket.h> 42 #include <sys/sysctl.h> 43 #include <sys/queue.h> 44 #include <sys/lock.h> 45 #include <sys/sx.h> 46 47 #include <net/if.h> 48 #include <net/if_var.h> 49 #include <net/if_arp.h> 50 #include <net/ethernet.h> 51 #include <net/if_dl.h> 52 #include <net/if_media.h> 53 54 #include <net/bpf.h> 55 56 #include <net/if_types.h> 57 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> 65 #endif 66 67 #include <vm/vm.h> 68 #include <vm/pmap.h> 69 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> 75 76 #include <sys/bus.h> 77 #include <sys/rman.h> 78 79 #include <machine/intr_machdep.h> 80 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> 88 89 #include <machine/xen/xenvar.h> 90 91 #include "xenbus_if.h" 92 93 /* Features supported by all backends. TSO and LRO can be negotiated */ 94 #define XN_CSUM_FEATURES (CSUM_TCP | CSUM_UDP) 95 96 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE) 97 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE) 98 99 #if __FreeBSD_version >= 700000 100 /* 101 * Should the driver do LRO on the RX end 102 * this can be toggled on the fly, but the 103 * interface must be reset (down/up) for it 104 * to take effect. 105 */ 106 static int xn_enable_lro = 1; 107 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro); 108 #else 109 110 #define IFCAP_TSO4 0 111 #define CSUM_TSO 0 112 113 #endif 114 115 #ifdef CONFIG_XEN 116 static int MODPARM_rx_copy = 0; 117 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0); 118 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)"); 119 static int MODPARM_rx_flip = 0; 120 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0); 121 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)"); 122 #else 123 static const int MODPARM_rx_copy = 1; 124 static const int MODPARM_rx_flip = 0; 125 #endif 126 127 /** 128 * \brief The maximum allowed data fragments in a single transmit 129 * request. 130 * 131 * This limit is imposed by the backend driver. We assume here that 132 * we are dealing with a Linux driver domain and have set our limit 133 * to mirror the Linux MAX_SKB_FRAGS constant. 134 */ 135 #define MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2) 136 137 #define RX_COPY_THRESHOLD 256 138 139 #define net_ratelimit() 0 140 141 struct netfront_info; 142 struct netfront_rx_info; 143 144 static void xn_txeof(struct netfront_info *); 145 static void xn_rxeof(struct netfront_info *); 146 static void network_alloc_rx_buffers(struct netfront_info *); 147 148 static void xn_tick_locked(struct netfront_info *); 149 static void xn_tick(void *); 150 151 static void xn_intr(void *); 152 static inline int xn_count_frags(struct mbuf *m); 153 static int xn_assemble_tx_request(struct netfront_info *sc, 154 struct mbuf *m_head); 155 static void xn_start_locked(struct ifnet *); 156 static void xn_start(struct ifnet *); 157 static int xn_ioctl(struct ifnet *, u_long, caddr_t); 158 static void xn_ifinit_locked(struct netfront_info *); 159 static void xn_ifinit(void *); 160 static void xn_stop(struct netfront_info *); 161 static void xn_query_features(struct netfront_info *np); 162 static int xn_configure_features(struct netfront_info *np); 163 #ifdef notyet 164 static void xn_watchdog(struct ifnet *); 165 #endif 166 167 #ifdef notyet 168 static void netfront_closing(device_t dev); 169 #endif 170 static void netif_free(struct netfront_info *info); 171 static int netfront_detach(device_t dev); 172 173 static int talk_to_backend(device_t dev, struct netfront_info *info); 174 static int create_netdev(device_t dev); 175 static void netif_disconnect_backend(struct netfront_info *info); 176 static int setup_device(device_t dev, struct netfront_info *info); 177 static void free_ring(int *ref, void *ring_ptr_ref); 178 179 static int xn_ifmedia_upd(struct ifnet *ifp); 180 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr); 181 182 /* Xenolinux helper functions */ 183 int network_connect(struct netfront_info *); 184 185 static void xn_free_rx_ring(struct netfront_info *); 186 187 static void xn_free_tx_ring(struct netfront_info *); 188 189 static int xennet_get_responses(struct netfront_info *np, 190 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, 191 struct mbuf **list, int *pages_flipped_p); 192 193 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT) 194 195 #define INVALID_P2M_ENTRY (~0UL) 196 197 /* 198 * Mbuf pointers. We need these to keep track of the virtual addresses 199 * of our mbuf chains since we can only convert from virtual to physical, 200 * not the other way around. The size must track the free index arrays. 201 */ 202 struct xn_chain_data { 203 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1]; 204 int xn_tx_chain_cnt; 205 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1]; 206 }; 207 208 struct net_device_stats 209 { 210 u_long rx_packets; /* total packets received */ 211 u_long tx_packets; /* total packets transmitted */ 212 u_long rx_bytes; /* total bytes received */ 213 u_long tx_bytes; /* total bytes transmitted */ 214 u_long rx_errors; /* bad packets received */ 215 u_long tx_errors; /* packet transmit problems */ 216 u_long rx_dropped; /* no space in linux buffers */ 217 u_long tx_dropped; /* no space available in linux */ 218 u_long multicast; /* multicast packets received */ 219 u_long collisions; 220 221 /* detailed rx_errors: */ 222 u_long rx_length_errors; 223 u_long rx_over_errors; /* receiver ring buff overflow */ 224 u_long rx_crc_errors; /* recved pkt with crc error */ 225 u_long rx_frame_errors; /* recv'd frame alignment error */ 226 u_long rx_fifo_errors; /* recv'r fifo overrun */ 227 u_long rx_missed_errors; /* receiver missed packet */ 228 229 /* detailed tx_errors */ 230 u_long tx_aborted_errors; 231 u_long tx_carrier_errors; 232 u_long tx_fifo_errors; 233 u_long tx_heartbeat_errors; 234 u_long tx_window_errors; 235 236 /* for cslip etc */ 237 u_long rx_compressed; 238 u_long tx_compressed; 239 }; 240 241 struct netfront_info { 242 struct ifnet *xn_ifp; 243 #if __FreeBSD_version >= 700000 244 struct lro_ctrl xn_lro; 245 #endif 246 247 struct net_device_stats stats; 248 u_int tx_full; 249 250 netif_tx_front_ring_t tx; 251 netif_rx_front_ring_t rx; 252 253 struct mtx tx_lock; 254 struct mtx rx_lock; 255 struct mtx sc_lock; 256 257 xen_intr_handle_t xen_intr_handle; 258 u_int copying_receiver; 259 u_int carrier; 260 u_int maxfrags; 261 262 /* Receive-ring batched refills. */ 263 #define RX_MIN_TARGET 32 264 #define RX_MAX_TARGET NET_RX_RING_SIZE 265 int rx_min_target; 266 int rx_max_target; 267 int rx_target; 268 269 grant_ref_t gref_tx_head; 270 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1]; 271 grant_ref_t gref_rx_head; 272 grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1]; 273 274 device_t xbdev; 275 int tx_ring_ref; 276 int rx_ring_ref; 277 uint8_t mac[ETHER_ADDR_LEN]; 278 struct xn_chain_data xn_cdata; /* mbufs */ 279 struct mbufq xn_rx_batch; /* batch queue */ 280 281 int xn_if_flags; 282 struct callout xn_stat_ch; 283 284 u_long rx_pfn_array[NET_RX_RING_SIZE]; 285 multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1]; 286 mmu_update_t rx_mmu[NET_RX_RING_SIZE]; 287 struct ifmedia sc_media; 288 }; 289 290 #define rx_mbufs xn_cdata.xn_rx_chain 291 #define tx_mbufs xn_cdata.xn_tx_chain 292 293 #define XN_LOCK_INIT(_sc, _name) \ 294 mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \ 295 mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF); \ 296 mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF) 297 298 #define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock) 299 #define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock) 300 301 #define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock) 302 #define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock) 303 304 #define XN_LOCK(_sc) mtx_lock(&(_sc)->sc_lock); 305 #define XN_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_lock); 306 307 #define XN_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->sc_lock, MA_OWNED); 308 #define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED); 309 #define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED); 310 #define XN_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->rx_lock); \ 311 mtx_destroy(&(_sc)->tx_lock); \ 312 mtx_destroy(&(_sc)->sc_lock); 313 314 struct netfront_rx_info { 315 struct netif_rx_response rx; 316 struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; 317 }; 318 319 #define netfront_carrier_on(netif) ((netif)->carrier = 1) 320 #define netfront_carrier_off(netif) ((netif)->carrier = 0) 321 #define netfront_carrier_ok(netif) ((netif)->carrier) 322 323 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */ 324 325 static inline void 326 add_id_to_freelist(struct mbuf **list, uintptr_t id) 327 { 328 KASSERT(id != 0, 329 ("%s: the head item (0) must always be free.", __func__)); 330 list[id] = list[0]; 331 list[0] = (struct mbuf *)id; 332 } 333 334 static inline unsigned short 335 get_id_from_freelist(struct mbuf **list) 336 { 337 uintptr_t id; 338 339 id = (uintptr_t)list[0]; 340 KASSERT(id != 0, 341 ("%s: the head item (0) must always remain free.", __func__)); 342 list[0] = list[id]; 343 return (id); 344 } 345 346 static inline int 347 xennet_rxidx(RING_IDX idx) 348 { 349 return idx & (NET_RX_RING_SIZE - 1); 350 } 351 352 static inline struct mbuf * 353 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri) 354 { 355 int i = xennet_rxidx(ri); 356 struct mbuf *m; 357 358 m = np->rx_mbufs[i]; 359 np->rx_mbufs[i] = NULL; 360 return (m); 361 } 362 363 static inline grant_ref_t 364 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri) 365 { 366 int i = xennet_rxidx(ri); 367 grant_ref_t ref = np->grant_rx_ref[i]; 368 KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n")); 369 np->grant_rx_ref[i] = GRANT_REF_INVALID; 370 return ref; 371 } 372 373 #define IPRINTK(fmt, args...) \ 374 printf("[XEN] " fmt, ##args) 375 #ifdef INVARIANTS 376 #define WPRINTK(fmt, args...) \ 377 printf("[XEN] " fmt, ##args) 378 #else 379 #define WPRINTK(fmt, args...) 380 #endif 381 #ifdef DEBUG 382 #define DPRINTK(fmt, args...) \ 383 printf("[XEN] %s: " fmt, __func__, ##args) 384 #else 385 #define DPRINTK(fmt, args...) 386 #endif 387 388 /** 389 * Read the 'mac' node at the given device's node in the store, and parse that 390 * as colon-separated octets, placing result the given mac array. mac must be 391 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h). 392 * Return 0 on success, or errno on error. 393 */ 394 static int 395 xen_net_read_mac(device_t dev, uint8_t mac[]) 396 { 397 int error, i; 398 char *s, *e, *macstr; 399 const char *path; 400 401 path = xenbus_get_node(dev); 402 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr); 403 if (error == ENOENT) { 404 /* 405 * Deal with missing mac XenStore nodes on devices with 406 * HVM emulation (the 'ioemu' configuration attribute) 407 * enabled. 408 * 409 * The HVM emulator may execute in a stub device model 410 * domain which lacks the permission, only given to Dom0, 411 * to update the guest's XenStore tree. For this reason, 412 * the HVM emulator doesn't even attempt to write the 413 * front-side mac node, even when operating in Dom0. 414 * However, there should always be a mac listed in the 415 * backend tree. Fallback to this version if our query 416 * of the front side XenStore location doesn't find 417 * anything. 418 */ 419 path = xenbus_get_otherend_path(dev); 420 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr); 421 } 422 if (error != 0) { 423 xenbus_dev_fatal(dev, error, "parsing %s/mac", path); 424 return (error); 425 } 426 427 s = macstr; 428 for (i = 0; i < ETHER_ADDR_LEN; i++) { 429 mac[i] = strtoul(s, &e, 16); 430 if (s == e || (e[0] != ':' && e[0] != 0)) { 431 free(macstr, M_XENBUS); 432 return (ENOENT); 433 } 434 s = &e[1]; 435 } 436 free(macstr, M_XENBUS); 437 return (0); 438 } 439 440 /** 441 * Entry point to this code when a new device is created. Allocate the basic 442 * structures and the ring buffers for communication with the backend, and 443 * inform the backend of the appropriate details for those. Switch to 444 * Connected state. 445 */ 446 static int 447 netfront_probe(device_t dev) 448 { 449 450 if (!strcmp(xenbus_get_type(dev), "vif")) { 451 device_set_desc(dev, "Virtual Network Interface"); 452 return (0); 453 } 454 455 return (ENXIO); 456 } 457 458 static int 459 netfront_attach(device_t dev) 460 { 461 int err; 462 463 err = create_netdev(dev); 464 if (err) { 465 xenbus_dev_fatal(dev, err, "creating netdev"); 466 return (err); 467 } 468 469 #if __FreeBSD_version >= 700000 470 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev), 471 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), 472 OID_AUTO, "enable_lro", CTLFLAG_RW, 473 &xn_enable_lro, 0, "Large Receive Offload"); 474 #endif 475 476 return (0); 477 } 478 479 static int 480 netfront_suspend(device_t dev) 481 { 482 struct netfront_info *info = device_get_softc(dev); 483 484 XN_RX_LOCK(info); 485 XN_TX_LOCK(info); 486 netfront_carrier_off(info); 487 XN_TX_UNLOCK(info); 488 XN_RX_UNLOCK(info); 489 return (0); 490 } 491 492 /** 493 * We are reconnecting to the backend, due to a suspend/resume, or a backend 494 * driver restart. We tear down our netif structure and recreate it, but 495 * leave the device-layer structures intact so that this is transparent to the 496 * rest of the kernel. 497 */ 498 static int 499 netfront_resume(device_t dev) 500 { 501 struct netfront_info *info = device_get_softc(dev); 502 503 netif_disconnect_backend(info); 504 return (0); 505 } 506 507 /* Common code used when first setting up, and when resuming. */ 508 static int 509 talk_to_backend(device_t dev, struct netfront_info *info) 510 { 511 const char *message; 512 struct xs_transaction xst; 513 const char *node = xenbus_get_node(dev); 514 int err; 515 516 err = xen_net_read_mac(dev, info->mac); 517 if (err) { 518 xenbus_dev_fatal(dev, err, "parsing %s/mac", node); 519 goto out; 520 } 521 522 /* Create shared ring, alloc event channel. */ 523 err = setup_device(dev, info); 524 if (err) 525 goto out; 526 527 again: 528 err = xs_transaction_start(&xst); 529 if (err) { 530 xenbus_dev_fatal(dev, err, "starting transaction"); 531 goto destroy_ring; 532 } 533 err = xs_printf(xst, node, "tx-ring-ref","%u", 534 info->tx_ring_ref); 535 if (err) { 536 message = "writing tx ring-ref"; 537 goto abort_transaction; 538 } 539 err = xs_printf(xst, node, "rx-ring-ref","%u", 540 info->rx_ring_ref); 541 if (err) { 542 message = "writing rx ring-ref"; 543 goto abort_transaction; 544 } 545 err = xs_printf(xst, node, 546 "event-channel", "%u", 547 xen_intr_port(info->xen_intr_handle)); 548 if (err) { 549 message = "writing event-channel"; 550 goto abort_transaction; 551 } 552 err = xs_printf(xst, node, "request-rx-copy", "%u", 553 info->copying_receiver); 554 if (err) { 555 message = "writing request-rx-copy"; 556 goto abort_transaction; 557 } 558 err = xs_printf(xst, node, "feature-rx-notify", "%d", 1); 559 if (err) { 560 message = "writing feature-rx-notify"; 561 goto abort_transaction; 562 } 563 err = xs_printf(xst, node, "feature-sg", "%d", 1); 564 if (err) { 565 message = "writing feature-sg"; 566 goto abort_transaction; 567 } 568 #if __FreeBSD_version >= 700000 569 err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1); 570 if (err) { 571 message = "writing feature-gso-tcpv4"; 572 goto abort_transaction; 573 } 574 #endif 575 576 err = xs_transaction_end(xst, 0); 577 if (err) { 578 if (err == EAGAIN) 579 goto again; 580 xenbus_dev_fatal(dev, err, "completing transaction"); 581 goto destroy_ring; 582 } 583 584 return 0; 585 586 abort_transaction: 587 xs_transaction_end(xst, 1); 588 xenbus_dev_fatal(dev, err, "%s", message); 589 destroy_ring: 590 netif_free(info); 591 out: 592 return err; 593 } 594 595 static int 596 setup_device(device_t dev, struct netfront_info *info) 597 { 598 netif_tx_sring_t *txs; 599 netif_rx_sring_t *rxs; 600 int error; 601 struct ifnet *ifp; 602 603 ifp = info->xn_ifp; 604 605 info->tx_ring_ref = GRANT_REF_INVALID; 606 info->rx_ring_ref = GRANT_REF_INVALID; 607 info->rx.sring = NULL; 608 info->tx.sring = NULL; 609 610 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO); 611 if (!txs) { 612 error = ENOMEM; 613 xenbus_dev_fatal(dev, error, "allocating tx ring page"); 614 goto fail; 615 } 616 SHARED_RING_INIT(txs); 617 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); 618 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref); 619 if (error) 620 goto fail; 621 622 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO); 623 if (!rxs) { 624 error = ENOMEM; 625 xenbus_dev_fatal(dev, error, "allocating rx ring page"); 626 goto fail; 627 } 628 SHARED_RING_INIT(rxs); 629 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); 630 631 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref); 632 if (error) 633 goto fail; 634 635 error = xen_intr_alloc_and_bind_local_port(dev, 636 xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info, 637 INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle); 638 639 if (error) { 640 xenbus_dev_fatal(dev, error, 641 "xen_intr_alloc_and_bind_local_port failed"); 642 goto fail; 643 } 644 645 return (0); 646 647 fail: 648 netif_free(info); 649 return (error); 650 } 651 652 #ifdef INET 653 /** 654 * If this interface has an ipv4 address, send an arp for it. This 655 * helps to get the network going again after migrating hosts. 656 */ 657 static void 658 netfront_send_fake_arp(device_t dev, struct netfront_info *info) 659 { 660 struct ifnet *ifp; 661 struct ifaddr *ifa; 662 663 ifp = info->xn_ifp; 664 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 665 if (ifa->ifa_addr->sa_family == AF_INET) { 666 arp_ifinit(ifp, ifa); 667 } 668 } 669 } 670 #endif 671 672 /** 673 * Callback received when the backend's state changes. 674 */ 675 static void 676 netfront_backend_changed(device_t dev, XenbusState newstate) 677 { 678 struct netfront_info *sc = device_get_softc(dev); 679 680 DPRINTK("newstate=%d\n", newstate); 681 682 switch (newstate) { 683 case XenbusStateInitialising: 684 case XenbusStateInitialised: 685 case XenbusStateConnected: 686 case XenbusStateUnknown: 687 case XenbusStateClosed: 688 case XenbusStateReconfigured: 689 case XenbusStateReconfiguring: 690 break; 691 case XenbusStateInitWait: 692 if (xenbus_get_state(dev) != XenbusStateInitialising) 693 break; 694 if (network_connect(sc) != 0) 695 break; 696 xenbus_set_state(dev, XenbusStateConnected); 697 #ifdef INET 698 netfront_send_fake_arp(dev, sc); 699 #endif 700 break; 701 case XenbusStateClosing: 702 xenbus_set_state(dev, XenbusStateClosed); 703 break; 704 } 705 } 706 707 static void 708 xn_free_rx_ring(struct netfront_info *sc) 709 { 710 #if 0 711 int i; 712 713 for (i = 0; i < NET_RX_RING_SIZE; i++) { 714 if (sc->xn_cdata.rx_mbufs[i] != NULL) { 715 m_freem(sc->rx_mbufs[i]); 716 sc->rx_mbufs[i] = NULL; 717 } 718 } 719 720 sc->rx.rsp_cons = 0; 721 sc->xn_rx_if->req_prod = 0; 722 sc->xn_rx_if->event = sc->rx.rsp_cons ; 723 #endif 724 } 725 726 static void 727 xn_free_tx_ring(struct netfront_info *sc) 728 { 729 #if 0 730 int i; 731 732 for (i = 0; i < NET_TX_RING_SIZE; i++) { 733 if (sc->tx_mbufs[i] != NULL) { 734 m_freem(sc->tx_mbufs[i]); 735 sc->xn_cdata.xn_tx_chain[i] = NULL; 736 } 737 } 738 739 return; 740 #endif 741 } 742 743 /** 744 * \brief Verify that there is sufficient space in the Tx ring 745 * buffer for a maximally sized request to be enqueued. 746 * 747 * A transmit request requires a transmit descriptor for each packet 748 * fragment, plus up to 2 entries for "options" (e.g. TSO). 749 */ 750 static inline int 751 xn_tx_slot_available(struct netfront_info *np) 752 { 753 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2)); 754 } 755 756 static void 757 netif_release_tx_bufs(struct netfront_info *np) 758 { 759 int i; 760 761 for (i = 1; i <= NET_TX_RING_SIZE; i++) { 762 struct mbuf *m; 763 764 m = np->tx_mbufs[i]; 765 766 /* 767 * We assume that no kernel addresses are 768 * less than NET_TX_RING_SIZE. Any entry 769 * in the table that is below this number 770 * must be an index from free-list tracking. 771 */ 772 if (((uintptr_t)m) <= NET_TX_RING_SIZE) 773 continue; 774 gnttab_end_foreign_access_ref(np->grant_tx_ref[i]); 775 gnttab_release_grant_reference(&np->gref_tx_head, 776 np->grant_tx_ref[i]); 777 np->grant_tx_ref[i] = GRANT_REF_INVALID; 778 add_id_to_freelist(np->tx_mbufs, i); 779 np->xn_cdata.xn_tx_chain_cnt--; 780 if (np->xn_cdata.xn_tx_chain_cnt < 0) { 781 panic("%s: tx_chain_cnt must be >= 0", __func__); 782 } 783 m_free(m); 784 } 785 } 786 787 static void 788 network_alloc_rx_buffers(struct netfront_info *sc) 789 { 790 int otherend_id = xenbus_get_otherend_id(sc->xbdev); 791 unsigned short id; 792 struct mbuf *m_new; 793 int i, batch_target, notify; 794 RING_IDX req_prod; 795 struct xen_memory_reservation reservation; 796 grant_ref_t ref; 797 int nr_flips; 798 netif_rx_request_t *req; 799 vm_offset_t vaddr; 800 u_long pfn; 801 802 req_prod = sc->rx.req_prod_pvt; 803 804 if (__predict_false(sc->carrier == 0)) 805 return; 806 807 /* 808 * Allocate mbufs greedily, even though we batch updates to the 809 * receive ring. This creates a less bursty demand on the memory 810 * allocator, and so should reduce the chance of failed allocation 811 * requests both for ourself and for other kernel subsystems. 812 * 813 * Here we attempt to maintain rx_target buffers in flight, counting 814 * buffers that we have yet to process in the receive ring. 815 */ 816 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons); 817 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) { 818 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE); 819 if (m_new == NULL) { 820 if (i != 0) 821 goto refill; 822 /* 823 * XXX set timer 824 */ 825 break; 826 } 827 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE; 828 829 /* queue the mbufs allocated */ 830 (void )mbufq_enqueue(&sc->xn_rx_batch, m_new); 831 } 832 833 /* 834 * If we've allocated at least half of our target number of entries, 835 * submit them to the backend - we have enough to make the overhead 836 * of submission worthwhile. Otherwise wait for more mbufs and 837 * request entries to become available. 838 */ 839 if (i < (sc->rx_target/2)) { 840 if (req_prod >sc->rx.sring->req_prod) 841 goto push; 842 return; 843 } 844 845 /* 846 * Double floating fill target if we risked having the backend 847 * run out of empty buffers for receive traffic. We define "running 848 * low" as having less than a fourth of our target buffers free 849 * at the time we refilled the queue. 850 */ 851 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) { 852 sc->rx_target *= 2; 853 if (sc->rx_target > sc->rx_max_target) 854 sc->rx_target = sc->rx_max_target; 855 } 856 857 refill: 858 for (nr_flips = i = 0; ; i++) { 859 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL) 860 break; 861 862 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)( 863 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT); 864 865 id = xennet_rxidx(req_prod + i); 866 867 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain")); 868 sc->rx_mbufs[id] = m_new; 869 870 ref = gnttab_claim_grant_reference(&sc->gref_rx_head); 871 KASSERT(ref != GNTTAB_LIST_END, 872 ("reserved grant references exhuasted")); 873 sc->grant_rx_ref[id] = ref; 874 875 vaddr = mtod(m_new, vm_offset_t); 876 pfn = vtophys(vaddr) >> PAGE_SHIFT; 877 req = RING_GET_REQUEST(&sc->rx, req_prod + i); 878 879 if (sc->copying_receiver == 0) { 880 gnttab_grant_foreign_transfer_ref(ref, 881 otherend_id, pfn); 882 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn); 883 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 884 /* Remove this page before passing 885 * back to Xen. 886 */ 887 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 888 MULTI_update_va_mapping(&sc->rx_mcl[i], 889 vaddr, 0, 0); 890 } 891 nr_flips++; 892 } else { 893 gnttab_grant_foreign_access_ref(ref, 894 otherend_id, 895 PFNTOMFN(pfn), 0); 896 } 897 req->id = id; 898 req->gref = ref; 899 900 sc->rx_pfn_array[i] = 901 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT; 902 } 903 904 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */ 905 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed")); 906 /* 907 * We may have allocated buffers which have entries outstanding 908 * in the page * update queue -- make sure we flush those first! 909 */ 910 PT_UPDATES_FLUSH(); 911 if (nr_flips != 0) { 912 #ifdef notyet 913 /* Tell the ballon driver what is going on. */ 914 balloon_update_driver_allowance(i); 915 #endif 916 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array); 917 reservation.nr_extents = i; 918 reservation.extent_order = 0; 919 reservation.address_bits = 0; 920 reservation.domid = DOMID_SELF; 921 922 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 923 /* After all PTEs have been zapped, flush the TLB. */ 924 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] = 925 UVMF_TLB_FLUSH|UVMF_ALL; 926 927 /* Give away a batch of pages. */ 928 sc->rx_mcl[i].op = __HYPERVISOR_memory_op; 929 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation; 930 sc->rx_mcl[i].args[1] = (u_long)&reservation; 931 /* Zap PTEs and give away pages in one big multicall. */ 932 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1); 933 934 if (__predict_false(sc->rx_mcl[i].result != i || 935 HYPERVISOR_memory_op(XENMEM_decrease_reservation, 936 &reservation) != i)) 937 panic("%s: unable to reduce memory " 938 "reservation\n", __func__); 939 } 940 } else { 941 wmb(); 942 } 943 944 /* Above is a suitable barrier to ensure backend will see requests. */ 945 sc->rx.req_prod_pvt = req_prod + i; 946 push: 947 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify); 948 if (notify) 949 xen_intr_signal(sc->xen_intr_handle); 950 } 951 952 static void 953 xn_rxeof(struct netfront_info *np) 954 { 955 struct ifnet *ifp; 956 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 957 struct lro_ctrl *lro = &np->xn_lro; 958 struct lro_entry *queued; 959 #endif 960 struct netfront_rx_info rinfo; 961 struct netif_rx_response *rx = &rinfo.rx; 962 struct netif_extra_info *extras = rinfo.extras; 963 RING_IDX i, rp; 964 multicall_entry_t *mcl; 965 struct mbuf *m; 966 struct mbufq rxq, errq; 967 int err, pages_flipped = 0, work_to_do; 968 969 do { 970 XN_RX_LOCK_ASSERT(np); 971 if (!netfront_carrier_ok(np)) 972 return; 973 974 /* XXX: there should be some sane limit. */ 975 mbufq_init(&errq, INT_MAX); 976 mbufq_init(&rxq, INT_MAX); 977 978 ifp = np->xn_ifp; 979 980 rp = np->rx.sring->rsp_prod; 981 rmb(); /* Ensure we see queued responses up to 'rp'. */ 982 983 i = np->rx.rsp_cons; 984 while ((i != rp)) { 985 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); 986 memset(extras, 0, sizeof(rinfo.extras)); 987 988 m = NULL; 989 err = xennet_get_responses(np, &rinfo, rp, &i, &m, 990 &pages_flipped); 991 992 if (__predict_false(err)) { 993 if (m) 994 (void )mbufq_enqueue(&errq, m); 995 np->stats.rx_errors++; 996 continue; 997 } 998 999 m->m_pkthdr.rcvif = ifp; 1000 if ( rx->flags & NETRXF_data_validated ) { 1001 /* Tell the stack the checksums are okay */ 1002 /* 1003 * XXX this isn't necessarily the case - need to add 1004 * check 1005 */ 1006 1007 m->m_pkthdr.csum_flags |= 1008 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID 1009 | CSUM_PSEUDO_HDR); 1010 m->m_pkthdr.csum_data = 0xffff; 1011 } 1012 1013 np->stats.rx_packets++; 1014 np->stats.rx_bytes += m->m_pkthdr.len; 1015 1016 (void )mbufq_enqueue(&rxq, m); 1017 np->rx.rsp_cons = i; 1018 } 1019 1020 if (pages_flipped) { 1021 /* Some pages are no longer absent... */ 1022 #ifdef notyet 1023 balloon_update_driver_allowance(-pages_flipped); 1024 #endif 1025 /* Do all the remapping work, and M->P updates, in one big 1026 * hypercall. 1027 */ 1028 if (!!xen_feature(XENFEAT_auto_translated_physmap)) { 1029 mcl = np->rx_mcl + pages_flipped; 1030 mcl->op = __HYPERVISOR_mmu_update; 1031 mcl->args[0] = (u_long)np->rx_mmu; 1032 mcl->args[1] = pages_flipped; 1033 mcl->args[2] = 0; 1034 mcl->args[3] = DOMID_SELF; 1035 (void)HYPERVISOR_multicall(np->rx_mcl, 1036 pages_flipped + 1); 1037 } 1038 } 1039 1040 mbufq_drain(&errq); 1041 1042 /* 1043 * Process all the mbufs after the remapping is complete. 1044 * Break the mbuf chain first though. 1045 */ 1046 while ((m = mbufq_dequeue(&rxq)) != NULL) { 1047 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); 1048 1049 /* 1050 * Do we really need to drop the rx lock? 1051 */ 1052 XN_RX_UNLOCK(np); 1053 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1054 /* Use LRO if possible */ 1055 if ((ifp->if_capenable & IFCAP_LRO) == 0 || 1056 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) { 1057 /* 1058 * If LRO fails, pass up to the stack 1059 * directly. 1060 */ 1061 (*ifp->if_input)(ifp, m); 1062 } 1063 #else 1064 (*ifp->if_input)(ifp, m); 1065 #endif 1066 XN_RX_LOCK(np); 1067 } 1068 1069 np->rx.rsp_cons = i; 1070 1071 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1072 /* 1073 * Flush any outstanding LRO work 1074 */ 1075 while (!SLIST_EMPTY(&lro->lro_active)) { 1076 queued = SLIST_FIRST(&lro->lro_active); 1077 SLIST_REMOVE_HEAD(&lro->lro_active, next); 1078 tcp_lro_flush(lro, queued); 1079 } 1080 #endif 1081 1082 #if 0 1083 /* If we get a callback with very few responses, reduce fill target. */ 1084 /* NB. Note exponential increase, linear decrease. */ 1085 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > 1086 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target)) 1087 np->rx_target = np->rx_min_target; 1088 #endif 1089 1090 network_alloc_rx_buffers(np); 1091 1092 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do); 1093 } while (work_to_do); 1094 } 1095 1096 static void 1097 xn_txeof(struct netfront_info *np) 1098 { 1099 RING_IDX i, prod; 1100 unsigned short id; 1101 struct ifnet *ifp; 1102 netif_tx_response_t *txr; 1103 struct mbuf *m; 1104 1105 XN_TX_LOCK_ASSERT(np); 1106 1107 if (!netfront_carrier_ok(np)) 1108 return; 1109 1110 ifp = np->xn_ifp; 1111 1112 do { 1113 prod = np->tx.sring->rsp_prod; 1114 rmb(); /* Ensure we see responses up to 'rp'. */ 1115 1116 for (i = np->tx.rsp_cons; i != prod; i++) { 1117 txr = RING_GET_RESPONSE(&np->tx, i); 1118 if (txr->status == NETIF_RSP_NULL) 1119 continue; 1120 1121 if (txr->status != NETIF_RSP_OKAY) { 1122 printf("%s: WARNING: response is %d!\n", 1123 __func__, txr->status); 1124 } 1125 id = txr->id; 1126 m = np->tx_mbufs[id]; 1127 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain")); 1128 KASSERT((uintptr_t)m > NET_TX_RING_SIZE, 1129 ("mbuf already on the free list, but we're " 1130 "trying to free it again!")); 1131 M_ASSERTVALID(m); 1132 1133 /* 1134 * Increment packet count if this is the last 1135 * mbuf of the chain. 1136 */ 1137 if (!m->m_next) 1138 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 1139 if (__predict_false(gnttab_query_foreign_access( 1140 np->grant_tx_ref[id]) != 0)) { 1141 panic("%s: grant id %u still in use by the " 1142 "backend", __func__, id); 1143 } 1144 gnttab_end_foreign_access_ref( 1145 np->grant_tx_ref[id]); 1146 gnttab_release_grant_reference( 1147 &np->gref_tx_head, np->grant_tx_ref[id]); 1148 np->grant_tx_ref[id] = GRANT_REF_INVALID; 1149 1150 np->tx_mbufs[id] = NULL; 1151 add_id_to_freelist(np->tx_mbufs, id); 1152 np->xn_cdata.xn_tx_chain_cnt--; 1153 m_free(m); 1154 /* Only mark the queue active if we've freed up at least one slot to try */ 1155 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1156 } 1157 np->tx.rsp_cons = prod; 1158 1159 /* 1160 * Set a new event, then check for race with update of 1161 * tx_cons. Note that it is essential to schedule a 1162 * callback, no matter how few buffers are pending. Even if 1163 * there is space in the transmit ring, higher layers may 1164 * be blocked because too much data is outstanding: in such 1165 * cases notification from Xen is likely to be the only kick 1166 * that we'll get. 1167 */ 1168 np->tx.sring->rsp_event = 1169 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; 1170 1171 mb(); 1172 } while (prod != np->tx.sring->rsp_prod); 1173 1174 if (np->tx_full && 1175 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) { 1176 np->tx_full = 0; 1177 #if 0 1178 if (np->user_state == UST_OPEN) 1179 netif_wake_queue(dev); 1180 #endif 1181 } 1182 } 1183 1184 static void 1185 xn_intr(void *xsc) 1186 { 1187 struct netfront_info *np = xsc; 1188 struct ifnet *ifp = np->xn_ifp; 1189 1190 #if 0 1191 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod && 1192 likely(netfront_carrier_ok(np)) && 1193 ifp->if_drv_flags & IFF_DRV_RUNNING)) 1194 return; 1195 #endif 1196 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) { 1197 XN_TX_LOCK(np); 1198 xn_txeof(np); 1199 XN_TX_UNLOCK(np); 1200 } 1201 1202 XN_RX_LOCK(np); 1203 xn_rxeof(np); 1204 XN_RX_UNLOCK(np); 1205 1206 if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1207 !IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 1208 xn_start(ifp); 1209 } 1210 1211 static void 1212 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m, 1213 grant_ref_t ref) 1214 { 1215 int new = xennet_rxidx(np->rx.req_prod_pvt); 1216 1217 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL")); 1218 np->rx_mbufs[new] = m; 1219 np->grant_rx_ref[new] = ref; 1220 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; 1221 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; 1222 np->rx.req_prod_pvt++; 1223 } 1224 1225 static int 1226 xennet_get_extras(struct netfront_info *np, 1227 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons) 1228 { 1229 struct netif_extra_info *extra; 1230 1231 int err = 0; 1232 1233 do { 1234 struct mbuf *m; 1235 grant_ref_t ref; 1236 1237 if (__predict_false(*cons + 1 == rp)) { 1238 #if 0 1239 if (net_ratelimit()) 1240 WPRINTK("Missing extra info\n"); 1241 #endif 1242 err = EINVAL; 1243 break; 1244 } 1245 1246 extra = (struct netif_extra_info *) 1247 RING_GET_RESPONSE(&np->rx, ++(*cons)); 1248 1249 if (__predict_false(!extra->type || 1250 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 1251 #if 0 1252 if (net_ratelimit()) 1253 WPRINTK("Invalid extra type: %d\n", 1254 extra->type); 1255 #endif 1256 err = EINVAL; 1257 } else { 1258 memcpy(&extras[extra->type - 1], extra, sizeof(*extra)); 1259 } 1260 1261 m = xennet_get_rx_mbuf(np, *cons); 1262 ref = xennet_get_rx_ref(np, *cons); 1263 xennet_move_rx_slot(np, m, ref); 1264 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 1265 1266 return err; 1267 } 1268 1269 static int 1270 xennet_get_responses(struct netfront_info *np, 1271 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, 1272 struct mbuf **list, 1273 int *pages_flipped_p) 1274 { 1275 int pages_flipped = *pages_flipped_p; 1276 struct mmu_update *mmu; 1277 struct multicall_entry *mcl; 1278 struct netif_rx_response *rx = &rinfo->rx; 1279 struct netif_extra_info *extras = rinfo->extras; 1280 struct mbuf *m, *m0, *m_prev; 1281 grant_ref_t ref = xennet_get_rx_ref(np, *cons); 1282 RING_IDX ref_cons = *cons; 1283 int frags = 1; 1284 int err = 0; 1285 u_long ret; 1286 1287 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons); 1288 1289 if (rx->flags & NETRXF_extra_info) { 1290 err = xennet_get_extras(np, extras, rp, cons); 1291 } 1292 1293 if (m0 != NULL) { 1294 m0->m_pkthdr.len = 0; 1295 m0->m_next = NULL; 1296 } 1297 1298 for (;;) { 1299 u_long mfn; 1300 1301 #if 0 1302 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n", 1303 rx->status, rx->offset, frags); 1304 #endif 1305 if (__predict_false(rx->status < 0 || 1306 rx->offset + rx->status > PAGE_SIZE)) { 1307 1308 #if 0 1309 if (net_ratelimit()) 1310 WPRINTK("rx->offset: %x, size: %u\n", 1311 rx->offset, rx->status); 1312 #endif 1313 xennet_move_rx_slot(np, m, ref); 1314 if (m0 == m) 1315 m0 = NULL; 1316 m = NULL; 1317 err = EINVAL; 1318 goto next_skip_queue; 1319 } 1320 1321 /* 1322 * This definitely indicates a bug, either in this driver or in 1323 * the backend driver. In future this should flag the bad 1324 * situation to the system controller to reboot the backed. 1325 */ 1326 if (ref == GRANT_REF_INVALID) { 1327 1328 #if 0 1329 if (net_ratelimit()) 1330 WPRINTK("Bad rx response id %d.\n", rx->id); 1331 #endif 1332 printf("%s: Bad rx response id %d.\n", __func__,rx->id); 1333 err = EINVAL; 1334 goto next; 1335 } 1336 1337 if (!np->copying_receiver) { 1338 /* Memory pressure, insufficient buffer 1339 * headroom, ... 1340 */ 1341 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) { 1342 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n", 1343 rx->id, rx->status); 1344 xennet_move_rx_slot(np, m, ref); 1345 err = ENOMEM; 1346 goto next; 1347 } 1348 1349 if (!xen_feature( XENFEAT_auto_translated_physmap)) { 1350 /* Remap the page. */ 1351 void *vaddr = mtod(m, void *); 1352 uint32_t pfn; 1353 1354 mcl = np->rx_mcl + pages_flipped; 1355 mmu = np->rx_mmu + pages_flipped; 1356 1357 MULTI_update_va_mapping(mcl, (u_long)vaddr, 1358 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW | 1359 PG_V | PG_M | PG_A, 0); 1360 pfn = (uintptr_t)m->m_ext.ext_arg1; 1361 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) | 1362 MMU_MACHPHYS_UPDATE; 1363 mmu->val = pfn; 1364 1365 set_phys_to_machine(pfn, mfn); 1366 } 1367 pages_flipped++; 1368 } else { 1369 ret = gnttab_end_foreign_access_ref(ref); 1370 KASSERT(ret, ("ret != 0")); 1371 } 1372 1373 gnttab_release_grant_reference(&np->gref_rx_head, ref); 1374 1375 next: 1376 if (m == NULL) 1377 break; 1378 1379 m->m_len = rx->status; 1380 m->m_data += rx->offset; 1381 m0->m_pkthdr.len += rx->status; 1382 1383 next_skip_queue: 1384 if (!(rx->flags & NETRXF_more_data)) 1385 break; 1386 1387 if (*cons + frags == rp) { 1388 if (net_ratelimit()) 1389 WPRINTK("Need more frags\n"); 1390 err = ENOENT; 1391 printf("%s: cons %u frags %u rp %u, not enough frags\n", 1392 __func__, *cons, frags, rp); 1393 break; 1394 } 1395 /* 1396 * Note that m can be NULL, if rx->status < 0 or if 1397 * rx->offset + rx->status > PAGE_SIZE above. 1398 */ 1399 m_prev = m; 1400 1401 rx = RING_GET_RESPONSE(&np->rx, *cons + frags); 1402 m = xennet_get_rx_mbuf(np, *cons + frags); 1403 1404 /* 1405 * m_prev == NULL can happen if rx->status < 0 or if 1406 * rx->offset + * rx->status > PAGE_SIZE above. 1407 */ 1408 if (m_prev != NULL) 1409 m_prev->m_next = m; 1410 1411 /* 1412 * m0 can be NULL if rx->status < 0 or if * rx->offset + 1413 * rx->status > PAGE_SIZE above. 1414 */ 1415 if (m0 == NULL) 1416 m0 = m; 1417 m->m_next = NULL; 1418 ref = xennet_get_rx_ref(np, *cons + frags); 1419 ref_cons = *cons + frags; 1420 frags++; 1421 } 1422 *list = m0; 1423 *cons += frags; 1424 *pages_flipped_p = pages_flipped; 1425 1426 return (err); 1427 } 1428 1429 static void 1430 xn_tick_locked(struct netfront_info *sc) 1431 { 1432 XN_RX_LOCK_ASSERT(sc); 1433 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); 1434 1435 /* XXX placeholder for printing debug information */ 1436 } 1437 1438 static void 1439 xn_tick(void *xsc) 1440 { 1441 struct netfront_info *sc; 1442 1443 sc = xsc; 1444 XN_RX_LOCK(sc); 1445 xn_tick_locked(sc); 1446 XN_RX_UNLOCK(sc); 1447 } 1448 1449 /** 1450 * \brief Count the number of fragments in an mbuf chain. 1451 * 1452 * Surprisingly, there isn't an M* macro for this. 1453 */ 1454 static inline int 1455 xn_count_frags(struct mbuf *m) 1456 { 1457 int nfrags; 1458 1459 for (nfrags = 0; m != NULL; m = m->m_next) 1460 nfrags++; 1461 1462 return (nfrags); 1463 } 1464 1465 /** 1466 * Given an mbuf chain, make sure we have enough room and then push 1467 * it onto the transmit ring. 1468 */ 1469 static int 1470 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) 1471 { 1472 struct ifnet *ifp; 1473 struct mbuf *m; 1474 u_int nfrags; 1475 int otherend_id; 1476 1477 ifp = sc->xn_ifp; 1478 1479 /** 1480 * Defragment the mbuf if necessary. 1481 */ 1482 nfrags = xn_count_frags(m_head); 1483 1484 /* 1485 * Check to see whether this request is longer than netback 1486 * can handle, and try to defrag it. 1487 */ 1488 /** 1489 * It is a bit lame, but the netback driver in Linux can't 1490 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of 1491 * the Linux network stack. 1492 */ 1493 if (nfrags > sc->maxfrags) { 1494 m = m_defrag(m_head, M_NOWAIT); 1495 if (!m) { 1496 /* 1497 * Defrag failed, so free the mbuf and 1498 * therefore drop the packet. 1499 */ 1500 m_freem(m_head); 1501 return (EMSGSIZE); 1502 } 1503 m_head = m; 1504 } 1505 1506 /* Determine how many fragments now exist */ 1507 nfrags = xn_count_frags(m_head); 1508 1509 /* 1510 * Check to see whether the defragmented packet has too many 1511 * segments for the Linux netback driver. 1512 */ 1513 /** 1514 * The FreeBSD TCP stack, with TSO enabled, can produce a chain 1515 * of mbufs longer than Linux can handle. Make sure we don't 1516 * pass a too-long chain over to the other side by dropping the 1517 * packet. It doesn't look like there is currently a way to 1518 * tell the TCP stack to generate a shorter chain of packets. 1519 */ 1520 if (nfrags > MAX_TX_REQ_FRAGS) { 1521 #ifdef DEBUG 1522 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback " 1523 "won't be able to handle it, dropping\n", 1524 __func__, nfrags, MAX_TX_REQ_FRAGS); 1525 #endif 1526 m_freem(m_head); 1527 return (EMSGSIZE); 1528 } 1529 1530 /* 1531 * This check should be redundant. We've already verified that we 1532 * have enough slots in the ring to handle a packet of maximum 1533 * size, and that our packet is less than the maximum size. Keep 1534 * it in here as an assert for now just to make certain that 1535 * xn_tx_chain_cnt is accurate. 1536 */ 1537 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE, 1538 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE " 1539 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt, 1540 (int) nfrags, (int) NET_TX_RING_SIZE)); 1541 1542 /* 1543 * Start packing the mbufs in this chain into 1544 * the fragment pointers. Stop when we run out 1545 * of fragments or hit the end of the mbuf chain. 1546 */ 1547 m = m_head; 1548 otherend_id = xenbus_get_otherend_id(sc->xbdev); 1549 for (m = m_head; m; m = m->m_next) { 1550 netif_tx_request_t *tx; 1551 uintptr_t id; 1552 grant_ref_t ref; 1553 u_long mfn; /* XXX Wrong type? */ 1554 1555 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt); 1556 id = get_id_from_freelist(sc->tx_mbufs); 1557 if (id == 0) 1558 panic("%s: was allocated the freelist head!\n", 1559 __func__); 1560 sc->xn_cdata.xn_tx_chain_cnt++; 1561 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE) 1562 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n", 1563 __func__); 1564 sc->tx_mbufs[id] = m; 1565 tx->id = id; 1566 ref = gnttab_claim_grant_reference(&sc->gref_tx_head); 1567 KASSERT((short)ref >= 0, ("Negative ref")); 1568 mfn = virt_to_mfn(mtod(m, vm_offset_t)); 1569 gnttab_grant_foreign_access_ref(ref, otherend_id, 1570 mfn, GNTMAP_readonly); 1571 tx->gref = sc->grant_tx_ref[id] = ref; 1572 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1); 1573 tx->flags = 0; 1574 if (m == m_head) { 1575 /* 1576 * The first fragment has the entire packet 1577 * size, subsequent fragments have just the 1578 * fragment size. The backend works out the 1579 * true size of the first fragment by 1580 * subtracting the sizes of the other 1581 * fragments. 1582 */ 1583 tx->size = m->m_pkthdr.len; 1584 1585 /* 1586 * The first fragment contains the checksum flags 1587 * and is optionally followed by extra data for 1588 * TSO etc. 1589 */ 1590 /** 1591 * CSUM_TSO requires checksum offloading. 1592 * Some versions of FreeBSD fail to 1593 * set CSUM_TCP in the CSUM_TSO case, 1594 * so we have to test for CSUM_TSO 1595 * explicitly. 1596 */ 1597 if (m->m_pkthdr.csum_flags 1598 & (CSUM_DELAY_DATA | CSUM_TSO)) { 1599 tx->flags |= (NETTXF_csum_blank 1600 | NETTXF_data_validated); 1601 } 1602 #if __FreeBSD_version >= 700000 1603 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 1604 struct netif_extra_info *gso = 1605 (struct netif_extra_info *) 1606 RING_GET_REQUEST(&sc->tx, 1607 ++sc->tx.req_prod_pvt); 1608 1609 tx->flags |= NETTXF_extra_info; 1610 1611 gso->u.gso.size = m->m_pkthdr.tso_segsz; 1612 gso->u.gso.type = 1613 XEN_NETIF_GSO_TYPE_TCPV4; 1614 gso->u.gso.pad = 0; 1615 gso->u.gso.features = 0; 1616 1617 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 1618 gso->flags = 0; 1619 } 1620 #endif 1621 } else { 1622 tx->size = m->m_len; 1623 } 1624 if (m->m_next) 1625 tx->flags |= NETTXF_more_data; 1626 1627 sc->tx.req_prod_pvt++; 1628 } 1629 BPF_MTAP(ifp, m_head); 1630 1631 sc->stats.tx_bytes += m_head->m_pkthdr.len; 1632 sc->stats.tx_packets++; 1633 1634 return (0); 1635 } 1636 1637 static void 1638 xn_start_locked(struct ifnet *ifp) 1639 { 1640 struct netfront_info *sc; 1641 struct mbuf *m_head; 1642 int notify; 1643 1644 sc = ifp->if_softc; 1645 1646 if (!netfront_carrier_ok(sc)) 1647 return; 1648 1649 /* 1650 * While we have enough transmit slots available for at least one 1651 * maximum-sized packet, pull mbufs off the queue and put them on 1652 * the transmit ring. 1653 */ 1654 while (xn_tx_slot_available(sc)) { 1655 IF_DEQUEUE(&ifp->if_snd, m_head); 1656 if (m_head == NULL) 1657 break; 1658 1659 if (xn_assemble_tx_request(sc, m_head) != 0) 1660 break; 1661 } 1662 1663 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify); 1664 if (notify) 1665 xen_intr_signal(sc->xen_intr_handle); 1666 1667 if (RING_FULL(&sc->tx)) { 1668 sc->tx_full = 1; 1669 #if 0 1670 netif_stop_queue(dev); 1671 #endif 1672 } 1673 } 1674 1675 static void 1676 xn_start(struct ifnet *ifp) 1677 { 1678 struct netfront_info *sc; 1679 sc = ifp->if_softc; 1680 XN_TX_LOCK(sc); 1681 xn_start_locked(ifp); 1682 XN_TX_UNLOCK(sc); 1683 } 1684 1685 /* equivalent of network_open() in Linux */ 1686 static void 1687 xn_ifinit_locked(struct netfront_info *sc) 1688 { 1689 struct ifnet *ifp; 1690 1691 XN_LOCK_ASSERT(sc); 1692 1693 ifp = sc->xn_ifp; 1694 1695 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1696 return; 1697 1698 xn_stop(sc); 1699 1700 network_alloc_rx_buffers(sc); 1701 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1; 1702 1703 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1704 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1705 if_link_state_change(ifp, LINK_STATE_UP); 1706 1707 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); 1708 } 1709 1710 static void 1711 xn_ifinit(void *xsc) 1712 { 1713 struct netfront_info *sc = xsc; 1714 1715 XN_LOCK(sc); 1716 xn_ifinit_locked(sc); 1717 XN_UNLOCK(sc); 1718 } 1719 1720 static int 1721 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1722 { 1723 struct netfront_info *sc = ifp->if_softc; 1724 struct ifreq *ifr = (struct ifreq *) data; 1725 #ifdef INET 1726 struct ifaddr *ifa = (struct ifaddr *)data; 1727 #endif 1728 1729 int mask, error = 0; 1730 switch(cmd) { 1731 case SIOCSIFADDR: 1732 #ifdef INET 1733 XN_LOCK(sc); 1734 if (ifa->ifa_addr->sa_family == AF_INET) { 1735 ifp->if_flags |= IFF_UP; 1736 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) 1737 xn_ifinit_locked(sc); 1738 arp_ifinit(ifp, ifa); 1739 XN_UNLOCK(sc); 1740 } else { 1741 XN_UNLOCK(sc); 1742 #endif 1743 error = ether_ioctl(ifp, cmd, data); 1744 #ifdef INET 1745 } 1746 #endif 1747 break; 1748 case SIOCSIFMTU: 1749 /* XXX can we alter the MTU on a VN ?*/ 1750 #ifdef notyet 1751 if (ifr->ifr_mtu > XN_JUMBO_MTU) 1752 error = EINVAL; 1753 else 1754 #endif 1755 { 1756 ifp->if_mtu = ifr->ifr_mtu; 1757 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1758 xn_ifinit(sc); 1759 } 1760 break; 1761 case SIOCSIFFLAGS: 1762 XN_LOCK(sc); 1763 if (ifp->if_flags & IFF_UP) { 1764 /* 1765 * If only the state of the PROMISC flag changed, 1766 * then just use the 'set promisc mode' command 1767 * instead of reinitializing the entire NIC. Doing 1768 * a full re-init means reloading the firmware and 1769 * waiting for it to start up, which may take a 1770 * second or two. 1771 */ 1772 #ifdef notyet 1773 /* No promiscuous mode with Xen */ 1774 if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1775 ifp->if_flags & IFF_PROMISC && 1776 !(sc->xn_if_flags & IFF_PROMISC)) { 1777 XN_SETBIT(sc, XN_RX_MODE, 1778 XN_RXMODE_RX_PROMISC); 1779 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1780 !(ifp->if_flags & IFF_PROMISC) && 1781 sc->xn_if_flags & IFF_PROMISC) { 1782 XN_CLRBIT(sc, XN_RX_MODE, 1783 XN_RXMODE_RX_PROMISC); 1784 } else 1785 #endif 1786 xn_ifinit_locked(sc); 1787 } else { 1788 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1789 xn_stop(sc); 1790 } 1791 } 1792 sc->xn_if_flags = ifp->if_flags; 1793 XN_UNLOCK(sc); 1794 error = 0; 1795 break; 1796 case SIOCSIFCAP: 1797 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1798 if (mask & IFCAP_TXCSUM) { 1799 if (IFCAP_TXCSUM & ifp->if_capenable) { 1800 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4); 1801 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP 1802 | CSUM_IP | CSUM_TSO); 1803 } else { 1804 ifp->if_capenable |= IFCAP_TXCSUM; 1805 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP 1806 | CSUM_IP); 1807 } 1808 } 1809 if (mask & IFCAP_RXCSUM) { 1810 ifp->if_capenable ^= IFCAP_RXCSUM; 1811 } 1812 #if __FreeBSD_version >= 700000 1813 if (mask & IFCAP_TSO4) { 1814 if (IFCAP_TSO4 & ifp->if_capenable) { 1815 ifp->if_capenable &= ~IFCAP_TSO4; 1816 ifp->if_hwassist &= ~CSUM_TSO; 1817 } else if (IFCAP_TXCSUM & ifp->if_capenable) { 1818 ifp->if_capenable |= IFCAP_TSO4; 1819 ifp->if_hwassist |= CSUM_TSO; 1820 } else { 1821 IPRINTK("Xen requires tx checksum offload" 1822 " be enabled to use TSO\n"); 1823 error = EINVAL; 1824 } 1825 } 1826 if (mask & IFCAP_LRO) { 1827 ifp->if_capenable ^= IFCAP_LRO; 1828 1829 } 1830 #endif 1831 error = 0; 1832 break; 1833 case SIOCADDMULTI: 1834 case SIOCDELMULTI: 1835 #ifdef notyet 1836 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1837 XN_LOCK(sc); 1838 xn_setmulti(sc); 1839 XN_UNLOCK(sc); 1840 error = 0; 1841 } 1842 #endif 1843 /* FALLTHROUGH */ 1844 case SIOCSIFMEDIA: 1845 case SIOCGIFMEDIA: 1846 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1847 break; 1848 default: 1849 error = ether_ioctl(ifp, cmd, data); 1850 } 1851 1852 return (error); 1853 } 1854 1855 static void 1856 xn_stop(struct netfront_info *sc) 1857 { 1858 struct ifnet *ifp; 1859 1860 XN_LOCK_ASSERT(sc); 1861 1862 ifp = sc->xn_ifp; 1863 1864 callout_stop(&sc->xn_stat_ch); 1865 1866 xn_free_rx_ring(sc); 1867 xn_free_tx_ring(sc); 1868 1869 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 1870 if_link_state_change(ifp, LINK_STATE_DOWN); 1871 } 1872 1873 /* START of Xenolinux helper functions adapted to FreeBSD */ 1874 int 1875 network_connect(struct netfront_info *np) 1876 { 1877 int i, requeue_idx, error; 1878 grant_ref_t ref; 1879 netif_rx_request_t *req; 1880 u_int feature_rx_copy, feature_rx_flip; 1881 1882 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1883 "feature-rx-copy", NULL, "%u", &feature_rx_copy); 1884 if (error) 1885 feature_rx_copy = 0; 1886 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1887 "feature-rx-flip", NULL, "%u", &feature_rx_flip); 1888 if (error) 1889 feature_rx_flip = 1; 1890 1891 /* 1892 * Copy packets on receive path if: 1893 * (a) This was requested by user, and the backend supports it; or 1894 * (b) Flipping was requested, but this is unsupported by the backend. 1895 */ 1896 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) || 1897 (MODPARM_rx_flip && !feature_rx_flip)); 1898 1899 /* Recovery procedure: */ 1900 error = talk_to_backend(np->xbdev, np); 1901 if (error) 1902 return (error); 1903 1904 /* Step 1: Reinitialise variables. */ 1905 xn_query_features(np); 1906 xn_configure_features(np); 1907 netif_release_tx_bufs(np); 1908 1909 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ 1910 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { 1911 struct mbuf *m; 1912 u_long pfn; 1913 1914 if (np->rx_mbufs[i] == NULL) 1915 continue; 1916 1917 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i); 1918 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); 1919 1920 req = RING_GET_REQUEST(&np->rx, requeue_idx); 1921 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT; 1922 1923 if (!np->copying_receiver) { 1924 gnttab_grant_foreign_transfer_ref(ref, 1925 xenbus_get_otherend_id(np->xbdev), 1926 pfn); 1927 } else { 1928 gnttab_grant_foreign_access_ref(ref, 1929 xenbus_get_otherend_id(np->xbdev), 1930 PFNTOMFN(pfn), 0); 1931 } 1932 req->gref = ref; 1933 req->id = requeue_idx; 1934 1935 requeue_idx++; 1936 } 1937 1938 np->rx.req_prod_pvt = requeue_idx; 1939 1940 /* Step 3: All public and private state should now be sane. Get 1941 * ready to start sending and receiving packets and give the driver 1942 * domain a kick because we've probably just requeued some 1943 * packets. 1944 */ 1945 netfront_carrier_on(np); 1946 xen_intr_signal(np->xen_intr_handle); 1947 XN_TX_LOCK(np); 1948 xn_txeof(np); 1949 XN_TX_UNLOCK(np); 1950 network_alloc_rx_buffers(np); 1951 1952 return (0); 1953 } 1954 1955 static void 1956 xn_query_features(struct netfront_info *np) 1957 { 1958 int val; 1959 1960 device_printf(np->xbdev, "backend features:"); 1961 1962 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1963 "feature-sg", NULL, "%d", &val) < 0) 1964 val = 0; 1965 1966 np->maxfrags = 1; 1967 if (val) { 1968 np->maxfrags = MAX_TX_REQ_FRAGS; 1969 printf(" feature-sg"); 1970 } 1971 1972 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1973 "feature-gso-tcpv4", NULL, "%d", &val) < 0) 1974 val = 0; 1975 1976 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO); 1977 if (val) { 1978 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO; 1979 printf(" feature-gso-tcp4"); 1980 } 1981 1982 printf("\n"); 1983 } 1984 1985 static int 1986 xn_configure_features(struct netfront_info *np) 1987 { 1988 int err; 1989 1990 err = 0; 1991 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1992 if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0) 1993 tcp_lro_free(&np->xn_lro); 1994 #endif 1995 np->xn_ifp->if_capenable = 1996 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4); 1997 np->xn_ifp->if_hwassist &= ~CSUM_TSO; 1998 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1999 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) { 2000 err = tcp_lro_init(&np->xn_lro); 2001 if (err) { 2002 device_printf(np->xbdev, "LRO initialization failed\n"); 2003 } else { 2004 np->xn_lro.ifp = np->xn_ifp; 2005 np->xn_ifp->if_capenable |= IFCAP_LRO; 2006 } 2007 } 2008 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) { 2009 np->xn_ifp->if_capenable |= IFCAP_TSO4; 2010 np->xn_ifp->if_hwassist |= CSUM_TSO; 2011 } 2012 #endif 2013 return (err); 2014 } 2015 2016 /** 2017 * Create a network device. 2018 * @param dev Newbus device representing this virtual NIC. 2019 */ 2020 int 2021 create_netdev(device_t dev) 2022 { 2023 int i; 2024 struct netfront_info *np; 2025 int err; 2026 struct ifnet *ifp; 2027 2028 np = device_get_softc(dev); 2029 2030 np->xbdev = dev; 2031 2032 XN_LOCK_INIT(np, xennetif); 2033 2034 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts); 2035 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL); 2036 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL); 2037 2038 np->rx_target = RX_MIN_TARGET; 2039 np->rx_min_target = RX_MIN_TARGET; 2040 np->rx_max_target = RX_MAX_TARGET; 2041 2042 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */ 2043 for (i = 0; i <= NET_TX_RING_SIZE; i++) { 2044 np->tx_mbufs[i] = (void *) ((u_long) i+1); 2045 np->grant_tx_ref[i] = GRANT_REF_INVALID; 2046 } 2047 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0; 2048 2049 for (i = 0; i <= NET_RX_RING_SIZE; i++) { 2050 2051 np->rx_mbufs[i] = NULL; 2052 np->grant_rx_ref[i] = GRANT_REF_INVALID; 2053 } 2054 2055 mbufq_init(&np->xn_rx_batch, INT_MAX); 2056 2057 /* A grant for every tx ring slot */ 2058 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, 2059 &np->gref_tx_head) != 0) { 2060 IPRINTK("#### netfront can't alloc tx grant refs\n"); 2061 err = ENOMEM; 2062 goto exit; 2063 } 2064 /* A grant for every rx ring slot */ 2065 if (gnttab_alloc_grant_references(RX_MAX_TARGET, 2066 &np->gref_rx_head) != 0) { 2067 WPRINTK("#### netfront can't alloc rx grant refs\n"); 2068 gnttab_free_grant_references(np->gref_tx_head); 2069 err = ENOMEM; 2070 goto exit; 2071 } 2072 2073 err = xen_net_read_mac(dev, np->mac); 2074 if (err) 2075 goto out; 2076 2077 /* Set up ifnet structure */ 2078 ifp = np->xn_ifp = if_alloc(IFT_ETHER); 2079 ifp->if_softc = np; 2080 if_initname(ifp, "xn", device_get_unit(dev)); 2081 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 2082 ifp->if_ioctl = xn_ioctl; 2083 ifp->if_output = ether_output; 2084 ifp->if_start = xn_start; 2085 #ifdef notyet 2086 ifp->if_watchdog = xn_watchdog; 2087 #endif 2088 ifp->if_init = xn_ifinit; 2089 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1; 2090 2091 ifp->if_hwassist = XN_CSUM_FEATURES; 2092 ifp->if_capabilities = IFCAP_HWCSUM; 2093 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); 2094 ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS; 2095 ifp->if_hw_tsomaxsegsize = PAGE_SIZE; 2096 2097 ether_ifattach(ifp, np->mac); 2098 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE); 2099 netfront_carrier_off(np); 2100 2101 return (0); 2102 2103 exit: 2104 gnttab_free_grant_references(np->gref_tx_head); 2105 out: 2106 return (err); 2107 } 2108 2109 /** 2110 * Handle the change of state of the backend to Closing. We must delete our 2111 * device-layer structures now, to ensure that writes are flushed through to 2112 * the backend. Once is this done, we can switch to Closed in 2113 * acknowledgement. 2114 */ 2115 #if 0 2116 static void 2117 netfront_closing(device_t dev) 2118 { 2119 #if 0 2120 struct netfront_info *info = dev->dev_driver_data; 2121 2122 DPRINTK("netfront_closing: %s removed\n", dev->nodename); 2123 2124 close_netdev(info); 2125 #endif 2126 xenbus_switch_state(dev, XenbusStateClosed); 2127 } 2128 #endif 2129 2130 static int 2131 netfront_detach(device_t dev) 2132 { 2133 struct netfront_info *info = device_get_softc(dev); 2134 2135 DPRINTK("%s\n", xenbus_get_node(dev)); 2136 2137 netif_free(info); 2138 2139 return 0; 2140 } 2141 2142 static void 2143 netif_free(struct netfront_info *info) 2144 { 2145 XN_LOCK(info); 2146 xn_stop(info); 2147 XN_UNLOCK(info); 2148 callout_drain(&info->xn_stat_ch); 2149 netif_disconnect_backend(info); 2150 if (info->xn_ifp != NULL) { 2151 ether_ifdetach(info->xn_ifp); 2152 if_free(info->xn_ifp); 2153 info->xn_ifp = NULL; 2154 } 2155 ifmedia_removeall(&info->sc_media); 2156 } 2157 2158 static void 2159 netif_disconnect_backend(struct netfront_info *info) 2160 { 2161 XN_RX_LOCK(info); 2162 XN_TX_LOCK(info); 2163 netfront_carrier_off(info); 2164 XN_TX_UNLOCK(info); 2165 XN_RX_UNLOCK(info); 2166 2167 free_ring(&info->tx_ring_ref, &info->tx.sring); 2168 free_ring(&info->rx_ring_ref, &info->rx.sring); 2169 2170 xen_intr_unbind(&info->xen_intr_handle); 2171 } 2172 2173 static void 2174 free_ring(int *ref, void *ring_ptr_ref) 2175 { 2176 void **ring_ptr_ptr = ring_ptr_ref; 2177 2178 if (*ref != GRANT_REF_INVALID) { 2179 /* This API frees the associated storage. */ 2180 gnttab_end_foreign_access(*ref, *ring_ptr_ptr); 2181 *ref = GRANT_REF_INVALID; 2182 } 2183 *ring_ptr_ptr = NULL; 2184 } 2185 2186 static int 2187 xn_ifmedia_upd(struct ifnet *ifp) 2188 { 2189 return (0); 2190 } 2191 2192 static void 2193 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 2194 { 2195 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE; 2196 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL; 2197 } 2198 2199 /* ** Driver registration ** */ 2200 static device_method_t netfront_methods[] = { 2201 /* Device interface */ 2202 DEVMETHOD(device_probe, netfront_probe), 2203 DEVMETHOD(device_attach, netfront_attach), 2204 DEVMETHOD(device_detach, netfront_detach), 2205 DEVMETHOD(device_shutdown, bus_generic_shutdown), 2206 DEVMETHOD(device_suspend, netfront_suspend), 2207 DEVMETHOD(device_resume, netfront_resume), 2208 2209 /* Xenbus interface */ 2210 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed), 2211 2212 DEVMETHOD_END 2213 }; 2214 2215 static driver_t netfront_driver = { 2216 "xn", 2217 netfront_methods, 2218 sizeof(struct netfront_info), 2219 }; 2220 devclass_t netfront_devclass; 2221 2222 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL, 2223 NULL); 2224