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 XenbusStateUnknown: 686 case XenbusStateClosed: 687 case XenbusStateReconfigured: 688 case XenbusStateReconfiguring: 689 break; 690 case XenbusStateInitWait: 691 if (xenbus_get_state(dev) != XenbusStateInitialising) 692 break; 693 if (network_connect(sc) != 0) 694 break; 695 xenbus_set_state(dev, XenbusStateConnected); 696 break; 697 case XenbusStateClosing: 698 xenbus_set_state(dev, XenbusStateClosed); 699 break; 700 case XenbusStateConnected: 701 #ifdef INET 702 netfront_send_fake_arp(dev, sc); 703 #endif 704 break; 705 } 706 } 707 708 static void 709 xn_free_rx_ring(struct netfront_info *sc) 710 { 711 #if 0 712 int i; 713 714 for (i = 0; i < NET_RX_RING_SIZE; i++) { 715 if (sc->xn_cdata.rx_mbufs[i] != NULL) { 716 m_freem(sc->rx_mbufs[i]); 717 sc->rx_mbufs[i] = NULL; 718 } 719 } 720 721 sc->rx.rsp_cons = 0; 722 sc->xn_rx_if->req_prod = 0; 723 sc->xn_rx_if->event = sc->rx.rsp_cons ; 724 #endif 725 } 726 727 static void 728 xn_free_tx_ring(struct netfront_info *sc) 729 { 730 #if 0 731 int i; 732 733 for (i = 0; i < NET_TX_RING_SIZE; i++) { 734 if (sc->tx_mbufs[i] != NULL) { 735 m_freem(sc->tx_mbufs[i]); 736 sc->xn_cdata.xn_tx_chain[i] = NULL; 737 } 738 } 739 740 return; 741 #endif 742 } 743 744 /** 745 * \brief Verify that there is sufficient space in the Tx ring 746 * buffer for a maximally sized request to be enqueued. 747 * 748 * A transmit request requires a transmit descriptor for each packet 749 * fragment, plus up to 2 entries for "options" (e.g. TSO). 750 */ 751 static inline int 752 xn_tx_slot_available(struct netfront_info *np) 753 { 754 return (RING_FREE_REQUESTS(&np->tx) > (MAX_TX_REQ_FRAGS + 2)); 755 } 756 757 static void 758 netif_release_tx_bufs(struct netfront_info *np) 759 { 760 int i; 761 762 for (i = 1; i <= NET_TX_RING_SIZE; i++) { 763 struct mbuf *m; 764 765 m = np->tx_mbufs[i]; 766 767 /* 768 * We assume that no kernel addresses are 769 * less than NET_TX_RING_SIZE. Any entry 770 * in the table that is below this number 771 * must be an index from free-list tracking. 772 */ 773 if (((uintptr_t)m) <= NET_TX_RING_SIZE) 774 continue; 775 gnttab_end_foreign_access_ref(np->grant_tx_ref[i]); 776 gnttab_release_grant_reference(&np->gref_tx_head, 777 np->grant_tx_ref[i]); 778 np->grant_tx_ref[i] = GRANT_REF_INVALID; 779 add_id_to_freelist(np->tx_mbufs, i); 780 np->xn_cdata.xn_tx_chain_cnt--; 781 if (np->xn_cdata.xn_tx_chain_cnt < 0) { 782 panic("%s: tx_chain_cnt must be >= 0", __func__); 783 } 784 m_free(m); 785 } 786 } 787 788 static void 789 network_alloc_rx_buffers(struct netfront_info *sc) 790 { 791 int otherend_id = xenbus_get_otherend_id(sc->xbdev); 792 unsigned short id; 793 struct mbuf *m_new; 794 int i, batch_target, notify; 795 RING_IDX req_prod; 796 struct xen_memory_reservation reservation; 797 grant_ref_t ref; 798 int nr_flips; 799 netif_rx_request_t *req; 800 vm_offset_t vaddr; 801 u_long pfn; 802 803 req_prod = sc->rx.req_prod_pvt; 804 805 if (__predict_false(sc->carrier == 0)) 806 return; 807 808 /* 809 * Allocate mbufs greedily, even though we batch updates to the 810 * receive ring. This creates a less bursty demand on the memory 811 * allocator, and so should reduce the chance of failed allocation 812 * requests both for ourself and for other kernel subsystems. 813 * 814 * Here we attempt to maintain rx_target buffers in flight, counting 815 * buffers that we have yet to process in the receive ring. 816 */ 817 batch_target = sc->rx_target - (req_prod - sc->rx.rsp_cons); 818 for (i = mbufq_len(&sc->xn_rx_batch); i < batch_target; i++) { 819 m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, MJUMPAGESIZE); 820 if (m_new == NULL) { 821 if (i != 0) 822 goto refill; 823 /* 824 * XXX set timer 825 */ 826 break; 827 } 828 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE; 829 830 /* queue the mbufs allocated */ 831 (void )mbufq_enqueue(&sc->xn_rx_batch, m_new); 832 } 833 834 /* 835 * If we've allocated at least half of our target number of entries, 836 * submit them to the backend - we have enough to make the overhead 837 * of submission worthwhile. Otherwise wait for more mbufs and 838 * request entries to become available. 839 */ 840 if (i < (sc->rx_target/2)) { 841 if (req_prod >sc->rx.sring->req_prod) 842 goto push; 843 return; 844 } 845 846 /* 847 * Double floating fill target if we risked having the backend 848 * run out of empty buffers for receive traffic. We define "running 849 * low" as having less than a fourth of our target buffers free 850 * at the time we refilled the queue. 851 */ 852 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) { 853 sc->rx_target *= 2; 854 if (sc->rx_target > sc->rx_max_target) 855 sc->rx_target = sc->rx_max_target; 856 } 857 858 refill: 859 for (nr_flips = i = 0; ; i++) { 860 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL) 861 break; 862 863 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)( 864 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT); 865 866 id = xennet_rxidx(req_prod + i); 867 868 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain")); 869 sc->rx_mbufs[id] = m_new; 870 871 ref = gnttab_claim_grant_reference(&sc->gref_rx_head); 872 KASSERT(ref != GNTTAB_LIST_END, 873 ("reserved grant references exhuasted")); 874 sc->grant_rx_ref[id] = ref; 875 876 vaddr = mtod(m_new, vm_offset_t); 877 pfn = vtophys(vaddr) >> PAGE_SHIFT; 878 req = RING_GET_REQUEST(&sc->rx, req_prod + i); 879 880 if (sc->copying_receiver == 0) { 881 gnttab_grant_foreign_transfer_ref(ref, 882 otherend_id, pfn); 883 sc->rx_pfn_array[nr_flips] = pfn; 884 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 885 /* Remove this page before passing 886 * back to Xen. 887 */ 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 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 if (nr_flips != 0) { 911 #ifdef notyet 912 /* Tell the ballon driver what is going on. */ 913 balloon_update_driver_allowance(i); 914 #endif 915 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array); 916 reservation.nr_extents = i; 917 reservation.extent_order = 0; 918 reservation.address_bits = 0; 919 reservation.domid = DOMID_SELF; 920 921 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 922 /* After all PTEs have been zapped, flush the TLB. */ 923 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] = 924 UVMF_TLB_FLUSH|UVMF_ALL; 925 926 /* Give away a batch of pages. */ 927 sc->rx_mcl[i].op = __HYPERVISOR_memory_op; 928 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation; 929 sc->rx_mcl[i].args[1] = (u_long)&reservation; 930 /* Zap PTEs and give away pages in one big multicall. */ 931 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1); 932 933 if (__predict_false(sc->rx_mcl[i].result != i || 934 HYPERVISOR_memory_op(XENMEM_decrease_reservation, 935 &reservation) != i)) 936 panic("%s: unable to reduce memory " 937 "reservation\n", __func__); 938 } 939 } else { 940 wmb(); 941 } 942 943 /* Above is a suitable barrier to ensure backend will see requests. */ 944 sc->rx.req_prod_pvt = req_prod + i; 945 push: 946 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify); 947 if (notify) 948 xen_intr_signal(sc->xen_intr_handle); 949 } 950 951 static void 952 xn_rxeof(struct netfront_info *np) 953 { 954 struct ifnet *ifp; 955 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 956 struct lro_ctrl *lro = &np->xn_lro; 957 struct lro_entry *queued; 958 #endif 959 struct netfront_rx_info rinfo; 960 struct netif_rx_response *rx = &rinfo.rx; 961 struct netif_extra_info *extras = rinfo.extras; 962 RING_IDX i, rp; 963 multicall_entry_t *mcl; 964 struct mbuf *m; 965 struct mbufq rxq, errq; 966 int err, pages_flipped = 0, work_to_do; 967 968 do { 969 XN_RX_LOCK_ASSERT(np); 970 if (!netfront_carrier_ok(np)) 971 return; 972 973 /* XXX: there should be some sane limit. */ 974 mbufq_init(&errq, INT_MAX); 975 mbufq_init(&rxq, INT_MAX); 976 977 ifp = np->xn_ifp; 978 979 rp = np->rx.sring->rsp_prod; 980 rmb(); /* Ensure we see queued responses up to 'rp'. */ 981 982 i = np->rx.rsp_cons; 983 while ((i != rp)) { 984 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); 985 memset(extras, 0, sizeof(rinfo.extras)); 986 987 m = NULL; 988 err = xennet_get_responses(np, &rinfo, rp, &i, &m, 989 &pages_flipped); 990 991 if (__predict_false(err)) { 992 if (m) 993 (void )mbufq_enqueue(&errq, m); 994 np->stats.rx_errors++; 995 continue; 996 } 997 998 m->m_pkthdr.rcvif = ifp; 999 if ( rx->flags & NETRXF_data_validated ) { 1000 /* Tell the stack the checksums are okay */ 1001 /* 1002 * XXX this isn't necessarily the case - need to add 1003 * check 1004 */ 1005 1006 m->m_pkthdr.csum_flags |= 1007 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID 1008 | CSUM_PSEUDO_HDR); 1009 m->m_pkthdr.csum_data = 0xffff; 1010 } 1011 1012 np->stats.rx_packets++; 1013 np->stats.rx_bytes += m->m_pkthdr.len; 1014 1015 (void )mbufq_enqueue(&rxq, m); 1016 np->rx.rsp_cons = i; 1017 } 1018 1019 if (pages_flipped) { 1020 /* Some pages are no longer absent... */ 1021 #ifdef notyet 1022 balloon_update_driver_allowance(-pages_flipped); 1023 #endif 1024 /* Do all the remapping work, and M->P updates, in one big 1025 * hypercall. 1026 */ 1027 if (!!xen_feature(XENFEAT_auto_translated_physmap)) { 1028 mcl = np->rx_mcl + pages_flipped; 1029 mcl->op = __HYPERVISOR_mmu_update; 1030 mcl->args[0] = (u_long)np->rx_mmu; 1031 mcl->args[1] = pages_flipped; 1032 mcl->args[2] = 0; 1033 mcl->args[3] = DOMID_SELF; 1034 (void)HYPERVISOR_multicall(np->rx_mcl, 1035 pages_flipped + 1); 1036 } 1037 } 1038 1039 mbufq_drain(&errq); 1040 1041 /* 1042 * Process all the mbufs after the remapping is complete. 1043 * Break the mbuf chain first though. 1044 */ 1045 while ((m = mbufq_dequeue(&rxq)) != NULL) { 1046 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); 1047 1048 /* 1049 * Do we really need to drop the rx lock? 1050 */ 1051 XN_RX_UNLOCK(np); 1052 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1053 /* Use LRO if possible */ 1054 if ((ifp->if_capenable & IFCAP_LRO) == 0 || 1055 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) { 1056 /* 1057 * If LRO fails, pass up to the stack 1058 * directly. 1059 */ 1060 (*ifp->if_input)(ifp, m); 1061 } 1062 #else 1063 (*ifp->if_input)(ifp, m); 1064 #endif 1065 XN_RX_LOCK(np); 1066 } 1067 1068 np->rx.rsp_cons = i; 1069 1070 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1071 /* 1072 * Flush any outstanding LRO work 1073 */ 1074 while (!SLIST_EMPTY(&lro->lro_active)) { 1075 queued = SLIST_FIRST(&lro->lro_active); 1076 SLIST_REMOVE_HEAD(&lro->lro_active, next); 1077 tcp_lro_flush(lro, queued); 1078 } 1079 #endif 1080 1081 #if 0 1082 /* If we get a callback with very few responses, reduce fill target. */ 1083 /* NB. Note exponential increase, linear decrease. */ 1084 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > 1085 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target)) 1086 np->rx_target = np->rx_min_target; 1087 #endif 1088 1089 network_alloc_rx_buffers(np); 1090 1091 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do); 1092 } while (work_to_do); 1093 } 1094 1095 static void 1096 xn_txeof(struct netfront_info *np) 1097 { 1098 RING_IDX i, prod; 1099 unsigned short id; 1100 struct ifnet *ifp; 1101 netif_tx_response_t *txr; 1102 struct mbuf *m; 1103 1104 XN_TX_LOCK_ASSERT(np); 1105 1106 if (!netfront_carrier_ok(np)) 1107 return; 1108 1109 ifp = np->xn_ifp; 1110 1111 do { 1112 prod = np->tx.sring->rsp_prod; 1113 rmb(); /* Ensure we see responses up to 'rp'. */ 1114 1115 for (i = np->tx.rsp_cons; i != prod; i++) { 1116 txr = RING_GET_RESPONSE(&np->tx, i); 1117 if (txr->status == NETIF_RSP_NULL) 1118 continue; 1119 1120 if (txr->status != NETIF_RSP_OKAY) { 1121 printf("%s: WARNING: response is %d!\n", 1122 __func__, txr->status); 1123 } 1124 id = txr->id; 1125 m = np->tx_mbufs[id]; 1126 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain")); 1127 KASSERT((uintptr_t)m > NET_TX_RING_SIZE, 1128 ("mbuf already on the free list, but we're " 1129 "trying to free it again!")); 1130 M_ASSERTVALID(m); 1131 1132 /* 1133 * Increment packet count if this is the last 1134 * mbuf of the chain. 1135 */ 1136 if (!m->m_next) 1137 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 1138 if (__predict_false(gnttab_query_foreign_access( 1139 np->grant_tx_ref[id]) != 0)) { 1140 panic("%s: grant id %u still in use by the " 1141 "backend", __func__, id); 1142 } 1143 gnttab_end_foreign_access_ref( 1144 np->grant_tx_ref[id]); 1145 gnttab_release_grant_reference( 1146 &np->gref_tx_head, np->grant_tx_ref[id]); 1147 np->grant_tx_ref[id] = GRANT_REF_INVALID; 1148 1149 np->tx_mbufs[id] = NULL; 1150 add_id_to_freelist(np->tx_mbufs, id); 1151 np->xn_cdata.xn_tx_chain_cnt--; 1152 m_free(m); 1153 /* Only mark the queue active if we've freed up at least one slot to try */ 1154 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1155 } 1156 np->tx.rsp_cons = prod; 1157 1158 /* 1159 * Set a new event, then check for race with update of 1160 * tx_cons. Note that it is essential to schedule a 1161 * callback, no matter how few buffers are pending. Even if 1162 * there is space in the transmit ring, higher layers may 1163 * be blocked because too much data is outstanding: in such 1164 * cases notification from Xen is likely to be the only kick 1165 * that we'll get. 1166 */ 1167 np->tx.sring->rsp_event = 1168 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; 1169 1170 mb(); 1171 } while (prod != np->tx.sring->rsp_prod); 1172 1173 if (np->tx_full && 1174 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) { 1175 np->tx_full = 0; 1176 #if 0 1177 if (np->user_state == UST_OPEN) 1178 netif_wake_queue(dev); 1179 #endif 1180 } 1181 } 1182 1183 static void 1184 xn_intr(void *xsc) 1185 { 1186 struct netfront_info *np = xsc; 1187 struct ifnet *ifp = np->xn_ifp; 1188 1189 #if 0 1190 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod && 1191 likely(netfront_carrier_ok(np)) && 1192 ifp->if_drv_flags & IFF_DRV_RUNNING)) 1193 return; 1194 #endif 1195 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) { 1196 XN_TX_LOCK(np); 1197 xn_txeof(np); 1198 XN_TX_UNLOCK(np); 1199 } 1200 1201 XN_RX_LOCK(np); 1202 xn_rxeof(np); 1203 XN_RX_UNLOCK(np); 1204 1205 if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1206 !IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 1207 xn_start(ifp); 1208 } 1209 1210 static void 1211 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m, 1212 grant_ref_t ref) 1213 { 1214 int new = xennet_rxidx(np->rx.req_prod_pvt); 1215 1216 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL")); 1217 np->rx_mbufs[new] = m; 1218 np->grant_rx_ref[new] = ref; 1219 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; 1220 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; 1221 np->rx.req_prod_pvt++; 1222 } 1223 1224 static int 1225 xennet_get_extras(struct netfront_info *np, 1226 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons) 1227 { 1228 struct netif_extra_info *extra; 1229 1230 int err = 0; 1231 1232 do { 1233 struct mbuf *m; 1234 grant_ref_t ref; 1235 1236 if (__predict_false(*cons + 1 == rp)) { 1237 #if 0 1238 if (net_ratelimit()) 1239 WPRINTK("Missing extra info\n"); 1240 #endif 1241 err = EINVAL; 1242 break; 1243 } 1244 1245 extra = (struct netif_extra_info *) 1246 RING_GET_RESPONSE(&np->rx, ++(*cons)); 1247 1248 if (__predict_false(!extra->type || 1249 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 1250 #if 0 1251 if (net_ratelimit()) 1252 WPRINTK("Invalid extra type: %d\n", 1253 extra->type); 1254 #endif 1255 err = EINVAL; 1256 } else { 1257 memcpy(&extras[extra->type - 1], extra, sizeof(*extra)); 1258 } 1259 1260 m = xennet_get_rx_mbuf(np, *cons); 1261 ref = xennet_get_rx_ref(np, *cons); 1262 xennet_move_rx_slot(np, m, ref); 1263 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 1264 1265 return err; 1266 } 1267 1268 static int 1269 xennet_get_responses(struct netfront_info *np, 1270 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, 1271 struct mbuf **list, 1272 int *pages_flipped_p) 1273 { 1274 int pages_flipped = *pages_flipped_p; 1275 struct mmu_update *mmu; 1276 struct multicall_entry *mcl; 1277 struct netif_rx_response *rx = &rinfo->rx; 1278 struct netif_extra_info *extras = rinfo->extras; 1279 struct mbuf *m, *m0, *m_prev; 1280 grant_ref_t ref = xennet_get_rx_ref(np, *cons); 1281 RING_IDX ref_cons = *cons; 1282 int frags = 1; 1283 int err = 0; 1284 u_long ret; 1285 1286 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons); 1287 1288 if (rx->flags & NETRXF_extra_info) { 1289 err = xennet_get_extras(np, extras, rp, cons); 1290 } 1291 1292 if (m0 != NULL) { 1293 m0->m_pkthdr.len = 0; 1294 m0->m_next = NULL; 1295 } 1296 1297 for (;;) { 1298 u_long mfn; 1299 1300 #if 0 1301 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n", 1302 rx->status, rx->offset, frags); 1303 #endif 1304 if (__predict_false(rx->status < 0 || 1305 rx->offset + rx->status > PAGE_SIZE)) { 1306 1307 #if 0 1308 if (net_ratelimit()) 1309 WPRINTK("rx->offset: %x, size: %u\n", 1310 rx->offset, rx->status); 1311 #endif 1312 xennet_move_rx_slot(np, m, ref); 1313 if (m0 == m) 1314 m0 = NULL; 1315 m = NULL; 1316 err = EINVAL; 1317 goto next_skip_queue; 1318 } 1319 1320 /* 1321 * This definitely indicates a bug, either in this driver or in 1322 * the backend driver. In future this should flag the bad 1323 * situation to the system controller to reboot the backed. 1324 */ 1325 if (ref == GRANT_REF_INVALID) { 1326 1327 #if 0 1328 if (net_ratelimit()) 1329 WPRINTK("Bad rx response id %d.\n", rx->id); 1330 #endif 1331 printf("%s: Bad rx response id %d.\n", __func__,rx->id); 1332 err = EINVAL; 1333 goto next; 1334 } 1335 1336 if (!np->copying_receiver) { 1337 /* Memory pressure, insufficient buffer 1338 * headroom, ... 1339 */ 1340 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) { 1341 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n", 1342 rx->id, rx->status); 1343 xennet_move_rx_slot(np, m, ref); 1344 err = ENOMEM; 1345 goto next; 1346 } 1347 1348 if (!xen_feature( XENFEAT_auto_translated_physmap)) { 1349 /* Remap the page. */ 1350 void *vaddr = mtod(m, void *); 1351 uint32_t pfn; 1352 1353 mcl = np->rx_mcl + pages_flipped; 1354 mmu = np->rx_mmu + pages_flipped; 1355 1356 MULTI_update_va_mapping(mcl, (u_long)vaddr, 1357 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW | 1358 PG_V | PG_M | PG_A, 0); 1359 pfn = (uintptr_t)m->m_ext.ext_arg1; 1360 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) | 1361 MMU_MACHPHYS_UPDATE; 1362 mmu->val = pfn; 1363 } 1364 pages_flipped++; 1365 } else { 1366 ret = gnttab_end_foreign_access_ref(ref); 1367 KASSERT(ret, ("ret != 0")); 1368 } 1369 1370 gnttab_release_grant_reference(&np->gref_rx_head, ref); 1371 1372 next: 1373 if (m == NULL) 1374 break; 1375 1376 m->m_len = rx->status; 1377 m->m_data += rx->offset; 1378 m0->m_pkthdr.len += rx->status; 1379 1380 next_skip_queue: 1381 if (!(rx->flags & NETRXF_more_data)) 1382 break; 1383 1384 if (*cons + frags == rp) { 1385 if (net_ratelimit()) 1386 WPRINTK("Need more frags\n"); 1387 err = ENOENT; 1388 printf("%s: cons %u frags %u rp %u, not enough frags\n", 1389 __func__, *cons, frags, rp); 1390 break; 1391 } 1392 /* 1393 * Note that m can be NULL, if rx->status < 0 or if 1394 * rx->offset + rx->status > PAGE_SIZE above. 1395 */ 1396 m_prev = m; 1397 1398 rx = RING_GET_RESPONSE(&np->rx, *cons + frags); 1399 m = xennet_get_rx_mbuf(np, *cons + frags); 1400 1401 /* 1402 * m_prev == NULL can happen if rx->status < 0 or if 1403 * rx->offset + * rx->status > PAGE_SIZE above. 1404 */ 1405 if (m_prev != NULL) 1406 m_prev->m_next = m; 1407 1408 /* 1409 * m0 can be NULL if rx->status < 0 or if * rx->offset + 1410 * rx->status > PAGE_SIZE above. 1411 */ 1412 if (m0 == NULL) 1413 m0 = m; 1414 m->m_next = NULL; 1415 ref = xennet_get_rx_ref(np, *cons + frags); 1416 ref_cons = *cons + frags; 1417 frags++; 1418 } 1419 *list = m0; 1420 *cons += frags; 1421 *pages_flipped_p = pages_flipped; 1422 1423 return (err); 1424 } 1425 1426 static void 1427 xn_tick_locked(struct netfront_info *sc) 1428 { 1429 XN_RX_LOCK_ASSERT(sc); 1430 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); 1431 1432 /* XXX placeholder for printing debug information */ 1433 } 1434 1435 static void 1436 xn_tick(void *xsc) 1437 { 1438 struct netfront_info *sc; 1439 1440 sc = xsc; 1441 XN_RX_LOCK(sc); 1442 xn_tick_locked(sc); 1443 XN_RX_UNLOCK(sc); 1444 } 1445 1446 /** 1447 * \brief Count the number of fragments in an mbuf chain. 1448 * 1449 * Surprisingly, there isn't an M* macro for this. 1450 */ 1451 static inline int 1452 xn_count_frags(struct mbuf *m) 1453 { 1454 int nfrags; 1455 1456 for (nfrags = 0; m != NULL; m = m->m_next) 1457 nfrags++; 1458 1459 return (nfrags); 1460 } 1461 1462 /** 1463 * Given an mbuf chain, make sure we have enough room and then push 1464 * it onto the transmit ring. 1465 */ 1466 static int 1467 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) 1468 { 1469 struct ifnet *ifp; 1470 struct mbuf *m; 1471 u_int nfrags; 1472 int otherend_id; 1473 1474 ifp = sc->xn_ifp; 1475 1476 /** 1477 * Defragment the mbuf if necessary. 1478 */ 1479 nfrags = xn_count_frags(m_head); 1480 1481 /* 1482 * Check to see whether this request is longer than netback 1483 * can handle, and try to defrag it. 1484 */ 1485 /** 1486 * It is a bit lame, but the netback driver in Linux can't 1487 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of 1488 * the Linux network stack. 1489 */ 1490 if (nfrags > sc->maxfrags) { 1491 m = m_defrag(m_head, M_NOWAIT); 1492 if (!m) { 1493 /* 1494 * Defrag failed, so free the mbuf and 1495 * therefore drop the packet. 1496 */ 1497 m_freem(m_head); 1498 return (EMSGSIZE); 1499 } 1500 m_head = m; 1501 } 1502 1503 /* Determine how many fragments now exist */ 1504 nfrags = xn_count_frags(m_head); 1505 1506 /* 1507 * Check to see whether the defragmented packet has too many 1508 * segments for the Linux netback driver. 1509 */ 1510 /** 1511 * The FreeBSD TCP stack, with TSO enabled, can produce a chain 1512 * of mbufs longer than Linux can handle. Make sure we don't 1513 * pass a too-long chain over to the other side by dropping the 1514 * packet. It doesn't look like there is currently a way to 1515 * tell the TCP stack to generate a shorter chain of packets. 1516 */ 1517 if (nfrags > MAX_TX_REQ_FRAGS) { 1518 #ifdef DEBUG 1519 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback " 1520 "won't be able to handle it, dropping\n", 1521 __func__, nfrags, MAX_TX_REQ_FRAGS); 1522 #endif 1523 m_freem(m_head); 1524 return (EMSGSIZE); 1525 } 1526 1527 /* 1528 * This check should be redundant. We've already verified that we 1529 * have enough slots in the ring to handle a packet of maximum 1530 * size, and that our packet is less than the maximum size. Keep 1531 * it in here as an assert for now just to make certain that 1532 * xn_tx_chain_cnt is accurate. 1533 */ 1534 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE, 1535 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE " 1536 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt, 1537 (int) nfrags, (int) NET_TX_RING_SIZE)); 1538 1539 /* 1540 * Start packing the mbufs in this chain into 1541 * the fragment pointers. Stop when we run out 1542 * of fragments or hit the end of the mbuf chain. 1543 */ 1544 m = m_head; 1545 otherend_id = xenbus_get_otherend_id(sc->xbdev); 1546 for (m = m_head; m; m = m->m_next) { 1547 netif_tx_request_t *tx; 1548 uintptr_t id; 1549 grant_ref_t ref; 1550 u_long mfn; /* XXX Wrong type? */ 1551 1552 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt); 1553 id = get_id_from_freelist(sc->tx_mbufs); 1554 if (id == 0) 1555 panic("%s: was allocated the freelist head!\n", 1556 __func__); 1557 sc->xn_cdata.xn_tx_chain_cnt++; 1558 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE) 1559 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n", 1560 __func__); 1561 sc->tx_mbufs[id] = m; 1562 tx->id = id; 1563 ref = gnttab_claim_grant_reference(&sc->gref_tx_head); 1564 KASSERT((short)ref >= 0, ("Negative ref")); 1565 mfn = virt_to_mfn(mtod(m, vm_offset_t)); 1566 gnttab_grant_foreign_access_ref(ref, otherend_id, 1567 mfn, GNTMAP_readonly); 1568 tx->gref = sc->grant_tx_ref[id] = ref; 1569 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1); 1570 tx->flags = 0; 1571 if (m == m_head) { 1572 /* 1573 * The first fragment has the entire packet 1574 * size, subsequent fragments have just the 1575 * fragment size. The backend works out the 1576 * true size of the first fragment by 1577 * subtracting the sizes of the other 1578 * fragments. 1579 */ 1580 tx->size = m->m_pkthdr.len; 1581 1582 /* 1583 * The first fragment contains the checksum flags 1584 * and is optionally followed by extra data for 1585 * TSO etc. 1586 */ 1587 /** 1588 * CSUM_TSO requires checksum offloading. 1589 * Some versions of FreeBSD fail to 1590 * set CSUM_TCP in the CSUM_TSO case, 1591 * so we have to test for CSUM_TSO 1592 * explicitly. 1593 */ 1594 if (m->m_pkthdr.csum_flags 1595 & (CSUM_DELAY_DATA | CSUM_TSO)) { 1596 tx->flags |= (NETTXF_csum_blank 1597 | NETTXF_data_validated); 1598 } 1599 #if __FreeBSD_version >= 700000 1600 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 1601 struct netif_extra_info *gso = 1602 (struct netif_extra_info *) 1603 RING_GET_REQUEST(&sc->tx, 1604 ++sc->tx.req_prod_pvt); 1605 1606 tx->flags |= NETTXF_extra_info; 1607 1608 gso->u.gso.size = m->m_pkthdr.tso_segsz; 1609 gso->u.gso.type = 1610 XEN_NETIF_GSO_TYPE_TCPV4; 1611 gso->u.gso.pad = 0; 1612 gso->u.gso.features = 0; 1613 1614 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 1615 gso->flags = 0; 1616 } 1617 #endif 1618 } else { 1619 tx->size = m->m_len; 1620 } 1621 if (m->m_next) 1622 tx->flags |= NETTXF_more_data; 1623 1624 sc->tx.req_prod_pvt++; 1625 } 1626 BPF_MTAP(ifp, m_head); 1627 1628 sc->stats.tx_bytes += m_head->m_pkthdr.len; 1629 sc->stats.tx_packets++; 1630 1631 return (0); 1632 } 1633 1634 static void 1635 xn_start_locked(struct ifnet *ifp) 1636 { 1637 struct netfront_info *sc; 1638 struct mbuf *m_head; 1639 int notify; 1640 1641 sc = ifp->if_softc; 1642 1643 if (!netfront_carrier_ok(sc)) 1644 return; 1645 1646 /* 1647 * While we have enough transmit slots available for at least one 1648 * maximum-sized packet, pull mbufs off the queue and put them on 1649 * the transmit ring. 1650 */ 1651 while (xn_tx_slot_available(sc)) { 1652 IF_DEQUEUE(&ifp->if_snd, m_head); 1653 if (m_head == NULL) 1654 break; 1655 1656 if (xn_assemble_tx_request(sc, m_head) != 0) 1657 break; 1658 } 1659 1660 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify); 1661 if (notify) 1662 xen_intr_signal(sc->xen_intr_handle); 1663 1664 if (RING_FULL(&sc->tx)) { 1665 sc->tx_full = 1; 1666 #if 0 1667 netif_stop_queue(dev); 1668 #endif 1669 } 1670 } 1671 1672 static void 1673 xn_start(struct ifnet *ifp) 1674 { 1675 struct netfront_info *sc; 1676 sc = ifp->if_softc; 1677 XN_TX_LOCK(sc); 1678 xn_start_locked(ifp); 1679 XN_TX_UNLOCK(sc); 1680 } 1681 1682 /* equivalent of network_open() in Linux */ 1683 static void 1684 xn_ifinit_locked(struct netfront_info *sc) 1685 { 1686 struct ifnet *ifp; 1687 1688 XN_LOCK_ASSERT(sc); 1689 1690 ifp = sc->xn_ifp; 1691 1692 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1693 return; 1694 1695 xn_stop(sc); 1696 1697 network_alloc_rx_buffers(sc); 1698 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1; 1699 1700 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1701 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1702 if_link_state_change(ifp, LINK_STATE_UP); 1703 1704 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); 1705 } 1706 1707 static void 1708 xn_ifinit(void *xsc) 1709 { 1710 struct netfront_info *sc = xsc; 1711 1712 XN_LOCK(sc); 1713 xn_ifinit_locked(sc); 1714 XN_UNLOCK(sc); 1715 } 1716 1717 static int 1718 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1719 { 1720 struct netfront_info *sc = ifp->if_softc; 1721 struct ifreq *ifr = (struct ifreq *) data; 1722 #ifdef INET 1723 struct ifaddr *ifa = (struct ifaddr *)data; 1724 #endif 1725 1726 int mask, error = 0; 1727 switch(cmd) { 1728 case SIOCSIFADDR: 1729 #ifdef INET 1730 XN_LOCK(sc); 1731 if (ifa->ifa_addr->sa_family == AF_INET) { 1732 ifp->if_flags |= IFF_UP; 1733 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) 1734 xn_ifinit_locked(sc); 1735 arp_ifinit(ifp, ifa); 1736 XN_UNLOCK(sc); 1737 } else { 1738 XN_UNLOCK(sc); 1739 #endif 1740 error = ether_ioctl(ifp, cmd, data); 1741 #ifdef INET 1742 } 1743 #endif 1744 break; 1745 case SIOCSIFMTU: 1746 /* XXX can we alter the MTU on a VN ?*/ 1747 #ifdef notyet 1748 if (ifr->ifr_mtu > XN_JUMBO_MTU) 1749 error = EINVAL; 1750 else 1751 #endif 1752 { 1753 ifp->if_mtu = ifr->ifr_mtu; 1754 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1755 xn_ifinit(sc); 1756 } 1757 break; 1758 case SIOCSIFFLAGS: 1759 XN_LOCK(sc); 1760 if (ifp->if_flags & IFF_UP) { 1761 /* 1762 * If only the state of the PROMISC flag changed, 1763 * then just use the 'set promisc mode' command 1764 * instead of reinitializing the entire NIC. Doing 1765 * a full re-init means reloading the firmware and 1766 * waiting for it to start up, which may take a 1767 * second or two. 1768 */ 1769 #ifdef notyet 1770 /* No promiscuous mode with Xen */ 1771 if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1772 ifp->if_flags & IFF_PROMISC && 1773 !(sc->xn_if_flags & IFF_PROMISC)) { 1774 XN_SETBIT(sc, XN_RX_MODE, 1775 XN_RXMODE_RX_PROMISC); 1776 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1777 !(ifp->if_flags & IFF_PROMISC) && 1778 sc->xn_if_flags & IFF_PROMISC) { 1779 XN_CLRBIT(sc, XN_RX_MODE, 1780 XN_RXMODE_RX_PROMISC); 1781 } else 1782 #endif 1783 xn_ifinit_locked(sc); 1784 } else { 1785 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1786 xn_stop(sc); 1787 } 1788 } 1789 sc->xn_if_flags = ifp->if_flags; 1790 XN_UNLOCK(sc); 1791 error = 0; 1792 break; 1793 case SIOCSIFCAP: 1794 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1795 if (mask & IFCAP_TXCSUM) { 1796 if (IFCAP_TXCSUM & ifp->if_capenable) { 1797 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4); 1798 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP 1799 | CSUM_IP | CSUM_TSO); 1800 } else { 1801 ifp->if_capenable |= IFCAP_TXCSUM; 1802 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP 1803 | CSUM_IP); 1804 } 1805 } 1806 if (mask & IFCAP_RXCSUM) { 1807 ifp->if_capenable ^= IFCAP_RXCSUM; 1808 } 1809 #if __FreeBSD_version >= 700000 1810 if (mask & IFCAP_TSO4) { 1811 if (IFCAP_TSO4 & ifp->if_capenable) { 1812 ifp->if_capenable &= ~IFCAP_TSO4; 1813 ifp->if_hwassist &= ~CSUM_TSO; 1814 } else if (IFCAP_TXCSUM & ifp->if_capenable) { 1815 ifp->if_capenable |= IFCAP_TSO4; 1816 ifp->if_hwassist |= CSUM_TSO; 1817 } else { 1818 IPRINTK("Xen requires tx checksum offload" 1819 " be enabled to use TSO\n"); 1820 error = EINVAL; 1821 } 1822 } 1823 if (mask & IFCAP_LRO) { 1824 ifp->if_capenable ^= IFCAP_LRO; 1825 1826 } 1827 #endif 1828 error = 0; 1829 break; 1830 case SIOCADDMULTI: 1831 case SIOCDELMULTI: 1832 #ifdef notyet 1833 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1834 XN_LOCK(sc); 1835 xn_setmulti(sc); 1836 XN_UNLOCK(sc); 1837 error = 0; 1838 } 1839 #endif 1840 /* FALLTHROUGH */ 1841 case SIOCSIFMEDIA: 1842 case SIOCGIFMEDIA: 1843 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1844 break; 1845 default: 1846 error = ether_ioctl(ifp, cmd, data); 1847 } 1848 1849 return (error); 1850 } 1851 1852 static void 1853 xn_stop(struct netfront_info *sc) 1854 { 1855 struct ifnet *ifp; 1856 1857 XN_LOCK_ASSERT(sc); 1858 1859 ifp = sc->xn_ifp; 1860 1861 callout_stop(&sc->xn_stat_ch); 1862 1863 xn_free_rx_ring(sc); 1864 xn_free_tx_ring(sc); 1865 1866 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 1867 if_link_state_change(ifp, LINK_STATE_DOWN); 1868 } 1869 1870 /* START of Xenolinux helper functions adapted to FreeBSD */ 1871 int 1872 network_connect(struct netfront_info *np) 1873 { 1874 int i, requeue_idx, error; 1875 grant_ref_t ref; 1876 netif_rx_request_t *req; 1877 u_int feature_rx_copy, feature_rx_flip; 1878 1879 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1880 "feature-rx-copy", NULL, "%u", &feature_rx_copy); 1881 if (error) 1882 feature_rx_copy = 0; 1883 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1884 "feature-rx-flip", NULL, "%u", &feature_rx_flip); 1885 if (error) 1886 feature_rx_flip = 1; 1887 1888 /* 1889 * Copy packets on receive path if: 1890 * (a) This was requested by user, and the backend supports it; or 1891 * (b) Flipping was requested, but this is unsupported by the backend. 1892 */ 1893 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) || 1894 (MODPARM_rx_flip && !feature_rx_flip)); 1895 1896 /* Recovery procedure: */ 1897 error = talk_to_backend(np->xbdev, np); 1898 if (error) 1899 return (error); 1900 1901 /* Step 1: Reinitialise variables. */ 1902 xn_query_features(np); 1903 xn_configure_features(np); 1904 netif_release_tx_bufs(np); 1905 1906 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ 1907 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { 1908 struct mbuf *m; 1909 u_long pfn; 1910 1911 if (np->rx_mbufs[i] == NULL) 1912 continue; 1913 1914 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i); 1915 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); 1916 1917 req = RING_GET_REQUEST(&np->rx, requeue_idx); 1918 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT; 1919 1920 if (!np->copying_receiver) { 1921 gnttab_grant_foreign_transfer_ref(ref, 1922 xenbus_get_otherend_id(np->xbdev), 1923 pfn); 1924 } else { 1925 gnttab_grant_foreign_access_ref(ref, 1926 xenbus_get_otherend_id(np->xbdev), 1927 pfn, 0); 1928 } 1929 req->gref = ref; 1930 req->id = requeue_idx; 1931 1932 requeue_idx++; 1933 } 1934 1935 np->rx.req_prod_pvt = requeue_idx; 1936 1937 /* Step 3: All public and private state should now be sane. Get 1938 * ready to start sending and receiving packets and give the driver 1939 * domain a kick because we've probably just requeued some 1940 * packets. 1941 */ 1942 netfront_carrier_on(np); 1943 xen_intr_signal(np->xen_intr_handle); 1944 XN_TX_LOCK(np); 1945 xn_txeof(np); 1946 XN_TX_UNLOCK(np); 1947 network_alloc_rx_buffers(np); 1948 1949 return (0); 1950 } 1951 1952 static void 1953 xn_query_features(struct netfront_info *np) 1954 { 1955 int val; 1956 1957 device_printf(np->xbdev, "backend features:"); 1958 1959 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1960 "feature-sg", NULL, "%d", &val) < 0) 1961 val = 0; 1962 1963 np->maxfrags = 1; 1964 if (val) { 1965 np->maxfrags = MAX_TX_REQ_FRAGS; 1966 printf(" feature-sg"); 1967 } 1968 1969 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1970 "feature-gso-tcpv4", NULL, "%d", &val) < 0) 1971 val = 0; 1972 1973 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO); 1974 if (val) { 1975 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO; 1976 printf(" feature-gso-tcp4"); 1977 } 1978 1979 printf("\n"); 1980 } 1981 1982 static int 1983 xn_configure_features(struct netfront_info *np) 1984 { 1985 int err; 1986 1987 err = 0; 1988 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1989 if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0) 1990 tcp_lro_free(&np->xn_lro); 1991 #endif 1992 np->xn_ifp->if_capenable = 1993 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4); 1994 np->xn_ifp->if_hwassist &= ~CSUM_TSO; 1995 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1996 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) { 1997 err = tcp_lro_init(&np->xn_lro); 1998 if (err) { 1999 device_printf(np->xbdev, "LRO initialization failed\n"); 2000 } else { 2001 np->xn_lro.ifp = np->xn_ifp; 2002 np->xn_ifp->if_capenable |= IFCAP_LRO; 2003 } 2004 } 2005 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) { 2006 np->xn_ifp->if_capenable |= IFCAP_TSO4; 2007 np->xn_ifp->if_hwassist |= CSUM_TSO; 2008 } 2009 #endif 2010 return (err); 2011 } 2012 2013 /** 2014 * Create a network device. 2015 * @param dev Newbus device representing this virtual NIC. 2016 */ 2017 int 2018 create_netdev(device_t dev) 2019 { 2020 int i; 2021 struct netfront_info *np; 2022 int err; 2023 struct ifnet *ifp; 2024 2025 np = device_get_softc(dev); 2026 2027 np->xbdev = dev; 2028 2029 XN_LOCK_INIT(np, xennetif); 2030 2031 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts); 2032 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL); 2033 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL); 2034 2035 np->rx_target = RX_MIN_TARGET; 2036 np->rx_min_target = RX_MIN_TARGET; 2037 np->rx_max_target = RX_MAX_TARGET; 2038 2039 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */ 2040 for (i = 0; i <= NET_TX_RING_SIZE; i++) { 2041 np->tx_mbufs[i] = (void *) ((u_long) i+1); 2042 np->grant_tx_ref[i] = GRANT_REF_INVALID; 2043 } 2044 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0; 2045 2046 for (i = 0; i <= NET_RX_RING_SIZE; i++) { 2047 2048 np->rx_mbufs[i] = NULL; 2049 np->grant_rx_ref[i] = GRANT_REF_INVALID; 2050 } 2051 2052 mbufq_init(&np->xn_rx_batch, INT_MAX); 2053 2054 /* A grant for every tx ring slot */ 2055 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, 2056 &np->gref_tx_head) != 0) { 2057 IPRINTK("#### netfront can't alloc tx grant refs\n"); 2058 err = ENOMEM; 2059 goto exit; 2060 } 2061 /* A grant for every rx ring slot */ 2062 if (gnttab_alloc_grant_references(RX_MAX_TARGET, 2063 &np->gref_rx_head) != 0) { 2064 WPRINTK("#### netfront can't alloc rx grant refs\n"); 2065 gnttab_free_grant_references(np->gref_tx_head); 2066 err = ENOMEM; 2067 goto exit; 2068 } 2069 2070 err = xen_net_read_mac(dev, np->mac); 2071 if (err) 2072 goto out; 2073 2074 /* Set up ifnet structure */ 2075 ifp = np->xn_ifp = if_alloc(IFT_ETHER); 2076 ifp->if_softc = np; 2077 if_initname(ifp, "xn", device_get_unit(dev)); 2078 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 2079 ifp->if_ioctl = xn_ioctl; 2080 ifp->if_output = ether_output; 2081 ifp->if_start = xn_start; 2082 #ifdef notyet 2083 ifp->if_watchdog = xn_watchdog; 2084 #endif 2085 ifp->if_init = xn_ifinit; 2086 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1; 2087 2088 ifp->if_hwassist = XN_CSUM_FEATURES; 2089 ifp->if_capabilities = IFCAP_HWCSUM; 2090 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); 2091 ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS; 2092 ifp->if_hw_tsomaxsegsize = PAGE_SIZE; 2093 2094 ether_ifattach(ifp, np->mac); 2095 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE); 2096 netfront_carrier_off(np); 2097 2098 return (0); 2099 2100 exit: 2101 gnttab_free_grant_references(np->gref_tx_head); 2102 out: 2103 return (err); 2104 } 2105 2106 /** 2107 * Handle the change of state of the backend to Closing. We must delete our 2108 * device-layer structures now, to ensure that writes are flushed through to 2109 * the backend. Once is this done, we can switch to Closed in 2110 * acknowledgement. 2111 */ 2112 #if 0 2113 static void 2114 netfront_closing(device_t dev) 2115 { 2116 #if 0 2117 struct netfront_info *info = dev->dev_driver_data; 2118 2119 DPRINTK("netfront_closing: %s removed\n", dev->nodename); 2120 2121 close_netdev(info); 2122 #endif 2123 xenbus_switch_state(dev, XenbusStateClosed); 2124 } 2125 #endif 2126 2127 static int 2128 netfront_detach(device_t dev) 2129 { 2130 struct netfront_info *info = device_get_softc(dev); 2131 2132 DPRINTK("%s\n", xenbus_get_node(dev)); 2133 2134 netif_free(info); 2135 2136 return 0; 2137 } 2138 2139 static void 2140 netif_free(struct netfront_info *info) 2141 { 2142 XN_LOCK(info); 2143 xn_stop(info); 2144 XN_UNLOCK(info); 2145 callout_drain(&info->xn_stat_ch); 2146 netif_disconnect_backend(info); 2147 if (info->xn_ifp != NULL) { 2148 ether_ifdetach(info->xn_ifp); 2149 if_free(info->xn_ifp); 2150 info->xn_ifp = NULL; 2151 } 2152 ifmedia_removeall(&info->sc_media); 2153 } 2154 2155 static void 2156 netif_disconnect_backend(struct netfront_info *info) 2157 { 2158 XN_RX_LOCK(info); 2159 XN_TX_LOCK(info); 2160 netfront_carrier_off(info); 2161 XN_TX_UNLOCK(info); 2162 XN_RX_UNLOCK(info); 2163 2164 free_ring(&info->tx_ring_ref, &info->tx.sring); 2165 free_ring(&info->rx_ring_ref, &info->rx.sring); 2166 2167 xen_intr_unbind(&info->xen_intr_handle); 2168 } 2169 2170 static void 2171 free_ring(int *ref, void *ring_ptr_ref) 2172 { 2173 void **ring_ptr_ptr = ring_ptr_ref; 2174 2175 if (*ref != GRANT_REF_INVALID) { 2176 /* This API frees the associated storage. */ 2177 gnttab_end_foreign_access(*ref, *ring_ptr_ptr); 2178 *ref = GRANT_REF_INVALID; 2179 } 2180 *ring_ptr_ptr = NULL; 2181 } 2182 2183 static int 2184 xn_ifmedia_upd(struct ifnet *ifp) 2185 { 2186 return (0); 2187 } 2188 2189 static void 2190 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 2191 { 2192 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE; 2193 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL; 2194 } 2195 2196 /* ** Driver registration ** */ 2197 static device_method_t netfront_methods[] = { 2198 /* Device interface */ 2199 DEVMETHOD(device_probe, netfront_probe), 2200 DEVMETHOD(device_attach, netfront_attach), 2201 DEVMETHOD(device_detach, netfront_detach), 2202 DEVMETHOD(device_shutdown, bus_generic_shutdown), 2203 DEVMETHOD(device_suspend, netfront_suspend), 2204 DEVMETHOD(device_resume, netfront_resume), 2205 2206 /* Xenbus interface */ 2207 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed), 2208 2209 DEVMETHOD_END 2210 }; 2211 2212 static driver_t netfront_driver = { 2213 "xn", 2214 netfront_methods, 2215 sizeof(struct netfront_info), 2216 }; 2217 devclass_t netfront_devclass; 2218 2219 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL, 2220 NULL); 2221