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/mbuf.h> 37 #include <sys/malloc.h> 38 #include <sys/module.h> 39 #include <sys/kernel.h> 40 #include <sys/socket.h> 41 #include <sys/sysctl.h> 42 #include <sys/queue.h> 43 #include <sys/lock.h> 44 #include <sys/sx.h> 45 46 #include <net/if.h> 47 #include <net/if_var.h> 48 #include <net/if_arp.h> 49 #include <net/ethernet.h> 50 #include <net/if_dl.h> 51 #include <net/if_media.h> 52 53 #include <net/bpf.h> 54 55 #include <net/if_types.h> 56 57 #include <netinet/in_systm.h> 58 #include <netinet/in.h> 59 #include <netinet/ip.h> 60 #include <netinet/if_ether.h> 61 #if __FreeBSD_version >= 700000 62 #include <netinet/tcp.h> 63 #include <netinet/tcp_lro.h> 64 #endif 65 66 #include <vm/vm.h> 67 #include <vm/pmap.h> 68 69 #include <machine/clock.h> /* for DELAY */ 70 #include <machine/bus.h> 71 #include <machine/resource.h> 72 #include <machine/frame.h> 73 #include <machine/vmparam.h> 74 75 #include <sys/bus.h> 76 #include <sys/rman.h> 77 78 #include <machine/intr_machdep.h> 79 80 #include <xen/xen-os.h> 81 #include <xen/hypervisor.h> 82 #include <xen/xen_intr.h> 83 #include <xen/gnttab.h> 84 #include <xen/interface/memory.h> 85 #include <xen/interface/io/netif.h> 86 #include <xen/xenbus/xenbusvar.h> 87 88 #include <machine/xen/xenvar.h> 89 90 #include <dev/xen/netfront/mbufq.h> 91 92 #include "xenbus_if.h" 93 94 /* Features supported by all backends. TSO and LRO can be negotiated */ 95 #define XN_CSUM_FEATURES (CSUM_TCP | CSUM_UDP) 96 97 #define NET_TX_RING_SIZE __RING_SIZE((netif_tx_sring_t *)0, PAGE_SIZE) 98 #define NET_RX_RING_SIZE __RING_SIZE((netif_rx_sring_t *)0, PAGE_SIZE) 99 100 #if __FreeBSD_version >= 700000 101 /* 102 * Should the driver do LRO on the RX end 103 * this can be toggled on the fly, but the 104 * interface must be reset (down/up) for it 105 * to take effect. 106 */ 107 static int xn_enable_lro = 1; 108 TUNABLE_INT("hw.xn.enable_lro", &xn_enable_lro); 109 #else 110 111 #define IFCAP_TSO4 0 112 #define CSUM_TSO 0 113 114 #endif 115 116 #ifdef CONFIG_XEN 117 static int MODPARM_rx_copy = 0; 118 module_param_named(rx_copy, MODPARM_rx_copy, bool, 0); 119 MODULE_PARM_DESC(rx_copy, "Copy packets from network card (rather than flip)"); 120 static int MODPARM_rx_flip = 0; 121 module_param_named(rx_flip, MODPARM_rx_flip, bool, 0); 122 MODULE_PARM_DESC(rx_flip, "Flip packets from network card (rather than copy)"); 123 #else 124 static const int MODPARM_rx_copy = 1; 125 static const int MODPARM_rx_flip = 0; 126 #endif 127 128 /** 129 * \brief The maximum allowed data fragments in a single transmit 130 * request. 131 * 132 * This limit is imposed by the backend driver. We assume here that 133 * we are dealing with a Linux driver domain and have set our limit 134 * to mirror the Linux MAX_SKB_FRAGS constant. 135 */ 136 #define MAX_TX_REQ_FRAGS (65536 / PAGE_SIZE + 2) 137 138 #define RX_COPY_THRESHOLD 256 139 140 #define net_ratelimit() 0 141 142 struct netfront_info; 143 struct netfront_rx_info; 144 145 static void xn_txeof(struct netfront_info *); 146 static void xn_rxeof(struct netfront_info *); 147 static void network_alloc_rx_buffers(struct netfront_info *); 148 149 static void xn_tick_locked(struct netfront_info *); 150 static void xn_tick(void *); 151 152 static void xn_intr(void *); 153 static inline int xn_count_frags(struct mbuf *m); 154 static int xn_assemble_tx_request(struct netfront_info *sc, 155 struct mbuf *m_head); 156 static void xn_start_locked(struct ifnet *); 157 static void xn_start(struct ifnet *); 158 static int xn_ioctl(struct ifnet *, u_long, caddr_t); 159 static void xn_ifinit_locked(struct netfront_info *); 160 static void xn_ifinit(void *); 161 static void xn_stop(struct netfront_info *); 162 static void xn_query_features(struct netfront_info *np); 163 static int xn_configure_features(struct netfront_info *np); 164 #ifdef notyet 165 static void xn_watchdog(struct ifnet *); 166 #endif 167 168 #ifdef notyet 169 static void netfront_closing(device_t dev); 170 #endif 171 static void netif_free(struct netfront_info *info); 172 static int netfront_detach(device_t dev); 173 174 static int talk_to_backend(device_t dev, struct netfront_info *info); 175 static int create_netdev(device_t dev); 176 static void netif_disconnect_backend(struct netfront_info *info); 177 static int setup_device(device_t dev, struct netfront_info *info); 178 static void free_ring(int *ref, void *ring_ptr_ref); 179 180 static int xn_ifmedia_upd(struct ifnet *ifp); 181 static void xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr); 182 183 /* Xenolinux helper functions */ 184 int network_connect(struct netfront_info *); 185 186 static void xn_free_rx_ring(struct netfront_info *); 187 188 static void xn_free_tx_ring(struct netfront_info *); 189 190 static int xennet_get_responses(struct netfront_info *np, 191 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, 192 struct mbuf **list, int *pages_flipped_p); 193 194 #define virt_to_mfn(x) (vtomach(x) >> PAGE_SHIFT) 195 196 #define INVALID_P2M_ENTRY (~0UL) 197 198 /* 199 * Mbuf pointers. We need these to keep track of the virtual addresses 200 * of our mbuf chains since we can only convert from virtual to physical, 201 * not the other way around. The size must track the free index arrays. 202 */ 203 struct xn_chain_data { 204 struct mbuf *xn_tx_chain[NET_TX_RING_SIZE+1]; 205 int xn_tx_chain_cnt; 206 struct mbuf *xn_rx_chain[NET_RX_RING_SIZE+1]; 207 }; 208 209 struct net_device_stats 210 { 211 u_long rx_packets; /* total packets received */ 212 u_long tx_packets; /* total packets transmitted */ 213 u_long rx_bytes; /* total bytes received */ 214 u_long tx_bytes; /* total bytes transmitted */ 215 u_long rx_errors; /* bad packets received */ 216 u_long tx_errors; /* packet transmit problems */ 217 u_long rx_dropped; /* no space in linux buffers */ 218 u_long tx_dropped; /* no space available in linux */ 219 u_long multicast; /* multicast packets received */ 220 u_long collisions; 221 222 /* detailed rx_errors: */ 223 u_long rx_length_errors; 224 u_long rx_over_errors; /* receiver ring buff overflow */ 225 u_long rx_crc_errors; /* recved pkt with crc error */ 226 u_long rx_frame_errors; /* recv'd frame alignment error */ 227 u_long rx_fifo_errors; /* recv'r fifo overrun */ 228 u_long rx_missed_errors; /* receiver missed packet */ 229 230 /* detailed tx_errors */ 231 u_long tx_aborted_errors; 232 u_long tx_carrier_errors; 233 u_long tx_fifo_errors; 234 u_long tx_heartbeat_errors; 235 u_long tx_window_errors; 236 237 /* for cslip etc */ 238 u_long rx_compressed; 239 u_long tx_compressed; 240 }; 241 242 struct netfront_info { 243 struct ifnet *xn_ifp; 244 #if __FreeBSD_version >= 700000 245 struct lro_ctrl xn_lro; 246 #endif 247 248 struct net_device_stats stats; 249 u_int tx_full; 250 251 netif_tx_front_ring_t tx; 252 netif_rx_front_ring_t rx; 253 254 struct mtx tx_lock; 255 struct mtx rx_lock; 256 struct mtx sc_lock; 257 258 xen_intr_handle_t xen_intr_handle; 259 u_int copying_receiver; 260 u_int carrier; 261 u_int maxfrags; 262 263 /* Receive-ring batched refills. */ 264 #define RX_MIN_TARGET 32 265 #define RX_MAX_TARGET NET_RX_RING_SIZE 266 int rx_min_target; 267 int rx_max_target; 268 int rx_target; 269 270 grant_ref_t gref_tx_head; 271 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE + 1]; 272 grant_ref_t gref_rx_head; 273 grant_ref_t grant_rx_ref[NET_TX_RING_SIZE + 1]; 274 275 device_t xbdev; 276 int tx_ring_ref; 277 int rx_ring_ref; 278 uint8_t mac[ETHER_ADDR_LEN]; 279 struct xn_chain_data xn_cdata; /* mbufs */ 280 struct mbuf_head xn_rx_batch; /* head of the batch queue */ 281 282 int xn_if_flags; 283 struct callout xn_stat_ch; 284 285 u_long rx_pfn_array[NET_RX_RING_SIZE]; 286 multicall_entry_t rx_mcl[NET_RX_RING_SIZE+1]; 287 mmu_update_t rx_mmu[NET_RX_RING_SIZE]; 288 struct ifmedia sc_media; 289 }; 290 291 #define rx_mbufs xn_cdata.xn_rx_chain 292 #define tx_mbufs xn_cdata.xn_tx_chain 293 294 #define XN_LOCK_INIT(_sc, _name) \ 295 mtx_init(&(_sc)->tx_lock, #_name"_tx", "network transmit lock", MTX_DEF); \ 296 mtx_init(&(_sc)->rx_lock, #_name"_rx", "network receive lock", MTX_DEF); \ 297 mtx_init(&(_sc)->sc_lock, #_name"_sc", "netfront softc lock", MTX_DEF) 298 299 #define XN_RX_LOCK(_sc) mtx_lock(&(_sc)->rx_lock) 300 #define XN_RX_UNLOCK(_sc) mtx_unlock(&(_sc)->rx_lock) 301 302 #define XN_TX_LOCK(_sc) mtx_lock(&(_sc)->tx_lock) 303 #define XN_TX_UNLOCK(_sc) mtx_unlock(&(_sc)->tx_lock) 304 305 #define XN_LOCK(_sc) mtx_lock(&(_sc)->sc_lock); 306 #define XN_UNLOCK(_sc) mtx_unlock(&(_sc)->sc_lock); 307 308 #define XN_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->sc_lock, MA_OWNED); 309 #define XN_RX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->rx_lock, MA_OWNED); 310 #define XN_TX_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->tx_lock, MA_OWNED); 311 #define XN_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->rx_lock); \ 312 mtx_destroy(&(_sc)->tx_lock); \ 313 mtx_destroy(&(_sc)->sc_lock); 314 315 struct netfront_rx_info { 316 struct netif_rx_response rx; 317 struct netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; 318 }; 319 320 #define netfront_carrier_on(netif) ((netif)->carrier = 1) 321 #define netfront_carrier_off(netif) ((netif)->carrier = 0) 322 #define netfront_carrier_ok(netif) ((netif)->carrier) 323 324 /* Access macros for acquiring freeing slots in xn_free_{tx,rx}_idxs[]. */ 325 326 static inline void 327 add_id_to_freelist(struct mbuf **list, uintptr_t id) 328 { 329 KASSERT(id != 0, 330 ("%s: the head item (0) must always be free.", __func__)); 331 list[id] = list[0]; 332 list[0] = (struct mbuf *)id; 333 } 334 335 static inline unsigned short 336 get_id_from_freelist(struct mbuf **list) 337 { 338 uintptr_t id; 339 340 id = (uintptr_t)list[0]; 341 KASSERT(id != 0, 342 ("%s: the head item (0) must always remain free.", __func__)); 343 list[0] = list[id]; 344 return (id); 345 } 346 347 static inline int 348 xennet_rxidx(RING_IDX idx) 349 { 350 return idx & (NET_RX_RING_SIZE - 1); 351 } 352 353 static inline struct mbuf * 354 xennet_get_rx_mbuf(struct netfront_info *np, RING_IDX ri) 355 { 356 int i = xennet_rxidx(ri); 357 struct mbuf *m; 358 359 m = np->rx_mbufs[i]; 360 np->rx_mbufs[i] = NULL; 361 return (m); 362 } 363 364 static inline grant_ref_t 365 xennet_get_rx_ref(struct netfront_info *np, RING_IDX ri) 366 { 367 int i = xennet_rxidx(ri); 368 grant_ref_t ref = np->grant_rx_ref[i]; 369 KASSERT(ref != GRANT_REF_INVALID, ("Invalid grant reference!\n")); 370 np->grant_rx_ref[i] = GRANT_REF_INVALID; 371 return ref; 372 } 373 374 #define IPRINTK(fmt, args...) \ 375 printf("[XEN] " fmt, ##args) 376 #ifdef INVARIANTS 377 #define WPRINTK(fmt, args...) \ 378 printf("[XEN] " fmt, ##args) 379 #else 380 #define WPRINTK(fmt, args...) 381 #endif 382 #ifdef DEBUG 383 #define DPRINTK(fmt, args...) \ 384 printf("[XEN] %s: " fmt, __func__, ##args) 385 #else 386 #define DPRINTK(fmt, args...) 387 #endif 388 389 /** 390 * Read the 'mac' node at the given device's node in the store, and parse that 391 * as colon-separated octets, placing result the given mac array. mac must be 392 * a preallocated array of length ETH_ALEN (as declared in linux/if_ether.h). 393 * Return 0 on success, or errno on error. 394 */ 395 static int 396 xen_net_read_mac(device_t dev, uint8_t mac[]) 397 { 398 int error, i; 399 char *s, *e, *macstr; 400 const char *path; 401 402 path = xenbus_get_node(dev); 403 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr); 404 if (error == ENOENT) { 405 /* 406 * Deal with missing mac XenStore nodes on devices with 407 * HVM emulation (the 'ioemu' configuration attribute) 408 * enabled. 409 * 410 * The HVM emulator may execute in a stub device model 411 * domain which lacks the permission, only given to Dom0, 412 * to update the guest's XenStore tree. For this reason, 413 * the HVM emulator doesn't even attempt to write the 414 * front-side mac node, even when operating in Dom0. 415 * However, there should always be a mac listed in the 416 * backend tree. Fallback to this version if our query 417 * of the front side XenStore location doesn't find 418 * anything. 419 */ 420 path = xenbus_get_otherend_path(dev); 421 error = xs_read(XST_NIL, path, "mac", NULL, (void **) &macstr); 422 } 423 if (error != 0) { 424 xenbus_dev_fatal(dev, error, "parsing %s/mac", path); 425 return (error); 426 } 427 428 s = macstr; 429 for (i = 0; i < ETHER_ADDR_LEN; i++) { 430 mac[i] = strtoul(s, &e, 16); 431 if (s == e || (e[0] != ':' && e[0] != 0)) { 432 free(macstr, M_XENBUS); 433 return (ENOENT); 434 } 435 s = &e[1]; 436 } 437 free(macstr, M_XENBUS); 438 return (0); 439 } 440 441 /** 442 * Entry point to this code when a new device is created. Allocate the basic 443 * structures and the ring buffers for communication with the backend, and 444 * inform the backend of the appropriate details for those. Switch to 445 * Connected state. 446 */ 447 static int 448 netfront_probe(device_t dev) 449 { 450 451 if (!strcmp(xenbus_get_type(dev), "vif")) { 452 device_set_desc(dev, "Virtual Network Interface"); 453 return (0); 454 } 455 456 return (ENXIO); 457 } 458 459 static int 460 netfront_attach(device_t dev) 461 { 462 int err; 463 464 err = create_netdev(dev); 465 if (err) { 466 xenbus_dev_fatal(dev, err, "creating netdev"); 467 return (err); 468 } 469 470 #if __FreeBSD_version >= 700000 471 SYSCTL_ADD_INT(device_get_sysctl_ctx(dev), 472 SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), 473 OID_AUTO, "enable_lro", CTLFLAG_RW, 474 &xn_enable_lro, 0, "Large Receive Offload"); 475 #endif 476 477 return (0); 478 } 479 480 static int 481 netfront_suspend(device_t dev) 482 { 483 struct netfront_info *info = device_get_softc(dev); 484 485 XN_RX_LOCK(info); 486 XN_TX_LOCK(info); 487 netfront_carrier_off(info); 488 XN_TX_UNLOCK(info); 489 XN_RX_UNLOCK(info); 490 return (0); 491 } 492 493 /** 494 * We are reconnecting to the backend, due to a suspend/resume, or a backend 495 * driver restart. We tear down our netif structure and recreate it, but 496 * leave the device-layer structures intact so that this is transparent to the 497 * rest of the kernel. 498 */ 499 static int 500 netfront_resume(device_t dev) 501 { 502 struct netfront_info *info = device_get_softc(dev); 503 504 netif_disconnect_backend(info); 505 return (0); 506 } 507 508 /* Common code used when first setting up, and when resuming. */ 509 static int 510 talk_to_backend(device_t dev, struct netfront_info *info) 511 { 512 const char *message; 513 struct xs_transaction xst; 514 const char *node = xenbus_get_node(dev); 515 int err; 516 517 err = xen_net_read_mac(dev, info->mac); 518 if (err) { 519 xenbus_dev_fatal(dev, err, "parsing %s/mac", node); 520 goto out; 521 } 522 523 /* Create shared ring, alloc event channel. */ 524 err = setup_device(dev, info); 525 if (err) 526 goto out; 527 528 again: 529 err = xs_transaction_start(&xst); 530 if (err) { 531 xenbus_dev_fatal(dev, err, "starting transaction"); 532 goto destroy_ring; 533 } 534 err = xs_printf(xst, node, "tx-ring-ref","%u", 535 info->tx_ring_ref); 536 if (err) { 537 message = "writing tx ring-ref"; 538 goto abort_transaction; 539 } 540 err = xs_printf(xst, node, "rx-ring-ref","%u", 541 info->rx_ring_ref); 542 if (err) { 543 message = "writing rx ring-ref"; 544 goto abort_transaction; 545 } 546 err = xs_printf(xst, node, 547 "event-channel", "%u", 548 xen_intr_port(info->xen_intr_handle)); 549 if (err) { 550 message = "writing event-channel"; 551 goto abort_transaction; 552 } 553 err = xs_printf(xst, node, "request-rx-copy", "%u", 554 info->copying_receiver); 555 if (err) { 556 message = "writing request-rx-copy"; 557 goto abort_transaction; 558 } 559 err = xs_printf(xst, node, "feature-rx-notify", "%d", 1); 560 if (err) { 561 message = "writing feature-rx-notify"; 562 goto abort_transaction; 563 } 564 err = xs_printf(xst, node, "feature-sg", "%d", 1); 565 if (err) { 566 message = "writing feature-sg"; 567 goto abort_transaction; 568 } 569 #if __FreeBSD_version >= 700000 570 err = xs_printf(xst, node, "feature-gso-tcpv4", "%d", 1); 571 if (err) { 572 message = "writing feature-gso-tcpv4"; 573 goto abort_transaction; 574 } 575 #endif 576 577 err = xs_transaction_end(xst, 0); 578 if (err) { 579 if (err == EAGAIN) 580 goto again; 581 xenbus_dev_fatal(dev, err, "completing transaction"); 582 goto destroy_ring; 583 } 584 585 return 0; 586 587 abort_transaction: 588 xs_transaction_end(xst, 1); 589 xenbus_dev_fatal(dev, err, "%s", message); 590 destroy_ring: 591 netif_free(info); 592 out: 593 return err; 594 } 595 596 static int 597 setup_device(device_t dev, struct netfront_info *info) 598 { 599 netif_tx_sring_t *txs; 600 netif_rx_sring_t *rxs; 601 int error; 602 struct ifnet *ifp; 603 604 ifp = info->xn_ifp; 605 606 info->tx_ring_ref = GRANT_REF_INVALID; 607 info->rx_ring_ref = GRANT_REF_INVALID; 608 info->rx.sring = NULL; 609 info->tx.sring = NULL; 610 611 txs = (netif_tx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO); 612 if (!txs) { 613 error = ENOMEM; 614 xenbus_dev_fatal(dev, error, "allocating tx ring page"); 615 goto fail; 616 } 617 SHARED_RING_INIT(txs); 618 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); 619 error = xenbus_grant_ring(dev, virt_to_mfn(txs), &info->tx_ring_ref); 620 if (error) 621 goto fail; 622 623 rxs = (netif_rx_sring_t *)malloc(PAGE_SIZE, M_DEVBUF, M_NOWAIT|M_ZERO); 624 if (!rxs) { 625 error = ENOMEM; 626 xenbus_dev_fatal(dev, error, "allocating rx ring page"); 627 goto fail; 628 } 629 SHARED_RING_INIT(rxs); 630 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); 631 632 error = xenbus_grant_ring(dev, virt_to_mfn(rxs), &info->rx_ring_ref); 633 if (error) 634 goto fail; 635 636 error = xen_intr_alloc_and_bind_local_port(dev, 637 xenbus_get_otherend_id(dev), /*filter*/NULL, xn_intr, info, 638 INTR_TYPE_NET | INTR_MPSAFE | INTR_ENTROPY, &info->xen_intr_handle); 639 640 if (error) { 641 xenbus_dev_fatal(dev, error, 642 "xen_intr_alloc_and_bind_local_port failed"); 643 goto fail; 644 } 645 646 return (0); 647 648 fail: 649 netif_free(info); 650 return (error); 651 } 652 653 #ifdef INET 654 /** 655 * If this interface has an ipv4 address, send an arp for it. This 656 * helps to get the network going again after migrating hosts. 657 */ 658 static void 659 netfront_send_fake_arp(device_t dev, struct netfront_info *info) 660 { 661 struct ifnet *ifp; 662 struct ifaddr *ifa; 663 664 ifp = info->xn_ifp; 665 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 666 if (ifa->ifa_addr->sa_family == AF_INET) { 667 arp_ifinit(ifp, ifa); 668 } 669 } 670 } 671 #endif 672 673 /** 674 * Callback received when the backend's state changes. 675 */ 676 static void 677 netfront_backend_changed(device_t dev, XenbusState newstate) 678 { 679 struct netfront_info *sc = device_get_softc(dev); 680 681 DPRINTK("newstate=%d\n", newstate); 682 683 switch (newstate) { 684 case XenbusStateInitialising: 685 case XenbusStateInitialised: 686 case XenbusStateConnected: 687 case XenbusStateUnknown: 688 case XenbusStateClosed: 689 case XenbusStateReconfigured: 690 case XenbusStateReconfiguring: 691 break; 692 case XenbusStateInitWait: 693 if (xenbus_get_state(dev) != XenbusStateInitialising) 694 break; 695 if (network_connect(sc) != 0) 696 break; 697 xenbus_set_state(dev, XenbusStateConnected); 698 #ifdef INET 699 netfront_send_fake_arp(dev, sc); 700 #endif 701 break; 702 case XenbusStateClosing: 703 xenbus_set_state(dev, XenbusStateClosed); 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 MGETHDR(m_new, M_NOWAIT, MT_DATA); 820 if (m_new == NULL) { 821 printf("%s: MGETHDR failed\n", __func__); 822 goto no_mbuf; 823 } 824 825 if (m_cljget(m_new, M_NOWAIT, MJUMPAGESIZE) == NULL) { 826 printf("%s: m_cljget failed\n", __func__); 827 m_freem(m_new); 828 829 no_mbuf: 830 if (i != 0) 831 goto refill; 832 /* 833 * XXX set timer 834 */ 835 break; 836 } 837 m_new->m_len = m_new->m_pkthdr.len = MJUMPAGESIZE; 838 839 /* queue the mbufs allocated */ 840 mbufq_tail(&sc->xn_rx_batch, m_new); 841 } 842 843 /* 844 * If we've allocated at least half of our target number of entries, 845 * submit them to the backend - we have enough to make the overhead 846 * of submission worthwhile. Otherwise wait for more mbufs and 847 * request entries to become available. 848 */ 849 if (i < (sc->rx_target/2)) { 850 if (req_prod >sc->rx.sring->req_prod) 851 goto push; 852 return; 853 } 854 855 /* 856 * Double floating fill target if we risked having the backend 857 * run out of empty buffers for receive traffic. We define "running 858 * low" as having less than a fourth of our target buffers free 859 * at the time we refilled the queue. 860 */ 861 if ((req_prod - sc->rx.sring->rsp_prod) < (sc->rx_target / 4)) { 862 sc->rx_target *= 2; 863 if (sc->rx_target > sc->rx_max_target) 864 sc->rx_target = sc->rx_max_target; 865 } 866 867 refill: 868 for (nr_flips = i = 0; ; i++) { 869 if ((m_new = mbufq_dequeue(&sc->xn_rx_batch)) == NULL) 870 break; 871 872 m_new->m_ext.ext_arg1 = (vm_paddr_t *)(uintptr_t)( 873 vtophys(m_new->m_ext.ext_buf) >> PAGE_SHIFT); 874 875 id = xennet_rxidx(req_prod + i); 876 877 KASSERT(sc->rx_mbufs[id] == NULL, ("non-NULL xm_rx_chain")); 878 sc->rx_mbufs[id] = m_new; 879 880 ref = gnttab_claim_grant_reference(&sc->gref_rx_head); 881 KASSERT(ref != GNTTAB_LIST_END, 882 ("reserved grant references exhuasted")); 883 sc->grant_rx_ref[id] = ref; 884 885 vaddr = mtod(m_new, vm_offset_t); 886 pfn = vtophys(vaddr) >> PAGE_SHIFT; 887 req = RING_GET_REQUEST(&sc->rx, req_prod + i); 888 889 if (sc->copying_receiver == 0) { 890 gnttab_grant_foreign_transfer_ref(ref, 891 otherend_id, pfn); 892 sc->rx_pfn_array[nr_flips] = PFNTOMFN(pfn); 893 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 894 /* Remove this page before passing 895 * back to Xen. 896 */ 897 set_phys_to_machine(pfn, INVALID_P2M_ENTRY); 898 MULTI_update_va_mapping(&sc->rx_mcl[i], 899 vaddr, 0, 0); 900 } 901 nr_flips++; 902 } else { 903 gnttab_grant_foreign_access_ref(ref, 904 otherend_id, 905 PFNTOMFN(pfn), 0); 906 } 907 req->id = id; 908 req->gref = ref; 909 910 sc->rx_pfn_array[i] = 911 vtomach(mtod(m_new,vm_offset_t)) >> PAGE_SHIFT; 912 } 913 914 KASSERT(i, ("no mbufs processed")); /* should have returned earlier */ 915 KASSERT(mbufq_len(&sc->xn_rx_batch) == 0, ("not all mbufs processed")); 916 /* 917 * We may have allocated buffers which have entries outstanding 918 * in the page * update queue -- make sure we flush those first! 919 */ 920 PT_UPDATES_FLUSH(); 921 if (nr_flips != 0) { 922 #ifdef notyet 923 /* Tell the ballon driver what is going on. */ 924 balloon_update_driver_allowance(i); 925 #endif 926 set_xen_guest_handle(reservation.extent_start, sc->rx_pfn_array); 927 reservation.nr_extents = i; 928 reservation.extent_order = 0; 929 reservation.address_bits = 0; 930 reservation.domid = DOMID_SELF; 931 932 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 933 /* After all PTEs have been zapped, flush the TLB. */ 934 sc->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] = 935 UVMF_TLB_FLUSH|UVMF_ALL; 936 937 /* Give away a batch of pages. */ 938 sc->rx_mcl[i].op = __HYPERVISOR_memory_op; 939 sc->rx_mcl[i].args[0] = XENMEM_decrease_reservation; 940 sc->rx_mcl[i].args[1] = (u_long)&reservation; 941 /* Zap PTEs and give away pages in one big multicall. */ 942 (void)HYPERVISOR_multicall(sc->rx_mcl, i+1); 943 944 if (__predict_false(sc->rx_mcl[i].result != i || 945 HYPERVISOR_memory_op(XENMEM_decrease_reservation, 946 &reservation) != i)) 947 panic("%s: unable to reduce memory " 948 "reservation\n", __func__); 949 } 950 } else { 951 wmb(); 952 } 953 954 /* Above is a suitable barrier to ensure backend will see requests. */ 955 sc->rx.req_prod_pvt = req_prod + i; 956 push: 957 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->rx, notify); 958 if (notify) 959 xen_intr_signal(sc->xen_intr_handle); 960 } 961 962 static void 963 xn_rxeof(struct netfront_info *np) 964 { 965 struct ifnet *ifp; 966 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 967 struct lro_ctrl *lro = &np->xn_lro; 968 struct lro_entry *queued; 969 #endif 970 struct netfront_rx_info rinfo; 971 struct netif_rx_response *rx = &rinfo.rx; 972 struct netif_extra_info *extras = rinfo.extras; 973 RING_IDX i, rp; 974 multicall_entry_t *mcl; 975 struct mbuf *m; 976 struct mbuf_head rxq, errq; 977 int err, pages_flipped = 0, work_to_do; 978 979 do { 980 XN_RX_LOCK_ASSERT(np); 981 if (!netfront_carrier_ok(np)) 982 return; 983 984 mbufq_init(&errq); 985 mbufq_init(&rxq); 986 987 ifp = np->xn_ifp; 988 989 rp = np->rx.sring->rsp_prod; 990 rmb(); /* Ensure we see queued responses up to 'rp'. */ 991 992 i = np->rx.rsp_cons; 993 while ((i != rp)) { 994 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); 995 memset(extras, 0, sizeof(rinfo.extras)); 996 997 m = NULL; 998 err = xennet_get_responses(np, &rinfo, rp, &i, &m, 999 &pages_flipped); 1000 1001 if (__predict_false(err)) { 1002 if (m) 1003 mbufq_tail(&errq, m); 1004 np->stats.rx_errors++; 1005 continue; 1006 } 1007 1008 m->m_pkthdr.rcvif = ifp; 1009 if ( rx->flags & NETRXF_data_validated ) { 1010 /* Tell the stack the checksums are okay */ 1011 /* 1012 * XXX this isn't necessarily the case - need to add 1013 * check 1014 */ 1015 1016 m->m_pkthdr.csum_flags |= 1017 (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID 1018 | CSUM_PSEUDO_HDR); 1019 m->m_pkthdr.csum_data = 0xffff; 1020 } 1021 1022 np->stats.rx_packets++; 1023 np->stats.rx_bytes += m->m_pkthdr.len; 1024 1025 mbufq_tail(&rxq, m); 1026 np->rx.rsp_cons = i; 1027 } 1028 1029 if (pages_flipped) { 1030 /* Some pages are no longer absent... */ 1031 #ifdef notyet 1032 balloon_update_driver_allowance(-pages_flipped); 1033 #endif 1034 /* Do all the remapping work, and M->P updates, in one big 1035 * hypercall. 1036 */ 1037 if (!!xen_feature(XENFEAT_auto_translated_physmap)) { 1038 mcl = np->rx_mcl + pages_flipped; 1039 mcl->op = __HYPERVISOR_mmu_update; 1040 mcl->args[0] = (u_long)np->rx_mmu; 1041 mcl->args[1] = pages_flipped; 1042 mcl->args[2] = 0; 1043 mcl->args[3] = DOMID_SELF; 1044 (void)HYPERVISOR_multicall(np->rx_mcl, 1045 pages_flipped + 1); 1046 } 1047 } 1048 1049 while ((m = mbufq_dequeue(&errq))) 1050 m_freem(m); 1051 1052 /* 1053 * Process all the mbufs after the remapping is complete. 1054 * Break the mbuf chain first though. 1055 */ 1056 while ((m = mbufq_dequeue(&rxq)) != NULL) { 1057 if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1); 1058 1059 /* 1060 * Do we really need to drop the rx lock? 1061 */ 1062 XN_RX_UNLOCK(np); 1063 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1064 /* Use LRO if possible */ 1065 if ((ifp->if_capenable & IFCAP_LRO) == 0 || 1066 lro->lro_cnt == 0 || tcp_lro_rx(lro, m, 0)) { 1067 /* 1068 * If LRO fails, pass up to the stack 1069 * directly. 1070 */ 1071 (*ifp->if_input)(ifp, m); 1072 } 1073 #else 1074 (*ifp->if_input)(ifp, m); 1075 #endif 1076 XN_RX_LOCK(np); 1077 } 1078 1079 np->rx.rsp_cons = i; 1080 1081 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 1082 /* 1083 * Flush any outstanding LRO work 1084 */ 1085 while (!SLIST_EMPTY(&lro->lro_active)) { 1086 queued = SLIST_FIRST(&lro->lro_active); 1087 SLIST_REMOVE_HEAD(&lro->lro_active, next); 1088 tcp_lro_flush(lro, queued); 1089 } 1090 #endif 1091 1092 #if 0 1093 /* If we get a callback with very few responses, reduce fill target. */ 1094 /* NB. Note exponential increase, linear decrease. */ 1095 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > 1096 ((3*np->rx_target) / 4)) && (--np->rx_target < np->rx_min_target)) 1097 np->rx_target = np->rx_min_target; 1098 #endif 1099 1100 network_alloc_rx_buffers(np); 1101 1102 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, work_to_do); 1103 } while (work_to_do); 1104 } 1105 1106 static void 1107 xn_txeof(struct netfront_info *np) 1108 { 1109 RING_IDX i, prod; 1110 unsigned short id; 1111 struct ifnet *ifp; 1112 netif_tx_response_t *txr; 1113 struct mbuf *m; 1114 1115 XN_TX_LOCK_ASSERT(np); 1116 1117 if (!netfront_carrier_ok(np)) 1118 return; 1119 1120 ifp = np->xn_ifp; 1121 1122 do { 1123 prod = np->tx.sring->rsp_prod; 1124 rmb(); /* Ensure we see responses up to 'rp'. */ 1125 1126 for (i = np->tx.rsp_cons; i != prod; i++) { 1127 txr = RING_GET_RESPONSE(&np->tx, i); 1128 if (txr->status == NETIF_RSP_NULL) 1129 continue; 1130 1131 if (txr->status != NETIF_RSP_OKAY) { 1132 printf("%s: WARNING: response is %d!\n", 1133 __func__, txr->status); 1134 } 1135 id = txr->id; 1136 m = np->tx_mbufs[id]; 1137 KASSERT(m != NULL, ("mbuf not found in xn_tx_chain")); 1138 KASSERT((uintptr_t)m > NET_TX_RING_SIZE, 1139 ("mbuf already on the free list, but we're " 1140 "trying to free it again!")); 1141 M_ASSERTVALID(m); 1142 1143 /* 1144 * Increment packet count if this is the last 1145 * mbuf of the chain. 1146 */ 1147 if (!m->m_next) 1148 if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); 1149 if (__predict_false(gnttab_query_foreign_access( 1150 np->grant_tx_ref[id]) != 0)) { 1151 panic("%s: grant id %u still in use by the " 1152 "backend", __func__, id); 1153 } 1154 gnttab_end_foreign_access_ref( 1155 np->grant_tx_ref[id]); 1156 gnttab_release_grant_reference( 1157 &np->gref_tx_head, np->grant_tx_ref[id]); 1158 np->grant_tx_ref[id] = GRANT_REF_INVALID; 1159 1160 np->tx_mbufs[id] = NULL; 1161 add_id_to_freelist(np->tx_mbufs, id); 1162 np->xn_cdata.xn_tx_chain_cnt--; 1163 m_free(m); 1164 /* Only mark the queue active if we've freed up at least one slot to try */ 1165 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1166 } 1167 np->tx.rsp_cons = prod; 1168 1169 /* 1170 * Set a new event, then check for race with update of 1171 * tx_cons. Note that it is essential to schedule a 1172 * callback, no matter how few buffers are pending. Even if 1173 * there is space in the transmit ring, higher layers may 1174 * be blocked because too much data is outstanding: in such 1175 * cases notification from Xen is likely to be the only kick 1176 * that we'll get. 1177 */ 1178 np->tx.sring->rsp_event = 1179 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; 1180 1181 mb(); 1182 } while (prod != np->tx.sring->rsp_prod); 1183 1184 if (np->tx_full && 1185 ((np->tx.sring->req_prod - prod) < NET_TX_RING_SIZE)) { 1186 np->tx_full = 0; 1187 #if 0 1188 if (np->user_state == UST_OPEN) 1189 netif_wake_queue(dev); 1190 #endif 1191 } 1192 } 1193 1194 static void 1195 xn_intr(void *xsc) 1196 { 1197 struct netfront_info *np = xsc; 1198 struct ifnet *ifp = np->xn_ifp; 1199 1200 #if 0 1201 if (!(np->rx.rsp_cons != np->rx.sring->rsp_prod && 1202 likely(netfront_carrier_ok(np)) && 1203 ifp->if_drv_flags & IFF_DRV_RUNNING)) 1204 return; 1205 #endif 1206 if (RING_HAS_UNCONSUMED_RESPONSES(&np->tx)) { 1207 XN_TX_LOCK(np); 1208 xn_txeof(np); 1209 XN_TX_UNLOCK(np); 1210 } 1211 1212 XN_RX_LOCK(np); 1213 xn_rxeof(np); 1214 XN_RX_UNLOCK(np); 1215 1216 if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1217 !IFQ_DRV_IS_EMPTY(&ifp->if_snd)) 1218 xn_start(ifp); 1219 } 1220 1221 static void 1222 xennet_move_rx_slot(struct netfront_info *np, struct mbuf *m, 1223 grant_ref_t ref) 1224 { 1225 int new = xennet_rxidx(np->rx.req_prod_pvt); 1226 1227 KASSERT(np->rx_mbufs[new] == NULL, ("rx_mbufs != NULL")); 1228 np->rx_mbufs[new] = m; 1229 np->grant_rx_ref[new] = ref; 1230 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; 1231 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; 1232 np->rx.req_prod_pvt++; 1233 } 1234 1235 static int 1236 xennet_get_extras(struct netfront_info *np, 1237 struct netif_extra_info *extras, RING_IDX rp, RING_IDX *cons) 1238 { 1239 struct netif_extra_info *extra; 1240 1241 int err = 0; 1242 1243 do { 1244 struct mbuf *m; 1245 grant_ref_t ref; 1246 1247 if (__predict_false(*cons + 1 == rp)) { 1248 #if 0 1249 if (net_ratelimit()) 1250 WPRINTK("Missing extra info\n"); 1251 #endif 1252 err = EINVAL; 1253 break; 1254 } 1255 1256 extra = (struct netif_extra_info *) 1257 RING_GET_RESPONSE(&np->rx, ++(*cons)); 1258 1259 if (__predict_false(!extra->type || 1260 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 1261 #if 0 1262 if (net_ratelimit()) 1263 WPRINTK("Invalid extra type: %d\n", 1264 extra->type); 1265 #endif 1266 err = EINVAL; 1267 } else { 1268 memcpy(&extras[extra->type - 1], extra, sizeof(*extra)); 1269 } 1270 1271 m = xennet_get_rx_mbuf(np, *cons); 1272 ref = xennet_get_rx_ref(np, *cons); 1273 xennet_move_rx_slot(np, m, ref); 1274 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 1275 1276 return err; 1277 } 1278 1279 static int 1280 xennet_get_responses(struct netfront_info *np, 1281 struct netfront_rx_info *rinfo, RING_IDX rp, RING_IDX *cons, 1282 struct mbuf **list, 1283 int *pages_flipped_p) 1284 { 1285 int pages_flipped = *pages_flipped_p; 1286 struct mmu_update *mmu; 1287 struct multicall_entry *mcl; 1288 struct netif_rx_response *rx = &rinfo->rx; 1289 struct netif_extra_info *extras = rinfo->extras; 1290 struct mbuf *m, *m0, *m_prev; 1291 grant_ref_t ref = xennet_get_rx_ref(np, *cons); 1292 RING_IDX ref_cons = *cons; 1293 int frags = 1; 1294 int err = 0; 1295 u_long ret; 1296 1297 m0 = m = m_prev = xennet_get_rx_mbuf(np, *cons); 1298 1299 if (rx->flags & NETRXF_extra_info) { 1300 err = xennet_get_extras(np, extras, rp, cons); 1301 } 1302 1303 if (m0 != NULL) { 1304 m0->m_pkthdr.len = 0; 1305 m0->m_next = NULL; 1306 } 1307 1308 for (;;) { 1309 u_long mfn; 1310 1311 #if 0 1312 DPRINTK("rx->status=%hd rx->offset=%hu frags=%u\n", 1313 rx->status, rx->offset, frags); 1314 #endif 1315 if (__predict_false(rx->status < 0 || 1316 rx->offset + rx->status > PAGE_SIZE)) { 1317 1318 #if 0 1319 if (net_ratelimit()) 1320 WPRINTK("rx->offset: %x, size: %u\n", 1321 rx->offset, rx->status); 1322 #endif 1323 xennet_move_rx_slot(np, m, ref); 1324 if (m0 == m) 1325 m0 = NULL; 1326 m = NULL; 1327 err = EINVAL; 1328 goto next_skip_queue; 1329 } 1330 1331 /* 1332 * This definitely indicates a bug, either in this driver or in 1333 * the backend driver. In future this should flag the bad 1334 * situation to the system controller to reboot the backed. 1335 */ 1336 if (ref == GRANT_REF_INVALID) { 1337 1338 #if 0 1339 if (net_ratelimit()) 1340 WPRINTK("Bad rx response id %d.\n", rx->id); 1341 #endif 1342 printf("%s: Bad rx response id %d.\n", __func__,rx->id); 1343 err = EINVAL; 1344 goto next; 1345 } 1346 1347 if (!np->copying_receiver) { 1348 /* Memory pressure, insufficient buffer 1349 * headroom, ... 1350 */ 1351 if (!(mfn = gnttab_end_foreign_transfer_ref(ref))) { 1352 WPRINTK("Unfulfilled rx req (id=%d, st=%d).\n", 1353 rx->id, rx->status); 1354 xennet_move_rx_slot(np, m, ref); 1355 err = ENOMEM; 1356 goto next; 1357 } 1358 1359 if (!xen_feature( XENFEAT_auto_translated_physmap)) { 1360 /* Remap the page. */ 1361 void *vaddr = mtod(m, void *); 1362 uint32_t pfn; 1363 1364 mcl = np->rx_mcl + pages_flipped; 1365 mmu = np->rx_mmu + pages_flipped; 1366 1367 MULTI_update_va_mapping(mcl, (u_long)vaddr, 1368 (((vm_paddr_t)mfn) << PAGE_SHIFT) | PG_RW | 1369 PG_V | PG_M | PG_A, 0); 1370 pfn = (uintptr_t)m->m_ext.ext_arg1; 1371 mmu->ptr = ((vm_paddr_t)mfn << PAGE_SHIFT) | 1372 MMU_MACHPHYS_UPDATE; 1373 mmu->val = pfn; 1374 1375 set_phys_to_machine(pfn, mfn); 1376 } 1377 pages_flipped++; 1378 } else { 1379 ret = gnttab_end_foreign_access_ref(ref); 1380 KASSERT(ret, ("ret != 0")); 1381 } 1382 1383 gnttab_release_grant_reference(&np->gref_rx_head, ref); 1384 1385 next: 1386 if (m == NULL) 1387 break; 1388 1389 m->m_len = rx->status; 1390 m->m_data += rx->offset; 1391 m0->m_pkthdr.len += rx->status; 1392 1393 next_skip_queue: 1394 if (!(rx->flags & NETRXF_more_data)) 1395 break; 1396 1397 if (*cons + frags == rp) { 1398 if (net_ratelimit()) 1399 WPRINTK("Need more frags\n"); 1400 err = ENOENT; 1401 printf("%s: cons %u frags %u rp %u, not enough frags\n", 1402 __func__, *cons, frags, rp); 1403 break; 1404 } 1405 /* 1406 * Note that m can be NULL, if rx->status < 0 or if 1407 * rx->offset + rx->status > PAGE_SIZE above. 1408 */ 1409 m_prev = m; 1410 1411 rx = RING_GET_RESPONSE(&np->rx, *cons + frags); 1412 m = xennet_get_rx_mbuf(np, *cons + frags); 1413 1414 /* 1415 * m_prev == NULL can happen if rx->status < 0 or if 1416 * rx->offset + * rx->status > PAGE_SIZE above. 1417 */ 1418 if (m_prev != NULL) 1419 m_prev->m_next = m; 1420 1421 /* 1422 * m0 can be NULL if rx->status < 0 or if * rx->offset + 1423 * rx->status > PAGE_SIZE above. 1424 */ 1425 if (m0 == NULL) 1426 m0 = m; 1427 m->m_next = NULL; 1428 ref = xennet_get_rx_ref(np, *cons + frags); 1429 ref_cons = *cons + frags; 1430 frags++; 1431 } 1432 *list = m0; 1433 *cons += frags; 1434 *pages_flipped_p = pages_flipped; 1435 1436 return (err); 1437 } 1438 1439 static void 1440 xn_tick_locked(struct netfront_info *sc) 1441 { 1442 XN_RX_LOCK_ASSERT(sc); 1443 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); 1444 1445 /* XXX placeholder for printing debug information */ 1446 } 1447 1448 static void 1449 xn_tick(void *xsc) 1450 { 1451 struct netfront_info *sc; 1452 1453 sc = xsc; 1454 XN_RX_LOCK(sc); 1455 xn_tick_locked(sc); 1456 XN_RX_UNLOCK(sc); 1457 } 1458 1459 /** 1460 * \brief Count the number of fragments in an mbuf chain. 1461 * 1462 * Surprisingly, there isn't an M* macro for this. 1463 */ 1464 static inline int 1465 xn_count_frags(struct mbuf *m) 1466 { 1467 int nfrags; 1468 1469 for (nfrags = 0; m != NULL; m = m->m_next) 1470 nfrags++; 1471 1472 return (nfrags); 1473 } 1474 1475 /** 1476 * Given an mbuf chain, make sure we have enough room and then push 1477 * it onto the transmit ring. 1478 */ 1479 static int 1480 xn_assemble_tx_request(struct netfront_info *sc, struct mbuf *m_head) 1481 { 1482 struct ifnet *ifp; 1483 struct mbuf *m; 1484 u_int nfrags; 1485 netif_extra_info_t *extra; 1486 int otherend_id; 1487 1488 ifp = sc->xn_ifp; 1489 1490 /** 1491 * Defragment the mbuf if necessary. 1492 */ 1493 nfrags = xn_count_frags(m_head); 1494 1495 /* 1496 * Check to see whether this request is longer than netback 1497 * can handle, and try to defrag it. 1498 */ 1499 /** 1500 * It is a bit lame, but the netback driver in Linux can't 1501 * deal with nfrags > MAX_TX_REQ_FRAGS, which is a quirk of 1502 * the Linux network stack. 1503 */ 1504 if (nfrags > sc->maxfrags) { 1505 m = m_defrag(m_head, M_NOWAIT); 1506 if (!m) { 1507 /* 1508 * Defrag failed, so free the mbuf and 1509 * therefore drop the packet. 1510 */ 1511 m_freem(m_head); 1512 return (EMSGSIZE); 1513 } 1514 m_head = m; 1515 } 1516 1517 /* Determine how many fragments now exist */ 1518 nfrags = xn_count_frags(m_head); 1519 1520 /* 1521 * Check to see whether the defragmented packet has too many 1522 * segments for the Linux netback driver. 1523 */ 1524 /** 1525 * The FreeBSD TCP stack, with TSO enabled, can produce a chain 1526 * of mbufs longer than Linux can handle. Make sure we don't 1527 * pass a too-long chain over to the other side by dropping the 1528 * packet. It doesn't look like there is currently a way to 1529 * tell the TCP stack to generate a shorter chain of packets. 1530 */ 1531 if (nfrags > MAX_TX_REQ_FRAGS) { 1532 #ifdef DEBUG 1533 printf("%s: nfrags %d > MAX_TX_REQ_FRAGS %d, netback " 1534 "won't be able to handle it, dropping\n", 1535 __func__, nfrags, MAX_TX_REQ_FRAGS); 1536 #endif 1537 m_freem(m_head); 1538 return (EMSGSIZE); 1539 } 1540 1541 /* 1542 * This check should be redundant. We've already verified that we 1543 * have enough slots in the ring to handle a packet of maximum 1544 * size, and that our packet is less than the maximum size. Keep 1545 * it in here as an assert for now just to make certain that 1546 * xn_tx_chain_cnt is accurate. 1547 */ 1548 KASSERT((sc->xn_cdata.xn_tx_chain_cnt + nfrags) <= NET_TX_RING_SIZE, 1549 ("%s: xn_tx_chain_cnt (%d) + nfrags (%d) > NET_TX_RING_SIZE " 1550 "(%d)!", __func__, (int) sc->xn_cdata.xn_tx_chain_cnt, 1551 (int) nfrags, (int) NET_TX_RING_SIZE)); 1552 1553 /* 1554 * Start packing the mbufs in this chain into 1555 * the fragment pointers. Stop when we run out 1556 * of fragments or hit the end of the mbuf chain. 1557 */ 1558 m = m_head; 1559 extra = NULL; 1560 otherend_id = xenbus_get_otherend_id(sc->xbdev); 1561 for (m = m_head; m; m = m->m_next) { 1562 netif_tx_request_t *tx; 1563 uintptr_t id; 1564 grant_ref_t ref; 1565 u_long mfn; /* XXX Wrong type? */ 1566 1567 tx = RING_GET_REQUEST(&sc->tx, sc->tx.req_prod_pvt); 1568 id = get_id_from_freelist(sc->tx_mbufs); 1569 if (id == 0) 1570 panic("%s: was allocated the freelist head!\n", 1571 __func__); 1572 sc->xn_cdata.xn_tx_chain_cnt++; 1573 if (sc->xn_cdata.xn_tx_chain_cnt > NET_TX_RING_SIZE) 1574 panic("%s: tx_chain_cnt must be <= NET_TX_RING_SIZE\n", 1575 __func__); 1576 sc->tx_mbufs[id] = m; 1577 tx->id = id; 1578 ref = gnttab_claim_grant_reference(&sc->gref_tx_head); 1579 KASSERT((short)ref >= 0, ("Negative ref")); 1580 mfn = virt_to_mfn(mtod(m, vm_offset_t)); 1581 gnttab_grant_foreign_access_ref(ref, otherend_id, 1582 mfn, GNTMAP_readonly); 1583 tx->gref = sc->grant_tx_ref[id] = ref; 1584 tx->offset = mtod(m, vm_offset_t) & (PAGE_SIZE - 1); 1585 tx->flags = 0; 1586 if (m == m_head) { 1587 /* 1588 * The first fragment has the entire packet 1589 * size, subsequent fragments have just the 1590 * fragment size. The backend works out the 1591 * true size of the first fragment by 1592 * subtracting the sizes of the other 1593 * fragments. 1594 */ 1595 tx->size = m->m_pkthdr.len; 1596 1597 /* 1598 * The first fragment contains the checksum flags 1599 * and is optionally followed by extra data for 1600 * TSO etc. 1601 */ 1602 /** 1603 * CSUM_TSO requires checksum offloading. 1604 * Some versions of FreeBSD fail to 1605 * set CSUM_TCP in the CSUM_TSO case, 1606 * so we have to test for CSUM_TSO 1607 * explicitly. 1608 */ 1609 if (m->m_pkthdr.csum_flags 1610 & (CSUM_DELAY_DATA | CSUM_TSO)) { 1611 tx->flags |= (NETTXF_csum_blank 1612 | NETTXF_data_validated); 1613 } 1614 #if __FreeBSD_version >= 700000 1615 if (m->m_pkthdr.csum_flags & CSUM_TSO) { 1616 struct netif_extra_info *gso = 1617 (struct netif_extra_info *) 1618 RING_GET_REQUEST(&sc->tx, 1619 ++sc->tx.req_prod_pvt); 1620 1621 tx->flags |= NETTXF_extra_info; 1622 1623 gso->u.gso.size = m->m_pkthdr.tso_segsz; 1624 gso->u.gso.type = 1625 XEN_NETIF_GSO_TYPE_TCPV4; 1626 gso->u.gso.pad = 0; 1627 gso->u.gso.features = 0; 1628 1629 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 1630 gso->flags = 0; 1631 } 1632 #endif 1633 } else { 1634 tx->size = m->m_len; 1635 } 1636 if (m->m_next) 1637 tx->flags |= NETTXF_more_data; 1638 1639 sc->tx.req_prod_pvt++; 1640 } 1641 BPF_MTAP(ifp, m_head); 1642 1643 sc->stats.tx_bytes += m_head->m_pkthdr.len; 1644 sc->stats.tx_packets++; 1645 1646 return (0); 1647 } 1648 1649 static void 1650 xn_start_locked(struct ifnet *ifp) 1651 { 1652 struct netfront_info *sc; 1653 struct mbuf *m_head; 1654 int notify; 1655 1656 sc = ifp->if_softc; 1657 1658 if (!netfront_carrier_ok(sc)) 1659 return; 1660 1661 /* 1662 * While we have enough transmit slots available for at least one 1663 * maximum-sized packet, pull mbufs off the queue and put them on 1664 * the transmit ring. 1665 */ 1666 while (xn_tx_slot_available(sc)) { 1667 IF_DEQUEUE(&ifp->if_snd, m_head); 1668 if (m_head == NULL) 1669 break; 1670 1671 if (xn_assemble_tx_request(sc, m_head) != 0) 1672 break; 1673 } 1674 1675 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&sc->tx, notify); 1676 if (notify) 1677 xen_intr_signal(sc->xen_intr_handle); 1678 1679 if (RING_FULL(&sc->tx)) { 1680 sc->tx_full = 1; 1681 #if 0 1682 netif_stop_queue(dev); 1683 #endif 1684 } 1685 } 1686 1687 static void 1688 xn_start(struct ifnet *ifp) 1689 { 1690 struct netfront_info *sc; 1691 sc = ifp->if_softc; 1692 XN_TX_LOCK(sc); 1693 xn_start_locked(ifp); 1694 XN_TX_UNLOCK(sc); 1695 } 1696 1697 /* equivalent of network_open() in Linux */ 1698 static void 1699 xn_ifinit_locked(struct netfront_info *sc) 1700 { 1701 struct ifnet *ifp; 1702 1703 XN_LOCK_ASSERT(sc); 1704 1705 ifp = sc->xn_ifp; 1706 1707 if (ifp->if_drv_flags & IFF_DRV_RUNNING) 1708 return; 1709 1710 xn_stop(sc); 1711 1712 network_alloc_rx_buffers(sc); 1713 sc->rx.sring->rsp_event = sc->rx.rsp_cons + 1; 1714 1715 ifp->if_drv_flags |= IFF_DRV_RUNNING; 1716 ifp->if_drv_flags &= ~IFF_DRV_OACTIVE; 1717 if_link_state_change(ifp, LINK_STATE_UP); 1718 1719 callout_reset(&sc->xn_stat_ch, hz, xn_tick, sc); 1720 } 1721 1722 static void 1723 xn_ifinit(void *xsc) 1724 { 1725 struct netfront_info *sc = xsc; 1726 1727 XN_LOCK(sc); 1728 xn_ifinit_locked(sc); 1729 XN_UNLOCK(sc); 1730 } 1731 1732 static int 1733 xn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data) 1734 { 1735 struct netfront_info *sc = ifp->if_softc; 1736 struct ifreq *ifr = (struct ifreq *) data; 1737 #ifdef INET 1738 struct ifaddr *ifa = (struct ifaddr *)data; 1739 #endif 1740 1741 int mask, error = 0; 1742 switch(cmd) { 1743 case SIOCSIFADDR: 1744 #ifdef INET 1745 XN_LOCK(sc); 1746 if (ifa->ifa_addr->sa_family == AF_INET) { 1747 ifp->if_flags |= IFF_UP; 1748 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) 1749 xn_ifinit_locked(sc); 1750 arp_ifinit(ifp, ifa); 1751 XN_UNLOCK(sc); 1752 } else { 1753 XN_UNLOCK(sc); 1754 #endif 1755 error = ether_ioctl(ifp, cmd, data); 1756 #ifdef INET 1757 } 1758 #endif 1759 break; 1760 case SIOCSIFMTU: 1761 /* XXX can we alter the MTU on a VN ?*/ 1762 #ifdef notyet 1763 if (ifr->ifr_mtu > XN_JUMBO_MTU) 1764 error = EINVAL; 1765 else 1766 #endif 1767 { 1768 ifp->if_mtu = ifr->ifr_mtu; 1769 ifp->if_drv_flags &= ~IFF_DRV_RUNNING; 1770 xn_ifinit(sc); 1771 } 1772 break; 1773 case SIOCSIFFLAGS: 1774 XN_LOCK(sc); 1775 if (ifp->if_flags & IFF_UP) { 1776 /* 1777 * If only the state of the PROMISC flag changed, 1778 * then just use the 'set promisc mode' command 1779 * instead of reinitializing the entire NIC. Doing 1780 * a full re-init means reloading the firmware and 1781 * waiting for it to start up, which may take a 1782 * second or two. 1783 */ 1784 #ifdef notyet 1785 /* No promiscuous mode with Xen */ 1786 if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1787 ifp->if_flags & IFF_PROMISC && 1788 !(sc->xn_if_flags & IFF_PROMISC)) { 1789 XN_SETBIT(sc, XN_RX_MODE, 1790 XN_RXMODE_RX_PROMISC); 1791 } else if (ifp->if_drv_flags & IFF_DRV_RUNNING && 1792 !(ifp->if_flags & IFF_PROMISC) && 1793 sc->xn_if_flags & IFF_PROMISC) { 1794 XN_CLRBIT(sc, XN_RX_MODE, 1795 XN_RXMODE_RX_PROMISC); 1796 } else 1797 #endif 1798 xn_ifinit_locked(sc); 1799 } else { 1800 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1801 xn_stop(sc); 1802 } 1803 } 1804 sc->xn_if_flags = ifp->if_flags; 1805 XN_UNLOCK(sc); 1806 error = 0; 1807 break; 1808 case SIOCSIFCAP: 1809 mask = ifr->ifr_reqcap ^ ifp->if_capenable; 1810 if (mask & IFCAP_TXCSUM) { 1811 if (IFCAP_TXCSUM & ifp->if_capenable) { 1812 ifp->if_capenable &= ~(IFCAP_TXCSUM|IFCAP_TSO4); 1813 ifp->if_hwassist &= ~(CSUM_TCP | CSUM_UDP 1814 | CSUM_IP | CSUM_TSO); 1815 } else { 1816 ifp->if_capenable |= IFCAP_TXCSUM; 1817 ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP 1818 | CSUM_IP); 1819 } 1820 } 1821 if (mask & IFCAP_RXCSUM) { 1822 ifp->if_capenable ^= IFCAP_RXCSUM; 1823 } 1824 #if __FreeBSD_version >= 700000 1825 if (mask & IFCAP_TSO4) { 1826 if (IFCAP_TSO4 & ifp->if_capenable) { 1827 ifp->if_capenable &= ~IFCAP_TSO4; 1828 ifp->if_hwassist &= ~CSUM_TSO; 1829 } else if (IFCAP_TXCSUM & ifp->if_capenable) { 1830 ifp->if_capenable |= IFCAP_TSO4; 1831 ifp->if_hwassist |= CSUM_TSO; 1832 } else { 1833 IPRINTK("Xen requires tx checksum offload" 1834 " be enabled to use TSO\n"); 1835 error = EINVAL; 1836 } 1837 } 1838 if (mask & IFCAP_LRO) { 1839 ifp->if_capenable ^= IFCAP_LRO; 1840 1841 } 1842 #endif 1843 error = 0; 1844 break; 1845 case SIOCADDMULTI: 1846 case SIOCDELMULTI: 1847 #ifdef notyet 1848 if (ifp->if_drv_flags & IFF_DRV_RUNNING) { 1849 XN_LOCK(sc); 1850 xn_setmulti(sc); 1851 XN_UNLOCK(sc); 1852 error = 0; 1853 } 1854 #endif 1855 /* FALLTHROUGH */ 1856 case SIOCSIFMEDIA: 1857 case SIOCGIFMEDIA: 1858 error = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd); 1859 break; 1860 default: 1861 error = ether_ioctl(ifp, cmd, data); 1862 } 1863 1864 return (error); 1865 } 1866 1867 static void 1868 xn_stop(struct netfront_info *sc) 1869 { 1870 struct ifnet *ifp; 1871 1872 XN_LOCK_ASSERT(sc); 1873 1874 ifp = sc->xn_ifp; 1875 1876 callout_stop(&sc->xn_stat_ch); 1877 1878 xn_free_rx_ring(sc); 1879 xn_free_tx_ring(sc); 1880 1881 ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE); 1882 if_link_state_change(ifp, LINK_STATE_DOWN); 1883 } 1884 1885 /* START of Xenolinux helper functions adapted to FreeBSD */ 1886 int 1887 network_connect(struct netfront_info *np) 1888 { 1889 int i, requeue_idx, error; 1890 grant_ref_t ref; 1891 netif_rx_request_t *req; 1892 u_int feature_rx_copy, feature_rx_flip; 1893 1894 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1895 "feature-rx-copy", NULL, "%u", &feature_rx_copy); 1896 if (error) 1897 feature_rx_copy = 0; 1898 error = xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1899 "feature-rx-flip", NULL, "%u", &feature_rx_flip); 1900 if (error) 1901 feature_rx_flip = 1; 1902 1903 /* 1904 * Copy packets on receive path if: 1905 * (a) This was requested by user, and the backend supports it; or 1906 * (b) Flipping was requested, but this is unsupported by the backend. 1907 */ 1908 np->copying_receiver = ((MODPARM_rx_copy && feature_rx_copy) || 1909 (MODPARM_rx_flip && !feature_rx_flip)); 1910 1911 /* Recovery procedure: */ 1912 error = talk_to_backend(np->xbdev, np); 1913 if (error) 1914 return (error); 1915 1916 /* Step 1: Reinitialise variables. */ 1917 xn_query_features(np); 1918 xn_configure_features(np); 1919 netif_release_tx_bufs(np); 1920 1921 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ 1922 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { 1923 struct mbuf *m; 1924 u_long pfn; 1925 1926 if (np->rx_mbufs[i] == NULL) 1927 continue; 1928 1929 m = np->rx_mbufs[requeue_idx] = xennet_get_rx_mbuf(np, i); 1930 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); 1931 1932 req = RING_GET_REQUEST(&np->rx, requeue_idx); 1933 pfn = vtophys(mtod(m, vm_offset_t)) >> PAGE_SHIFT; 1934 1935 if (!np->copying_receiver) { 1936 gnttab_grant_foreign_transfer_ref(ref, 1937 xenbus_get_otherend_id(np->xbdev), 1938 pfn); 1939 } else { 1940 gnttab_grant_foreign_access_ref(ref, 1941 xenbus_get_otherend_id(np->xbdev), 1942 PFNTOMFN(pfn), 0); 1943 } 1944 req->gref = ref; 1945 req->id = requeue_idx; 1946 1947 requeue_idx++; 1948 } 1949 1950 np->rx.req_prod_pvt = requeue_idx; 1951 1952 /* Step 3: All public and private state should now be sane. Get 1953 * ready to start sending and receiving packets and give the driver 1954 * domain a kick because we've probably just requeued some 1955 * packets. 1956 */ 1957 netfront_carrier_on(np); 1958 xen_intr_signal(np->xen_intr_handle); 1959 XN_TX_LOCK(np); 1960 xn_txeof(np); 1961 XN_TX_UNLOCK(np); 1962 network_alloc_rx_buffers(np); 1963 1964 return (0); 1965 } 1966 1967 static void 1968 xn_query_features(struct netfront_info *np) 1969 { 1970 int val; 1971 1972 device_printf(np->xbdev, "backend features:"); 1973 1974 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1975 "feature-sg", NULL, "%d", &val) < 0) 1976 val = 0; 1977 1978 np->maxfrags = 1; 1979 if (val) { 1980 np->maxfrags = MAX_TX_REQ_FRAGS; 1981 printf(" feature-sg"); 1982 } 1983 1984 if (xs_scanf(XST_NIL, xenbus_get_otherend_path(np->xbdev), 1985 "feature-gso-tcpv4", NULL, "%d", &val) < 0) 1986 val = 0; 1987 1988 np->xn_ifp->if_capabilities &= ~(IFCAP_TSO4|IFCAP_LRO); 1989 if (val) { 1990 np->xn_ifp->if_capabilities |= IFCAP_TSO4|IFCAP_LRO; 1991 printf(" feature-gso-tcp4"); 1992 } 1993 1994 printf("\n"); 1995 } 1996 1997 static int 1998 xn_configure_features(struct netfront_info *np) 1999 { 2000 int err; 2001 2002 err = 0; 2003 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 2004 if ((np->xn_ifp->if_capenable & IFCAP_LRO) != 0) 2005 tcp_lro_free(&np->xn_lro); 2006 #endif 2007 np->xn_ifp->if_capenable = 2008 np->xn_ifp->if_capabilities & ~(IFCAP_LRO|IFCAP_TSO4); 2009 np->xn_ifp->if_hwassist &= ~CSUM_TSO; 2010 #if __FreeBSD_version >= 700000 && (defined(INET) || defined(INET6)) 2011 if (xn_enable_lro && (np->xn_ifp->if_capabilities & IFCAP_LRO) != 0) { 2012 err = tcp_lro_init(&np->xn_lro); 2013 if (err) { 2014 device_printf(np->xbdev, "LRO initialization failed\n"); 2015 } else { 2016 np->xn_lro.ifp = np->xn_ifp; 2017 np->xn_ifp->if_capenable |= IFCAP_LRO; 2018 } 2019 } 2020 if ((np->xn_ifp->if_capabilities & IFCAP_TSO4) != 0) { 2021 np->xn_ifp->if_capenable |= IFCAP_TSO4; 2022 np->xn_ifp->if_hwassist |= CSUM_TSO; 2023 } 2024 #endif 2025 return (err); 2026 } 2027 2028 /** 2029 * Create a network device. 2030 * @param dev Newbus device representing this virtual NIC. 2031 */ 2032 int 2033 create_netdev(device_t dev) 2034 { 2035 int i; 2036 struct netfront_info *np; 2037 int err; 2038 struct ifnet *ifp; 2039 2040 np = device_get_softc(dev); 2041 2042 np->xbdev = dev; 2043 2044 XN_LOCK_INIT(np, xennetif); 2045 2046 ifmedia_init(&np->sc_media, 0, xn_ifmedia_upd, xn_ifmedia_sts); 2047 ifmedia_add(&np->sc_media, IFM_ETHER|IFM_MANUAL, 0, NULL); 2048 ifmedia_set(&np->sc_media, IFM_ETHER|IFM_MANUAL); 2049 2050 np->rx_target = RX_MIN_TARGET; 2051 np->rx_min_target = RX_MIN_TARGET; 2052 np->rx_max_target = RX_MAX_TARGET; 2053 2054 /* Initialise {tx,rx}_skbs to be a free chain containing every entry. */ 2055 for (i = 0; i <= NET_TX_RING_SIZE; i++) { 2056 np->tx_mbufs[i] = (void *) ((u_long) i+1); 2057 np->grant_tx_ref[i] = GRANT_REF_INVALID; 2058 } 2059 np->tx_mbufs[NET_TX_RING_SIZE] = (void *)0; 2060 2061 for (i = 0; i <= NET_RX_RING_SIZE; i++) { 2062 2063 np->rx_mbufs[i] = NULL; 2064 np->grant_rx_ref[i] = GRANT_REF_INVALID; 2065 } 2066 /* A grant for every tx ring slot */ 2067 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, 2068 &np->gref_tx_head) != 0) { 2069 IPRINTK("#### netfront can't alloc tx grant refs\n"); 2070 err = ENOMEM; 2071 goto exit; 2072 } 2073 /* A grant for every rx ring slot */ 2074 if (gnttab_alloc_grant_references(RX_MAX_TARGET, 2075 &np->gref_rx_head) != 0) { 2076 WPRINTK("#### netfront can't alloc rx grant refs\n"); 2077 gnttab_free_grant_references(np->gref_tx_head); 2078 err = ENOMEM; 2079 goto exit; 2080 } 2081 2082 err = xen_net_read_mac(dev, np->mac); 2083 if (err) 2084 goto out; 2085 2086 /* Set up ifnet structure */ 2087 ifp = np->xn_ifp = if_alloc(IFT_ETHER); 2088 ifp->if_softc = np; 2089 if_initname(ifp, "xn", device_get_unit(dev)); 2090 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 2091 ifp->if_ioctl = xn_ioctl; 2092 ifp->if_output = ether_output; 2093 ifp->if_start = xn_start; 2094 #ifdef notyet 2095 ifp->if_watchdog = xn_watchdog; 2096 #endif 2097 ifp->if_init = xn_ifinit; 2098 ifp->if_snd.ifq_maxlen = NET_TX_RING_SIZE - 1; 2099 2100 ifp->if_hwassist = XN_CSUM_FEATURES; 2101 ifp->if_capabilities = IFCAP_HWCSUM; 2102 ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); 2103 ifp->if_hw_tsomaxsegcount = MAX_TX_REQ_FRAGS; 2104 ifp->if_hw_tsomaxsegsize = PAGE_SIZE; 2105 2106 ether_ifattach(ifp, np->mac); 2107 callout_init(&np->xn_stat_ch, CALLOUT_MPSAFE); 2108 netfront_carrier_off(np); 2109 2110 return (0); 2111 2112 exit: 2113 gnttab_free_grant_references(np->gref_tx_head); 2114 out: 2115 return (err); 2116 } 2117 2118 /** 2119 * Handle the change of state of the backend to Closing. We must delete our 2120 * device-layer structures now, to ensure that writes are flushed through to 2121 * the backend. Once is this done, we can switch to Closed in 2122 * acknowledgement. 2123 */ 2124 #if 0 2125 static void 2126 netfront_closing(device_t dev) 2127 { 2128 #if 0 2129 struct netfront_info *info = dev->dev_driver_data; 2130 2131 DPRINTK("netfront_closing: %s removed\n", dev->nodename); 2132 2133 close_netdev(info); 2134 #endif 2135 xenbus_switch_state(dev, XenbusStateClosed); 2136 } 2137 #endif 2138 2139 static int 2140 netfront_detach(device_t dev) 2141 { 2142 struct netfront_info *info = device_get_softc(dev); 2143 2144 DPRINTK("%s\n", xenbus_get_node(dev)); 2145 2146 netif_free(info); 2147 2148 return 0; 2149 } 2150 2151 static void 2152 netif_free(struct netfront_info *info) 2153 { 2154 XN_LOCK(info); 2155 xn_stop(info); 2156 XN_UNLOCK(info); 2157 callout_drain(&info->xn_stat_ch); 2158 netif_disconnect_backend(info); 2159 if (info->xn_ifp != NULL) { 2160 ether_ifdetach(info->xn_ifp); 2161 if_free(info->xn_ifp); 2162 info->xn_ifp = NULL; 2163 } 2164 ifmedia_removeall(&info->sc_media); 2165 } 2166 2167 static void 2168 netif_disconnect_backend(struct netfront_info *info) 2169 { 2170 XN_RX_LOCK(info); 2171 XN_TX_LOCK(info); 2172 netfront_carrier_off(info); 2173 XN_TX_UNLOCK(info); 2174 XN_RX_UNLOCK(info); 2175 2176 free_ring(&info->tx_ring_ref, &info->tx.sring); 2177 free_ring(&info->rx_ring_ref, &info->rx.sring); 2178 2179 xen_intr_unbind(&info->xen_intr_handle); 2180 } 2181 2182 static void 2183 free_ring(int *ref, void *ring_ptr_ref) 2184 { 2185 void **ring_ptr_ptr = ring_ptr_ref; 2186 2187 if (*ref != GRANT_REF_INVALID) { 2188 /* This API frees the associated storage. */ 2189 gnttab_end_foreign_access(*ref, *ring_ptr_ptr); 2190 *ref = GRANT_REF_INVALID; 2191 } 2192 *ring_ptr_ptr = NULL; 2193 } 2194 2195 static int 2196 xn_ifmedia_upd(struct ifnet *ifp) 2197 { 2198 return (0); 2199 } 2200 2201 static void 2202 xn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr) 2203 { 2204 ifmr->ifm_status = IFM_AVALID|IFM_ACTIVE; 2205 ifmr->ifm_active = IFM_ETHER|IFM_MANUAL; 2206 } 2207 2208 /* ** Driver registration ** */ 2209 static device_method_t netfront_methods[] = { 2210 /* Device interface */ 2211 DEVMETHOD(device_probe, netfront_probe), 2212 DEVMETHOD(device_attach, netfront_attach), 2213 DEVMETHOD(device_detach, netfront_detach), 2214 DEVMETHOD(device_shutdown, bus_generic_shutdown), 2215 DEVMETHOD(device_suspend, netfront_suspend), 2216 DEVMETHOD(device_resume, netfront_resume), 2217 2218 /* Xenbus interface */ 2219 DEVMETHOD(xenbus_otherend_changed, netfront_backend_changed), 2220 2221 DEVMETHOD_END 2222 }; 2223 2224 static driver_t netfront_driver = { 2225 "xn", 2226 netfront_methods, 2227 sizeof(struct netfront_info), 2228 }; 2229 devclass_t netfront_devclass; 2230 2231 DRIVER_MODULE(xe, xenbusb_front, netfront_driver, netfront_devclass, NULL, 2232 NULL); 2233