1/* 2 * Virtual network driver for conversing with remote driver backends. 3 * 4 * Copyright (c) 2002-2005, K A Fraser 5 * Copyright (c) 2005, XenSource Ltd 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License version 2 9 * as published by the Free Software Foundation; or, when distributed 10 * separately from the Linux kernel or incorporated into other 11 * software packages, subject to the following license: 12 * 13 * Permission is hereby granted, free of charge, to any person obtaining a copy 14 * of this source file (the "Software"), to deal in the Software without 15 * restriction, including without limitation the rights to use, copy, modify, 16 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 17 * and to permit persons to whom the Software is furnished to do so, subject to 18 * the following conditions: 19 * 20 * The above copyright notice and this permission notice shall be included in 21 * all copies or substantial portions of the Software. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 29 * IN THE SOFTWARE. 30 */ 31 32#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 33 34#include <linux/module.h> 35#include <linux/kernel.h> 36#include <linux/netdevice.h> 37#include <linux/etherdevice.h> 38#include <linux/skbuff.h> 39#include <linux/ethtool.h> 40#include <linux/if_ether.h> 41#include <net/tcp.h> 42#include <linux/udp.h> 43#include <linux/moduleparam.h> 44#include <linux/mm.h> 45#include <linux/slab.h> 46#include <net/ip.h> 47 48#include <asm/xen/page.h> 49#include <xen/xen.h> 50#include <xen/xenbus.h> 51#include <xen/events.h> 52#include <xen/page.h> 53#include <xen/platform_pci.h> 54#include <xen/grant_table.h> 55 56#include <xen/interface/io/netif.h> 57#include <xen/interface/memory.h> 58#include <xen/interface/grant_table.h> 59 60/* Module parameters */ 61static unsigned int xennet_max_queues; 62module_param_named(max_queues, xennet_max_queues, uint, 0644); 63MODULE_PARM_DESC(max_queues, 64 "Maximum number of queues per virtual interface"); 65 66static const struct ethtool_ops xennet_ethtool_ops; 67 68struct netfront_cb { 69 int pull_to; 70}; 71 72#define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) 73 74#define RX_COPY_THRESHOLD 256 75 76#define GRANT_INVALID_REF 0 77 78#define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) 79#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE) 80#define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256) 81 82/* Queue name is interface name with "-qNNN" appended */ 83#define QUEUE_NAME_SIZE (IFNAMSIZ + 6) 84 85/* IRQ name is queue name with "-tx" or "-rx" appended */ 86#define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3) 87 88struct netfront_stats { 89 u64 rx_packets; 90 u64 tx_packets; 91 u64 rx_bytes; 92 u64 tx_bytes; 93 struct u64_stats_sync syncp; 94}; 95 96struct netfront_info; 97 98struct netfront_queue { 99 unsigned int id; /* Queue ID, 0-based */ 100 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ 101 struct netfront_info *info; 102 103 struct napi_struct napi; 104 105 /* Split event channels support, tx_* == rx_* when using 106 * single event channel. 107 */ 108 unsigned int tx_evtchn, rx_evtchn; 109 unsigned int tx_irq, rx_irq; 110 /* Only used when split event channels support is enabled */ 111 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ 112 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ 113 114 spinlock_t tx_lock; 115 struct xen_netif_tx_front_ring tx; 116 int tx_ring_ref; 117 118 /* 119 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries 120 * are linked from tx_skb_freelist through skb_entry.link. 121 * 122 * NB. Freelist index entries are always going to be less than 123 * PAGE_OFFSET, whereas pointers to skbs will always be equal or 124 * greater than PAGE_OFFSET: we use this property to distinguish 125 * them. 126 */ 127 union skb_entry { 128 struct sk_buff *skb; 129 unsigned long link; 130 } tx_skbs[NET_TX_RING_SIZE]; 131 grant_ref_t gref_tx_head; 132 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; 133 struct page *grant_tx_page[NET_TX_RING_SIZE]; 134 unsigned tx_skb_freelist; 135 136 spinlock_t rx_lock ____cacheline_aligned_in_smp; 137 struct xen_netif_rx_front_ring rx; 138 int rx_ring_ref; 139 140 /* Receive-ring batched refills. */ 141#define RX_MIN_TARGET 8 142#define RX_DFL_MIN_TARGET 64 143#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) 144 unsigned rx_min_target, rx_max_target, rx_target; 145 struct sk_buff_head rx_batch; 146 147 struct timer_list rx_refill_timer; 148 149 struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; 150 grant_ref_t gref_rx_head; 151 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; 152 153 unsigned long rx_pfn_array[NET_RX_RING_SIZE]; 154 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; 155 struct mmu_update rx_mmu[NET_RX_RING_SIZE]; 156}; 157 158struct netfront_info { 159 struct list_head list; 160 struct net_device *netdev; 161 162 struct xenbus_device *xbdev; 163 164 /* Multi-queue support */ 165 struct netfront_queue *queues; 166 167 /* Statistics */ 168 struct netfront_stats __percpu *stats; 169 170 atomic_t rx_gso_checksum_fixup; 171}; 172 173struct netfront_rx_info { 174 struct xen_netif_rx_response rx; 175 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; 176}; 177 178static void skb_entry_set_link(union skb_entry *list, unsigned short id) 179{ 180 list->link = id; 181} 182 183static int skb_entry_is_link(const union skb_entry *list) 184{ 185 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); 186 return (unsigned long)list->skb < PAGE_OFFSET; 187} 188 189/* 190 * Access macros for acquiring freeing slots in tx_skbs[]. 191 */ 192 193static void add_id_to_freelist(unsigned *head, union skb_entry *list, 194 unsigned short id) 195{ 196 skb_entry_set_link(&list[id], *head); 197 *head = id; 198} 199 200static unsigned short get_id_from_freelist(unsigned *head, 201 union skb_entry *list) 202{ 203 unsigned int id = *head; 204 *head = list[id].link; 205 return id; 206} 207 208static int xennet_rxidx(RING_IDX idx) 209{ 210 return idx & (NET_RX_RING_SIZE - 1); 211} 212 213static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, 214 RING_IDX ri) 215{ 216 int i = xennet_rxidx(ri); 217 struct sk_buff *skb = queue->rx_skbs[i]; 218 queue->rx_skbs[i] = NULL; 219 return skb; 220} 221 222static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, 223 RING_IDX ri) 224{ 225 int i = xennet_rxidx(ri); 226 grant_ref_t ref = queue->grant_rx_ref[i]; 227 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 228 return ref; 229} 230 231#ifdef CONFIG_SYSFS 232static int xennet_sysfs_addif(struct net_device *netdev); 233static void xennet_sysfs_delif(struct net_device *netdev); 234#else /* !CONFIG_SYSFS */ 235#define xennet_sysfs_addif(dev) (0) 236#define xennet_sysfs_delif(dev) do { } while (0) 237#endif 238 239static bool xennet_can_sg(struct net_device *dev) 240{ 241 return dev->features & NETIF_F_SG; 242} 243 244 245static void rx_refill_timeout(unsigned long data) 246{ 247 struct netfront_queue *queue = (struct netfront_queue *)data; 248 napi_schedule(&queue->napi); 249} 250 251static int netfront_tx_slot_available(struct netfront_queue *queue) 252{ 253 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < 254 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2); 255} 256 257static void xennet_maybe_wake_tx(struct netfront_queue *queue) 258{ 259 struct net_device *dev = queue->info->netdev; 260 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); 261 262 if (unlikely(netif_tx_queue_stopped(dev_queue)) && 263 netfront_tx_slot_available(queue) && 264 likely(netif_running(dev))) 265 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); 266} 267 268static void xennet_alloc_rx_buffers(struct netfront_queue *queue) 269{ 270 unsigned short id; 271 struct sk_buff *skb; 272 struct page *page; 273 int i, batch_target, notify; 274 RING_IDX req_prod = queue->rx.req_prod_pvt; 275 grant_ref_t ref; 276 unsigned long pfn; 277 void *vaddr; 278 struct xen_netif_rx_request *req; 279 280 if (unlikely(!netif_carrier_ok(queue->info->netdev))) 281 return; 282 283 /* 284 * Allocate skbuffs greedily, even though we batch updates to the 285 * receive ring. This creates a less bursty demand on the memory 286 * allocator, so should reduce the chance of failed allocation requests 287 * both for ourself and for other kernel subsystems. 288 */ 289 batch_target = queue->rx_target - (req_prod - queue->rx.rsp_cons); 290 for (i = skb_queue_len(&queue->rx_batch); i < batch_target; i++) { 291 skb = __netdev_alloc_skb(queue->info->netdev, 292 RX_COPY_THRESHOLD + NET_IP_ALIGN, 293 GFP_ATOMIC | __GFP_NOWARN); 294 if (unlikely(!skb)) 295 goto no_skb; 296 297 /* Align ip header to a 16 bytes boundary */ 298 skb_reserve(skb, NET_IP_ALIGN); 299 300 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); 301 if (!page) { 302 kfree_skb(skb); 303no_skb: 304 /* Could not allocate any skbuffs. Try again later. */ 305 mod_timer(&queue->rx_refill_timer, 306 jiffies + (HZ/10)); 307 308 /* Any skbuffs queued for refill? Force them out. */ 309 if (i != 0) 310 goto refill; 311 break; 312 } 313 314 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); 315 __skb_queue_tail(&queue->rx_batch, skb); 316 } 317 318 /* Is the batch large enough to be worthwhile? */ 319 if (i < (queue->rx_target/2)) { 320 if (req_prod > queue->rx.sring->req_prod) 321 goto push; 322 return; 323 } 324 325 /* Adjust our fill target if we risked running out of buffers. */ 326 if (((req_prod - queue->rx.sring->rsp_prod) < (queue->rx_target / 4)) && 327 ((queue->rx_target *= 2) > queue->rx_max_target)) 328 queue->rx_target = queue->rx_max_target; 329 330 refill: 331 for (i = 0; ; i++) { 332 skb = __skb_dequeue(&queue->rx_batch); 333 if (skb == NULL) 334 break; 335 336 skb->dev = queue->info->netdev; 337 338 id = xennet_rxidx(req_prod + i); 339 340 BUG_ON(queue->rx_skbs[id]); 341 queue->rx_skbs[id] = skb; 342 343 ref = gnttab_claim_grant_reference(&queue->gref_rx_head); 344 BUG_ON((signed short)ref < 0); 345 queue->grant_rx_ref[id] = ref; 346 347 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0])); 348 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0])); 349 350 req = RING_GET_REQUEST(&queue->rx, req_prod + i); 351 gnttab_grant_foreign_access_ref(ref, 352 queue->info->xbdev->otherend_id, 353 pfn_to_mfn(pfn), 354 0); 355 356 req->id = id; 357 req->gref = ref; 358 } 359 360 wmb(); /* barrier so backend seens requests */ 361 362 /* Above is a suitable barrier to ensure backend will see requests. */ 363 queue->rx.req_prod_pvt = req_prod + i; 364 push: 365 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); 366 if (notify) 367 notify_remote_via_irq(queue->rx_irq); 368} 369 370static int xennet_open(struct net_device *dev) 371{ 372 struct netfront_info *np = netdev_priv(dev); 373 unsigned int num_queues = dev->real_num_tx_queues; 374 unsigned int i = 0; 375 struct netfront_queue *queue = NULL; 376 377 for (i = 0; i < num_queues; ++i) { 378 queue = &np->queues[i]; 379 napi_enable(&queue->napi); 380 381 spin_lock_bh(&queue->rx_lock); 382 if (netif_carrier_ok(dev)) { 383 xennet_alloc_rx_buffers(queue); 384 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; 385 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) 386 napi_schedule(&queue->napi); 387 } 388 spin_unlock_bh(&queue->rx_lock); 389 } 390 391 netif_tx_start_all_queues(dev); 392 393 return 0; 394} 395 396static void xennet_tx_buf_gc(struct netfront_queue *queue) 397{ 398 RING_IDX cons, prod; 399 unsigned short id; 400 struct sk_buff *skb; 401 402 BUG_ON(!netif_carrier_ok(queue->info->netdev)); 403 404 do { 405 prod = queue->tx.sring->rsp_prod; 406 rmb(); /* Ensure we see responses up to 'rp'. */ 407 408 for (cons = queue->tx.rsp_cons; cons != prod; cons++) { 409 struct xen_netif_tx_response *txrsp; 410 411 txrsp = RING_GET_RESPONSE(&queue->tx, cons); 412 if (txrsp->status == XEN_NETIF_RSP_NULL) 413 continue; 414 415 id = txrsp->id; 416 skb = queue->tx_skbs[id].skb; 417 if (unlikely(gnttab_query_foreign_access( 418 queue->grant_tx_ref[id]) != 0)) { 419 pr_alert("%s: warning -- grant still in use by backend domain\n", 420 __func__); 421 BUG(); 422 } 423 gnttab_end_foreign_access_ref( 424 queue->grant_tx_ref[id], GNTMAP_readonly); 425 gnttab_release_grant_reference( 426 &queue->gref_tx_head, queue->grant_tx_ref[id]); 427 queue->grant_tx_ref[id] = GRANT_INVALID_REF; 428 queue->grant_tx_page[id] = NULL; 429 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); 430 dev_kfree_skb_irq(skb); 431 } 432 433 queue->tx.rsp_cons = prod; 434 435 /* 436 * Set a new event, then check for race with update of tx_cons. 437 * Note that it is essential to schedule a callback, no matter 438 * how few buffers are pending. Even if there is space in the 439 * transmit ring, higher layers may be blocked because too much 440 * data is outstanding: in such cases notification from Xen is 441 * likely to be the only kick that we'll get. 442 */ 443 queue->tx.sring->rsp_event = 444 prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1; 445 mb(); /* update shared area */ 446 } while ((cons == prod) && (prod != queue->tx.sring->rsp_prod)); 447 448 xennet_maybe_wake_tx(queue); 449} 450 451static void xennet_make_frags(struct sk_buff *skb, struct netfront_queue *queue, 452 struct xen_netif_tx_request *tx) 453{ 454 char *data = skb->data; 455 unsigned long mfn; 456 RING_IDX prod = queue->tx.req_prod_pvt; 457 int frags = skb_shinfo(skb)->nr_frags; 458 unsigned int offset = offset_in_page(data); 459 unsigned int len = skb_headlen(skb); 460 unsigned int id; 461 grant_ref_t ref; 462 int i; 463 464 /* While the header overlaps a page boundary (including being 465 larger than a page), split it it into page-sized chunks. */ 466 while (len > PAGE_SIZE - offset) { 467 tx->size = PAGE_SIZE - offset; 468 tx->flags |= XEN_NETTXF_more_data; 469 len -= tx->size; 470 data += tx->size; 471 offset = 0; 472 473 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); 474 queue->tx_skbs[id].skb = skb_get(skb); 475 tx = RING_GET_REQUEST(&queue->tx, prod++); 476 tx->id = id; 477 ref = gnttab_claim_grant_reference(&queue->gref_tx_head); 478 BUG_ON((signed short)ref < 0); 479 480 mfn = virt_to_mfn(data); 481 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, 482 mfn, GNTMAP_readonly); 483 484 queue->grant_tx_page[id] = virt_to_page(data); 485 tx->gref = queue->grant_tx_ref[id] = ref; 486 tx->offset = offset; 487 tx->size = len; 488 tx->flags = 0; 489 } 490 491 /* Grant backend access to each skb fragment page. */ 492 for (i = 0; i < frags; i++) { 493 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 494 struct page *page = skb_frag_page(frag); 495 496 len = skb_frag_size(frag); 497 offset = frag->page_offset; 498 499 /* Skip unused frames from start of page */ 500 page += offset >> PAGE_SHIFT; 501 offset &= ~PAGE_MASK; 502 503 while (len > 0) { 504 unsigned long bytes; 505 506 bytes = PAGE_SIZE - offset; 507 if (bytes > len) 508 bytes = len; 509 510 tx->flags |= XEN_NETTXF_more_data; 511 512 id = get_id_from_freelist(&queue->tx_skb_freelist, 513 queue->tx_skbs); 514 queue->tx_skbs[id].skb = skb_get(skb); 515 tx = RING_GET_REQUEST(&queue->tx, prod++); 516 tx->id = id; 517 ref = gnttab_claim_grant_reference(&queue->gref_tx_head); 518 BUG_ON((signed short)ref < 0); 519 520 mfn = pfn_to_mfn(page_to_pfn(page)); 521 gnttab_grant_foreign_access_ref(ref, 522 queue->info->xbdev->otherend_id, 523 mfn, GNTMAP_readonly); 524 525 queue->grant_tx_page[id] = page; 526 tx->gref = queue->grant_tx_ref[id] = ref; 527 tx->offset = offset; 528 tx->size = bytes; 529 tx->flags = 0; 530 531 offset += bytes; 532 len -= bytes; 533 534 /* Next frame */ 535 if (offset == PAGE_SIZE && len) { 536 BUG_ON(!PageCompound(page)); 537 page++; 538 offset = 0; 539 } 540 } 541 } 542 543 queue->tx.req_prod_pvt = prod; 544} 545 546/* 547 * Count how many ring slots are required to send the frags of this 548 * skb. Each frag might be a compound page. 549 */ 550static int xennet_count_skb_frag_slots(struct sk_buff *skb) 551{ 552 int i, frags = skb_shinfo(skb)->nr_frags; 553 int pages = 0; 554 555 for (i = 0; i < frags; i++) { 556 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 557 unsigned long size = skb_frag_size(frag); 558 unsigned long offset = frag->page_offset; 559 560 /* Skip unused frames from start of page */ 561 offset &= ~PAGE_MASK; 562 563 pages += PFN_UP(offset + size); 564 } 565 566 return pages; 567} 568 569static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, 570 void *accel_priv, select_queue_fallback_t fallback) 571{ 572 unsigned int num_queues = dev->real_num_tx_queues; 573 u32 hash; 574 u16 queue_idx; 575 576 /* First, check if there is only one queue */ 577 if (num_queues == 1) { 578 queue_idx = 0; 579 } else { 580 hash = skb_get_hash(skb); 581 queue_idx = hash % num_queues; 582 } 583 584 return queue_idx; 585} 586 587static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) 588{ 589 unsigned short id; 590 struct netfront_info *np = netdev_priv(dev); 591 struct netfront_stats *stats = this_cpu_ptr(np->stats); 592 struct xen_netif_tx_request *tx; 593 char *data = skb->data; 594 RING_IDX i; 595 grant_ref_t ref; 596 unsigned long mfn; 597 int notify; 598 int slots; 599 unsigned int offset = offset_in_page(data); 600 unsigned int len = skb_headlen(skb); 601 unsigned long flags; 602 struct netfront_queue *queue = NULL; 603 unsigned int num_queues = dev->real_num_tx_queues; 604 u16 queue_index; 605 606 /* Drop the packet if no queues are set up */ 607 if (num_queues < 1) 608 goto drop; 609 /* Determine which queue to transmit this SKB on */ 610 queue_index = skb_get_queue_mapping(skb); 611 queue = &np->queues[queue_index]; 612 613 /* If skb->len is too big for wire format, drop skb and alert 614 * user about misconfiguration. 615 */ 616 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { 617 net_alert_ratelimited( 618 "xennet: skb->len = %u, too big for wire format\n", 619 skb->len); 620 goto drop; 621 } 622 623 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) + 624 xennet_count_skb_frag_slots(skb); 625 if (unlikely(slots > MAX_SKB_FRAGS + 1)) { 626 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", 627 slots, skb->len); 628 if (skb_linearize(skb)) 629 goto drop; 630 } 631 632 spin_lock_irqsave(&queue->tx_lock, flags); 633 634 if (unlikely(!netif_carrier_ok(dev) || 635 (slots > 1 && !xennet_can_sg(dev)) || 636 netif_needs_gso(dev, skb, netif_skb_features(skb)))) { 637 spin_unlock_irqrestore(&queue->tx_lock, flags); 638 goto drop; 639 } 640 641 i = queue->tx.req_prod_pvt; 642 643 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); 644 queue->tx_skbs[id].skb = skb; 645 646 tx = RING_GET_REQUEST(&queue->tx, i); 647 648 tx->id = id; 649 ref = gnttab_claim_grant_reference(&queue->gref_tx_head); 650 BUG_ON((signed short)ref < 0); 651 mfn = virt_to_mfn(data); 652 gnttab_grant_foreign_access_ref( 653 ref, queue->info->xbdev->otherend_id, mfn, GNTMAP_readonly); 654 queue->grant_tx_page[id] = virt_to_page(data); 655 tx->gref = queue->grant_tx_ref[id] = ref; 656 tx->offset = offset; 657 tx->size = len; 658 659 tx->flags = 0; 660 if (skb->ip_summed == CHECKSUM_PARTIAL) 661 /* local packet? */ 662 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; 663 else if (skb->ip_summed == CHECKSUM_UNNECESSARY) 664 /* remote but checksummed. */ 665 tx->flags |= XEN_NETTXF_data_validated; 666 667 if (skb_shinfo(skb)->gso_size) { 668 struct xen_netif_extra_info *gso; 669 670 gso = (struct xen_netif_extra_info *) 671 RING_GET_REQUEST(&queue->tx, ++i); 672 673 tx->flags |= XEN_NETTXF_extra_info; 674 675 gso->u.gso.size = skb_shinfo(skb)->gso_size; 676 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? 677 XEN_NETIF_GSO_TYPE_TCPV6 : 678 XEN_NETIF_GSO_TYPE_TCPV4; 679 gso->u.gso.pad = 0; 680 gso->u.gso.features = 0; 681 682 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 683 gso->flags = 0; 684 } 685 686 queue->tx.req_prod_pvt = i + 1; 687 688 xennet_make_frags(skb, queue, tx); 689 tx->size = skb->len; 690 691 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); 692 if (notify) 693 notify_remote_via_irq(queue->tx_irq); 694 695 u64_stats_update_begin(&stats->syncp); 696 stats->tx_bytes += skb->len; 697 stats->tx_packets++; 698 u64_stats_update_end(&stats->syncp); 699 700 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ 701 xennet_tx_buf_gc(queue); 702 703 if (!netfront_tx_slot_available(queue)) 704 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); 705 706 spin_unlock_irqrestore(&queue->tx_lock, flags); 707 708 return NETDEV_TX_OK; 709 710 drop: 711 dev->stats.tx_dropped++; 712 dev_kfree_skb_any(skb); 713 return NETDEV_TX_OK; 714} 715 716static int xennet_close(struct net_device *dev) 717{ 718 struct netfront_info *np = netdev_priv(dev); 719 unsigned int num_queues = dev->real_num_tx_queues; 720 unsigned int i; 721 struct netfront_queue *queue; 722 netif_tx_stop_all_queues(np->netdev); 723 for (i = 0; i < num_queues; ++i) { 724 queue = &np->queues[i]; 725 napi_disable(&queue->napi); 726 } 727 return 0; 728} 729 730static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, 731 grant_ref_t ref) 732{ 733 int new = xennet_rxidx(queue->rx.req_prod_pvt); 734 735 BUG_ON(queue->rx_skbs[new]); 736 queue->rx_skbs[new] = skb; 737 queue->grant_rx_ref[new] = ref; 738 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; 739 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; 740 queue->rx.req_prod_pvt++; 741} 742 743static int xennet_get_extras(struct netfront_queue *queue, 744 struct xen_netif_extra_info *extras, 745 RING_IDX rp) 746 747{ 748 struct xen_netif_extra_info *extra; 749 struct device *dev = &queue->info->netdev->dev; 750 RING_IDX cons = queue->rx.rsp_cons; 751 int err = 0; 752 753 do { 754 struct sk_buff *skb; 755 grant_ref_t ref; 756 757 if (unlikely(cons + 1 == rp)) { 758 if (net_ratelimit()) 759 dev_warn(dev, "Missing extra info\n"); 760 err = -EBADR; 761 break; 762 } 763 764 extra = (struct xen_netif_extra_info *) 765 RING_GET_RESPONSE(&queue->rx, ++cons); 766 767 if (unlikely(!extra->type || 768 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 769 if (net_ratelimit()) 770 dev_warn(dev, "Invalid extra type: %d\n", 771 extra->type); 772 err = -EINVAL; 773 } else { 774 memcpy(&extras[extra->type - 1], extra, 775 sizeof(*extra)); 776 } 777 778 skb = xennet_get_rx_skb(queue, cons); 779 ref = xennet_get_rx_ref(queue, cons); 780 xennet_move_rx_slot(queue, skb, ref); 781 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 782 783 queue->rx.rsp_cons = cons; 784 return err; 785} 786 787static int xennet_get_responses(struct netfront_queue *queue, 788 struct netfront_rx_info *rinfo, RING_IDX rp, 789 struct sk_buff_head *list) 790{ 791 struct xen_netif_rx_response *rx = &rinfo->rx; 792 struct xen_netif_extra_info *extras = rinfo->extras; 793 struct device *dev = &queue->info->netdev->dev; 794 RING_IDX cons = queue->rx.rsp_cons; 795 struct sk_buff *skb = xennet_get_rx_skb(queue, cons); 796 grant_ref_t ref = xennet_get_rx_ref(queue, cons); 797 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); 798 int slots = 1; 799 int err = 0; 800 unsigned long ret; 801 802 if (rx->flags & XEN_NETRXF_extra_info) { 803 err = xennet_get_extras(queue, extras, rp); 804 cons = queue->rx.rsp_cons; 805 } 806 807 for (;;) { 808 if (unlikely(rx->status < 0 || 809 rx->offset + rx->status > PAGE_SIZE)) { 810 if (net_ratelimit()) 811 dev_warn(dev, "rx->offset: %x, size: %u\n", 812 rx->offset, rx->status); 813 xennet_move_rx_slot(queue, skb, ref); 814 err = -EINVAL; 815 goto next; 816 } 817 818 /* 819 * This definitely indicates a bug, either in this driver or in 820 * the backend driver. In future this should flag the bad 821 * situation to the system controller to reboot the backend. 822 */ 823 if (ref == GRANT_INVALID_REF) { 824 if (net_ratelimit()) 825 dev_warn(dev, "Bad rx response id %d.\n", 826 rx->id); 827 err = -EINVAL; 828 goto next; 829 } 830 831 ret = gnttab_end_foreign_access_ref(ref, 0); 832 BUG_ON(!ret); 833 834 gnttab_release_grant_reference(&queue->gref_rx_head, ref); 835 836 __skb_queue_tail(list, skb); 837 838next: 839 if (!(rx->flags & XEN_NETRXF_more_data)) 840 break; 841 842 if (cons + slots == rp) { 843 if (net_ratelimit()) 844 dev_warn(dev, "Need more slots\n"); 845 err = -ENOENT; 846 break; 847 } 848 849 rx = RING_GET_RESPONSE(&queue->rx, cons + slots); 850 skb = xennet_get_rx_skb(queue, cons + slots); 851 ref = xennet_get_rx_ref(queue, cons + slots); 852 slots++; 853 } 854 855 if (unlikely(slots > max)) { 856 if (net_ratelimit()) 857 dev_warn(dev, "Too many slots\n"); 858 err = -E2BIG; 859 } 860 861 if (unlikely(err)) 862 queue->rx.rsp_cons = cons + slots; 863 864 return err; 865} 866 867static int xennet_set_skb_gso(struct sk_buff *skb, 868 struct xen_netif_extra_info *gso) 869{ 870 if (!gso->u.gso.size) { 871 if (net_ratelimit()) 872 pr_warn("GSO size must not be zero\n"); 873 return -EINVAL; 874 } 875 876 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && 877 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { 878 if (net_ratelimit()) 879 pr_warn("Bad GSO type %d\n", gso->u.gso.type); 880 return -EINVAL; 881 } 882 883 skb_shinfo(skb)->gso_size = gso->u.gso.size; 884 skb_shinfo(skb)->gso_type = 885 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? 886 SKB_GSO_TCPV4 : 887 SKB_GSO_TCPV6; 888 889 /* Header must be checked, and gso_segs computed. */ 890 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 891 skb_shinfo(skb)->gso_segs = 0; 892 893 return 0; 894} 895 896static RING_IDX xennet_fill_frags(struct netfront_queue *queue, 897 struct sk_buff *skb, 898 struct sk_buff_head *list) 899{ 900 struct skb_shared_info *shinfo = skb_shinfo(skb); 901 RING_IDX cons = queue->rx.rsp_cons; 902 struct sk_buff *nskb; 903 904 while ((nskb = __skb_dequeue(list))) { 905 struct xen_netif_rx_response *rx = 906 RING_GET_RESPONSE(&queue->rx, ++cons); 907 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; 908 909 if (shinfo->nr_frags == MAX_SKB_FRAGS) { 910 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 911 912 BUG_ON(pull_to <= skb_headlen(skb)); 913 __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 914 } 915 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS); 916 917 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag), 918 rx->offset, rx->status, PAGE_SIZE); 919 920 skb_shinfo(nskb)->nr_frags = 0; 921 kfree_skb(nskb); 922 } 923 924 return cons; 925} 926 927static int checksum_setup(struct net_device *dev, struct sk_buff *skb) 928{ 929 bool recalculate_partial_csum = false; 930 931 /* 932 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy 933 * peers can fail to set NETRXF_csum_blank when sending a GSO 934 * frame. In this case force the SKB to CHECKSUM_PARTIAL and 935 * recalculate the partial checksum. 936 */ 937 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { 938 struct netfront_info *np = netdev_priv(dev); 939 atomic_inc(&np->rx_gso_checksum_fixup); 940 skb->ip_summed = CHECKSUM_PARTIAL; 941 recalculate_partial_csum = true; 942 } 943 944 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ 945 if (skb->ip_summed != CHECKSUM_PARTIAL) 946 return 0; 947 948 return skb_checksum_setup(skb, recalculate_partial_csum); 949} 950 951static int handle_incoming_queue(struct netfront_queue *queue, 952 struct sk_buff_head *rxq) 953{ 954 struct netfront_stats *stats = this_cpu_ptr(queue->info->stats); 955 int packets_dropped = 0; 956 struct sk_buff *skb; 957 958 while ((skb = __skb_dequeue(rxq)) != NULL) { 959 int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 960 961 if (pull_to > skb_headlen(skb)) 962 __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 963 964 /* Ethernet work: Delayed to here as it peeks the header. */ 965 skb->protocol = eth_type_trans(skb, queue->info->netdev); 966 skb_reset_network_header(skb); 967 968 if (checksum_setup(queue->info->netdev, skb)) { 969 kfree_skb(skb); 970 packets_dropped++; 971 queue->info->netdev->stats.rx_errors++; 972 continue; 973 } 974 975 u64_stats_update_begin(&stats->syncp); 976 stats->rx_packets++; 977 stats->rx_bytes += skb->len; 978 u64_stats_update_end(&stats->syncp); 979 980 /* Pass it up. */ 981 napi_gro_receive(&queue->napi, skb); 982 } 983 984 return packets_dropped; 985} 986 987static int xennet_poll(struct napi_struct *napi, int budget) 988{ 989 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); 990 struct net_device *dev = queue->info->netdev; 991 struct sk_buff *skb; 992 struct netfront_rx_info rinfo; 993 struct xen_netif_rx_response *rx = &rinfo.rx; 994 struct xen_netif_extra_info *extras = rinfo.extras; 995 RING_IDX i, rp; 996 int work_done; 997 struct sk_buff_head rxq; 998 struct sk_buff_head errq; 999 struct sk_buff_head tmpq; 1000 unsigned long flags; 1001 int err; 1002 1003 spin_lock(&queue->rx_lock); 1004 1005 skb_queue_head_init(&rxq); 1006 skb_queue_head_init(&errq); 1007 skb_queue_head_init(&tmpq); 1008 1009 rp = queue->rx.sring->rsp_prod; 1010 rmb(); /* Ensure we see queued responses up to 'rp'. */ 1011 1012 i = queue->rx.rsp_cons; 1013 work_done = 0; 1014 while ((i != rp) && (work_done < budget)) { 1015 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); 1016 memset(extras, 0, sizeof(rinfo.extras)); 1017 1018 err = xennet_get_responses(queue, &rinfo, rp, &tmpq); 1019 1020 if (unlikely(err)) { 1021err: 1022 while ((skb = __skb_dequeue(&tmpq))) 1023 __skb_queue_tail(&errq, skb); 1024 dev->stats.rx_errors++; 1025 i = queue->rx.rsp_cons; 1026 continue; 1027 } 1028 1029 skb = __skb_dequeue(&tmpq); 1030 1031 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { 1032 struct xen_netif_extra_info *gso; 1033 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; 1034 1035 if (unlikely(xennet_set_skb_gso(skb, gso))) { 1036 __skb_queue_head(&tmpq, skb); 1037 queue->rx.rsp_cons += skb_queue_len(&tmpq); 1038 goto err; 1039 } 1040 } 1041 1042 NETFRONT_SKB_CB(skb)->pull_to = rx->status; 1043 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) 1044 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; 1045 1046 skb_shinfo(skb)->frags[0].page_offset = rx->offset; 1047 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); 1048 skb->data_len = rx->status; 1049 skb->len += rx->status; 1050 1051 i = xennet_fill_frags(queue, skb, &tmpq); 1052 1053 if (rx->flags & XEN_NETRXF_csum_blank) 1054 skb->ip_summed = CHECKSUM_PARTIAL; 1055 else if (rx->flags & XEN_NETRXF_data_validated) 1056 skb->ip_summed = CHECKSUM_UNNECESSARY; 1057 1058 __skb_queue_tail(&rxq, skb); 1059 1060 queue->rx.rsp_cons = ++i; 1061 work_done++; 1062 } 1063 1064 __skb_queue_purge(&errq); 1065 1066 work_done -= handle_incoming_queue(queue, &rxq); 1067 1068 /* If we get a callback with very few responses, reduce fill target. */ 1069 /* NB. Note exponential increase, linear decrease. */ 1070 if (((queue->rx.req_prod_pvt - queue->rx.sring->rsp_prod) > 1071 ((3*queue->rx_target) / 4)) && 1072 (--queue->rx_target < queue->rx_min_target)) 1073 queue->rx_target = queue->rx_min_target; 1074 1075 xennet_alloc_rx_buffers(queue); 1076 1077 if (work_done < budget) { 1078 int more_to_do = 0; 1079 1080 napi_gro_flush(napi, false); 1081 1082 local_irq_save(flags); 1083 1084 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); 1085 if (!more_to_do) 1086 __napi_complete(napi); 1087 1088 local_irq_restore(flags); 1089 } 1090 1091 spin_unlock(&queue->rx_lock); 1092 1093 return work_done; 1094} 1095 1096static int xennet_change_mtu(struct net_device *dev, int mtu) 1097{ 1098 int max = xennet_can_sg(dev) ? 1099 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN; 1100 1101 if (mtu > max) 1102 return -EINVAL; 1103 dev->mtu = mtu; 1104 return 0; 1105} 1106 1107static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev, 1108 struct rtnl_link_stats64 *tot) 1109{ 1110 struct netfront_info *np = netdev_priv(dev); 1111 int cpu; 1112 1113 for_each_possible_cpu(cpu) { 1114 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu); 1115 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 1116 unsigned int start; 1117 1118 do { 1119 start = u64_stats_fetch_begin_irq(&stats->syncp); 1120 1121 rx_packets = stats->rx_packets; 1122 tx_packets = stats->tx_packets; 1123 rx_bytes = stats->rx_bytes; 1124 tx_bytes = stats->tx_bytes; 1125 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1126 1127 tot->rx_packets += rx_packets; 1128 tot->tx_packets += tx_packets; 1129 tot->rx_bytes += rx_bytes; 1130 tot->tx_bytes += tx_bytes; 1131 } 1132 1133 tot->rx_errors = dev->stats.rx_errors; 1134 tot->tx_dropped = dev->stats.tx_dropped; 1135 1136 return tot; 1137} 1138 1139static void xennet_release_tx_bufs(struct netfront_queue *queue) 1140{ 1141 struct sk_buff *skb; 1142 int i; 1143 1144 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1145 /* Skip over entries which are actually freelist references */ 1146 if (skb_entry_is_link(&queue->tx_skbs[i])) 1147 continue; 1148 1149 skb = queue->tx_skbs[i].skb; 1150 get_page(queue->grant_tx_page[i]); 1151 gnttab_end_foreign_access(queue->grant_tx_ref[i], 1152 GNTMAP_readonly, 1153 (unsigned long)page_address(queue->grant_tx_page[i])); 1154 queue->grant_tx_page[i] = NULL; 1155 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1156 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); 1157 dev_kfree_skb_irq(skb); 1158 } 1159} 1160 1161static void xennet_release_rx_bufs(struct netfront_queue *queue) 1162{ 1163 int id, ref; 1164 1165 spin_lock_bh(&queue->rx_lock); 1166 1167 for (id = 0; id < NET_RX_RING_SIZE; id++) { 1168 struct sk_buff *skb; 1169 struct page *page; 1170 1171 skb = queue->rx_skbs[id]; 1172 if (!skb) 1173 continue; 1174 1175 ref = queue->grant_rx_ref[id]; 1176 if (ref == GRANT_INVALID_REF) 1177 continue; 1178 1179 page = skb_frag_page(&skb_shinfo(skb)->frags[0]); 1180 1181 /* gnttab_end_foreign_access() needs a page ref until 1182 * foreign access is ended (which may be deferred). 1183 */ 1184 get_page(page); 1185 gnttab_end_foreign_access(ref, 0, 1186 (unsigned long)page_address(page)); 1187 queue->grant_rx_ref[id] = GRANT_INVALID_REF; 1188 1189 kfree_skb(skb); 1190 } 1191 1192 spin_unlock_bh(&queue->rx_lock); 1193} 1194 1195static netdev_features_t xennet_fix_features(struct net_device *dev, 1196 netdev_features_t features) 1197{ 1198 struct netfront_info *np = netdev_priv(dev); 1199 int val; 1200 1201 if (features & NETIF_F_SG) { 1202 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", 1203 "%d", &val) < 0) 1204 val = 0; 1205 1206 if (!val) 1207 features &= ~NETIF_F_SG; 1208 } 1209 1210 if (features & NETIF_F_IPV6_CSUM) { 1211 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, 1212 "feature-ipv6-csum-offload", "%d", &val) < 0) 1213 val = 0; 1214 1215 if (!val) 1216 features &= ~NETIF_F_IPV6_CSUM; 1217 } 1218 1219 if (features & NETIF_F_TSO) { 1220 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, 1221 "feature-gso-tcpv4", "%d", &val) < 0) 1222 val = 0; 1223 1224 if (!val) 1225 features &= ~NETIF_F_TSO; 1226 } 1227 1228 if (features & NETIF_F_TSO6) { 1229 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, 1230 "feature-gso-tcpv6", "%d", &val) < 0) 1231 val = 0; 1232 1233 if (!val) 1234 features &= ~NETIF_F_TSO6; 1235 } 1236 1237 return features; 1238} 1239 1240static int xennet_set_features(struct net_device *dev, 1241 netdev_features_t features) 1242{ 1243 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { 1244 netdev_info(dev, "Reducing MTU because no SG offload"); 1245 dev->mtu = ETH_DATA_LEN; 1246 } 1247 1248 return 0; 1249} 1250 1251static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) 1252{ 1253 struct netfront_queue *queue = dev_id; 1254 unsigned long flags; 1255 1256 spin_lock_irqsave(&queue->tx_lock, flags); 1257 xennet_tx_buf_gc(queue); 1258 spin_unlock_irqrestore(&queue->tx_lock, flags); 1259 1260 return IRQ_HANDLED; 1261} 1262 1263static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) 1264{ 1265 struct netfront_queue *queue = dev_id; 1266 struct net_device *dev = queue->info->netdev; 1267 1268 if (likely(netif_carrier_ok(dev) && 1269 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) 1270 napi_schedule(&queue->napi); 1271 1272 return IRQ_HANDLED; 1273} 1274 1275static irqreturn_t xennet_interrupt(int irq, void *dev_id) 1276{ 1277 xennet_tx_interrupt(irq, dev_id); 1278 xennet_rx_interrupt(irq, dev_id); 1279 return IRQ_HANDLED; 1280} 1281 1282#ifdef CONFIG_NET_POLL_CONTROLLER 1283static void xennet_poll_controller(struct net_device *dev) 1284{ 1285 /* Poll each queue */ 1286 struct netfront_info *info = netdev_priv(dev); 1287 unsigned int num_queues = dev->real_num_tx_queues; 1288 unsigned int i; 1289 for (i = 0; i < num_queues; ++i) 1290 xennet_interrupt(0, &info->queues[i]); 1291} 1292#endif 1293 1294static const struct net_device_ops xennet_netdev_ops = { 1295 .ndo_open = xennet_open, 1296 .ndo_stop = xennet_close, 1297 .ndo_start_xmit = xennet_start_xmit, 1298 .ndo_change_mtu = xennet_change_mtu, 1299 .ndo_get_stats64 = xennet_get_stats64, 1300 .ndo_set_mac_address = eth_mac_addr, 1301 .ndo_validate_addr = eth_validate_addr, 1302 .ndo_fix_features = xennet_fix_features, 1303 .ndo_set_features = xennet_set_features, 1304 .ndo_select_queue = xennet_select_queue, 1305#ifdef CONFIG_NET_POLL_CONTROLLER 1306 .ndo_poll_controller = xennet_poll_controller, 1307#endif 1308}; 1309 1310static struct net_device *xennet_create_dev(struct xenbus_device *dev) 1311{ 1312 int err; 1313 struct net_device *netdev; 1314 struct netfront_info *np; 1315 1316 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); 1317 if (!netdev) 1318 return ERR_PTR(-ENOMEM); 1319 1320 np = netdev_priv(netdev); 1321 np->xbdev = dev; 1322 1323 /* No need to use rtnl_lock() before the call below as it 1324 * happens before register_netdev(). 1325 */ 1326 netif_set_real_num_tx_queues(netdev, 0); 1327 np->queues = NULL; 1328 1329 err = -ENOMEM; 1330 np->stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1331 if (np->stats == NULL) 1332 goto exit; 1333 1334 netdev->netdev_ops = &xennet_netdev_ops; 1335 1336 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | 1337 NETIF_F_GSO_ROBUST; 1338 netdev->hw_features = NETIF_F_SG | 1339 NETIF_F_IPV6_CSUM | 1340 NETIF_F_TSO | NETIF_F_TSO6; 1341 1342 /* 1343 * Assume that all hw features are available for now. This set 1344 * will be adjusted by the call to netdev_update_features() in 1345 * xennet_connect() which is the earliest point where we can 1346 * negotiate with the backend regarding supported features. 1347 */ 1348 netdev->features |= netdev->hw_features; 1349 1350 netdev->ethtool_ops = &xennet_ethtool_ops; 1351 SET_NETDEV_DEV(netdev, &dev->dev); 1352 1353 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER); 1354 1355 np->netdev = netdev; 1356 1357 netif_carrier_off(netdev); 1358 1359 return netdev; 1360 1361 exit: 1362 free_netdev(netdev); 1363 return ERR_PTR(err); 1364} 1365 1366/** 1367 * Entry point to this code when a new device is created. Allocate the basic 1368 * structures and the ring buffers for communication with the backend, and 1369 * inform the backend of the appropriate details for those. 1370 */ 1371static int netfront_probe(struct xenbus_device *dev, 1372 const struct xenbus_device_id *id) 1373{ 1374 int err; 1375 struct net_device *netdev; 1376 struct netfront_info *info; 1377 1378 netdev = xennet_create_dev(dev); 1379 if (IS_ERR(netdev)) { 1380 err = PTR_ERR(netdev); 1381 xenbus_dev_fatal(dev, err, "creating netdev"); 1382 return err; 1383 } 1384 1385 info = netdev_priv(netdev); 1386 dev_set_drvdata(&dev->dev, info); 1387 1388 err = register_netdev(info->netdev); 1389 if (err) { 1390 pr_warn("%s: register_netdev err=%d\n", __func__, err); 1391 goto fail; 1392 } 1393 1394 err = xennet_sysfs_addif(info->netdev); 1395 if (err) { 1396 unregister_netdev(info->netdev); 1397 pr_warn("%s: add sysfs failed err=%d\n", __func__, err); 1398 goto fail; 1399 } 1400 1401 return 0; 1402 1403 fail: 1404 free_netdev(netdev); 1405 dev_set_drvdata(&dev->dev, NULL); 1406 return err; 1407} 1408 1409static void xennet_end_access(int ref, void *page) 1410{ 1411 /* This frees the page as a side-effect */ 1412 if (ref != GRANT_INVALID_REF) 1413 gnttab_end_foreign_access(ref, 0, (unsigned long)page); 1414} 1415 1416static void xennet_disconnect_backend(struct netfront_info *info) 1417{ 1418 unsigned int i = 0; 1419 unsigned int num_queues = info->netdev->real_num_tx_queues; 1420 1421 netif_carrier_off(info->netdev); 1422 1423 for (i = 0; i < num_queues; ++i) { 1424 struct netfront_queue *queue = &info->queues[i]; 1425 1426 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) 1427 unbind_from_irqhandler(queue->tx_irq, queue); 1428 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { 1429 unbind_from_irqhandler(queue->tx_irq, queue); 1430 unbind_from_irqhandler(queue->rx_irq, queue); 1431 } 1432 queue->tx_evtchn = queue->rx_evtchn = 0; 1433 queue->tx_irq = queue->rx_irq = 0; 1434 1435 napi_synchronize(&queue->napi); 1436 1437 xennet_release_tx_bufs(queue); 1438 xennet_release_rx_bufs(queue); 1439 gnttab_free_grant_references(queue->gref_tx_head); 1440 gnttab_free_grant_references(queue->gref_rx_head); 1441 1442 /* End access and free the pages */ 1443 xennet_end_access(queue->tx_ring_ref, queue->tx.sring); 1444 xennet_end_access(queue->rx_ring_ref, queue->rx.sring); 1445 1446 queue->tx_ring_ref = GRANT_INVALID_REF; 1447 queue->rx_ring_ref = GRANT_INVALID_REF; 1448 queue->tx.sring = NULL; 1449 queue->rx.sring = NULL; 1450 } 1451} 1452 1453/** 1454 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1455 * driver restart. We tear down our netif structure and recreate it, but 1456 * leave the device-layer structures intact so that this is transparent to the 1457 * rest of the kernel. 1458 */ 1459static int netfront_resume(struct xenbus_device *dev) 1460{ 1461 struct netfront_info *info = dev_get_drvdata(&dev->dev); 1462 1463 dev_dbg(&dev->dev, "%s\n", dev->nodename); 1464 1465 xennet_disconnect_backend(info); 1466 return 0; 1467} 1468 1469static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) 1470{ 1471 char *s, *e, *macstr; 1472 int i; 1473 1474 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); 1475 if (IS_ERR(macstr)) 1476 return PTR_ERR(macstr); 1477 1478 for (i = 0; i < ETH_ALEN; i++) { 1479 mac[i] = simple_strtoul(s, &e, 16); 1480 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { 1481 kfree(macstr); 1482 return -ENOENT; 1483 } 1484 s = e+1; 1485 } 1486 1487 kfree(macstr); 1488 return 0; 1489} 1490 1491static int setup_netfront_single(struct netfront_queue *queue) 1492{ 1493 int err; 1494 1495 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1496 if (err < 0) 1497 goto fail; 1498 1499 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1500 xennet_interrupt, 1501 0, queue->info->netdev->name, queue); 1502 if (err < 0) 1503 goto bind_fail; 1504 queue->rx_evtchn = queue->tx_evtchn; 1505 queue->rx_irq = queue->tx_irq = err; 1506 1507 return 0; 1508 1509bind_fail: 1510 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1511 queue->tx_evtchn = 0; 1512fail: 1513 return err; 1514} 1515 1516static int setup_netfront_split(struct netfront_queue *queue) 1517{ 1518 int err; 1519 1520 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1521 if (err < 0) 1522 goto fail; 1523 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); 1524 if (err < 0) 1525 goto alloc_rx_evtchn_fail; 1526 1527 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), 1528 "%s-tx", queue->name); 1529 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1530 xennet_tx_interrupt, 1531 0, queue->tx_irq_name, queue); 1532 if (err < 0) 1533 goto bind_tx_fail; 1534 queue->tx_irq = err; 1535 1536 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), 1537 "%s-rx", queue->name); 1538 err = bind_evtchn_to_irqhandler(queue->rx_evtchn, 1539 xennet_rx_interrupt, 1540 0, queue->rx_irq_name, queue); 1541 if (err < 0) 1542 goto bind_rx_fail; 1543 queue->rx_irq = err; 1544 1545 return 0; 1546 1547bind_rx_fail: 1548 unbind_from_irqhandler(queue->tx_irq, queue); 1549 queue->tx_irq = 0; 1550bind_tx_fail: 1551 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); 1552 queue->rx_evtchn = 0; 1553alloc_rx_evtchn_fail: 1554 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1555 queue->tx_evtchn = 0; 1556fail: 1557 return err; 1558} 1559 1560static int setup_netfront(struct xenbus_device *dev, 1561 struct netfront_queue *queue, unsigned int feature_split_evtchn) 1562{ 1563 struct xen_netif_tx_sring *txs; 1564 struct xen_netif_rx_sring *rxs; 1565 int err; 1566 1567 queue->tx_ring_ref = GRANT_INVALID_REF; 1568 queue->rx_ring_ref = GRANT_INVALID_REF; 1569 queue->rx.sring = NULL; 1570 queue->tx.sring = NULL; 1571 1572 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1573 if (!txs) { 1574 err = -ENOMEM; 1575 xenbus_dev_fatal(dev, err, "allocating tx ring page"); 1576 goto fail; 1577 } 1578 SHARED_RING_INIT(txs); 1579 FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE); 1580 1581 err = xenbus_grant_ring(dev, virt_to_mfn(txs)); 1582 if (err < 0) 1583 goto grant_tx_ring_fail; 1584 queue->tx_ring_ref = err; 1585 1586 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1587 if (!rxs) { 1588 err = -ENOMEM; 1589 xenbus_dev_fatal(dev, err, "allocating rx ring page"); 1590 goto alloc_rx_ring_fail; 1591 } 1592 SHARED_RING_INIT(rxs); 1593 FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE); 1594 1595 err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); 1596 if (err < 0) 1597 goto grant_rx_ring_fail; 1598 queue->rx_ring_ref = err; 1599 1600 if (feature_split_evtchn) 1601 err = setup_netfront_split(queue); 1602 /* setup single event channel if 1603 * a) feature-split-event-channels == 0 1604 * b) feature-split-event-channels == 1 but failed to setup 1605 */ 1606 if (!feature_split_evtchn || (feature_split_evtchn && err)) 1607 err = setup_netfront_single(queue); 1608 1609 if (err) 1610 goto alloc_evtchn_fail; 1611 1612 return 0; 1613 1614 /* If we fail to setup netfront, it is safe to just revoke access to 1615 * granted pages because backend is not accessing it at this point. 1616 */ 1617alloc_evtchn_fail: 1618 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); 1619grant_rx_ring_fail: 1620 free_page((unsigned long)rxs); 1621alloc_rx_ring_fail: 1622 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); 1623grant_tx_ring_fail: 1624 free_page((unsigned long)txs); 1625fail: 1626 return err; 1627} 1628 1629/* Queue-specific initialisation 1630 * This used to be done in xennet_create_dev() but must now 1631 * be run per-queue. 1632 */ 1633static int xennet_init_queue(struct netfront_queue *queue) 1634{ 1635 unsigned short i; 1636 int err = 0; 1637 1638 spin_lock_init(&queue->tx_lock); 1639 spin_lock_init(&queue->rx_lock); 1640 1641 skb_queue_head_init(&queue->rx_batch); 1642 queue->rx_target = RX_DFL_MIN_TARGET; 1643 queue->rx_min_target = RX_DFL_MIN_TARGET; 1644 queue->rx_max_target = RX_MAX_TARGET; 1645 1646 init_timer(&queue->rx_refill_timer); 1647 queue->rx_refill_timer.data = (unsigned long)queue; 1648 queue->rx_refill_timer.function = rx_refill_timeout; 1649 1650 snprintf(queue->name, sizeof(queue->name), "%s-q%u", 1651 queue->info->netdev->name, queue->id); 1652 1653 /* Initialise tx_skbs as a free chain containing every entry. */ 1654 queue->tx_skb_freelist = 0; 1655 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1656 skb_entry_set_link(&queue->tx_skbs[i], i+1); 1657 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1658 queue->grant_tx_page[i] = NULL; 1659 } 1660 1661 /* Clear out rx_skbs */ 1662 for (i = 0; i < NET_RX_RING_SIZE; i++) { 1663 queue->rx_skbs[i] = NULL; 1664 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 1665 } 1666 1667 /* A grant for every tx ring slot */ 1668 if (gnttab_alloc_grant_references(TX_MAX_TARGET, 1669 &queue->gref_tx_head) < 0) { 1670 pr_alert("can't alloc tx grant refs\n"); 1671 err = -ENOMEM; 1672 goto exit; 1673 } 1674 1675 /* A grant for every rx ring slot */ 1676 if (gnttab_alloc_grant_references(RX_MAX_TARGET, 1677 &queue->gref_rx_head) < 0) { 1678 pr_alert("can't alloc rx grant refs\n"); 1679 err = -ENOMEM; 1680 goto exit_free_tx; 1681 } 1682 1683 return 0; 1684 1685 exit_free_tx: 1686 gnttab_free_grant_references(queue->gref_tx_head); 1687 exit: 1688 return err; 1689} 1690 1691static int write_queue_xenstore_keys(struct netfront_queue *queue, 1692 struct xenbus_transaction *xbt, int write_hierarchical) 1693{ 1694 /* Write the queue-specific keys into XenStore in the traditional 1695 * way for a single queue, or in a queue subkeys for multiple 1696 * queues. 1697 */ 1698 struct xenbus_device *dev = queue->info->xbdev; 1699 int err; 1700 const char *message; 1701 char *path; 1702 size_t pathsize; 1703 1704 /* Choose the correct place to write the keys */ 1705 if (write_hierarchical) { 1706 pathsize = strlen(dev->nodename) + 10; 1707 path = kzalloc(pathsize, GFP_KERNEL); 1708 if (!path) { 1709 err = -ENOMEM; 1710 message = "out of memory while writing ring references"; 1711 goto error; 1712 } 1713 snprintf(path, pathsize, "%s/queue-%u", 1714 dev->nodename, queue->id); 1715 } else { 1716 path = (char *)dev->nodename; 1717 } 1718 1719 /* Write ring references */ 1720 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", 1721 queue->tx_ring_ref); 1722 if (err) { 1723 message = "writing tx-ring-ref"; 1724 goto error; 1725 } 1726 1727 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", 1728 queue->rx_ring_ref); 1729 if (err) { 1730 message = "writing rx-ring-ref"; 1731 goto error; 1732 } 1733 1734 /* Write event channels; taking into account both shared 1735 * and split event channel scenarios. 1736 */ 1737 if (queue->tx_evtchn == queue->rx_evtchn) { 1738 /* Shared event channel */ 1739 err = xenbus_printf(*xbt, path, 1740 "event-channel", "%u", queue->tx_evtchn); 1741 if (err) { 1742 message = "writing event-channel"; 1743 goto error; 1744 } 1745 } else { 1746 /* Split event channels */ 1747 err = xenbus_printf(*xbt, path, 1748 "event-channel-tx", "%u", queue->tx_evtchn); 1749 if (err) { 1750 message = "writing event-channel-tx"; 1751 goto error; 1752 } 1753 1754 err = xenbus_printf(*xbt, path, 1755 "event-channel-rx", "%u", queue->rx_evtchn); 1756 if (err) { 1757 message = "writing event-channel-rx"; 1758 goto error; 1759 } 1760 } 1761 1762 if (write_hierarchical) 1763 kfree(path); 1764 return 0; 1765 1766error: 1767 if (write_hierarchical) 1768 kfree(path); 1769 xenbus_dev_fatal(dev, err, "%s", message); 1770 return err; 1771} 1772 1773static void xennet_destroy_queues(struct netfront_info *info) 1774{ 1775 unsigned int i; 1776 1777 rtnl_lock(); 1778 1779 for (i = 0; i < info->netdev->real_num_tx_queues; i++) { 1780 struct netfront_queue *queue = &info->queues[i]; 1781 1782 if (netif_running(info->netdev)) 1783 napi_disable(&queue->napi); 1784 netif_napi_del(&queue->napi); 1785 } 1786 1787 rtnl_unlock(); 1788 1789 kfree(info->queues); 1790 info->queues = NULL; 1791} 1792 1793static int xennet_create_queues(struct netfront_info *info, 1794 unsigned int num_queues) 1795{ 1796 unsigned int i; 1797 int ret; 1798 1799 info->queues = kcalloc(num_queues, sizeof(struct netfront_queue), 1800 GFP_KERNEL); 1801 if (!info->queues) 1802 return -ENOMEM; 1803 1804 rtnl_lock(); 1805 1806 for (i = 0; i < num_queues; i++) { 1807 struct netfront_queue *queue = &info->queues[i]; 1808 1809 queue->id = i; 1810 queue->info = info; 1811 1812 ret = xennet_init_queue(queue); 1813 if (ret < 0) { 1814 dev_warn(&info->netdev->dev, 1815 "only created %d queues\n", i); 1816 num_queues = i; 1817 break; 1818 } 1819 1820 netif_napi_add(queue->info->netdev, &queue->napi, 1821 xennet_poll, 64); 1822 if (netif_running(info->netdev)) 1823 napi_enable(&queue->napi); 1824 } 1825 1826 netif_set_real_num_tx_queues(info->netdev, num_queues); 1827 1828 rtnl_unlock(); 1829 1830 if (num_queues == 0) { 1831 dev_err(&info->netdev->dev, "no queues\n"); 1832 return -EINVAL; 1833 } 1834 return 0; 1835} 1836 1837/* Common code used when first setting up, and when resuming. */ 1838static int talk_to_netback(struct xenbus_device *dev, 1839 struct netfront_info *info) 1840{ 1841 const char *message; 1842 struct xenbus_transaction xbt; 1843 int err; 1844 unsigned int feature_split_evtchn; 1845 unsigned int i = 0; 1846 unsigned int max_queues = 0; 1847 struct netfront_queue *queue = NULL; 1848 unsigned int num_queues = 1; 1849 1850 info->netdev->irq = 0; 1851 1852 /* Check if backend supports multiple queues */ 1853 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 1854 "multi-queue-max-queues", "%u", &max_queues); 1855 if (err < 0) 1856 max_queues = 1; 1857 num_queues = min(max_queues, xennet_max_queues); 1858 1859 /* Check feature-split-event-channels */ 1860 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend, 1861 "feature-split-event-channels", "%u", 1862 &feature_split_evtchn); 1863 if (err < 0) 1864 feature_split_evtchn = 0; 1865 1866 /* Read mac addr. */ 1867 err = xen_net_read_mac(dev, info->netdev->dev_addr); 1868 if (err) { 1869 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); 1870 goto out; 1871 } 1872 1873 if (info->queues) 1874 xennet_destroy_queues(info); 1875 1876 err = xennet_create_queues(info, num_queues); 1877 if (err < 0) 1878 goto destroy_ring; 1879 1880 /* Create shared ring, alloc event channel -- for each queue */ 1881 for (i = 0; i < num_queues; ++i) { 1882 queue = &info->queues[i]; 1883 err = setup_netfront(dev, queue, feature_split_evtchn); 1884 if (err) { 1885 /* setup_netfront() will tidy up the current 1886 * queue on error, but we need to clean up 1887 * those already allocated. 1888 */ 1889 if (i > 0) { 1890 rtnl_lock(); 1891 netif_set_real_num_tx_queues(info->netdev, i); 1892 rtnl_unlock(); 1893 goto destroy_ring; 1894 } else { 1895 goto out; 1896 } 1897 } 1898 } 1899 1900again: 1901 err = xenbus_transaction_start(&xbt); 1902 if (err) { 1903 xenbus_dev_fatal(dev, err, "starting transaction"); 1904 goto destroy_ring; 1905 } 1906 1907 if (num_queues == 1) { 1908 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ 1909 if (err) 1910 goto abort_transaction_no_dev_fatal; 1911 } else { 1912 /* Write the number of queues */ 1913 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues", 1914 "%u", num_queues); 1915 if (err) { 1916 message = "writing multi-queue-num-queues"; 1917 goto abort_transaction_no_dev_fatal; 1918 } 1919 1920 /* Write the keys for each queue */ 1921 for (i = 0; i < num_queues; ++i) { 1922 queue = &info->queues[i]; 1923 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ 1924 if (err) 1925 goto abort_transaction_no_dev_fatal; 1926 } 1927 } 1928 1929 /* The remaining keys are not queue-specific */ 1930 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", 1931 1); 1932 if (err) { 1933 message = "writing request-rx-copy"; 1934 goto abort_transaction; 1935 } 1936 1937 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); 1938 if (err) { 1939 message = "writing feature-rx-notify"; 1940 goto abort_transaction; 1941 } 1942 1943 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); 1944 if (err) { 1945 message = "writing feature-sg"; 1946 goto abort_transaction; 1947 } 1948 1949 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); 1950 if (err) { 1951 message = "writing feature-gso-tcpv4"; 1952 goto abort_transaction; 1953 } 1954 1955 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); 1956 if (err) { 1957 message = "writing feature-gso-tcpv6"; 1958 goto abort_transaction; 1959 } 1960 1961 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", 1962 "1"); 1963 if (err) { 1964 message = "writing feature-ipv6-csum-offload"; 1965 goto abort_transaction; 1966 } 1967 1968 err = xenbus_transaction_end(xbt, 0); 1969 if (err) { 1970 if (err == -EAGAIN) 1971 goto again; 1972 xenbus_dev_fatal(dev, err, "completing transaction"); 1973 goto destroy_ring; 1974 } 1975 1976 return 0; 1977 1978 abort_transaction: 1979 xenbus_dev_fatal(dev, err, "%s", message); 1980abort_transaction_no_dev_fatal: 1981 xenbus_transaction_end(xbt, 1); 1982 destroy_ring: 1983 xennet_disconnect_backend(info); 1984 kfree(info->queues); 1985 info->queues = NULL; 1986 rtnl_lock(); 1987 netif_set_real_num_tx_queues(info->netdev, 0); 1988 rtnl_unlock(); 1989 out: 1990 return err; 1991} 1992 1993static int xennet_connect(struct net_device *dev) 1994{ 1995 struct netfront_info *np = netdev_priv(dev); 1996 unsigned int num_queues = 0; 1997 int err; 1998 unsigned int feature_rx_copy; 1999 unsigned int j = 0; 2000 struct netfront_queue *queue = NULL; 2001 2002 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, 2003 "feature-rx-copy", "%u", &feature_rx_copy); 2004 if (err != 1) 2005 feature_rx_copy = 0; 2006 2007 if (!feature_rx_copy) { 2008 dev_info(&dev->dev, 2009 "backend does not support copying receive path\n"); 2010 return -ENODEV; 2011 } 2012 2013 err = talk_to_netback(np->xbdev, np); 2014 if (err) 2015 return err; 2016 2017 /* talk_to_netback() sets the correct number of queues */ 2018 num_queues = dev->real_num_tx_queues; 2019 2020 rtnl_lock(); 2021 netdev_update_features(dev); 2022 rtnl_unlock(); 2023 2024 /* 2025 * All public and private state should now be sane. Get 2026 * ready to start sending and receiving packets and give the driver 2027 * domain a kick because we've probably just requeued some 2028 * packets. 2029 */ 2030 netif_carrier_on(np->netdev); 2031 for (j = 0; j < num_queues; ++j) { 2032 queue = &np->queues[j]; 2033 2034 notify_remote_via_irq(queue->tx_irq); 2035 if (queue->tx_irq != queue->rx_irq) 2036 notify_remote_via_irq(queue->rx_irq); 2037 2038 spin_lock_irq(&queue->tx_lock); 2039 xennet_tx_buf_gc(queue); 2040 spin_unlock_irq(&queue->tx_lock); 2041 2042 spin_lock_bh(&queue->rx_lock); 2043 xennet_alloc_rx_buffers(queue); 2044 spin_unlock_bh(&queue->rx_lock); 2045 } 2046 2047 return 0; 2048} 2049 2050/** 2051 * Callback received when the backend's state changes. 2052 */ 2053static void netback_changed(struct xenbus_device *dev, 2054 enum xenbus_state backend_state) 2055{ 2056 struct netfront_info *np = dev_get_drvdata(&dev->dev); 2057 struct net_device *netdev = np->netdev; 2058 2059 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); 2060 2061 switch (backend_state) { 2062 case XenbusStateInitialising: 2063 case XenbusStateInitialised: 2064 case XenbusStateReconfiguring: 2065 case XenbusStateReconfigured: 2066 case XenbusStateUnknown: 2067 break; 2068 2069 case XenbusStateInitWait: 2070 if (dev->state != XenbusStateInitialising) 2071 break; 2072 if (xennet_connect(netdev) != 0) 2073 break; 2074 xenbus_switch_state(dev, XenbusStateConnected); 2075 break; 2076 2077 case XenbusStateConnected: 2078 netdev_notify_peers(netdev); 2079 break; 2080 2081 case XenbusStateClosed: 2082 if (dev->state == XenbusStateClosed) 2083 break; 2084 /* Missed the backend's CLOSING state -- fallthrough */ 2085 case XenbusStateClosing: 2086 xenbus_frontend_closed(dev); 2087 break; 2088 } 2089} 2090 2091static const struct xennet_stat { 2092 char name[ETH_GSTRING_LEN]; 2093 u16 offset; 2094} xennet_stats[] = { 2095 { 2096 "rx_gso_checksum_fixup", 2097 offsetof(struct netfront_info, rx_gso_checksum_fixup) 2098 }, 2099}; 2100 2101static int xennet_get_sset_count(struct net_device *dev, int string_set) 2102{ 2103 switch (string_set) { 2104 case ETH_SS_STATS: 2105 return ARRAY_SIZE(xennet_stats); 2106 default: 2107 return -EINVAL; 2108 } 2109} 2110 2111static void xennet_get_ethtool_stats(struct net_device *dev, 2112 struct ethtool_stats *stats, u64 * data) 2113{ 2114 void *np = netdev_priv(dev); 2115 int i; 2116 2117 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2118 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); 2119} 2120 2121static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) 2122{ 2123 int i; 2124 2125 switch (stringset) { 2126 case ETH_SS_STATS: 2127 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2128 memcpy(data + i * ETH_GSTRING_LEN, 2129 xennet_stats[i].name, ETH_GSTRING_LEN); 2130 break; 2131 } 2132} 2133 2134static const struct ethtool_ops xennet_ethtool_ops = 2135{ 2136 .get_link = ethtool_op_get_link, 2137 2138 .get_sset_count = xennet_get_sset_count, 2139 .get_ethtool_stats = xennet_get_ethtool_stats, 2140 .get_strings = xennet_get_strings, 2141}; 2142 2143#ifdef CONFIG_SYSFS 2144static ssize_t show_rxbuf_min(struct device *dev, 2145 struct device_attribute *attr, char *buf) 2146{ 2147 struct net_device *netdev = to_net_dev(dev); 2148 struct netfront_info *info = netdev_priv(netdev); 2149 unsigned int num_queues = netdev->real_num_tx_queues; 2150 2151 if (num_queues) 2152 return sprintf(buf, "%u\n", info->queues[0].rx_min_target); 2153 else 2154 return sprintf(buf, "%u\n", RX_MIN_TARGET); 2155} 2156 2157static ssize_t store_rxbuf_min(struct device *dev, 2158 struct device_attribute *attr, 2159 const char *buf, size_t len) 2160{ 2161 struct net_device *netdev = to_net_dev(dev); 2162 struct netfront_info *np = netdev_priv(netdev); 2163 unsigned int num_queues = netdev->real_num_tx_queues; 2164 char *endp; 2165 unsigned long target; 2166 unsigned int i; 2167 struct netfront_queue *queue; 2168 2169 if (!capable(CAP_NET_ADMIN)) 2170 return -EPERM; 2171 2172 target = simple_strtoul(buf, &endp, 0); 2173 if (endp == buf) 2174 return -EBADMSG; 2175 2176 if (target < RX_MIN_TARGET) 2177 target = RX_MIN_TARGET; 2178 if (target > RX_MAX_TARGET) 2179 target = RX_MAX_TARGET; 2180 2181 for (i = 0; i < num_queues; ++i) { 2182 queue = &np->queues[i]; 2183 spin_lock_bh(&queue->rx_lock); 2184 if (target > queue->rx_max_target) 2185 queue->rx_max_target = target; 2186 queue->rx_min_target = target; 2187 if (target > queue->rx_target) 2188 queue->rx_target = target; 2189 2190 xennet_alloc_rx_buffers(queue); 2191 2192 spin_unlock_bh(&queue->rx_lock); 2193 } 2194 return len; 2195} 2196 2197static ssize_t show_rxbuf_max(struct device *dev, 2198 struct device_attribute *attr, char *buf) 2199{ 2200 struct net_device *netdev = to_net_dev(dev); 2201 struct netfront_info *info = netdev_priv(netdev); 2202 unsigned int num_queues = netdev->real_num_tx_queues; 2203 2204 if (num_queues) 2205 return sprintf(buf, "%u\n", info->queues[0].rx_max_target); 2206 else 2207 return sprintf(buf, "%u\n", RX_MAX_TARGET); 2208} 2209 2210static ssize_t store_rxbuf_max(struct device *dev, 2211 struct device_attribute *attr, 2212 const char *buf, size_t len) 2213{ 2214 struct net_device *netdev = to_net_dev(dev); 2215 struct netfront_info *np = netdev_priv(netdev); 2216 unsigned int num_queues = netdev->real_num_tx_queues; 2217 char *endp; 2218 unsigned long target; 2219 unsigned int i = 0; 2220 struct netfront_queue *queue = NULL; 2221 2222 if (!capable(CAP_NET_ADMIN)) 2223 return -EPERM; 2224 2225 target = simple_strtoul(buf, &endp, 0); 2226 if (endp == buf) 2227 return -EBADMSG; 2228 2229 if (target < RX_MIN_TARGET) 2230 target = RX_MIN_TARGET; 2231 if (target > RX_MAX_TARGET) 2232 target = RX_MAX_TARGET; 2233 2234 for (i = 0; i < num_queues; ++i) { 2235 queue = &np->queues[i]; 2236 spin_lock_bh(&queue->rx_lock); 2237 if (target < queue->rx_min_target) 2238 queue->rx_min_target = target; 2239 queue->rx_max_target = target; 2240 if (target < queue->rx_target) 2241 queue->rx_target = target; 2242 2243 xennet_alloc_rx_buffers(queue); 2244 2245 spin_unlock_bh(&queue->rx_lock); 2246 } 2247 return len; 2248} 2249 2250static ssize_t show_rxbuf_cur(struct device *dev, 2251 struct device_attribute *attr, char *buf) 2252{ 2253 struct net_device *netdev = to_net_dev(dev); 2254 struct netfront_info *info = netdev_priv(netdev); 2255 unsigned int num_queues = netdev->real_num_tx_queues; 2256 2257 if (num_queues) 2258 return sprintf(buf, "%u\n", info->queues[0].rx_target); 2259 else 2260 return sprintf(buf, "0\n"); 2261} 2262 2263static struct device_attribute xennet_attrs[] = { 2264 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), 2265 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), 2266 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), 2267}; 2268 2269static int xennet_sysfs_addif(struct net_device *netdev) 2270{ 2271 int i; 2272 int err; 2273 2274 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { 2275 err = device_create_file(&netdev->dev, 2276 &xennet_attrs[i]); 2277 if (err) 2278 goto fail; 2279 } 2280 return 0; 2281 2282 fail: 2283 while (--i >= 0) 2284 device_remove_file(&netdev->dev, &xennet_attrs[i]); 2285 return err; 2286} 2287 2288static void xennet_sysfs_delif(struct net_device *netdev) 2289{ 2290 int i; 2291 2292 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) 2293 device_remove_file(&netdev->dev, &xennet_attrs[i]); 2294} 2295 2296#endif /* CONFIG_SYSFS */ 2297 2298static int xennet_remove(struct xenbus_device *dev) 2299{ 2300 struct netfront_info *info = dev_get_drvdata(&dev->dev); 2301 unsigned int num_queues = info->netdev->real_num_tx_queues; 2302 struct netfront_queue *queue = NULL; 2303 unsigned int i = 0; 2304 2305 dev_dbg(&dev->dev, "%s\n", dev->nodename); 2306 2307 xennet_disconnect_backend(info); 2308 2309 xennet_sysfs_delif(info->netdev); 2310 2311 unregister_netdev(info->netdev); 2312 2313 for (i = 0; i < num_queues; ++i) { 2314 queue = &info->queues[i]; 2315 del_timer_sync(&queue->rx_refill_timer); 2316 } 2317 2318 if (num_queues) { 2319 kfree(info->queues); 2320 info->queues = NULL; 2321 } 2322 2323 free_percpu(info->stats); 2324 2325 free_netdev(info->netdev); 2326 2327 return 0; 2328} 2329 2330static const struct xenbus_device_id netfront_ids[] = { 2331 { "vif" }, 2332 { "" } 2333}; 2334 2335static struct xenbus_driver netfront_driver = { 2336 .ids = netfront_ids, 2337 .probe = netfront_probe, 2338 .remove = xennet_remove, 2339 .resume = netfront_resume, 2340 .otherend_changed = netback_changed, 2341}; 2342 2343static int __init netif_init(void) 2344{ 2345 if (!xen_domain()) 2346 return -ENODEV; 2347 2348 if (!xen_has_pv_nic_devices()) 2349 return -ENODEV; 2350 2351 pr_info("Initialising Xen virtual ethernet driver\n"); 2352 2353 /* Allow as many queues as there are CPUs, by default */ 2354 xennet_max_queues = num_online_cpus(); 2355 2356 return xenbus_register_frontend(&netfront_driver); 2357} 2358module_init(netif_init); 2359 2360 2361static void __exit netif_exit(void) 2362{ 2363 xenbus_unregister_driver(&netfront_driver); 2364} 2365module_exit(netif_exit); 2366 2367MODULE_DESCRIPTION("Xen virtual network device frontend"); 2368MODULE_LICENSE("GPL"); 2369MODULE_ALIAS("xen:vif"); 2370MODULE_ALIAS("xennet"); 2371