1/* A network driver using virtio. 2 * 3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation 4 * 5 * This program is free software; you can redistribute it and/or modify 6 * it under the terms of the GNU General Public License as published by 7 * the Free Software Foundation; either version 2 of the License, or 8 * (at your option) any later version. 9 * 10 * This program is distributed in the hope that it will be useful, 11 * but WITHOUT ANY WARRANTY; without even the implied warranty of 12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 13 * GNU General Public License for more details. 14 * 15 * You should have received a copy of the GNU General Public License 16 * along with this program; if not, write to the Free Software 17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 18 */ 19//#define DEBUG 20#include <linux/netdevice.h> 21#include <linux/etherdevice.h> 22#include <linux/ethtool.h> 23#include <linux/module.h> 24#include <linux/virtio.h> 25#include <linux/virtio_net.h> 26#include <linux/scatterlist.h> 27#include <linux/if_vlan.h> 28#include <linux/slab.h> 29 30static int napi_weight = 128; 31module_param(napi_weight, int, 0444); 32 33static int csum = 1, gso = 1; 34module_param(csum, bool, 0444); 35module_param(gso, bool, 0444); 36 37#define MAX_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN) 38#define GOOD_COPY_LEN 128 39 40#define VIRTNET_SEND_COMMAND_SG_MAX 2 41 42struct virtnet_info { 43 struct virtio_device *vdev; 44 struct virtqueue *rvq, *svq, *cvq; 45 struct net_device *dev; 46 struct napi_struct napi; 47 unsigned int status; 48 49 /* Number of input buffers, and max we've ever had. */ 50 unsigned int num, max; 51 52 /* I like... big packets and I cannot lie! */ 53 bool big_packets; 54 55 /* Host will merge rx buffers for big packets (shake it! shake it!) */ 56 bool mergeable_rx_bufs; 57 58 /* Work struct for refilling if we run low on memory. */ 59 struct delayed_work refill; 60 61 /* Chain pages by the private ptr. */ 62 struct page *pages; 63 64 /* fragments + linear part + virtio header */ 65 struct scatterlist rx_sg[MAX_SKB_FRAGS + 2]; 66 struct scatterlist tx_sg[MAX_SKB_FRAGS + 2]; 67}; 68 69struct skb_vnet_hdr { 70 union { 71 struct virtio_net_hdr hdr; 72 struct virtio_net_hdr_mrg_rxbuf mhdr; 73 }; 74 unsigned int num_sg; 75}; 76 77struct padded_vnet_hdr { 78 struct virtio_net_hdr hdr; 79 /* 80 * virtio_net_hdr should be in a separated sg buffer because of a 81 * QEMU bug, and data sg buffer shares same page with this header sg. 82 * This padding makes next sg 16 byte aligned after virtio_net_hdr. 83 */ 84 char padding[6]; 85}; 86 87static inline struct skb_vnet_hdr *skb_vnet_hdr(struct sk_buff *skb) 88{ 89 return (struct skb_vnet_hdr *)skb->cb; 90} 91 92/* 93 * private is used to chain pages for big packets, put the whole 94 * most recent used list in the beginning for reuse 95 */ 96static void give_pages(struct virtnet_info *vi, struct page *page) 97{ 98 struct page *end; 99 100 /* Find end of list, sew whole thing into vi->pages. */ 101 for (end = page; end->private; end = (struct page *)end->private); 102 end->private = (unsigned long)vi->pages; 103 vi->pages = page; 104} 105 106static struct page *get_a_page(struct virtnet_info *vi, gfp_t gfp_mask) 107{ 108 struct page *p = vi->pages; 109 110 if (p) { 111 vi->pages = (struct page *)p->private; 112 /* clear private here, it is used to chain pages */ 113 p->private = 0; 114 } else 115 p = alloc_page(gfp_mask); 116 return p; 117} 118 119static void skb_xmit_done(struct virtqueue *svq) 120{ 121 struct virtnet_info *vi = svq->vdev->priv; 122 123 /* Suppress further interrupts. */ 124 virtqueue_disable_cb(svq); 125 126 /* We were probably waiting for more output buffers. */ 127 netif_wake_queue(vi->dev); 128} 129 130static void set_skb_frag(struct sk_buff *skb, struct page *page, 131 unsigned int offset, unsigned int *len) 132{ 133 int i = skb_shinfo(skb)->nr_frags; 134 skb_frag_t *f; 135 136 f = &skb_shinfo(skb)->frags[i]; 137 f->size = min((unsigned)PAGE_SIZE - offset, *len); 138 f->page_offset = offset; 139 f->page = page; 140 141 skb->data_len += f->size; 142 skb->len += f->size; 143 skb_shinfo(skb)->nr_frags++; 144 *len -= f->size; 145} 146 147static struct sk_buff *page_to_skb(struct virtnet_info *vi, 148 struct page *page, unsigned int len) 149{ 150 struct sk_buff *skb; 151 struct skb_vnet_hdr *hdr; 152 unsigned int copy, hdr_len, offset; 153 char *p; 154 155 p = page_address(page); 156 157 /* copy small packet so we can reuse these pages for small data */ 158 skb = netdev_alloc_skb_ip_align(vi->dev, GOOD_COPY_LEN); 159 if (unlikely(!skb)) 160 return NULL; 161 162 hdr = skb_vnet_hdr(skb); 163 164 if (vi->mergeable_rx_bufs) { 165 hdr_len = sizeof hdr->mhdr; 166 offset = hdr_len; 167 } else { 168 hdr_len = sizeof hdr->hdr; 169 offset = sizeof(struct padded_vnet_hdr); 170 } 171 172 memcpy(hdr, p, hdr_len); 173 174 len -= hdr_len; 175 p += offset; 176 177 copy = len; 178 if (copy > skb_tailroom(skb)) 179 copy = skb_tailroom(skb); 180 memcpy(skb_put(skb, copy), p, copy); 181 182 len -= copy; 183 offset += copy; 184 185 while (len) { 186 set_skb_frag(skb, page, offset, &len); 187 page = (struct page *)page->private; 188 offset = 0; 189 } 190 191 if (page) 192 give_pages(vi, page); 193 194 return skb; 195} 196 197static int receive_mergeable(struct virtnet_info *vi, struct sk_buff *skb) 198{ 199 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); 200 struct page *page; 201 int num_buf, i, len; 202 203 num_buf = hdr->mhdr.num_buffers; 204 while (--num_buf) { 205 i = skb_shinfo(skb)->nr_frags; 206 if (i >= MAX_SKB_FRAGS) { 207 pr_debug("%s: packet too long\n", skb->dev->name); 208 skb->dev->stats.rx_length_errors++; 209 return -EINVAL; 210 } 211 212 page = virtqueue_get_buf(vi->rvq, &len); 213 if (!page) { 214 pr_debug("%s: rx error: %d buffers missing\n", 215 skb->dev->name, hdr->mhdr.num_buffers); 216 skb->dev->stats.rx_length_errors++; 217 return -EINVAL; 218 } 219 if (len > PAGE_SIZE) 220 len = PAGE_SIZE; 221 222 set_skb_frag(skb, page, 0, &len); 223 224 --vi->num; 225 } 226 return 0; 227} 228 229static void receive_buf(struct net_device *dev, void *buf, unsigned int len) 230{ 231 struct virtnet_info *vi = netdev_priv(dev); 232 struct sk_buff *skb; 233 struct page *page; 234 struct skb_vnet_hdr *hdr; 235 236 if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) { 237 pr_debug("%s: short packet %i\n", dev->name, len); 238 dev->stats.rx_length_errors++; 239 if (vi->mergeable_rx_bufs || vi->big_packets) 240 give_pages(vi, buf); 241 else 242 dev_kfree_skb(buf); 243 return; 244 } 245 246 if (!vi->mergeable_rx_bufs && !vi->big_packets) { 247 skb = buf; 248 len -= sizeof(struct virtio_net_hdr); 249 skb_trim(skb, len); 250 } else { 251 page = buf; 252 skb = page_to_skb(vi, page, len); 253 if (unlikely(!skb)) { 254 dev->stats.rx_dropped++; 255 give_pages(vi, page); 256 return; 257 } 258 if (vi->mergeable_rx_bufs) 259 if (receive_mergeable(vi, skb)) { 260 dev_kfree_skb(skb); 261 return; 262 } 263 } 264 265 hdr = skb_vnet_hdr(skb); 266 skb->truesize += skb->data_len; 267 dev->stats.rx_bytes += skb->len; 268 dev->stats.rx_packets++; 269 270 if (hdr->hdr.flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) { 271 pr_debug("Needs csum!\n"); 272 if (!skb_partial_csum_set(skb, 273 hdr->hdr.csum_start, 274 hdr->hdr.csum_offset)) 275 goto frame_err; 276 } 277 278 skb->protocol = eth_type_trans(skb, dev); 279 pr_debug("Receiving skb proto 0x%04x len %i type %i\n", 280 ntohs(skb->protocol), skb->len, skb->pkt_type); 281 282 if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) { 283 pr_debug("GSO!\n"); 284 switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) { 285 case VIRTIO_NET_HDR_GSO_TCPV4: 286 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 287 break; 288 case VIRTIO_NET_HDR_GSO_UDP: 289 skb_shinfo(skb)->gso_type = SKB_GSO_UDP; 290 break; 291 case VIRTIO_NET_HDR_GSO_TCPV6: 292 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6; 293 break; 294 default: 295 if (net_ratelimit()) 296 printk(KERN_WARNING "%s: bad gso type %u.\n", 297 dev->name, hdr->hdr.gso_type); 298 goto frame_err; 299 } 300 301 if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN) 302 skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN; 303 304 skb_shinfo(skb)->gso_size = hdr->hdr.gso_size; 305 if (skb_shinfo(skb)->gso_size == 0) { 306 if (net_ratelimit()) 307 printk(KERN_WARNING "%s: zero gso size.\n", 308 dev->name); 309 goto frame_err; 310 } 311 312 /* Header must be checked, and gso_segs computed. */ 313 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 314 skb_shinfo(skb)->gso_segs = 0; 315 } 316 317 netif_receive_skb(skb); 318 return; 319 320frame_err: 321 dev->stats.rx_frame_errors++; 322 dev_kfree_skb(skb); 323} 324 325static int add_recvbuf_small(struct virtnet_info *vi, gfp_t gfp) 326{ 327 struct sk_buff *skb; 328 struct skb_vnet_hdr *hdr; 329 int err; 330 331 skb = netdev_alloc_skb_ip_align(vi->dev, MAX_PACKET_LEN); 332 if (unlikely(!skb)) 333 return -ENOMEM; 334 335 skb_put(skb, MAX_PACKET_LEN); 336 337 hdr = skb_vnet_hdr(skb); 338 sg_set_buf(vi->rx_sg, &hdr->hdr, sizeof hdr->hdr); 339 340 skb_to_sgvec(skb, vi->rx_sg + 1, 0, skb->len); 341 342 err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 2, skb, gfp); 343 if (err < 0) 344 dev_kfree_skb(skb); 345 346 return err; 347} 348 349static int add_recvbuf_big(struct virtnet_info *vi, gfp_t gfp) 350{ 351 struct page *first, *list = NULL; 352 char *p; 353 int i, err, offset; 354 355 /* page in vi->rx_sg[MAX_SKB_FRAGS + 1] is list tail */ 356 for (i = MAX_SKB_FRAGS + 1; i > 1; --i) { 357 first = get_a_page(vi, gfp); 358 if (!first) { 359 if (list) 360 give_pages(vi, list); 361 return -ENOMEM; 362 } 363 sg_set_buf(&vi->rx_sg[i], page_address(first), PAGE_SIZE); 364 365 /* chain new page in list head to match sg */ 366 first->private = (unsigned long)list; 367 list = first; 368 } 369 370 first = get_a_page(vi, gfp); 371 if (!first) { 372 give_pages(vi, list); 373 return -ENOMEM; 374 } 375 p = page_address(first); 376 377 /* vi->rx_sg[0], vi->rx_sg[1] share the same page */ 378 /* a separated vi->rx_sg[0] for virtio_net_hdr only due to QEMU bug */ 379 sg_set_buf(&vi->rx_sg[0], p, sizeof(struct virtio_net_hdr)); 380 381 /* vi->rx_sg[1] for data packet, from offset */ 382 offset = sizeof(struct padded_vnet_hdr); 383 sg_set_buf(&vi->rx_sg[1], p + offset, PAGE_SIZE - offset); 384 385 /* chain first in list head */ 386 first->private = (unsigned long)list; 387 err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, MAX_SKB_FRAGS + 2, 388 first, gfp); 389 if (err < 0) 390 give_pages(vi, first); 391 392 return err; 393} 394 395static int add_recvbuf_mergeable(struct virtnet_info *vi, gfp_t gfp) 396{ 397 struct page *page; 398 int err; 399 400 page = get_a_page(vi, gfp); 401 if (!page) 402 return -ENOMEM; 403 404 sg_init_one(vi->rx_sg, page_address(page), PAGE_SIZE); 405 406 err = virtqueue_add_buf_gfp(vi->rvq, vi->rx_sg, 0, 1, page, gfp); 407 if (err < 0) 408 give_pages(vi, page); 409 410 return err; 411} 412 413/* Returns false if we couldn't fill entirely (OOM). */ 414static bool try_fill_recv(struct virtnet_info *vi, gfp_t gfp) 415{ 416 int err; 417 bool oom; 418 419 do { 420 if (vi->mergeable_rx_bufs) 421 err = add_recvbuf_mergeable(vi, gfp); 422 else if (vi->big_packets) 423 err = add_recvbuf_big(vi, gfp); 424 else 425 err = add_recvbuf_small(vi, gfp); 426 427 oom = err == -ENOMEM; 428 if (err < 0) 429 break; 430 ++vi->num; 431 } while (err > 0); 432 if (unlikely(vi->num > vi->max)) 433 vi->max = vi->num; 434 virtqueue_kick(vi->rvq); 435 return !oom; 436} 437 438static void skb_recv_done(struct virtqueue *rvq) 439{ 440 struct virtnet_info *vi = rvq->vdev->priv; 441 /* Schedule NAPI, Suppress further interrupts if successful. */ 442 if (napi_schedule_prep(&vi->napi)) { 443 virtqueue_disable_cb(rvq); 444 __napi_schedule(&vi->napi); 445 } 446} 447 448static void virtnet_napi_enable(struct virtnet_info *vi) 449{ 450 napi_enable(&vi->napi); 451 452 /* If all buffers were filled by other side before we napi_enabled, we 453 * won't get another interrupt, so process any outstanding packets 454 * now. virtnet_poll wants re-enable the queue, so we disable here. 455 * We synchronize against interrupts via NAPI_STATE_SCHED */ 456 if (napi_schedule_prep(&vi->napi)) { 457 virtqueue_disable_cb(vi->rvq); 458 __napi_schedule(&vi->napi); 459 } 460} 461 462static void refill_work(struct work_struct *work) 463{ 464 struct virtnet_info *vi; 465 bool still_empty; 466 467 vi = container_of(work, struct virtnet_info, refill.work); 468 napi_disable(&vi->napi); 469 still_empty = !try_fill_recv(vi, GFP_KERNEL); 470 virtnet_napi_enable(vi); 471 472 /* In theory, this can happen: if we don't get any buffers in 473 * we will *never* try to fill again. */ 474 if (still_empty) 475 schedule_delayed_work(&vi->refill, HZ/2); 476} 477 478static int virtnet_poll(struct napi_struct *napi, int budget) 479{ 480 struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi); 481 void *buf; 482 unsigned int len, received = 0; 483 484again: 485 while (received < budget && 486 (buf = virtqueue_get_buf(vi->rvq, &len)) != NULL) { 487 receive_buf(vi->dev, buf, len); 488 --vi->num; 489 received++; 490 } 491 492 if (vi->num < vi->max / 2) { 493 if (!try_fill_recv(vi, GFP_ATOMIC)) 494 schedule_delayed_work(&vi->refill, 0); 495 } 496 497 /* Out of packets? */ 498 if (received < budget) { 499 napi_complete(napi); 500 if (unlikely(!virtqueue_enable_cb(vi->rvq)) && 501 napi_schedule_prep(napi)) { 502 virtqueue_disable_cb(vi->rvq); 503 __napi_schedule(napi); 504 goto again; 505 } 506 } 507 508 return received; 509} 510 511static unsigned int free_old_xmit_skbs(struct virtnet_info *vi) 512{ 513 struct sk_buff *skb; 514 unsigned int len, tot_sgs = 0; 515 516 while ((skb = virtqueue_get_buf(vi->svq, &len)) != NULL) { 517 pr_debug("Sent skb %p\n", skb); 518 vi->dev->stats.tx_bytes += skb->len; 519 vi->dev->stats.tx_packets++; 520 tot_sgs += skb_vnet_hdr(skb)->num_sg; 521 dev_kfree_skb_any(skb); 522 } 523 return tot_sgs; 524} 525 526static int xmit_skb(struct virtnet_info *vi, struct sk_buff *skb) 527{ 528 struct skb_vnet_hdr *hdr = skb_vnet_hdr(skb); 529 const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest; 530 531 pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest); 532 533 if (skb->ip_summed == CHECKSUM_PARTIAL) { 534 hdr->hdr.flags = VIRTIO_NET_HDR_F_NEEDS_CSUM; 535 hdr->hdr.csum_start = skb->csum_start - skb_headroom(skb); 536 hdr->hdr.csum_offset = skb->csum_offset; 537 } else { 538 hdr->hdr.flags = 0; 539 hdr->hdr.csum_offset = hdr->hdr.csum_start = 0; 540 } 541 542 if (skb_is_gso(skb)) { 543 hdr->hdr.hdr_len = skb_headlen(skb); 544 hdr->hdr.gso_size = skb_shinfo(skb)->gso_size; 545 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4) 546 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV4; 547 else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) 548 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_TCPV6; 549 else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP) 550 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_UDP; 551 else 552 BUG(); 553 if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN) 554 hdr->hdr.gso_type |= VIRTIO_NET_HDR_GSO_ECN; 555 } else { 556 hdr->hdr.gso_type = VIRTIO_NET_HDR_GSO_NONE; 557 hdr->hdr.gso_size = hdr->hdr.hdr_len = 0; 558 } 559 560 hdr->mhdr.num_buffers = 0; 561 562 /* Encode metadata header at front. */ 563 if (vi->mergeable_rx_bufs) 564 sg_set_buf(vi->tx_sg, &hdr->mhdr, sizeof hdr->mhdr); 565 else 566 sg_set_buf(vi->tx_sg, &hdr->hdr, sizeof hdr->hdr); 567 568 hdr->num_sg = skb_to_sgvec(skb, vi->tx_sg + 1, 0, skb->len) + 1; 569 return virtqueue_add_buf(vi->svq, vi->tx_sg, hdr->num_sg, 570 0, skb); 571} 572 573static netdev_tx_t start_xmit(struct sk_buff *skb, struct net_device *dev) 574{ 575 struct virtnet_info *vi = netdev_priv(dev); 576 int capacity; 577 578 /* Free up any pending old buffers before queueing new ones. */ 579 free_old_xmit_skbs(vi); 580 581 /* Try to transmit */ 582 capacity = xmit_skb(vi, skb); 583 584 /* This can happen with OOM and indirect buffers. */ 585 if (unlikely(capacity < 0)) { 586 if (net_ratelimit()) { 587 if (likely(capacity == -ENOMEM)) { 588 dev_warn(&dev->dev, 589 "TX queue failure: out of memory\n"); 590 } else { 591 dev->stats.tx_fifo_errors++; 592 dev_warn(&dev->dev, 593 "Unexpected TX queue failure: %d\n", 594 capacity); 595 } 596 } 597 dev->stats.tx_dropped++; 598 kfree_skb(skb); 599 return NETDEV_TX_OK; 600 } 601 virtqueue_kick(vi->svq); 602 603 /* Don't wait up for transmitted skbs to be freed. */ 604 skb_orphan(skb); 605 nf_reset(skb); 606 607 /* Apparently nice girls don't return TX_BUSY; stop the queue 608 * before it gets out of hand. Naturally, this wastes entries. */ 609 if (capacity < 2+MAX_SKB_FRAGS) { 610 netif_stop_queue(dev); 611 if (unlikely(!virtqueue_enable_cb(vi->svq))) { 612 /* More just got used, free them then recheck. */ 613 capacity += free_old_xmit_skbs(vi); 614 if (capacity >= 2+MAX_SKB_FRAGS) { 615 netif_start_queue(dev); 616 virtqueue_disable_cb(vi->svq); 617 } 618 } 619 } 620 621 return NETDEV_TX_OK; 622} 623 624static int virtnet_set_mac_address(struct net_device *dev, void *p) 625{ 626 struct virtnet_info *vi = netdev_priv(dev); 627 struct virtio_device *vdev = vi->vdev; 628 int ret; 629 630 ret = eth_mac_addr(dev, p); 631 if (ret) 632 return ret; 633 634 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) 635 vdev->config->set(vdev, offsetof(struct virtio_net_config, mac), 636 dev->dev_addr, dev->addr_len); 637 638 return 0; 639} 640 641#ifdef CONFIG_NET_POLL_CONTROLLER 642static void virtnet_netpoll(struct net_device *dev) 643{ 644 struct virtnet_info *vi = netdev_priv(dev); 645 646 napi_schedule(&vi->napi); 647} 648#endif 649 650static int virtnet_open(struct net_device *dev) 651{ 652 struct virtnet_info *vi = netdev_priv(dev); 653 654 virtnet_napi_enable(vi); 655 return 0; 656} 657 658/* 659 * Send command via the control virtqueue and check status. Commands 660 * supported by the hypervisor, as indicated by feature bits, should 661 * never fail unless improperly formated. 662 */ 663static bool virtnet_send_command(struct virtnet_info *vi, u8 class, u8 cmd, 664 struct scatterlist *data, int out, int in) 665{ 666 struct scatterlist *s, sg[VIRTNET_SEND_COMMAND_SG_MAX + 2]; 667 struct virtio_net_ctrl_hdr ctrl; 668 virtio_net_ctrl_ack status = ~0; 669 unsigned int tmp; 670 int i; 671 672 /* Caller should know better */ 673 BUG_ON(!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) || 674 (out + in > VIRTNET_SEND_COMMAND_SG_MAX)); 675 676 out++; /* Add header */ 677 in++; /* Add return status */ 678 679 ctrl.class = class; 680 ctrl.cmd = cmd; 681 682 sg_init_table(sg, out + in); 683 684 sg_set_buf(&sg[0], &ctrl, sizeof(ctrl)); 685 for_each_sg(data, s, out + in - 2, i) 686 sg_set_buf(&sg[i + 1], sg_virt(s), s->length); 687 sg_set_buf(&sg[out + in - 1], &status, sizeof(status)); 688 689 BUG_ON(virtqueue_add_buf(vi->cvq, sg, out, in, vi) < 0); 690 691 virtqueue_kick(vi->cvq); 692 693 /* 694 * Spin for a response, the kick causes an ioport write, trapping 695 * into the hypervisor, so the request should be handled immediately. 696 */ 697 while (!virtqueue_get_buf(vi->cvq, &tmp)) 698 cpu_relax(); 699 700 return status == VIRTIO_NET_OK; 701} 702 703static int virtnet_close(struct net_device *dev) 704{ 705 struct virtnet_info *vi = netdev_priv(dev); 706 707 napi_disable(&vi->napi); 708 709 return 0; 710} 711 712static void virtnet_get_drvinfo(struct net_device *dev, 713 struct ethtool_drvinfo *drvinfo) 714{ 715 struct virtnet_info *vi = netdev_priv(dev); 716 struct virtio_device *vdev = vi->vdev; 717 718 strncpy(drvinfo->driver, KBUILD_MODNAME, ARRAY_SIZE(drvinfo->driver)); 719 strncpy(drvinfo->version, "N/A", ARRAY_SIZE(drvinfo->version)); 720 strncpy(drvinfo->fw_version, "N/A", ARRAY_SIZE(drvinfo->fw_version)); 721 strncpy(drvinfo->bus_info, dev_name(&vdev->dev), 722 ARRAY_SIZE(drvinfo->bus_info)); 723} 724 725static int virtnet_set_tx_csum(struct net_device *dev, u32 data) 726{ 727 struct virtnet_info *vi = netdev_priv(dev); 728 struct virtio_device *vdev = vi->vdev; 729 730 if (data && !virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) 731 return -ENOSYS; 732 733 return ethtool_op_set_tx_hw_csum(dev, data); 734} 735 736static void virtnet_set_rx_mode(struct net_device *dev) 737{ 738 struct virtnet_info *vi = netdev_priv(dev); 739 struct scatterlist sg[2]; 740 u8 promisc, allmulti; 741 struct virtio_net_ctrl_mac *mac_data; 742 struct netdev_hw_addr *ha; 743 int uc_count; 744 int mc_count; 745 void *buf; 746 int i; 747 748 /* We can't dynamicaly set ndo_set_rx_mode, so return gracefully */ 749 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_RX)) 750 return; 751 752 promisc = ((dev->flags & IFF_PROMISC) != 0); 753 allmulti = ((dev->flags & IFF_ALLMULTI) != 0); 754 755 sg_init_one(sg, &promisc, sizeof(promisc)); 756 757 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, 758 VIRTIO_NET_CTRL_RX_PROMISC, 759 sg, 1, 0)) 760 dev_warn(&dev->dev, "Failed to %sable promisc mode.\n", 761 promisc ? "en" : "dis"); 762 763 sg_init_one(sg, &allmulti, sizeof(allmulti)); 764 765 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_RX, 766 VIRTIO_NET_CTRL_RX_ALLMULTI, 767 sg, 1, 0)) 768 dev_warn(&dev->dev, "Failed to %sable allmulti mode.\n", 769 allmulti ? "en" : "dis"); 770 771 uc_count = netdev_uc_count(dev); 772 mc_count = netdev_mc_count(dev); 773 /* MAC filter - use one buffer for both lists */ 774 buf = kzalloc(((uc_count + mc_count) * ETH_ALEN) + 775 (2 * sizeof(mac_data->entries)), GFP_ATOMIC); 776 mac_data = buf; 777 if (!buf) { 778 dev_warn(&dev->dev, "No memory for MAC address buffer\n"); 779 return; 780 } 781 782 sg_init_table(sg, 2); 783 784 /* Store the unicast list and count in the front of the buffer */ 785 mac_data->entries = uc_count; 786 i = 0; 787 netdev_for_each_uc_addr(ha, dev) 788 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); 789 790 sg_set_buf(&sg[0], mac_data, 791 sizeof(mac_data->entries) + (uc_count * ETH_ALEN)); 792 793 /* multicast list and count fill the end */ 794 mac_data = (void *)&mac_data->macs[uc_count][0]; 795 796 mac_data->entries = mc_count; 797 i = 0; 798 netdev_for_each_mc_addr(ha, dev) 799 memcpy(&mac_data->macs[i++][0], ha->addr, ETH_ALEN); 800 801 sg_set_buf(&sg[1], mac_data, 802 sizeof(mac_data->entries) + (mc_count * ETH_ALEN)); 803 804 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_MAC, 805 VIRTIO_NET_CTRL_MAC_TABLE_SET, 806 sg, 2, 0)) 807 dev_warn(&dev->dev, "Failed to set MAC fitler table.\n"); 808 809 kfree(buf); 810} 811 812static void virtnet_vlan_rx_add_vid(struct net_device *dev, u16 vid) 813{ 814 struct virtnet_info *vi = netdev_priv(dev); 815 struct scatterlist sg; 816 817 sg_init_one(&sg, &vid, sizeof(vid)); 818 819 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, 820 VIRTIO_NET_CTRL_VLAN_ADD, &sg, 1, 0)) 821 dev_warn(&dev->dev, "Failed to add VLAN ID %d.\n", vid); 822} 823 824static void virtnet_vlan_rx_kill_vid(struct net_device *dev, u16 vid) 825{ 826 struct virtnet_info *vi = netdev_priv(dev); 827 struct scatterlist sg; 828 829 sg_init_one(&sg, &vid, sizeof(vid)); 830 831 if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_VLAN, 832 VIRTIO_NET_CTRL_VLAN_DEL, &sg, 1, 0)) 833 dev_warn(&dev->dev, "Failed to kill VLAN ID %d.\n", vid); 834} 835 836static const struct ethtool_ops virtnet_ethtool_ops = { 837 .get_drvinfo = virtnet_get_drvinfo, 838 .set_tx_csum = virtnet_set_tx_csum, 839 .set_sg = ethtool_op_set_sg, 840 .set_tso = ethtool_op_set_tso, 841 .set_ufo = ethtool_op_set_ufo, 842 .get_link = ethtool_op_get_link, 843}; 844 845#define MIN_MTU 68 846#define MAX_MTU 65535 847 848static int virtnet_change_mtu(struct net_device *dev, int new_mtu) 849{ 850 if (new_mtu < MIN_MTU || new_mtu > MAX_MTU) 851 return -EINVAL; 852 dev->mtu = new_mtu; 853 return 0; 854} 855 856static const struct net_device_ops virtnet_netdev = { 857 .ndo_open = virtnet_open, 858 .ndo_stop = virtnet_close, 859 .ndo_start_xmit = start_xmit, 860 .ndo_validate_addr = eth_validate_addr, 861 .ndo_set_mac_address = virtnet_set_mac_address, 862 .ndo_set_rx_mode = virtnet_set_rx_mode, 863 .ndo_change_mtu = virtnet_change_mtu, 864 .ndo_vlan_rx_add_vid = virtnet_vlan_rx_add_vid, 865 .ndo_vlan_rx_kill_vid = virtnet_vlan_rx_kill_vid, 866#ifdef CONFIG_NET_POLL_CONTROLLER 867 .ndo_poll_controller = virtnet_netpoll, 868#endif 869}; 870 871static void virtnet_update_status(struct virtnet_info *vi) 872{ 873 u16 v; 874 875 if (!virtio_has_feature(vi->vdev, VIRTIO_NET_F_STATUS)) 876 return; 877 878 vi->vdev->config->get(vi->vdev, 879 offsetof(struct virtio_net_config, status), 880 &v, sizeof(v)); 881 882 /* Ignore unknown (future) status bits */ 883 v &= VIRTIO_NET_S_LINK_UP; 884 885 if (vi->status == v) 886 return; 887 888 vi->status = v; 889 890 if (vi->status & VIRTIO_NET_S_LINK_UP) { 891 netif_carrier_on(vi->dev); 892 netif_wake_queue(vi->dev); 893 } else { 894 netif_carrier_off(vi->dev); 895 netif_stop_queue(vi->dev); 896 } 897} 898 899static void virtnet_config_changed(struct virtio_device *vdev) 900{ 901 struct virtnet_info *vi = vdev->priv; 902 903 virtnet_update_status(vi); 904} 905 906static int virtnet_probe(struct virtio_device *vdev) 907{ 908 int err; 909 struct net_device *dev; 910 struct virtnet_info *vi; 911 struct virtqueue *vqs[3]; 912 vq_callback_t *callbacks[] = { skb_recv_done, skb_xmit_done, NULL}; 913 const char *names[] = { "input", "output", "control" }; 914 int nvqs; 915 916 /* Allocate ourselves a network device with room for our info */ 917 dev = alloc_etherdev(sizeof(struct virtnet_info)); 918 if (!dev) 919 return -ENOMEM; 920 921 /* Set up network device as normal. */ 922 dev->netdev_ops = &virtnet_netdev; 923 dev->features = NETIF_F_HIGHDMA; 924 SET_ETHTOOL_OPS(dev, &virtnet_ethtool_ops); 925 SET_NETDEV_DEV(dev, &vdev->dev); 926 927 /* Do we support "hardware" checksums? */ 928 if (csum && virtio_has_feature(vdev, VIRTIO_NET_F_CSUM)) { 929 /* This opens up the world of extra features. */ 930 dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST; 931 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_GSO)) { 932 dev->features |= NETIF_F_TSO | NETIF_F_UFO 933 | NETIF_F_TSO_ECN | NETIF_F_TSO6; 934 } 935 /* Individual feature bits: what can host handle? */ 936 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO4)) 937 dev->features |= NETIF_F_TSO; 938 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_TSO6)) 939 dev->features |= NETIF_F_TSO6; 940 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_ECN)) 941 dev->features |= NETIF_F_TSO_ECN; 942 if (gso && virtio_has_feature(vdev, VIRTIO_NET_F_HOST_UFO)) 943 dev->features |= NETIF_F_UFO; 944 } 945 946 /* Configuration may specify what MAC to use. Otherwise random. */ 947 if (virtio_has_feature(vdev, VIRTIO_NET_F_MAC)) { 948 vdev->config->get(vdev, 949 offsetof(struct virtio_net_config, mac), 950 dev->dev_addr, dev->addr_len); 951 } else 952 random_ether_addr(dev->dev_addr); 953 954 /* Set up our device-specific information */ 955 vi = netdev_priv(dev); 956 netif_napi_add(dev, &vi->napi, virtnet_poll, napi_weight); 957 vi->dev = dev; 958 vi->vdev = vdev; 959 vdev->priv = vi; 960 vi->pages = NULL; 961 INIT_DELAYED_WORK(&vi->refill, refill_work); 962 sg_init_table(vi->rx_sg, ARRAY_SIZE(vi->rx_sg)); 963 sg_init_table(vi->tx_sg, ARRAY_SIZE(vi->tx_sg)); 964 965 /* If we can receive ANY GSO packets, we must allocate large ones. */ 966 if (virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO4) || 967 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_TSO6) || 968 virtio_has_feature(vdev, VIRTIO_NET_F_GUEST_ECN)) 969 vi->big_packets = true; 970 971 if (virtio_has_feature(vdev, VIRTIO_NET_F_MRG_RXBUF)) 972 vi->mergeable_rx_bufs = true; 973 974 /* We expect two virtqueues, receive then send, 975 * and optionally control. */ 976 nvqs = virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ) ? 3 : 2; 977 978 err = vdev->config->find_vqs(vdev, nvqs, vqs, callbacks, names); 979 if (err) 980 goto free; 981 982 vi->rvq = vqs[0]; 983 vi->svq = vqs[1]; 984 985 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VQ)) { 986 vi->cvq = vqs[2]; 987 988 if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_CTRL_VLAN)) 989 dev->features |= NETIF_F_HW_VLAN_FILTER; 990 } 991 992 err = register_netdev(dev); 993 if (err) { 994 pr_debug("virtio_net: registering device failed\n"); 995 goto free_vqs; 996 } 997 998 /* Last of all, set up some receive buffers. */ 999 try_fill_recv(vi, GFP_KERNEL); 1000 1001 /* If we didn't even get one input buffer, we're useless. */ 1002 if (vi->num == 0) { 1003 err = -ENOMEM; 1004 goto unregister; 1005 } 1006 1007 vi->status = VIRTIO_NET_S_LINK_UP; 1008 virtnet_update_status(vi); 1009 netif_carrier_on(dev); 1010 1011 pr_debug("virtnet: registered device %s\n", dev->name); 1012 return 0; 1013 1014unregister: 1015 unregister_netdev(dev); 1016 cancel_delayed_work_sync(&vi->refill); 1017free_vqs: 1018 vdev->config->del_vqs(vdev); 1019free: 1020 free_netdev(dev); 1021 return err; 1022} 1023 1024static void free_unused_bufs(struct virtnet_info *vi) 1025{ 1026 void *buf; 1027 while (1) { 1028 buf = virtqueue_detach_unused_buf(vi->svq); 1029 if (!buf) 1030 break; 1031 dev_kfree_skb(buf); 1032 } 1033 while (1) { 1034 buf = virtqueue_detach_unused_buf(vi->rvq); 1035 if (!buf) 1036 break; 1037 if (vi->mergeable_rx_bufs || vi->big_packets) 1038 give_pages(vi, buf); 1039 else 1040 dev_kfree_skb(buf); 1041 --vi->num; 1042 } 1043 BUG_ON(vi->num != 0); 1044} 1045 1046static void __devexit virtnet_remove(struct virtio_device *vdev) 1047{ 1048 struct virtnet_info *vi = vdev->priv; 1049 1050 /* Stop all the virtqueues. */ 1051 vdev->config->reset(vdev); 1052 1053 1054 unregister_netdev(vi->dev); 1055 cancel_delayed_work_sync(&vi->refill); 1056 1057 /* Free unused buffers in both send and recv, if any. */ 1058 free_unused_bufs(vi); 1059 1060 vdev->config->del_vqs(vi->vdev); 1061 1062 while (vi->pages) 1063 __free_pages(get_a_page(vi, GFP_KERNEL), 0); 1064 1065 free_netdev(vi->dev); 1066} 1067 1068static struct virtio_device_id id_table[] = { 1069 { VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID }, 1070 { 0 }, 1071}; 1072 1073static unsigned int features[] = { 1074 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, 1075 VIRTIO_NET_F_GSO, VIRTIO_NET_F_MAC, 1076 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, 1077 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, 1078 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, 1079 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, 1080 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, 1081}; 1082 1083static struct virtio_driver virtio_net_driver = { 1084 .feature_table = features, 1085 .feature_table_size = ARRAY_SIZE(features), 1086 .driver.name = KBUILD_MODNAME, 1087 .driver.owner = THIS_MODULE, 1088 .id_table = id_table, 1089 .probe = virtnet_probe, 1090 .remove = __devexit_p(virtnet_remove), 1091 .config_changed = virtnet_config_changed, 1092}; 1093 1094static int __init init(void) 1095{ 1096 return register_virtio_driver(&virtio_net_driver); 1097} 1098 1099static void __exit fini(void) 1100{ 1101 unregister_virtio_driver(&virtio_net_driver); 1102} 1103module_init(init); 1104module_exit(fini); 1105 1106MODULE_DEVICE_TABLE(virtio, id_table); 1107MODULE_DESCRIPTION("Virtio network driver"); 1108MODULE_LICENSE("GPL"); 1109