drivers/staging/hv/netvsc_drv.c

   1 /*
   2  * Copyright (c) 2009, Microsoft Corporation.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms and conditions of the GNU General Public License,
   6  * version 2, as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope it will be useful, but WITHOUT
   9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  11  * more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along with
  14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/init.h>
  24 #include <linux/module.h>
  25 #include <linux/highmem.h>
  26 #include <linux/device.h>
  27 #include <linux/io.h>
  28 #include <linux/delay.h>
  29 #include <linux/netdevice.h>
  30 #include <linux/inetdevice.h>
  31 #include <linux/etherdevice.h>
  32 #include <linux/skbuff.h>
  33 #include <linux/in.h>
  34 #include <linux/slab.h>
  35 #include <linux/dmi.h>
  36 #include <linux/pci.h>
  37 #include <net/arp.h>
  38 #include <net/route.h>
  39 #include <net/sock.h>
  40 #include <net/pkt_sched.h>
  41 #include "hv_api.h"
  42 #include "logging.h"
  43 #include "version_info.h"
  44 #include "vmbus.h"
  45 #include "netvsc_api.h"
  46
  47 struct net_device_context {
  48         /* point back to our device context */
  49         struct hv_device *device_ctx;
  50         unsigned long avail;
  51         struct work_struct work;
  52 };
  53
  54
  55 #define PACKET_PAGES_LOWATER  8
  56 /* Need this many pages to handle worst case fragmented packet */
  57 #define PACKET_PAGES_HIWATER  (MAX_SKB_FRAGS + 2)
  58
  59 static int ring_size = 128;
  60 module_param(ring_size, int, S_IRUGO);
  61 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  62
  63 /* The one and only one */
  64 static struct  netvsc_driver g_netvsc_drv;
  65
  66 /* no-op so the netdev core doesn't return -EINVAL when modifying the the
  67  * multicast address list in SIOCADDMULTI. hv is setup to get all multicast
  68  * when it calls RndisFilterOnOpen() */
  69 static void netvsc_set_multicast_list(struct net_device *net)
  70 {
  71 }
  72
  73 static int netvsc_open(struct net_device *net)
  74 {
  75         struct net_device_context *net_device_ctx = netdev_priv(net);
  76         struct hv_device *device_obj = net_device_ctx->device_ctx;
  77         int ret = 0;
  78
  79         if (netif_carrier_ok(net)) {
  80                 /* Open up the device */
  81                 ret = rndis_filter_open(device_obj);
  82                 if (ret != 0) {
  83                         netdev_err(net, "unable to open device (ret %d).\n",
  84                                    ret);
  85                         return ret;
  86                 }
  87
  88                 netif_start_queue(net);
  89         } else {
  90                 netdev_err(net, "unable to open device...link is down.\n");
  91         }
  92
  93         return ret;
  94 }
  95
  96 static int netvsc_close(struct net_device *net)
  97 {
  98         struct net_device_context *net_device_ctx = netdev_priv(net);
  99         struct hv_device *device_obj = net_device_ctx->device_ctx;
 100         int ret;
 101
 102         netif_stop_queue(net);
 103
 104         ret = rndis_filter_close(device_obj);
 105         if (ret != 0)
 106                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 107
 108         return ret;
 109 }
 110
 111 static void netvsc_xmit_completion(void *context)
 112 {
 113         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 114         struct sk_buff *skb = (struct sk_buff *)
 115                 (unsigned long)packet->completion.send.send_completion_tid;
 116
 117         kfree(packet);
 118
 119         if (skb) {
 120                 struct net_device *net = skb->dev;
 121                 struct net_device_context *net_device_ctx = netdev_priv(net);
 122                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
 123
 124                 dev_kfree_skb_any(skb);
 125
 126                 net_device_ctx->avail += num_pages;
 127                 if (net_device_ctx->avail >= PACKET_PAGES_HIWATER)
 128                         netif_wake_queue(net);
 129         }
 130 }
 131
 132 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 133 {
 134         struct net_device_context *net_device_ctx = netdev_priv(net);
 135         struct hv_driver *drv =
 136             drv_to_hv_drv(net_device_ctx->device_ctx->device.driver);
 137         struct netvsc_driver *net_drv_obj = drv->priv;
 138         struct hv_netvsc_packet *packet;
 139         int ret;
 140         unsigned int i, num_pages;
 141
 142         /* Add 1 for skb->data and additional one for RNDIS */
 143         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
 144         if (num_pages > net_device_ctx->avail)
 145                 return NETDEV_TX_BUSY;
 146
 147         /* Allocate a netvsc packet based on # of frags. */
 148         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 149                          (num_pages * sizeof(struct hv_page_buffer)) +
 150                          net_drv_obj->req_ext_size, GFP_ATOMIC);
 151         if (!packet) {
 152                 /* out of memory, silently drop packet */
 153                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 154
 155                 dev_kfree_skb(skb);
 156                 net->stats.tx_dropped++;
 157                 return NETDEV_TX_OK;
 158         }
 159
 160         packet->extension = (void *)(unsigned long)packet +
 161                                 sizeof(struct hv_netvsc_packet) +
 162                                     (num_pages * sizeof(struct hv_page_buffer));
 163
 164         /* Setup the rndis header */
 165         packet->page_buf_cnt = num_pages;
 166
 167         /* TODO: Flush all write buffers/ memory fence ??? */
 168         /* wmb(); */
 169
 170         /* Initialize it from the skb */
 171         packet->total_data_buflen       = skb->len;
 172
 173         /* Start filling in the page buffers starting after RNDIS buffer. */
 174         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 175         packet->page_buf[1].offset
 176                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 177         packet->page_buf[1].len = skb_headlen(skb);
 178
 179         /* Additional fragments are after SKB data */
 180         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 181                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 182
 183                 packet->page_buf[i+2].pfn = page_to_pfn(f->page);
 184                 packet->page_buf[i+2].offset = f->page_offset;
 185                 packet->page_buf[i+2].len = f->size;
 186         }
 187
 188         /* Set the completion routine */
 189         packet->completion.send.send_completion = netvsc_xmit_completion;
 190         packet->completion.send.send_completion_ctx = packet;
 191         packet->completion.send.send_completion_tid = (unsigned long)skb;
 192
 193         ret = net_drv_obj->send(net_device_ctx->device_ctx,
 194                                   packet);
 195         if (ret == 0) {
 196                 net->stats.tx_bytes += skb->len;
 197                 net->stats.tx_packets++;
 198
 199                 net_device_ctx->avail -= num_pages;
 200                 if (net_device_ctx->avail < PACKET_PAGES_LOWATER)
 201                         netif_stop_queue(net);
 202         } else {
 203                 /* we are shutting down or bus overloaded, just drop packet */
 204                 net->stats.tx_dropped++;
 205                 netvsc_xmit_completion(packet);
 206         }
 207
 208         return NETDEV_TX_OK;
 209 }
 210
 211 /*
 212  * netvsc_linkstatus_callback - Link up/down notification
 213  */
 214 static void netvsc_linkstatus_callback(struct hv_device *device_obj,
 215                                        unsigned int status)
 216 {
 217         struct net_device *net = dev_get_drvdata(&device_obj->device);
 218         struct net_device_context *ndev_ctx;
 219
 220         if (!net) {
 221                 netdev_err(net, "got link status but net device "
 222                                 "not initialized yet\n");
 223                 return;
 224         }
 225
 226         if (status == 1) {
 227                 netif_carrier_on(net);
 228                 netif_wake_queue(net);
 229                 netif_notify_peers(net);
 230                 ndev_ctx = netdev_priv(net);
 231                 schedule_work(&ndev_ctx->work);
 232         } else {
 233                 netif_carrier_off(net);
 234                 netif_stop_queue(net);
 235         }
 236 }
 237
 238 /*
 239  * netvsc_recv_callback -  Callback when we receive a packet from the
 240  * "wire" on the specified device.
 241  */
 242 static int netvsc_recv_callback(struct hv_device *device_obj,
 243                                 struct hv_netvsc_packet *packet)
 244 {
 245         struct net_device *net = dev_get_drvdata(&device_obj->device);
 246         struct sk_buff *skb;
 247         void *data;
 248         int i;
 249         unsigned long flags;
 250
 251         if (!net) {
 252                 netdev_err(net, "got receive callback but net device"
 253                         " not initialized yet\n");
 254                 return 0;
 255         }
 256
 257         /* Allocate a skb - TODO direct I/O to pages? */
 258         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 259         if (unlikely(!skb)) {
 260                 ++net->stats.rx_dropped;
 261                 return 0;
 262         }
 263
 264         /* for kmap_atomic */
 265         local_irq_save(flags);
 266
 267         /*
 268          * Copy to skb. This copy is needed here since the memory pointed by
 269          * hv_netvsc_packet cannot be deallocated
 270          */
 271         for (i = 0; i < packet->page_buf_cnt; i++) {
 272                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
 273                                                KM_IRQ1);
 274                 data = (void *)(unsigned long)data +
 275                                 packet->page_buf[i].offset;
 276
 277                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
 278                        packet->page_buf[i].len);
 279
 280                 kunmap_atomic((void *)((unsigned long)data -
 281                                        packet->page_buf[i].offset), KM_IRQ1);
 282         }
 283
 284         local_irq_restore(flags);
 285
 286         skb->protocol = eth_type_trans(skb, net);
 287         skb->ip_summed = CHECKSUM_NONE;
 288
 289         net->stats.rx_packets++;
 290         net->stats.rx_bytes += skb->len;
 291
 292         /*
 293          * Pass the skb back up. Network stack will deallocate the skb when it
 294          * is done.
 295          * TODO - use NAPI?
 296          */
 297         netif_rx(skb);
 298
 299         return 0;
 300 }
 301
 302 static void netvsc_get_drvinfo(struct net_device *net,
 303                                struct ethtool_drvinfo *info)
 304 {
 305         strcpy(info->driver, "hv_netvsc");
 306         strcpy(info->version, HV_DRV_VERSION);
 307         strcpy(info->fw_version, "N/A");
 308 }
 309
 310 static const struct ethtool_ops ethtool_ops = {
 311         .get_drvinfo    = netvsc_get_drvinfo,
 312         .get_sg         = ethtool_op_get_sg,
 313         .set_sg         = ethtool_op_set_sg,
 314         .get_link       = ethtool_op_get_link,
 315 };
 316
 317 static const struct net_device_ops device_ops = {
 318         .ndo_open =                     netvsc_open,
 319         .ndo_stop =                     netvsc_close,
 320         .ndo_start_xmit =               netvsc_start_xmit,
 321         .ndo_set_multicast_list =       netvsc_set_multicast_list,
 322         .ndo_change_mtu =               eth_change_mtu,
 323         .ndo_validate_addr =            eth_validate_addr,
 324         .ndo_set_mac_address =          eth_mac_addr,
 325 };
 326
 327 /*
 328  * Send GARP packet to network peers after migrations.
 329  * After Quick Migration, the network is not immediately operational in the
 330  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 331  * another netif_notify_peers() into a scheduled work, otherwise GARP packet
 332  * will not be sent after quick migration, and cause network disconnection.
 333  */
 334 static void netvsc_send_garp(struct work_struct *w)
 335 {
 336         struct net_device_context *ndev_ctx;
 337         struct net_device *net;
 338
 339         msleep(20);
 340         ndev_ctx = container_of(w, struct net_device_context, work);
 341         net = dev_get_drvdata(&ndev_ctx->device_ctx->device);
 342         netif_notify_peers(net);
 343 }
 344
 345
 346 static int netvsc_probe(struct device *device)
 347 {
 348         struct hv_driver *drv =
 349                 drv_to_hv_drv(device->driver);
 350         struct netvsc_driver *net_drv_obj = drv->priv;
 351         struct hv_device *device_obj = device_to_hv_device(device);
 352         struct net_device *net = NULL;
 353         struct net_device_context *net_device_ctx;
 354         struct netvsc_device_info device_info;
 355         int ret;
 356
 357         if (!net_drv_obj->base.dev_add)
 358                 return -1;
 359
 360         net = alloc_etherdev(sizeof(struct net_device_context));
 361         if (!net)
 362                 return -1;
 363
 364         /* Set initial state */
 365         netif_carrier_off(net);
 366
 367         net_device_ctx = netdev_priv(net);
 368         net_device_ctx->device_ctx = device_obj;
 369         net_device_ctx->avail = ring_size;
 370         dev_set_drvdata(device, net);
 371         INIT_WORK(&net_device_ctx->work, netvsc_send_garp);
 372
 373         /* Notify the netvsc driver of the new device */
 374         ret = net_drv_obj->base.dev_add(device_obj, &device_info);
 375         if (ret != 0) {
 376                 free_netdev(net);
 377                 dev_set_drvdata(device, NULL);
 378
 379                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 380                 return ret;
 381         }
 382
 383         /*
 384          * If carrier is still off ie we did not get a link status callback,
 385          * update it if necessary
 386          */
 387         /*
 388          * FIXME: We should use a atomic or test/set instead to avoid getting
 389          * out of sync with the device's link status
 390          */
 391         if (!netif_carrier_ok(net))
 392                 if (!device_info.link_state)
 393                         netif_carrier_on(net);
 394
 395         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 396
 397         net->netdev_ops = &device_ops;
 398
 399         /* TODO: Add GSO and Checksum offload */
 400         net->features = NETIF_F_SG;
 401
 402         SET_ETHTOOL_OPS(net, &ethtool_ops);
 403         SET_NETDEV_DEV(net, device);
 404
 405         ret = register_netdev(net);
 406         if (ret != 0) {
 407                 /* Remove the device and release the resource */
 408                 net_drv_obj->base.dev_rm(device_obj);
 409                 free_netdev(net);
 410         }
 411
 412         return ret;
 413 }
 414
 415 static int netvsc_remove(struct device *device)
 416 {
 417         struct hv_driver *drv =
 418                 drv_to_hv_drv(device->driver);
 419         struct netvsc_driver *net_drv_obj = drv->priv;
 420         struct hv_device *device_obj = device_to_hv_device(device);
 421         struct net_device *net = dev_get_drvdata(&device_obj->device);
 422         int ret;
 423
 424         if (net == NULL) {
 425                 dev_err(device, "No net device to remove\n");
 426                 return 0;
 427         }
 428
 429         if (!net_drv_obj->base.dev_rm)
 430                 return -1;
 431
 432         /* Stop outbound asap */
 433         netif_stop_queue(net);
 434         /* netif_carrier_off(net); */
 435
 436         unregister_netdev(net);
 437
 438         /*
 439          * Call to the vsc driver to let it know that the device is being
 440          * removed
 441          */
 442         ret = net_drv_obj->base.dev_rm(device_obj);
 443         if (ret != 0) {
 444                 /* TODO: */
 445                 netdev_err(net, "unable to remove vsc device (ret %d)\n", ret);
 446         }
 447
 448         free_netdev(net);
 449         return ret;
 450 }
 451
 452 static int netvsc_drv_exit_cb(struct device *dev, void *data)
 453 {
 454         struct device **curr = (struct device **)data;
 455
 456         *curr = dev;
 457         /* stop iterating */
 458         return 1;
 459 }
 460
 461 static void netvsc_drv_exit(void)
 462 {
 463         struct netvsc_driver *netvsc_drv_obj = &g_netvsc_drv;
 464         struct hv_driver *drv = &g_netvsc_drv.base;
 465         struct device *current_dev;
 466         int ret;
 467
 468         while (1) {
 469                 current_dev = NULL;
 470
 471                 /* Get the device */
 472                 ret = driver_for_each_device(&drv->driver, NULL,
 473                                              &current_dev, netvsc_drv_exit_cb);
 474
 475                 if (current_dev == NULL)
 476                         break;
 477
 478                 /* Initiate removal from the top-down */
 479                 dev_err(current_dev, "unregistering device (%s)...\n",
 480                         dev_name(current_dev));
 481
 482                 device_unregister(current_dev);
 483         }
 484
 485         if (netvsc_drv_obj->base.cleanup)
 486                 netvsc_drv_obj->base.cleanup(&netvsc_drv_obj->base);
 487
 488         vmbus_child_driver_unregister(&drv->driver);
 489
 490         return;
 491 }
 492
 493 static int netvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
 494 {
 495         struct netvsc_driver *net_drv_obj = &g_netvsc_drv;
 496         struct hv_driver *drv = &g_netvsc_drv.base;
 497         int ret;
 498
 499         net_drv_obj->ring_buf_size = ring_size * PAGE_SIZE;
 500         net_drv_obj->recv_cb = netvsc_recv_callback;
 501         net_drv_obj->link_status_change = netvsc_linkstatus_callback;
 502         drv->priv = net_drv_obj;
 503
 504         /* Callback to client driver to complete the initialization */
 505         drv_init(&net_drv_obj->base);
 506
 507         drv->driver.name = net_drv_obj->base.name;
 508
 509         drv->driver.probe = netvsc_probe;
 510         drv->driver.remove = netvsc_remove;
 511
 512         /* The driver belongs to vmbus */
 513         ret = vmbus_child_driver_register(&drv->driver);
 514
 515         return ret;
 516 }
 517
 518 static const struct dmi_system_id __initconst
 519 hv_netvsc_dmi_table[] __maybe_unused  = {
 520         {
 521                 .ident = "Hyper-V",
 522                 .matches = {
 523                         DMI_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),
 524                         DMI_MATCH(DMI_PRODUCT_NAME, "Virtual Machine"),
 525                         DMI_MATCH(DMI_BOARD_NAME, "Virtual Machine"),
 526                 },
 527         },
 528         { },
 529 };
 530 MODULE_DEVICE_TABLE(dmi, hv_netvsc_dmi_table);
 531
 532 static int __init netvsc_init(void)
 533 {
 534         pr_info("initializing....");
 535
 536         if (!dmi_check_system(hv_netvsc_dmi_table))
 537                 return -ENODEV;
 538
 539         return netvsc_drv_init(netvsc_initialize);
 540 }
 541
 542 static void __exit netvsc_exit(void)
 543 {
 544         netvsc_drv_exit();
 545 }
 546
 547 static const struct pci_device_id __initconst
 548 hv_netvsc_pci_table[] __maybe_unused = {
 549         { PCI_DEVICE(0x1414, 0x5353) }, /* VGA compatible controller */
 550         { 0 }
 551 };
 552 MODULE_DEVICE_TABLE(pci, hv_netvsc_pci_table);
 553
 554 MODULE_LICENSE("GPL");
 555 MODULE_VERSION(HV_DRV_VERSION);
 556 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 557
 558 module_init(netvsc_init);
 559 module_exit(netvsc_exit);