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/atomic.h>
  25 #include <linux/module.h>
  26 #include <linux/highmem.h>
  27 #include <linux/device.h>
  28 #include <linux/io.h>
  29 #include <linux/delay.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/inetdevice.h>
  32 #include <linux/etherdevice.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/in.h>
  35 #include <linux/slab.h>
  36 #include <linux/dmi.h>
  37 #include <linux/pci.h>
  38 #include <net/arp.h>
  39 #include <net/route.h>
  40 #include <net/sock.h>
  41 #include <net/pkt_sched.h>
  42
  43 #include "hyperv.h"
  44 #include "hyperv_net.h"
  45
  46 struct net_device_context {
  47         /* point back to our device context */
  48         struct hv_device *device_ctx;
  49         atomic_t avail;
  50         struct delayed_work dwork;
  51 };
  52
  53
  54 #define PACKET_PAGES_LOWATER  8
  55 /* Need this many pages to handle worst case fragmented packet */
  56 #define PACKET_PAGES_HIWATER  (MAX_SKB_FRAGS + 2)
  57
  58 static int ring_size = 128;
  59 module_param(ring_size, int, S_IRUGO);
  60 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  61
  62 /* no-op so the netdev core doesn't return -EINVAL when modifying the the
  63  * multicast address list in SIOCADDMULTI. hv is setup to get all multicast
  64  * when it calls RndisFilterOnOpen() */
  65 static void netvsc_set_multicast_list(struct net_device *net)
  66 {
  67 }
  68
  69 static int netvsc_open(struct net_device *net)
  70 {
  71         struct net_device_context *net_device_ctx = netdev_priv(net);
  72         struct hv_device *device_obj = net_device_ctx->device_ctx;
  73         int ret = 0;
  74
  75         if (netif_carrier_ok(net)) {
  76                 /* Open up the device */
  77                 ret = rndis_filter_open(device_obj);
  78                 if (ret != 0) {
  79                         netdev_err(net, "unable to open device (ret %d).\n",
  80                                    ret);
  81                         return ret;
  82                 }
  83
  84                 netif_start_queue(net);
  85         } else {
  86                 netdev_err(net, "unable to open device...link is down.\n");
  87         }
  88
  89         return ret;
  90 }
  91
  92 static int netvsc_close(struct net_device *net)
  93 {
  94         struct net_device_context *net_device_ctx = netdev_priv(net);
  95         struct hv_device *device_obj = net_device_ctx->device_ctx;
  96         int ret;
  97
  98         netif_stop_queue(net);
  99
 100         ret = rndis_filter_close(device_obj);
 101         if (ret != 0)
 102                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 103
 104         return ret;
 105 }
 106
 107 static void netvsc_xmit_completion(void *context)
 108 {
 109         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 110         struct sk_buff *skb = (struct sk_buff *)
 111                 (unsigned long)packet->completion.send.send_completion_tid;
 112
 113         kfree(packet);
 114
 115         if (skb) {
 116                 struct net_device *net = skb->dev;
 117                 struct net_device_context *net_device_ctx = netdev_priv(net);
 118                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
 119
 120                 dev_kfree_skb_any(skb);
 121
 122                 atomic_add(num_pages, &net_device_ctx->avail);
 123                 if (atomic_read(&net_device_ctx->avail) >=
 124                                 PACKET_PAGES_HIWATER)
 125                         netif_wake_queue(net);
 126         }
 127 }
 128
 129 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 130 {
 131         struct net_device_context *net_device_ctx = netdev_priv(net);
 132         struct hv_netvsc_packet *packet;
 133         int ret;
 134         unsigned int i, num_pages;
 135
 136         /* Add 1 for skb->data and additional one for RNDIS */
 137         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
 138         if (num_pages > atomic_read(&net_device_ctx->avail))
 139                 return NETDEV_TX_BUSY;
 140
 141         /* Allocate a netvsc packet based on # of frags. */
 142         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 143                          (num_pages * sizeof(struct hv_page_buffer)) +
 144                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
 145         if (!packet) {
 146                 /* out of memory, silently drop packet */
 147                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 148
 149                 dev_kfree_skb(skb);
 150                 net->stats.tx_dropped++;
 151                 return NETDEV_TX_OK;
 152         }
 153
 154         packet->extension = (void *)(unsigned long)packet +
 155                                 sizeof(struct hv_netvsc_packet) +
 156                                     (num_pages * sizeof(struct hv_page_buffer));
 157
 158         /* Setup the rndis header */
 159         packet->page_buf_cnt = num_pages;
 160
 161         /* Initialize it from the skb */
 162         packet->total_data_buflen       = skb->len;
 163
 164         /* Start filling in the page buffers starting after RNDIS buffer. */
 165         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 166         packet->page_buf[1].offset
 167                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 168         packet->page_buf[1].len = skb_headlen(skb);
 169
 170         /* Additional fragments are after SKB data */
 171         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 172                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 173
 174                 packet->page_buf[i+2].pfn = page_to_pfn(f->page);
 175                 packet->page_buf[i+2].offset = f->page_offset;
 176                 packet->page_buf[i+2].len = f->size;
 177         }
 178
 179         /* Set the completion routine */
 180         packet->completion.send.send_completion = netvsc_xmit_completion;
 181         packet->completion.send.send_completion_ctx = packet;
 182         packet->completion.send.send_completion_tid = (unsigned long)skb;
 183
 184         ret = rndis_filter_send(net_device_ctx->device_ctx,
 185                                   packet);
 186         if (ret == 0) {
 187                 net->stats.tx_bytes += skb->len;
 188                 net->stats.tx_packets++;
 189
 190                 atomic_sub(num_pages, &net_device_ctx->avail);
 191                 if (atomic_read(&net_device_ctx->avail) < PACKET_PAGES_LOWATER)
 192                         netif_stop_queue(net);
 193         } else {
 194                 /* we are shutting down or bus overloaded, just drop packet */
 195                 net->stats.tx_dropped++;
 196                 netvsc_xmit_completion(packet);
 197         }
 198
 199         return NETDEV_TX_OK;
 200 }
 201
 202 /*
 203  * netvsc_linkstatus_callback - Link up/down notification
 204  */
 205 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 206                                        unsigned int status)
 207 {
 208         struct net_device *net = dev_get_drvdata(&device_obj->device);
 209         struct net_device_context *ndev_ctx;
 210
 211         if (!net) {
 212                 netdev_err(net, "got link status but net device "
 213                                 "not initialized yet\n");
 214                 return;
 215         }
 216
 217         if (status == 1) {
 218                 netif_carrier_on(net);
 219                 netif_wake_queue(net);
 220                 netif_notify_peers(net);
 221                 ndev_ctx = netdev_priv(net);
 222                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 223         } else {
 224                 netif_carrier_off(net);
 225                 netif_stop_queue(net);
 226         }
 227 }
 228
 229 /*
 230  * netvsc_recv_callback -  Callback when we receive a packet from the
 231  * "wire" on the specified device.
 232  */
 233 int netvsc_recv_callback(struct hv_device *device_obj,
 234                                 struct hv_netvsc_packet *packet)
 235 {
 236         struct net_device *net = dev_get_drvdata(&device_obj->device);
 237         struct sk_buff *skb;
 238         void *data;
 239         int i;
 240         unsigned long flags;
 241
 242         if (!net) {
 243                 netdev_err(net, "got receive callback but net device"
 244                         " not initialized yet\n");
 245                 return 0;
 246         }
 247
 248         /* Allocate a skb - TODO direct I/O to pages? */
 249         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 250         if (unlikely(!skb)) {
 251                 ++net->stats.rx_dropped;
 252                 return 0;
 253         }
 254
 255         /* for kmap_atomic */
 256         local_irq_save(flags);
 257
 258         /*
 259          * Copy to skb. This copy is needed here since the memory pointed by
 260          * hv_netvsc_packet cannot be deallocated
 261          */
 262         for (i = 0; i < packet->page_buf_cnt; i++) {
 263                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
 264                                                KM_IRQ1);
 265                 data = (void *)(unsigned long)data +
 266                                 packet->page_buf[i].offset;
 267
 268                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
 269                        packet->page_buf[i].len);
 270
 271                 kunmap_atomic((void *)((unsigned long)data -
 272                                        packet->page_buf[i].offset), KM_IRQ1);
 273         }
 274
 275         local_irq_restore(flags);
 276
 277         skb->protocol = eth_type_trans(skb, net);
 278         skb->ip_summed = CHECKSUM_NONE;
 279
 280         net->stats.rx_packets++;
 281         net->stats.rx_bytes += skb->len;
 282
 283         /*
 284          * Pass the skb back up. Network stack will deallocate the skb when it
 285          * is done.
 286          * TODO - use NAPI?
 287          */
 288         netif_rx(skb);
 289
 290         return 0;
 291 }
 292
 293 static void netvsc_get_drvinfo(struct net_device *net,
 294                                struct ethtool_drvinfo *info)
 295 {
 296         strcpy(info->driver, "hv_netvsc");
 297         strcpy(info->version, HV_DRV_VERSION);
 298         strcpy(info->fw_version, "N/A");
 299 }
 300
 301 static const struct ethtool_ops ethtool_ops = {
 302         .get_drvinfo    = netvsc_get_drvinfo,
 303         .get_link       = ethtool_op_get_link,
 304 };
 305
 306 static const struct net_device_ops device_ops = {
 307         .ndo_open =                     netvsc_open,
 308         .ndo_stop =                     netvsc_close,
 309         .ndo_start_xmit =               netvsc_start_xmit,
 310         .ndo_set_multicast_list =       netvsc_set_multicast_list,
 311         .ndo_change_mtu =               eth_change_mtu,
 312         .ndo_validate_addr =            eth_validate_addr,
 313         .ndo_set_mac_address =          eth_mac_addr,
 314 };
 315
 316 /*
 317  * Send GARP packet to network peers after migrations.
 318  * After Quick Migration, the network is not immediately operational in the
 319  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 320  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 321  * will not be sent after quick migration, and cause network disconnection.
 322  */
 323 static void netvsc_send_garp(struct work_struct *w)
 324 {
 325         struct net_device_context *ndev_ctx;
 326         struct net_device *net;
 327
 328         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 329         net = dev_get_drvdata(&ndev_ctx->device_ctx->device);
 330         netif_notify_peers(net);
 331 }
 332
 333
 334 static int netvsc_probe(struct hv_device *dev)
 335 {
 336         struct net_device *net = NULL;
 337         struct net_device_context *net_device_ctx;
 338         struct netvsc_device_info device_info;
 339         int ret;
 340
 341         net = alloc_etherdev(sizeof(struct net_device_context));
 342         if (!net)
 343                 return -1;
 344
 345         /* Set initial state */
 346         netif_carrier_off(net);
 347
 348         net_device_ctx = netdev_priv(net);
 349         net_device_ctx->device_ctx = dev;
 350         atomic_set(&net_device_ctx->avail, ring_size);
 351         dev_set_drvdata(&dev->device, net);
 352         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
 353
 354         /* Notify the netvsc driver of the new device */
 355         device_info.ring_size = ring_size;
 356         ret = rndis_filter_device_add(dev, &device_info);
 357         if (ret != 0) {
 358                 free_netdev(net);
 359                 dev_set_drvdata(&dev->device, NULL);
 360
 361                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 362                 return ret;
 363         }
 364
 365         netif_carrier_on(net);
 366
 367         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 368
 369         net->netdev_ops = &device_ops;
 370
 371         /* TODO: Add GSO and Checksum offload */
 372         net->hw_features = NETIF_F_SG;
 373         net->features = NETIF_F_SG;
 374
 375         SET_ETHTOOL_OPS(net, &ethtool_ops);
 376         SET_NETDEV_DEV(net, &dev->device);
 377
 378         ret = register_netdev(net);
 379         if (ret != 0) {
 380                 /* Remove the device and release the resource */
 381                 rndis_filter_device_remove(dev);
 382                 free_netdev(net);
 383         }
 384
 385         return ret;
 386 }
 387
 388 static int netvsc_remove(struct hv_device *dev)
 389 {
 390         struct net_device *net = dev_get_drvdata(&dev->device);
 391         struct net_device_context *ndev_ctx;
 392
 393         if (net == NULL) {
 394                 dev_err(&dev->device, "No net device to remove\n");
 395                 return 0;
 396         }
 397
 398         ndev_ctx = netdev_priv(net);
 399         cancel_delayed_work_sync(&ndev_ctx->dwork);
 400
 401         /* Stop outbound asap */
 402         netif_stop_queue(net);
 403
 404         unregister_netdev(net);
 405
 406         /*
 407          * Call to the vsc driver to let it know that the device is being
 408          * removed
 409          */
 410         rndis_filter_device_remove(dev);
 411
 412         free_netdev(net);
 413         return 0;
 414 }
 415
 416 /* The one and only one */
 417 static struct  hv_driver netvsc_drv = {
 418         .probe = netvsc_probe,
 419         .remove = netvsc_remove,
 420 };
 421
 422 static void __exit netvsc_drv_exit(void)
 423 {
 424         vmbus_child_driver_unregister(&netvsc_drv.driver);
 425 }
 426
 427
 428 static const struct dmi_system_id __initconst
 429 hv_netvsc_dmi_table[] __maybe_unused  = {
 430         {
 431                 .ident = "Hyper-V",
 432                 .matches = {
 433                         DMI_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),
 434                         DMI_MATCH(DMI_PRODUCT_NAME, "Virtual Machine"),
 435                         DMI_MATCH(DMI_BOARD_NAME, "Virtual Machine"),
 436                 },
 437         },
 438         { },
 439 };
 440 MODULE_DEVICE_TABLE(dmi, hv_netvsc_dmi_table);
 441
 442 static int __init netvsc_drv_init(void)
 443 {
 444         struct hv_driver *drv = &netvsc_drv;
 445         int ret;
 446
 447         pr_info("initializing....");
 448
 449         if (!dmi_check_system(hv_netvsc_dmi_table))
 450                 return -ENODEV;
 451
 452
 453         /* Callback to client driver to complete the initialization */
 454         netvsc_initialize(drv);
 455
 456         drv->driver.name = drv->name;
 457
 458         /* The driver belongs to vmbus */
 459         ret = vmbus_child_driver_register(&drv->driver);
 460
 461         return ret;
 462 }
 463
 464 static const struct pci_device_id __initconst
 465 hv_netvsc_pci_table[] __maybe_unused = {
 466         { PCI_DEVICE(0x1414, 0x5353) }, /* VGA compatible controller */
 467         { 0 }
 468 };
 469 MODULE_DEVICE_TABLE(pci, hv_netvsc_pci_table);
 470
 471 MODULE_LICENSE("GPL");
 472 MODULE_VERSION(HV_DRV_VERSION);
 473 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 474
 475 module_init(netvsc_drv_init);
 476 module_exit(netvsc_drv_exit);