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 <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43 struct net_device_context {
  44         /* point back to our device context */
  45         struct hv_device *device_ctx;
  46         atomic_t avail;
  47         struct delayed_work dwork;
  48 };
  49
  50
  51 #define PACKET_PAGES_LOWATER  8
  52 /* Need this many pages to handle worst case fragmented packet */
  53 #define PACKET_PAGES_HIWATER  (MAX_SKB_FRAGS + 2)
  54
  55 /* Restrict GSO size to account for NVGRE */
  56 #define NETVSC_GSO_MAX_SIZE     62768
  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         /* Open up the device */
  76         ret = rndis_filter_open(device_obj);
  77         if (ret != 0) {
  78                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  79                 return ret;
  80         }
  81
  82         netif_start_queue(net);
  83
  84         return ret;
  85 }
  86
  87 static int netvsc_close(struct net_device *net)
  88 {
  89         struct net_device_context *net_device_ctx = netdev_priv(net);
  90         struct hv_device *device_obj = net_device_ctx->device_ctx;
  91         int ret;
  92
  93         netif_stop_queue(net);
  94
  95         ret = rndis_filter_close(device_obj);
  96         if (ret != 0)
  97                 netdev_err(net, "unable to close device (ret %d).\n", ret);
  98
  99         return ret;
 100 }
 101
 102 static void netvsc_xmit_completion(void *context)
 103 {
 104         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 105         struct sk_buff *skb = (struct sk_buff *)
 106                 (unsigned long)packet->completion.send.send_completion_tid;
 107
 108         kfree(packet);
 109
 110         if (skb) {
 111                 struct net_device *net = skb->dev;
 112                 struct net_device_context *net_device_ctx = netdev_priv(net);
 113                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
 114
 115                 dev_kfree_skb_any(skb);
 116
 117                 atomic_add(num_pages, &net_device_ctx->avail);
 118                 if (atomic_read(&net_device_ctx->avail) >=
 119                                 PACKET_PAGES_HIWATER)
 120                         netif_wake_queue(net);
 121         }
 122 }
 123
 124 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 125 {
 126         struct net_device_context *net_device_ctx = netdev_priv(net);
 127         struct hv_netvsc_packet *packet;
 128         int ret;
 129         unsigned int i, num_pages;
 130
 131         /* Add 1 for skb->data and additional one for RNDIS */
 132         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
 133         if (num_pages > atomic_read(&net_device_ctx->avail))
 134                 return NETDEV_TX_BUSY;
 135
 136         /* Allocate a netvsc packet based on # of frags. */
 137         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 138                          (num_pages * sizeof(struct hv_page_buffer)) +
 139                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
 140         if (!packet) {
 141                 /* out of memory, drop packet */
 142                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 143
 144                 dev_kfree_skb(skb);
 145                 net->stats.tx_dropped++;
 146                 return NETDEV_TX_BUSY;
 147         }
 148
 149         packet->extension = (void *)(unsigned long)packet +
 150                                 sizeof(struct hv_netvsc_packet) +
 151                                     (num_pages * sizeof(struct hv_page_buffer));
 152
 153         /* Setup the rndis header */
 154         packet->page_buf_cnt = num_pages;
 155
 156         /* Initialize it from the skb */
 157         packet->total_data_buflen       = skb->len;
 158
 159         /* Start filling in the page buffers starting after RNDIS buffer. */
 160         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 161         packet->page_buf[1].offset
 162                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 163         packet->page_buf[1].len = skb_headlen(skb);
 164
 165         /* Additional fragments are after SKB data */
 166         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 167                 const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 168
 169                 packet->page_buf[i+2].pfn = page_to_pfn(skb_frag_page(f));
 170                 packet->page_buf[i+2].offset = f->page_offset;
 171                 packet->page_buf[i+2].len = skb_frag_size(f);
 172         }
 173
 174         /* Set the completion routine */
 175         packet->completion.send.send_completion = netvsc_xmit_completion;
 176         packet->completion.send.send_completion_ctx = packet;
 177         packet->completion.send.send_completion_tid = (unsigned long)skb;
 178
 179         ret = rndis_filter_send(net_device_ctx->device_ctx,
 180                                   packet);
 181         if (ret == 0) {
 182                 net->stats.tx_bytes += skb->len;
 183                 net->stats.tx_packets++;
 184
 185                 atomic_sub(num_pages, &net_device_ctx->avail);
 186                 if (atomic_read(&net_device_ctx->avail) < PACKET_PAGES_LOWATER)
 187                         netif_stop_queue(net);
 188         } else {
 189                 /* we are shutting down or bus overloaded, just drop packet */
 190                 net->stats.tx_dropped++;
 191                 kfree(packet);
 192                 dev_kfree_skb_any(skb);
 193         }
 194
 195         return ret ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 196 }
 197
 198 /*
 199  * netvsc_linkstatus_callback - Link up/down notification
 200  */
 201 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 202                                        unsigned int status)
 203 {
 204         struct net_device *net;
 205         struct net_device_context *ndev_ctx;
 206         struct netvsc_device *net_device;
 207
 208         net_device = hv_get_drvdata(device_obj);
 209         net = net_device->ndev;
 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                 ndev_ctx = netdev_priv(net);
 221                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 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         struct netvsc_device *net_device;
 242
 243         net_device = hv_get_drvdata(device_obj);
 244         net = net_device->ndev;
 245
 246         if (!net) {
 247                 netdev_err(net, "got receive callback but net device"
 248                         " not initialized yet\n");
 249                 return 0;
 250         }
 251
 252         /* Allocate a skb - TODO direct I/O to pages? */
 253         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 254         if (unlikely(!skb)) {
 255                 ++net->stats.rx_dropped;
 256                 return 0;
 257         }
 258
 259         /* for kmap_atomic */
 260         local_irq_save(flags);
 261
 262         /*
 263          * Copy to skb. This copy is needed here since the memory pointed by
 264          * hv_netvsc_packet cannot be deallocated
 265          */
 266         for (i = 0; i < packet->page_buf_cnt; i++) {
 267                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
 268                                                KM_IRQ1);
 269                 data = (void *)(unsigned long)data +
 270                                 packet->page_buf[i].offset;
 271
 272                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
 273                        packet->page_buf[i].len);
 274
 275                 kunmap_atomic((void *)((unsigned long)data -
 276                                        packet->page_buf[i].offset), KM_IRQ1);
 277         }
 278
 279         local_irq_restore(flags);
 280
 281         skb->protocol = eth_type_trans(skb, net);
 282         skb->ip_summed = CHECKSUM_NONE;
 283
 284         net->stats.rx_packets++;
 285         net->stats.rx_bytes += skb->len;
 286
 287         /*
 288          * Pass the skb back up. Network stack will deallocate the skb when it
 289          * is done.
 290          * TODO - use NAPI?
 291          */
 292         netif_rx(skb);
 293
 294         return 0;
 295 }
 296
 297 static void netvsc_get_drvinfo(struct net_device *net,
 298                                struct ethtool_drvinfo *info)
 299 {
 300         strcpy(info->driver, "hv_netvsc");
 301         strcpy(info->version, HV_DRV_VERSION);
 302         strcpy(info->fw_version, "N/A");
 303 }
 304
 305 static const struct ethtool_ops ethtool_ops = {
 306         .get_drvinfo    = netvsc_get_drvinfo,
 307         .get_link       = ethtool_op_get_link,
 308 };
 309
 310 static const struct net_device_ops device_ops = {
 311         .ndo_open =                     netvsc_open,
 312         .ndo_stop =                     netvsc_close,
 313         .ndo_start_xmit =               netvsc_start_xmit,
 314         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 315         .ndo_change_mtu =               eth_change_mtu,
 316         .ndo_validate_addr =            eth_validate_addr,
 317         .ndo_set_mac_address =          eth_mac_addr,
 318 };
 319
 320 /*
 321  * Send GARP packet to network peers after migrations.
 322  * After Quick Migration, the network is not immediately operational in the
 323  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 324  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 325  * will not be sent after quick migration, and cause network disconnection.
 326  */
 327 static void netvsc_send_garp(struct work_struct *w)
 328 {
 329         struct net_device_context *ndev_ctx;
 330         struct net_device *net;
 331         struct netvsc_device *net_device;
 332
 333         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 334         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 335         net = net_device->ndev;
 336         netif_notify_peers(net);
 337 }
 338
 339
 340 static int netvsc_probe(struct hv_device *dev,
 341                         const struct hv_vmbus_device_id *dev_id)
 342 {
 343         struct net_device *net = NULL;
 344         struct net_device_context *net_device_ctx;
 345         struct netvsc_device_info device_info;
 346         int ret;
 347
 348         net = alloc_etherdev(sizeof(struct net_device_context));
 349         if (!net)
 350                 return -ENOMEM;
 351
 352         /* Set initial state */
 353         netif_carrier_off(net);
 354
 355         net_device_ctx = netdev_priv(net);
 356         net_device_ctx->device_ctx = dev;
 357         atomic_set(&net_device_ctx->avail, ring_size);
 358         hv_set_drvdata(dev, net);
 359         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
 360
 361         net->netdev_ops = &device_ops;
 362
 363         /* TODO: Add GSO and Checksum offload */
 364         net->hw_features = NETIF_F_SG;
 365         net->features = NETIF_F_SG;
 366
 367         SET_ETHTOOL_OPS(net, &ethtool_ops);
 368         SET_NETDEV_DEV(net, &dev->device);
 369         netif_set_gso_max_size(net, NETVSC_GSO_MAX_SIZE);
 370
 371         ret = register_netdev(net);
 372         if (ret != 0) {
 373                 pr_err("Unable to register netdev.\n");
 374                 free_netdev(net);
 375                 goto out;
 376         }
 377
 378         /* Notify the netvsc driver of the new device */
 379         device_info.ring_size = ring_size;
 380         ret = rndis_filter_device_add(dev, &device_info);
 381         if (ret != 0) {
 382                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 383                 unregister_netdev(net);
 384                 free_netdev(net);
 385                 hv_set_drvdata(dev, NULL);
 386                 return ret;
 387         }
 388         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 389
 390         netif_carrier_on(net);
 391
 392 out:
 393         return ret;
 394 }
 395
 396 static int netvsc_remove(struct hv_device *dev)
 397 {
 398         struct net_device *net;
 399         struct net_device_context *ndev_ctx;
 400         struct netvsc_device *net_device;
 401
 402         net_device = hv_get_drvdata(dev);
 403         net = net_device->ndev;
 404
 405         if (net == NULL) {
 406                 dev_err(&dev->device, "No net device to remove\n");
 407                 return 0;
 408         }
 409
 410         ndev_ctx = netdev_priv(net);
 411         cancel_delayed_work_sync(&ndev_ctx->dwork);
 412
 413         /* Stop outbound asap */
 414         netif_stop_queue(net);
 415
 416         unregister_netdev(net);
 417
 418         /*
 419          * Call to the vsc driver to let it know that the device is being
 420          * removed
 421          */
 422         rndis_filter_device_remove(dev);
 423
 424         free_netdev(net);
 425         return 0;
 426 }
 427
 428 static const struct hv_vmbus_device_id id_table[] = {
 429         /* Network guid */
 430         { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
 431                        0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
 432         { },
 433 };
 434
 435 MODULE_DEVICE_TABLE(vmbus, id_table);
 436
 437 /* The one and only one */
 438 static struct  hv_driver netvsc_drv = {
 439         .name = "netvsc",
 440         .id_table = id_table,
 441         .probe = netvsc_probe,
 442         .remove = netvsc_remove,
 443 };
 444
 445 static void __exit netvsc_drv_exit(void)
 446 {
 447         vmbus_driver_unregister(&netvsc_drv);
 448 }
 449
 450 static int __init netvsc_drv_init(void)
 451 {
 452         return vmbus_driver_register(&netvsc_drv);
 453 }
 454
 455 MODULE_LICENSE("GPL");
 456 MODULE_VERSION(HV_DRV_VERSION);
 457 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 458
 459 module_init(netvsc_drv_init);
 460 module_exit(netvsc_drv_exit);