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 static int ring_size = 128;
  56 module_param(ring_size, int, S_IRUGO);
  57 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  58
  59 /* no-op so the netdev core doesn't return -EINVAL when modifying the the
  60  * multicast address list in SIOCADDMULTI. hv is setup to get all multicast
  61  * when it calls RndisFilterOnOpen() */
  62 static void netvsc_set_multicast_list(struct net_device *net)
  63 {
  64 }
  65
  66 static int netvsc_open(struct net_device *net)
  67 {
  68         struct net_device_context *net_device_ctx = netdev_priv(net);
  69         struct hv_device *device_obj = net_device_ctx->device_ctx;
  70         int ret = 0;
  71
  72         if (netif_carrier_ok(net)) {
  73                 /* Open up the device */
  74                 ret = rndis_filter_open(device_obj);
  75                 if (ret != 0) {
  76                         netdev_err(net, "unable to open device (ret %d).\n",
  77                                    ret);
  78                         return ret;
  79                 }
  80
  81                 netif_start_queue(net);
  82         } else {
  83                 netdev_err(net, "unable to open device...link is down.\n");
  84         }
  85
  86         return ret;
  87 }
  88
  89 static int netvsc_close(struct net_device *net)
  90 {
  91         struct net_device_context *net_device_ctx = netdev_priv(net);
  92         struct hv_device *device_obj = net_device_ctx->device_ctx;
  93         int ret;
  94
  95         netif_stop_queue(net);
  96
  97         ret = rndis_filter_close(device_obj);
  98         if (ret != 0)
  99                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 100
 101         return ret;
 102 }
 103
 104 static void netvsc_xmit_completion(void *context)
 105 {
 106         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 107         struct sk_buff *skb = (struct sk_buff *)
 108                 (unsigned long)packet->completion.send.send_completion_tid;
 109
 110         kfree(packet);
 111
 112         if (skb) {
 113                 struct net_device *net = skb->dev;
 114                 struct net_device_context *net_device_ctx = netdev_priv(net);
 115                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
 116
 117                 dev_kfree_skb_any(skb);
 118
 119                 atomic_add(num_pages, &net_device_ctx->avail);
 120                 if (atomic_read(&net_device_ctx->avail) >=
 121                                 PACKET_PAGES_HIWATER)
 122                         netif_wake_queue(net);
 123         }
 124 }
 125
 126 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 127 {
 128         struct net_device_context *net_device_ctx = netdev_priv(net);
 129         struct hv_netvsc_packet *packet;
 130         int ret;
 131         unsigned int i, num_pages;
 132
 133         /* Add 1 for skb->data and additional one for RNDIS */
 134         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
 135         if (num_pages > atomic_read(&net_device_ctx->avail))
 136                 return NETDEV_TX_BUSY;
 137
 138         /* Allocate a netvsc packet based on # of frags. */
 139         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 140                          (num_pages * sizeof(struct hv_page_buffer)) +
 141                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
 142         if (!packet) {
 143                 /* out of memory, silently drop packet */
 144                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 145
 146                 dev_kfree_skb(skb);
 147                 net->stats.tx_dropped++;
 148                 return NETDEV_TX_OK;
 149         }
 150
 151         packet->extension = (void *)(unsigned long)packet +
 152                                 sizeof(struct hv_netvsc_packet) +
 153                                     (num_pages * sizeof(struct hv_page_buffer));
 154
 155         /* Setup the rndis header */
 156         packet->page_buf_cnt = num_pages;
 157
 158         /* Initialize it from the skb */
 159         packet->total_data_buflen       = skb->len;
 160
 161         /* Start filling in the page buffers starting after RNDIS buffer. */
 162         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 163         packet->page_buf[1].offset
 164                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 165         packet->page_buf[1].len = skb_headlen(skb);
 166
 167         /* Additional fragments are after SKB data */
 168         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 169                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 170
 171                 packet->page_buf[i+2].pfn = page_to_pfn(f->page);
 172                 packet->page_buf[i+2].offset = f->page_offset;
 173                 packet->page_buf[i+2].len = f->size;
 174         }
 175
 176         /* Set the completion routine */
 177         packet->completion.send.send_completion = netvsc_xmit_completion;
 178         packet->completion.send.send_completion_ctx = packet;
 179         packet->completion.send.send_completion_tid = (unsigned long)skb;
 180
 181         ret = rndis_filter_send(net_device_ctx->device_ctx,
 182                                   packet);
 183         if (ret == 0) {
 184                 net->stats.tx_bytes += skb->len;
 185                 net->stats.tx_packets++;
 186
 187                 atomic_sub(num_pages, &net_device_ctx->avail);
 188                 if (atomic_read(&net_device_ctx->avail) < PACKET_PAGES_LOWATER)
 189                         netif_stop_queue(net);
 190         } else {
 191                 /* we are shutting down or bus overloaded, just drop packet */
 192                 net->stats.tx_dropped++;
 193                 kfree(packet);
 194                 dev_kfree_skb_any(skb);
 195         }
 196
 197         return NETDEV_TX_OK;
 198 }
 199
 200 /*
 201  * netvsc_linkstatus_callback - Link up/down notification
 202  */
 203 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 204                                        unsigned int status)
 205 {
 206         struct net_device *net = dev_get_drvdata(&device_obj->device);
 207         struct net_device_context *ndev_ctx;
 208
 209         if (!net) {
 210                 netdev_err(net, "got link status but net device "
 211                                 "not initialized yet\n");
 212                 return;
 213         }
 214
 215         if (status == 1) {
 216                 netif_carrier_on(net);
 217                 netif_wake_queue(net);
 218                 ndev_ctx = netdev_priv(net);
 219                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 220                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 221         } else {
 222                 netif_carrier_off(net);
 223                 netif_stop_queue(net);
 224         }
 225 }
 226
 227 /*
 228  * netvsc_recv_callback -  Callback when we receive a packet from the
 229  * "wire" on the specified device.
 230  */
 231 int netvsc_recv_callback(struct hv_device *device_obj,
 232                                 struct hv_netvsc_packet *packet)
 233 {
 234         struct net_device *net = dev_get_drvdata(&device_obj->device);
 235         struct sk_buff *skb;
 236         void *data;
 237         int i;
 238         unsigned long flags;
 239
 240         if (!net) {
 241                 netdev_err(net, "got receive callback but net device"
 242                         " not initialized yet\n");
 243                 return 0;
 244         }
 245
 246         /* Allocate a skb - TODO direct I/O to pages? */
 247         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 248         if (unlikely(!skb)) {
 249                 ++net->stats.rx_dropped;
 250                 return 0;
 251         }
 252
 253         /* for kmap_atomic */
 254         local_irq_save(flags);
 255
 256         /*
 257          * Copy to skb. This copy is needed here since the memory pointed by
 258          * hv_netvsc_packet cannot be deallocated
 259          */
 260         for (i = 0; i < packet->page_buf_cnt; i++) {
 261                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
 262                                                KM_IRQ1);
 263                 data = (void *)(unsigned long)data +
 264                                 packet->page_buf[i].offset;
 265
 266                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
 267                        packet->page_buf[i].len);
 268
 269                 kunmap_atomic((void *)((unsigned long)data -
 270                                        packet->page_buf[i].offset), KM_IRQ1);
 271         }
 272
 273         local_irq_restore(flags);
 274
 275         skb->protocol = eth_type_trans(skb, net);
 276         skb->ip_summed = CHECKSUM_NONE;
 277
 278         net->stats.rx_packets++;
 279         net->stats.rx_bytes += skb->len;
 280
 281         /*
 282          * Pass the skb back up. Network stack will deallocate the skb when it
 283          * is done.
 284          * TODO - use NAPI?
 285          */
 286         netif_rx(skb);
 287
 288         return 0;
 289 }
 290
 291 static void netvsc_get_drvinfo(struct net_device *net,
 292                                struct ethtool_drvinfo *info)
 293 {
 294         strcpy(info->driver, "hv_netvsc");
 295         strcpy(info->version, HV_DRV_VERSION);
 296         strcpy(info->fw_version, "N/A");
 297 }
 298
 299 static const struct ethtool_ops ethtool_ops = {
 300         .get_drvinfo    = netvsc_get_drvinfo,
 301         .get_link       = ethtool_op_get_link,
 302 };
 303
 304 static const struct net_device_ops device_ops = {
 305         .ndo_open =                     netvsc_open,
 306         .ndo_stop =                     netvsc_close,
 307         .ndo_start_xmit =               netvsc_start_xmit,
 308         .ndo_set_multicast_list =       netvsc_set_multicast_list,
 309         .ndo_change_mtu =               eth_change_mtu,
 310         .ndo_validate_addr =            eth_validate_addr,
 311         .ndo_set_mac_address =          eth_mac_addr,
 312 };
 313
 314 /*
 315  * Send GARP packet to network peers after migrations.
 316  * After Quick Migration, the network is not immediately operational in the
 317  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 318  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 319  * will not be sent after quick migration, and cause network disconnection.
 320  */
 321 static void netvsc_send_garp(struct work_struct *w)
 322 {
 323         struct net_device_context *ndev_ctx;
 324         struct net_device *net;
 325
 326         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 327         net = dev_get_drvdata(&ndev_ctx->device_ctx->device);
 328         netif_notify_peers(net);
 329 }
 330
 331
 332 static int netvsc_probe(struct hv_device *dev)
 333 {
 334         struct net_device *net = NULL;
 335         struct net_device_context *net_device_ctx;
 336         struct netvsc_device_info device_info;
 337         int ret;
 338
 339         net = alloc_etherdev(sizeof(struct net_device_context));
 340         if (!net)
 341                 return -ENOMEM;
 342
 343         /* Set initial state */
 344         netif_carrier_off(net);
 345
 346         net_device_ctx = netdev_priv(net);
 347         net_device_ctx->device_ctx = dev;
 348         atomic_set(&net_device_ctx->avail, ring_size);
 349         dev_set_drvdata(&dev->device, net);
 350         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
 351
 352         net->netdev_ops = &device_ops;
 353
 354         /* TODO: Add GSO and Checksum offload */
 355         net->hw_features = NETIF_F_SG;
 356         net->features = NETIF_F_SG;
 357
 358         SET_ETHTOOL_OPS(net, &ethtool_ops);
 359         SET_NETDEV_DEV(net, &dev->device);
 360
 361         ret = register_netdev(net);
 362         if (ret != 0) {
 363                 pr_err("Unable to register netdev.\n");
 364                 free_netdev(net);
 365                 goto out;
 366         }
 367
 368         /* Notify the netvsc driver of the new device */
 369         device_info.ring_size = ring_size;
 370         ret = rndis_filter_device_add(dev, &device_info);
 371         if (ret != 0) {
 372                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 373                 unregister_netdev(net);
 374                 free_netdev(net);
 375                 dev_set_drvdata(&dev->device, NULL);
 376                 return ret;
 377         }
 378         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 379
 380         netif_carrier_on(net);
 381
 382 out:
 383         return ret;
 384 }
 385
 386 static int netvsc_remove(struct hv_device *dev)
 387 {
 388         struct net_device *net = dev_get_drvdata(&dev->device);
 389         struct net_device_context *ndev_ctx;
 390
 391         if (net == NULL) {
 392                 dev_err(&dev->device, "No net device to remove\n");
 393                 return 0;
 394         }
 395
 396         ndev_ctx = netdev_priv(net);
 397         cancel_delayed_work_sync(&ndev_ctx->dwork);
 398
 399         /* Stop outbound asap */
 400         netif_stop_queue(net);
 401
 402         unregister_netdev(net);
 403
 404         /*
 405          * Call to the vsc driver to let it know that the device is being
 406          * removed
 407          */
 408         rndis_filter_device_remove(dev);
 409
 410         free_netdev(net);
 411         return 0;
 412 }
 413
 414 static const struct hv_vmbus_device_id id_table[] = {
 415         /* Network guid */
 416         { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
 417                        0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
 418         { },
 419 };
 420
 421 MODULE_DEVICE_TABLE(vmbus, id_table);
 422
 423 /* The one and only one */
 424 static struct  hv_driver netvsc_drv = {
 425         .name = "netvsc",
 426         .id_table = id_table,
 427         .probe = netvsc_probe,
 428         .remove = netvsc_remove,
 429 };
 430
 431 static void __exit netvsc_drv_exit(void)
 432 {
 433         vmbus_driver_unregister(&netvsc_drv);
 434 }
 435
 436 static int __init netvsc_drv_init(void)
 437 {
 438         return vmbus_driver_register(&netvsc_drv);
 439 }
 440
 441 MODULE_LICENSE("GPL");
 442 MODULE_VERSION(HV_DRV_VERSION);
 443 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 444
 445 module_init(netvsc_drv_init);
 446 module_exit(netvsc_drv_exit);