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         /* Open up the device */
  73         ret = rndis_filter_open(device_obj);
  74         if (ret != 0) {
  75                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  76                 return ret;
  77         }
  78
  79         netif_start_queue(net);
  80
  81         return ret;
  82 }
  83
  84 static int netvsc_close(struct net_device *net)
  85 {
  86         struct net_device_context *net_device_ctx = netdev_priv(net);
  87         struct hv_device *device_obj = net_device_ctx->device_ctx;
  88         int ret;
  89
  90         netif_stop_queue(net);
  91
  92         ret = rndis_filter_close(device_obj);
  93         if (ret != 0)
  94                 netdev_err(net, "unable to close device (ret %d).\n", ret);
  95
  96         return ret;
  97 }
  98
  99 static void netvsc_xmit_completion(void *context)
 100 {
 101         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 102         struct sk_buff *skb = (struct sk_buff *)
 103                 (unsigned long)packet->completion.send.send_completion_tid;
 104
 105         kfree(packet);
 106
 107         if (skb) {
 108                 struct net_device *net = skb->dev;
 109                 struct net_device_context *net_device_ctx = netdev_priv(net);
 110                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
 111
 112                 dev_kfree_skb_any(skb);
 113
 114                 atomic_add(num_pages, &net_device_ctx->avail);
 115                 if (atomic_read(&net_device_ctx->avail) >=
 116                                 PACKET_PAGES_HIWATER)
 117                         netif_wake_queue(net);
 118         }
 119 }
 120
 121 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 122 {
 123         struct net_device_context *net_device_ctx = netdev_priv(net);
 124         struct hv_netvsc_packet *packet;
 125         int ret;
 126         unsigned int i, num_pages;
 127
 128         /* Add 1 for skb->data and additional one for RNDIS */
 129         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
 130         if (num_pages > atomic_read(&net_device_ctx->avail))
 131                 return NETDEV_TX_BUSY;
 132
 133         /* Allocate a netvsc packet based on # of frags. */
 134         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 135                          (num_pages * sizeof(struct hv_page_buffer)) +
 136                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
 137         if (!packet) {
 138                 /* out of memory, drop packet */
 139                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 140
 141                 dev_kfree_skb(skb);
 142                 net->stats.tx_dropped++;
 143                 return NETDEV_TX_BUSY;
 144         }
 145
 146         packet->extension = (void *)(unsigned long)packet +
 147                                 sizeof(struct hv_netvsc_packet) +
 148                                     (num_pages * sizeof(struct hv_page_buffer));
 149
 150         /* Setup the rndis header */
 151         packet->page_buf_cnt = num_pages;
 152
 153         /* Initialize it from the skb */
 154         packet->total_data_buflen       = skb->len;
 155
 156         /* Start filling in the page buffers starting after RNDIS buffer. */
 157         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 158         packet->page_buf[1].offset
 159                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 160         packet->page_buf[1].len = skb_headlen(skb);
 161
 162         /* Additional fragments are after SKB data */
 163         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 164                 const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 165
 166                 packet->page_buf[i+2].pfn = page_to_pfn(skb_frag_page(f));
 167                 packet->page_buf[i+2].offset = f->page_offset;
 168                 packet->page_buf[i+2].len = skb_frag_size(f);
 169         }
 170
 171         /* Set the completion routine */
 172         packet->completion.send.send_completion = netvsc_xmit_completion;
 173         packet->completion.send.send_completion_ctx = packet;
 174         packet->completion.send.send_completion_tid = (unsigned long)skb;
 175
 176         ret = rndis_filter_send(net_device_ctx->device_ctx,
 177                                   packet);
 178         if (ret == 0) {
 179                 net->stats.tx_bytes += skb->len;
 180                 net->stats.tx_packets++;
 181
 182                 atomic_sub(num_pages, &net_device_ctx->avail);
 183                 if (atomic_read(&net_device_ctx->avail) < PACKET_PAGES_LOWATER)
 184                         netif_stop_queue(net);
 185         } else {
 186                 /* we are shutting down or bus overloaded, just drop packet */
 187                 net->stats.tx_dropped++;
 188                 kfree(packet);
 189                 dev_kfree_skb_any(skb);
 190         }
 191
 192         return ret ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 193 }
 194
 195 /*
 196  * netvsc_linkstatus_callback - Link up/down notification
 197  */
 198 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 199                                        unsigned int status)
 200 {
 201         struct net_device *net;
 202         struct net_device_context *ndev_ctx;
 203         struct netvsc_device *net_device;
 204
 205         net_device = hv_get_drvdata(device_obj);
 206         net = net_device->ndev;
 207
 208         if (!net) {
 209                 netdev_err(net, "got link status but net device "
 210                                 "not initialized yet\n");
 211                 return;
 212         }
 213
 214         if (status == 1) {
 215                 netif_carrier_on(net);
 216                 netif_wake_queue(net);
 217                 ndev_ctx = netdev_priv(net);
 218                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 219                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 220         } else {
 221                 netif_carrier_off(net);
 222                 netif_stop_queue(net);
 223         }
 224 }
 225
 226 /*
 227  * netvsc_recv_callback -  Callback when we receive a packet from the
 228  * "wire" on the specified device.
 229  */
 230 int netvsc_recv_callback(struct hv_device *device_obj,
 231                                 struct hv_netvsc_packet *packet)
 232 {
 233         struct net_device *net = dev_get_drvdata(&device_obj->device);
 234         struct sk_buff *skb;
 235         void *data;
 236         int i;
 237         unsigned long flags;
 238         struct netvsc_device *net_device;
 239
 240         net_device = hv_get_drvdata(device_obj);
 241         net = net_device->ndev;
 242
 243         if (!net) {
 244                 netdev_err(net, "got receive callback but net device"
 245                         " not initialized yet\n");
 246                 return 0;
 247         }
 248
 249         /* Allocate a skb - TODO direct I/O to pages? */
 250         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 251         if (unlikely(!skb)) {
 252                 ++net->stats.rx_dropped;
 253                 return 0;
 254         }
 255
 256         /* for kmap_atomic */
 257         local_irq_save(flags);
 258
 259         /*
 260          * Copy to skb. This copy is needed here since the memory pointed by
 261          * hv_netvsc_packet cannot be deallocated
 262          */
 263         for (i = 0; i < packet->page_buf_cnt; i++) {
 264                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
 265                                                KM_IRQ1);
 266                 data = (void *)(unsigned long)data +
 267                                 packet->page_buf[i].offset;
 268
 269                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
 270                        packet->page_buf[i].len);
 271
 272                 kunmap_atomic((void *)((unsigned long)data -
 273                                        packet->page_buf[i].offset), KM_IRQ1);
 274         }
 275
 276         local_irq_restore(flags);
 277
 278         skb->protocol = eth_type_trans(skb, net);
 279         skb->ip_summed = CHECKSUM_NONE;
 280
 281         net->stats.rx_packets++;
 282         net->stats.rx_bytes += skb->len;
 283
 284         /*
 285          * Pass the skb back up. Network stack will deallocate the skb when it
 286          * is done.
 287          * TODO - use NAPI?
 288          */
 289         netif_rx(skb);
 290
 291         return 0;
 292 }
 293
 294 static void netvsc_get_drvinfo(struct net_device *net,
 295                                struct ethtool_drvinfo *info)
 296 {
 297         strcpy(info->driver, "hv_netvsc");
 298         strcpy(info->version, HV_DRV_VERSION);
 299         strcpy(info->fw_version, "N/A");
 300 }
 301
 302 static const struct ethtool_ops ethtool_ops = {
 303         .get_drvinfo    = netvsc_get_drvinfo,
 304         .get_link       = ethtool_op_get_link,
 305 };
 306
 307 static const struct net_device_ops device_ops = {
 308         .ndo_open =                     netvsc_open,
 309         .ndo_stop =                     netvsc_close,
 310         .ndo_start_xmit =               netvsc_start_xmit,
 311         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 312         .ndo_change_mtu =               eth_change_mtu,
 313         .ndo_validate_addr =            eth_validate_addr,
 314         .ndo_set_mac_address =          eth_mac_addr,
 315 };
 316
 317 /*
 318  * Send GARP packet to network peers after migrations.
 319  * After Quick Migration, the network is not immediately operational in the
 320  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 321  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 322  * will not be sent after quick migration, and cause network disconnection.
 323  */
 324 static void netvsc_send_garp(struct work_struct *w)
 325 {
 326         struct net_device_context *ndev_ctx;
 327         struct net_device *net;
 328         struct netvsc_device *net_device;
 329
 330         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 331         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 332         net = net_device->ndev;
 333         netif_notify_peers(net);
 334 }
 335
 336
 337 static int netvsc_probe(struct hv_device *dev,
 338                         const struct hv_vmbus_device_id *dev_id)
 339 {
 340         struct net_device *net = NULL;
 341         struct net_device_context *net_device_ctx;
 342         struct netvsc_device_info device_info;
 343         int ret;
 344
 345         net = alloc_etherdev(sizeof(struct net_device_context));
 346         if (!net)
 347                 return -ENOMEM;
 348
 349         /* Set initial state */
 350         netif_carrier_off(net);
 351
 352         net_device_ctx = netdev_priv(net);
 353         net_device_ctx->device_ctx = dev;
 354         atomic_set(&net_device_ctx->avail, ring_size);
 355         hv_set_drvdata(dev, net);
 356         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
 357
 358         net->netdev_ops = &device_ops;
 359
 360         /* TODO: Add GSO and Checksum offload */
 361         net->hw_features = NETIF_F_SG;
 362         net->features = NETIF_F_SG;
 363
 364         SET_ETHTOOL_OPS(net, &ethtool_ops);
 365         SET_NETDEV_DEV(net, &dev->device);
 366
 367         ret = register_netdev(net);
 368         if (ret != 0) {
 369                 pr_err("Unable to register netdev.\n");
 370                 free_netdev(net);
 371                 goto out;
 372         }
 373
 374         /* Notify the netvsc driver of the new device */
 375         device_info.ring_size = ring_size;
 376         ret = rndis_filter_device_add(dev, &device_info);
 377         if (ret != 0) {
 378                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 379                 unregister_netdev(net);
 380                 free_netdev(net);
 381                 hv_set_drvdata(dev, NULL);
 382                 return ret;
 383         }
 384         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 385
 386         netif_carrier_on(net);
 387
 388 out:
 389         return ret;
 390 }
 391
 392 static int netvsc_remove(struct hv_device *dev)
 393 {
 394         struct net_device *net;
 395         struct net_device_context *ndev_ctx;
 396         struct netvsc_device *net_device;
 397
 398         net_device = hv_get_drvdata(dev);
 399         net = net_device->ndev;
 400
 401         if (net == NULL) {
 402                 dev_err(&dev->device, "No net device to remove\n");
 403                 return 0;
 404         }
 405
 406         ndev_ctx = netdev_priv(net);
 407         cancel_delayed_work_sync(&ndev_ctx->dwork);
 408
 409         /* Stop outbound asap */
 410         netif_stop_queue(net);
 411
 412         unregister_netdev(net);
 413
 414         /*
 415          * Call to the vsc driver to let it know that the device is being
 416          * removed
 417          */
 418         rndis_filter_device_remove(dev);
 419
 420         free_netdev(net);
 421         return 0;
 422 }
 423
 424 static const struct hv_vmbus_device_id id_table[] = {
 425         /* Network guid */
 426         { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
 427                        0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
 428         { },
 429 };
 430
 431 MODULE_DEVICE_TABLE(vmbus, id_table);
 432
 433 /* The one and only one */
 434 static struct  hv_driver netvsc_drv = {
 435         .name = "netvsc",
 436         .id_table = id_table,
 437         .probe = netvsc_probe,
 438         .remove = netvsc_remove,
 439 };
 440
 441 static void __exit netvsc_drv_exit(void)
 442 {
 443         vmbus_driver_unregister(&netvsc_drv);
 444 }
 445
 446 static int __init netvsc_drv_init(void)
 447 {
 448         return vmbus_driver_register(&netvsc_drv);
 449 }
 450
 451 MODULE_LICENSE("GPL");
 452 MODULE_VERSION(HV_DRV_VERSION);
 453 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 454
 455 module_init(netvsc_drv_init);
 456 module_exit(netvsc_drv_exit);