drivers/net/xen-netback/interface.c

   1 /*
   2  * Network-device interface management.
   3  *
   4  * Copyright (c) 2004-2005, Keir Fraser
   5  *
   6  * This program is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU General Public License version 2
   8  * as published by the Free Software Foundation; or, when distributed
   9  * separately from the Linux kernel or incorporated into other
  10  * software packages, subject to the following license:
  11  *
  12  * Permission is hereby granted, free of charge, to any person obtaining a copy
  13  * of this source file (the "Software"), to deal in the Software without
  14  * restriction, including without limitation the rights to use, copy, modify,
  15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
  16  * and to permit persons to whom the Software is furnished to do so, subject to
  17  * the following conditions:
  18  *
  19  * The above copyright notice and this permission notice shall be included in
  20  * all copies or substantial portions of the Software.
  21  *
  22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  28  * IN THE SOFTWARE.
  29  */
  30
  31 #include "common.h"
  32
  33 #include <linux/kthread.h>
  34 #include <linux/ethtool.h>
  35 #include <linux/rtnetlink.h>
  36 #include <linux/if_vlan.h>
  37
  38 #include <xen/events.h>
  39 #include <asm/xen/hypercall.h>
  40
  41 #define XENVIF_QUEUE_LENGTH 32
  42 #define XENVIF_NAPI_WEIGHT  64
  43
  44 int xenvif_schedulable(struct xenvif *vif)
  45 {
  46         return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
  47 }
  48
  49 static int xenvif_rx_schedulable(struct xenvif *vif)
  50 {
  51         return xenvif_schedulable(vif) && !xenvif_rx_ring_full(vif);
  52 }
  53
  54 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
  55 {
  56         struct xenvif *vif = dev_id;
  57
  58         if (RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))
  59                 napi_schedule(&vif->napi);
  60
  61         return IRQ_HANDLED;
  62 }
  63
  64 static int xenvif_poll(struct napi_struct *napi, int budget)
  65 {
  66         struct xenvif *vif = container_of(napi, struct xenvif, napi);
  67         int work_done;
  68
  69         work_done = xenvif_tx_action(vif, budget);
  70
  71         if (work_done < budget) {
  72                 int more_to_do = 0;
  73                 unsigned long flags;
  74
  75                 /* It is necessary to disable IRQ before calling
  76                  * RING_HAS_UNCONSUMED_REQUESTS. Otherwise we might
  77                  * lose event from the frontend.
  78                  *
  79                  * Consider:
  80                  *   RING_HAS_UNCONSUMED_REQUESTS
  81                  *   <frontend generates event to trigger napi_schedule>
  82                  *   __napi_complete
  83                  *
  84                  * This handler is still in scheduled state so the
  85                  * event has no effect at all. After __napi_complete
  86                  * this handler is descheduled and cannot get
  87                  * scheduled again. We lose event in this case and the ring
  88                  * will be completely stalled.
  89                  */
  90
  91                 local_irq_save(flags);
  92
  93                 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
  94                 if (!more_to_do)
  95                         __napi_complete(napi);
  96
  97                 local_irq_restore(flags);
  98         }
  99
 100         return work_done;
 101 }
 102
 103 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
 104 {
 105         struct xenvif *vif = dev_id;
 106
 107         if (xenvif_rx_schedulable(vif))
 108                 netif_wake_queue(vif->dev);
 109
 110         return IRQ_HANDLED;
 111 }
 112
 113 static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
 114 {
 115         xenvif_tx_interrupt(irq, dev_id);
 116         xenvif_rx_interrupt(irq, dev_id);
 117
 118         return IRQ_HANDLED;
 119 }
 120
 121 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
 122 {
 123         struct xenvif *vif = netdev_priv(dev);
 124
 125         BUG_ON(skb->dev != dev);
 126
 127         /* Drop the packet if vif is not ready */
 128         if (vif->task == NULL)
 129                 goto drop;
 130
 131         /* Drop the packet if the target domain has no receive buffers. */
 132         if (!xenvif_rx_schedulable(vif))
 133                 goto drop;
 134
 135         /* Reserve ring slots for the worst-case number of fragments. */
 136         vif->rx_req_cons_peek += xenvif_count_skb_slots(vif, skb);
 137
 138         if (vif->can_queue && xenvif_must_stop_queue(vif))
 139                 netif_stop_queue(dev);
 140
 141         xenvif_queue_tx_skb(vif, skb);
 142
 143         return NETDEV_TX_OK;
 144
 145  drop:
 146         vif->dev->stats.tx_dropped++;
 147         dev_kfree_skb(skb);
 148         return NETDEV_TX_OK;
 149 }
 150
 151 void xenvif_notify_tx_completion(struct xenvif *vif)
 152 {
 153         if (netif_queue_stopped(vif->dev) && xenvif_rx_schedulable(vif))
 154                 netif_wake_queue(vif->dev);
 155 }
 156
 157 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
 158 {
 159         struct xenvif *vif = netdev_priv(dev);
 160         return &vif->dev->stats;
 161 }
 162
 163 static void xenvif_up(struct xenvif *vif)
 164 {
 165         napi_enable(&vif->napi);
 166         enable_irq(vif->tx_irq);
 167         if (vif->tx_irq != vif->rx_irq)
 168                 enable_irq(vif->rx_irq);
 169         xenvif_check_rx_xenvif(vif);
 170 }
 171
 172 static void xenvif_down(struct xenvif *vif)
 173 {
 174         napi_disable(&vif->napi);
 175         disable_irq(vif->tx_irq);
 176         if (vif->tx_irq != vif->rx_irq)
 177                 disable_irq(vif->rx_irq);
 178         del_timer_sync(&vif->credit_timeout);
 179 }
 180
 181 static int xenvif_open(struct net_device *dev)
 182 {
 183         struct xenvif *vif = netdev_priv(dev);
 184         if (netif_carrier_ok(dev))
 185                 xenvif_up(vif);
 186         netif_start_queue(dev);
 187         return 0;
 188 }
 189
 190 static int xenvif_close(struct net_device *dev)
 191 {
 192         struct xenvif *vif = netdev_priv(dev);
 193         if (netif_carrier_ok(dev))
 194                 xenvif_down(vif);
 195         netif_stop_queue(dev);
 196         return 0;
 197 }
 198
 199 static int xenvif_change_mtu(struct net_device *dev, int mtu)
 200 {
 201         struct xenvif *vif = netdev_priv(dev);
 202         int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
 203
 204         if (mtu > max)
 205                 return -EINVAL;
 206         dev->mtu = mtu;
 207         return 0;
 208 }
 209
 210 static netdev_features_t xenvif_fix_features(struct net_device *dev,
 211         netdev_features_t features)
 212 {
 213         struct xenvif *vif = netdev_priv(dev);
 214
 215         if (!vif->can_sg)
 216                 features &= ~NETIF_F_SG;
 217         if (!vif->gso && !vif->gso_prefix)
 218                 features &= ~NETIF_F_TSO;
 219         if (!vif->csum)
 220                 features &= ~NETIF_F_IP_CSUM;
 221
 222         return features;
 223 }
 224
 225 static const struct xenvif_stat {
 226         char name[ETH_GSTRING_LEN];
 227         u16 offset;
 228 } xenvif_stats[] = {
 229         {
 230                 "rx_gso_checksum_fixup",
 231                 offsetof(struct xenvif, rx_gso_checksum_fixup)
 232         },
 233 };
 234
 235 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
 236 {
 237         switch (string_set) {
 238         case ETH_SS_STATS:
 239                 return ARRAY_SIZE(xenvif_stats);
 240         default:
 241                 return -EINVAL;
 242         }
 243 }
 244
 245 static void xenvif_get_ethtool_stats(struct net_device *dev,
 246                                      struct ethtool_stats *stats, u64 * data)
 247 {
 248         void *vif = netdev_priv(dev);
 249         int i;
 250
 251         for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
 252                 data[i] = *(unsigned long *)(vif + xenvif_stats[i].offset);
 253 }
 254
 255 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
 256 {
 257         int i;
 258
 259         switch (stringset) {
 260         case ETH_SS_STATS:
 261                 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
 262                         memcpy(data + i * ETH_GSTRING_LEN,
 263                                xenvif_stats[i].name, ETH_GSTRING_LEN);
 264                 break;
 265         }
 266 }
 267
 268 static const struct ethtool_ops xenvif_ethtool_ops = {
 269         .get_link       = ethtool_op_get_link,
 270
 271         .get_sset_count = xenvif_get_sset_count,
 272         .get_ethtool_stats = xenvif_get_ethtool_stats,
 273         .get_strings = xenvif_get_strings,
 274 };
 275
 276 static const struct net_device_ops xenvif_netdev_ops = {
 277         .ndo_start_xmit = xenvif_start_xmit,
 278         .ndo_get_stats  = xenvif_get_stats,
 279         .ndo_open       = xenvif_open,
 280         .ndo_stop       = xenvif_close,
 281         .ndo_change_mtu = xenvif_change_mtu,
 282         .ndo_fix_features = xenvif_fix_features,
 283         .ndo_set_mac_address = eth_mac_addr,
 284         .ndo_validate_addr   = eth_validate_addr,
 285 };
 286
 287 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
 288                             unsigned int handle)
 289 {
 290         int err;
 291         struct net_device *dev;
 292         struct xenvif *vif;
 293         char name[IFNAMSIZ] = {};
 294         int i;
 295
 296         snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
 297         dev = alloc_netdev(sizeof(struct xenvif), name, ether_setup);
 298         if (dev == NULL) {
 299                 pr_warn("Could not allocate netdev for %s\n", name);
 300                 return ERR_PTR(-ENOMEM);
 301         }
 302
 303         SET_NETDEV_DEV(dev, parent);
 304
 305         vif = netdev_priv(dev);
 306         vif->domid  = domid;
 307         vif->handle = handle;
 308         vif->can_sg = 1;
 309         vif->csum = 1;
 310         vif->dev = dev;
 311
 312         vif->credit_bytes = vif->remaining_credit = ~0UL;
 313         vif->credit_usec  = 0UL;
 314         init_timer(&vif->credit_timeout);
 315         /* Initialize 'expires' now: it's used to track the credit window. */
 316         vif->credit_timeout.expires = jiffies;
 317
 318         dev->netdev_ops = &xenvif_netdev_ops;
 319         dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
 320         dev->features = dev->hw_features;
 321         SET_ETHTOOL_OPS(dev, &xenvif_ethtool_ops);
 322
 323         dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
 324
 325         skb_queue_head_init(&vif->rx_queue);
 326         skb_queue_head_init(&vif->tx_queue);
 327
 328         vif->pending_cons = 0;
 329         vif->pending_prod = MAX_PENDING_REQS;
 330         for (i = 0; i < MAX_PENDING_REQS; i++)
 331                 vif->pending_ring[i] = i;
 332         for (i = 0; i < MAX_PENDING_REQS; i++)
 333                 vif->mmap_pages[i] = NULL;
 334
 335         /*
 336          * Initialise a dummy MAC address. We choose the numerically
 337          * largest non-broadcast address to prevent the address getting
 338          * stolen by an Ethernet bridge for STP purposes.
 339          * (FE:FF:FF:FF:FF:FF)
 340          */
 341         memset(dev->dev_addr, 0xFF, ETH_ALEN);
 342         dev->dev_addr[0] &= ~0x01;
 343
 344         netif_napi_add(dev, &vif->napi, xenvif_poll, XENVIF_NAPI_WEIGHT);
 345
 346         netif_carrier_off(dev);
 347
 348         err = register_netdev(dev);
 349         if (err) {
 350                 netdev_warn(dev, "Could not register device: err=%d\n", err);
 351                 free_netdev(dev);
 352                 return ERR_PTR(err);
 353         }
 354
 355         netdev_dbg(dev, "Successfully created xenvif\n");
 356
 357         __module_get(THIS_MODULE);
 358
 359         return vif;
 360 }
 361
 362 int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
 363                    unsigned long rx_ring_ref, unsigned int tx_evtchn,
 364                    unsigned int rx_evtchn)
 365 {
 366         int err = -ENOMEM;
 367
 368         /* Already connected through? */
 369         if (vif->tx_irq)
 370                 return 0;
 371
 372         err = xenvif_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
 373         if (err < 0)
 374                 goto err;
 375
 376         if (tx_evtchn == rx_evtchn) {
 377                 /* feature-split-event-channels == 0 */
 378                 err = bind_interdomain_evtchn_to_irqhandler(
 379                         vif->domid, tx_evtchn, xenvif_interrupt, 0,
 380                         vif->dev->name, vif);
 381                 if (err < 0)
 382                         goto err_unmap;
 383                 vif->tx_irq = vif->rx_irq = err;
 384                 disable_irq(vif->tx_irq);
 385         } else {
 386                 /* feature-split-event-channels == 1 */
 387                 snprintf(vif->tx_irq_name, sizeof(vif->tx_irq_name),
 388                          "%s-tx", vif->dev->name);
 389                 err = bind_interdomain_evtchn_to_irqhandler(
 390                         vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
 391                         vif->tx_irq_name, vif);
 392                 if (err < 0)
 393                         goto err_unmap;
 394                 vif->tx_irq = err;
 395                 disable_irq(vif->tx_irq);
 396
 397                 snprintf(vif->rx_irq_name, sizeof(vif->rx_irq_name),
 398                          "%s-rx", vif->dev->name);
 399                 err = bind_interdomain_evtchn_to_irqhandler(
 400                         vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
 401                         vif->rx_irq_name, vif);
 402                 if (err < 0)
 403                         goto err_tx_unbind;
 404                 vif->rx_irq = err;
 405                 disable_irq(vif->rx_irq);
 406         }
 407
 408         init_waitqueue_head(&vif->wq);
 409         vif->task = kthread_create(xenvif_kthread,
 410                                    (void *)vif, "%s", vif->dev->name);
 411         if (IS_ERR(vif->task)) {
 412                 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
 413                 err = PTR_ERR(vif->task);
 414                 goto err_rx_unbind;
 415         }
 416
 417         rtnl_lock();
 418         if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
 419                 dev_set_mtu(vif->dev, ETH_DATA_LEN);
 420         netdev_update_features(vif->dev);
 421         netif_carrier_on(vif->dev);
 422         if (netif_running(vif->dev))
 423                 xenvif_up(vif);
 424         rtnl_unlock();
 425
 426         wake_up_process(vif->task);
 427
 428         return 0;
 429
 430 err_rx_unbind:
 431         unbind_from_irqhandler(vif->rx_irq, vif);
 432         vif->rx_irq = 0;
 433 err_tx_unbind:
 434         unbind_from_irqhandler(vif->tx_irq, vif);
 435         vif->tx_irq = 0;
 436 err_unmap:
 437         xenvif_unmap_frontend_rings(vif);
 438 err:
 439         module_put(THIS_MODULE);
 440         return err;
 441 }
 442
 443 void xenvif_carrier_off(struct xenvif *vif)
 444 {
 445         struct net_device *dev = vif->dev;
 446
 447         rtnl_lock();
 448         netif_carrier_off(dev); /* discard queued packets */
 449         if (netif_running(dev))
 450                 xenvif_down(vif);
 451         rtnl_unlock();
 452 }
 453
 454 void xenvif_disconnect(struct xenvif *vif)
 455 {
 456         if (netif_carrier_ok(vif->dev))
 457                 xenvif_carrier_off(vif);
 458
 459         if (vif->tx_irq) {
 460                 if (vif->tx_irq == vif->rx_irq)
 461                         unbind_from_irqhandler(vif->tx_irq, vif);
 462                 else {
 463                         unbind_from_irqhandler(vif->tx_irq, vif);
 464                         unbind_from_irqhandler(vif->rx_irq, vif);
 465                 }
 466                 vif->tx_irq = 0;
 467         }
 468
 469         if (vif->task)
 470                 kthread_stop(vif->task);
 471
 472         xenvif_unmap_frontend_rings(vif);
 473 }
 474
 475 void xenvif_free(struct xenvif *vif)
 476 {
 477         netif_napi_del(&vif->napi);
 478
 479         unregister_netdev(vif->dev);
 480
 481         free_netdev(vif->dev);
 482
 483         module_put(THIS_MODULE);
 484 }