drivers/net/xen-netback/xenbus.c

   1 /*
   2  * Xenbus code for netif backend
   3  *
   4  * Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   5  * Copyright (C) 2005 XenSource Ltd
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "common.h"
  22 #include <linux/vmalloc.h>
  23 #include <linux/rtnetlink.h>
  24
  25 struct backend_info {
  26         struct xenbus_device *dev;
  27         struct xenvif *vif;
  28
  29         /* This is the state that will be reflected in xenstore when any
  30          * active hotplug script completes.
  31          */
  32         enum xenbus_state state;
  33
  34         enum xenbus_state frontend_state;
  35         struct xenbus_watch hotplug_status_watch;
  36         u8 have_hotplug_status_watch:1;
  37 };
  38
  39 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue);
  40 static void connect(struct backend_info *be);
  41 static int read_xenbus_vif_flags(struct backend_info *be);
  42 static void backend_create_xenvif(struct backend_info *be);
  43 static void unregister_hotplug_status_watch(struct backend_info *be);
  44 static void set_backend_state(struct backend_info *be,
  45                               enum xenbus_state state);
  46
  47 #ifdef CONFIG_DEBUG_FS
  48 struct dentry *xen_netback_dbg_root = NULL;
  49
  50 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  51 {
  52         struct xenvif_queue *queue = m->private;
  53         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  54         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  55         struct netdev_queue *dev_queue;
  56
  57         if (tx_ring->sring) {
  58                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  59
  60                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  61                            tx_ring->nr_ents);
  62                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  63                            sring->req_prod,
  64                            sring->req_prod - sring->rsp_prod,
  65                            tx_ring->req_cons,
  66                            tx_ring->req_cons - sring->rsp_prod,
  67                            sring->req_event,
  68                            sring->req_event - sring->rsp_prod);
  69                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  70                            sring->rsp_prod,
  71                            tx_ring->rsp_prod_pvt,
  72                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  73                            sring->rsp_event,
  74                            sring->rsp_event - sring->rsp_prod);
  75                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  76                            queue->pending_prod,
  77                            queue->pending_cons,
  78                            nr_pending_reqs(queue));
  79                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  80                            queue->dealloc_prod,
  81                            queue->dealloc_cons,
  82                            queue->dealloc_prod - queue->dealloc_cons);
  83         }
  84
  85         if (rx_ring->sring) {
  86                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  87
  88                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  89                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  90                            sring->req_prod,
  91                            sring->req_prod - sring->rsp_prod,
  92                            rx_ring->req_cons,
  93                            rx_ring->req_cons - sring->rsp_prod,
  94                            sring->req_event,
  95                            sring->req_event - sring->rsp_prod);
  96                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
  97                            sring->rsp_prod,
  98                            rx_ring->rsp_prod_pvt,
  99                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
 100                            sring->rsp_event,
 101                            sring->rsp_event - sring->rsp_prod);
 102         }
 103
 104         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 105                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 106                    "remaining: %lu, expires: %lu, now: %lu\n",
 107                    queue->napi.state, queue->napi.weight,
 108                    skb_queue_len(&queue->tx_queue),
 109                    timer_pending(&queue->credit_timeout),
 110                    queue->credit_bytes,
 111                    queue->credit_usec,
 112                    queue->remaining_credit,
 113                    queue->credit_timeout.expires,
 114                    jiffies);
 115
 116         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
 117
 118         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
 119                    queue->rx_queue_len, queue->rx_queue_max,
 120                    skb_queue_len(&queue->rx_queue),
 121                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
 122
 123         return 0;
 124 }
 125
 126 #define XENVIF_KICK_STR "kick"
 127 #define BUFFER_SIZE     32
 128
 129 static ssize_t
 130 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 131                      loff_t *ppos)
 132 {
 133         struct xenvif_queue *queue =
 134                 ((struct seq_file *)filp->private_data)->private;
 135         int len;
 136         char write[BUFFER_SIZE];
 137
 138         /* don't allow partial writes and check the length */
 139         if (*ppos != 0)
 140                 return 0;
 141         if (count >= sizeof(write))
 142                 return -ENOSPC;
 143
 144         len = simple_write_to_buffer(write,
 145                                      sizeof(write) - 1,
 146                                      ppos,
 147                                      buf,
 148                                      count);
 149         if (len < 0)
 150                 return len;
 151
 152         write[len] = '\0';
 153
 154         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 155                 xenvif_interrupt(0, (void *)queue);
 156         else {
 157                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 158                         queue->id);
 159                 count = -EINVAL;
 160         }
 161         return count;
 162 }
 163
 164 static int xenvif_dump_open(struct inode *inode, struct file *filp)
 165 {
 166         int ret;
 167         void *queue = NULL;
 168
 169         if (inode->i_private)
 170                 queue = inode->i_private;
 171         ret = single_open(filp, xenvif_read_io_ring, queue);
 172         filp->f_mode |= FMODE_PWRITE;
 173         return ret;
 174 }
 175
 176 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 177         .owner = THIS_MODULE,
 178         .open = xenvif_dump_open,
 179         .read = seq_read,
 180         .llseek = seq_lseek,
 181         .release = single_release,
 182         .write = xenvif_write_io_ring,
 183 };
 184
 185 static void xenvif_debugfs_addif(struct xenvif *vif)
 186 {
 187         struct dentry *pfile;
 188         int i;
 189
 190         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 191                 return;
 192
 193         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 194                                                   xen_netback_dbg_root);
 195         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 196                 for (i = 0; i < vif->num_queues; ++i) {
 197                         char filename[sizeof("io_ring_q") + 4];
 198
 199                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 200                         pfile = debugfs_create_file(filename,
 201                                                     S_IRUSR | S_IWUSR,
 202                                                     vif->xenvif_dbg_root,
 203                                                     &vif->queues[i],
 204                                                     &xenvif_dbg_io_ring_ops_fops);
 205                         if (IS_ERR_OR_NULL(pfile))
 206                                 pr_warn("Creation of io_ring file returned %ld!\n",
 207                                         PTR_ERR(pfile));
 208                 }
 209         } else
 210                 netdev_warn(vif->dev,
 211                             "Creation of vif debugfs dir returned %ld!\n",
 212                             PTR_ERR(vif->xenvif_dbg_root));
 213 }
 214
 215 static void xenvif_debugfs_delif(struct xenvif *vif)
 216 {
 217         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 218                 return;
 219
 220         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 221                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 222         vif->xenvif_dbg_root = NULL;
 223 }
 224 #endif /* CONFIG_DEBUG_FS */
 225
 226 static int netback_remove(struct xenbus_device *dev)
 227 {
 228         struct backend_info *be = dev_get_drvdata(&dev->dev);
 229
 230         set_backend_state(be, XenbusStateClosed);
 231
 232         unregister_hotplug_status_watch(be);
 233         if (be->vif) {
 234                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 235                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 236                 xenvif_free(be->vif);
 237                 be->vif = NULL;
 238         }
 239         kfree(be);
 240         dev_set_drvdata(&dev->dev, NULL);
 241         return 0;
 242 }
 243
 244
 245 /**
 246  * Entry point to this code when a new device is created.  Allocate the basic
 247  * structures and switch to InitWait.
 248  */
 249 static int netback_probe(struct xenbus_device *dev,
 250                          const struct xenbus_device_id *id)
 251 {
 252         const char *message;
 253         struct xenbus_transaction xbt;
 254         int err;
 255         int sg;
 256         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 257                                           GFP_KERNEL);
 258         if (!be) {
 259                 xenbus_dev_fatal(dev, -ENOMEM,
 260                                  "allocating backend structure");
 261                 return -ENOMEM;
 262         }
 263
 264         be->dev = dev;
 265         dev_set_drvdata(&dev->dev, be);
 266
 267         sg = 1;
 268
 269         do {
 270                 err = xenbus_transaction_start(&xbt);
 271                 if (err) {
 272                         xenbus_dev_fatal(dev, err, "starting transaction");
 273                         goto fail;
 274                 }
 275
 276                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 277                 if (err) {
 278                         message = "writing feature-sg";
 279                         goto abort_transaction;
 280                 }
 281
 282                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 283                                     "%d", sg);
 284                 if (err) {
 285                         message = "writing feature-gso-tcpv4";
 286                         goto abort_transaction;
 287                 }
 288
 289                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 290                                     "%d", sg);
 291                 if (err) {
 292                         message = "writing feature-gso-tcpv6";
 293                         goto abort_transaction;
 294                 }
 295
 296                 /* We support partial checksum setup for IPv6 packets */
 297                 err = xenbus_printf(xbt, dev->nodename,
 298                                     "feature-ipv6-csum-offload",
 299                                     "%d", 1);
 300                 if (err) {
 301                         message = "writing feature-ipv6-csum-offload";
 302                         goto abort_transaction;
 303                 }
 304
 305                 /* We support rx-copy path. */
 306                 err = xenbus_printf(xbt, dev->nodename,
 307                                     "feature-rx-copy", "%d", 1);
 308                 if (err) {
 309                         message = "writing feature-rx-copy";
 310                         goto abort_transaction;
 311                 }
 312
 313                 /*
 314                  * We don't support rx-flip path (except old guests who don't
 315                  * grok this feature flag).
 316                  */
 317                 err = xenbus_printf(xbt, dev->nodename,
 318                                     "feature-rx-flip", "%d", 0);
 319                 if (err) {
 320                         message = "writing feature-rx-flip";
 321                         goto abort_transaction;
 322                 }
 323
 324                 err = xenbus_transaction_end(xbt, 0);
 325         } while (err == -EAGAIN);
 326
 327         if (err) {
 328                 xenbus_dev_fatal(dev, err, "completing transaction");
 329                 goto fail;
 330         }
 331
 332         /*
 333          * Split event channels support, this is optional so it is not
 334          * put inside the above loop.
 335          */
 336         err = xenbus_printf(XBT_NIL, dev->nodename,
 337                             "feature-split-event-channels",
 338                             "%u", separate_tx_rx_irq);
 339         if (err)
 340                 pr_debug("Error writing feature-split-event-channels\n");
 341
 342         /* Multi-queue support: This is an optional feature. */
 343         err = xenbus_printf(XBT_NIL, dev->nodename,
 344                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 345         if (err)
 346                 pr_debug("Error writing multi-queue-max-queues\n");
 347
 348         err = xenbus_switch_state(dev, XenbusStateInitWait);
 349         if (err)
 350                 goto fail;
 351
 352         be->state = XenbusStateInitWait;
 353
 354         /* This kicks hotplug scripts, so do it immediately. */
 355         backend_create_xenvif(be);
 356
 357         return 0;
 358
 359 abort_transaction:
 360         xenbus_transaction_end(xbt, 1);
 361         xenbus_dev_fatal(dev, err, "%s", message);
 362 fail:
 363         pr_debug("failed\n");
 364         netback_remove(dev);
 365         return err;
 366 }
 367
 368
 369 /*
 370  * Handle the creation of the hotplug script environment.  We add the script
 371  * and vif variables to the environment, for the benefit of the vif-* hotplug
 372  * scripts.
 373  */
 374 static int netback_uevent(struct xenbus_device *xdev,
 375                           struct kobj_uevent_env *env)
 376 {
 377         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 378         char *val;
 379
 380         val = xenbus_read(XBT_NIL, xdev->nodename, "script", NULL);
 381         if (IS_ERR(val)) {
 382                 int err = PTR_ERR(val);
 383                 xenbus_dev_fatal(xdev, err, "reading script");
 384                 return err;
 385         } else {
 386                 if (add_uevent_var(env, "script=%s", val)) {
 387                         kfree(val);
 388                         return -ENOMEM;
 389                 }
 390                 kfree(val);
 391         }
 392
 393         if (!be || !be->vif)
 394                 return 0;
 395
 396         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 397 }
 398
 399
 400 static void backend_create_xenvif(struct backend_info *be)
 401 {
 402         int err;
 403         long handle;
 404         struct xenbus_device *dev = be->dev;
 405
 406         if (be->vif != NULL)
 407                 return;
 408
 409         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 410         if (err != 1) {
 411                 xenbus_dev_fatal(dev, err, "reading handle");
 412                 return;
 413         }
 414
 415         be->vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 416         if (IS_ERR(be->vif)) {
 417                 err = PTR_ERR(be->vif);
 418                 be->vif = NULL;
 419                 xenbus_dev_fatal(dev, err, "creating interface");
 420                 return;
 421         }
 422
 423         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 424 }
 425
 426 static void backend_disconnect(struct backend_info *be)
 427 {
 428         if (be->vif) {
 429 #ifdef CONFIG_DEBUG_FS
 430                 xenvif_debugfs_delif(be->vif);
 431 #endif /* CONFIG_DEBUG_FS */
 432                 xenvif_disconnect(be->vif);
 433         }
 434 }
 435
 436 static void backend_connect(struct backend_info *be)
 437 {
 438         if (be->vif)
 439                 connect(be);
 440 }
 441
 442 static inline void backend_switch_state(struct backend_info *be,
 443                                         enum xenbus_state state)
 444 {
 445         struct xenbus_device *dev = be->dev;
 446
 447         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 448         be->state = state;
 449
 450         /* If we are waiting for a hotplug script then defer the
 451          * actual xenbus state change.
 452          */
 453         if (!be->have_hotplug_status_watch)
 454                 xenbus_switch_state(dev, state);
 455 }
 456
 457 /* Handle backend state transitions:
 458  *
 459  * The backend state starts in InitWait and the following transitions are
 460  * allowed.
 461  *
 462  * InitWait -> Connected
 463  *
 464  *    ^    \         |
 465  *    |     \        |
 466  *    |      \       |
 467  *    |       \      |
 468  *    |        \     |
 469  *    |         \    |
 470  *    |          V   V
 471  *
 472  *  Closed  <-> Closing
 473  *
 474  * The state argument specifies the eventual state of the backend and the
 475  * function transitions to that state via the shortest path.
 476  */
 477 static void set_backend_state(struct backend_info *be,
 478                               enum xenbus_state state)
 479 {
 480         while (be->state != state) {
 481                 switch (be->state) {
 482                 case XenbusStateClosed:
 483                         switch (state) {
 484                         case XenbusStateInitWait:
 485                         case XenbusStateConnected:
 486                                 pr_info("%s: prepare for reconnect\n",
 487                                         be->dev->nodename);
 488                                 backend_switch_state(be, XenbusStateInitWait);
 489                                 break;
 490                         case XenbusStateClosing:
 491                                 backend_switch_state(be, XenbusStateClosing);
 492                                 break;
 493                         default:
 494                                 BUG();
 495                         }
 496                         break;
 497                 case XenbusStateInitWait:
 498                         switch (state) {
 499                         case XenbusStateConnected:
 500                                 backend_connect(be);
 501                                 backend_switch_state(be, XenbusStateConnected);
 502                                 break;
 503                         case XenbusStateClosing:
 504                         case XenbusStateClosed:
 505                                 backend_switch_state(be, XenbusStateClosing);
 506                                 break;
 507                         default:
 508                                 BUG();
 509                         }
 510                         break;
 511                 case XenbusStateConnected:
 512                         switch (state) {
 513                         case XenbusStateInitWait:
 514                         case XenbusStateClosing:
 515                         case XenbusStateClosed:
 516                                 backend_disconnect(be);
 517                                 backend_switch_state(be, XenbusStateClosing);
 518                                 break;
 519                         default:
 520                                 BUG();
 521                         }
 522                         break;
 523                 case XenbusStateClosing:
 524                         switch (state) {
 525                         case XenbusStateInitWait:
 526                         case XenbusStateConnected:
 527                         case XenbusStateClosed:
 528                                 backend_switch_state(be, XenbusStateClosed);
 529                                 break;
 530                         default:
 531                                 BUG();
 532                         }
 533                         break;
 534                 default:
 535                         BUG();
 536                 }
 537         }
 538 }
 539
 540 /**
 541  * Callback received when the frontend's state changes.
 542  */
 543 static void frontend_changed(struct xenbus_device *dev,
 544                              enum xenbus_state frontend_state)
 545 {
 546         struct backend_info *be = dev_get_drvdata(&dev->dev);
 547
 548         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 549
 550         be->frontend_state = frontend_state;
 551
 552         switch (frontend_state) {
 553         case XenbusStateInitialising:
 554                 set_backend_state(be, XenbusStateInitWait);
 555                 break;
 556
 557         case XenbusStateInitialised:
 558                 break;
 559
 560         case XenbusStateConnected:
 561                 set_backend_state(be, XenbusStateConnected);
 562                 break;
 563
 564         case XenbusStateClosing:
 565                 set_backend_state(be, XenbusStateClosing);
 566                 break;
 567
 568         case XenbusStateClosed:
 569                 set_backend_state(be, XenbusStateClosed);
 570                 if (xenbus_dev_is_online(dev))
 571                         break;
 572                 /* fall through if not online */
 573         case XenbusStateUnknown:
 574                 set_backend_state(be, XenbusStateClosed);
 575                 device_unregister(&dev->dev);
 576                 break;
 577
 578         default:
 579                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 580                                  frontend_state);
 581                 break;
 582         }
 583 }
 584
 585
 586 static void xen_net_read_rate(struct xenbus_device *dev,
 587                               unsigned long *bytes, unsigned long *usec)
 588 {
 589         char *s, *e;
 590         unsigned long b, u;
 591         char *ratestr;
 592
 593         /* Default to unlimited bandwidth. */
 594         *bytes = ~0UL;
 595         *usec = 0;
 596
 597         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 598         if (IS_ERR(ratestr))
 599                 return;
 600
 601         s = ratestr;
 602         b = simple_strtoul(s, &e, 10);
 603         if ((s == e) || (*e != ','))
 604                 goto fail;
 605
 606         s = e + 1;
 607         u = simple_strtoul(s, &e, 10);
 608         if ((s == e) || (*e != '\0'))
 609                 goto fail;
 610
 611         *bytes = b;
 612         *usec = u;
 613
 614         kfree(ratestr);
 615         return;
 616
 617  fail:
 618         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 619         kfree(ratestr);
 620 }
 621
 622 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 623 {
 624         char *s, *e, *macstr;
 625         int i;
 626
 627         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 628         if (IS_ERR(macstr))
 629                 return PTR_ERR(macstr);
 630
 631         for (i = 0; i < ETH_ALEN; i++) {
 632                 mac[i] = simple_strtoul(s, &e, 16);
 633                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 634                         kfree(macstr);
 635                         return -ENOENT;
 636                 }
 637                 s = e+1;
 638         }
 639
 640         kfree(macstr);
 641         return 0;
 642 }
 643
 644 static void unregister_hotplug_status_watch(struct backend_info *be)
 645 {
 646         if (be->have_hotplug_status_watch) {
 647                 unregister_xenbus_watch(&be->hotplug_status_watch);
 648                 kfree(be->hotplug_status_watch.node);
 649         }
 650         be->have_hotplug_status_watch = 0;
 651 }
 652
 653 static void hotplug_status_changed(struct xenbus_watch *watch,
 654                                    const char **vec,
 655                                    unsigned int vec_size)
 656 {
 657         struct backend_info *be = container_of(watch,
 658                                                struct backend_info,
 659                                                hotplug_status_watch);
 660         char *str;
 661         unsigned int len;
 662
 663         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 664         if (IS_ERR(str))
 665                 return;
 666         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 667                 /* Complete any pending state change */
 668                 xenbus_switch_state(be->dev, be->state);
 669
 670                 /* Not interested in this watch anymore. */
 671                 unregister_hotplug_status_watch(be);
 672         }
 673         kfree(str);
 674 }
 675
 676 static void connect(struct backend_info *be)
 677 {
 678         int err;
 679         struct xenbus_device *dev = be->dev;
 680         unsigned long credit_bytes, credit_usec;
 681         unsigned int queue_index;
 682         unsigned int requested_num_queues;
 683         struct xenvif_queue *queue;
 684
 685         /* Check whether the frontend requested multiple queues
 686          * and read the number requested.
 687          */
 688         err = xenbus_scanf(XBT_NIL, dev->otherend,
 689                            "multi-queue-num-queues",
 690                            "%u", &requested_num_queues);
 691         if (err < 0) {
 692                 requested_num_queues = 1; /* Fall back to single queue */
 693         } else if (requested_num_queues > xenvif_max_queues) {
 694                 /* buggy or malicious guest */
 695                 xenbus_dev_fatal(dev, err,
 696                                  "guest requested %u queues, exceeding the maximum of %u.",
 697                                  requested_num_queues, xenvif_max_queues);
 698                 return;
 699         }
 700
 701         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 702         if (err) {
 703                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 704                 return;
 705         }
 706
 707         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 708         read_xenbus_vif_flags(be);
 709
 710         /* Use the number of queues requested by the frontend */
 711         be->vif->queues = vzalloc(requested_num_queues *
 712                                   sizeof(struct xenvif_queue));
 713         be->vif->num_queues = requested_num_queues;
 714         be->vif->stalled_queues = requested_num_queues;
 715
 716         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 717                 queue = &be->vif->queues[queue_index];
 718                 queue->vif = be->vif;
 719                 queue->id = queue_index;
 720                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 721                                 be->vif->dev->name, queue->id);
 722
 723                 err = xenvif_init_queue(queue);
 724                 if (err) {
 725                         /* xenvif_init_queue() cleans up after itself on
 726                          * failure, but we need to clean up any previously
 727                          * initialised queues. Set num_queues to i so that
 728                          * earlier queues can be destroyed using the regular
 729                          * disconnect logic.
 730                          */
 731                         be->vif->num_queues = queue_index;
 732                         goto err;
 733                 }
 734
 735                 queue->remaining_credit = credit_bytes;
 736
 737                 err = connect_rings(be, queue);
 738                 if (err) {
 739                         /* connect_rings() cleans up after itself on failure,
 740                          * but we need to clean up after xenvif_init_queue() here,
 741                          * and also clean up any previously initialised queues.
 742                          */
 743                         xenvif_deinit_queue(queue);
 744                         be->vif->num_queues = queue_index;
 745                         goto err;
 746                 }
 747         }
 748
 749 #ifdef CONFIG_DEBUG_FS
 750         xenvif_debugfs_addif(be->vif);
 751 #endif /* CONFIG_DEBUG_FS */
 752
 753         /* Initialisation completed, tell core driver the number of
 754          * active queues.
 755          */
 756         rtnl_lock();
 757         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 758         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 759         rtnl_unlock();
 760
 761         xenvif_carrier_on(be->vif);
 762
 763         unregister_hotplug_status_watch(be);
 764         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 765                                    hotplug_status_changed,
 766                                    "%s/%s", dev->nodename, "hotplug-status");
 767         if (!err)
 768                 be->have_hotplug_status_watch = 1;
 769
 770         netif_tx_wake_all_queues(be->vif->dev);
 771
 772         return;
 773
 774 err:
 775         if (be->vif->num_queues > 0)
 776                 xenvif_disconnect(be->vif); /* Clean up existing queues */
 777         vfree(be->vif->queues);
 778         be->vif->queues = NULL;
 779         be->vif->num_queues = 0;
 780         return;
 781 }
 782
 783
 784 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue)
 785 {
 786         struct xenbus_device *dev = be->dev;
 787         unsigned int num_queues = queue->vif->num_queues;
 788         unsigned long tx_ring_ref, rx_ring_ref;
 789         unsigned int tx_evtchn, rx_evtchn;
 790         int err;
 791         char *xspath;
 792         size_t xspathsize;
 793         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 794
 795         /* If the frontend requested 1 queue, or we have fallen back
 796          * to single queue due to lack of frontend support for multi-
 797          * queue, expect the remaining XenStore keys in the toplevel
 798          * directory. Otherwise, expect them in a subdirectory called
 799          * queue-N.
 800          */
 801         if (num_queues == 1) {
 802                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 803                 if (!xspath) {
 804                         xenbus_dev_fatal(dev, -ENOMEM,
 805                                          "reading ring references");
 806                         return -ENOMEM;
 807                 }
 808                 strcpy(xspath, dev->otherend);
 809         } else {
 810                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 811                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 812                 if (!xspath) {
 813                         xenbus_dev_fatal(dev, -ENOMEM,
 814                                          "reading ring references");
 815                         return -ENOMEM;
 816                 }
 817                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 818                          queue->id);
 819         }
 820
 821         err = xenbus_gather(XBT_NIL, xspath,
 822                             "tx-ring-ref", "%lu", &tx_ring_ref,
 823                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 824         if (err) {
 825                 xenbus_dev_fatal(dev, err,
 826                                  "reading %s/ring-ref",
 827                                  xspath);
 828                 goto err;
 829         }
 830
 831         /* Try split event channels first, then single event channel. */
 832         err = xenbus_gather(XBT_NIL, xspath,
 833                             "event-channel-tx", "%u", &tx_evtchn,
 834                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 835         if (err < 0) {
 836                 err = xenbus_scanf(XBT_NIL, xspath,
 837                                    "event-channel", "%u", &tx_evtchn);
 838                 if (err < 0) {
 839                         xenbus_dev_fatal(dev, err,
 840                                          "reading %s/event-channel(-tx/rx)",
 841                                          xspath);
 842                         goto err;
 843                 }
 844                 rx_evtchn = tx_evtchn;
 845         }
 846
 847         /* Map the shared frame, irq etc. */
 848         err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref,
 849                              tx_evtchn, rx_evtchn);
 850         if (err) {
 851                 xenbus_dev_fatal(dev, err,
 852                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 853                                  tx_ring_ref, rx_ring_ref,
 854                                  tx_evtchn, rx_evtchn);
 855                 goto err;
 856         }
 857
 858         err = 0;
 859 err: /* Regular return falls through with err == 0 */
 860         kfree(xspath);
 861         return err;
 862 }
 863
 864 static int read_xenbus_vif_flags(struct backend_info *be)
 865 {
 866         struct xenvif *vif = be->vif;
 867         struct xenbus_device *dev = be->dev;
 868         unsigned int rx_copy;
 869         int err, val;
 870
 871         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 872                            &rx_copy);
 873         if (err == -ENOENT) {
 874                 err = 0;
 875                 rx_copy = 0;
 876         }
 877         if (err < 0) {
 878                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 879                                  dev->otherend);
 880                 return err;
 881         }
 882         if (!rx_copy)
 883                 return -EOPNOTSUPP;
 884
 885         if (xenbus_scanf(XBT_NIL, dev->otherend,
 886                          "feature-rx-notify", "%d", &val) < 0 || val == 0) {
 887                 xenbus_dev_fatal(dev, -EINVAL, "feature-rx-notify is mandatory");
 888                 return -EINVAL;
 889         }
 890
 891         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
 892                          "%d", &val) < 0)
 893                 val = 0;
 894         vif->can_sg = !!val;
 895
 896         vif->gso_mask = 0;
 897         vif->gso_prefix_mask = 0;
 898
 899         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
 900                          "%d", &val) < 0)
 901                 val = 0;
 902         if (val)
 903                 vif->gso_mask |= GSO_BIT(TCPV4);
 904
 905         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
 906                          "%d", &val) < 0)
 907                 val = 0;
 908         if (val)
 909                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
 910
 911         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
 912                          "%d", &val) < 0)
 913                 val = 0;
 914         if (val)
 915                 vif->gso_mask |= GSO_BIT(TCPV6);
 916
 917         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
 918                          "%d", &val) < 0)
 919                 val = 0;
 920         if (val)
 921                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
 922
 923         if (vif->gso_mask & vif->gso_prefix_mask) {
 924                 xenbus_dev_fatal(dev, err,
 925                                  "%s: gso and gso prefix flags are not "
 926                                  "mutually exclusive",
 927                                  dev->otherend);
 928                 return -EOPNOTSUPP;
 929         }
 930
 931         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
 932                          "%d", &val) < 0)
 933                 val = 0;
 934         vif->ip_csum = !val;
 935
 936         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
 937                          "%d", &val) < 0)
 938                 val = 0;
 939         vif->ipv6_csum = !!val;
 940
 941         return 0;
 942 }
 943
 944 static const struct xenbus_device_id netback_ids[] = {
 945         { "vif" },
 946         { "" }
 947 };
 948
 949 static struct xenbus_driver netback_driver = {
 950         .ids = netback_ids,
 951         .probe = netback_probe,
 952         .remove = netback_remove,
 953         .uevent = netback_uevent,
 954         .otherend_changed = frontend_changed,
 955 };
 956
 957 int xenvif_xenbus_init(void)
 958 {
 959         return xenbus_register_backend(&netback_driver);
 960 }
 961
 962 void xenvif_xenbus_fini(void)
 963 {
 964         return xenbus_unregister_driver(&netback_driver);
 965 }