drivers/virtio/virtio_pci.c

   1 /*
   2  * Virtio PCI driver
   3  *
   4  * This module allows virtio devices to be used over a virtual PCI device.
   5  * This can be used with QEMU based VMMs like KVM or Xen.
   6  *
   7  * Copyright IBM Corp. 2007
   8  *
   9  * Authors:
  10  *  Anthony Liguori  <aliguori@us.ibm.com>
  11  *
  12  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  13  * See the COPYING file in the top-level directory.
  14  *
  15  */
  16
  17 #include <linux/module.h>
  18 #include <linux/list.h>
  19 #include <linux/pci.h>
  20 #include <linux/slab.h>
  21 #include <linux/interrupt.h>
  22 #include <linux/virtio.h>
  23 #include <linux/virtio_config.h>
  24 #include <linux/virtio_ring.h>
  25 #include <linux/virtio_pci.h>
  26 #include <linux/highmem.h>
  27 #include <linux/spinlock.h>
  28
  29 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
  30 MODULE_DESCRIPTION("virtio-pci");
  31 MODULE_LICENSE("GPL");
  32 MODULE_VERSION("1");
  33
  34 /* Our device structure */
  35 struct virtio_pci_device
  36 {
  37         struct virtio_device vdev;
  38         struct pci_dev *pci_dev;
  39
  40         /* the IO mapping for the PCI config space */
  41         void __iomem *ioaddr;
  42
  43         /* a list of queues so we can dispatch IRQs */
  44         spinlock_t lock;
  45         struct list_head virtqueues;
  46
  47         /* MSI-X support */
  48         int msix_enabled;
  49         int intx_enabled;
  50         struct msix_entry *msix_entries;
  51         /* Name strings for interrupts. This size should be enough,
  52          * and I'm too lazy to allocate each name separately. */
  53         char (*msix_names)[256];
  54         /* Number of available vectors */
  55         unsigned msix_vectors;
  56         /* Vectors allocated, excluding per-vq vectors if any */
  57         unsigned msix_used_vectors;
  58         /* Whether we have vector per vq */
  59         bool per_vq_vectors;
  60 };
  61
  62 /* Constants for MSI-X */
  63 /* Use first vector for configuration changes, second and the rest for
  64  * virtqueues Thus, we need at least 2 vectors for MSI. */
  65 enum {
  66         VP_MSIX_CONFIG_VECTOR = 0,
  67         VP_MSIX_VQ_VECTOR = 1,
  68 };
  69
  70 struct virtio_pci_vq_info
  71 {
  72         /* the actual virtqueue */
  73         struct virtqueue *vq;
  74
  75         /* the number of entries in the queue */
  76         int num;
  77
  78         /* the index of the queue */
  79         int queue_index;
  80
  81         /* the virtual address of the ring queue */
  82         void *queue;
  83
  84         /* the list node for the virtqueues list */
  85         struct list_head node;
  86
  87         /* MSI-X vector (or none) */
  88         unsigned msix_vector;
  89 };
  90
  91 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
  92 static struct pci_device_id virtio_pci_id_table[] = {
  93         { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
  94         { 0 },
  95 };
  96
  97 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
  98
  99 /* Convert a generic virtio device to our structure */
 100 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
 101 {
 102         return container_of(vdev, struct virtio_pci_device, vdev);
 103 }
 104
 105 /* virtio config->get_features() implementation */
 106 static u32 vp_get_features(struct virtio_device *vdev)
 107 {
 108         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 109
 110         /* When someone needs more than 32 feature bits, we'll need to
 111          * steal a bit to indicate that the rest are somewhere else. */
 112         return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
 113 }
 114
 115 /* virtio config->finalize_features() implementation */
 116 static void vp_finalize_features(struct virtio_device *vdev)
 117 {
 118         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 119
 120         /* Give virtio_ring a chance to accept features. */
 121         vring_transport_features(vdev);
 122
 123         /* We only support 32 feature bits. */
 124         BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
 125         iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
 126 }
 127
 128 /* virtio config->get() implementation */
 129 static void vp_get(struct virtio_device *vdev, unsigned offset,
 130                    void *buf, unsigned len)
 131 {
 132         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 133         void __iomem *ioaddr = vp_dev->ioaddr +
 134                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 135         u8 *ptr = buf;
 136         int i;
 137
 138         for (i = 0; i < len; i++)
 139                 ptr[i] = ioread8(ioaddr + i);
 140 }
 141
 142 /* the config->set() implementation.  it's symmetric to the config->get()
 143  * implementation */
 144 static void vp_set(struct virtio_device *vdev, unsigned offset,
 145                    const void *buf, unsigned len)
 146 {
 147         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 148         void __iomem *ioaddr = vp_dev->ioaddr +
 149                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 150         const u8 *ptr = buf;
 151         int i;
 152
 153         for (i = 0; i < len; i++)
 154                 iowrite8(ptr[i], ioaddr + i);
 155 }
 156
 157 /* config->{get,set}_status() implementations */
 158 static u8 vp_get_status(struct virtio_device *vdev)
 159 {
 160         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 161         return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 162 }
 163
 164 static void vp_set_status(struct virtio_device *vdev, u8 status)
 165 {
 166         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 167         /* We should never be setting status to 0. */
 168         BUG_ON(status == 0);
 169         iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 170 }
 171
 172 static void vp_reset(struct virtio_device *vdev)
 173 {
 174         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 175         /* 0 status means a reset. */
 176         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 177 }
 178
 179 /* the notify function used when creating a virt queue */
 180 static void vp_notify(struct virtqueue *vq)
 181 {
 182         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 183         struct virtio_pci_vq_info *info = vq->priv;
 184
 185         /* we write the queue's selector into the notification register to
 186          * signal the other end */
 187         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 188 }
 189
 190 /* Handle a configuration change: Tell driver if it wants to know. */
 191 static irqreturn_t vp_config_changed(int irq, void *opaque)
 192 {
 193         struct virtio_pci_device *vp_dev = opaque;
 194         struct virtio_driver *drv;
 195         drv = container_of(vp_dev->vdev.dev.driver,
 196                            struct virtio_driver, driver);
 197
 198         if (drv && drv->config_changed)
 199                 drv->config_changed(&vp_dev->vdev);
 200         return IRQ_HANDLED;
 201 }
 202
 203 /* Notify all virtqueues on an interrupt. */
 204 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 205 {
 206         struct virtio_pci_device *vp_dev = opaque;
 207         struct virtio_pci_vq_info *info;
 208         irqreturn_t ret = IRQ_NONE;
 209         unsigned long flags;
 210
 211         spin_lock_irqsave(&vp_dev->lock, flags);
 212         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 213                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 214                         ret = IRQ_HANDLED;
 215         }
 216         spin_unlock_irqrestore(&vp_dev->lock, flags);
 217
 218         return ret;
 219 }
 220
 221 /* A small wrapper to also acknowledge the interrupt when it's handled.
 222  * I really need an EIO hook for the vring so I can ack the interrupt once we
 223  * know that we'll be handling the IRQ but before we invoke the callback since
 224  * the callback may notify the host which results in the host attempting to
 225  * raise an interrupt that we would then mask once we acknowledged the
 226  * interrupt. */
 227 static irqreturn_t vp_interrupt(int irq, void *opaque)
 228 {
 229         struct virtio_pci_device *vp_dev = opaque;
 230         u8 isr;
 231
 232         /* reading the ISR has the effect of also clearing it so it's very
 233          * important to save off the value. */
 234         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 235
 236         /* It's definitely not us if the ISR was not high */
 237         if (!isr)
 238                 return IRQ_NONE;
 239
 240         /* Configuration change?  Tell driver if it wants to know. */
 241         if (isr & VIRTIO_PCI_ISR_CONFIG)
 242                 vp_config_changed(irq, opaque);
 243
 244         return vp_vring_interrupt(irq, opaque);
 245 }
 246
 247 static void vp_free_vectors(struct virtio_device *vdev)
 248 {
 249         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 250         int i;
 251
 252         if (vp_dev->intx_enabled) {
 253                 free_irq(vp_dev->pci_dev->irq, vp_dev);
 254                 vp_dev->intx_enabled = 0;
 255         }
 256
 257         for (i = 0; i < vp_dev->msix_used_vectors; ++i)
 258                 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
 259
 260         if (vp_dev->msix_enabled) {
 261                 /* Disable the vector used for configuration */
 262                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 263                           vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 264                 /* Flush the write out to device */
 265                 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 266
 267                 pci_disable_msix(vp_dev->pci_dev);
 268                 vp_dev->msix_enabled = 0;
 269                 vp_dev->msix_vectors = 0;
 270         }
 271
 272         vp_dev->msix_used_vectors = 0;
 273         kfree(vp_dev->msix_names);
 274         vp_dev->msix_names = NULL;
 275         kfree(vp_dev->msix_entries);
 276         vp_dev->msix_entries = NULL;
 277 }
 278
 279 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
 280                                    bool per_vq_vectors)
 281 {
 282         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 283         const char *name = dev_name(&vp_dev->vdev.dev);
 284         unsigned i, v;
 285         int err = -ENOMEM;
 286
 287         vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 288                                        GFP_KERNEL);
 289         if (!vp_dev->msix_entries)
 290                 goto error;
 291         vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 292                                      GFP_KERNEL);
 293         if (!vp_dev->msix_names)
 294                 goto error;
 295
 296         for (i = 0; i < nvectors; ++i)
 297                 vp_dev->msix_entries[i].entry = i;
 298
 299         /* pci_enable_msix returns positive if we can't get this many. */
 300         err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
 301         if (err > 0)
 302                 err = -ENOSPC;
 303         if (err)
 304                 goto error;
 305         vp_dev->msix_vectors = nvectors;
 306         vp_dev->msix_enabled = 1;
 307
 308         /* Set the vector used for configuration */
 309         v = vp_dev->msix_used_vectors;
 310         snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 311                  "%s-config", name);
 312         err = request_irq(vp_dev->msix_entries[v].vector,
 313                           vp_config_changed, 0, vp_dev->msix_names[v],
 314                           vp_dev);
 315         if (err)
 316                 goto error;
 317         ++vp_dev->msix_used_vectors;
 318
 319         iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 320         /* Verify we had enough resources to assign the vector */
 321         v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 322         if (v == VIRTIO_MSI_NO_VECTOR) {
 323                 err = -EBUSY;
 324                 goto error;
 325         }
 326
 327         if (!per_vq_vectors) {
 328                 /* Shared vector for all VQs */
 329                 v = vp_dev->msix_used_vectors;
 330                 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 331                          "%s-virtqueues", name);
 332                 err = request_irq(vp_dev->msix_entries[v].vector,
 333                                   vp_vring_interrupt, 0, vp_dev->msix_names[v],
 334                                   vp_dev);
 335                 if (err)
 336                         goto error;
 337                 ++vp_dev->msix_used_vectors;
 338         }
 339         return 0;
 340 error:
 341         vp_free_vectors(vdev);
 342         return err;
 343 }
 344
 345 static int vp_request_intx(struct virtio_device *vdev)
 346 {
 347         int err;
 348         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 349
 350         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 351                           IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 352         if (!err)
 353                 vp_dev->intx_enabled = 1;
 354         return err;
 355 }
 356
 357 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 358                                   void (*callback)(struct virtqueue *vq),
 359                                   const char *name,
 360                                   u16 msix_vec)
 361 {
 362         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 363         struct virtio_pci_vq_info *info;
 364         struct virtqueue *vq;
 365         unsigned long flags, size;
 366         u16 num;
 367         int err;
 368
 369         /* Select the queue we're interested in */
 370         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 371
 372         /* Check if queue is either not available or already active. */
 373         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 374         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 375                 return ERR_PTR(-ENOENT);
 376
 377         /* allocate and fill out our structure the represents an active
 378          * queue */
 379         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 380         if (!info)
 381                 return ERR_PTR(-ENOMEM);
 382
 383         info->queue_index = index;
 384         info->num = num;
 385         info->msix_vector = msix_vec;
 386
 387         size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 388         info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 389         if (info->queue == NULL) {
 390                 err = -ENOMEM;
 391                 goto out_info;
 392         }
 393
 394         /* activate the queue */
 395         iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 396                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 397
 398         /* create the vring */
 399         vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
 400                                  vdev, info->queue, vp_notify, callback, name);
 401         if (!vq) {
 402                 err = -ENOMEM;
 403                 goto out_activate_queue;
 404         }
 405
 406         vq->priv = info;
 407         info->vq = vq;
 408
 409         if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 410                 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 411                 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 412                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 413                         err = -EBUSY;
 414                         goto out_assign;
 415                 }
 416         }
 417
 418         spin_lock_irqsave(&vp_dev->lock, flags);
 419         list_add(&info->node, &vp_dev->virtqueues);
 420         spin_unlock_irqrestore(&vp_dev->lock, flags);
 421
 422         return vq;
 423
 424 out_assign:
 425         vring_del_virtqueue(vq);
 426 out_activate_queue:
 427         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 428         free_pages_exact(info->queue, size);
 429 out_info:
 430         kfree(info);
 431         return ERR_PTR(err);
 432 }
 433
 434 static void vp_del_vq(struct virtqueue *vq)
 435 {
 436         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 437         struct virtio_pci_vq_info *info = vq->priv;
 438         unsigned long flags, size;
 439
 440         spin_lock_irqsave(&vp_dev->lock, flags);
 441         list_del(&info->node);
 442         spin_unlock_irqrestore(&vp_dev->lock, flags);
 443
 444         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 445
 446         if (vp_dev->msix_enabled) {
 447                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 448                           vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 449                 /* Flush the write out to device */
 450                 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 451         }
 452
 453         vring_del_virtqueue(vq);
 454
 455         /* Select and deactivate the queue */
 456         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 457
 458         size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 459         free_pages_exact(info->queue, size);
 460         kfree(info);
 461 }
 462
 463 /* the config->del_vqs() implementation */
 464 static void vp_del_vqs(struct virtio_device *vdev)
 465 {
 466         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 467         struct virtqueue *vq, *n;
 468         struct virtio_pci_vq_info *info;
 469
 470         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 471                 info = vq->priv;
 472                 if (vp_dev->per_vq_vectors &&
 473                         info->msix_vector != VIRTIO_MSI_NO_VECTOR)
 474                         free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 475                                  vq);
 476                 vp_del_vq(vq);
 477         }
 478         vp_dev->per_vq_vectors = false;
 479
 480         vp_free_vectors(vdev);
 481 }
 482
 483 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 484                               struct virtqueue *vqs[],
 485                               vq_callback_t *callbacks[],
 486                               const char *names[],
 487                               bool use_msix,
 488                               bool per_vq_vectors)
 489 {
 490         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 491         u16 msix_vec;
 492         int i, err, nvectors, allocated_vectors;
 493
 494         if (!use_msix) {
 495                 /* Old style: one normal interrupt for change and all vqs. */
 496                 err = vp_request_intx(vdev);
 497                 if (err)
 498                         goto error_request;
 499         } else {
 500                 if (per_vq_vectors) {
 501                         /* Best option: one for change interrupt, one per vq. */
 502                         nvectors = 1;
 503                         for (i = 0; i < nvqs; ++i)
 504                                 if (callbacks[i])
 505                                         ++nvectors;
 506                 } else {
 507                         /* Second best: one for change, shared for all vqs. */
 508                         nvectors = 2;
 509                 }
 510
 511                 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 512                 if (err)
 513                         goto error_request;
 514         }
 515
 516         vp_dev->per_vq_vectors = per_vq_vectors;
 517         allocated_vectors = vp_dev->msix_used_vectors;
 518         for (i = 0; i < nvqs; ++i) {
 519                 if (!callbacks[i] || !vp_dev->msix_enabled)
 520                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 521                 else if (vp_dev->per_vq_vectors)
 522                         msix_vec = allocated_vectors++;
 523                 else
 524                         msix_vec = VP_MSIX_VQ_VECTOR;
 525                 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 526                 if (IS_ERR(vqs[i])) {
 527                         err = PTR_ERR(vqs[i]);
 528                         goto error_find;
 529                 }
 530
 531                 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 532                         continue;
 533
 534                 /* allocate per-vq irq if available and necessary */
 535                 snprintf(vp_dev->msix_names[msix_vec],
 536                          sizeof *vp_dev->msix_names,
 537                          "%s-%s",
 538                          dev_name(&vp_dev->vdev.dev), names[i]);
 539                 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 540                                   vring_interrupt, 0,
 541                                   vp_dev->msix_names[msix_vec],
 542                                   vqs[i]);
 543                 if (err) {
 544                         vp_del_vq(vqs[i]);
 545                         goto error_find;
 546                 }
 547         }
 548         return 0;
 549
 550 error_find:
 551         vp_del_vqs(vdev);
 552
 553 error_request:
 554         return err;
 555 }
 556
 557 /* the config->find_vqs() implementation */
 558 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 559                        struct virtqueue *vqs[],
 560                        vq_callback_t *callbacks[],
 561                        const char *names[])
 562 {
 563         int err;
 564
 565         /* Try MSI-X with one vector per queue. */
 566         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 567         if (!err)
 568                 return 0;
 569         /* Fallback: MSI-X with one vector for config, one shared for queues. */
 570         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 571                                  true, false);
 572         if (!err)
 573                 return 0;
 574         /* Finally fall back to regular interrupts. */
 575         return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 576                                   false, false);
 577 }
 578
 579 static struct virtio_config_ops virtio_pci_config_ops = {
 580         .get            = vp_get,
 581         .set            = vp_set,
 582         .get_status     = vp_get_status,
 583         .set_status     = vp_set_status,
 584         .reset          = vp_reset,
 585         .find_vqs       = vp_find_vqs,
 586         .del_vqs        = vp_del_vqs,
 587         .get_features   = vp_get_features,
 588         .finalize_features = vp_finalize_features,
 589 };
 590
 591 static void virtio_pci_release_dev(struct device *_d)
 592 {
 593         struct virtio_device *dev = container_of(_d, struct virtio_device, dev);
 594         struct virtio_pci_device *vp_dev = to_vp_device(dev);
 595         struct pci_dev *pci_dev = vp_dev->pci_dev;
 596
 597         vp_del_vqs(dev);
 598         pci_set_drvdata(pci_dev, NULL);
 599         pci_iounmap(pci_dev, vp_dev->ioaddr);
 600         pci_release_regions(pci_dev);
 601         pci_disable_device(pci_dev);
 602         kfree(vp_dev);
 603 }
 604
 605 /* the PCI probing function */
 606 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
 607                                       const struct pci_device_id *id)
 608 {
 609         struct virtio_pci_device *vp_dev;
 610         int err;
 611
 612         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 613         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 614                 return -ENODEV;
 615
 616         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 617                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 618                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 619                 return -ENODEV;
 620         }
 621
 622         /* allocate our structure and fill it out */
 623         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 624         if (vp_dev == NULL)
 625                 return -ENOMEM;
 626
 627         vp_dev->vdev.dev.parent = &pci_dev->dev;
 628         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 629         vp_dev->vdev.config = &virtio_pci_config_ops;
 630         vp_dev->pci_dev = pci_dev;
 631         INIT_LIST_HEAD(&vp_dev->virtqueues);
 632         spin_lock_init(&vp_dev->lock);
 633
 634         /* Disable MSI/MSIX to bring device to a known good state. */
 635         pci_msi_off(pci_dev);
 636
 637         /* enable the device */
 638         err = pci_enable_device(pci_dev);
 639         if (err)
 640                 goto out;
 641
 642         err = pci_request_regions(pci_dev, "virtio-pci");
 643         if (err)
 644                 goto out_enable_device;
 645
 646         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 647         if (vp_dev->ioaddr == NULL)
 648                 goto out_req_regions;
 649
 650         pci_set_drvdata(pci_dev, vp_dev);
 651         pci_set_master(pci_dev);
 652
 653         /* we use the subsystem vendor/device id as the virtio vendor/device
 654          * id.  this allows us to use the same PCI vendor/device id for all
 655          * virtio devices and to identify the particular virtio driver by
 656          * the subsystem ids */
 657         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 658         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 659
 660         /* finally register the virtio device */
 661         err = register_virtio_device(&vp_dev->vdev);
 662         if (err)
 663                 goto out_set_drvdata;
 664
 665         return 0;
 666
 667 out_set_drvdata:
 668         pci_set_drvdata(pci_dev, NULL);
 669         pci_iounmap(pci_dev, vp_dev->ioaddr);
 670 out_req_regions:
 671         pci_release_regions(pci_dev);
 672 out_enable_device:
 673         pci_disable_device(pci_dev);
 674 out:
 675         kfree(vp_dev);
 676         return err;
 677 }
 678
 679 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
 680 {
 681         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 682
 683         unregister_virtio_device(&vp_dev->vdev);
 684 }
 685
 686 #ifdef CONFIG_PM
 687 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
 688 {
 689         pci_save_state(pci_dev);
 690         pci_set_power_state(pci_dev, PCI_D3hot);
 691         return 0;
 692 }
 693
 694 static int virtio_pci_resume(struct pci_dev *pci_dev)
 695 {
 696         pci_restore_state(pci_dev);
 697         pci_set_power_state(pci_dev, PCI_D0);
 698         return 0;
 699 }
 700 #endif
 701
 702 static struct pci_driver virtio_pci_driver = {
 703         .name           = "virtio-pci",
 704         .id_table       = virtio_pci_id_table,
 705         .probe          = virtio_pci_probe,
 706         .remove         = __devexit_p(virtio_pci_remove),
 707 #ifdef CONFIG_PM
 708         .suspend        = virtio_pci_suspend,
 709         .resume         = virtio_pci_resume,
 710 #endif
 711 };
 712
 713 static int __init virtio_pci_init(void)
 714 {
 715         return pci_register_driver(&virtio_pci_driver);
 716 }
 717
 718 module_init(virtio_pci_init);
 719
 720 static void __exit virtio_pci_exit(void)
 721 {
 722         pci_unregister_driver(&virtio_pci_driver);
 723 }
 724
 725 module_exit(virtio_pci_exit);