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/interrupt.h>
  21 #include <linux/virtio.h>
  22 #include <linux/virtio_config.h>
  23 #include <linux/virtio_ring.h>
  24 #include <linux/virtio_pci.h>
  25 #include <linux/highmem.h>
  26 #include <linux/spinlock.h>
  27
  28 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
  29 MODULE_DESCRIPTION("virtio-pci");
  30 MODULE_LICENSE("GPL");
  31 MODULE_VERSION("1");
  32
  33 /* Our device structure */
  34 struct virtio_pci_device
  35 {
  36         struct virtio_device vdev;
  37         struct pci_dev *pci_dev;
  38
  39         /* the IO mapping for the PCI config space */
  40         void __iomem *ioaddr;
  41
  42         /* a list of queues so we can dispatch IRQs */
  43         spinlock_t lock;
  44         struct list_head virtqueues;
  45
  46         /* MSI-X support */
  47         int msix_enabled;
  48         int intx_enabled;
  49         struct msix_entry *msix_entries;
  50         /* Name strings for interrupts. This size should be enough,
  51          * and I'm too lazy to allocate each name separately. */
  52         char (*msix_names)[256];
  53         /* Number of available vectors */
  54         unsigned msix_vectors;
  55         /* Vectors allocated, excluding per-vq vectors if any */
  56         unsigned msix_used_vectors;
  57         /* Whether we have vector per vq */
  58         bool per_vq_vectors;
  59 };
  60
  61 /* Constants for MSI-X */
  62 /* Use first vector for configuration changes, second and the rest for
  63  * virtqueues Thus, we need at least 2 vectors for MSI. */
  64 enum {
  65         VP_MSIX_CONFIG_VECTOR = 0,
  66         VP_MSIX_VQ_VECTOR = 1,
  67 };
  68
  69 struct virtio_pci_vq_info
  70 {
  71         /* the actual virtqueue */
  72         struct virtqueue *vq;
  73
  74         /* the number of entries in the queue */
  75         int num;
  76
  77         /* the index of the queue */
  78         int queue_index;
  79
  80         /* the virtual address of the ring queue */
  81         void *queue;
  82
  83         /* the list node for the virtqueues list */
  84         struct list_head node;
  85
  86         /* MSI-X vector (or none) */
  87         unsigned msix_vector;
  88 };
  89
  90 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
  91 static struct pci_device_id virtio_pci_id_table[] = {
  92         { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
  93         { 0 },
  94 };
  95
  96 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
  97
  98 /* A PCI device has it's own struct device and so does a virtio device so
  99  * we create a place for the virtio devices to show up in sysfs.  I think it
 100  * would make more sense for virtio to not insist on having it's own device. */
 101 static struct device *virtio_pci_root;
 102
 103 /* Convert a generic virtio device to our structure */
 104 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
 105 {
 106         return container_of(vdev, struct virtio_pci_device, vdev);
 107 }
 108
 109 /* virtio config->get_features() implementation */
 110 static u32 vp_get_features(struct virtio_device *vdev)
 111 {
 112         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 113
 114         /* When someone needs more than 32 feature bits, we'll need to
 115          * steal a bit to indicate that the rest are somewhere else. */
 116         return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
 117 }
 118
 119 /* virtio config->finalize_features() implementation */
 120 static void vp_finalize_features(struct virtio_device *vdev)
 121 {
 122         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 123
 124         /* Give virtio_ring a chance to accept features. */
 125         vring_transport_features(vdev);
 126
 127         /* We only support 32 feature bits. */
 128         BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
 129         iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
 130 }
 131
 132 /* virtio config->get() implementation */
 133 static void vp_get(struct virtio_device *vdev, unsigned offset,
 134                    void *buf, unsigned len)
 135 {
 136         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 137         void __iomem *ioaddr = vp_dev->ioaddr +
 138                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 139         u8 *ptr = buf;
 140         int i;
 141
 142         for (i = 0; i < len; i++)
 143                 ptr[i] = ioread8(ioaddr + i);
 144 }
 145
 146 /* the config->set() implementation.  it's symmetric to the config->get()
 147  * implementation */
 148 static void vp_set(struct virtio_device *vdev, unsigned offset,
 149                    const void *buf, unsigned len)
 150 {
 151         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 152         void __iomem *ioaddr = vp_dev->ioaddr +
 153                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 154         const u8 *ptr = buf;
 155         int i;
 156
 157         for (i = 0; i < len; i++)
 158                 iowrite8(ptr[i], ioaddr + i);
 159 }
 160
 161 /* config->{get,set}_status() implementations */
 162 static u8 vp_get_status(struct virtio_device *vdev)
 163 {
 164         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 165         return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 166 }
 167
 168 static void vp_set_status(struct virtio_device *vdev, u8 status)
 169 {
 170         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 171         /* We should never be setting status to 0. */
 172         BUG_ON(status == 0);
 173         iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 174 }
 175
 176 static void vp_reset(struct virtio_device *vdev)
 177 {
 178         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 179         /* 0 status means a reset. */
 180         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 181 }
 182
 183 /* the notify function used when creating a virt queue */
 184 static void vp_notify(struct virtqueue *vq)
 185 {
 186         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 187         struct virtio_pci_vq_info *info = vq->priv;
 188
 189         /* we write the queue's selector into the notification register to
 190          * signal the other end */
 191         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 192 }
 193
 194 /* Handle a configuration change: Tell driver if it wants to know. */
 195 static irqreturn_t vp_config_changed(int irq, void *opaque)
 196 {
 197         struct virtio_pci_device *vp_dev = opaque;
 198         struct virtio_driver *drv;
 199         drv = container_of(vp_dev->vdev.dev.driver,
 200                            struct virtio_driver, driver);
 201
 202         if (drv && drv->config_changed)
 203                 drv->config_changed(&vp_dev->vdev);
 204         return IRQ_HANDLED;
 205 }
 206
 207 /* Notify all virtqueues on an interrupt. */
 208 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 209 {
 210         struct virtio_pci_device *vp_dev = opaque;
 211         struct virtio_pci_vq_info *info;
 212         irqreturn_t ret = IRQ_NONE;
 213         unsigned long flags;
 214
 215         spin_lock_irqsave(&vp_dev->lock, flags);
 216         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 217                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 218                         ret = IRQ_HANDLED;
 219         }
 220         spin_unlock_irqrestore(&vp_dev->lock, flags);
 221
 222         return ret;
 223 }
 224
 225 /* A small wrapper to also acknowledge the interrupt when it's handled.
 226  * I really need an EIO hook for the vring so I can ack the interrupt once we
 227  * know that we'll be handling the IRQ but before we invoke the callback since
 228  * the callback may notify the host which results in the host attempting to
 229  * raise an interrupt that we would then mask once we acknowledged the
 230  * interrupt. */
 231 static irqreturn_t vp_interrupt(int irq, void *opaque)
 232 {
 233         struct virtio_pci_device *vp_dev = opaque;
 234         u8 isr;
 235
 236         /* reading the ISR has the effect of also clearing it so it's very
 237          * important to save off the value. */
 238         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 239
 240         /* It's definitely not us if the ISR was not high */
 241         if (!isr)
 242                 return IRQ_NONE;
 243
 244         /* Configuration change?  Tell driver if it wants to know. */
 245         if (isr & VIRTIO_PCI_ISR_CONFIG)
 246                 vp_config_changed(irq, opaque);
 247
 248         return vp_vring_interrupt(irq, opaque);
 249 }
 250
 251 static void vp_free_vectors(struct virtio_device *vdev)
 252 {
 253         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 254         int i;
 255
 256         if (vp_dev->intx_enabled) {
 257                 free_irq(vp_dev->pci_dev->irq, vp_dev);
 258                 vp_dev->intx_enabled = 0;
 259         }
 260
 261         for (i = 0; i < vp_dev->msix_used_vectors; ++i)
 262                 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
 263
 264         if (vp_dev->msix_enabled) {
 265                 /* Disable the vector used for configuration */
 266                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 267                           vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 268                 /* Flush the write out to device */
 269                 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 270
 271                 pci_disable_msix(vp_dev->pci_dev);
 272                 vp_dev->msix_enabled = 0;
 273                 vp_dev->msix_vectors = 0;
 274         }
 275
 276         vp_dev->msix_used_vectors = 0;
 277         kfree(vp_dev->msix_names);
 278         vp_dev->msix_names = NULL;
 279         kfree(vp_dev->msix_entries);
 280         vp_dev->msix_entries = NULL;
 281 }
 282
 283 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
 284                                    bool per_vq_vectors)
 285 {
 286         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 287         const char *name = dev_name(&vp_dev->vdev.dev);
 288         unsigned i, v;
 289         int err = -ENOMEM;
 290
 291         vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 292                                        GFP_KERNEL);
 293         if (!vp_dev->msix_entries)
 294                 goto error;
 295         vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 296                                      GFP_KERNEL);
 297         if (!vp_dev->msix_names)
 298                 goto error;
 299
 300         for (i = 0; i < nvectors; ++i)
 301                 vp_dev->msix_entries[i].entry = i;
 302
 303         /* pci_enable_msix returns positive if we can't get this many. */
 304         err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
 305         if (err > 0)
 306                 err = -ENOSPC;
 307         if (err)
 308                 goto error;
 309         vp_dev->msix_vectors = nvectors;
 310         vp_dev->msix_enabled = 1;
 311
 312         /* Set the vector used for configuration */
 313         v = vp_dev->msix_used_vectors;
 314         snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 315                  "%s-config", name);
 316         err = request_irq(vp_dev->msix_entries[v].vector,
 317                           vp_config_changed, 0, vp_dev->msix_names[v],
 318                           vp_dev);
 319         if (err)
 320                 goto error;
 321         ++vp_dev->msix_used_vectors;
 322
 323         iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 324         /* Verify we had enough resources to assign the vector */
 325         v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 326         if (v == VIRTIO_MSI_NO_VECTOR) {
 327                 err = -EBUSY;
 328                 goto error;
 329         }
 330
 331         if (!per_vq_vectors) {
 332                 /* Shared vector for all VQs */
 333                 v = vp_dev->msix_used_vectors;
 334                 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 335                          "%s-virtqueues", name);
 336                 err = request_irq(vp_dev->msix_entries[v].vector,
 337                                   vp_vring_interrupt, 0, vp_dev->msix_names[v],
 338                                   vp_dev);
 339                 if (err)
 340                         goto error;
 341                 ++vp_dev->msix_used_vectors;
 342         }
 343         return 0;
 344 error:
 345         vp_free_vectors(vdev);
 346         return err;
 347 }
 348
 349 static int vp_request_intx(struct virtio_device *vdev)
 350 {
 351         int err;
 352         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 353
 354         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 355                           IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 356         if (!err)
 357                 vp_dev->intx_enabled = 1;
 358         return err;
 359 }
 360
 361 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 362                                   void (*callback)(struct virtqueue *vq),
 363                                   const char *name,
 364                                   u16 msix_vec)
 365 {
 366         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 367         struct virtio_pci_vq_info *info;
 368         struct virtqueue *vq;
 369         unsigned long flags, size;
 370         u16 num;
 371         int err;
 372
 373         /* Select the queue we're interested in */
 374         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 375
 376         /* Check if queue is either not available or already active. */
 377         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 378         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 379                 return ERR_PTR(-ENOENT);
 380
 381         /* allocate and fill out our structure the represents an active
 382          * queue */
 383         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 384         if (!info)
 385                 return ERR_PTR(-ENOMEM);
 386
 387         info->queue_index = index;
 388         info->num = num;
 389         info->msix_vector = msix_vec;
 390
 391         size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 392         info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 393         if (info->queue == NULL) {
 394                 err = -ENOMEM;
 395                 goto out_info;
 396         }
 397
 398         /* activate the queue */
 399         iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 400                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 401
 402         /* create the vring */
 403         vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
 404                                  vdev, info->queue, vp_notify, callback, name);
 405         if (!vq) {
 406                 err = -ENOMEM;
 407                 goto out_activate_queue;
 408         }
 409
 410         vq->priv = info;
 411         info->vq = vq;
 412
 413         if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 414                 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 415                 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 416                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 417                         err = -EBUSY;
 418                         goto out_assign;
 419                 }
 420         }
 421
 422         spin_lock_irqsave(&vp_dev->lock, flags);
 423         list_add(&info->node, &vp_dev->virtqueues);
 424         spin_unlock_irqrestore(&vp_dev->lock, flags);
 425
 426         return vq;
 427
 428 out_assign:
 429         vring_del_virtqueue(vq);
 430 out_activate_queue:
 431         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 432         free_pages_exact(info->queue, size);
 433 out_info:
 434         kfree(info);
 435         return ERR_PTR(err);
 436 }
 437
 438 static void vp_del_vq(struct virtqueue *vq)
 439 {
 440         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 441         struct virtio_pci_vq_info *info = vq->priv;
 442         unsigned long flags, size;
 443
 444         spin_lock_irqsave(&vp_dev->lock, flags);
 445         list_del(&info->node);
 446         spin_unlock_irqrestore(&vp_dev->lock, flags);
 447
 448         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 449
 450         if (vp_dev->msix_enabled) {
 451                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 452                           vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 453                 /* Flush the write out to device */
 454                 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 455         }
 456
 457         vring_del_virtqueue(vq);
 458
 459         /* Select and deactivate the queue */
 460         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 461
 462         size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 463         free_pages_exact(info->queue, size);
 464         kfree(info);
 465 }
 466
 467 /* the config->del_vqs() implementation */
 468 static void vp_del_vqs(struct virtio_device *vdev)
 469 {
 470         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 471         struct virtqueue *vq, *n;
 472         struct virtio_pci_vq_info *info;
 473
 474         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 475                 info = vq->priv;
 476                 if (vp_dev->per_vq_vectors)
 477                         free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 478                                  vq);
 479                 vp_del_vq(vq);
 480         }
 481         vp_dev->per_vq_vectors = false;
 482
 483         vp_free_vectors(vdev);
 484 }
 485
 486 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 487                               struct virtqueue *vqs[],
 488                               vq_callback_t *callbacks[],
 489                               const char *names[],
 490                               bool use_msix,
 491                               bool per_vq_vectors)
 492 {
 493         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 494         u16 msix_vec;
 495         int i, err, nvectors, allocated_vectors;
 496
 497         if (!use_msix) {
 498                 /* Old style: one normal interrupt for change and all vqs. */
 499                 err = vp_request_intx(vdev);
 500                 if (err)
 501                         goto error_request;
 502         } else {
 503                 if (per_vq_vectors) {
 504                         /* Best option: one for change interrupt, one per vq. */
 505                         nvectors = 1;
 506                         for (i = 0; i < nvqs; ++i)
 507                                 if (callbacks[i])
 508                                         ++nvectors;
 509                 } else {
 510                         /* Second best: one for change, shared for all vqs. */
 511                         nvectors = 2;
 512                 }
 513
 514                 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 515                 if (err)
 516                         goto error_request;
 517         }
 518
 519         vp_dev->per_vq_vectors = per_vq_vectors;
 520         allocated_vectors = vp_dev->msix_used_vectors;
 521         for (i = 0; i < nvqs; ++i) {
 522                 if (!callbacks[i] || !vp_dev->msix_enabled)
 523                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 524                 else if (vp_dev->per_vq_vectors)
 525                         msix_vec = allocated_vectors++;
 526                 else
 527                         msix_vec = VP_MSIX_VQ_VECTOR;
 528                 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 529                 if (IS_ERR(vqs[i])) {
 530                         err = PTR_ERR(vqs[i]);
 531                         goto error_find;
 532                 }
 533
 534                 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 535                         continue;
 536
 537                 /* allocate per-vq irq if available and necessary */
 538                 snprintf(vp_dev->msix_names[msix_vec],
 539                          sizeof *vp_dev->msix_names,
 540                          "%s-%s",
 541                          dev_name(&vp_dev->vdev.dev), names[i]);
 542                 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 543                                   vring_interrupt, 0,
 544                                   vp_dev->msix_names[msix_vec],
 545                                   vqs[i]);
 546                 if (err) {
 547                         vp_del_vq(vqs[i]);
 548                         goto error_find;
 549                 }
 550         }
 551         return 0;
 552
 553 error_find:
 554         vp_del_vqs(vdev);
 555
 556 error_request:
 557         return err;
 558 }
 559
 560 /* the config->find_vqs() implementation */
 561 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 562                        struct virtqueue *vqs[],
 563                        vq_callback_t *callbacks[],
 564                        const char *names[])
 565 {
 566         int err;
 567
 568         /* Try MSI-X with one vector per queue. */
 569         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 570         if (!err)
 571                 return 0;
 572         /* Fallback: MSI-X with one vector for config, one shared for queues. */
 573         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 574                                  true, false);
 575         if (!err)
 576                 return 0;
 577         /* Finally fall back to regular interrupts. */
 578         return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 579                                   false, false);
 580 }
 581
 582 static struct virtio_config_ops virtio_pci_config_ops = {
 583         .get            = vp_get,
 584         .set            = vp_set,
 585         .get_status     = vp_get_status,
 586         .set_status     = vp_set_status,
 587         .reset          = vp_reset,
 588         .find_vqs       = vp_find_vqs,
 589         .del_vqs        = vp_del_vqs,
 590         .get_features   = vp_get_features,
 591         .finalize_features = vp_finalize_features,
 592 };
 593
 594 static void virtio_pci_release_dev(struct device *_d)
 595 {
 596         struct virtio_device *dev = container_of(_d, struct virtio_device, dev);
 597         struct virtio_pci_device *vp_dev = to_vp_device(dev);
 598         struct pci_dev *pci_dev = vp_dev->pci_dev;
 599
 600         vp_del_vqs(dev);
 601         pci_set_drvdata(pci_dev, NULL);
 602         pci_iounmap(pci_dev, vp_dev->ioaddr);
 603         pci_release_regions(pci_dev);
 604         pci_disable_device(pci_dev);
 605         kfree(vp_dev);
 606 }
 607
 608 /* the PCI probing function */
 609 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
 610                                       const struct pci_device_id *id)
 611 {
 612         struct virtio_pci_device *vp_dev;
 613         int err;
 614
 615         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 616         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 617                 return -ENODEV;
 618
 619         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 620                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 621                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 622                 return -ENODEV;
 623         }
 624
 625         /* allocate our structure and fill it out */
 626         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 627         if (vp_dev == NULL)
 628                 return -ENOMEM;
 629
 630         vp_dev->vdev.dev.parent = virtio_pci_root;
 631         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 632         vp_dev->vdev.config = &virtio_pci_config_ops;
 633         vp_dev->pci_dev = pci_dev;
 634         INIT_LIST_HEAD(&vp_dev->virtqueues);
 635         spin_lock_init(&vp_dev->lock);
 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
 652         /* we use the subsystem vendor/device id as the virtio vendor/device
 653          * id.  this allows us to use the same PCI vendor/device id for all
 654          * virtio devices and to identify the particular virtio driver by
 655          * the subsytem ids */
 656         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 657         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 658
 659         /* finally register the virtio device */
 660         err = register_virtio_device(&vp_dev->vdev);
 661         if (err)
 662                 goto out_set_drvdata;
 663
 664         return 0;
 665
 666 out_set_drvdata:
 667         pci_set_drvdata(pci_dev, NULL);
 668         pci_iounmap(pci_dev, vp_dev->ioaddr);
 669 out_req_regions:
 670         pci_release_regions(pci_dev);
 671 out_enable_device:
 672         pci_disable_device(pci_dev);
 673 out:
 674         kfree(vp_dev);
 675         return err;
 676 }
 677
 678 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
 679 {
 680         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 681
 682         unregister_virtio_device(&vp_dev->vdev);
 683 }
 684
 685 #ifdef CONFIG_PM
 686 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
 687 {
 688         pci_save_state(pci_dev);
 689         pci_set_power_state(pci_dev, PCI_D3hot);
 690         return 0;
 691 }
 692
 693 static int virtio_pci_resume(struct pci_dev *pci_dev)
 694 {
 695         pci_restore_state(pci_dev);
 696         pci_set_power_state(pci_dev, PCI_D0);
 697         return 0;
 698 }
 699 #endif
 700
 701 static struct pci_driver virtio_pci_driver = {
 702         .name           = "virtio-pci",
 703         .id_table       = virtio_pci_id_table,
 704         .probe          = virtio_pci_probe,
 705         .remove         = virtio_pci_remove,
 706 #ifdef CONFIG_PM
 707         .suspend        = virtio_pci_suspend,
 708         .resume         = virtio_pci_resume,
 709 #endif
 710 };
 711
 712 static int __init virtio_pci_init(void)
 713 {
 714         int err;
 715
 716         virtio_pci_root = root_device_register("virtio-pci");
 717         if (IS_ERR(virtio_pci_root))
 718                 return PTR_ERR(virtio_pci_root);
 719
 720         err = pci_register_driver(&virtio_pci_driver);
 721         if (err)
 722                 root_device_unregister(virtio_pci_root);
 723
 724         return err;
 725 }
 726
 727 module_init(virtio_pci_init);
 728
 729 static void __exit virtio_pci_exit(void)
 730 {
 731         pci_unregister_driver(&virtio_pci_driver);
 732         root_device_unregister(virtio_pci_root);
 733 }
 734
 735 module_exit(virtio_pci_exit);