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 /* wait for pending irq handlers */
 173 static void vp_synchronize_vectors(struct virtio_device *vdev)
 174 {
 175         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 176         int i;
 177
 178         if (vp_dev->intx_enabled)
 179                 synchronize_irq(vp_dev->pci_dev->irq);
 180
 181         for (i = 0; i < vp_dev->msix_vectors; ++i)
 182                 synchronize_irq(vp_dev->msix_entries[i].vector);
 183 }
 184
 185 static void vp_reset(struct virtio_device *vdev)
 186 {
 187         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 188         /* 0 status means a reset. */
 189         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 190         /* Flush out the status write, and flush in device writes,
 191          * including MSi-X interrupts, if any. */
 192         ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 193         /* Flush pending VQ/configuration callbacks. */
 194         vp_synchronize_vectors(vdev);
 195 }
 196
 197 /* the notify function used when creating a virt queue */
 198 static void vp_notify(struct virtqueue *vq)
 199 {
 200         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 201         struct virtio_pci_vq_info *info = vq->priv;
 202
 203         /* we write the queue's selector into the notification register to
 204          * signal the other end */
 205         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 206 }
 207
 208 /* Handle a configuration change: Tell driver if it wants to know. */
 209 static irqreturn_t vp_config_changed(int irq, void *opaque)
 210 {
 211         struct virtio_pci_device *vp_dev = opaque;
 212         struct virtio_driver *drv;
 213         drv = container_of(vp_dev->vdev.dev.driver,
 214                            struct virtio_driver, driver);
 215
 216         if (drv && drv->config_changed)
 217                 drv->config_changed(&vp_dev->vdev);
 218         return IRQ_HANDLED;
 219 }
 220
 221 /* Notify all virtqueues on an interrupt. */
 222 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 223 {
 224         struct virtio_pci_device *vp_dev = opaque;
 225         struct virtio_pci_vq_info *info;
 226         irqreturn_t ret = IRQ_NONE;
 227         unsigned long flags;
 228
 229         spin_lock_irqsave(&vp_dev->lock, flags);
 230         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 231                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 232                         ret = IRQ_HANDLED;
 233         }
 234         spin_unlock_irqrestore(&vp_dev->lock, flags);
 235
 236         return ret;
 237 }
 238
 239 /* A small wrapper to also acknowledge the interrupt when it's handled.
 240  * I really need an EIO hook for the vring so I can ack the interrupt once we
 241  * know that we'll be handling the IRQ but before we invoke the callback since
 242  * the callback may notify the host which results in the host attempting to
 243  * raise an interrupt that we would then mask once we acknowledged the
 244  * interrupt. */
 245 static irqreturn_t vp_interrupt(int irq, void *opaque)
 246 {
 247         struct virtio_pci_device *vp_dev = opaque;
 248         u8 isr;
 249
 250         /* reading the ISR has the effect of also clearing it so it's very
 251          * important to save off the value. */
 252         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 253
 254         /* It's definitely not us if the ISR was not high */
 255         if (!isr)
 256                 return IRQ_NONE;
 257
 258         /* Configuration change?  Tell driver if it wants to know. */
 259         if (isr & VIRTIO_PCI_ISR_CONFIG)
 260                 vp_config_changed(irq, opaque);
 261
 262         return vp_vring_interrupt(irq, opaque);
 263 }
 264
 265 static void vp_free_vectors(struct virtio_device *vdev)
 266 {
 267         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 268         int i;
 269
 270         if (vp_dev->intx_enabled) {
 271                 free_irq(vp_dev->pci_dev->irq, vp_dev);
 272                 vp_dev->intx_enabled = 0;
 273         }
 274
 275         for (i = 0; i < vp_dev->msix_used_vectors; ++i)
 276                 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
 277
 278         if (vp_dev->msix_enabled) {
 279                 /* Disable the vector used for configuration */
 280                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 281                           vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 282                 /* Flush the write out to device */
 283                 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 284
 285                 pci_disable_msix(vp_dev->pci_dev);
 286                 vp_dev->msix_enabled = 0;
 287                 vp_dev->msix_vectors = 0;
 288         }
 289
 290         vp_dev->msix_used_vectors = 0;
 291         kfree(vp_dev->msix_names);
 292         vp_dev->msix_names = NULL;
 293         kfree(vp_dev->msix_entries);
 294         vp_dev->msix_entries = NULL;
 295 }
 296
 297 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
 298                                    bool per_vq_vectors)
 299 {
 300         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 301         const char *name = dev_name(&vp_dev->vdev.dev);
 302         unsigned i, v;
 303         int err = -ENOMEM;
 304
 305         vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 306                                        GFP_KERNEL);
 307         if (!vp_dev->msix_entries)
 308                 goto error;
 309         vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 310                                      GFP_KERNEL);
 311         if (!vp_dev->msix_names)
 312                 goto error;
 313
 314         for (i = 0; i < nvectors; ++i)
 315                 vp_dev->msix_entries[i].entry = i;
 316
 317         /* pci_enable_msix returns positive if we can't get this many. */
 318         err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
 319         if (err > 0)
 320                 err = -ENOSPC;
 321         if (err)
 322                 goto error;
 323         vp_dev->msix_vectors = nvectors;
 324         vp_dev->msix_enabled = 1;
 325
 326         /* Set the vector used for configuration */
 327         v = vp_dev->msix_used_vectors;
 328         snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 329                  "%s-config", name);
 330         err = request_irq(vp_dev->msix_entries[v].vector,
 331                           vp_config_changed, 0, vp_dev->msix_names[v],
 332                           vp_dev);
 333         if (err)
 334                 goto error;
 335         ++vp_dev->msix_used_vectors;
 336
 337         iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 338         /* Verify we had enough resources to assign the vector */
 339         v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 340         if (v == VIRTIO_MSI_NO_VECTOR) {
 341                 err = -EBUSY;
 342                 goto error;
 343         }
 344
 345         if (!per_vq_vectors) {
 346                 /* Shared vector for all VQs */
 347                 v = vp_dev->msix_used_vectors;
 348                 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 349                          "%s-virtqueues", name);
 350                 err = request_irq(vp_dev->msix_entries[v].vector,
 351                                   vp_vring_interrupt, 0, vp_dev->msix_names[v],
 352                                   vp_dev);
 353                 if (err)
 354                         goto error;
 355                 ++vp_dev->msix_used_vectors;
 356         }
 357         return 0;
 358 error:
 359         vp_free_vectors(vdev);
 360         return err;
 361 }
 362
 363 static int vp_request_intx(struct virtio_device *vdev)
 364 {
 365         int err;
 366         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 367
 368         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 369                           IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 370         if (!err)
 371                 vp_dev->intx_enabled = 1;
 372         return err;
 373 }
 374
 375 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 376                                   void (*callback)(struct virtqueue *vq),
 377                                   const char *name,
 378                                   u16 msix_vec)
 379 {
 380         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 381         struct virtio_pci_vq_info *info;
 382         struct virtqueue *vq;
 383         unsigned long flags, size;
 384         u16 num;
 385         int err;
 386
 387         /* Select the queue we're interested in */
 388         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 389
 390         /* Check if queue is either not available or already active. */
 391         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 392         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 393                 return ERR_PTR(-ENOENT);
 394
 395         /* allocate and fill out our structure the represents an active
 396          * queue */
 397         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 398         if (!info)
 399                 return ERR_PTR(-ENOMEM);
 400
 401         info->queue_index = index;
 402         info->num = num;
 403         info->msix_vector = msix_vec;
 404
 405         size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 406         info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 407         if (info->queue == NULL) {
 408                 err = -ENOMEM;
 409                 goto out_info;
 410         }
 411
 412         /* activate the queue */
 413         iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 414                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 415
 416         /* create the vring */
 417         vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
 418                                  vdev, info->queue, vp_notify, callback, name);
 419         if (!vq) {
 420                 err = -ENOMEM;
 421                 goto out_activate_queue;
 422         }
 423
 424         vq->priv = info;
 425         info->vq = vq;
 426
 427         if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 428                 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 429                 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 430                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 431                         err = -EBUSY;
 432                         goto out_assign;
 433                 }
 434         }
 435
 436         if (callback) {
 437                 spin_lock_irqsave(&vp_dev->lock, flags);
 438                 list_add(&info->node, &vp_dev->virtqueues);
 439                 spin_unlock_irqrestore(&vp_dev->lock, flags);
 440         } else {
 441                 INIT_LIST_HEAD(&info->node);
 442         }
 443
 444         return vq;
 445
 446 out_assign:
 447         vring_del_virtqueue(vq);
 448 out_activate_queue:
 449         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 450         free_pages_exact(info->queue, size);
 451 out_info:
 452         kfree(info);
 453         return ERR_PTR(err);
 454 }
 455
 456 static void vp_del_vq(struct virtqueue *vq)
 457 {
 458         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 459         struct virtio_pci_vq_info *info = vq->priv;
 460         unsigned long flags, size;
 461
 462         spin_lock_irqsave(&vp_dev->lock, flags);
 463         list_del(&info->node);
 464         spin_unlock_irqrestore(&vp_dev->lock, flags);
 465
 466         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 467
 468         if (vp_dev->msix_enabled) {
 469                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 470                           vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 471                 /* Flush the write out to device */
 472                 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 473         }
 474
 475         vring_del_virtqueue(vq);
 476
 477         /* Select and deactivate the queue */
 478         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 479
 480         size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 481         free_pages_exact(info->queue, size);
 482         kfree(info);
 483 }
 484
 485 /* the config->del_vqs() implementation */
 486 static void vp_del_vqs(struct virtio_device *vdev)
 487 {
 488         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 489         struct virtqueue *vq, *n;
 490         struct virtio_pci_vq_info *info;
 491
 492         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 493                 info = vq->priv;
 494                 if (vp_dev->per_vq_vectors &&
 495                         info->msix_vector != VIRTIO_MSI_NO_VECTOR)
 496                         free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 497                                  vq);
 498                 vp_del_vq(vq);
 499         }
 500         vp_dev->per_vq_vectors = false;
 501
 502         vp_free_vectors(vdev);
 503 }
 504
 505 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 506                               struct virtqueue *vqs[],
 507                               vq_callback_t *callbacks[],
 508                               const char *names[],
 509                               bool use_msix,
 510                               bool per_vq_vectors)
 511 {
 512         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 513         u16 msix_vec;
 514         int i, err, nvectors, allocated_vectors;
 515
 516         if (!use_msix) {
 517                 /* Old style: one normal interrupt for change and all vqs. */
 518                 err = vp_request_intx(vdev);
 519                 if (err)
 520                         goto error_request;
 521         } else {
 522                 if (per_vq_vectors) {
 523                         /* Best option: one for change interrupt, one per vq. */
 524                         nvectors = 1;
 525                         for (i = 0; i < nvqs; ++i)
 526                                 if (callbacks[i])
 527                                         ++nvectors;
 528                 } else {
 529                         /* Second best: one for change, shared for all vqs. */
 530                         nvectors = 2;
 531                 }
 532
 533                 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 534                 if (err)
 535                         goto error_request;
 536         }
 537
 538         vp_dev->per_vq_vectors = per_vq_vectors;
 539         allocated_vectors = vp_dev->msix_used_vectors;
 540         for (i = 0; i < nvqs; ++i) {
 541                 if (!callbacks[i] || !vp_dev->msix_enabled)
 542                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 543                 else if (vp_dev->per_vq_vectors)
 544                         msix_vec = allocated_vectors++;
 545                 else
 546                         msix_vec = VP_MSIX_VQ_VECTOR;
 547                 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 548                 if (IS_ERR(vqs[i])) {
 549                         err = PTR_ERR(vqs[i]);
 550                         goto error_find;
 551                 }
 552
 553                 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 554                         continue;
 555
 556                 /* allocate per-vq irq if available and necessary */
 557                 snprintf(vp_dev->msix_names[msix_vec],
 558                          sizeof *vp_dev->msix_names,
 559                          "%s-%s",
 560                          dev_name(&vp_dev->vdev.dev), names[i]);
 561                 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 562                                   vring_interrupt, 0,
 563                                   vp_dev->msix_names[msix_vec],
 564                                   vqs[i]);
 565                 if (err) {
 566                         vp_del_vq(vqs[i]);
 567                         goto error_find;
 568                 }
 569         }
 570         return 0;
 571
 572 error_find:
 573         vp_del_vqs(vdev);
 574
 575 error_request:
 576         return err;
 577 }
 578
 579 /* the config->find_vqs() implementation */
 580 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 581                        struct virtqueue *vqs[],
 582                        vq_callback_t *callbacks[],
 583                        const char *names[])
 584 {
 585         int err;
 586
 587         /* Try MSI-X with one vector per queue. */
 588         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 589         if (!err)
 590                 return 0;
 591         /* Fallback: MSI-X with one vector for config, one shared for queues. */
 592         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 593                                  true, false);
 594         if (!err)
 595                 return 0;
 596         /* Finally fall back to regular interrupts. */
 597         return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 598                                   false, false);
 599 }
 600
 601 static struct virtio_config_ops virtio_pci_config_ops = {
 602         .get            = vp_get,
 603         .set            = vp_set,
 604         .get_status     = vp_get_status,
 605         .set_status     = vp_set_status,
 606         .reset          = vp_reset,
 607         .find_vqs       = vp_find_vqs,
 608         .del_vqs        = vp_del_vqs,
 609         .get_features   = vp_get_features,
 610         .finalize_features = vp_finalize_features,
 611 };
 612
 613 static void virtio_pci_release_dev(struct device *_d)
 614 {
 615         /*
 616          * No need for a release method as we allocate/free
 617          * all devices together with the pci devices.
 618          * Provide an empty one to avoid getting a warning from core.
 619          */
 620 }
 621
 622 /* the PCI probing function */
 623 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
 624                                       const struct pci_device_id *id)
 625 {
 626         struct virtio_pci_device *vp_dev;
 627         int err;
 628
 629         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 630         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 631                 return -ENODEV;
 632
 633         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 634                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 635                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 636                 return -ENODEV;
 637         }
 638
 639         /* allocate our structure and fill it out */
 640         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 641         if (vp_dev == NULL)
 642                 return -ENOMEM;
 643
 644         vp_dev->vdev.dev.parent = &pci_dev->dev;
 645         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 646         vp_dev->vdev.config = &virtio_pci_config_ops;
 647         vp_dev->pci_dev = pci_dev;
 648         INIT_LIST_HEAD(&vp_dev->virtqueues);
 649         spin_lock_init(&vp_dev->lock);
 650
 651         /* Disable MSI/MSIX to bring device to a known good state. */
 652         pci_msi_off(pci_dev);
 653
 654         /* enable the device */
 655         err = pci_enable_device(pci_dev);
 656         if (err)
 657                 goto out;
 658
 659         err = pci_request_regions(pci_dev, "virtio-pci");
 660         if (err)
 661                 goto out_enable_device;
 662
 663         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 664         if (vp_dev->ioaddr == NULL)
 665                 goto out_req_regions;
 666
 667         pci_set_drvdata(pci_dev, vp_dev);
 668         pci_set_master(pci_dev);
 669
 670         /* we use the subsystem vendor/device id as the virtio vendor/device
 671          * id.  this allows us to use the same PCI vendor/device id for all
 672          * virtio devices and to identify the particular virtio driver by
 673          * the subsystem ids */
 674         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 675         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 676
 677         /* finally register the virtio device */
 678         err = register_virtio_device(&vp_dev->vdev);
 679         if (err)
 680                 goto out_set_drvdata;
 681
 682         return 0;
 683
 684 out_set_drvdata:
 685         pci_set_drvdata(pci_dev, NULL);
 686         pci_iounmap(pci_dev, vp_dev->ioaddr);
 687 out_req_regions:
 688         pci_release_regions(pci_dev);
 689 out_enable_device:
 690         pci_disable_device(pci_dev);
 691 out:
 692         kfree(vp_dev);
 693         return err;
 694 }
 695
 696 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
 697 {
 698         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 699
 700         unregister_virtio_device(&vp_dev->vdev);
 701
 702         vp_del_vqs(&vp_dev->vdev);
 703         pci_set_drvdata(pci_dev, NULL);
 704         pci_iounmap(pci_dev, vp_dev->ioaddr);
 705         pci_release_regions(pci_dev);
 706         pci_disable_device(pci_dev);
 707         kfree(vp_dev);
 708 }
 709
 710 #ifdef CONFIG_PM
 711 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
 712 {
 713         pci_save_state(pci_dev);
 714         pci_set_power_state(pci_dev, PCI_D3hot);
 715         return 0;
 716 }
 717
 718 static int virtio_pci_resume(struct pci_dev *pci_dev)
 719 {
 720         pci_restore_state(pci_dev);
 721         pci_set_power_state(pci_dev, PCI_D0);
 722         return 0;
 723 }
 724 #endif
 725
 726 static struct pci_driver virtio_pci_driver = {
 727         .name           = "virtio-pci",
 728         .id_table       = virtio_pci_id_table,
 729         .probe          = virtio_pci_probe,
 730         .remove         = __devexit_p(virtio_pci_remove),
 731 #ifdef CONFIG_PM
 732         .suspend        = virtio_pci_suspend,
 733         .resume         = virtio_pci_resume,
 734 #endif
 735 };
 736
 737 static int __init virtio_pci_init(void)
 738 {
 739         return pci_register_driver(&virtio_pci_driver);
 740 }
 741
 742 module_init(virtio_pci_init);
 743
 744 static void __exit virtio_pci_exit(void)
 745 {
 746         pci_unregister_driver(&virtio_pci_driver);
 747 }
 748
 749 module_exit(virtio_pci_exit);