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
  59         /* Status saved during hibernate/restore */
  60         u8 saved_status;
  61
  62         /* Whether we have vector per vq */
  63         bool per_vq_vectors;
  64 };
  65
  66 /* Constants for MSI-X */
  67 /* Use first vector for configuration changes, second and the rest for
  68  * virtqueues Thus, we need at least 2 vectors for MSI. */
  69 enum {
  70         VP_MSIX_CONFIG_VECTOR = 0,
  71         VP_MSIX_VQ_VECTOR = 1,
  72 };
  73
  74 struct virtio_pci_vq_info
  75 {
  76         /* the actual virtqueue */
  77         struct virtqueue *vq;
  78
  79         /* the number of entries in the queue */
  80         int num;
  81
  82         /* the index of the queue */
  83         int queue_index;
  84
  85         /* the virtual address of the ring queue */
  86         void *queue;
  87
  88         /* the list node for the virtqueues list */
  89         struct list_head node;
  90
  91         /* MSI-X vector (or none) */
  92         unsigned msix_vector;
  93 };
  94
  95 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
  96 static struct pci_device_id virtio_pci_id_table[] = {
  97         { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
  98         { 0 },
  99 };
 100
 101 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
 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 /* wait for pending irq handlers */
 177 static void vp_synchronize_vectors(struct virtio_device *vdev)
 178 {
 179         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 180         int i;
 181
 182         if (vp_dev->intx_enabled)
 183                 synchronize_irq(vp_dev->pci_dev->irq);
 184
 185         for (i = 0; i < vp_dev->msix_vectors; ++i)
 186                 synchronize_irq(vp_dev->msix_entries[i].vector);
 187 }
 188
 189 static void vp_reset(struct virtio_device *vdev)
 190 {
 191         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 192         /* 0 status means a reset. */
 193         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 194         /* Flush out the status write, and flush in device writes,
 195          * including MSi-X interrupts, if any. */
 196         ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 197         /* Flush pending VQ/configuration callbacks. */
 198         vp_synchronize_vectors(vdev);
 199 }
 200
 201 /* the notify function used when creating a virt queue */
 202 static void vp_notify(struct virtqueue *vq)
 203 {
 204         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 205         struct virtio_pci_vq_info *info = vq->priv;
 206
 207         /* we write the queue's selector into the notification register to
 208          * signal the other end */
 209         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 210 }
 211
 212 /* Handle a configuration change: Tell driver if it wants to know. */
 213 static irqreturn_t vp_config_changed(int irq, void *opaque)
 214 {
 215         struct virtio_pci_device *vp_dev = opaque;
 216         struct virtio_driver *drv;
 217         drv = container_of(vp_dev->vdev.dev.driver,
 218                            struct virtio_driver, driver);
 219
 220         if (drv && drv->config_changed)
 221                 drv->config_changed(&vp_dev->vdev);
 222         return IRQ_HANDLED;
 223 }
 224
 225 /* Notify all virtqueues on an interrupt. */
 226 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 227 {
 228         struct virtio_pci_device *vp_dev = opaque;
 229         struct virtio_pci_vq_info *info;
 230         irqreturn_t ret = IRQ_NONE;
 231         unsigned long flags;
 232
 233         spin_lock_irqsave(&vp_dev->lock, flags);
 234         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 235                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 236                         ret = IRQ_HANDLED;
 237         }
 238         spin_unlock_irqrestore(&vp_dev->lock, flags);
 239
 240         return ret;
 241 }
 242
 243 /* A small wrapper to also acknowledge the interrupt when it's handled.
 244  * I really need an EIO hook for the vring so I can ack the interrupt once we
 245  * know that we'll be handling the IRQ but before we invoke the callback since
 246  * the callback may notify the host which results in the host attempting to
 247  * raise an interrupt that we would then mask once we acknowledged the
 248  * interrupt. */
 249 static irqreturn_t vp_interrupt(int irq, void *opaque)
 250 {
 251         struct virtio_pci_device *vp_dev = opaque;
 252         u8 isr;
 253
 254         /* reading the ISR has the effect of also clearing it so it's very
 255          * important to save off the value. */
 256         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 257
 258         /* It's definitely not us if the ISR was not high */
 259         if (!isr)
 260                 return IRQ_NONE;
 261
 262         /* Configuration change?  Tell driver if it wants to know. */
 263         if (isr & VIRTIO_PCI_ISR_CONFIG)
 264                 vp_config_changed(irq, opaque);
 265
 266         return vp_vring_interrupt(irq, opaque);
 267 }
 268
 269 static void vp_free_vectors(struct virtio_device *vdev)
 270 {
 271         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 272         int i;
 273
 274         if (vp_dev->intx_enabled) {
 275                 free_irq(vp_dev->pci_dev->irq, vp_dev);
 276                 vp_dev->intx_enabled = 0;
 277         }
 278
 279         for (i = 0; i < vp_dev->msix_used_vectors; ++i)
 280                 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
 281
 282         if (vp_dev->msix_enabled) {
 283                 /* Disable the vector used for configuration */
 284                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 285                           vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 286                 /* Flush the write out to device */
 287                 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 288
 289                 pci_disable_msix(vp_dev->pci_dev);
 290                 vp_dev->msix_enabled = 0;
 291                 vp_dev->msix_vectors = 0;
 292         }
 293
 294         vp_dev->msix_used_vectors = 0;
 295         kfree(vp_dev->msix_names);
 296         vp_dev->msix_names = NULL;
 297         kfree(vp_dev->msix_entries);
 298         vp_dev->msix_entries = NULL;
 299 }
 300
 301 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
 302                                    bool per_vq_vectors)
 303 {
 304         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 305         const char *name = dev_name(&vp_dev->vdev.dev);
 306         unsigned i, v;
 307         int err = -ENOMEM;
 308
 309         vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 310                                        GFP_KERNEL);
 311         if (!vp_dev->msix_entries)
 312                 goto error;
 313         vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 314                                      GFP_KERNEL);
 315         if (!vp_dev->msix_names)
 316                 goto error;
 317
 318         for (i = 0; i < nvectors; ++i)
 319                 vp_dev->msix_entries[i].entry = i;
 320
 321         /* pci_enable_msix returns positive if we can't get this many. */
 322         err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
 323         if (err > 0)
 324                 err = -ENOSPC;
 325         if (err)
 326                 goto error;
 327         vp_dev->msix_vectors = nvectors;
 328         vp_dev->msix_enabled = 1;
 329
 330         /* Set the vector used for configuration */
 331         v = vp_dev->msix_used_vectors;
 332         snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 333                  "%s-config", name);
 334         err = request_irq(vp_dev->msix_entries[v].vector,
 335                           vp_config_changed, 0, vp_dev->msix_names[v],
 336                           vp_dev);
 337         if (err)
 338                 goto error;
 339         ++vp_dev->msix_used_vectors;
 340
 341         iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 342         /* Verify we had enough resources to assign the vector */
 343         v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 344         if (v == VIRTIO_MSI_NO_VECTOR) {
 345                 err = -EBUSY;
 346                 goto error;
 347         }
 348
 349         if (!per_vq_vectors) {
 350                 /* Shared vector for all VQs */
 351                 v = vp_dev->msix_used_vectors;
 352                 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 353                          "%s-virtqueues", name);
 354                 err = request_irq(vp_dev->msix_entries[v].vector,
 355                                   vp_vring_interrupt, 0, vp_dev->msix_names[v],
 356                                   vp_dev);
 357                 if (err)
 358                         goto error;
 359                 ++vp_dev->msix_used_vectors;
 360         }
 361         return 0;
 362 error:
 363         vp_free_vectors(vdev);
 364         return err;
 365 }
 366
 367 static int vp_request_intx(struct virtio_device *vdev)
 368 {
 369         int err;
 370         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 371
 372         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 373                           IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 374         if (!err)
 375                 vp_dev->intx_enabled = 1;
 376         return err;
 377 }
 378
 379 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 380                                   void (*callback)(struct virtqueue *vq),
 381                                   const char *name,
 382                                   u16 msix_vec)
 383 {
 384         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 385         struct virtio_pci_vq_info *info;
 386         struct virtqueue *vq;
 387         unsigned long flags, size;
 388         u16 num;
 389         int err;
 390
 391         /* Select the queue we're interested in */
 392         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 393
 394         /* Check if queue is either not available or already active. */
 395         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 396         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 397                 return ERR_PTR(-ENOENT);
 398
 399         /* allocate and fill out our structure the represents an active
 400          * queue */
 401         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 402         if (!info)
 403                 return ERR_PTR(-ENOMEM);
 404
 405         info->queue_index = index;
 406         info->num = num;
 407         info->msix_vector = msix_vec;
 408
 409         size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 410         info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 411         if (info->queue == NULL) {
 412                 err = -ENOMEM;
 413                 goto out_info;
 414         }
 415
 416         /* activate the queue */
 417         iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 418                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 419
 420         /* create the vring */
 421         vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
 422                                  true, info->queue, vp_notify, callback, name);
 423         if (!vq) {
 424                 err = -ENOMEM;
 425                 goto out_activate_queue;
 426         }
 427
 428         vq->priv = info;
 429         info->vq = vq;
 430
 431         if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 432                 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 433                 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 434                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 435                         err = -EBUSY;
 436                         goto out_assign;
 437                 }
 438         }
 439
 440         if (callback) {
 441                 spin_lock_irqsave(&vp_dev->lock, flags);
 442                 list_add(&info->node, &vp_dev->virtqueues);
 443                 spin_unlock_irqrestore(&vp_dev->lock, flags);
 444         } else {
 445                 INIT_LIST_HEAD(&info->node);
 446         }
 447
 448         return vq;
 449
 450 out_assign:
 451         vring_del_virtqueue(vq);
 452 out_activate_queue:
 453         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 454         free_pages_exact(info->queue, size);
 455 out_info:
 456         kfree(info);
 457         return ERR_PTR(err);
 458 }
 459
 460 static void vp_del_vq(struct virtqueue *vq)
 461 {
 462         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 463         struct virtio_pci_vq_info *info = vq->priv;
 464         unsigned long flags, size;
 465
 466         spin_lock_irqsave(&vp_dev->lock, flags);
 467         list_del(&info->node);
 468         spin_unlock_irqrestore(&vp_dev->lock, flags);
 469
 470         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 471
 472         if (vp_dev->msix_enabled) {
 473                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 474                           vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 475                 /* Flush the write out to device */
 476                 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 477         }
 478
 479         vring_del_virtqueue(vq);
 480
 481         /* Select and deactivate the queue */
 482         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 483
 484         size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 485         free_pages_exact(info->queue, size);
 486         kfree(info);
 487 }
 488
 489 /* the config->del_vqs() implementation */
 490 static void vp_del_vqs(struct virtio_device *vdev)
 491 {
 492         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 493         struct virtqueue *vq, *n;
 494         struct virtio_pci_vq_info *info;
 495
 496         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 497                 info = vq->priv;
 498                 if (vp_dev->per_vq_vectors &&
 499                         info->msix_vector != VIRTIO_MSI_NO_VECTOR)
 500                         free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 501                                  vq);
 502                 vp_del_vq(vq);
 503         }
 504         vp_dev->per_vq_vectors = false;
 505
 506         vp_free_vectors(vdev);
 507 }
 508
 509 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 510                               struct virtqueue *vqs[],
 511                               vq_callback_t *callbacks[],
 512                               const char *names[],
 513                               bool use_msix,
 514                               bool per_vq_vectors)
 515 {
 516         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 517         u16 msix_vec;
 518         int i, err, nvectors, allocated_vectors;
 519
 520         if (!use_msix) {
 521                 /* Old style: one normal interrupt for change and all vqs. */
 522                 err = vp_request_intx(vdev);
 523                 if (err)
 524                         goto error_request;
 525         } else {
 526                 if (per_vq_vectors) {
 527                         /* Best option: one for change interrupt, one per vq. */
 528                         nvectors = 1;
 529                         for (i = 0; i < nvqs; ++i)
 530                                 if (callbacks[i])
 531                                         ++nvectors;
 532                 } else {
 533                         /* Second best: one for change, shared for all vqs. */
 534                         nvectors = 2;
 535                 }
 536
 537                 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 538                 if (err)
 539                         goto error_request;
 540         }
 541
 542         vp_dev->per_vq_vectors = per_vq_vectors;
 543         allocated_vectors = vp_dev->msix_used_vectors;
 544         for (i = 0; i < nvqs; ++i) {
 545                 if (!callbacks[i] || !vp_dev->msix_enabled)
 546                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 547                 else if (vp_dev->per_vq_vectors)
 548                         msix_vec = allocated_vectors++;
 549                 else
 550                         msix_vec = VP_MSIX_VQ_VECTOR;
 551                 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 552                 if (IS_ERR(vqs[i])) {
 553                         err = PTR_ERR(vqs[i]);
 554                         goto error_find;
 555                 }
 556
 557                 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 558                         continue;
 559
 560                 /* allocate per-vq irq if available and necessary */
 561                 snprintf(vp_dev->msix_names[msix_vec],
 562                          sizeof *vp_dev->msix_names,
 563                          "%s-%s",
 564                          dev_name(&vp_dev->vdev.dev), names[i]);
 565                 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 566                                   vring_interrupt, 0,
 567                                   vp_dev->msix_names[msix_vec],
 568                                   vqs[i]);
 569                 if (err) {
 570                         vp_del_vq(vqs[i]);
 571                         goto error_find;
 572                 }
 573         }
 574         return 0;
 575
 576 error_find:
 577         vp_del_vqs(vdev);
 578
 579 error_request:
 580         return err;
 581 }
 582
 583 /* the config->find_vqs() implementation */
 584 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 585                        struct virtqueue *vqs[],
 586                        vq_callback_t *callbacks[],
 587                        const char *names[])
 588 {
 589         int err;
 590
 591         /* Try MSI-X with one vector per queue. */
 592         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 593         if (!err)
 594                 return 0;
 595         /* Fallback: MSI-X with one vector for config, one shared for queues. */
 596         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 597                                  true, false);
 598         if (!err)
 599                 return 0;
 600         /* Finally fall back to regular interrupts. */
 601         return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 602                                   false, false);
 603 }
 604
 605 static const char *vp_bus_name(struct virtio_device *vdev)
 606 {
 607         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 608
 609         return pci_name(vp_dev->pci_dev);
 610 }
 611
 612 static struct virtio_config_ops virtio_pci_config_ops = {
 613         .get            = vp_get,
 614         .set            = vp_set,
 615         .get_status     = vp_get_status,
 616         .set_status     = vp_set_status,
 617         .reset          = vp_reset,
 618         .find_vqs       = vp_find_vqs,
 619         .del_vqs        = vp_del_vqs,
 620         .get_features   = vp_get_features,
 621         .finalize_features = vp_finalize_features,
 622         .bus_name       = vp_bus_name,
 623 };
 624
 625 static void virtio_pci_release_dev(struct device *_d)
 626 {
 627         /*
 628          * No need for a release method as we allocate/free
 629          * all devices together with the pci devices.
 630          * Provide an empty one to avoid getting a warning from core.
 631          */
 632 }
 633
 634 /* the PCI probing function */
 635 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
 636                                       const struct pci_device_id *id)
 637 {
 638         struct virtio_pci_device *vp_dev;
 639         int err;
 640
 641         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 642         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 643                 return -ENODEV;
 644
 645         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 646                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 647                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 648                 return -ENODEV;
 649         }
 650
 651         /* allocate our structure and fill it out */
 652         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 653         if (vp_dev == NULL)
 654                 return -ENOMEM;
 655
 656         vp_dev->vdev.dev.parent = &pci_dev->dev;
 657         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 658         vp_dev->vdev.config = &virtio_pci_config_ops;
 659         vp_dev->pci_dev = pci_dev;
 660         INIT_LIST_HEAD(&vp_dev->virtqueues);
 661         spin_lock_init(&vp_dev->lock);
 662
 663         /* Disable MSI/MSIX to bring device to a known good state. */
 664         pci_msi_off(pci_dev);
 665
 666         /* enable the device */
 667         err = pci_enable_device(pci_dev);
 668         if (err)
 669                 goto out;
 670
 671         err = pci_request_regions(pci_dev, "virtio-pci");
 672         if (err)
 673                 goto out_enable_device;
 674
 675         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 676         if (vp_dev->ioaddr == NULL)
 677                 goto out_req_regions;
 678
 679         pci_set_drvdata(pci_dev, vp_dev);
 680         pci_set_master(pci_dev);
 681
 682         /* we use the subsystem vendor/device id as the virtio vendor/device
 683          * id.  this allows us to use the same PCI vendor/device id for all
 684          * virtio devices and to identify the particular virtio driver by
 685          * the subsystem ids */
 686         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 687         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 688
 689         /* finally register the virtio device */
 690         err = register_virtio_device(&vp_dev->vdev);
 691         if (err)
 692                 goto out_set_drvdata;
 693
 694         return 0;
 695
 696 out_set_drvdata:
 697         pci_set_drvdata(pci_dev, NULL);
 698         pci_iounmap(pci_dev, vp_dev->ioaddr);
 699 out_req_regions:
 700         pci_release_regions(pci_dev);
 701 out_enable_device:
 702         pci_disable_device(pci_dev);
 703 out:
 704         kfree(vp_dev);
 705         return err;
 706 }
 707
 708 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
 709 {
 710         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 711
 712         unregister_virtio_device(&vp_dev->vdev);
 713
 714         vp_del_vqs(&vp_dev->vdev);
 715         pci_set_drvdata(pci_dev, NULL);
 716         pci_iounmap(pci_dev, vp_dev->ioaddr);
 717         pci_release_regions(pci_dev);
 718         pci_disable_device(pci_dev);
 719         kfree(vp_dev);
 720 }
 721
 722 #ifdef CONFIG_PM
 723 static int virtio_pci_suspend(struct device *dev)
 724 {
 725         struct pci_dev *pci_dev = to_pci_dev(dev);
 726
 727         pci_save_state(pci_dev);
 728         pci_set_power_state(pci_dev, PCI_D3hot);
 729         return 0;
 730 }
 731
 732 static int virtio_pci_resume(struct device *dev)
 733 {
 734         struct pci_dev *pci_dev = to_pci_dev(dev);
 735
 736         pci_restore_state(pci_dev);
 737         pci_set_power_state(pci_dev, PCI_D0);
 738         return 0;
 739 }
 740
 741 static int virtio_pci_freeze(struct device *dev)
 742 {
 743         struct pci_dev *pci_dev = to_pci_dev(dev);
 744         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 745         struct virtio_driver *drv;
 746         int ret;
 747
 748         drv = container_of(vp_dev->vdev.dev.driver,
 749                            struct virtio_driver, driver);
 750
 751         ret = 0;
 752         vp_dev->saved_status = vp_get_status(&vp_dev->vdev);
 753         if (drv && drv->freeze)
 754                 ret = drv->freeze(&vp_dev->vdev);
 755
 756         if (!ret)
 757                 pci_disable_device(pci_dev);
 758         return ret;
 759 }
 760
 761 static int restore_common(struct device *dev)
 762 {
 763         struct pci_dev *pci_dev = to_pci_dev(dev);
 764         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 765         int ret;
 766
 767         ret = pci_enable_device(pci_dev);
 768         if (ret)
 769                 return ret;
 770         pci_set_master(pci_dev);
 771         vp_finalize_features(&vp_dev->vdev);
 772
 773         return ret;
 774 }
 775
 776 static int virtio_pci_thaw(struct device *dev)
 777 {
 778         struct pci_dev *pci_dev = to_pci_dev(dev);
 779         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 780         struct virtio_driver *drv;
 781         int ret;
 782
 783         ret = restore_common(dev);
 784         if (ret)
 785                 return ret;
 786
 787         drv = container_of(vp_dev->vdev.dev.driver,
 788                            struct virtio_driver, driver);
 789
 790         if (drv && drv->thaw)
 791                 ret = drv->thaw(&vp_dev->vdev);
 792         else if (drv && drv->restore)
 793                 ret = drv->restore(&vp_dev->vdev);
 794
 795         /* Finally, tell the device we're all set */
 796         if (!ret)
 797                 vp_set_status(&vp_dev->vdev, vp_dev->saved_status);
 798
 799         return ret;
 800 }
 801
 802 static int virtio_pci_restore(struct device *dev)
 803 {
 804         struct pci_dev *pci_dev = to_pci_dev(dev);
 805         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 806         struct virtio_driver *drv;
 807         int ret;
 808
 809         drv = container_of(vp_dev->vdev.dev.driver,
 810                            struct virtio_driver, driver);
 811
 812         ret = restore_common(dev);
 813         if (!ret && drv && drv->restore)
 814                 ret = drv->restore(&vp_dev->vdev);
 815
 816         /* Finally, tell the device we're all set */
 817         if (!ret)
 818                 vp_set_status(&vp_dev->vdev, vp_dev->saved_status);
 819
 820         return ret;
 821 }
 822
 823 static const struct dev_pm_ops virtio_pci_pm_ops = {
 824         .suspend        = virtio_pci_suspend,
 825         .resume         = virtio_pci_resume,
 826         .freeze         = virtio_pci_freeze,
 827         .thaw           = virtio_pci_thaw,
 828         .restore        = virtio_pci_restore,
 829         .poweroff       = virtio_pci_suspend,
 830 };
 831 #endif
 832
 833 static struct pci_driver virtio_pci_driver = {
 834         .name           = "virtio-pci",
 835         .id_table       = virtio_pci_id_table,
 836         .probe          = virtio_pci_probe,
 837         .remove         = __devexit_p(virtio_pci_remove),
 838 #ifdef CONFIG_PM
 839         .driver.pm      = &virtio_pci_pm_ops,
 840 #endif
 841 };
 842
 843 static int __init virtio_pci_init(void)
 844 {
 845         return pci_register_driver(&virtio_pci_driver);
 846 }
 847
 848 module_init(virtio_pci_init);
 849
 850 static void __exit virtio_pci_exit(void)
 851 {
 852         pci_unregister_driver(&virtio_pci_driver);
 853 }
 854
 855 module_exit(virtio_pci_exit);