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         cpumask_var_t *msix_affinity_masks;
  52         /* Name strings for interrupts. This size should be enough,
  53          * and I'm too lazy to allocate each name separately. */
  54         char (*msix_names)[256];
  55         /* Number of available vectors */
  56         unsigned msix_vectors;
  57         /* Vectors allocated, excluding per-vq vectors if any */
  58         unsigned msix_used_vectors;
  59
  60         /* Whether we have vector per vq */
  61         bool per_vq_vectors;
  62 };
  63
  64 /* Constants for MSI-X */
  65 /* Use first vector for configuration changes, second and the rest for
  66  * virtqueues Thus, we need at least 2 vectors for MSI. */
  67 enum {
  68         VP_MSIX_CONFIG_VECTOR = 0,
  69         VP_MSIX_VQ_VECTOR = 1,
  70 };
  71
  72 struct virtio_pci_vq_info
  73 {
  74         /* the actual virtqueue */
  75         struct virtqueue *vq;
  76
  77         /* the number of entries in the queue */
  78         int num;
  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 const struct pci_device_id virtio_pci_id_table[] = {
  92         { PCI_DEVICE(0x1af4, PCI_ANY_ID) },
  93         { 0 }
  94 };
  95
  96 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
  97
  98 /* Convert a generic virtio device to our structure */
  99 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
 100 {
 101         return container_of(vdev, struct virtio_pci_device, vdev);
 102 }
 103
 104 /* virtio config->get_features() implementation */
 105 static u32 vp_get_features(struct virtio_device *vdev)
 106 {
 107         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 108
 109         /* When someone needs more than 32 feature bits, we'll need to
 110          * steal a bit to indicate that the rest are somewhere else. */
 111         return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
 112 }
 113
 114 /* virtio config->finalize_features() implementation */
 115 static void vp_finalize_features(struct virtio_device *vdev)
 116 {
 117         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 118
 119         /* Give virtio_ring a chance to accept features. */
 120         vring_transport_features(vdev);
 121
 122         /* We only support 32 feature bits. */
 123         BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
 124         iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
 125 }
 126
 127 /* virtio config->get() implementation */
 128 static void vp_get(struct virtio_device *vdev, unsigned offset,
 129                    void *buf, unsigned len)
 130 {
 131         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 132         void __iomem *ioaddr = vp_dev->ioaddr +
 133                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 134         u8 *ptr = buf;
 135         int i;
 136
 137         for (i = 0; i < len; i++)
 138                 ptr[i] = ioread8(ioaddr + i);
 139 }
 140
 141 /* the config->set() implementation.  it's symmetric to the config->get()
 142  * implementation */
 143 static void vp_set(struct virtio_device *vdev, unsigned offset,
 144                    const void *buf, unsigned len)
 145 {
 146         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 147         void __iomem *ioaddr = vp_dev->ioaddr +
 148                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 149         const u8 *ptr = buf;
 150         int i;
 151
 152         for (i = 0; i < len; i++)
 153                 iowrite8(ptr[i], ioaddr + i);
 154 }
 155
 156 /* config->{get,set}_status() implementations */
 157 static u8 vp_get_status(struct virtio_device *vdev)
 158 {
 159         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 160         return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 161 }
 162
 163 static void vp_set_status(struct virtio_device *vdev, u8 status)
 164 {
 165         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 166         /* We should never be setting status to 0. */
 167         BUG_ON(status == 0);
 168         iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 169 }
 170
 171 /* wait for pending irq handlers */
 172 static void vp_synchronize_vectors(struct virtio_device *vdev)
 173 {
 174         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 175         int i;
 176
 177         if (vp_dev->intx_enabled)
 178                 synchronize_irq(vp_dev->pci_dev->irq);
 179
 180         for (i = 0; i < vp_dev->msix_vectors; ++i)
 181                 synchronize_irq(vp_dev->msix_entries[i].vector);
 182 }
 183
 184 static void vp_reset(struct virtio_device *vdev)
 185 {
 186         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 187         /* 0 status means a reset. */
 188         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 189         /* Flush out the status write, and flush in device writes,
 190          * including MSi-X interrupts, if any. */
 191         ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 192         /* Flush pending VQ/configuration callbacks. */
 193         vp_synchronize_vectors(vdev);
 194 }
 195
 196 /* the notify function used when creating a virt queue */
 197 static bool vp_notify(struct virtqueue *vq)
 198 {
 199         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 200
 201         /* we write the queue's selector into the notification register to
 202          * signal the other end */
 203         iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 204         return true;
 205 }
 206
 207 /* Handle a configuration change: Tell driver if it wants to know. */
 208 static irqreturn_t vp_config_changed(int irq, void *opaque)
 209 {
 210         struct virtio_pci_device *vp_dev = opaque;
 211
 212         virtio_config_changed(&vp_dev->vdev);
 213         return IRQ_HANDLED;
 214 }
 215
 216 /* Notify all virtqueues on an interrupt. */
 217 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 218 {
 219         struct virtio_pci_device *vp_dev = opaque;
 220         struct virtio_pci_vq_info *info;
 221         irqreturn_t ret = IRQ_NONE;
 222         unsigned long flags;
 223
 224         spin_lock_irqsave(&vp_dev->lock, flags);
 225         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 226                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 227                         ret = IRQ_HANDLED;
 228         }
 229         spin_unlock_irqrestore(&vp_dev->lock, flags);
 230
 231         return ret;
 232 }
 233
 234 /* A small wrapper to also acknowledge the interrupt when it's handled.
 235  * I really need an EIO hook for the vring so I can ack the interrupt once we
 236  * know that we'll be handling the IRQ but before we invoke the callback since
 237  * the callback may notify the host which results in the host attempting to
 238  * raise an interrupt that we would then mask once we acknowledged the
 239  * interrupt. */
 240 static irqreturn_t vp_interrupt(int irq, void *opaque)
 241 {
 242         struct virtio_pci_device *vp_dev = opaque;
 243         u8 isr;
 244
 245         /* reading the ISR has the effect of also clearing it so it's very
 246          * important to save off the value. */
 247         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 248
 249         /* It's definitely not us if the ISR was not high */
 250         if (!isr)
 251                 return IRQ_NONE;
 252
 253         /* Configuration change?  Tell driver if it wants to know. */
 254         if (isr & VIRTIO_PCI_ISR_CONFIG)
 255                 vp_config_changed(irq, opaque);
 256
 257         return vp_vring_interrupt(irq, opaque);
 258 }
 259
 260 static void vp_free_vectors(struct virtio_device *vdev)
 261 {
 262         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 263         int i;
 264
 265         if (vp_dev->intx_enabled) {
 266                 free_irq(vp_dev->pci_dev->irq, vp_dev);
 267                 vp_dev->intx_enabled = 0;
 268         }
 269
 270         for (i = 0; i < vp_dev->msix_used_vectors; ++i)
 271                 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
 272
 273         for (i = 0; i < vp_dev->msix_vectors; i++)
 274                 if (vp_dev->msix_affinity_masks[i])
 275                         free_cpumask_var(vp_dev->msix_affinity_masks[i]);
 276
 277         if (vp_dev->msix_enabled) {
 278                 /* Disable the vector used for configuration */
 279                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 280                           vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 281                 /* Flush the write out to device */
 282                 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 283
 284                 pci_disable_msix(vp_dev->pci_dev);
 285                 vp_dev->msix_enabled = 0;
 286         }
 287
 288         vp_dev->msix_vectors = 0;
 289         vp_dev->msix_used_vectors = 0;
 290         kfree(vp_dev->msix_names);
 291         vp_dev->msix_names = NULL;
 292         kfree(vp_dev->msix_entries);
 293         vp_dev->msix_entries = NULL;
 294         kfree(vp_dev->msix_affinity_masks);
 295         vp_dev->msix_affinity_masks = NULL;
 296 }
 297
 298 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
 299                                    bool per_vq_vectors)
 300 {
 301         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 302         const char *name = dev_name(&vp_dev->vdev.dev);
 303         unsigned i, v;
 304         int err = -ENOMEM;
 305
 306         vp_dev->msix_vectors = nvectors;
 307
 308         vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 309                                        GFP_KERNEL);
 310         if (!vp_dev->msix_entries)
 311                 goto error;
 312         vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 313                                      GFP_KERNEL);
 314         if (!vp_dev->msix_names)
 315                 goto error;
 316         vp_dev->msix_affinity_masks
 317                 = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
 318                           GFP_KERNEL);
 319         if (!vp_dev->msix_affinity_masks)
 320                 goto error;
 321         for (i = 0; i < nvectors; ++i)
 322                 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
 323                                         GFP_KERNEL))
 324                         goto error;
 325
 326         for (i = 0; i < nvectors; ++i)
 327                 vp_dev->msix_entries[i].entry = i;
 328
 329         err = pci_enable_msix_exact(vp_dev->pci_dev,
 330                                     vp_dev->msix_entries, nvectors);
 331         if (err)
 332                 goto error;
 333         vp_dev->msix_enabled = 1;
 334
 335         /* Set the vector used for configuration */
 336         v = vp_dev->msix_used_vectors;
 337         snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 338                  "%s-config", name);
 339         err = request_irq(vp_dev->msix_entries[v].vector,
 340                           vp_config_changed, 0, vp_dev->msix_names[v],
 341                           vp_dev);
 342         if (err)
 343                 goto error;
 344         ++vp_dev->msix_used_vectors;
 345
 346         iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 347         /* Verify we had enough resources to assign the vector */
 348         v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 349         if (v == VIRTIO_MSI_NO_VECTOR) {
 350                 err = -EBUSY;
 351                 goto error;
 352         }
 353
 354         if (!per_vq_vectors) {
 355                 /* Shared vector for all VQs */
 356                 v = vp_dev->msix_used_vectors;
 357                 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 358                          "%s-virtqueues", name);
 359                 err = request_irq(vp_dev->msix_entries[v].vector,
 360                                   vp_vring_interrupt, 0, vp_dev->msix_names[v],
 361                                   vp_dev);
 362                 if (err)
 363                         goto error;
 364                 ++vp_dev->msix_used_vectors;
 365         }
 366         return 0;
 367 error:
 368         vp_free_vectors(vdev);
 369         return err;
 370 }
 371
 372 static int vp_request_intx(struct virtio_device *vdev)
 373 {
 374         int err;
 375         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 376
 377         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 378                           IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 379         if (!err)
 380                 vp_dev->intx_enabled = 1;
 381         return err;
 382 }
 383
 384 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 385                                   void (*callback)(struct virtqueue *vq),
 386                                   const char *name,
 387                                   u16 msix_vec)
 388 {
 389         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 390         struct virtio_pci_vq_info *info;
 391         struct virtqueue *vq;
 392         unsigned long flags, size;
 393         u16 num;
 394         int err;
 395
 396         /* Select the queue we're interested in */
 397         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 398
 399         /* Check if queue is either not available or already active. */
 400         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 401         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 402                 return ERR_PTR(-ENOENT);
 403
 404         /* allocate and fill out our structure the represents an active
 405          * queue */
 406         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 407         if (!info)
 408                 return ERR_PTR(-ENOMEM);
 409
 410         info->num = num;
 411         info->msix_vector = msix_vec;
 412
 413         size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 414         info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 415         if (info->queue == NULL) {
 416                 err = -ENOMEM;
 417                 goto out_info;
 418         }
 419
 420         /* activate the queue */
 421         iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 422                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 423
 424         /* create the vring */
 425         vq = vring_new_virtqueue(index, info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
 426                                  true, info->queue, vp_notify, callback, name);
 427         if (!vq) {
 428                 err = -ENOMEM;
 429                 goto out_activate_queue;
 430         }
 431
 432         vq->priv = info;
 433         info->vq = vq;
 434
 435         if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 436                 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 437                 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 438                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 439                         err = -EBUSY;
 440                         goto out_assign;
 441                 }
 442         }
 443
 444         if (callback) {
 445                 spin_lock_irqsave(&vp_dev->lock, flags);
 446                 list_add(&info->node, &vp_dev->virtqueues);
 447                 spin_unlock_irqrestore(&vp_dev->lock, flags);
 448         } else {
 449                 INIT_LIST_HEAD(&info->node);
 450         }
 451
 452         return vq;
 453
 454 out_assign:
 455         vring_del_virtqueue(vq);
 456 out_activate_queue:
 457         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 458         free_pages_exact(info->queue, size);
 459 out_info:
 460         kfree(info);
 461         return ERR_PTR(err);
 462 }
 463
 464 static void vp_del_vq(struct virtqueue *vq)
 465 {
 466         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 467         struct virtio_pci_vq_info *info = vq->priv;
 468         unsigned long flags, size;
 469
 470         spin_lock_irqsave(&vp_dev->lock, flags);
 471         list_del(&info->node);
 472         spin_unlock_irqrestore(&vp_dev->lock, flags);
 473
 474         iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 475
 476         if (vp_dev->msix_enabled) {
 477                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 478                           vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 479                 /* Flush the write out to device */
 480                 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 481         }
 482
 483         vring_del_virtqueue(vq);
 484
 485         /* Select and deactivate the queue */
 486         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 487
 488         size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 489         free_pages_exact(info->queue, size);
 490         kfree(info);
 491 }
 492
 493 /* the config->del_vqs() implementation */
 494 static void vp_del_vqs(struct virtio_device *vdev)
 495 {
 496         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 497         struct virtqueue *vq, *n;
 498         struct virtio_pci_vq_info *info;
 499
 500         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 501                 info = vq->priv;
 502                 if (vp_dev->per_vq_vectors &&
 503                         info->msix_vector != VIRTIO_MSI_NO_VECTOR)
 504                         free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 505                                  vq);
 506                 vp_del_vq(vq);
 507         }
 508         vp_dev->per_vq_vectors = false;
 509
 510         vp_free_vectors(vdev);
 511 }
 512
 513 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 514                               struct virtqueue *vqs[],
 515                               vq_callback_t *callbacks[],
 516                               const char *names[],
 517                               bool use_msix,
 518                               bool per_vq_vectors)
 519 {
 520         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 521         u16 msix_vec;
 522         int i, err, nvectors, allocated_vectors;
 523
 524         if (!use_msix) {
 525                 /* Old style: one normal interrupt for change and all vqs. */
 526                 err = vp_request_intx(vdev);
 527                 if (err)
 528                         goto error_request;
 529         } else {
 530                 if (per_vq_vectors) {
 531                         /* Best option: one for change interrupt, one per vq. */
 532                         nvectors = 1;
 533                         for (i = 0; i < nvqs; ++i)
 534                                 if (callbacks[i])
 535                                         ++nvectors;
 536                 } else {
 537                         /* Second best: one for change, shared for all vqs. */
 538                         nvectors = 2;
 539                 }
 540
 541                 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 542                 if (err)
 543                         goto error_request;
 544         }
 545
 546         vp_dev->per_vq_vectors = per_vq_vectors;
 547         allocated_vectors = vp_dev->msix_used_vectors;
 548         for (i = 0; i < nvqs; ++i) {
 549                 if (!names[i]) {
 550                         vqs[i] = NULL;
 551                         continue;
 552                 } else if (!callbacks[i] || !vp_dev->msix_enabled)
 553                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 554                 else if (vp_dev->per_vq_vectors)
 555                         msix_vec = allocated_vectors++;
 556                 else
 557                         msix_vec = VP_MSIX_VQ_VECTOR;
 558                 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 559                 if (IS_ERR(vqs[i])) {
 560                         err = PTR_ERR(vqs[i]);
 561                         goto error_find;
 562                 }
 563
 564                 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 565                         continue;
 566
 567                 /* allocate per-vq irq if available and necessary */
 568                 snprintf(vp_dev->msix_names[msix_vec],
 569                          sizeof *vp_dev->msix_names,
 570                          "%s-%s",
 571                          dev_name(&vp_dev->vdev.dev), names[i]);
 572                 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 573                                   vring_interrupt, 0,
 574                                   vp_dev->msix_names[msix_vec],
 575                                   vqs[i]);
 576                 if (err) {
 577                         vp_del_vq(vqs[i]);
 578                         goto error_find;
 579                 }
 580         }
 581         return 0;
 582
 583 error_find:
 584         vp_del_vqs(vdev);
 585
 586 error_request:
 587         return err;
 588 }
 589
 590 /* the config->find_vqs() implementation */
 591 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 592                        struct virtqueue *vqs[],
 593                        vq_callback_t *callbacks[],
 594                        const char *names[])
 595 {
 596         int err;
 597
 598         /* Try MSI-X with one vector per queue. */
 599         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 600         if (!err)
 601                 return 0;
 602         /* Fallback: MSI-X with one vector for config, one shared for queues. */
 603         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 604                                  true, false);
 605         if (!err)
 606                 return 0;
 607         /* Finally fall back to regular interrupts. */
 608         return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 609                                   false, false);
 610 }
 611
 612 static const char *vp_bus_name(struct virtio_device *vdev)
 613 {
 614         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 615
 616         return pci_name(vp_dev->pci_dev);
 617 }
 618
 619 /* Setup the affinity for a virtqueue:
 620  * - force the affinity for per vq vector
 621  * - OR over all affinities for shared MSI
 622  * - ignore the affinity request if we're using INTX
 623  */
 624 static int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 625 {
 626         struct virtio_device *vdev = vq->vdev;
 627         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 628         struct virtio_pci_vq_info *info = vq->priv;
 629         struct cpumask *mask;
 630         unsigned int irq;
 631
 632         if (!vq->callback)
 633                 return -EINVAL;
 634
 635         if (vp_dev->msix_enabled) {
 636                 mask = vp_dev->msix_affinity_masks[info->msix_vector];
 637                 irq = vp_dev->msix_entries[info->msix_vector].vector;
 638                 if (cpu == -1)
 639                         irq_set_affinity_hint(irq, NULL);
 640                 else {
 641                         cpumask_set_cpu(cpu, mask);
 642                         irq_set_affinity_hint(irq, mask);
 643                 }
 644         }
 645         return 0;
 646 }
 647
 648 static const struct virtio_config_ops virtio_pci_config_ops = {
 649         .get            = vp_get,
 650         .set            = vp_set,
 651         .get_status     = vp_get_status,
 652         .set_status     = vp_set_status,
 653         .reset          = vp_reset,
 654         .find_vqs       = vp_find_vqs,
 655         .del_vqs        = vp_del_vqs,
 656         .get_features   = vp_get_features,
 657         .finalize_features = vp_finalize_features,
 658         .bus_name       = vp_bus_name,
 659         .set_vq_affinity = vp_set_vq_affinity,
 660 };
 661
 662 static void virtio_pci_release_dev(struct device *_d)
 663 {
 664         /*
 665          * No need for a release method as we allocate/free
 666          * all devices together with the pci devices.
 667          * Provide an empty one to avoid getting a warning from core.
 668          */
 669 }
 670
 671 /* the PCI probing function */
 672 static int virtio_pci_probe(struct pci_dev *pci_dev,
 673                             const struct pci_device_id *id)
 674 {
 675         struct virtio_pci_device *vp_dev;
 676         int err;
 677
 678         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 679         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 680                 return -ENODEV;
 681
 682         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 683                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 684                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 685                 return -ENODEV;
 686         }
 687
 688         /* allocate our structure and fill it out */
 689         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 690         if (vp_dev == NULL)
 691                 return -ENOMEM;
 692
 693         vp_dev->vdev.dev.parent = &pci_dev->dev;
 694         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 695         vp_dev->vdev.config = &virtio_pci_config_ops;
 696         vp_dev->pci_dev = pci_dev;
 697         INIT_LIST_HEAD(&vp_dev->virtqueues);
 698         spin_lock_init(&vp_dev->lock);
 699
 700         /* Disable MSI/MSIX to bring device to a known good state. */
 701         pci_msi_off(pci_dev);
 702
 703         /* enable the device */
 704         err = pci_enable_device(pci_dev);
 705         if (err)
 706                 goto out;
 707
 708         err = pci_request_regions(pci_dev, "virtio-pci");
 709         if (err)
 710                 goto out_enable_device;
 711
 712         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 713         if (vp_dev->ioaddr == NULL) {
 714                 err = -ENOMEM;
 715                 goto out_req_regions;
 716         }
 717
 718         pci_set_drvdata(pci_dev, vp_dev);
 719         pci_set_master(pci_dev);
 720
 721         /* we use the subsystem vendor/device id as the virtio vendor/device
 722          * id.  this allows us to use the same PCI vendor/device id for all
 723          * virtio devices and to identify the particular virtio driver by
 724          * the subsystem ids */
 725         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 726         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 727
 728         /* finally register the virtio device */
 729         err = register_virtio_device(&vp_dev->vdev);
 730         if (err)
 731                 goto out_set_drvdata;
 732
 733         return 0;
 734
 735 out_set_drvdata:
 736         pci_iounmap(pci_dev, vp_dev->ioaddr);
 737 out_req_regions:
 738         pci_release_regions(pci_dev);
 739 out_enable_device:
 740         pci_disable_device(pci_dev);
 741 out:
 742         kfree(vp_dev);
 743         return err;
 744 }
 745
 746 static void virtio_pci_remove(struct pci_dev *pci_dev)
 747 {
 748         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 749
 750         unregister_virtio_device(&vp_dev->vdev);
 751
 752         vp_del_vqs(&vp_dev->vdev);
 753         pci_iounmap(pci_dev, vp_dev->ioaddr);
 754         pci_release_regions(pci_dev);
 755         pci_disable_device(pci_dev);
 756         kfree(vp_dev);
 757 }
 758
 759 #ifdef CONFIG_PM_SLEEP
 760 static int virtio_pci_freeze(struct device *dev)
 761 {
 762         struct pci_dev *pci_dev = to_pci_dev(dev);
 763         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 764         int ret;
 765
 766         ret = virtio_device_freeze(&vp_dev->vdev);
 767
 768         if (!ret)
 769                 pci_disable_device(pci_dev);
 770         return ret;
 771 }
 772
 773 static int virtio_pci_restore(struct device *dev)
 774 {
 775         struct pci_dev *pci_dev = to_pci_dev(dev);
 776         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 777         int ret;
 778
 779         ret = pci_enable_device(pci_dev);
 780         if (ret)
 781                 return ret;
 782
 783         pci_set_master(pci_dev);
 784         return virtio_device_restore(&vp_dev->vdev);
 785 }
 786
 787 static const struct dev_pm_ops virtio_pci_pm_ops = {
 788         SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
 789 };
 790 #endif
 791
 792 static struct pci_driver virtio_pci_driver = {
 793         .name           = "virtio-pci",
 794         .id_table       = virtio_pci_id_table,
 795         .probe          = virtio_pci_probe,
 796         .remove         = virtio_pci_remove,
 797 #ifdef CONFIG_PM_SLEEP
 798         .driver.pm      = &virtio_pci_pm_ops,
 799 #endif
 800 };
 801
 802 module_pci_driver(virtio_pci_driver);