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         /* Status saved during hibernate/restore */
  61         u8 saved_status;
  62
  63         /* Whether we have vector per vq */
  64         bool per_vq_vectors;
  65 };
  66
  67 /* Constants for MSI-X */
  68 /* Use first vector for configuration changes, second and the rest for
  69  * virtqueues Thus, we need at least 2 vectors for MSI. */
  70 enum {
  71         VP_MSIX_CONFIG_VECTOR = 0,
  72         VP_MSIX_VQ_VECTOR = 1,
  73 };
  74
  75 struct virtio_pci_vq_info
  76 {
  77         /* the actual virtqueue */
  78         struct virtqueue *vq;
  79
  80         /* the number of entries in the queue */
  81         int num;
  82
  83         /* the virtual address of the ring queue */
  84         void *queue;
  85
  86         /* the list node for the virtqueues list */
  87         struct list_head node;
  88
  89         /* MSI-X vector (or none) */
  90         unsigned msix_vector;
  91 };
  92
  93 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
  94 static const struct pci_device_id virtio_pci_id_table[] = {
  95         { PCI_DEVICE(0x1af4, PCI_ANY_ID) },
  96         { 0 }
  97 };
  98
  99 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
 100
 101 /* Convert a generic virtio device to our structure */
 102 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
 103 {
 104         return container_of(vdev, struct virtio_pci_device, vdev);
 105 }
 106
 107 /* virtio config->get_features() implementation */
 108 static u32 vp_get_features(struct virtio_device *vdev)
 109 {
 110         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 111
 112         /* When someone needs more than 32 feature bits, we'll need to
 113          * steal a bit to indicate that the rest are somewhere else. */
 114         return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
 115 }
 116
 117 /* virtio config->finalize_features() implementation */
 118 static void vp_finalize_features(struct virtio_device *vdev)
 119 {
 120         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 121
 122         /* Give virtio_ring a chance to accept features. */
 123         vring_transport_features(vdev);
 124
 125         /* We only support 32 feature bits. */
 126         BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
 127         iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
 128 }
 129
 130 /* virtio config->get() implementation */
 131 static void vp_get(struct virtio_device *vdev, unsigned offset,
 132                    void *buf, unsigned len)
 133 {
 134         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 135         void __iomem *ioaddr = vp_dev->ioaddr +
 136                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 137         u8 *ptr = buf;
 138         int i;
 139
 140         for (i = 0; i < len; i++)
 141                 ptr[i] = ioread8(ioaddr + i);
 142 }
 143
 144 /* the config->set() implementation.  it's symmetric to the config->get()
 145  * implementation */
 146 static void vp_set(struct virtio_device *vdev, unsigned offset,
 147                    const void *buf, unsigned len)
 148 {
 149         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 150         void __iomem *ioaddr = vp_dev->ioaddr +
 151                                 VIRTIO_PCI_CONFIG(vp_dev) + offset;
 152         const u8 *ptr = buf;
 153         int i;
 154
 155         for (i = 0; i < len; i++)
 156                 iowrite8(ptr[i], ioaddr + i);
 157 }
 158
 159 /* config->{get,set}_status() implementations */
 160 static u8 vp_get_status(struct virtio_device *vdev)
 161 {
 162         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 163         return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 164 }
 165
 166 static void vp_set_status(struct virtio_device *vdev, u8 status)
 167 {
 168         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 169         /* We should never be setting status to 0. */
 170         BUG_ON(status == 0);
 171         iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 172 }
 173
 174 /* wait for pending irq handlers */
 175 static void vp_synchronize_vectors(struct virtio_device *vdev)
 176 {
 177         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 178         int i;
 179
 180         if (vp_dev->intx_enabled)
 181                 synchronize_irq(vp_dev->pci_dev->irq);
 182
 183         for (i = 0; i < vp_dev->msix_vectors; ++i)
 184                 synchronize_irq(vp_dev->msix_entries[i].vector);
 185 }
 186
 187 static void vp_reset(struct virtio_device *vdev)
 188 {
 189         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 190         /* 0 status means a reset. */
 191         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 192         /* Flush out the status write, and flush in device writes,
 193          * including MSi-X interrupts, if any. */
 194         ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 195         /* Flush pending VQ/configuration callbacks. */
 196         vp_synchronize_vectors(vdev);
 197 }
 198
 199 /* the notify function used when creating a virt queue */
 200 static bool vp_notify(struct virtqueue *vq)
 201 {
 202         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 203
 204         /* we write the queue's selector into the notification register to
 205          * signal the other end */
 206         iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 207         return true;
 208 }
 209
 210 /* Handle a configuration change: Tell driver if it wants to know. */
 211 static irqreturn_t vp_config_changed(int irq, void *opaque)
 212 {
 213         struct virtio_pci_device *vp_dev = opaque;
 214
 215         virtio_config_changed(&vp_dev->vdev);
 216         return IRQ_HANDLED;
 217 }
 218
 219 /* Notify all virtqueues on an interrupt. */
 220 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
 221 {
 222         struct virtio_pci_device *vp_dev = opaque;
 223         struct virtio_pci_vq_info *info;
 224         irqreturn_t ret = IRQ_NONE;
 225         unsigned long flags;
 226
 227         spin_lock_irqsave(&vp_dev->lock, flags);
 228         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 229                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 230                         ret = IRQ_HANDLED;
 231         }
 232         spin_unlock_irqrestore(&vp_dev->lock, flags);
 233
 234         return ret;
 235 }
 236
 237 /* A small wrapper to also acknowledge the interrupt when it's handled.
 238  * I really need an EIO hook for the vring so I can ack the interrupt once we
 239  * know that we'll be handling the IRQ but before we invoke the callback since
 240  * the callback may notify the host which results in the host attempting to
 241  * raise an interrupt that we would then mask once we acknowledged the
 242  * interrupt. */
 243 static irqreturn_t vp_interrupt(int irq, void *opaque)
 244 {
 245         struct virtio_pci_device *vp_dev = opaque;
 246         u8 isr;
 247
 248         /* reading the ISR has the effect of also clearing it so it's very
 249          * important to save off the value. */
 250         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 251
 252         /* It's definitely not us if the ISR was not high */
 253         if (!isr)
 254                 return IRQ_NONE;
 255
 256         /* Configuration change?  Tell driver if it wants to know. */
 257         if (isr & VIRTIO_PCI_ISR_CONFIG)
 258                 vp_config_changed(irq, opaque);
 259
 260         return vp_vring_interrupt(irq, opaque);
 261 }
 262
 263 static void vp_free_vectors(struct virtio_device *vdev)
 264 {
 265         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 266         int i;
 267
 268         if (vp_dev->intx_enabled) {
 269                 free_irq(vp_dev->pci_dev->irq, vp_dev);
 270                 vp_dev->intx_enabled = 0;
 271         }
 272
 273         for (i = 0; i < vp_dev->msix_used_vectors; ++i)
 274                 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
 275
 276         for (i = 0; i < vp_dev->msix_vectors; i++)
 277                 if (vp_dev->msix_affinity_masks[i])
 278                         free_cpumask_var(vp_dev->msix_affinity_masks[i]);
 279
 280         if (vp_dev->msix_enabled) {
 281                 /* Disable the vector used for configuration */
 282                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 283                           vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 284                 /* Flush the write out to device */
 285                 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 286
 287                 pci_disable_msix(vp_dev->pci_dev);
 288                 vp_dev->msix_enabled = 0;
 289         }
 290
 291         vp_dev->msix_vectors = 0;
 292         vp_dev->msix_used_vectors = 0;
 293         kfree(vp_dev->msix_names);
 294         vp_dev->msix_names = NULL;
 295         kfree(vp_dev->msix_entries);
 296         vp_dev->msix_entries = NULL;
 297         kfree(vp_dev->msix_affinity_masks);
 298         vp_dev->msix_affinity_masks = 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_vectors = nvectors;
 310
 311         vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 312                                        GFP_KERNEL);
 313         if (!vp_dev->msix_entries)
 314                 goto error;
 315         vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 316                                      GFP_KERNEL);
 317         if (!vp_dev->msix_names)
 318                 goto error;
 319         vp_dev->msix_affinity_masks
 320                 = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
 321                           GFP_KERNEL);
 322         if (!vp_dev->msix_affinity_masks)
 323                 goto error;
 324         for (i = 0; i < nvectors; ++i)
 325                 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
 326                                         GFP_KERNEL))
 327                         goto error;
 328
 329         for (i = 0; i < nvectors; ++i)
 330                 vp_dev->msix_entries[i].entry = i;
 331
 332         err = pci_enable_msix_exact(vp_dev->pci_dev,
 333                                     vp_dev->msix_entries, nvectors);
 334         if (err)
 335                 goto error;
 336         vp_dev->msix_enabled = 1;
 337
 338         /* Set the vector used for configuration */
 339         v = vp_dev->msix_used_vectors;
 340         snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 341                  "%s-config", name);
 342         err = request_irq(vp_dev->msix_entries[v].vector,
 343                           vp_config_changed, 0, vp_dev->msix_names[v],
 344                           vp_dev);
 345         if (err)
 346                 goto error;
 347         ++vp_dev->msix_used_vectors;
 348
 349         iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 350         /* Verify we had enough resources to assign the vector */
 351         v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 352         if (v == VIRTIO_MSI_NO_VECTOR) {
 353                 err = -EBUSY;
 354                 goto error;
 355         }
 356
 357         if (!per_vq_vectors) {
 358                 /* Shared vector for all VQs */
 359                 v = vp_dev->msix_used_vectors;
 360                 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 361                          "%s-virtqueues", name);
 362                 err = request_irq(vp_dev->msix_entries[v].vector,
 363                                   vp_vring_interrupt, 0, vp_dev->msix_names[v],
 364                                   vp_dev);
 365                 if (err)
 366                         goto error;
 367                 ++vp_dev->msix_used_vectors;
 368         }
 369         return 0;
 370 error:
 371         vp_free_vectors(vdev);
 372         return err;
 373 }
 374
 375 static int vp_request_intx(struct virtio_device *vdev)
 376 {
 377         int err;
 378         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 379
 380         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 381                           IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 382         if (!err)
 383                 vp_dev->intx_enabled = 1;
 384         return err;
 385 }
 386
 387 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 388                                   void (*callback)(struct virtqueue *vq),
 389                                   const char *name,
 390                                   u16 msix_vec)
 391 {
 392         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 393         struct virtio_pci_vq_info *info;
 394         struct virtqueue *vq;
 395         unsigned long flags, size;
 396         u16 num;
 397         int err;
 398
 399         /* Select the queue we're interested in */
 400         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 401
 402         /* Check if queue is either not available or already active. */
 403         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 404         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 405                 return ERR_PTR(-ENOENT);
 406
 407         /* allocate and fill out our structure the represents an active
 408          * queue */
 409         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 410         if (!info)
 411                 return ERR_PTR(-ENOMEM);
 412
 413         info->num = num;
 414         info->msix_vector = msix_vec;
 415
 416         size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 417         info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 418         if (info->queue == NULL) {
 419                 err = -ENOMEM;
 420                 goto out_info;
 421         }
 422
 423         /* activate the queue */
 424         iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 425                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 426
 427         /* create the vring */
 428         vq = vring_new_virtqueue(index, info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
 429                                  true, info->queue, vp_notify, callback, name);
 430         if (!vq) {
 431                 err = -ENOMEM;
 432                 goto out_activate_queue;
 433         }
 434
 435         vq->priv = info;
 436         info->vq = vq;
 437
 438         if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 439                 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 440                 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 441                 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 442                         err = -EBUSY;
 443                         goto out_assign;
 444                 }
 445         }
 446
 447         if (callback) {
 448                 spin_lock_irqsave(&vp_dev->lock, flags);
 449                 list_add(&info->node, &vp_dev->virtqueues);
 450                 spin_unlock_irqrestore(&vp_dev->lock, flags);
 451         } else {
 452                 INIT_LIST_HEAD(&info->node);
 453         }
 454
 455         return vq;
 456
 457 out_assign:
 458         vring_del_virtqueue(vq);
 459 out_activate_queue:
 460         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 461         free_pages_exact(info->queue, size);
 462 out_info:
 463         kfree(info);
 464         return ERR_PTR(err);
 465 }
 466
 467 static void vp_del_vq(struct virtqueue *vq)
 468 {
 469         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 470         struct virtio_pci_vq_info *info = vq->priv;
 471         unsigned long flags, size;
 472
 473         spin_lock_irqsave(&vp_dev->lock, flags);
 474         list_del(&info->node);
 475         spin_unlock_irqrestore(&vp_dev->lock, flags);
 476
 477         iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 478
 479         if (vp_dev->msix_enabled) {
 480                 iowrite16(VIRTIO_MSI_NO_VECTOR,
 481                           vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 482                 /* Flush the write out to device */
 483                 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 484         }
 485
 486         vring_del_virtqueue(vq);
 487
 488         /* Select and deactivate the queue */
 489         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 490
 491         size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 492         free_pages_exact(info->queue, size);
 493         kfree(info);
 494 }
 495
 496 /* the config->del_vqs() implementation */
 497 static void vp_del_vqs(struct virtio_device *vdev)
 498 {
 499         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 500         struct virtqueue *vq, *n;
 501         struct virtio_pci_vq_info *info;
 502
 503         list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 504                 info = vq->priv;
 505                 if (vp_dev->per_vq_vectors &&
 506                         info->msix_vector != VIRTIO_MSI_NO_VECTOR)
 507                         free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 508                                  vq);
 509                 vp_del_vq(vq);
 510         }
 511         vp_dev->per_vq_vectors = false;
 512
 513         vp_free_vectors(vdev);
 514 }
 515
 516 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 517                               struct virtqueue *vqs[],
 518                               vq_callback_t *callbacks[],
 519                               const char *names[],
 520                               bool use_msix,
 521                               bool per_vq_vectors)
 522 {
 523         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 524         u16 msix_vec;
 525         int i, err, nvectors, allocated_vectors;
 526
 527         if (!use_msix) {
 528                 /* Old style: one normal interrupt for change and all vqs. */
 529                 err = vp_request_intx(vdev);
 530                 if (err)
 531                         goto error_request;
 532         } else {
 533                 if (per_vq_vectors) {
 534                         /* Best option: one for change interrupt, one per vq. */
 535                         nvectors = 1;
 536                         for (i = 0; i < nvqs; ++i)
 537                                 if (callbacks[i])
 538                                         ++nvectors;
 539                 } else {
 540                         /* Second best: one for change, shared for all vqs. */
 541                         nvectors = 2;
 542                 }
 543
 544                 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 545                 if (err)
 546                         goto error_request;
 547         }
 548
 549         vp_dev->per_vq_vectors = per_vq_vectors;
 550         allocated_vectors = vp_dev->msix_used_vectors;
 551         for (i = 0; i < nvqs; ++i) {
 552                 if (!names[i]) {
 553                         vqs[i] = NULL;
 554                         continue;
 555                 } else if (!callbacks[i] || !vp_dev->msix_enabled)
 556                         msix_vec = VIRTIO_MSI_NO_VECTOR;
 557                 else if (vp_dev->per_vq_vectors)
 558                         msix_vec = allocated_vectors++;
 559                 else
 560                         msix_vec = VP_MSIX_VQ_VECTOR;
 561                 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 562                 if (IS_ERR(vqs[i])) {
 563                         err = PTR_ERR(vqs[i]);
 564                         goto error_find;
 565                 }
 566
 567                 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 568                         continue;
 569
 570                 /* allocate per-vq irq if available and necessary */
 571                 snprintf(vp_dev->msix_names[msix_vec],
 572                          sizeof *vp_dev->msix_names,
 573                          "%s-%s",
 574                          dev_name(&vp_dev->vdev.dev), names[i]);
 575                 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 576                                   vring_interrupt, 0,
 577                                   vp_dev->msix_names[msix_vec],
 578                                   vqs[i]);
 579                 if (err) {
 580                         vp_del_vq(vqs[i]);
 581                         goto error_find;
 582                 }
 583         }
 584         return 0;
 585
 586 error_find:
 587         vp_del_vqs(vdev);
 588
 589 error_request:
 590         return err;
 591 }
 592
 593 /* the config->find_vqs() implementation */
 594 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 595                        struct virtqueue *vqs[],
 596                        vq_callback_t *callbacks[],
 597                        const char *names[])
 598 {
 599         int err;
 600
 601         /* Try MSI-X with one vector per queue. */
 602         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 603         if (!err)
 604                 return 0;
 605         /* Fallback: MSI-X with one vector for config, one shared for queues. */
 606         err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 607                                  true, false);
 608         if (!err)
 609                 return 0;
 610         /* Finally fall back to regular interrupts. */
 611         return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 612                                   false, false);
 613 }
 614
 615 static const char *vp_bus_name(struct virtio_device *vdev)
 616 {
 617         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 618
 619         return pci_name(vp_dev->pci_dev);
 620 }
 621
 622 /* Setup the affinity for a virtqueue:
 623  * - force the affinity for per vq vector
 624  * - OR over all affinities for shared MSI
 625  * - ignore the affinity request if we're using INTX
 626  */
 627 static int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 628 {
 629         struct virtio_device *vdev = vq->vdev;
 630         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 631         struct virtio_pci_vq_info *info = vq->priv;
 632         struct cpumask *mask;
 633         unsigned int irq;
 634
 635         if (!vq->callback)
 636                 return -EINVAL;
 637
 638         if (vp_dev->msix_enabled) {
 639                 mask = vp_dev->msix_affinity_masks[info->msix_vector];
 640                 irq = vp_dev->msix_entries[info->msix_vector].vector;
 641                 if (cpu == -1)
 642                         irq_set_affinity_hint(irq, NULL);
 643                 else {
 644                         cpumask_set_cpu(cpu, mask);
 645                         irq_set_affinity_hint(irq, mask);
 646                 }
 647         }
 648         return 0;
 649 }
 650
 651 static const struct virtio_config_ops virtio_pci_config_ops = {
 652         .get            = vp_get,
 653         .set            = vp_set,
 654         .get_status     = vp_get_status,
 655         .set_status     = vp_set_status,
 656         .reset          = vp_reset,
 657         .find_vqs       = vp_find_vqs,
 658         .del_vqs        = vp_del_vqs,
 659         .get_features   = vp_get_features,
 660         .finalize_features = vp_finalize_features,
 661         .bus_name       = vp_bus_name,
 662         .set_vq_affinity = vp_set_vq_affinity,
 663 };
 664
 665 static void virtio_pci_release_dev(struct device *_d)
 666 {
 667         /*
 668          * No need for a release method as we allocate/free
 669          * all devices together with the pci devices.
 670          * Provide an empty one to avoid getting a warning from core.
 671          */
 672 }
 673
 674 /* the PCI probing function */
 675 static int virtio_pci_probe(struct pci_dev *pci_dev,
 676                             const struct pci_device_id *id)
 677 {
 678         struct virtio_pci_device *vp_dev;
 679         int err;
 680
 681         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 682         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 683                 return -ENODEV;
 684
 685         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 686                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 687                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 688                 return -ENODEV;
 689         }
 690
 691         /* allocate our structure and fill it out */
 692         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 693         if (vp_dev == NULL)
 694                 return -ENOMEM;
 695
 696         vp_dev->vdev.dev.parent = &pci_dev->dev;
 697         vp_dev->vdev.dev.release = virtio_pci_release_dev;
 698         vp_dev->vdev.config = &virtio_pci_config_ops;
 699         vp_dev->pci_dev = pci_dev;
 700         INIT_LIST_HEAD(&vp_dev->virtqueues);
 701         spin_lock_init(&vp_dev->lock);
 702
 703         /* Disable MSI/MSIX to bring device to a known good state. */
 704         pci_msi_off(pci_dev);
 705
 706         /* enable the device */
 707         err = pci_enable_device(pci_dev);
 708         if (err)
 709                 goto out;
 710
 711         err = pci_request_regions(pci_dev, "virtio-pci");
 712         if (err)
 713                 goto out_enable_device;
 714
 715         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 716         if (vp_dev->ioaddr == NULL) {
 717                 err = -ENOMEM;
 718                 goto out_req_regions;
 719         }
 720
 721         pci_set_drvdata(pci_dev, vp_dev);
 722         pci_set_master(pci_dev);
 723
 724         /* we use the subsystem vendor/device id as the virtio vendor/device
 725          * id.  this allows us to use the same PCI vendor/device id for all
 726          * virtio devices and to identify the particular virtio driver by
 727          * the subsystem ids */
 728         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 729         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 730
 731         /* finally register the virtio device */
 732         err = register_virtio_device(&vp_dev->vdev);
 733         if (err)
 734                 goto out_set_drvdata;
 735
 736         return 0;
 737
 738 out_set_drvdata:
 739         pci_iounmap(pci_dev, vp_dev->ioaddr);
 740 out_req_regions:
 741         pci_release_regions(pci_dev);
 742 out_enable_device:
 743         pci_disable_device(pci_dev);
 744 out:
 745         kfree(vp_dev);
 746         return err;
 747 }
 748
 749 static void virtio_pci_remove(struct pci_dev *pci_dev)
 750 {
 751         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 752
 753         unregister_virtio_device(&vp_dev->vdev);
 754
 755         vp_del_vqs(&vp_dev->vdev);
 756         pci_iounmap(pci_dev, vp_dev->ioaddr);
 757         pci_release_regions(pci_dev);
 758         pci_disable_device(pci_dev);
 759         kfree(vp_dev);
 760 }
 761
 762 #ifdef CONFIG_PM_SLEEP
 763 static int virtio_pci_freeze(struct device *dev)
 764 {
 765         struct pci_dev *pci_dev = to_pci_dev(dev);
 766         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 767         struct virtio_driver *drv;
 768         int ret;
 769
 770         drv = container_of(vp_dev->vdev.dev.driver,
 771                            struct virtio_driver, driver);
 772
 773         ret = 0;
 774         vp_dev->saved_status = vp_get_status(&vp_dev->vdev);
 775         if (drv && drv->freeze)
 776                 ret = drv->freeze(&vp_dev->vdev);
 777
 778         if (!ret)
 779                 pci_disable_device(pci_dev);
 780         return ret;
 781 }
 782
 783 static int virtio_pci_restore(struct device *dev)
 784 {
 785         struct pci_dev *pci_dev = to_pci_dev(dev);
 786         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 787         struct virtio_driver *drv;
 788         unsigned status = 0;
 789         int ret;
 790
 791         drv = container_of(vp_dev->vdev.dev.driver,
 792                            struct virtio_driver, driver);
 793
 794         ret = pci_enable_device(pci_dev);
 795         if (ret)
 796                 return ret;
 797
 798         pci_set_master(pci_dev);
 799         /* We always start by resetting the device, in case a previous
 800          * driver messed it up. */
 801         vp_reset(&vp_dev->vdev);
 802
 803         /* Acknowledge that we've seen the device. */
 804         status |= VIRTIO_CONFIG_S_ACKNOWLEDGE;
 805         vp_set_status(&vp_dev->vdev, status);
 806
 807         /* Maybe driver failed before freeze.
 808          * Restore the failed status, for debugging. */
 809         status |= vp_dev->saved_status & VIRTIO_CONFIG_S_FAILED;
 810         vp_set_status(&vp_dev->vdev, status);
 811
 812         if (!drv)
 813                 return 0;
 814
 815         /* We have a driver! */
 816         status |= VIRTIO_CONFIG_S_DRIVER;
 817         vp_set_status(&vp_dev->vdev, status);
 818
 819         vp_finalize_features(&vp_dev->vdev);
 820
 821         if (drv->restore) {
 822                 ret = drv->restore(&vp_dev->vdev);
 823                 if (ret) {
 824                         status |= VIRTIO_CONFIG_S_FAILED;
 825                         vp_set_status(&vp_dev->vdev, status);
 826                         return ret;
 827                 }
 828         }
 829
 830         /* Finally, tell the device we're all set */
 831         status |= VIRTIO_CONFIG_S_DRIVER_OK;
 832         vp_set_status(&vp_dev->vdev, status);
 833
 834         return ret;
 835 }
 836
 837 static const struct dev_pm_ops virtio_pci_pm_ops = {
 838         SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
 839 };
 840 #endif
 841
 842 static struct pci_driver virtio_pci_driver = {
 843         .name           = "virtio-pci",
 844         .id_table       = virtio_pci_id_table,
 845         .probe          = virtio_pci_probe,
 846         .remove         = virtio_pci_remove,
 847 #ifdef CONFIG_PM_SLEEP
 848         .driver.pm      = &virtio_pci_pm_ops,
 849 #endif
 850 };
 851
 852 module_pci_driver(virtio_pci_driver);