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