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