linux/drivers/virtio/virtio_pci.c
<<
>>
Prefs
   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
  29MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
  30MODULE_DESCRIPTION("virtio-pci");
  31MODULE_LICENSE("GPL");
  32MODULE_VERSION("1");
  33
  34/* Our device structure */
  35struct 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. */
  70enum {
  71        VP_MSIX_CONFIG_VECTOR = 0,
  72        VP_MSIX_VQ_VECTOR = 1,
  73};
  74
  75struct 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. */
  94static DEFINE_PCI_DEVICE_TABLE(virtio_pci_id_table) = {
  95        { PCI_DEVICE(0x1af4, PCI_ANY_ID) },
  96        { 0 }
  97};
  98
  99MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
 100
 101/* Convert a generic virtio device to our structure */
 102static 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 */
 108static 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 */
 118static 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 */
 131static 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 */
 146static 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 */
 160static 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
 166static 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 */
 175static 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
 187static 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 */
 200static 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. */
 210static 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. */
 223static 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. */
 246static 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
 266static 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                vp_dev->msix_vectors = 0;
 293        }
 294
 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
 304static 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_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
 313                                       GFP_KERNEL);
 314        if (!vp_dev->msix_entries)
 315                goto error;
 316        vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
 317                                     GFP_KERNEL);
 318        if (!vp_dev->msix_names)
 319                goto error;
 320        vp_dev->msix_affinity_masks
 321                = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
 322                          GFP_KERNEL);
 323        if (!vp_dev->msix_affinity_masks)
 324                goto error;
 325        for (i = 0; i < nvectors; ++i)
 326                if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
 327                                        GFP_KERNEL))
 328                        goto error;
 329
 330        for (i = 0; i < nvectors; ++i)
 331                vp_dev->msix_entries[i].entry = i;
 332
 333        /* pci_enable_msix returns positive if we can't get this many. */
 334        err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
 335        if (err > 0)
 336                err = -ENOSPC;
 337        if (err)
 338                goto error;
 339        vp_dev->msix_vectors = nvectors;
 340        vp_dev->msix_enabled = 1;
 341
 342        /* Set the vector used for configuration */
 343        v = vp_dev->msix_used_vectors;
 344        snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 345                 "%s-config", name);
 346        err = request_irq(vp_dev->msix_entries[v].vector,
 347                          vp_config_changed, 0, vp_dev->msix_names[v],
 348                          vp_dev);
 349        if (err)
 350                goto error;
 351        ++vp_dev->msix_used_vectors;
 352
 353        iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 354        /* Verify we had enough resources to assign the vector */
 355        v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
 356        if (v == VIRTIO_MSI_NO_VECTOR) {
 357                err = -EBUSY;
 358                goto error;
 359        }
 360
 361        if (!per_vq_vectors) {
 362                /* Shared vector for all VQs */
 363                v = vp_dev->msix_used_vectors;
 364                snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
 365                         "%s-virtqueues", name);
 366                err = request_irq(vp_dev->msix_entries[v].vector,
 367                                  vp_vring_interrupt, 0, vp_dev->msix_names[v],
 368                                  vp_dev);
 369                if (err)
 370                        goto error;
 371                ++vp_dev->msix_used_vectors;
 372        }
 373        return 0;
 374error:
 375        vp_free_vectors(vdev);
 376        return err;
 377}
 378
 379static int vp_request_intx(struct virtio_device *vdev)
 380{
 381        int err;
 382        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 383
 384        err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
 385                          IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
 386        if (!err)
 387                vp_dev->intx_enabled = 1;
 388        return err;
 389}
 390
 391static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
 392                                  void (*callback)(struct virtqueue *vq),
 393                                  const char *name,
 394                                  u16 msix_vec)
 395{
 396        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 397        struct virtio_pci_vq_info *info;
 398        struct virtqueue *vq;
 399        unsigned long flags, size;
 400        u16 num;
 401        int err;
 402
 403        /* Select the queue we're interested in */
 404        iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 405
 406        /* Check if queue is either not available or already active. */
 407        num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 408        if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 409                return ERR_PTR(-ENOENT);
 410
 411        /* allocate and fill out our structure the represents an active
 412         * queue */
 413        info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 414        if (!info)
 415                return ERR_PTR(-ENOMEM);
 416
 417        info->num = num;
 418        info->msix_vector = msix_vec;
 419
 420        size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
 421        info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
 422        if (info->queue == NULL) {
 423                err = -ENOMEM;
 424                goto out_info;
 425        }
 426
 427        /* activate the queue */
 428        iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
 429                  vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 430
 431        /* create the vring */
 432        vq = vring_new_virtqueue(index, info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
 433                                 true, info->queue, vp_notify, callback, name);
 434        if (!vq) {
 435                err = -ENOMEM;
 436                goto out_activate_queue;
 437        }
 438
 439        vq->priv = info;
 440        info->vq = vq;
 441
 442        if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
 443                iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 444                msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 445                if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
 446                        err = -EBUSY;
 447                        goto out_assign;
 448                }
 449        }
 450
 451        if (callback) {
 452                spin_lock_irqsave(&vp_dev->lock, flags);
 453                list_add(&info->node, &vp_dev->virtqueues);
 454                spin_unlock_irqrestore(&vp_dev->lock, flags);
 455        } else {
 456                INIT_LIST_HEAD(&info->node);
 457        }
 458
 459        return vq;
 460
 461out_assign:
 462        vring_del_virtqueue(vq);
 463out_activate_queue:
 464        iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 465        free_pages_exact(info->queue, size);
 466out_info:
 467        kfree(info);
 468        return ERR_PTR(err);
 469}
 470
 471static void vp_del_vq(struct virtqueue *vq)
 472{
 473        struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 474        struct virtio_pci_vq_info *info = vq->priv;
 475        unsigned long flags, size;
 476
 477        spin_lock_irqsave(&vp_dev->lock, flags);
 478        list_del(&info->node);
 479        spin_unlock_irqrestore(&vp_dev->lock, flags);
 480
 481        iowrite16(vq->index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 482
 483        if (vp_dev->msix_enabled) {
 484                iowrite16(VIRTIO_MSI_NO_VECTOR,
 485                          vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
 486                /* Flush the write out to device */
 487                ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 488        }
 489
 490        vring_del_virtqueue(vq);
 491
 492        /* Select and deactivate the queue */
 493        iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 494
 495        size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
 496        free_pages_exact(info->queue, size);
 497        kfree(info);
 498}
 499
 500/* the config->del_vqs() implementation */
 501static void vp_del_vqs(struct virtio_device *vdev)
 502{
 503        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 504        struct virtqueue *vq, *n;
 505        struct virtio_pci_vq_info *info;
 506
 507        list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
 508                info = vq->priv;
 509                if (vp_dev->per_vq_vectors &&
 510                        info->msix_vector != VIRTIO_MSI_NO_VECTOR)
 511                        free_irq(vp_dev->msix_entries[info->msix_vector].vector,
 512                                 vq);
 513                vp_del_vq(vq);
 514        }
 515        vp_dev->per_vq_vectors = false;
 516
 517        vp_free_vectors(vdev);
 518}
 519
 520static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 521                              struct virtqueue *vqs[],
 522                              vq_callback_t *callbacks[],
 523                              const char *names[],
 524                              bool use_msix,
 525                              bool per_vq_vectors)
 526{
 527        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 528        u16 msix_vec;
 529        int i, err, nvectors, allocated_vectors;
 530
 531        if (!use_msix) {
 532                /* Old style: one normal interrupt for change and all vqs. */
 533                err = vp_request_intx(vdev);
 534                if (err)
 535                        goto error_request;
 536        } else {
 537                if (per_vq_vectors) {
 538                        /* Best option: one for change interrupt, one per vq. */
 539                        nvectors = 1;
 540                        for (i = 0; i < nvqs; ++i)
 541                                if (callbacks[i])
 542                                        ++nvectors;
 543                } else {
 544                        /* Second best: one for change, shared for all vqs. */
 545                        nvectors = 2;
 546                }
 547
 548                err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
 549                if (err)
 550                        goto error_request;
 551        }
 552
 553        vp_dev->per_vq_vectors = per_vq_vectors;
 554        allocated_vectors = vp_dev->msix_used_vectors;
 555        for (i = 0; i < nvqs; ++i) {
 556                if (!names[i]) {
 557                        vqs[i] = NULL;
 558                        continue;
 559                } else if (!callbacks[i] || !vp_dev->msix_enabled)
 560                        msix_vec = VIRTIO_MSI_NO_VECTOR;
 561                else if (vp_dev->per_vq_vectors)
 562                        msix_vec = allocated_vectors++;
 563                else
 564                        msix_vec = VP_MSIX_VQ_VECTOR;
 565                vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
 566                if (IS_ERR(vqs[i])) {
 567                        err = PTR_ERR(vqs[i]);
 568                        goto error_find;
 569                }
 570
 571                if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
 572                        continue;
 573
 574                /* allocate per-vq irq if available and necessary */
 575                snprintf(vp_dev->msix_names[msix_vec],
 576                         sizeof *vp_dev->msix_names,
 577                         "%s-%s",
 578                         dev_name(&vp_dev->vdev.dev), names[i]);
 579                err = request_irq(vp_dev->msix_entries[msix_vec].vector,
 580                                  vring_interrupt, 0,
 581                                  vp_dev->msix_names[msix_vec],
 582                                  vqs[i]);
 583                if (err) {
 584                        vp_del_vq(vqs[i]);
 585                        goto error_find;
 586                }
 587        }
 588        return 0;
 589
 590error_find:
 591        vp_del_vqs(vdev);
 592
 593error_request:
 594        return err;
 595}
 596
 597/* the config->find_vqs() implementation */
 598static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
 599                       struct virtqueue *vqs[],
 600                       vq_callback_t *callbacks[],
 601                       const char *names[])
 602{
 603        int err;
 604
 605        /* Try MSI-X with one vector per queue. */
 606        err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
 607        if (!err)
 608                return 0;
 609        /* Fallback: MSI-X with one vector for config, one shared for queues. */
 610        err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 611                                 true, false);
 612        if (!err)
 613                return 0;
 614        /* Finally fall back to regular interrupts. */
 615        return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
 616                                  false, false);
 617}
 618
 619static const char *vp_bus_name(struct virtio_device *vdev)
 620{
 621        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 622
 623        return pci_name(vp_dev->pci_dev);
 624}
 625
 626/* Setup the affinity for a virtqueue:
 627 * - force the affinity for per vq vector
 628 * - OR over all affinities for shared MSI
 629 * - ignore the affinity request if we're using INTX
 630 */
 631static int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
 632{
 633        struct virtio_device *vdev = vq->vdev;
 634        struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 635        struct virtio_pci_vq_info *info = vq->priv;
 636        struct cpumask *mask;
 637        unsigned int irq;
 638
 639        if (!vq->callback)
 640                return -EINVAL;
 641
 642        if (vp_dev->msix_enabled) {
 643                mask = vp_dev->msix_affinity_masks[info->msix_vector];
 644                irq = vp_dev->msix_entries[info->msix_vector].vector;
 645                if (cpu == -1)
 646                        irq_set_affinity_hint(irq, NULL);
 647                else {
 648                        cpumask_set_cpu(cpu, mask);
 649                        irq_set_affinity_hint(irq, mask);
 650                }
 651        }
 652        return 0;
 653}
 654
 655static const struct virtio_config_ops virtio_pci_config_ops = {
 656        .get            = vp_get,
 657        .set            = vp_set,
 658        .get_status     = vp_get_status,
 659        .set_status     = vp_set_status,
 660        .reset          = vp_reset,
 661        .find_vqs       = vp_find_vqs,
 662        .del_vqs        = vp_del_vqs,
 663        .get_features   = vp_get_features,
 664        .finalize_features = vp_finalize_features,
 665        .bus_name       = vp_bus_name,
 666        .set_vq_affinity = vp_set_vq_affinity,
 667};
 668
 669static void virtio_pci_release_dev(struct device *_d)
 670{
 671        /*
 672         * No need for a release method as we allocate/free
 673         * all devices together with the pci devices.
 674         * Provide an empty one to avoid getting a warning from core.
 675         */
 676}
 677
 678/* the PCI probing function */
 679static int virtio_pci_probe(struct pci_dev *pci_dev,
 680                            const struct pci_device_id *id)
 681{
 682        struct virtio_pci_device *vp_dev;
 683        int err;
 684
 685        /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 686        if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 687                return -ENODEV;
 688
 689        if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 690                printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 691                       VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 692                return -ENODEV;
 693        }
 694
 695        /* allocate our structure and fill it out */
 696        vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 697        if (vp_dev == NULL)
 698                return -ENOMEM;
 699
 700        vp_dev->vdev.dev.parent = &pci_dev->dev;
 701        vp_dev->vdev.dev.release = virtio_pci_release_dev;
 702        vp_dev->vdev.config = &virtio_pci_config_ops;
 703        vp_dev->pci_dev = pci_dev;
 704        INIT_LIST_HEAD(&vp_dev->virtqueues);
 705        spin_lock_init(&vp_dev->lock);
 706
 707        /* Disable MSI/MSIX to bring device to a known good state. */
 708        pci_msi_off(pci_dev);
 709
 710        /* enable the device */
 711        err = pci_enable_device(pci_dev);
 712        if (err)
 713                goto out;
 714
 715        err = pci_request_regions(pci_dev, "virtio-pci");
 716        if (err)
 717                goto out_enable_device;
 718
 719        vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 720        if (vp_dev->ioaddr == NULL) {
 721                err = -ENOMEM;
 722                goto out_req_regions;
 723        }
 724
 725        pci_set_drvdata(pci_dev, vp_dev);
 726        pci_set_master(pci_dev);
 727
 728        /* we use the subsystem vendor/device id as the virtio vendor/device
 729         * id.  this allows us to use the same PCI vendor/device id for all
 730         * virtio devices and to identify the particular virtio driver by
 731         * the subsystem ids */
 732        vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 733        vp_dev->vdev.id.device = pci_dev->subsystem_device;
 734
 735        /* finally register the virtio device */
 736        err = register_virtio_device(&vp_dev->vdev);
 737        if (err)
 738                goto out_set_drvdata;
 739
 740        return 0;
 741
 742out_set_drvdata:
 743        pci_set_drvdata(pci_dev, NULL);
 744        pci_iounmap(pci_dev, vp_dev->ioaddr);
 745out_req_regions:
 746        pci_release_regions(pci_dev);
 747out_enable_device:
 748        pci_disable_device(pci_dev);
 749out:
 750        kfree(vp_dev);
 751        return err;
 752}
 753
 754static void virtio_pci_remove(struct pci_dev *pci_dev)
 755{
 756        struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 757
 758        unregister_virtio_device(&vp_dev->vdev);
 759
 760        vp_del_vqs(&vp_dev->vdev);
 761        pci_set_drvdata(pci_dev, NULL);
 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
 769static 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
 789static 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
 816static 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
 821static 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
 827        .driver.pm      = &virtio_pci_pm_ops,
 828#endif
 829};
 830
 831module_pci_driver(virtio_pci_driver);
 832
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.