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/interrupt.h>
  21 #include <linux/virtio.h>
  22 #include <linux/virtio_config.h>
  23 #include <linux/virtio_ring.h>
  24 #include <linux/virtio_pci.h>
  25 #include <linux/highmem.h>
  26 #include <linux/spinlock.h>
  27
  28 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
  29 MODULE_DESCRIPTION("virtio-pci");
  30 MODULE_LICENSE("GPL");
  31 MODULE_VERSION("1");
  32
  33 /* Our device structure */
  34 struct virtio_pci_device
  35 {
  36         struct virtio_device vdev;
  37         struct pci_dev *pci_dev;
  38
  39         /* the IO mapping for the PCI config space */
  40         void __iomem *ioaddr;
  41
  42         /* a list of queues so we can dispatch IRQs */
  43         spinlock_t lock;
  44         struct list_head virtqueues;
  45 };
  46
  47 struct virtio_pci_vq_info
  48 {
  49         /* the actual virtqueue */
  50         struct virtqueue *vq;
  51
  52         /* the number of entries in the queue */
  53         int num;
  54
  55         /* the index of the queue */
  56         int queue_index;
  57
  58         /* the virtual address of the ring queue */
  59         void *queue;
  60
  61         /* the list node for the virtqueues list */
  62         struct list_head node;
  63 };
  64
  65 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
  66 static struct pci_device_id virtio_pci_id_table[] = {
  67         { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
  68         { 0 },
  69 };
  70
  71 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
  72
  73 /* A PCI device has it's own struct device and so does a virtio device so
  74  * we create a place for the virtio devices to show up in sysfs.  I think it
  75  * would make more sense for virtio to not insist on having it's own device. */
  76 static struct device virtio_pci_root = {
  77         .parent         = NULL,
  78         .bus_id         = "virtio-pci",
  79 };
  80
  81 /* Convert a generic virtio device to our structure */
  82 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
  83 {
  84         return container_of(vdev, struct virtio_pci_device, vdev);
  85 }
  86
  87 /* virtio config->get_features() implementation */
  88 static u32 vp_get_features(struct virtio_device *vdev)
  89 {
  90         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
  91
  92         /* When someone needs more than 32 feature bits, we'll need to
  93          * steal a bit to indicate that the rest are somewhere else. */
  94         return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
  95 }
  96
  97 /* virtio config->finalize_features() implementation */
  98 static void vp_finalize_features(struct virtio_device *vdev)
  99 {
 100         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 101
 102         /* Give virtio_ring a chance to accept features. */
 103         vring_transport_features(vdev);
 104
 105         /* We only support 32 feature bits. */
 106         BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
 107         iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
 108 }
 109
 110 /* virtio config->get() implementation */
 111 static void vp_get(struct virtio_device *vdev, unsigned offset,
 112                    void *buf, unsigned len)
 113 {
 114         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 115         void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
 116         u8 *ptr = buf;
 117         int i;
 118
 119         for (i = 0; i < len; i++)
 120                 ptr[i] = ioread8(ioaddr + i);
 121 }
 122
 123 /* the config->set() implementation.  it's symmetric to the config->get()
 124  * implementation */
 125 static void vp_set(struct virtio_device *vdev, unsigned offset,
 126                    const void *buf, unsigned len)
 127 {
 128         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 129         void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
 130         const u8 *ptr = buf;
 131         int i;
 132
 133         for (i = 0; i < len; i++)
 134                 iowrite8(ptr[i], ioaddr + i);
 135 }
 136
 137 /* config->{get,set}_status() implementations */
 138 static u8 vp_get_status(struct virtio_device *vdev)
 139 {
 140         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 141         return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 142 }
 143
 144 static void vp_set_status(struct virtio_device *vdev, u8 status)
 145 {
 146         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 147         /* We should never be setting status to 0. */
 148         BUG_ON(status == 0);
 149         iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 150 }
 151
 152 static void vp_reset(struct virtio_device *vdev)
 153 {
 154         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 155         /* 0 status means a reset. */
 156         iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 157 }
 158
 159 /* the notify function used when creating a virt queue */
 160 static void vp_notify(struct virtqueue *vq)
 161 {
 162         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 163         struct virtio_pci_vq_info *info = vq->priv;
 164
 165         /* we write the queue's selector into the notification register to
 166          * signal the other end */
 167         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 168 }
 169
 170 /* A small wrapper to also acknowledge the interrupt when it's handled.
 171  * I really need an EIO hook for the vring so I can ack the interrupt once we
 172  * know that we'll be handling the IRQ but before we invoke the callback since
 173  * the callback may notify the host which results in the host attempting to
 174  * raise an interrupt that we would then mask once we acknowledged the
 175  * interrupt. */
 176 static irqreturn_t vp_interrupt(int irq, void *opaque)
 177 {
 178         struct virtio_pci_device *vp_dev = opaque;
 179         struct virtio_pci_vq_info *info;
 180         irqreturn_t ret = IRQ_NONE;
 181         unsigned long flags;
 182         u8 isr;
 183
 184         /* reading the ISR has the effect of also clearing it so it's very
 185          * important to save off the value. */
 186         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 187
 188         /* It's definitely not us if the ISR was not high */
 189         if (!isr)
 190                 return IRQ_NONE;
 191
 192         /* Configuration change?  Tell driver if it wants to know. */
 193         if (isr & VIRTIO_PCI_ISR_CONFIG) {
 194                 struct virtio_driver *drv;
 195                 drv = container_of(vp_dev->vdev.dev.driver,
 196                                    struct virtio_driver, driver);
 197
 198                 if (drv->config_changed)
 199                         drv->config_changed(&vp_dev->vdev);
 200         }
 201
 202         spin_lock_irqsave(&vp_dev->lock, flags);
 203         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 204                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 205                         ret = IRQ_HANDLED;
 206         }
 207         spin_unlock_irqrestore(&vp_dev->lock, flags);
 208
 209         return ret;
 210 }
 211
 212 /* the config->find_vq() implementation */
 213 static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
 214                                     void (*callback)(struct virtqueue *vq))
 215 {
 216         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 217         struct virtio_pci_vq_info *info;
 218         struct virtqueue *vq;
 219         unsigned long flags;
 220         u16 num;
 221         int err;
 222
 223         /* Select the queue we're interested in */
 224         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 225
 226         /* Check if queue is either not available or already active. */
 227         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 228         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 229                 return ERR_PTR(-ENOENT);
 230
 231         /* allocate and fill out our structure the represents an active
 232          * queue */
 233         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 234         if (!info)
 235                 return ERR_PTR(-ENOMEM);
 236
 237         info->queue_index = index;
 238         info->num = num;
 239
 240         info->queue = kzalloc(PAGE_ALIGN(vring_size(num,PAGE_SIZE)), GFP_KERNEL);
 241         if (info->queue == NULL) {
 242                 err = -ENOMEM;
 243                 goto out_info;
 244         }
 245
 246         /* activate the queue */
 247         iowrite32(virt_to_phys(info->queue) >> PAGE_SHIFT,
 248                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 249
 250         /* create the vring */
 251         vq = vring_new_virtqueue(info->num, vdev, info->queue,
 252                                  vp_notify, callback);
 253         if (!vq) {
 254                 err = -ENOMEM;
 255                 goto out_activate_queue;
 256         }
 257
 258         vq->priv = info;
 259         info->vq = vq;
 260
 261         spin_lock_irqsave(&vp_dev->lock, flags);
 262         list_add(&info->node, &vp_dev->virtqueues);
 263         spin_unlock_irqrestore(&vp_dev->lock, flags);
 264
 265         return vq;
 266
 267 out_activate_queue:
 268         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 269         kfree(info->queue);
 270 out_info:
 271         kfree(info);
 272         return ERR_PTR(err);
 273 }
 274
 275 /* the config->del_vq() implementation */
 276 static void vp_del_vq(struct virtqueue *vq)
 277 {
 278         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 279         struct virtio_pci_vq_info *info = vq->priv;
 280         unsigned long flags;
 281
 282         spin_lock_irqsave(&vp_dev->lock, flags);
 283         list_del(&info->node);
 284         spin_unlock_irqrestore(&vp_dev->lock, flags);
 285
 286         vring_del_virtqueue(vq);
 287
 288         /* Select and deactivate the queue */
 289         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 290         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 291
 292         kfree(info->queue);
 293         kfree(info);
 294 }
 295
 296 static struct virtio_config_ops virtio_pci_config_ops = {
 297         .get            = vp_get,
 298         .set            = vp_set,
 299         .get_status     = vp_get_status,
 300         .set_status     = vp_set_status,
 301         .reset          = vp_reset,
 302         .find_vq        = vp_find_vq,
 303         .del_vq         = vp_del_vq,
 304         .get_features   = vp_get_features,
 305         .finalize_features = vp_finalize_features,
 306 };
 307
 308 /* the PCI probing function */
 309 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
 310                                       const struct pci_device_id *id)
 311 {
 312         struct virtio_pci_device *vp_dev;
 313         int err;
 314
 315         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 316         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 317                 return -ENODEV;
 318
 319         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 320                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 321                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 322                 return -ENODEV;
 323         }
 324
 325         /* allocate our structure and fill it out */
 326         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 327         if (vp_dev == NULL)
 328                 return -ENOMEM;
 329
 330         vp_dev->vdev.dev.parent = &virtio_pci_root;
 331         vp_dev->vdev.config = &virtio_pci_config_ops;
 332         vp_dev->pci_dev = pci_dev;
 333         INIT_LIST_HEAD(&vp_dev->virtqueues);
 334         spin_lock_init(&vp_dev->lock);
 335
 336         /* enable the device */
 337         err = pci_enable_device(pci_dev);
 338         if (err)
 339                 goto out;
 340
 341         err = pci_request_regions(pci_dev, "virtio-pci");
 342         if (err)
 343                 goto out_enable_device;
 344
 345         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 346         if (vp_dev->ioaddr == NULL)
 347                 goto out_req_regions;
 348
 349         pci_set_drvdata(pci_dev, vp_dev);
 350
 351         /* we use the subsystem vendor/device id as the virtio vendor/device
 352          * id.  this allows us to use the same PCI vendor/device id for all
 353          * virtio devices and to identify the particular virtio driver by
 354          * the subsytem ids */
 355         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 356         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 357
 358         /* register a handler for the queue with the PCI device's interrupt */
 359         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
 360                           vp_dev->vdev.dev.bus_id, vp_dev);
 361         if (err)
 362                 goto out_set_drvdata;
 363
 364         /* finally register the virtio device */
 365         err = register_virtio_device(&vp_dev->vdev);
 366         if (err)
 367                 goto out_req_irq;
 368
 369         return 0;
 370
 371 out_req_irq:
 372         free_irq(pci_dev->irq, vp_dev);
 373 out_set_drvdata:
 374         pci_set_drvdata(pci_dev, NULL);
 375         pci_iounmap(pci_dev, vp_dev->ioaddr);
 376 out_req_regions:
 377         pci_release_regions(pci_dev);
 378 out_enable_device:
 379         pci_disable_device(pci_dev);
 380 out:
 381         kfree(vp_dev);
 382         return err;
 383 }
 384
 385 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
 386 {
 387         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 388
 389         unregister_virtio_device(&vp_dev->vdev);
 390         free_irq(pci_dev->irq, vp_dev);
 391         pci_set_drvdata(pci_dev, NULL);
 392         pci_iounmap(pci_dev, vp_dev->ioaddr);
 393         pci_release_regions(pci_dev);
 394         pci_disable_device(pci_dev);
 395         kfree(vp_dev);
 396 }
 397
 398 #ifdef CONFIG_PM
 399 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
 400 {
 401         pci_save_state(pci_dev);
 402         pci_set_power_state(pci_dev, PCI_D3hot);
 403         return 0;
 404 }
 405
 406 static int virtio_pci_resume(struct pci_dev *pci_dev)
 407 {
 408         pci_restore_state(pci_dev);
 409         pci_set_power_state(pci_dev, PCI_D0);
 410         return 0;
 411 }
 412 #endif
 413
 414 static struct pci_driver virtio_pci_driver = {
 415         .name           = "virtio-pci",
 416         .id_table       = virtio_pci_id_table,
 417         .probe          = virtio_pci_probe,
 418         .remove         = virtio_pci_remove,
 419 #ifdef CONFIG_PM
 420         .suspend        = virtio_pci_suspend,
 421         .resume         = virtio_pci_resume,
 422 #endif
 423 };
 424
 425 static int __init virtio_pci_init(void)
 426 {
 427         int err;
 428
 429         err = device_register(&virtio_pci_root);
 430         if (err)
 431                 return err;
 432
 433         err = pci_register_driver(&virtio_pci_driver);
 434         if (err)
 435                 device_unregister(&virtio_pci_root);
 436
 437         return err;
 438 }
 439
 440 module_init(virtio_pci_init);
 441
 442 static void __exit virtio_pci_exit(void)
 443 {
 444         device_unregister(&virtio_pci_root);
 445         pci_unregister_driver(&virtio_pci_driver);
 446 }
 447
 448 module_exit(virtio_pci_exit);