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 /* Unique numbering for devices under the kvm root */
  82 static unsigned int dev_index;
  83
  84 /* Convert a generic virtio device to our structure */
  85 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
  86 {
  87         return container_of(vdev, struct virtio_pci_device, vdev);
  88 }
  89
  90 /* virtio config->feature() implementation */
  91 static bool vp_feature(struct virtio_device *vdev, unsigned bit)
  92 {
  93         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
  94         u32 mask;
  95
  96         /* Since this function is supposed to have the side effect of
  97          * enabling a queried feature, we simulate that by doing a read
  98          * from the host feature bitmask and then writing to the guest
  99          * feature bitmask */
 100         mask = ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
 101         if (mask & (1 << bit)) {
 102                 mask |= (1 << bit);
 103                 iowrite32(mask, vp_dev->ioaddr + VIRTIO_PCI_GUEST_FEATURES);
 104         }
 105
 106         return !!(mask & (1 << bit));
 107 }
 108
 109 /* virtio config->get() implementation */
 110 static void vp_get(struct virtio_device *vdev, unsigned offset,
 111                    void *buf, unsigned len)
 112 {
 113         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 114         void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
 115         u8 *ptr = buf;
 116         int i;
 117
 118         for (i = 0; i < len; i++)
 119                 ptr[i] = ioread8(ioaddr + i);
 120 }
 121
 122 /* the config->set() implementation.  it's symmetric to the config->get()
 123  * implementation */
 124 static void vp_set(struct virtio_device *vdev, unsigned offset,
 125                    const void *buf, unsigned len)
 126 {
 127         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 128         void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
 129         const u8 *ptr = buf;
 130         int i;
 131
 132         for (i = 0; i < len; i++)
 133                 iowrite8(ptr[i], ioaddr + i);
 134 }
 135
 136 /* config->{get,set}_status() implementations */
 137 static u8 vp_get_status(struct virtio_device *vdev)
 138 {
 139         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 140         return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 141 }
 142
 143 static void vp_set_status(struct virtio_device *vdev, u8 status)
 144 {
 145         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 146         /* We should never be setting status to 0. */
 147         BUG_ON(status == 0);
 148         return iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 149 }
 150
 151 static void vp_reset(struct virtio_device *vdev)
 152 {
 153         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 154         /* 0 status means a reset. */
 155         return iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
 156 }
 157
 158 /* the notify function used when creating a virt queue */
 159 static void vp_notify(struct virtqueue *vq)
 160 {
 161         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 162         struct virtio_pci_vq_info *info = vq->priv;
 163
 164         /* we write the queue's selector into the notification register to
 165          * signal the other end */
 166         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
 167 }
 168
 169 /* A small wrapper to also acknowledge the interrupt when it's handled.
 170  * I really need an EIO hook for the vring so I can ack the interrupt once we
 171  * know that we'll be handling the IRQ but before we invoke the callback since
 172  * the callback may notify the host which results in the host attempting to
 173  * raise an interrupt that we would then mask once we acknowledged the
 174  * interrupt. */
 175 static irqreturn_t vp_interrupt(int irq, void *opaque)
 176 {
 177         struct virtio_pci_device *vp_dev = opaque;
 178         struct virtio_pci_vq_info *info;
 179         irqreturn_t ret = IRQ_NONE;
 180         unsigned long flags;
 181         u8 isr;
 182
 183         /* reading the ISR has the effect of also clearing it so it's very
 184          * important to save off the value. */
 185         isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
 186
 187         /* It's definitely not us if the ISR was not high */
 188         if (!isr)
 189                 return IRQ_NONE;
 190
 191         /* Configuration change?  Tell driver if it wants to know. */
 192         if (isr & VIRTIO_PCI_ISR_CONFIG) {
 193                 struct virtio_driver *drv;
 194                 drv = container_of(vp_dev->vdev.dev.driver,
 195                                    struct virtio_driver, driver);
 196
 197                 if (drv->config_changed)
 198                         drv->config_changed(&vp_dev->vdev);
 199         }
 200
 201         spin_lock_irqsave(&vp_dev->lock, flags);
 202         list_for_each_entry(info, &vp_dev->virtqueues, node) {
 203                 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
 204                         ret = IRQ_HANDLED;
 205         }
 206         spin_unlock_irqrestore(&vp_dev->lock, flags);
 207
 208         return ret;
 209 }
 210
 211 /* the config->find_vq() implementation */
 212 static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
 213                                     void (*callback)(struct virtqueue *vq))
 214 {
 215         struct virtio_pci_device *vp_dev = to_vp_device(vdev);
 216         struct virtio_pci_vq_info *info;
 217         struct virtqueue *vq;
 218         unsigned long flags;
 219         u16 num;
 220         int err;
 221
 222         /* Select the queue we're interested in */
 223         iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 224
 225         /* Check if queue is either not available or already active. */
 226         num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
 227         if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
 228                 return ERR_PTR(-ENOENT);
 229
 230         /* allocate and fill out our structure the represents an active
 231          * queue */
 232         info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
 233         if (!info)
 234                 return ERR_PTR(-ENOMEM);
 235
 236         info->queue_index = index;
 237         info->num = num;
 238
 239         info->queue = kzalloc(PAGE_ALIGN(vring_size(num,PAGE_SIZE)), GFP_KERNEL);
 240         if (info->queue == NULL) {
 241                 err = -ENOMEM;
 242                 goto out_info;
 243         }
 244
 245         /* activate the queue */
 246         iowrite32(virt_to_phys(info->queue) >> PAGE_SHIFT,
 247                   vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 248
 249         /* create the vring */
 250         vq = vring_new_virtqueue(info->num, vdev, info->queue,
 251                                  vp_notify, callback);
 252         if (!vq) {
 253                 err = -ENOMEM;
 254                 goto out_activate_queue;
 255         }
 256
 257         vq->priv = info;
 258         info->vq = vq;
 259
 260         spin_lock_irqsave(&vp_dev->lock, flags);
 261         list_add(&info->node, &vp_dev->virtqueues);
 262         spin_unlock_irqrestore(&vp_dev->lock, flags);
 263
 264         return vq;
 265
 266 out_activate_queue:
 267         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 268         kfree(info->queue);
 269 out_info:
 270         kfree(info);
 271         return ERR_PTR(err);
 272 }
 273
 274 /* the config->del_vq() implementation */
 275 static void vp_del_vq(struct virtqueue *vq)
 276 {
 277         struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
 278         struct virtio_pci_vq_info *info = vq->priv;
 279         unsigned long flags;
 280
 281         spin_lock_irqsave(&vp_dev->lock, flags);
 282         list_del(&info->node);
 283         spin_unlock_irqrestore(&vp_dev->lock, flags);
 284
 285         vring_del_virtqueue(vq);
 286
 287         /* Select and deactivate the queue */
 288         iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
 289         iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 290
 291         kfree(info->queue);
 292         kfree(info);
 293 }
 294
 295 static struct virtio_config_ops virtio_pci_config_ops = {
 296         .feature        = vp_feature,
 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 };
 305
 306 /* the PCI probing function */
 307 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
 308                                       const struct pci_device_id *id)
 309 {
 310         struct virtio_pci_device *vp_dev;
 311         int err;
 312
 313         /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
 314         if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
 315                 return -ENODEV;
 316
 317         if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
 318                 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
 319                        VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
 320                 return -ENODEV;
 321         }
 322
 323         /* allocate our structure and fill it out */
 324         vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
 325         if (vp_dev == NULL)
 326                 return -ENOMEM;
 327
 328         snprintf(vp_dev->vdev.dev.bus_id, BUS_ID_SIZE, "virtio%d", dev_index);
 329         vp_dev->vdev.index = dev_index;
 330         dev_index++;
 331
 332         vp_dev->vdev.dev.parent = &virtio_pci_root;
 333         vp_dev->vdev.config = &virtio_pci_config_ops;
 334         vp_dev->pci_dev = pci_dev;
 335         INIT_LIST_HEAD(&vp_dev->virtqueues);
 336         spin_lock_init(&vp_dev->lock);
 337
 338         /* enable the device */
 339         err = pci_enable_device(pci_dev);
 340         if (err)
 341                 goto out;
 342
 343         err = pci_request_regions(pci_dev, "virtio-pci");
 344         if (err)
 345                 goto out_enable_device;
 346
 347         vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
 348         if (vp_dev->ioaddr == NULL)
 349                 goto out_req_regions;
 350
 351         pci_set_drvdata(pci_dev, vp_dev);
 352
 353         /* we use the subsystem vendor/device id as the virtio vendor/device
 354          * id.  this allows us to use the same PCI vendor/device id for all
 355          * virtio devices and to identify the particular virtio driver by
 356          * the subsytem ids */
 357         vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
 358         vp_dev->vdev.id.device = pci_dev->subsystem_device;
 359
 360         /* register a handler for the queue with the PCI device's interrupt */
 361         err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
 362                           vp_dev->vdev.dev.bus_id, vp_dev);
 363         if (err)
 364                 goto out_set_drvdata;
 365
 366         /* finally register the virtio device */
 367         err = register_virtio_device(&vp_dev->vdev);
 368         if (err)
 369                 goto out_req_irq;
 370
 371         return 0;
 372
 373 out_req_irq:
 374         free_irq(pci_dev->irq, vp_dev);
 375 out_set_drvdata:
 376         pci_set_drvdata(pci_dev, NULL);
 377         pci_iounmap(pci_dev, vp_dev->ioaddr);
 378 out_req_regions:
 379         pci_release_regions(pci_dev);
 380 out_enable_device:
 381         pci_disable_device(pci_dev);
 382 out:
 383         kfree(vp_dev);
 384         return err;
 385 }
 386
 387 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
 388 {
 389         struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
 390
 391         unregister_virtio_device(&vp_dev->vdev);
 392         free_irq(pci_dev->irq, vp_dev);
 393         pci_set_drvdata(pci_dev, NULL);
 394         pci_iounmap(pci_dev, vp_dev->ioaddr);
 395         pci_release_regions(pci_dev);
 396         pci_disable_device(pci_dev);
 397         kfree(vp_dev);
 398 }
 399
 400 #ifdef CONFIG_PM
 401 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
 402 {
 403         pci_save_state(pci_dev);
 404         pci_set_power_state(pci_dev, PCI_D3hot);
 405         return 0;
 406 }
 407
 408 static int virtio_pci_resume(struct pci_dev *pci_dev)
 409 {
 410         pci_restore_state(pci_dev);
 411         pci_set_power_state(pci_dev, PCI_D0);
 412         return 0;
 413 }
 414 #endif
 415
 416 static struct pci_driver virtio_pci_driver = {
 417         .name           = "virtio-pci",
 418         .id_table       = virtio_pci_id_table,
 419         .probe          = virtio_pci_probe,
 420         .remove         = virtio_pci_remove,
 421 #ifdef CONFIG_PM
 422         .suspend        = virtio_pci_suspend,
 423         .resume         = virtio_pci_resume,
 424 #endif
 425 };
 426
 427 static int __init virtio_pci_init(void)
 428 {
 429         int err;
 430
 431         err = device_register(&virtio_pci_root);
 432         if (err)
 433                 return err;
 434
 435         err = pci_register_driver(&virtio_pci_driver);
 436         if (err)
 437                 device_unregister(&virtio_pci_root);
 438
 439         return err;
 440 }
 441
 442 module_init(virtio_pci_init);
 443
 444 static void __exit virtio_pci_exit(void)
 445 {
 446         device_unregister(&virtio_pci_root);
 447         pci_unregister_driver(&virtio_pci_driver);
 448 }
 449
 450 module_exit(virtio_pci_exit);