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