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.
7 * Copyright IBM Corp. 2007
10 * Anthony Liguori <aliguori@us.ibm.com>
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.
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>
28 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
29 MODULE_DESCRIPTION("virtio-pci");
30 MODULE_LICENSE("GPL");
33 /* Our device structure */
34 struct virtio_pci_device
36 struct virtio_device vdev;
37 struct pci_dev *pci_dev;
39 /* the IO mapping for the PCI config space */
42 /* a list of queues so we can dispatch IRQs */
44 struct list_head virtqueues;
47 struct virtio_pci_vq_info
49 /* the actual virtqueue */
52 /* the number of entries in the queue */
55 /* the index of the queue */
58 /* the virtual address of the ring queue */
61 /* the list node for the virtqueues list */
62 struct list_head node;
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 },
71 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
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 = {
78 .bus_id = "virtio-pci",
81 /* Unique numbering for devices under the kvm root */
82 static unsigned int dev_index;
84 /* Convert a generic virtio device to our structure */
85 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
87 return container_of(vdev, struct virtio_pci_device, vdev);
90 /* virtio config->get_features() implementation */
91 static u32 vp_get_features(struct virtio_device *vdev)
93 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
95 /* When someone needs more than 32 feature bits, we'll need to
96 * steal a bit to indicate that the rest are somewhere else. */
97 return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
100 /* virtio config->set_features() implementation */
101 static void vp_set_features(struct virtio_device *vdev, u32 features)
103 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
105 iowrite32(features, vp_dev->ioaddr + VIRTIO_PCI_GUEST_FEATURES);
108 /* virtio config->get() implementation */
109 static void vp_get(struct virtio_device *vdev, unsigned offset,
110 void *buf, unsigned len)
112 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
113 void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
117 for (i = 0; i < len; i++)
118 ptr[i] = ioread8(ioaddr + i);
121 /* the config->set() implementation. it's symmetric to the config->get()
123 static void vp_set(struct virtio_device *vdev, unsigned offset,
124 const void *buf, unsigned len)
126 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
127 void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
131 for (i = 0; i < len; i++)
132 iowrite8(ptr[i], ioaddr + i);
135 /* config->{get,set}_status() implementations */
136 static u8 vp_get_status(struct virtio_device *vdev)
138 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
139 return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
142 static void vp_set_status(struct virtio_device *vdev, u8 status)
144 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
145 /* We should never be setting status to 0. */
147 iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
150 static void vp_reset(struct virtio_device *vdev)
152 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
153 /* 0 status means a reset. */
154 iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
157 /* the notify function used when creating a virt queue */
158 static void vp_notify(struct virtqueue *vq)
160 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
161 struct virtio_pci_vq_info *info = vq->priv;
163 /* we write the queue's selector into the notification register to
164 * signal the other end */
165 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
168 /* A small wrapper to also acknowledge the interrupt when it's handled.
169 * I really need an EIO hook for the vring so I can ack the interrupt once we
170 * know that we'll be handling the IRQ but before we invoke the callback since
171 * the callback may notify the host which results in the host attempting to
172 * raise an interrupt that we would then mask once we acknowledged the
174 static irqreturn_t vp_interrupt(int irq, void *opaque)
176 struct virtio_pci_device *vp_dev = opaque;
177 struct virtio_pci_vq_info *info;
178 irqreturn_t ret = IRQ_NONE;
182 /* reading the ISR has the effect of also clearing it so it's very
183 * important to save off the value. */
184 isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
186 /* It's definitely not us if the ISR was not high */
190 /* Configuration change? Tell driver if it wants to know. */
191 if (isr & VIRTIO_PCI_ISR_CONFIG) {
192 struct virtio_driver *drv;
193 drv = container_of(vp_dev->vdev.dev.driver,
194 struct virtio_driver, driver);
196 if (drv->config_changed)
197 drv->config_changed(&vp_dev->vdev);
200 spin_lock_irqsave(&vp_dev->lock, flags);
201 list_for_each_entry(info, &vp_dev->virtqueues, node) {
202 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
205 spin_unlock_irqrestore(&vp_dev->lock, flags);
210 /* the config->find_vq() implementation */
211 static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
212 void (*callback)(struct virtqueue *vq))
214 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
215 struct virtio_pci_vq_info *info;
216 struct virtqueue *vq;
221 /* Select the queue we're interested in */
222 iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
224 /* Check if queue is either not available or already active. */
225 num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
226 if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
227 return ERR_PTR(-ENOENT);
229 /* allocate and fill out our structure the represents an active
231 info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
233 return ERR_PTR(-ENOMEM);
235 info->queue_index = index;
238 info->queue = kzalloc(PAGE_ALIGN(vring_size(num,PAGE_SIZE)), GFP_KERNEL);
239 if (info->queue == NULL) {
244 /* activate the queue */
245 iowrite32(virt_to_phys(info->queue) >> PAGE_SHIFT,
246 vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
248 /* create the vring */
249 vq = vring_new_virtqueue(info->num, vdev, info->queue,
250 vp_notify, callback);
253 goto out_activate_queue;
259 spin_lock_irqsave(&vp_dev->lock, flags);
260 list_add(&info->node, &vp_dev->virtqueues);
261 spin_unlock_irqrestore(&vp_dev->lock, flags);
266 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
273 /* the config->del_vq() implementation */
274 static void vp_del_vq(struct virtqueue *vq)
276 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
277 struct virtio_pci_vq_info *info = vq->priv;
280 spin_lock_irqsave(&vp_dev->lock, flags);
281 list_del(&info->node);
282 spin_unlock_irqrestore(&vp_dev->lock, flags);
284 vring_del_virtqueue(vq);
286 /* Select and deactivate the queue */
287 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
288 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
294 static struct virtio_config_ops virtio_pci_config_ops = {
297 .get_status = vp_get_status,
298 .set_status = vp_set_status,
300 .find_vq = vp_find_vq,
302 .get_features = vp_get_features,
303 .set_features = vp_set_features,
306 /* the PCI probing function */
307 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
308 const struct pci_device_id *id)
310 struct virtio_pci_device *vp_dev;
313 /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
314 if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
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);
323 /* allocate our structure and fill it out */
324 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
328 snprintf(vp_dev->vdev.dev.bus_id, BUS_ID_SIZE, "virtio%d", dev_index);
329 vp_dev->vdev.index = dev_index;
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);
338 /* enable the device */
339 err = pci_enable_device(pci_dev);
343 err = pci_request_regions(pci_dev, "virtio-pci");
345 goto out_enable_device;
347 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
348 if (vp_dev->ioaddr == NULL)
349 goto out_req_regions;
351 pci_set_drvdata(pci_dev, vp_dev);
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;
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);
364 goto out_set_drvdata;
366 /* finally register the virtio device */
367 err = register_virtio_device(&vp_dev->vdev);
374 free_irq(pci_dev->irq, vp_dev);
376 pci_set_drvdata(pci_dev, NULL);
377 pci_iounmap(pci_dev, vp_dev->ioaddr);
379 pci_release_regions(pci_dev);
381 pci_disable_device(pci_dev);
387 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
389 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
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);
401 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
403 pci_save_state(pci_dev);
404 pci_set_power_state(pci_dev, PCI_D3hot);
408 static int virtio_pci_resume(struct pci_dev *pci_dev)
410 pci_restore_state(pci_dev);
411 pci_set_power_state(pci_dev, PCI_D0);
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,
422 .suspend = virtio_pci_suspend,
423 .resume = virtio_pci_resume,
427 static int __init virtio_pci_init(void)
431 err = device_register(&virtio_pci_root);
435 err = pci_register_driver(&virtio_pci_driver);
437 device_unregister(&virtio_pci_root);
442 module_init(virtio_pci_init);
444 static void __exit virtio_pci_exit(void)
446 device_unregister(&virtio_pci_root);
447 pci_unregister_driver(&virtio_pci_driver);
450 module_exit(virtio_pci_exit);