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 /* Convert a generic virtio device to our structure */
79 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
81 return container_of(vdev, struct virtio_pci_device, vdev);
84 /* virtio config->get_features() implementation */
85 static u32 vp_get_features(struct virtio_device *vdev)
87 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
89 /* When someone needs more than 32 feature bits, we'll need to
90 * steal a bit to indicate that the rest are somewhere else. */
91 return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
94 /* virtio config->finalize_features() implementation */
95 static void vp_finalize_features(struct virtio_device *vdev)
97 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
99 /* Give virtio_ring a chance to accept features. */
100 vring_transport_features(vdev);
102 /* We only support 32 feature bits. */
103 BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
104 iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
107 /* virtio config->get() implementation */
108 static void vp_get(struct virtio_device *vdev, unsigned offset,
109 void *buf, unsigned len)
111 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
112 void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
116 for (i = 0; i < len; i++)
117 ptr[i] = ioread8(ioaddr + i);
120 /* the config->set() implementation. it's symmetric to the config->get()
122 static void vp_set(struct virtio_device *vdev, unsigned offset,
123 const void *buf, unsigned len)
125 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
126 void __iomem *ioaddr = vp_dev->ioaddr + VIRTIO_PCI_CONFIG + offset;
130 for (i = 0; i < len; i++)
131 iowrite8(ptr[i], ioaddr + i);
134 /* config->{get,set}_status() implementations */
135 static u8 vp_get_status(struct virtio_device *vdev)
137 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
138 return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
141 static void vp_set_status(struct virtio_device *vdev, u8 status)
143 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
144 /* We should never be setting status to 0. */
146 iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
149 static void vp_reset(struct virtio_device *vdev)
151 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
152 /* 0 status means a reset. */
153 iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
156 /* the notify function used when creating a virt queue */
157 static void vp_notify(struct virtqueue *vq)
159 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
160 struct virtio_pci_vq_info *info = vq->priv;
162 /* we write the queue's selector into the notification register to
163 * signal the other end */
164 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
167 /* A small wrapper to also acknowledge the interrupt when it's handled.
168 * I really need an EIO hook for the vring so I can ack the interrupt once we
169 * know that we'll be handling the IRQ but before we invoke the callback since
170 * the callback may notify the host which results in the host attempting to
171 * raise an interrupt that we would then mask once we acknowledged the
173 static irqreturn_t vp_interrupt(int irq, void *opaque)
175 struct virtio_pci_device *vp_dev = opaque;
176 struct virtio_pci_vq_info *info;
177 irqreturn_t ret = IRQ_NONE;
181 /* reading the ISR has the effect of also clearing it so it's very
182 * important to save off the value. */
183 isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
185 /* It's definitely not us if the ISR was not high */
189 /* Configuration change? Tell driver if it wants to know. */
190 if (isr & VIRTIO_PCI_ISR_CONFIG) {
191 struct virtio_driver *drv;
192 drv = container_of(vp_dev->vdev.dev.driver,
193 struct virtio_driver, driver);
195 if (drv && drv->config_changed)
196 drv->config_changed(&vp_dev->vdev);
199 spin_lock_irqsave(&vp_dev->lock, flags);
200 list_for_each_entry(info, &vp_dev->virtqueues, node) {
201 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
204 spin_unlock_irqrestore(&vp_dev->lock, flags);
209 /* the config->find_vq() implementation */
210 static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
211 void (*callback)(struct virtqueue *vq))
213 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
214 struct virtio_pci_vq_info *info;
215 struct virtqueue *vq;
216 unsigned long flags, size;
220 /* Select the queue we're interested in */
221 iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
223 /* Check if queue is either not available or already active. */
224 num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
225 if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
226 return ERR_PTR(-ENOENT);
228 /* allocate and fill out our structure the represents an active
230 info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
232 return ERR_PTR(-ENOMEM);
234 info->queue_index = index;
237 size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
238 info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
239 if (info->queue == NULL) {
244 /* activate the queue */
245 iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
246 vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
248 /* create the vring */
249 vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
250 vdev, info->queue, 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);
267 free_pages_exact(info->queue, size);
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;
278 unsigned long flags, size;
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);
290 size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
291 free_pages_exact(info->queue, size);
295 static struct virtio_config_ops virtio_pci_config_ops = {
298 .get_status = vp_get_status,
299 .set_status = vp_set_status,
301 .find_vq = vp_find_vq,
303 .get_features = vp_get_features,
304 .finalize_features = vp_finalize_features,
307 static void virtio_pci_release_dev(struct device *_d)
309 struct virtio_device *dev = container_of(_d, struct virtio_device, dev);
310 struct virtio_pci_device *vp_dev = to_vp_device(dev);
311 struct pci_dev *pci_dev = vp_dev->pci_dev;
313 free_irq(pci_dev->irq, vp_dev);
314 pci_set_drvdata(pci_dev, NULL);
315 pci_iounmap(pci_dev, vp_dev->ioaddr);
316 pci_release_regions(pci_dev);
317 pci_disable_device(pci_dev);
321 /* the PCI probing function */
322 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
323 const struct pci_device_id *id)
325 struct virtio_pci_device *vp_dev;
328 /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
329 if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
332 if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
333 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
334 VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
338 /* allocate our structure and fill it out */
339 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
343 vp_dev->vdev.dev.parent = virtio_pci_root;
344 vp_dev->vdev.dev.release = virtio_pci_release_dev;
345 vp_dev->vdev.config = &virtio_pci_config_ops;
346 vp_dev->pci_dev = pci_dev;
347 INIT_LIST_HEAD(&vp_dev->virtqueues);
348 spin_lock_init(&vp_dev->lock);
350 /* enable the device */
351 err = pci_enable_device(pci_dev);
355 err = pci_request_regions(pci_dev, "virtio-pci");
357 goto out_enable_device;
359 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
360 if (vp_dev->ioaddr == NULL)
361 goto out_req_regions;
363 pci_set_drvdata(pci_dev, vp_dev);
365 /* we use the subsystem vendor/device id as the virtio vendor/device
366 * id. this allows us to use the same PCI vendor/device id for all
367 * virtio devices and to identify the particular virtio driver by
368 * the subsytem ids */
369 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
370 vp_dev->vdev.id.device = pci_dev->subsystem_device;
372 /* register a handler for the queue with the PCI device's interrupt */
373 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
374 dev_name(&vp_dev->vdev.dev), vp_dev);
376 goto out_set_drvdata;
378 /* finally register the virtio device */
379 err = register_virtio_device(&vp_dev->vdev);
386 free_irq(pci_dev->irq, vp_dev);
388 pci_set_drvdata(pci_dev, NULL);
389 pci_iounmap(pci_dev, vp_dev->ioaddr);
391 pci_release_regions(pci_dev);
393 pci_disable_device(pci_dev);
399 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
401 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
403 unregister_virtio_device(&vp_dev->vdev);
407 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
409 pci_save_state(pci_dev);
410 pci_set_power_state(pci_dev, PCI_D3hot);
414 static int virtio_pci_resume(struct pci_dev *pci_dev)
416 pci_restore_state(pci_dev);
417 pci_set_power_state(pci_dev, PCI_D0);
422 static struct pci_driver virtio_pci_driver = {
423 .name = "virtio-pci",
424 .id_table = virtio_pci_id_table,
425 .probe = virtio_pci_probe,
426 .remove = virtio_pci_remove,
428 .suspend = virtio_pci_suspend,
429 .resume = virtio_pci_resume,
433 static int __init virtio_pci_init(void)
437 virtio_pci_root = root_device_register("virtio-pci");
438 if (IS_ERR(virtio_pci_root))
439 return PTR_ERR(virtio_pci_root);
441 err = pci_register_driver(&virtio_pci_driver);
443 device_unregister(virtio_pci_root);
448 module_init(virtio_pci_init);
450 static void __exit virtio_pci_exit(void)
452 pci_unregister_driver(&virtio_pci_driver);
453 root_device_unregister(virtio_pci_root);
456 module_exit(virtio_pci_exit);