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;
49 struct msix_entry *msix_entries;
50 /* Name strings for interrupts. This size should be enough,
51 * and I'm too lazy to allocate each name separately. */
52 char (*msix_names)[256];
53 /* Number of available vectors */
54 unsigned msix_vectors;
55 /* Vectors allocated */
56 unsigned msix_used_vectors;
59 /* Constants for MSI-X */
60 /* Use first vector for configuration changes, second and the rest for
61 * virtqueues Thus, we need at least 2 vectors for MSI. */
63 VP_MSIX_CONFIG_VECTOR = 0,
64 VP_MSIX_VQ_VECTOR = 1,
67 struct virtio_pci_vq_info
69 /* the actual virtqueue */
72 /* the number of entries in the queue */
75 /* the index of the queue */
78 /* the virtual address of the ring queue */
81 /* the list node for the virtqueues list */
82 struct list_head node;
84 /* MSI-X vector (or none) */
88 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
89 static struct pci_device_id virtio_pci_id_table[] = {
90 { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
94 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
96 /* A PCI device has it's own struct device and so does a virtio device so
97 * we create a place for the virtio devices to show up in sysfs. I think it
98 * would make more sense for virtio to not insist on having it's own device. */
99 static struct device *virtio_pci_root;
101 /* Convert a generic virtio device to our structure */
102 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
104 return container_of(vdev, struct virtio_pci_device, vdev);
107 /* virtio config->get_features() implementation */
108 static u32 vp_get_features(struct virtio_device *vdev)
110 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
112 /* When someone needs more than 32 feature bits, we'll need to
113 * steal a bit to indicate that the rest are somewhere else. */
114 return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
117 /* virtio config->finalize_features() implementation */
118 static void vp_finalize_features(struct virtio_device *vdev)
120 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
122 /* Give virtio_ring a chance to accept features. */
123 vring_transport_features(vdev);
125 /* We only support 32 feature bits. */
126 BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
127 iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
130 /* virtio config->get() implementation */
131 static void vp_get(struct virtio_device *vdev, unsigned offset,
132 void *buf, unsigned len)
134 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
135 void __iomem *ioaddr = vp_dev->ioaddr +
136 VIRTIO_PCI_CONFIG(vp_dev) + offset;
140 for (i = 0; i < len; i++)
141 ptr[i] = ioread8(ioaddr + i);
144 /* the config->set() implementation. it's symmetric to the config->get()
146 static void vp_set(struct virtio_device *vdev, unsigned offset,
147 const void *buf, unsigned len)
149 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
150 void __iomem *ioaddr = vp_dev->ioaddr +
151 VIRTIO_PCI_CONFIG(vp_dev) + offset;
155 for (i = 0; i < len; i++)
156 iowrite8(ptr[i], ioaddr + i);
159 /* config->{get,set}_status() implementations */
160 static u8 vp_get_status(struct virtio_device *vdev)
162 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
163 return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
166 static void vp_set_status(struct virtio_device *vdev, u8 status)
168 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
169 /* We should never be setting status to 0. */
171 iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
174 static void vp_reset(struct virtio_device *vdev)
176 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
177 /* 0 status means a reset. */
178 iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
181 /* the notify function used when creating a virt queue */
182 static void vp_notify(struct virtqueue *vq)
184 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
185 struct virtio_pci_vq_info *info = vq->priv;
187 /* we write the queue's selector into the notification register to
188 * signal the other end */
189 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
192 /* Handle a configuration change: Tell driver if it wants to know. */
193 static irqreturn_t vp_config_changed(int irq, void *opaque)
195 struct virtio_pci_device *vp_dev = opaque;
196 struct virtio_driver *drv;
197 drv = container_of(vp_dev->vdev.dev.driver,
198 struct virtio_driver, driver);
200 if (drv && drv->config_changed)
201 drv->config_changed(&vp_dev->vdev);
205 /* Notify all virtqueues on an interrupt. */
206 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
208 struct virtio_pci_device *vp_dev = opaque;
209 struct virtio_pci_vq_info *info;
210 irqreturn_t ret = IRQ_NONE;
213 spin_lock_irqsave(&vp_dev->lock, flags);
214 list_for_each_entry(info, &vp_dev->virtqueues, node) {
215 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
218 spin_unlock_irqrestore(&vp_dev->lock, flags);
223 /* A small wrapper to also acknowledge the interrupt when it's handled.
224 * I really need an EIO hook for the vring so I can ack the interrupt once we
225 * know that we'll be handling the IRQ but before we invoke the callback since
226 * the callback may notify the host which results in the host attempting to
227 * raise an interrupt that we would then mask once we acknowledged the
229 static irqreturn_t vp_interrupt(int irq, void *opaque)
231 struct virtio_pci_device *vp_dev = opaque;
234 /* reading the ISR has the effect of also clearing it so it's very
235 * important to save off the value. */
236 isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
238 /* It's definitely not us if the ISR was not high */
242 /* Configuration change? Tell driver if it wants to know. */
243 if (isr & VIRTIO_PCI_ISR_CONFIG)
244 vp_config_changed(irq, opaque);
246 return vp_vring_interrupt(irq, opaque);
249 static void vp_free_vectors(struct virtio_device *vdev)
251 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
254 if (vp_dev->intx_enabled) {
255 free_irq(vp_dev->pci_dev->irq, vp_dev);
256 vp_dev->intx_enabled = 0;
259 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
260 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
262 if (vp_dev->msix_enabled) {
263 /* Disable the vector used for configuration */
264 iowrite16(VIRTIO_MSI_NO_VECTOR,
265 vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
266 /* Flush the write out to device */
267 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
269 pci_disable_msix(vp_dev->pci_dev);
270 vp_dev->msix_enabled = 0;
271 vp_dev->msix_vectors = 0;
274 vp_dev->msix_used_vectors = 0;
275 kfree(vp_dev->msix_names);
276 vp_dev->msix_names = NULL;
277 kfree(vp_dev->msix_entries);
278 vp_dev->msix_entries = NULL;
281 static int vp_enable_msix(struct pci_dev *dev, struct msix_entry *entries,
282 int *options, int noptions)
285 for (i = 0; i < noptions; ++i)
286 if (!pci_enable_msix(dev, entries, options[i]))
291 static int vp_request_vectors(struct virtio_device *vdev, unsigned max_vqs)
293 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
294 const char *name = dev_name(&vp_dev->vdev.dev);
297 /* We want at most one vector per queue and one for config changes.
298 * Fallback to separate vectors for config and a shared for queues.
299 * Finally fall back to regular interrupts. */
300 int options[] = { max_vqs + 1, 2 };
301 int nvectors = max(options[0], options[1]);
303 vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
305 if (!vp_dev->msix_entries)
307 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
309 if (!vp_dev->msix_names)
312 for (i = 0; i < nvectors; ++i)
313 vp_dev->msix_entries[i].entry = i;
315 err = vp_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries,
316 options, ARRAY_SIZE(options));
318 /* Can't allocate enough MSI-X vectors, use regular interrupt */
319 vp_dev->msix_vectors = 0;
320 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
321 IRQF_SHARED, name, vp_dev);
324 vp_dev->intx_enabled = 1;
326 vp_dev->msix_vectors = err;
327 vp_dev->msix_enabled = 1;
329 /* Set the vector used for configuration */
330 v = vp_dev->msix_used_vectors;
331 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
333 err = request_irq(vp_dev->msix_entries[v].vector,
334 vp_config_changed, 0, vp_dev->msix_names[v],
338 ++vp_dev->msix_used_vectors;
340 iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
341 /* Verify we had enough resources to assign the vector */
342 v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
343 if (v == VIRTIO_MSI_NO_VECTOR) {
349 if (vp_dev->msix_vectors && vp_dev->msix_vectors != max_vqs + 1) {
350 /* Shared vector for all VQs */
351 v = vp_dev->msix_used_vectors;
352 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
353 "%s-virtqueues", name);
354 err = request_irq(vp_dev->msix_entries[v].vector,
355 vp_vring_interrupt, 0, vp_dev->msix_names[v],
359 ++vp_dev->msix_used_vectors;
363 vp_free_vectors(vdev);
367 static struct virtqueue *vp_find_vq(struct virtio_device *vdev, unsigned index,
368 void (*callback)(struct virtqueue *vq),
371 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
372 struct virtio_pci_vq_info *info;
373 struct virtqueue *vq;
374 unsigned long flags, size;
378 /* Select the queue we're interested in */
379 iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
381 /* Check if queue is either not available or already active. */
382 num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
383 if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
384 return ERR_PTR(-ENOENT);
386 /* allocate and fill out our structure the represents an active
388 info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
390 return ERR_PTR(-ENOMEM);
392 info->queue_index = index;
394 info->vector = VIRTIO_MSI_NO_VECTOR;
396 size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
397 info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
398 if (info->queue == NULL) {
403 /* activate the queue */
404 iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
405 vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
407 /* create the vring */
408 vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
409 vdev, info->queue, vp_notify, callback, name);
412 goto out_activate_queue;
418 /* allocate per-vq vector if available and necessary */
419 if (callback && vp_dev->msix_used_vectors < vp_dev->msix_vectors) {
420 vector = vp_dev->msix_used_vectors;
421 snprintf(vp_dev->msix_names[vector], sizeof *vp_dev->msix_names,
422 "%s-%s", dev_name(&vp_dev->vdev.dev), name);
423 err = request_irq(vp_dev->msix_entries[vector].vector,
425 vp_dev->msix_names[vector], vq);
427 goto out_request_irq;
428 info->vector = vector;
429 ++vp_dev->msix_used_vectors;
431 vector = VP_MSIX_VQ_VECTOR;
433 if (callback && vp_dev->msix_enabled) {
434 iowrite16(vector, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
435 vector = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
436 if (vector == VIRTIO_MSI_NO_VECTOR) {
442 spin_lock_irqsave(&vp_dev->lock, flags);
443 list_add(&info->node, &vp_dev->virtqueues);
444 spin_unlock_irqrestore(&vp_dev->lock, flags);
449 if (info->vector != VIRTIO_MSI_NO_VECTOR) {
450 free_irq(vp_dev->msix_entries[info->vector].vector, vq);
451 --vp_dev->msix_used_vectors;
454 vring_del_virtqueue(vq);
456 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
457 free_pages_exact(info->queue, size);
463 static void vp_del_vq(struct virtqueue *vq)
465 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
466 struct virtio_pci_vq_info *info = vq->priv;
467 unsigned long flags, size;
469 spin_lock_irqsave(&vp_dev->lock, flags);
470 list_del(&info->node);
471 spin_unlock_irqrestore(&vp_dev->lock, flags);
473 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
475 if (info->vector != VIRTIO_MSI_NO_VECTOR)
476 free_irq(vp_dev->msix_entries[info->vector].vector, vq);
478 if (vp_dev->msix_enabled) {
479 iowrite16(VIRTIO_MSI_NO_VECTOR,
480 vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
481 /* Flush the write out to device */
482 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
485 vring_del_virtqueue(vq);
487 /* Select and deactivate the queue */
488 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
490 size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
491 free_pages_exact(info->queue, size);
495 /* the config->del_vqs() implementation */
496 static void vp_del_vqs(struct virtio_device *vdev)
498 struct virtqueue *vq, *n;
500 list_for_each_entry_safe(vq, n, &vdev->vqs, list)
503 vp_free_vectors(vdev);
506 /* the config->find_vqs() implementation */
507 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
508 struct virtqueue *vqs[],
509 vq_callback_t *callbacks[],
515 /* How many vectors would we like? */
516 for (i = 0; i < nvqs; ++i)
520 err = vp_request_vectors(vdev, vectors);
524 for (i = 0; i < nvqs; ++i) {
525 vqs[i] = vp_find_vq(vdev, i, callbacks[i], names[i]);
535 return PTR_ERR(vqs[i]);
538 static struct virtio_config_ops virtio_pci_config_ops = {
541 .get_status = vp_get_status,
542 .set_status = vp_set_status,
544 .find_vqs = vp_find_vqs,
545 .del_vqs = vp_del_vqs,
546 .get_features = vp_get_features,
547 .finalize_features = vp_finalize_features,
550 static void virtio_pci_release_dev(struct device *_d)
552 struct virtio_device *dev = container_of(_d, struct virtio_device, dev);
553 struct virtio_pci_device *vp_dev = to_vp_device(dev);
554 struct pci_dev *pci_dev = vp_dev->pci_dev;
557 pci_set_drvdata(pci_dev, NULL);
558 pci_iounmap(pci_dev, vp_dev->ioaddr);
559 pci_release_regions(pci_dev);
560 pci_disable_device(pci_dev);
564 /* the PCI probing function */
565 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
566 const struct pci_device_id *id)
568 struct virtio_pci_device *vp_dev;
571 /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
572 if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
575 if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
576 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
577 VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
581 /* allocate our structure and fill it out */
582 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
586 vp_dev->vdev.dev.parent = virtio_pci_root;
587 vp_dev->vdev.dev.release = virtio_pci_release_dev;
588 vp_dev->vdev.config = &virtio_pci_config_ops;
589 vp_dev->pci_dev = pci_dev;
590 INIT_LIST_HEAD(&vp_dev->virtqueues);
591 spin_lock_init(&vp_dev->lock);
593 /* enable the device */
594 err = pci_enable_device(pci_dev);
598 err = pci_request_regions(pci_dev, "virtio-pci");
600 goto out_enable_device;
602 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
603 if (vp_dev->ioaddr == NULL)
604 goto out_req_regions;
606 pci_set_drvdata(pci_dev, vp_dev);
608 /* we use the subsystem vendor/device id as the virtio vendor/device
609 * id. this allows us to use the same PCI vendor/device id for all
610 * virtio devices and to identify the particular virtio driver by
611 * the subsytem ids */
612 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
613 vp_dev->vdev.id.device = pci_dev->subsystem_device;
615 /* finally register the virtio device */
616 err = register_virtio_device(&vp_dev->vdev);
618 goto out_set_drvdata;
623 pci_set_drvdata(pci_dev, NULL);
624 pci_iounmap(pci_dev, vp_dev->ioaddr);
626 pci_release_regions(pci_dev);
628 pci_disable_device(pci_dev);
634 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
636 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
638 unregister_virtio_device(&vp_dev->vdev);
642 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
644 pci_save_state(pci_dev);
645 pci_set_power_state(pci_dev, PCI_D3hot);
649 static int virtio_pci_resume(struct pci_dev *pci_dev)
651 pci_restore_state(pci_dev);
652 pci_set_power_state(pci_dev, PCI_D0);
657 static struct pci_driver virtio_pci_driver = {
658 .name = "virtio-pci",
659 .id_table = virtio_pci_id_table,
660 .probe = virtio_pci_probe,
661 .remove = virtio_pci_remove,
663 .suspend = virtio_pci_suspend,
664 .resume = virtio_pci_resume,
668 static int __init virtio_pci_init(void)
672 virtio_pci_root = root_device_register("virtio-pci");
673 if (IS_ERR(virtio_pci_root))
674 return PTR_ERR(virtio_pci_root);
676 err = pci_register_driver(&virtio_pci_driver);
678 root_device_unregister(virtio_pci_root);
683 module_init(virtio_pci_init);
685 static void __exit virtio_pci_exit(void)
687 pci_unregister_driver(&virtio_pci_driver);
688 root_device_unregister(virtio_pci_root);
691 module_exit(virtio_pci_exit);