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/slab.h>
21 #include <linux/interrupt.h>
22 #include <linux/virtio.h>
23 #include <linux/virtio_config.h>
24 #include <linux/virtio_ring.h>
25 #include <linux/virtio_pci.h>
26 #include <linux/highmem.h>
27 #include <linux/spinlock.h>
29 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
30 MODULE_DESCRIPTION("virtio-pci");
31 MODULE_LICENSE("GPL");
34 /* Our device structure */
35 struct virtio_pci_device
37 struct virtio_device vdev;
38 struct pci_dev *pci_dev;
40 /* the IO mapping for the PCI config space */
43 /* a list of queues so we can dispatch IRQs */
45 struct list_head virtqueues;
50 struct msix_entry *msix_entries;
51 /* Name strings for interrupts. This size should be enough,
52 * and I'm too lazy to allocate each name separately. */
53 char (*msix_names)[256];
54 /* Number of available vectors */
55 unsigned msix_vectors;
56 /* Vectors allocated, excluding per-vq vectors if any */
57 unsigned msix_used_vectors;
58 /* Whether we have vector per vq */
62 /* Constants for MSI-X */
63 /* Use first vector for configuration changes, second and the rest for
64 * virtqueues Thus, we need at least 2 vectors for MSI. */
66 VP_MSIX_CONFIG_VECTOR = 0,
67 VP_MSIX_VQ_VECTOR = 1,
70 struct virtio_pci_vq_info
72 /* the actual virtqueue */
75 /* the number of entries in the queue */
78 /* the index of the queue */
81 /* the virtual address of the ring queue */
84 /* the list node for the virtqueues list */
85 struct list_head node;
87 /* MSI-X vector (or none) */
91 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
92 static struct pci_device_id virtio_pci_id_table[] = {
93 { 0x1af4, PCI_ANY_ID, PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0 },
97 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
99 /* A PCI device has it's own struct device and so does a virtio device so
100 * we create a place for the virtio devices to show up in sysfs. I think it
101 * would make more sense for virtio to not insist on having it's own device. */
102 static struct device *virtio_pci_root;
104 /* Convert a generic virtio device to our structure */
105 static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)
107 return container_of(vdev, struct virtio_pci_device, vdev);
110 /* virtio config->get_features() implementation */
111 static u32 vp_get_features(struct virtio_device *vdev)
113 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
115 /* When someone needs more than 32 feature bits, we'll need to
116 * steal a bit to indicate that the rest are somewhere else. */
117 return ioread32(vp_dev->ioaddr + VIRTIO_PCI_HOST_FEATURES);
120 /* virtio config->finalize_features() implementation */
121 static void vp_finalize_features(struct virtio_device *vdev)
123 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
125 /* Give virtio_ring a chance to accept features. */
126 vring_transport_features(vdev);
128 /* We only support 32 feature bits. */
129 BUILD_BUG_ON(ARRAY_SIZE(vdev->features) != 1);
130 iowrite32(vdev->features[0], vp_dev->ioaddr+VIRTIO_PCI_GUEST_FEATURES);
133 /* virtio config->get() implementation */
134 static void vp_get(struct virtio_device *vdev, unsigned offset,
135 void *buf, unsigned len)
137 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
138 void __iomem *ioaddr = vp_dev->ioaddr +
139 VIRTIO_PCI_CONFIG(vp_dev) + offset;
143 for (i = 0; i < len; i++)
144 ptr[i] = ioread8(ioaddr + i);
147 /* the config->set() implementation. it's symmetric to the config->get()
149 static void vp_set(struct virtio_device *vdev, unsigned offset,
150 const void *buf, unsigned len)
152 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
153 void __iomem *ioaddr = vp_dev->ioaddr +
154 VIRTIO_PCI_CONFIG(vp_dev) + offset;
158 for (i = 0; i < len; i++)
159 iowrite8(ptr[i], ioaddr + i);
162 /* config->{get,set}_status() implementations */
163 static u8 vp_get_status(struct virtio_device *vdev)
165 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
166 return ioread8(vp_dev->ioaddr + VIRTIO_PCI_STATUS);
169 static void vp_set_status(struct virtio_device *vdev, u8 status)
171 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
172 /* We should never be setting status to 0. */
174 iowrite8(status, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
177 static void vp_reset(struct virtio_device *vdev)
179 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
180 /* 0 status means a reset. */
181 iowrite8(0, vp_dev->ioaddr + VIRTIO_PCI_STATUS);
184 /* the notify function used when creating a virt queue */
185 static void vp_notify(struct virtqueue *vq)
187 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
188 struct virtio_pci_vq_info *info = vq->priv;
190 /* we write the queue's selector into the notification register to
191 * signal the other end */
192 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NOTIFY);
195 /* Handle a configuration change: Tell driver if it wants to know. */
196 static irqreturn_t vp_config_changed(int irq, void *opaque)
198 struct virtio_pci_device *vp_dev = opaque;
199 struct virtio_driver *drv;
200 drv = container_of(vp_dev->vdev.dev.driver,
201 struct virtio_driver, driver);
203 if (drv && drv->config_changed)
204 drv->config_changed(&vp_dev->vdev);
208 /* Notify all virtqueues on an interrupt. */
209 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
211 struct virtio_pci_device *vp_dev = opaque;
212 struct virtio_pci_vq_info *info;
213 irqreturn_t ret = IRQ_NONE;
216 spin_lock_irqsave(&vp_dev->lock, flags);
217 list_for_each_entry(info, &vp_dev->virtqueues, node) {
218 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
221 spin_unlock_irqrestore(&vp_dev->lock, flags);
226 /* A small wrapper to also acknowledge the interrupt when it's handled.
227 * I really need an EIO hook for the vring so I can ack the interrupt once we
228 * know that we'll be handling the IRQ but before we invoke the callback since
229 * the callback may notify the host which results in the host attempting to
230 * raise an interrupt that we would then mask once we acknowledged the
232 static irqreturn_t vp_interrupt(int irq, void *opaque)
234 struct virtio_pci_device *vp_dev = opaque;
237 /* reading the ISR has the effect of also clearing it so it's very
238 * important to save off the value. */
239 isr = ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
241 /* It's definitely not us if the ISR was not high */
245 /* Configuration change? Tell driver if it wants to know. */
246 if (isr & VIRTIO_PCI_ISR_CONFIG)
247 vp_config_changed(irq, opaque);
249 return vp_vring_interrupt(irq, opaque);
252 static void vp_free_vectors(struct virtio_device *vdev)
254 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
257 if (vp_dev->intx_enabled) {
258 free_irq(vp_dev->pci_dev->irq, vp_dev);
259 vp_dev->intx_enabled = 0;
262 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
263 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
265 if (vp_dev->msix_enabled) {
266 /* Disable the vector used for configuration */
267 iowrite16(VIRTIO_MSI_NO_VECTOR,
268 vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
269 /* Flush the write out to device */
270 ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
272 pci_disable_msix(vp_dev->pci_dev);
273 vp_dev->msix_enabled = 0;
274 vp_dev->msix_vectors = 0;
277 vp_dev->msix_used_vectors = 0;
278 kfree(vp_dev->msix_names);
279 vp_dev->msix_names = NULL;
280 kfree(vp_dev->msix_entries);
281 vp_dev->msix_entries = NULL;
284 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
287 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
288 const char *name = dev_name(&vp_dev->vdev.dev);
292 vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
294 if (!vp_dev->msix_entries)
296 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
298 if (!vp_dev->msix_names)
301 for (i = 0; i < nvectors; ++i)
302 vp_dev->msix_entries[i].entry = i;
304 /* pci_enable_msix returns positive if we can't get this many. */
305 err = pci_enable_msix(vp_dev->pci_dev, vp_dev->msix_entries, nvectors);
310 vp_dev->msix_vectors = nvectors;
311 vp_dev->msix_enabled = 1;
313 /* Set the vector used for configuration */
314 v = vp_dev->msix_used_vectors;
315 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
317 err = request_irq(vp_dev->msix_entries[v].vector,
318 vp_config_changed, 0, vp_dev->msix_names[v],
322 ++vp_dev->msix_used_vectors;
324 iowrite16(v, vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
325 /* Verify we had enough resources to assign the vector */
326 v = ioread16(vp_dev->ioaddr + VIRTIO_MSI_CONFIG_VECTOR);
327 if (v == VIRTIO_MSI_NO_VECTOR) {
332 if (!per_vq_vectors) {
333 /* Shared vector for all VQs */
334 v = vp_dev->msix_used_vectors;
335 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
336 "%s-virtqueues", name);
337 err = request_irq(vp_dev->msix_entries[v].vector,
338 vp_vring_interrupt, 0, vp_dev->msix_names[v],
342 ++vp_dev->msix_used_vectors;
346 vp_free_vectors(vdev);
350 static int vp_request_intx(struct virtio_device *vdev)
353 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
355 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
356 IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
358 vp_dev->intx_enabled = 1;
362 static struct virtqueue *setup_vq(struct virtio_device *vdev, unsigned index,
363 void (*callback)(struct virtqueue *vq),
367 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
368 struct virtio_pci_vq_info *info;
369 struct virtqueue *vq;
370 unsigned long flags, size;
374 /* Select the queue we're interested in */
375 iowrite16(index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
377 /* Check if queue is either not available or already active. */
378 num = ioread16(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_NUM);
379 if (!num || ioread32(vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN))
380 return ERR_PTR(-ENOENT);
382 /* allocate and fill out our structure the represents an active
384 info = kmalloc(sizeof(struct virtio_pci_vq_info), GFP_KERNEL);
386 return ERR_PTR(-ENOMEM);
388 info->queue_index = index;
390 info->msix_vector = msix_vec;
392 size = PAGE_ALIGN(vring_size(num, VIRTIO_PCI_VRING_ALIGN));
393 info->queue = alloc_pages_exact(size, GFP_KERNEL|__GFP_ZERO);
394 if (info->queue == NULL) {
399 /* activate the queue */
400 iowrite32(virt_to_phys(info->queue) >> VIRTIO_PCI_QUEUE_ADDR_SHIFT,
401 vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
403 /* create the vring */
404 vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
405 vdev, info->queue, vp_notify, callback, name);
408 goto out_activate_queue;
414 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
415 iowrite16(msix_vec, vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
416 msix_vec = ioread16(vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
417 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
423 spin_lock_irqsave(&vp_dev->lock, flags);
424 list_add(&info->node, &vp_dev->virtqueues);
425 spin_unlock_irqrestore(&vp_dev->lock, flags);
430 vring_del_virtqueue(vq);
432 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
433 free_pages_exact(info->queue, size);
439 static void vp_del_vq(struct virtqueue *vq)
441 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
442 struct virtio_pci_vq_info *info = vq->priv;
443 unsigned long flags, size;
445 spin_lock_irqsave(&vp_dev->lock, flags);
446 list_del(&info->node);
447 spin_unlock_irqrestore(&vp_dev->lock, flags);
449 iowrite16(info->queue_index, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_SEL);
451 if (vp_dev->msix_enabled) {
452 iowrite16(VIRTIO_MSI_NO_VECTOR,
453 vp_dev->ioaddr + VIRTIO_MSI_QUEUE_VECTOR);
454 /* Flush the write out to device */
455 ioread8(vp_dev->ioaddr + VIRTIO_PCI_ISR);
458 vring_del_virtqueue(vq);
460 /* Select and deactivate the queue */
461 iowrite32(0, vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
463 size = PAGE_ALIGN(vring_size(info->num, VIRTIO_PCI_VRING_ALIGN));
464 free_pages_exact(info->queue, size);
468 /* the config->del_vqs() implementation */
469 static void vp_del_vqs(struct virtio_device *vdev)
471 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
472 struct virtqueue *vq, *n;
473 struct virtio_pci_vq_info *info;
475 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
477 if (vp_dev->per_vq_vectors &&
478 info->msix_vector != VIRTIO_MSI_NO_VECTOR)
479 free_irq(vp_dev->msix_entries[info->msix_vector].vector,
483 vp_dev->per_vq_vectors = false;
485 vp_free_vectors(vdev);
488 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
489 struct virtqueue *vqs[],
490 vq_callback_t *callbacks[],
495 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
497 int i, err, nvectors, allocated_vectors;
500 /* Old style: one normal interrupt for change and all vqs. */
501 err = vp_request_intx(vdev);
505 if (per_vq_vectors) {
506 /* Best option: one for change interrupt, one per vq. */
508 for (i = 0; i < nvqs; ++i)
512 /* Second best: one for change, shared for all vqs. */
516 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
521 vp_dev->per_vq_vectors = per_vq_vectors;
522 allocated_vectors = vp_dev->msix_used_vectors;
523 for (i = 0; i < nvqs; ++i) {
524 if (!callbacks[i] || !vp_dev->msix_enabled)
525 msix_vec = VIRTIO_MSI_NO_VECTOR;
526 else if (vp_dev->per_vq_vectors)
527 msix_vec = allocated_vectors++;
529 msix_vec = VP_MSIX_VQ_VECTOR;
530 vqs[i] = setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
531 if (IS_ERR(vqs[i])) {
532 err = PTR_ERR(vqs[i]);
536 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
539 /* allocate per-vq irq if available and necessary */
540 snprintf(vp_dev->msix_names[msix_vec],
541 sizeof *vp_dev->msix_names,
543 dev_name(&vp_dev->vdev.dev), names[i]);
544 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
546 vp_dev->msix_names[msix_vec],
562 /* the config->find_vqs() implementation */
563 static int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
564 struct virtqueue *vqs[],
565 vq_callback_t *callbacks[],
570 /* Try MSI-X with one vector per queue. */
571 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
574 /* Fallback: MSI-X with one vector for config, one shared for queues. */
575 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
579 /* Finally fall back to regular interrupts. */
580 return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
584 static struct virtio_config_ops virtio_pci_config_ops = {
587 .get_status = vp_get_status,
588 .set_status = vp_set_status,
590 .find_vqs = vp_find_vqs,
591 .del_vqs = vp_del_vqs,
592 .get_features = vp_get_features,
593 .finalize_features = vp_finalize_features,
596 static void virtio_pci_release_dev(struct device *_d)
598 struct virtio_device *dev = container_of(_d, struct virtio_device, dev);
599 struct virtio_pci_device *vp_dev = to_vp_device(dev);
600 struct pci_dev *pci_dev = vp_dev->pci_dev;
603 pci_set_drvdata(pci_dev, NULL);
604 pci_iounmap(pci_dev, vp_dev->ioaddr);
605 pci_release_regions(pci_dev);
606 pci_disable_device(pci_dev);
610 /* the PCI probing function */
611 static int __devinit virtio_pci_probe(struct pci_dev *pci_dev,
612 const struct pci_device_id *id)
614 struct virtio_pci_device *vp_dev;
617 /* We only own devices >= 0x1000 and <= 0x103f: leave the rest. */
618 if (pci_dev->device < 0x1000 || pci_dev->device > 0x103f)
621 if (pci_dev->revision != VIRTIO_PCI_ABI_VERSION) {
622 printk(KERN_ERR "virtio_pci: expected ABI version %d, got %d\n",
623 VIRTIO_PCI_ABI_VERSION, pci_dev->revision);
627 /* allocate our structure and fill it out */
628 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
632 vp_dev->vdev.dev.parent = virtio_pci_root;
633 vp_dev->vdev.dev.release = virtio_pci_release_dev;
634 vp_dev->vdev.config = &virtio_pci_config_ops;
635 vp_dev->pci_dev = pci_dev;
636 INIT_LIST_HEAD(&vp_dev->virtqueues);
637 spin_lock_init(&vp_dev->lock);
639 /* Disable MSI/MSIX to bring device to a known good state. */
640 pci_msi_off(pci_dev);
642 /* enable the device */
643 err = pci_enable_device(pci_dev);
647 err = pci_request_regions(pci_dev, "virtio-pci");
649 goto out_enable_device;
651 vp_dev->ioaddr = pci_iomap(pci_dev, 0, 0);
652 if (vp_dev->ioaddr == NULL)
653 goto out_req_regions;
655 pci_set_drvdata(pci_dev, vp_dev);
656 pci_set_master(pci_dev);
658 /* we use the subsystem vendor/device id as the virtio vendor/device
659 * id. this allows us to use the same PCI vendor/device id for all
660 * virtio devices and to identify the particular virtio driver by
661 * the subsystem ids */
662 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
663 vp_dev->vdev.id.device = pci_dev->subsystem_device;
665 /* finally register the virtio device */
666 err = register_virtio_device(&vp_dev->vdev);
668 goto out_set_drvdata;
673 pci_set_drvdata(pci_dev, NULL);
674 pci_iounmap(pci_dev, vp_dev->ioaddr);
676 pci_release_regions(pci_dev);
678 pci_disable_device(pci_dev);
684 static void __devexit virtio_pci_remove(struct pci_dev *pci_dev)
686 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
688 unregister_virtio_device(&vp_dev->vdev);
692 static int virtio_pci_suspend(struct pci_dev *pci_dev, pm_message_t state)
694 pci_save_state(pci_dev);
695 pci_set_power_state(pci_dev, PCI_D3hot);
699 static int virtio_pci_resume(struct pci_dev *pci_dev)
701 pci_restore_state(pci_dev);
702 pci_set_power_state(pci_dev, PCI_D0);
707 static struct pci_driver virtio_pci_driver = {
708 .name = "virtio-pci",
709 .id_table = virtio_pci_id_table,
710 .probe = virtio_pci_probe,
711 .remove = __devexit_p(virtio_pci_remove),
713 .suspend = virtio_pci_suspend,
714 .resume = virtio_pci_resume,
718 static int __init virtio_pci_init(void)
722 virtio_pci_root = root_device_register("virtio-pci");
723 if (IS_ERR(virtio_pci_root))
724 return PTR_ERR(virtio_pci_root);
726 err = pci_register_driver(&virtio_pci_driver);
728 root_device_unregister(virtio_pci_root);
733 module_init(virtio_pci_init);
735 static void __exit virtio_pci_exit(void)
737 pci_unregister_driver(&virtio_pci_driver);
738 root_device_unregister(virtio_pci_root);
741 module_exit(virtio_pci_exit);