2 * Virtio PCI driver - common functionality for all device versions
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
8 * Copyright Red Hat, Inc. 2014
11 * Anthony Liguori <aliguori@us.ibm.com>
12 * Rusty Russell <rusty@rustcorp.com.au>
13 * Michael S. Tsirkin <mst@redhat.com>
15 * This work is licensed under the terms of the GNU GPL, version 2 or later.
16 * See the COPYING file in the top-level directory.
20 #include "virtio_pci_common.h"
22 /* wait for pending irq handlers */
23 void vp_synchronize_vectors(struct virtio_device *vdev)
25 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
28 if (vp_dev->intx_enabled)
29 synchronize_irq(vp_dev->pci_dev->irq);
31 for (i = 0; i < vp_dev->msix_vectors; ++i)
32 synchronize_irq(vp_dev->msix_entries[i].vector);
35 /* the notify function used when creating a virt queue */
36 bool vp_notify(struct virtqueue *vq)
38 /* we write the queue's selector into the notification register to
39 * signal the other end */
40 iowrite16(vq->index, (void __iomem *)vq->priv);
44 /* Handle a configuration change: Tell driver if it wants to know. */
45 static irqreturn_t vp_config_changed(int irq, void *opaque)
47 struct virtio_pci_device *vp_dev = opaque;
49 virtio_config_changed(&vp_dev->vdev);
53 /* Notify all virtqueues on an interrupt. */
54 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
56 struct virtio_pci_device *vp_dev = opaque;
57 struct virtio_pci_vq_info *info;
58 irqreturn_t ret = IRQ_NONE;
61 spin_lock_irqsave(&vp_dev->lock, flags);
62 list_for_each_entry(info, &vp_dev->virtqueues, node) {
63 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
66 spin_unlock_irqrestore(&vp_dev->lock, flags);
71 /* A small wrapper to also acknowledge the interrupt when it's handled.
72 * I really need an EIO hook for the vring so I can ack the interrupt once we
73 * know that we'll be handling the IRQ but before we invoke the callback since
74 * the callback may notify the host which results in the host attempting to
75 * raise an interrupt that we would then mask once we acknowledged the
77 static irqreturn_t vp_interrupt(int irq, void *opaque)
79 struct virtio_pci_device *vp_dev = opaque;
82 /* reading the ISR has the effect of also clearing it so it's very
83 * important to save off the value. */
84 isr = ioread8(vp_dev->isr);
86 /* It's definitely not us if the ISR was not high */
90 /* Configuration change? Tell driver if it wants to know. */
91 if (isr & VIRTIO_PCI_ISR_CONFIG)
92 vp_config_changed(irq, opaque);
94 return vp_vring_interrupt(irq, opaque);
97 static void vp_free_vectors(struct virtio_device *vdev)
99 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
102 if (vp_dev->intx_enabled) {
103 free_irq(vp_dev->pci_dev->irq, vp_dev);
104 vp_dev->intx_enabled = 0;
107 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
108 free_irq(vp_dev->msix_entries[i].vector, vp_dev);
110 for (i = 0; i < vp_dev->msix_vectors; i++)
111 if (vp_dev->msix_affinity_masks[i])
112 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
114 if (vp_dev->msix_enabled) {
115 /* Disable the vector used for configuration */
116 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
118 pci_disable_msix(vp_dev->pci_dev);
119 vp_dev->msix_enabled = 0;
122 vp_dev->msix_vectors = 0;
123 vp_dev->msix_used_vectors = 0;
124 kfree(vp_dev->msix_names);
125 vp_dev->msix_names = NULL;
126 kfree(vp_dev->msix_entries);
127 vp_dev->msix_entries = NULL;
128 kfree(vp_dev->msix_affinity_masks);
129 vp_dev->msix_affinity_masks = NULL;
132 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
135 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
136 const char *name = dev_name(&vp_dev->vdev.dev);
140 vp_dev->msix_vectors = nvectors;
142 vp_dev->msix_entries = kmalloc(nvectors * sizeof *vp_dev->msix_entries,
144 if (!vp_dev->msix_entries)
146 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
148 if (!vp_dev->msix_names)
150 vp_dev->msix_affinity_masks
151 = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
153 if (!vp_dev->msix_affinity_masks)
155 for (i = 0; i < nvectors; ++i)
156 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
160 for (i = 0; i < nvectors; ++i)
161 vp_dev->msix_entries[i].entry = i;
163 err = pci_enable_msix_exact(vp_dev->pci_dev,
164 vp_dev->msix_entries, nvectors);
167 vp_dev->msix_enabled = 1;
169 /* Set the vector used for configuration */
170 v = vp_dev->msix_used_vectors;
171 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
173 err = request_irq(vp_dev->msix_entries[v].vector,
174 vp_config_changed, 0, vp_dev->msix_names[v],
178 ++vp_dev->msix_used_vectors;
180 v = vp_dev->config_vector(vp_dev, v);
181 /* Verify we had enough resources to assign the vector */
182 if (v == VIRTIO_MSI_NO_VECTOR) {
187 if (!per_vq_vectors) {
188 /* Shared vector for all VQs */
189 v = vp_dev->msix_used_vectors;
190 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
191 "%s-virtqueues", name);
192 err = request_irq(vp_dev->msix_entries[v].vector,
193 vp_vring_interrupt, 0, vp_dev->msix_names[v],
197 ++vp_dev->msix_used_vectors;
201 vp_free_vectors(vdev);
205 static int vp_request_intx(struct virtio_device *vdev)
208 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
210 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt,
211 IRQF_SHARED, dev_name(&vdev->dev), vp_dev);
213 vp_dev->intx_enabled = 1;
217 static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned index,
218 void (*callback)(struct virtqueue *vq),
222 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
223 struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
224 struct virtqueue *vq;
227 /* fill out our structure that represents an active queue */
229 return ERR_PTR(-ENOMEM);
231 vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, msix_vec);
237 spin_lock_irqsave(&vp_dev->lock, flags);
238 list_add(&info->node, &vp_dev->virtqueues);
239 spin_unlock_irqrestore(&vp_dev->lock, flags);
241 INIT_LIST_HEAD(&info->node);
244 vp_dev->vqs[index] = info;
252 static void vp_del_vq(struct virtqueue *vq)
254 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
255 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
258 spin_lock_irqsave(&vp_dev->lock, flags);
259 list_del(&info->node);
260 spin_unlock_irqrestore(&vp_dev->lock, flags);
262 vp_dev->del_vq(info);
266 /* the config->del_vqs() implementation */
267 void vp_del_vqs(struct virtio_device *vdev)
269 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
270 struct virtqueue *vq, *n;
271 struct virtio_pci_vq_info *info;
273 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
274 info = vp_dev->vqs[vq->index];
275 if (vp_dev->per_vq_vectors &&
276 info->msix_vector != VIRTIO_MSI_NO_VECTOR)
277 free_irq(vp_dev->msix_entries[info->msix_vector].vector,
281 vp_dev->per_vq_vectors = false;
283 vp_free_vectors(vdev);
288 static int vp_try_to_find_vqs(struct virtio_device *vdev, unsigned nvqs,
289 struct virtqueue *vqs[],
290 vq_callback_t *callbacks[],
295 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
297 int i, err, nvectors, allocated_vectors;
299 vp_dev->vqs = kmalloc(nvqs * sizeof *vp_dev->vqs, GFP_KERNEL);
304 /* Old style: one normal interrupt for change and all vqs. */
305 err = vp_request_intx(vdev);
309 if (per_vq_vectors) {
310 /* Best option: one for change interrupt, one per vq. */
312 for (i = 0; i < nvqs; ++i)
316 /* Second best: one for change, shared for all vqs. */
320 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
325 vp_dev->per_vq_vectors = per_vq_vectors;
326 allocated_vectors = vp_dev->msix_used_vectors;
327 for (i = 0; i < nvqs; ++i) {
331 } else if (!callbacks[i] || !vp_dev->msix_enabled)
332 msix_vec = VIRTIO_MSI_NO_VECTOR;
333 else if (vp_dev->per_vq_vectors)
334 msix_vec = allocated_vectors++;
336 msix_vec = VP_MSIX_VQ_VECTOR;
337 vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i], msix_vec);
338 if (IS_ERR(vqs[i])) {
339 err = PTR_ERR(vqs[i]);
343 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
346 /* allocate per-vq irq if available and necessary */
347 snprintf(vp_dev->msix_names[msix_vec],
348 sizeof *vp_dev->msix_names,
350 dev_name(&vp_dev->vdev.dev), names[i]);
351 err = request_irq(vp_dev->msix_entries[msix_vec].vector,
353 vp_dev->msix_names[msix_vec],
367 /* the config->find_vqs() implementation */
368 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
369 struct virtqueue *vqs[],
370 vq_callback_t *callbacks[],
375 /* Try MSI-X with one vector per queue. */
376 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names, true, true);
379 /* Fallback: MSI-X with one vector for config, one shared for queues. */
380 err = vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
384 /* Finally fall back to regular interrupts. */
385 return vp_try_to_find_vqs(vdev, nvqs, vqs, callbacks, names,
389 const char *vp_bus_name(struct virtio_device *vdev)
391 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
393 return pci_name(vp_dev->pci_dev);
396 /* Setup the affinity for a virtqueue:
397 * - force the affinity for per vq vector
398 * - OR over all affinities for shared MSI
399 * - ignore the affinity request if we're using INTX
401 int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
403 struct virtio_device *vdev = vq->vdev;
404 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
405 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
406 struct cpumask *mask;
412 if (vp_dev->msix_enabled) {
413 mask = vp_dev->msix_affinity_masks[info->msix_vector];
414 irq = vp_dev->msix_entries[info->msix_vector].vector;
416 irq_set_affinity_hint(irq, NULL);
418 cpumask_set_cpu(cpu, mask);
419 irq_set_affinity_hint(irq, mask);
425 #ifdef CONFIG_PM_SLEEP
426 static int virtio_pci_freeze(struct device *dev)
428 struct pci_dev *pci_dev = to_pci_dev(dev);
429 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
432 ret = virtio_device_freeze(&vp_dev->vdev);
435 pci_disable_device(pci_dev);
439 static int virtio_pci_restore(struct device *dev)
441 struct pci_dev *pci_dev = to_pci_dev(dev);
442 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
445 ret = pci_enable_device(pci_dev);
449 pci_set_master(pci_dev);
450 return virtio_device_restore(&vp_dev->vdev);
453 static const struct dev_pm_ops virtio_pci_pm_ops = {
454 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
459 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
460 static const struct pci_device_id virtio_pci_id_table[] = {
461 { PCI_DEVICE(0x1af4, PCI_ANY_ID) },
465 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
467 static int virtio_pci_probe(struct pci_dev *pci_dev,
468 const struct pci_device_id *id)
470 return virtio_pci_legacy_probe(pci_dev, id);
473 static void virtio_pci_remove(struct pci_dev *pci_dev)
475 virtio_pci_legacy_remove(pci_dev);
478 static struct pci_driver virtio_pci_driver = {
479 .name = "virtio-pci",
480 .id_table = virtio_pci_id_table,
481 .probe = virtio_pci_probe,
482 .remove = virtio_pci_remove,
483 #ifdef CONFIG_PM_SLEEP
484 .driver.pm = &virtio_pci_pm_ops,
488 module_pci_driver(virtio_pci_driver);
490 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
491 MODULE_DESCRIPTION("virtio-pci");
492 MODULE_LICENSE("GPL");