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 static bool force_legacy = false;
24 #if IS_ENABLED(CONFIG_VIRTIO_PCI_LEGACY)
25 module_param(force_legacy, bool, 0444);
26 MODULE_PARM_DESC(force_legacy,
27 "Force legacy mode for transitional virtio 1 devices");
30 /* wait for pending irq handlers */
31 void vp_synchronize_vectors(struct virtio_device *vdev)
33 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
36 if (vp_dev->intx_enabled)
37 synchronize_irq(vp_dev->pci_dev->irq);
39 for (i = 0; i < vp_dev->msix_vectors; ++i)
40 synchronize_irq(pci_irq_vector(vp_dev->pci_dev, i));
43 /* the notify function used when creating a virt queue */
44 bool vp_notify(struct virtqueue *vq)
46 /* we write the queue's selector into the notification register to
47 * signal the other end */
48 iowrite16(vq->index, (void __iomem *)vq->priv);
52 /* Handle a configuration change: Tell driver if it wants to know. */
53 static irqreturn_t vp_config_changed(int irq, void *opaque)
55 struct virtio_pci_device *vp_dev = opaque;
57 virtio_config_changed(&vp_dev->vdev);
61 /* Notify all virtqueues on an interrupt. */
62 static irqreturn_t vp_vring_interrupt(int irq, void *opaque)
64 struct virtio_pci_device *vp_dev = opaque;
65 struct virtio_pci_vq_info *info;
66 irqreturn_t ret = IRQ_NONE;
69 spin_lock_irqsave(&vp_dev->lock, flags);
70 list_for_each_entry(info, &vp_dev->virtqueues, node) {
71 if (vring_interrupt(irq, info->vq) == IRQ_HANDLED)
74 spin_unlock_irqrestore(&vp_dev->lock, flags);
79 /* A small wrapper to also acknowledge the interrupt when it's handled.
80 * I really need an EIO hook for the vring so I can ack the interrupt once we
81 * know that we'll be handling the IRQ but before we invoke the callback since
82 * the callback may notify the host which results in the host attempting to
83 * raise an interrupt that we would then mask once we acknowledged the
85 static irqreturn_t vp_interrupt(int irq, void *opaque)
87 struct virtio_pci_device *vp_dev = opaque;
90 /* reading the ISR has the effect of also clearing it so it's very
91 * important to save off the value. */
92 isr = ioread8(vp_dev->isr);
94 /* It's definitely not us if the ISR was not high */
98 /* Configuration change? Tell driver if it wants to know. */
99 if (isr & VIRTIO_PCI_ISR_CONFIG)
100 vp_config_changed(irq, opaque);
102 return vp_vring_interrupt(irq, opaque);
105 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
106 bool per_vq_vectors, struct irq_affinity *desc)
108 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
109 const char *name = dev_name(&vp_dev->vdev.dev);
110 unsigned flags = PCI_IRQ_MSIX;
114 vp_dev->msix_vectors = nvectors;
116 vp_dev->msix_names = kmalloc_array(nvectors,
117 sizeof(*vp_dev->msix_names),
119 if (!vp_dev->msix_names)
121 vp_dev->msix_affinity_masks
122 = kcalloc(nvectors, sizeof(*vp_dev->msix_affinity_masks),
124 if (!vp_dev->msix_affinity_masks)
126 for (i = 0; i < nvectors; ++i)
127 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
132 flags |= PCI_IRQ_AFFINITY;
133 desc->pre_vectors++; /* virtio config vector */
136 err = pci_alloc_irq_vectors_affinity(vp_dev->pci_dev, nvectors,
137 nvectors, flags, desc);
140 vp_dev->msix_enabled = 1;
142 /* Set the vector used for configuration */
143 v = vp_dev->msix_used_vectors;
144 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
146 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
147 vp_config_changed, 0, vp_dev->msix_names[v],
151 ++vp_dev->msix_used_vectors;
153 v = vp_dev->config_vector(vp_dev, v);
154 /* Verify we had enough resources to assign the vector */
155 if (v == VIRTIO_MSI_NO_VECTOR) {
160 if (!per_vq_vectors) {
161 /* Shared vector for all VQs */
162 v = vp_dev->msix_used_vectors;
163 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
164 "%s-virtqueues", name);
165 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
166 vp_vring_interrupt, 0, vp_dev->msix_names[v],
170 ++vp_dev->msix_used_vectors;
177 static struct virtqueue *vp_setup_vq(struct virtio_device *vdev, unsigned index,
178 void (*callback)(struct virtqueue *vq),
183 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
184 struct virtio_pci_vq_info *info = kmalloc(sizeof *info, GFP_KERNEL);
185 struct virtqueue *vq;
188 /* fill out our structure that represents an active queue */
190 return ERR_PTR(-ENOMEM);
192 vq = vp_dev->setup_vq(vp_dev, info, index, callback, name, ctx,
199 spin_lock_irqsave(&vp_dev->lock, flags);
200 list_add(&info->node, &vp_dev->virtqueues);
201 spin_unlock_irqrestore(&vp_dev->lock, flags);
203 INIT_LIST_HEAD(&info->node);
206 vp_dev->vqs[index] = info;
214 static void vp_del_vq(struct virtqueue *vq)
216 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
217 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
220 spin_lock_irqsave(&vp_dev->lock, flags);
221 list_del(&info->node);
222 spin_unlock_irqrestore(&vp_dev->lock, flags);
224 vp_dev->del_vq(info);
228 /* the config->del_vqs() implementation */
229 void vp_del_vqs(struct virtio_device *vdev)
231 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
232 struct virtqueue *vq, *n;
235 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
236 if (vp_dev->per_vq_vectors) {
237 int v = vp_dev->vqs[vq->index]->msix_vector;
239 if (v != VIRTIO_MSI_NO_VECTOR) {
240 int irq = pci_irq_vector(vp_dev->pci_dev, v);
242 irq_set_affinity_hint(irq, NULL);
248 vp_dev->per_vq_vectors = false;
250 if (vp_dev->intx_enabled) {
251 free_irq(vp_dev->pci_dev->irq, vp_dev);
252 vp_dev->intx_enabled = 0;
255 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
256 free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
258 if (vp_dev->msix_affinity_masks) {
259 for (i = 0; i < vp_dev->msix_vectors; i++)
260 if (vp_dev->msix_affinity_masks[i])
261 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
264 if (vp_dev->msix_enabled) {
265 /* Disable the vector used for configuration */
266 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
268 pci_free_irq_vectors(vp_dev->pci_dev);
269 vp_dev->msix_enabled = 0;
272 vp_dev->msix_vectors = 0;
273 vp_dev->msix_used_vectors = 0;
274 kfree(vp_dev->msix_names);
275 vp_dev->msix_names = NULL;
276 kfree(vp_dev->msix_affinity_masks);
277 vp_dev->msix_affinity_masks = NULL;
282 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
283 struct virtqueue *vqs[], vq_callback_t *callbacks[],
284 const char * const names[], bool per_vq_vectors,
286 struct irq_affinity *desc)
288 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
290 int i, err, nvectors, allocated_vectors;
292 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
296 if (per_vq_vectors) {
297 /* Best option: one for change interrupt, one per vq. */
299 for (i = 0; i < nvqs; ++i)
303 /* Second best: one for change, shared for all vqs. */
307 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors,
308 per_vq_vectors ? desc : NULL);
312 vp_dev->per_vq_vectors = per_vq_vectors;
313 allocated_vectors = vp_dev->msix_used_vectors;
314 for (i = 0; i < nvqs; ++i) {
321 msix_vec = VIRTIO_MSI_NO_VECTOR;
322 else if (vp_dev->per_vq_vectors)
323 msix_vec = allocated_vectors++;
325 msix_vec = VP_MSIX_VQ_VECTOR;
326 vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i],
327 ctx ? ctx[i] : false,
329 if (IS_ERR(vqs[i])) {
330 err = PTR_ERR(vqs[i]);
334 if (!vp_dev->per_vq_vectors || msix_vec == VIRTIO_MSI_NO_VECTOR)
337 /* allocate per-vq irq if available and necessary */
338 snprintf(vp_dev->msix_names[msix_vec],
339 sizeof *vp_dev->msix_names,
341 dev_name(&vp_dev->vdev.dev), names[i]);
342 err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
344 vp_dev->msix_names[msix_vec],
356 static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned nvqs,
357 struct virtqueue *vqs[], vq_callback_t *callbacks[],
358 const char * const names[], const bool *ctx)
360 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
363 vp_dev->vqs = kcalloc(nvqs, sizeof(*vp_dev->vqs), GFP_KERNEL);
367 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
368 dev_name(&vdev->dev), vp_dev);
372 vp_dev->intx_enabled = 1;
373 vp_dev->per_vq_vectors = false;
374 for (i = 0; i < nvqs; ++i) {
379 vqs[i] = vp_setup_vq(vdev, i, callbacks[i], names[i],
380 ctx ? ctx[i] : false,
381 VIRTIO_MSI_NO_VECTOR);
382 if (IS_ERR(vqs[i])) {
383 err = PTR_ERR(vqs[i]);
394 /* the config->find_vqs() implementation */
395 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
396 struct virtqueue *vqs[], vq_callback_t *callbacks[],
397 const char * const names[], const bool *ctx,
398 struct irq_affinity *desc)
402 /* Try MSI-X with one vector per queue. */
403 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true, ctx, desc);
406 /* Fallback: MSI-X with one vector for config, one shared for queues. */
407 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false, ctx, desc);
410 /* Finally fall back to regular interrupts. */
411 return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names, ctx);
414 const char *vp_bus_name(struct virtio_device *vdev)
416 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
418 return pci_name(vp_dev->pci_dev);
421 /* Setup the affinity for a virtqueue:
422 * - force the affinity for per vq vector
423 * - OR over all affinities for shared MSI
424 * - ignore the affinity request if we're using INTX
426 int vp_set_vq_affinity(struct virtqueue *vq, const struct cpumask *cpu_mask)
428 struct virtio_device *vdev = vq->vdev;
429 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
430 struct virtio_pci_vq_info *info = vp_dev->vqs[vq->index];
431 struct cpumask *mask;
437 if (vp_dev->msix_enabled) {
438 mask = vp_dev->msix_affinity_masks[info->msix_vector];
439 irq = pci_irq_vector(vp_dev->pci_dev, info->msix_vector);
441 irq_set_affinity_hint(irq, NULL);
443 cpumask_copy(mask, cpu_mask);
444 irq_set_affinity_hint(irq, mask);
450 const struct cpumask *vp_get_vq_affinity(struct virtio_device *vdev, int index)
452 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
454 if (!vp_dev->per_vq_vectors ||
455 vp_dev->vqs[index]->msix_vector == VIRTIO_MSI_NO_VECTOR)
458 return pci_irq_get_affinity(vp_dev->pci_dev,
459 vp_dev->vqs[index]->msix_vector);
462 #ifdef CONFIG_PM_SLEEP
463 static int virtio_pci_freeze(struct device *dev)
465 struct pci_dev *pci_dev = to_pci_dev(dev);
466 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
469 ret = virtio_device_freeze(&vp_dev->vdev);
472 pci_disable_device(pci_dev);
476 static int virtio_pci_restore(struct device *dev)
478 struct pci_dev *pci_dev = to_pci_dev(dev);
479 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
482 ret = pci_enable_device(pci_dev);
486 pci_set_master(pci_dev);
487 return virtio_device_restore(&vp_dev->vdev);
490 static const struct dev_pm_ops virtio_pci_pm_ops = {
491 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
496 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
497 static const struct pci_device_id virtio_pci_id_table[] = {
498 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
502 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
504 static void virtio_pci_release_dev(struct device *_d)
506 struct virtio_device *vdev = dev_to_virtio(_d);
507 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
509 /* As struct device is a kobject, it's not safe to
510 * free the memory (including the reference counter itself)
511 * until it's release callback. */
515 static int virtio_pci_probe(struct pci_dev *pci_dev,
516 const struct pci_device_id *id)
518 struct virtio_pci_device *vp_dev, *reg_dev = NULL;
521 /* allocate our structure and fill it out */
522 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
526 pci_set_drvdata(pci_dev, vp_dev);
527 vp_dev->vdev.dev.parent = &pci_dev->dev;
528 vp_dev->vdev.dev.release = virtio_pci_release_dev;
529 vp_dev->pci_dev = pci_dev;
530 INIT_LIST_HEAD(&vp_dev->virtqueues);
531 spin_lock_init(&vp_dev->lock);
533 /* enable the device */
534 rc = pci_enable_device(pci_dev);
536 goto err_enable_device;
539 rc = virtio_pci_legacy_probe(vp_dev);
540 /* Also try modern mode if we can't map BAR0 (no IO space). */
541 if (rc == -ENODEV || rc == -ENOMEM)
542 rc = virtio_pci_modern_probe(vp_dev);
546 rc = virtio_pci_modern_probe(vp_dev);
548 rc = virtio_pci_legacy_probe(vp_dev);
553 pci_set_master(pci_dev);
555 rc = register_virtio_device(&vp_dev->vdev);
564 virtio_pci_legacy_remove(vp_dev);
566 virtio_pci_modern_remove(vp_dev);
568 pci_disable_device(pci_dev);
571 put_device(&vp_dev->vdev.dev);
577 static void virtio_pci_remove(struct pci_dev *pci_dev)
579 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
580 struct device *dev = get_device(&vp_dev->vdev.dev);
583 * Device is marked broken on surprise removal so that virtio upper
584 * layers can abort any ongoing operation.
586 if (!pci_device_is_present(pci_dev))
587 virtio_break_device(&vp_dev->vdev);
589 pci_disable_sriov(pci_dev);
591 unregister_virtio_device(&vp_dev->vdev);
594 virtio_pci_legacy_remove(vp_dev);
596 virtio_pci_modern_remove(vp_dev);
598 pci_disable_device(pci_dev);
602 static int virtio_pci_sriov_configure(struct pci_dev *pci_dev, int num_vfs)
604 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
605 struct virtio_device *vdev = &vp_dev->vdev;
608 if (!(vdev->config->get_status(vdev) & VIRTIO_CONFIG_S_DRIVER_OK))
611 if (!__virtio_test_bit(vdev, VIRTIO_F_SR_IOV))
614 if (pci_vfs_assigned(pci_dev))
618 pci_disable_sriov(pci_dev);
622 ret = pci_enable_sriov(pci_dev, num_vfs);
629 static struct pci_driver virtio_pci_driver = {
630 .name = "virtio-pci",
631 .id_table = virtio_pci_id_table,
632 .probe = virtio_pci_probe,
633 .remove = virtio_pci_remove,
634 #ifdef CONFIG_PM_SLEEP
635 .driver.pm = &virtio_pci_pm_ops,
637 .sriov_configure = virtio_pci_sriov_configure,
640 module_pci_driver(virtio_pci_driver);
642 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
643 MODULE_DESCRIPTION("virtio-pci");
644 MODULE_LICENSE("GPL");