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 irqreturn_t ret = IRQ_NONE;
68 list_for_each_entry(vq, &vp_dev->vdev.vqs, list) {
69 if (vq->callback && vring_interrupt(irq, vq) == IRQ_HANDLED)
76 /* A small wrapper to also acknowledge the interrupt when it's handled.
77 * I really need an EIO hook for the vring so I can ack the interrupt once we
78 * know that we'll be handling the IRQ but before we invoke the callback since
79 * the callback may notify the host which results in the host attempting to
80 * raise an interrupt that we would then mask once we acknowledged the
82 static irqreturn_t vp_interrupt(int irq, void *opaque)
84 struct virtio_pci_device *vp_dev = opaque;
87 /* reading the ISR has the effect of also clearing it so it's very
88 * important to save off the value. */
89 isr = ioread8(vp_dev->isr);
91 /* It's definitely not us if the ISR was not high */
95 /* Configuration change? Tell driver if it wants to know. */
96 if (isr & VIRTIO_PCI_ISR_CONFIG)
97 vp_config_changed(irq, opaque);
99 return vp_vring_interrupt(irq, opaque);
102 static int vp_request_msix_vectors(struct virtio_device *vdev, int nvectors,
105 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
106 const char *name = dev_name(&vp_dev->vdev.dev);
110 vp_dev->msix_vectors = nvectors;
112 vp_dev->msix_names = kmalloc(nvectors * sizeof *vp_dev->msix_names,
114 if (!vp_dev->msix_names)
116 vp_dev->msix_affinity_masks
117 = kzalloc(nvectors * sizeof *vp_dev->msix_affinity_masks,
119 if (!vp_dev->msix_affinity_masks)
121 for (i = 0; i < nvectors; ++i)
122 if (!alloc_cpumask_var(&vp_dev->msix_affinity_masks[i],
126 err = pci_alloc_irq_vectors(vp_dev->pci_dev, nvectors, nvectors,
130 vp_dev->msix_enabled = 1;
132 /* Set the vector used for configuration */
133 v = vp_dev->msix_used_vectors;
134 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
136 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
137 vp_config_changed, 0, vp_dev->msix_names[v],
141 ++vp_dev->msix_used_vectors;
143 v = vp_dev->config_vector(vp_dev, v);
144 /* Verify we had enough resources to assign the vector */
145 if (v == VIRTIO_MSI_NO_VECTOR) {
150 if (!per_vq_vectors) {
151 /* Shared vector for all VQs */
152 v = vp_dev->msix_used_vectors;
153 snprintf(vp_dev->msix_names[v], sizeof *vp_dev->msix_names,
154 "%s-virtqueues", name);
155 err = request_irq(pci_irq_vector(vp_dev->pci_dev, v),
156 vp_vring_interrupt, 0, vp_dev->msix_names[v],
160 ++vp_dev->msix_used_vectors;
167 /* the config->del_vqs() implementation */
168 void vp_del_vqs(struct virtio_device *vdev)
170 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
171 struct virtqueue *vq, *n;
174 list_for_each_entry_safe(vq, n, &vdev->vqs, list) {
175 if (vp_dev->msix_vector_map) {
176 int v = vp_dev->msix_vector_map[vq->index];
178 if (v != VIRTIO_MSI_NO_VECTOR)
179 free_irq(pci_irq_vector(vp_dev->pci_dev, v),
185 if (vp_dev->intx_enabled) {
186 free_irq(vp_dev->pci_dev->irq, vp_dev);
187 vp_dev->intx_enabled = 0;
190 for (i = 0; i < vp_dev->msix_used_vectors; ++i)
191 free_irq(pci_irq_vector(vp_dev->pci_dev, i), vp_dev);
193 for (i = 0; i < vp_dev->msix_vectors; i++)
194 if (vp_dev->msix_affinity_masks[i])
195 free_cpumask_var(vp_dev->msix_affinity_masks[i]);
197 if (vp_dev->msix_enabled) {
198 /* Disable the vector used for configuration */
199 vp_dev->config_vector(vp_dev, VIRTIO_MSI_NO_VECTOR);
201 pci_free_irq_vectors(vp_dev->pci_dev);
202 vp_dev->msix_enabled = 0;
205 vp_dev->msix_vectors = 0;
206 vp_dev->msix_used_vectors = 0;
207 kfree(vp_dev->msix_names);
208 vp_dev->msix_names = NULL;
209 kfree(vp_dev->msix_affinity_masks);
210 vp_dev->msix_affinity_masks = NULL;
211 kfree(vp_dev->msix_vector_map);
212 vp_dev->msix_vector_map = NULL;
215 static int vp_find_vqs_msix(struct virtio_device *vdev, unsigned nvqs,
216 struct virtqueue *vqs[],
217 vq_callback_t *callbacks[],
218 const char * const names[],
221 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
223 int i, err, nvectors, allocated_vectors;
225 if (per_vq_vectors) {
226 /* Best option: one for change interrupt, one per vq. */
228 for (i = 0; i < nvqs; ++i)
232 /* Second best: one for change, shared for all vqs. */
236 err = vp_request_msix_vectors(vdev, nvectors, per_vq_vectors);
240 if (per_vq_vectors) {
241 vp_dev->msix_vector_map = kmalloc_array(nvqs,
242 sizeof(*vp_dev->msix_vector_map), GFP_KERNEL);
243 if (!vp_dev->msix_vector_map)
247 allocated_vectors = vp_dev->msix_used_vectors;
248 for (i = 0; i < nvqs; ++i) {
255 msix_vec = VIRTIO_MSI_NO_VECTOR;
256 else if (per_vq_vectors)
257 msix_vec = allocated_vectors++;
259 msix_vec = VP_MSIX_VQ_VECTOR;
260 vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
262 if (IS_ERR(vqs[i])) {
263 err = PTR_ERR(vqs[i]);
270 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
271 vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
275 /* allocate per-vq irq if available and necessary */
276 snprintf(vp_dev->msix_names[msix_vec],
277 sizeof *vp_dev->msix_names,
279 dev_name(&vp_dev->vdev.dev), names[i]);
280 err = request_irq(pci_irq_vector(vp_dev->pci_dev, msix_vec),
282 vp_dev->msix_names[msix_vec],
285 /* don't free this irq on error */
286 vp_dev->msix_vector_map[i] = VIRTIO_MSI_NO_VECTOR;
289 vp_dev->msix_vector_map[i] = msix_vec;
298 static int vp_find_vqs_intx(struct virtio_device *vdev, unsigned nvqs,
299 struct virtqueue *vqs[], vq_callback_t *callbacks[],
300 const char * const names[])
302 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
305 err = request_irq(vp_dev->pci_dev->irq, vp_interrupt, IRQF_SHARED,
306 dev_name(&vdev->dev), vp_dev);
310 vp_dev->intx_enabled = 1;
311 for (i = 0; i < nvqs; ++i) {
316 vqs[i] = vp_dev->setup_vq(vp_dev, i, callbacks[i], names[i],
317 VIRTIO_MSI_NO_VECTOR);
318 if (IS_ERR(vqs[i])) {
319 err = PTR_ERR(vqs[i]);
330 /* the config->find_vqs() implementation */
331 int vp_find_vqs(struct virtio_device *vdev, unsigned nvqs,
332 struct virtqueue *vqs[],
333 vq_callback_t *callbacks[],
334 const char * const names[])
338 /* Try MSI-X with one vector per queue. */
339 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, true);
342 /* Fallback: MSI-X with one vector for config, one shared for queues. */
343 err = vp_find_vqs_msix(vdev, nvqs, vqs, callbacks, names, false);
346 /* Finally fall back to regular interrupts. */
347 return vp_find_vqs_intx(vdev, nvqs, vqs, callbacks, names);
350 const char *vp_bus_name(struct virtio_device *vdev)
352 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
354 return pci_name(vp_dev->pci_dev);
357 /* Setup the affinity for a virtqueue:
358 * - force the affinity for per vq vector
359 * - OR over all affinities for shared MSI
360 * - ignore the affinity request if we're using INTX
362 int vp_set_vq_affinity(struct virtqueue *vq, int cpu)
364 struct virtio_device *vdev = vq->vdev;
365 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
370 if (vp_dev->msix_enabled) {
371 int vec = vp_dev->msix_vector_map[vq->index];
372 struct cpumask *mask = vp_dev->msix_affinity_masks[vec];
373 unsigned int irq = pci_irq_vector(vp_dev->pci_dev, vec);
376 irq_set_affinity_hint(irq, NULL);
379 cpumask_set_cpu(cpu, mask);
380 irq_set_affinity_hint(irq, mask);
386 #ifdef CONFIG_PM_SLEEP
387 static int virtio_pci_freeze(struct device *dev)
389 struct pci_dev *pci_dev = to_pci_dev(dev);
390 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
393 ret = virtio_device_freeze(&vp_dev->vdev);
396 pci_disable_device(pci_dev);
400 static int virtio_pci_restore(struct device *dev)
402 struct pci_dev *pci_dev = to_pci_dev(dev);
403 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
406 ret = pci_enable_device(pci_dev);
410 pci_set_master(pci_dev);
411 return virtio_device_restore(&vp_dev->vdev);
414 static const struct dev_pm_ops virtio_pci_pm_ops = {
415 SET_SYSTEM_SLEEP_PM_OPS(virtio_pci_freeze, virtio_pci_restore)
420 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */
421 static const struct pci_device_id virtio_pci_id_table[] = {
422 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) },
426 MODULE_DEVICE_TABLE(pci, virtio_pci_id_table);
428 static void virtio_pci_release_dev(struct device *_d)
430 struct virtio_device *vdev = dev_to_virtio(_d);
431 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
433 /* As struct device is a kobject, it's not safe to
434 * free the memory (including the reference counter itself)
435 * until it's release callback. */
439 static int virtio_pci_probe(struct pci_dev *pci_dev,
440 const struct pci_device_id *id)
442 struct virtio_pci_device *vp_dev;
445 /* allocate our structure and fill it out */
446 vp_dev = kzalloc(sizeof(struct virtio_pci_device), GFP_KERNEL);
450 pci_set_drvdata(pci_dev, vp_dev);
451 vp_dev->vdev.dev.parent = &pci_dev->dev;
452 vp_dev->vdev.dev.release = virtio_pci_release_dev;
453 vp_dev->pci_dev = pci_dev;
455 /* enable the device */
456 rc = pci_enable_device(pci_dev);
458 goto err_enable_device;
461 rc = virtio_pci_legacy_probe(vp_dev);
462 /* Also try modern mode if we can't map BAR0 (no IO space). */
463 if (rc == -ENODEV || rc == -ENOMEM)
464 rc = virtio_pci_modern_probe(vp_dev);
468 rc = virtio_pci_modern_probe(vp_dev);
470 rc = virtio_pci_legacy_probe(vp_dev);
475 pci_set_master(pci_dev);
477 rc = register_virtio_device(&vp_dev->vdev);
485 virtio_pci_legacy_remove(vp_dev);
487 virtio_pci_modern_remove(vp_dev);
489 pci_disable_device(pci_dev);
495 static void virtio_pci_remove(struct pci_dev *pci_dev)
497 struct virtio_pci_device *vp_dev = pci_get_drvdata(pci_dev);
498 struct device *dev = get_device(&vp_dev->vdev.dev);
500 unregister_virtio_device(&vp_dev->vdev);
503 virtio_pci_legacy_remove(vp_dev);
505 virtio_pci_modern_remove(vp_dev);
507 pci_disable_device(pci_dev);
511 static struct pci_driver virtio_pci_driver = {
512 .name = "virtio-pci",
513 .id_table = virtio_pci_id_table,
514 .probe = virtio_pci_probe,
515 .remove = virtio_pci_remove,
516 #ifdef CONFIG_PM_SLEEP
517 .driver.pm = &virtio_pci_pm_ops,
521 module_pci_driver(virtio_pci_driver);
523 MODULE_AUTHOR("Anthony Liguori <aliguori@us.ibm.com>");
524 MODULE_DESCRIPTION("virtio-pci");
525 MODULE_LICENSE("GPL");