2 * Kernel-based Virtual Machine - device assignment support
4 * Copyright (C) 2010 Red Hat, Inc. and/or its affiliates.
6 * This work is licensed under the terms of the GNU GPL, version 2. See
7 * the COPYING file in the top-level directory.
11 #include <linux/kvm_host.h>
12 #include <linux/kvm.h>
13 #include <linux/uaccess.h>
14 #include <linux/vmalloc.h>
15 #include <linux/errno.h>
16 #include <linux/spinlock.h>
17 #include <linux/pci.h>
18 #include <linux/interrupt.h>
19 #include <linux/slab.h>
22 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
25 struct list_head *ptr;
26 struct kvm_assigned_dev_kernel *match;
28 list_for_each(ptr, head) {
29 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
30 if (match->assigned_dev_id == assigned_dev_id)
36 static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
37 *assigned_dev, int irq)
40 struct msix_entry *host_msix_entries;
42 host_msix_entries = assigned_dev->host_msix_entries;
45 for (i = 0; i < assigned_dev->entries_nr; i++)
46 if (irq == host_msix_entries[i].vector) {
51 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
58 static irqreturn_t kvm_assigned_dev_thread(int irq, void *dev_id)
60 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
62 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_INTX) {
63 spin_lock(&assigned_dev->intx_lock);
64 disable_irq_nosync(irq);
65 assigned_dev->host_irq_disabled = true;
66 spin_unlock(&assigned_dev->intx_lock);
69 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
70 assigned_dev->guest_irq, 1);
75 #ifdef __KVM_HAVE_MSIX
76 static irqreturn_t kvm_assigned_dev_thread_msix(int irq, void *dev_id)
78 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
79 int index = find_index_from_host_irq(assigned_dev, irq);
83 vector = assigned_dev->guest_msix_entries[index].vector;
84 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
92 /* Ack the irq line for an assigned device */
93 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
95 struct kvm_assigned_dev_kernel *dev =
96 container_of(kian, struct kvm_assigned_dev_kernel,
99 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
101 /* The guest irq may be shared so this ack may be
102 * from another device.
104 spin_lock(&dev->intx_lock);
105 if (dev->host_irq_disabled) {
106 enable_irq(dev->host_irq);
107 dev->host_irq_disabled = false;
109 spin_unlock(&dev->intx_lock);
112 static void deassign_guest_irq(struct kvm *kvm,
113 struct kvm_assigned_dev_kernel *assigned_dev)
115 if (assigned_dev->ack_notifier.gsi != -1)
116 kvm_unregister_irq_ack_notifier(kvm,
117 &assigned_dev->ack_notifier);
119 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
120 assigned_dev->guest_irq, 0);
122 if (assigned_dev->irq_source_id != -1)
123 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
124 assigned_dev->irq_source_id = -1;
125 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
128 /* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
129 static void deassign_host_irq(struct kvm *kvm,
130 struct kvm_assigned_dev_kernel *assigned_dev)
133 * We disable irq here to prevent further events.
135 * Notice this maybe result in nested disable if the interrupt type is
136 * INTx, but it's OK for we are going to free it.
138 * If this function is a part of VM destroy, please ensure that till
139 * now, the kvm state is still legal for probably we also have to wait
140 * on a currently running IRQ handler.
142 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
144 for (i = 0; i < assigned_dev->entries_nr; i++)
145 disable_irq(assigned_dev->host_msix_entries[i].vector);
147 for (i = 0; i < assigned_dev->entries_nr; i++)
148 free_irq(assigned_dev->host_msix_entries[i].vector,
151 assigned_dev->entries_nr = 0;
152 kfree(assigned_dev->host_msix_entries);
153 kfree(assigned_dev->guest_msix_entries);
154 pci_disable_msix(assigned_dev->dev);
156 /* Deal with MSI and INTx */
157 disable_irq(assigned_dev->host_irq);
159 free_irq(assigned_dev->host_irq, assigned_dev);
161 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
162 pci_disable_msi(assigned_dev->dev);
165 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
168 static int kvm_deassign_irq(struct kvm *kvm,
169 struct kvm_assigned_dev_kernel *assigned_dev,
170 unsigned long irq_requested_type)
172 unsigned long guest_irq_type, host_irq_type;
174 if (!irqchip_in_kernel(kvm))
176 /* no irq assignment to deassign */
177 if (!assigned_dev->irq_requested_type)
180 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
181 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
184 deassign_host_irq(kvm, assigned_dev);
186 deassign_guest_irq(kvm, assigned_dev);
191 static void kvm_free_assigned_irq(struct kvm *kvm,
192 struct kvm_assigned_dev_kernel *assigned_dev)
194 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
197 static void kvm_free_assigned_device(struct kvm *kvm,
198 struct kvm_assigned_dev_kernel
201 kvm_free_assigned_irq(kvm, assigned_dev);
203 pci_reset_function(assigned_dev->dev);
204 if (pci_load_and_free_saved_state(assigned_dev->dev,
205 &assigned_dev->pci_saved_state))
206 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
207 __func__, dev_name(&assigned_dev->dev->dev));
209 pci_restore_state(assigned_dev->dev);
211 pci_release_regions(assigned_dev->dev);
212 pci_disable_device(assigned_dev->dev);
213 pci_dev_put(assigned_dev->dev);
215 list_del(&assigned_dev->list);
219 void kvm_free_all_assigned_devices(struct kvm *kvm)
221 struct list_head *ptr, *ptr2;
222 struct kvm_assigned_dev_kernel *assigned_dev;
224 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
225 assigned_dev = list_entry(ptr,
226 struct kvm_assigned_dev_kernel,
229 kvm_free_assigned_device(kvm, assigned_dev);
233 static int assigned_device_enable_host_intx(struct kvm *kvm,
234 struct kvm_assigned_dev_kernel *dev)
236 dev->host_irq = dev->dev->irq;
237 /* Even though this is PCI, we don't want to use shared
238 * interrupts. Sharing host devices with guest-assigned devices
239 * on the same interrupt line is not a happy situation: there
240 * are going to be long delays in accepting, acking, etc.
242 if (request_threaded_irq(dev->host_irq, NULL, kvm_assigned_dev_thread,
243 IRQF_ONESHOT, dev->irq_name, dev))
248 #ifdef __KVM_HAVE_MSI
249 static int assigned_device_enable_host_msi(struct kvm *kvm,
250 struct kvm_assigned_dev_kernel *dev)
254 if (!dev->dev->msi_enabled) {
255 r = pci_enable_msi(dev->dev);
260 dev->host_irq = dev->dev->irq;
261 if (request_threaded_irq(dev->host_irq, NULL, kvm_assigned_dev_thread,
262 0, dev->irq_name, dev)) {
263 pci_disable_msi(dev->dev);
271 #ifdef __KVM_HAVE_MSIX
272 static int assigned_device_enable_host_msix(struct kvm *kvm,
273 struct kvm_assigned_dev_kernel *dev)
277 /* host_msix_entries and guest_msix_entries should have been
279 if (dev->entries_nr == 0)
282 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
286 for (i = 0; i < dev->entries_nr; i++) {
287 r = request_threaded_irq(dev->host_msix_entries[i].vector,
288 NULL, kvm_assigned_dev_thread_msix,
289 0, dev->irq_name, dev);
296 for (i -= 1; i >= 0; i--)
297 free_irq(dev->host_msix_entries[i].vector, dev);
298 pci_disable_msix(dev->dev);
304 static int assigned_device_enable_guest_intx(struct kvm *kvm,
305 struct kvm_assigned_dev_kernel *dev,
306 struct kvm_assigned_irq *irq)
308 dev->guest_irq = irq->guest_irq;
309 dev->ack_notifier.gsi = irq->guest_irq;
313 #ifdef __KVM_HAVE_MSI
314 static int assigned_device_enable_guest_msi(struct kvm *kvm,
315 struct kvm_assigned_dev_kernel *dev,
316 struct kvm_assigned_irq *irq)
318 dev->guest_irq = irq->guest_irq;
319 dev->ack_notifier.gsi = -1;
320 dev->host_irq_disabled = false;
325 #ifdef __KVM_HAVE_MSIX
326 static int assigned_device_enable_guest_msix(struct kvm *kvm,
327 struct kvm_assigned_dev_kernel *dev,
328 struct kvm_assigned_irq *irq)
330 dev->guest_irq = irq->guest_irq;
331 dev->ack_notifier.gsi = -1;
332 dev->host_irq_disabled = false;
337 static int assign_host_irq(struct kvm *kvm,
338 struct kvm_assigned_dev_kernel *dev,
343 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
346 snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
349 switch (host_irq_type) {
350 case KVM_DEV_IRQ_HOST_INTX:
351 r = assigned_device_enable_host_intx(kvm, dev);
353 #ifdef __KVM_HAVE_MSI
354 case KVM_DEV_IRQ_HOST_MSI:
355 r = assigned_device_enable_host_msi(kvm, dev);
358 #ifdef __KVM_HAVE_MSIX
359 case KVM_DEV_IRQ_HOST_MSIX:
360 r = assigned_device_enable_host_msix(kvm, dev);
368 dev->irq_requested_type |= host_irq_type;
373 static int assign_guest_irq(struct kvm *kvm,
374 struct kvm_assigned_dev_kernel *dev,
375 struct kvm_assigned_irq *irq,
376 unsigned long guest_irq_type)
381 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
384 id = kvm_request_irq_source_id(kvm);
388 dev->irq_source_id = id;
390 switch (guest_irq_type) {
391 case KVM_DEV_IRQ_GUEST_INTX:
392 r = assigned_device_enable_guest_intx(kvm, dev, irq);
394 #ifdef __KVM_HAVE_MSI
395 case KVM_DEV_IRQ_GUEST_MSI:
396 r = assigned_device_enable_guest_msi(kvm, dev, irq);
399 #ifdef __KVM_HAVE_MSIX
400 case KVM_DEV_IRQ_GUEST_MSIX:
401 r = assigned_device_enable_guest_msix(kvm, dev, irq);
409 dev->irq_requested_type |= guest_irq_type;
410 if (dev->ack_notifier.gsi != -1)
411 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
413 kvm_free_irq_source_id(kvm, dev->irq_source_id);
418 /* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
419 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
420 struct kvm_assigned_irq *assigned_irq)
423 struct kvm_assigned_dev_kernel *match;
424 unsigned long host_irq_type, guest_irq_type;
426 if (!irqchip_in_kernel(kvm))
429 mutex_lock(&kvm->lock);
431 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
432 assigned_irq->assigned_dev_id);
436 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
437 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
440 /* can only assign one type at a time */
441 if (hweight_long(host_irq_type) > 1)
443 if (hweight_long(guest_irq_type) > 1)
445 if (host_irq_type == 0 && guest_irq_type == 0)
450 r = assign_host_irq(kvm, match, host_irq_type);
455 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
457 mutex_unlock(&kvm->lock);
461 static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
462 struct kvm_assigned_irq
466 struct kvm_assigned_dev_kernel *match;
468 mutex_lock(&kvm->lock);
470 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
471 assigned_irq->assigned_dev_id);
475 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
477 mutex_unlock(&kvm->lock);
481 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
482 struct kvm_assigned_pci_dev *assigned_dev)
485 struct kvm_assigned_dev_kernel *match;
488 mutex_lock(&kvm->lock);
489 idx = srcu_read_lock(&kvm->srcu);
491 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
492 assigned_dev->assigned_dev_id);
494 /* device already assigned */
499 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
501 printk(KERN_INFO "%s: Couldn't allocate memory\n",
506 dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
508 assigned_dev->devfn);
510 printk(KERN_INFO "%s: host device not found\n", __func__);
514 if (pci_enable_device(dev)) {
515 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
519 r = pci_request_regions(dev, "kvm_assigned_device");
521 printk(KERN_INFO "%s: Could not get access to device regions\n",
526 pci_reset_function(dev);
528 match->pci_saved_state = pci_store_saved_state(dev);
529 if (!match->pci_saved_state)
530 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
531 __func__, dev_name(&dev->dev));
532 match->assigned_dev_id = assigned_dev->assigned_dev_id;
533 match->host_segnr = assigned_dev->segnr;
534 match->host_busnr = assigned_dev->busnr;
535 match->host_devfn = assigned_dev->devfn;
536 match->flags = assigned_dev->flags;
538 spin_lock_init(&match->intx_lock);
539 match->irq_source_id = -1;
541 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
543 list_add(&match->list, &kvm->arch.assigned_dev_head);
545 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
546 if (!kvm->arch.iommu_domain) {
547 r = kvm_iommu_map_guest(kvm);
551 r = kvm_assign_device(kvm, match);
557 srcu_read_unlock(&kvm->srcu, idx);
558 mutex_unlock(&kvm->lock);
561 if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
562 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
563 __func__, dev_name(&dev->dev));
564 list_del(&match->list);
565 pci_release_regions(dev);
567 pci_disable_device(dev);
572 srcu_read_unlock(&kvm->srcu, idx);
573 mutex_unlock(&kvm->lock);
577 static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
578 struct kvm_assigned_pci_dev *assigned_dev)
581 struct kvm_assigned_dev_kernel *match;
583 mutex_lock(&kvm->lock);
585 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
586 assigned_dev->assigned_dev_id);
588 printk(KERN_INFO "%s: device hasn't been assigned before, "
589 "so cannot be deassigned\n", __func__);
594 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
595 kvm_deassign_device(kvm, match);
597 kvm_free_assigned_device(kvm, match);
600 mutex_unlock(&kvm->lock);
605 #ifdef __KVM_HAVE_MSIX
606 static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
607 struct kvm_assigned_msix_nr *entry_nr)
610 struct kvm_assigned_dev_kernel *adev;
612 mutex_lock(&kvm->lock);
614 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
615 entry_nr->assigned_dev_id);
621 if (adev->entries_nr == 0) {
622 adev->entries_nr = entry_nr->entry_nr;
623 if (adev->entries_nr == 0 ||
624 adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
629 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
632 if (!adev->host_msix_entries) {
636 adev->guest_msix_entries =
637 kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
639 if (!adev->guest_msix_entries) {
640 kfree(adev->host_msix_entries);
644 } else /* Not allowed set MSI-X number twice */
647 mutex_unlock(&kvm->lock);
651 static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
652 struct kvm_assigned_msix_entry *entry)
655 struct kvm_assigned_dev_kernel *adev;
657 mutex_lock(&kvm->lock);
659 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
660 entry->assigned_dev_id);
667 for (i = 0; i < adev->entries_nr; i++)
668 if (adev->guest_msix_entries[i].vector == 0 ||
669 adev->guest_msix_entries[i].entry == entry->entry) {
670 adev->guest_msix_entries[i].entry = entry->entry;
671 adev->guest_msix_entries[i].vector = entry->gsi;
672 adev->host_msix_entries[i].entry = entry->entry;
675 if (i == adev->entries_nr) {
681 mutex_unlock(&kvm->lock);
687 long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
690 void __user *argp = (void __user *)arg;
694 case KVM_ASSIGN_PCI_DEVICE: {
695 struct kvm_assigned_pci_dev assigned_dev;
698 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
700 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
705 case KVM_ASSIGN_IRQ: {
709 case KVM_ASSIGN_DEV_IRQ: {
710 struct kvm_assigned_irq assigned_irq;
713 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
715 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
720 case KVM_DEASSIGN_DEV_IRQ: {
721 struct kvm_assigned_irq assigned_irq;
724 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
726 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
731 case KVM_DEASSIGN_PCI_DEVICE: {
732 struct kvm_assigned_pci_dev assigned_dev;
735 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
737 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
742 #ifdef KVM_CAP_IRQ_ROUTING
743 case KVM_SET_GSI_ROUTING: {
744 struct kvm_irq_routing routing;
745 struct kvm_irq_routing __user *urouting;
746 struct kvm_irq_routing_entry *entries;
749 if (copy_from_user(&routing, argp, sizeof(routing)))
752 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
757 entries = vmalloc(routing.nr * sizeof(*entries));
762 if (copy_from_user(entries, urouting->entries,
763 routing.nr * sizeof(*entries)))
764 goto out_free_irq_routing;
765 r = kvm_set_irq_routing(kvm, entries, routing.nr,
767 out_free_irq_routing:
771 #endif /* KVM_CAP_IRQ_ROUTING */
772 #ifdef __KVM_HAVE_MSIX
773 case KVM_ASSIGN_SET_MSIX_NR: {
774 struct kvm_assigned_msix_nr entry_nr;
776 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
778 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
783 case KVM_ASSIGN_SET_MSIX_ENTRY: {
784 struct kvm_assigned_msix_entry entry;
786 if (copy_from_user(&entry, argp, sizeof entry))
788 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);