Loading...
1/*
2 * Kernel-based Virtual Machine - device assignment support
3 *
4 * Copyright (C) 2010 Red Hat, Inc. and/or its affiliates.
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2. See
7 * the COPYING file in the top-level directory.
8 *
9 */
10
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>
20#include "irq.h"
21
22static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
23 int assigned_dev_id)
24{
25 struct list_head *ptr;
26 struct kvm_assigned_dev_kernel *match;
27
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)
31 return match;
32 }
33 return NULL;
34}
35
36static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
37 *assigned_dev, int irq)
38{
39 int i, index;
40 struct msix_entry *host_msix_entries;
41
42 host_msix_entries = assigned_dev->host_msix_entries;
43
44 index = -1;
45 for (i = 0; i < assigned_dev->entries_nr; i++)
46 if (irq == host_msix_entries[i].vector) {
47 index = i;
48 break;
49 }
50 if (index < 0) {
51 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
52 return 0;
53 }
54
55 return index;
56}
57
58static irqreturn_t kvm_assigned_dev_thread(int irq, void *dev_id)
59{
60 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
61 u32 vector;
62 int index;
63
64 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_INTX) {
65 spin_lock(&assigned_dev->intx_lock);
66 disable_irq_nosync(irq);
67 assigned_dev->host_irq_disabled = true;
68 spin_unlock(&assigned_dev->intx_lock);
69 }
70
71 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
72 index = find_index_from_host_irq(assigned_dev, irq);
73 if (index >= 0) {
74 vector = assigned_dev->
75 guest_msix_entries[index].vector;
76 kvm_set_irq(assigned_dev->kvm,
77 assigned_dev->irq_source_id, vector, 1);
78 }
79 } else
80 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
81 assigned_dev->guest_irq, 1);
82
83 return IRQ_HANDLED;
84}
85
86/* Ack the irq line for an assigned device */
87static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
88{
89 struct kvm_assigned_dev_kernel *dev;
90
91 if (kian->gsi == -1)
92 return;
93
94 dev = container_of(kian, struct kvm_assigned_dev_kernel,
95 ack_notifier);
96
97 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
98
99 /* The guest irq may be shared so this ack may be
100 * from another device.
101 */
102 spin_lock(&dev->intx_lock);
103 if (dev->host_irq_disabled) {
104 enable_irq(dev->host_irq);
105 dev->host_irq_disabled = false;
106 }
107 spin_unlock(&dev->intx_lock);
108}
109
110static void deassign_guest_irq(struct kvm *kvm,
111 struct kvm_assigned_dev_kernel *assigned_dev)
112{
113 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
114 assigned_dev->ack_notifier.gsi = -1;
115
116 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
117 assigned_dev->guest_irq, 0);
118
119 if (assigned_dev->irq_source_id != -1)
120 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
121 assigned_dev->irq_source_id = -1;
122 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
123}
124
125/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
126static void deassign_host_irq(struct kvm *kvm,
127 struct kvm_assigned_dev_kernel *assigned_dev)
128{
129 /*
130 * We disable irq here to prevent further events.
131 *
132 * Notice this maybe result in nested disable if the interrupt type is
133 * INTx, but it's OK for we are going to free it.
134 *
135 * If this function is a part of VM destroy, please ensure that till
136 * now, the kvm state is still legal for probably we also have to wait
137 * on a currently running IRQ handler.
138 */
139 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
140 int i;
141 for (i = 0; i < assigned_dev->entries_nr; i++)
142 disable_irq(assigned_dev->host_msix_entries[i].vector);
143
144 for (i = 0; i < assigned_dev->entries_nr; i++)
145 free_irq(assigned_dev->host_msix_entries[i].vector,
146 (void *)assigned_dev);
147
148 assigned_dev->entries_nr = 0;
149 kfree(assigned_dev->host_msix_entries);
150 kfree(assigned_dev->guest_msix_entries);
151 pci_disable_msix(assigned_dev->dev);
152 } else {
153 /* Deal with MSI and INTx */
154 disable_irq(assigned_dev->host_irq);
155
156 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
157
158 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
159 pci_disable_msi(assigned_dev->dev);
160 }
161
162 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
163}
164
165static int kvm_deassign_irq(struct kvm *kvm,
166 struct kvm_assigned_dev_kernel *assigned_dev,
167 unsigned long irq_requested_type)
168{
169 unsigned long guest_irq_type, host_irq_type;
170
171 if (!irqchip_in_kernel(kvm))
172 return -EINVAL;
173 /* no irq assignment to deassign */
174 if (!assigned_dev->irq_requested_type)
175 return -ENXIO;
176
177 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
178 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
179
180 if (host_irq_type)
181 deassign_host_irq(kvm, assigned_dev);
182 if (guest_irq_type)
183 deassign_guest_irq(kvm, assigned_dev);
184
185 return 0;
186}
187
188static void kvm_free_assigned_irq(struct kvm *kvm,
189 struct kvm_assigned_dev_kernel *assigned_dev)
190{
191 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
192}
193
194static void kvm_free_assigned_device(struct kvm *kvm,
195 struct kvm_assigned_dev_kernel
196 *assigned_dev)
197{
198 kvm_free_assigned_irq(kvm, assigned_dev);
199
200 pci_reset_function(assigned_dev->dev);
201 if (pci_load_and_free_saved_state(assigned_dev->dev,
202 &assigned_dev->pci_saved_state))
203 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
204 __func__, dev_name(&assigned_dev->dev->dev));
205 else
206 pci_restore_state(assigned_dev->dev);
207
208 pci_release_regions(assigned_dev->dev);
209 pci_disable_device(assigned_dev->dev);
210 pci_dev_put(assigned_dev->dev);
211
212 list_del(&assigned_dev->list);
213 kfree(assigned_dev);
214}
215
216void kvm_free_all_assigned_devices(struct kvm *kvm)
217{
218 struct list_head *ptr, *ptr2;
219 struct kvm_assigned_dev_kernel *assigned_dev;
220
221 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
222 assigned_dev = list_entry(ptr,
223 struct kvm_assigned_dev_kernel,
224 list);
225
226 kvm_free_assigned_device(kvm, assigned_dev);
227 }
228}
229
230static int assigned_device_enable_host_intx(struct kvm *kvm,
231 struct kvm_assigned_dev_kernel *dev)
232{
233 dev->host_irq = dev->dev->irq;
234 /* Even though this is PCI, we don't want to use shared
235 * interrupts. Sharing host devices with guest-assigned devices
236 * on the same interrupt line is not a happy situation: there
237 * are going to be long delays in accepting, acking, etc.
238 */
239 if (request_threaded_irq(dev->host_irq, NULL, kvm_assigned_dev_thread,
240 IRQF_ONESHOT, dev->irq_name, (void *)dev))
241 return -EIO;
242 return 0;
243}
244
245#ifdef __KVM_HAVE_MSI
246static int assigned_device_enable_host_msi(struct kvm *kvm,
247 struct kvm_assigned_dev_kernel *dev)
248{
249 int r;
250
251 if (!dev->dev->msi_enabled) {
252 r = pci_enable_msi(dev->dev);
253 if (r)
254 return r;
255 }
256
257 dev->host_irq = dev->dev->irq;
258 if (request_threaded_irq(dev->host_irq, NULL, kvm_assigned_dev_thread,
259 0, dev->irq_name, (void *)dev)) {
260 pci_disable_msi(dev->dev);
261 return -EIO;
262 }
263
264 return 0;
265}
266#endif
267
268#ifdef __KVM_HAVE_MSIX
269static int assigned_device_enable_host_msix(struct kvm *kvm,
270 struct kvm_assigned_dev_kernel *dev)
271{
272 int i, r = -EINVAL;
273
274 /* host_msix_entries and guest_msix_entries should have been
275 * initialized */
276 if (dev->entries_nr == 0)
277 return r;
278
279 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
280 if (r)
281 return r;
282
283 for (i = 0; i < dev->entries_nr; i++) {
284 r = request_threaded_irq(dev->host_msix_entries[i].vector,
285 NULL, kvm_assigned_dev_thread,
286 0, dev->irq_name, (void *)dev);
287 if (r)
288 goto err;
289 }
290
291 return 0;
292err:
293 for (i -= 1; i >= 0; i--)
294 free_irq(dev->host_msix_entries[i].vector, (void *)dev);
295 pci_disable_msix(dev->dev);
296 return r;
297}
298
299#endif
300
301static int assigned_device_enable_guest_intx(struct kvm *kvm,
302 struct kvm_assigned_dev_kernel *dev,
303 struct kvm_assigned_irq *irq)
304{
305 dev->guest_irq = irq->guest_irq;
306 dev->ack_notifier.gsi = irq->guest_irq;
307 return 0;
308}
309
310#ifdef __KVM_HAVE_MSI
311static int assigned_device_enable_guest_msi(struct kvm *kvm,
312 struct kvm_assigned_dev_kernel *dev,
313 struct kvm_assigned_irq *irq)
314{
315 dev->guest_irq = irq->guest_irq;
316 dev->ack_notifier.gsi = -1;
317 dev->host_irq_disabled = false;
318 return 0;
319}
320#endif
321
322#ifdef __KVM_HAVE_MSIX
323static int assigned_device_enable_guest_msix(struct kvm *kvm,
324 struct kvm_assigned_dev_kernel *dev,
325 struct kvm_assigned_irq *irq)
326{
327 dev->guest_irq = irq->guest_irq;
328 dev->ack_notifier.gsi = -1;
329 dev->host_irq_disabled = false;
330 return 0;
331}
332#endif
333
334static int assign_host_irq(struct kvm *kvm,
335 struct kvm_assigned_dev_kernel *dev,
336 __u32 host_irq_type)
337{
338 int r = -EEXIST;
339
340 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
341 return r;
342
343 snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
344 pci_name(dev->dev));
345
346 switch (host_irq_type) {
347 case KVM_DEV_IRQ_HOST_INTX:
348 r = assigned_device_enable_host_intx(kvm, dev);
349 break;
350#ifdef __KVM_HAVE_MSI
351 case KVM_DEV_IRQ_HOST_MSI:
352 r = assigned_device_enable_host_msi(kvm, dev);
353 break;
354#endif
355#ifdef __KVM_HAVE_MSIX
356 case KVM_DEV_IRQ_HOST_MSIX:
357 r = assigned_device_enable_host_msix(kvm, dev);
358 break;
359#endif
360 default:
361 r = -EINVAL;
362 }
363
364 if (!r)
365 dev->irq_requested_type |= host_irq_type;
366
367 return r;
368}
369
370static int assign_guest_irq(struct kvm *kvm,
371 struct kvm_assigned_dev_kernel *dev,
372 struct kvm_assigned_irq *irq,
373 unsigned long guest_irq_type)
374{
375 int id;
376 int r = -EEXIST;
377
378 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
379 return r;
380
381 id = kvm_request_irq_source_id(kvm);
382 if (id < 0)
383 return id;
384
385 dev->irq_source_id = id;
386
387 switch (guest_irq_type) {
388 case KVM_DEV_IRQ_GUEST_INTX:
389 r = assigned_device_enable_guest_intx(kvm, dev, irq);
390 break;
391#ifdef __KVM_HAVE_MSI
392 case KVM_DEV_IRQ_GUEST_MSI:
393 r = assigned_device_enable_guest_msi(kvm, dev, irq);
394 break;
395#endif
396#ifdef __KVM_HAVE_MSIX
397 case KVM_DEV_IRQ_GUEST_MSIX:
398 r = assigned_device_enable_guest_msix(kvm, dev, irq);
399 break;
400#endif
401 default:
402 r = -EINVAL;
403 }
404
405 if (!r) {
406 dev->irq_requested_type |= guest_irq_type;
407 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
408 } else
409 kvm_free_irq_source_id(kvm, dev->irq_source_id);
410
411 return r;
412}
413
414/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
415static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
416 struct kvm_assigned_irq *assigned_irq)
417{
418 int r = -EINVAL;
419 struct kvm_assigned_dev_kernel *match;
420 unsigned long host_irq_type, guest_irq_type;
421
422 if (!irqchip_in_kernel(kvm))
423 return r;
424
425 mutex_lock(&kvm->lock);
426 r = -ENODEV;
427 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
428 assigned_irq->assigned_dev_id);
429 if (!match)
430 goto out;
431
432 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
433 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
434
435 r = -EINVAL;
436 /* can only assign one type at a time */
437 if (hweight_long(host_irq_type) > 1)
438 goto out;
439 if (hweight_long(guest_irq_type) > 1)
440 goto out;
441 if (host_irq_type == 0 && guest_irq_type == 0)
442 goto out;
443
444 r = 0;
445 if (host_irq_type)
446 r = assign_host_irq(kvm, match, host_irq_type);
447 if (r)
448 goto out;
449
450 if (guest_irq_type)
451 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
452out:
453 mutex_unlock(&kvm->lock);
454 return r;
455}
456
457static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
458 struct kvm_assigned_irq
459 *assigned_irq)
460{
461 int r = -ENODEV;
462 struct kvm_assigned_dev_kernel *match;
463
464 mutex_lock(&kvm->lock);
465
466 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
467 assigned_irq->assigned_dev_id);
468 if (!match)
469 goto out;
470
471 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
472out:
473 mutex_unlock(&kvm->lock);
474 return r;
475}
476
477static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
478 struct kvm_assigned_pci_dev *assigned_dev)
479{
480 int r = 0, idx;
481 struct kvm_assigned_dev_kernel *match;
482 struct pci_dev *dev;
483
484 mutex_lock(&kvm->lock);
485 idx = srcu_read_lock(&kvm->srcu);
486
487 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
488 assigned_dev->assigned_dev_id);
489 if (match) {
490 /* device already assigned */
491 r = -EEXIST;
492 goto out;
493 }
494
495 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
496 if (match == NULL) {
497 printk(KERN_INFO "%s: Couldn't allocate memory\n",
498 __func__);
499 r = -ENOMEM;
500 goto out;
501 }
502 dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
503 assigned_dev->busnr,
504 assigned_dev->devfn);
505 if (!dev) {
506 printk(KERN_INFO "%s: host device not found\n", __func__);
507 r = -EINVAL;
508 goto out_free;
509 }
510 if (pci_enable_device(dev)) {
511 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
512 r = -EBUSY;
513 goto out_put;
514 }
515 r = pci_request_regions(dev, "kvm_assigned_device");
516 if (r) {
517 printk(KERN_INFO "%s: Could not get access to device regions\n",
518 __func__);
519 goto out_disable;
520 }
521
522 pci_reset_function(dev);
523 pci_save_state(dev);
524 match->pci_saved_state = pci_store_saved_state(dev);
525 if (!match->pci_saved_state)
526 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
527 __func__, dev_name(&dev->dev));
528 match->assigned_dev_id = assigned_dev->assigned_dev_id;
529 match->host_segnr = assigned_dev->segnr;
530 match->host_busnr = assigned_dev->busnr;
531 match->host_devfn = assigned_dev->devfn;
532 match->flags = assigned_dev->flags;
533 match->dev = dev;
534 spin_lock_init(&match->intx_lock);
535 match->irq_source_id = -1;
536 match->kvm = kvm;
537 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
538
539 list_add(&match->list, &kvm->arch.assigned_dev_head);
540
541 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
542 if (!kvm->arch.iommu_domain) {
543 r = kvm_iommu_map_guest(kvm);
544 if (r)
545 goto out_list_del;
546 }
547 r = kvm_assign_device(kvm, match);
548 if (r)
549 goto out_list_del;
550 }
551
552out:
553 srcu_read_unlock(&kvm->srcu, idx);
554 mutex_unlock(&kvm->lock);
555 return r;
556out_list_del:
557 if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
558 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
559 __func__, dev_name(&dev->dev));
560 list_del(&match->list);
561 pci_release_regions(dev);
562out_disable:
563 pci_disable_device(dev);
564out_put:
565 pci_dev_put(dev);
566out_free:
567 kfree(match);
568 srcu_read_unlock(&kvm->srcu, idx);
569 mutex_unlock(&kvm->lock);
570 return r;
571}
572
573static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
574 struct kvm_assigned_pci_dev *assigned_dev)
575{
576 int r = 0;
577 struct kvm_assigned_dev_kernel *match;
578
579 mutex_lock(&kvm->lock);
580
581 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
582 assigned_dev->assigned_dev_id);
583 if (!match) {
584 printk(KERN_INFO "%s: device hasn't been assigned before, "
585 "so cannot be deassigned\n", __func__);
586 r = -EINVAL;
587 goto out;
588 }
589
590 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
591 kvm_deassign_device(kvm, match);
592
593 kvm_free_assigned_device(kvm, match);
594
595out:
596 mutex_unlock(&kvm->lock);
597 return r;
598}
599
600
601#ifdef __KVM_HAVE_MSIX
602static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
603 struct kvm_assigned_msix_nr *entry_nr)
604{
605 int r = 0;
606 struct kvm_assigned_dev_kernel *adev;
607
608 mutex_lock(&kvm->lock);
609
610 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
611 entry_nr->assigned_dev_id);
612 if (!adev) {
613 r = -EINVAL;
614 goto msix_nr_out;
615 }
616
617 if (adev->entries_nr == 0) {
618 adev->entries_nr = entry_nr->entry_nr;
619 if (adev->entries_nr == 0 ||
620 adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
621 r = -EINVAL;
622 goto msix_nr_out;
623 }
624
625 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
626 entry_nr->entry_nr,
627 GFP_KERNEL);
628 if (!adev->host_msix_entries) {
629 r = -ENOMEM;
630 goto msix_nr_out;
631 }
632 adev->guest_msix_entries =
633 kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
634 GFP_KERNEL);
635 if (!adev->guest_msix_entries) {
636 kfree(adev->host_msix_entries);
637 r = -ENOMEM;
638 goto msix_nr_out;
639 }
640 } else /* Not allowed set MSI-X number twice */
641 r = -EINVAL;
642msix_nr_out:
643 mutex_unlock(&kvm->lock);
644 return r;
645}
646
647static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
648 struct kvm_assigned_msix_entry *entry)
649{
650 int r = 0, i;
651 struct kvm_assigned_dev_kernel *adev;
652
653 mutex_lock(&kvm->lock);
654
655 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
656 entry->assigned_dev_id);
657
658 if (!adev) {
659 r = -EINVAL;
660 goto msix_entry_out;
661 }
662
663 for (i = 0; i < adev->entries_nr; i++)
664 if (adev->guest_msix_entries[i].vector == 0 ||
665 adev->guest_msix_entries[i].entry == entry->entry) {
666 adev->guest_msix_entries[i].entry = entry->entry;
667 adev->guest_msix_entries[i].vector = entry->gsi;
668 adev->host_msix_entries[i].entry = entry->entry;
669 break;
670 }
671 if (i == adev->entries_nr) {
672 r = -ENOSPC;
673 goto msix_entry_out;
674 }
675
676msix_entry_out:
677 mutex_unlock(&kvm->lock);
678
679 return r;
680}
681#endif
682
683long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
684 unsigned long arg)
685{
686 void __user *argp = (void __user *)arg;
687 int r;
688
689 switch (ioctl) {
690 case KVM_ASSIGN_PCI_DEVICE: {
691 struct kvm_assigned_pci_dev assigned_dev;
692
693 r = -EFAULT;
694 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
695 goto out;
696 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
697 if (r)
698 goto out;
699 break;
700 }
701 case KVM_ASSIGN_IRQ: {
702 r = -EOPNOTSUPP;
703 break;
704 }
705 case KVM_ASSIGN_DEV_IRQ: {
706 struct kvm_assigned_irq assigned_irq;
707
708 r = -EFAULT;
709 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
710 goto out;
711 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
712 if (r)
713 goto out;
714 break;
715 }
716 case KVM_DEASSIGN_DEV_IRQ: {
717 struct kvm_assigned_irq assigned_irq;
718
719 r = -EFAULT;
720 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
721 goto out;
722 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
723 if (r)
724 goto out;
725 break;
726 }
727 case KVM_DEASSIGN_PCI_DEVICE: {
728 struct kvm_assigned_pci_dev assigned_dev;
729
730 r = -EFAULT;
731 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
732 goto out;
733 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
734 if (r)
735 goto out;
736 break;
737 }
738#ifdef KVM_CAP_IRQ_ROUTING
739 case KVM_SET_GSI_ROUTING: {
740 struct kvm_irq_routing routing;
741 struct kvm_irq_routing __user *urouting;
742 struct kvm_irq_routing_entry *entries;
743
744 r = -EFAULT;
745 if (copy_from_user(&routing, argp, sizeof(routing)))
746 goto out;
747 r = -EINVAL;
748 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
749 goto out;
750 if (routing.flags)
751 goto out;
752 r = -ENOMEM;
753 entries = vmalloc(routing.nr * sizeof(*entries));
754 if (!entries)
755 goto out;
756 r = -EFAULT;
757 urouting = argp;
758 if (copy_from_user(entries, urouting->entries,
759 routing.nr * sizeof(*entries)))
760 goto out_free_irq_routing;
761 r = kvm_set_irq_routing(kvm, entries, routing.nr,
762 routing.flags);
763 out_free_irq_routing:
764 vfree(entries);
765 break;
766 }
767#endif /* KVM_CAP_IRQ_ROUTING */
768#ifdef __KVM_HAVE_MSIX
769 case KVM_ASSIGN_SET_MSIX_NR: {
770 struct kvm_assigned_msix_nr entry_nr;
771 r = -EFAULT;
772 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
773 goto out;
774 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
775 if (r)
776 goto out;
777 break;
778 }
779 case KVM_ASSIGN_SET_MSIX_ENTRY: {
780 struct kvm_assigned_msix_entry entry;
781 r = -EFAULT;
782 if (copy_from_user(&entry, argp, sizeof entry))
783 goto out;
784 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
785 if (r)
786 goto out;
787 break;
788 }
789#endif
790 default:
791 r = -ENOTTY;
792 break;
793 }
794out:
795 return r;
796}
797
1/*
2 * Kernel-based Virtual Machine - device assignment support
3 *
4 * Copyright (C) 2010 Red Hat, Inc. and/or its affiliates.
5 *
6 * This work is licensed under the terms of the GNU GPL, version 2. See
7 * the COPYING file in the top-level directory.
8 *
9 */
10
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>
20#include <linux/namei.h>
21#include <linux/fs.h>
22#include "irq.h"
23
24static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
25 int assigned_dev_id)
26{
27 struct list_head *ptr;
28 struct kvm_assigned_dev_kernel *match;
29
30 list_for_each(ptr, head) {
31 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
32 if (match->assigned_dev_id == assigned_dev_id)
33 return match;
34 }
35 return NULL;
36}
37
38static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
39 *assigned_dev, int irq)
40{
41 int i, index;
42 struct msix_entry *host_msix_entries;
43
44 host_msix_entries = assigned_dev->host_msix_entries;
45
46 index = -1;
47 for (i = 0; i < assigned_dev->entries_nr; i++)
48 if (irq == host_msix_entries[i].vector) {
49 index = i;
50 break;
51 }
52 if (index < 0)
53 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
54
55 return index;
56}
57
58static irqreturn_t kvm_assigned_dev_intx(int irq, void *dev_id)
59{
60 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
61 int ret;
62
63 spin_lock(&assigned_dev->intx_lock);
64 if (pci_check_and_mask_intx(assigned_dev->dev)) {
65 assigned_dev->host_irq_disabled = true;
66 ret = IRQ_WAKE_THREAD;
67 } else
68 ret = IRQ_NONE;
69 spin_unlock(&assigned_dev->intx_lock);
70
71 return ret;
72}
73
74static void
75kvm_assigned_dev_raise_guest_irq(struct kvm_assigned_dev_kernel *assigned_dev,
76 int vector)
77{
78 if (unlikely(assigned_dev->irq_requested_type &
79 KVM_DEV_IRQ_GUEST_INTX)) {
80 spin_lock(&assigned_dev->intx_mask_lock);
81 if (!(assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX))
82 kvm_set_irq(assigned_dev->kvm,
83 assigned_dev->irq_source_id, vector, 1,
84 false);
85 spin_unlock(&assigned_dev->intx_mask_lock);
86 } else
87 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
88 vector, 1, false);
89}
90
91static irqreturn_t kvm_assigned_dev_thread_intx(int irq, void *dev_id)
92{
93 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
94
95 if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
96 spin_lock_irq(&assigned_dev->intx_lock);
97 disable_irq_nosync(irq);
98 assigned_dev->host_irq_disabled = true;
99 spin_unlock_irq(&assigned_dev->intx_lock);
100 }
101
102 kvm_assigned_dev_raise_guest_irq(assigned_dev,
103 assigned_dev->guest_irq);
104
105 return IRQ_HANDLED;
106}
107
108#ifdef __KVM_HAVE_MSI
109static irqreturn_t kvm_assigned_dev_msi(int irq, void *dev_id)
110{
111 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
112 int ret = kvm_set_irq_inatomic(assigned_dev->kvm,
113 assigned_dev->irq_source_id,
114 assigned_dev->guest_irq, 1);
115 return unlikely(ret == -EWOULDBLOCK) ? IRQ_WAKE_THREAD : IRQ_HANDLED;
116}
117
118static irqreturn_t kvm_assigned_dev_thread_msi(int irq, void *dev_id)
119{
120 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
121
122 kvm_assigned_dev_raise_guest_irq(assigned_dev,
123 assigned_dev->guest_irq);
124
125 return IRQ_HANDLED;
126}
127#endif
128
129#ifdef __KVM_HAVE_MSIX
130static irqreturn_t kvm_assigned_dev_msix(int irq, void *dev_id)
131{
132 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
133 int index = find_index_from_host_irq(assigned_dev, irq);
134 u32 vector;
135 int ret = 0;
136
137 if (index >= 0) {
138 vector = assigned_dev->guest_msix_entries[index].vector;
139 ret = kvm_set_irq_inatomic(assigned_dev->kvm,
140 assigned_dev->irq_source_id,
141 vector, 1);
142 }
143
144 return unlikely(ret == -EWOULDBLOCK) ? IRQ_WAKE_THREAD : IRQ_HANDLED;
145}
146
147static irqreturn_t kvm_assigned_dev_thread_msix(int irq, void *dev_id)
148{
149 struct kvm_assigned_dev_kernel *assigned_dev = dev_id;
150 int index = find_index_from_host_irq(assigned_dev, irq);
151 u32 vector;
152
153 if (index >= 0) {
154 vector = assigned_dev->guest_msix_entries[index].vector;
155 kvm_assigned_dev_raise_guest_irq(assigned_dev, vector);
156 }
157
158 return IRQ_HANDLED;
159}
160#endif
161
162/* Ack the irq line for an assigned device */
163static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
164{
165 struct kvm_assigned_dev_kernel *dev =
166 container_of(kian, struct kvm_assigned_dev_kernel,
167 ack_notifier);
168
169 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0, false);
170
171 spin_lock(&dev->intx_mask_lock);
172
173 if (!(dev->flags & KVM_DEV_ASSIGN_MASK_INTX)) {
174 bool reassert = false;
175
176 spin_lock_irq(&dev->intx_lock);
177 /*
178 * The guest IRQ may be shared so this ack can come from an
179 * IRQ for another guest device.
180 */
181 if (dev->host_irq_disabled) {
182 if (!(dev->flags & KVM_DEV_ASSIGN_PCI_2_3))
183 enable_irq(dev->host_irq);
184 else if (!pci_check_and_unmask_intx(dev->dev))
185 reassert = true;
186 dev->host_irq_disabled = reassert;
187 }
188 spin_unlock_irq(&dev->intx_lock);
189
190 if (reassert)
191 kvm_set_irq(dev->kvm, dev->irq_source_id,
192 dev->guest_irq, 1, false);
193 }
194
195 spin_unlock(&dev->intx_mask_lock);
196}
197
198static void deassign_guest_irq(struct kvm *kvm,
199 struct kvm_assigned_dev_kernel *assigned_dev)
200{
201 if (assigned_dev->ack_notifier.gsi != -1)
202 kvm_unregister_irq_ack_notifier(kvm,
203 &assigned_dev->ack_notifier);
204
205 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
206 assigned_dev->guest_irq, 0, false);
207
208 if (assigned_dev->irq_source_id != -1)
209 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
210 assigned_dev->irq_source_id = -1;
211 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
212}
213
214/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
215static void deassign_host_irq(struct kvm *kvm,
216 struct kvm_assigned_dev_kernel *assigned_dev)
217{
218 /*
219 * We disable irq here to prevent further events.
220 *
221 * Notice this maybe result in nested disable if the interrupt type is
222 * INTx, but it's OK for we are going to free it.
223 *
224 * If this function is a part of VM destroy, please ensure that till
225 * now, the kvm state is still legal for probably we also have to wait
226 * on a currently running IRQ handler.
227 */
228 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
229 int i;
230 for (i = 0; i < assigned_dev->entries_nr; i++)
231 disable_irq(assigned_dev->host_msix_entries[i].vector);
232
233 for (i = 0; i < assigned_dev->entries_nr; i++)
234 free_irq(assigned_dev->host_msix_entries[i].vector,
235 assigned_dev);
236
237 assigned_dev->entries_nr = 0;
238 kfree(assigned_dev->host_msix_entries);
239 kfree(assigned_dev->guest_msix_entries);
240 pci_disable_msix(assigned_dev->dev);
241 } else {
242 /* Deal with MSI and INTx */
243 if ((assigned_dev->irq_requested_type &
244 KVM_DEV_IRQ_HOST_INTX) &&
245 (assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
246 spin_lock_irq(&assigned_dev->intx_lock);
247 pci_intx(assigned_dev->dev, false);
248 spin_unlock_irq(&assigned_dev->intx_lock);
249 synchronize_irq(assigned_dev->host_irq);
250 } else
251 disable_irq(assigned_dev->host_irq);
252
253 free_irq(assigned_dev->host_irq, assigned_dev);
254
255 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
256 pci_disable_msi(assigned_dev->dev);
257 }
258
259 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
260}
261
262static int kvm_deassign_irq(struct kvm *kvm,
263 struct kvm_assigned_dev_kernel *assigned_dev,
264 unsigned long irq_requested_type)
265{
266 unsigned long guest_irq_type, host_irq_type;
267
268 if (!irqchip_in_kernel(kvm))
269 return -EINVAL;
270 /* no irq assignment to deassign */
271 if (!assigned_dev->irq_requested_type)
272 return -ENXIO;
273
274 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
275 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
276
277 if (host_irq_type)
278 deassign_host_irq(kvm, assigned_dev);
279 if (guest_irq_type)
280 deassign_guest_irq(kvm, assigned_dev);
281
282 return 0;
283}
284
285static void kvm_free_assigned_irq(struct kvm *kvm,
286 struct kvm_assigned_dev_kernel *assigned_dev)
287{
288 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
289}
290
291static void kvm_free_assigned_device(struct kvm *kvm,
292 struct kvm_assigned_dev_kernel
293 *assigned_dev)
294{
295 kvm_free_assigned_irq(kvm, assigned_dev);
296
297 pci_reset_function(assigned_dev->dev);
298 if (pci_load_and_free_saved_state(assigned_dev->dev,
299 &assigned_dev->pci_saved_state))
300 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
301 __func__, dev_name(&assigned_dev->dev->dev));
302 else
303 pci_restore_state(assigned_dev->dev);
304
305 assigned_dev->dev->dev_flags &= ~PCI_DEV_FLAGS_ASSIGNED;
306
307 pci_release_regions(assigned_dev->dev);
308 pci_disable_device(assigned_dev->dev);
309 pci_dev_put(assigned_dev->dev);
310
311 list_del(&assigned_dev->list);
312 kfree(assigned_dev);
313}
314
315void kvm_free_all_assigned_devices(struct kvm *kvm)
316{
317 struct list_head *ptr, *ptr2;
318 struct kvm_assigned_dev_kernel *assigned_dev;
319
320 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
321 assigned_dev = list_entry(ptr,
322 struct kvm_assigned_dev_kernel,
323 list);
324
325 kvm_free_assigned_device(kvm, assigned_dev);
326 }
327}
328
329static int assigned_device_enable_host_intx(struct kvm *kvm,
330 struct kvm_assigned_dev_kernel *dev)
331{
332 irq_handler_t irq_handler;
333 unsigned long flags;
334
335 dev->host_irq = dev->dev->irq;
336
337 /*
338 * We can only share the IRQ line with other host devices if we are
339 * able to disable the IRQ source at device-level - independently of
340 * the guest driver. Otherwise host devices may suffer from unbounded
341 * IRQ latencies when the guest keeps the line asserted.
342 */
343 if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
344 irq_handler = kvm_assigned_dev_intx;
345 flags = IRQF_SHARED;
346 } else {
347 irq_handler = NULL;
348 flags = IRQF_ONESHOT;
349 }
350 if (request_threaded_irq(dev->host_irq, irq_handler,
351 kvm_assigned_dev_thread_intx, flags,
352 dev->irq_name, dev))
353 return -EIO;
354
355 if (dev->flags & KVM_DEV_ASSIGN_PCI_2_3) {
356 spin_lock_irq(&dev->intx_lock);
357 pci_intx(dev->dev, true);
358 spin_unlock_irq(&dev->intx_lock);
359 }
360 return 0;
361}
362
363#ifdef __KVM_HAVE_MSI
364static int assigned_device_enable_host_msi(struct kvm *kvm,
365 struct kvm_assigned_dev_kernel *dev)
366{
367 int r;
368
369 if (!dev->dev->msi_enabled) {
370 r = pci_enable_msi(dev->dev);
371 if (r)
372 return r;
373 }
374
375 dev->host_irq = dev->dev->irq;
376 if (request_threaded_irq(dev->host_irq, kvm_assigned_dev_msi,
377 kvm_assigned_dev_thread_msi, 0,
378 dev->irq_name, dev)) {
379 pci_disable_msi(dev->dev);
380 return -EIO;
381 }
382
383 return 0;
384}
385#endif
386
387#ifdef __KVM_HAVE_MSIX
388static int assigned_device_enable_host_msix(struct kvm *kvm,
389 struct kvm_assigned_dev_kernel *dev)
390{
391 int i, r = -EINVAL;
392
393 /* host_msix_entries and guest_msix_entries should have been
394 * initialized */
395 if (dev->entries_nr == 0)
396 return r;
397
398 r = pci_enable_msix_exact(dev->dev,
399 dev->host_msix_entries, dev->entries_nr);
400 if (r)
401 return r;
402
403 for (i = 0; i < dev->entries_nr; i++) {
404 r = request_threaded_irq(dev->host_msix_entries[i].vector,
405 kvm_assigned_dev_msix,
406 kvm_assigned_dev_thread_msix,
407 0, dev->irq_name, dev);
408 if (r)
409 goto err;
410 }
411
412 return 0;
413err:
414 for (i -= 1; i >= 0; i--)
415 free_irq(dev->host_msix_entries[i].vector, dev);
416 pci_disable_msix(dev->dev);
417 return r;
418}
419
420#endif
421
422static int assigned_device_enable_guest_intx(struct kvm *kvm,
423 struct kvm_assigned_dev_kernel *dev,
424 struct kvm_assigned_irq *irq)
425{
426 dev->guest_irq = irq->guest_irq;
427 dev->ack_notifier.gsi = irq->guest_irq;
428 return 0;
429}
430
431#ifdef __KVM_HAVE_MSI
432static int assigned_device_enable_guest_msi(struct kvm *kvm,
433 struct kvm_assigned_dev_kernel *dev,
434 struct kvm_assigned_irq *irq)
435{
436 dev->guest_irq = irq->guest_irq;
437 dev->ack_notifier.gsi = -1;
438 return 0;
439}
440#endif
441
442#ifdef __KVM_HAVE_MSIX
443static int assigned_device_enable_guest_msix(struct kvm *kvm,
444 struct kvm_assigned_dev_kernel *dev,
445 struct kvm_assigned_irq *irq)
446{
447 dev->guest_irq = irq->guest_irq;
448 dev->ack_notifier.gsi = -1;
449 return 0;
450}
451#endif
452
453static int assign_host_irq(struct kvm *kvm,
454 struct kvm_assigned_dev_kernel *dev,
455 __u32 host_irq_type)
456{
457 int r = -EEXIST;
458
459 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
460 return r;
461
462 snprintf(dev->irq_name, sizeof(dev->irq_name), "kvm:%s",
463 pci_name(dev->dev));
464
465 switch (host_irq_type) {
466 case KVM_DEV_IRQ_HOST_INTX:
467 r = assigned_device_enable_host_intx(kvm, dev);
468 break;
469#ifdef __KVM_HAVE_MSI
470 case KVM_DEV_IRQ_HOST_MSI:
471 r = assigned_device_enable_host_msi(kvm, dev);
472 break;
473#endif
474#ifdef __KVM_HAVE_MSIX
475 case KVM_DEV_IRQ_HOST_MSIX:
476 r = assigned_device_enable_host_msix(kvm, dev);
477 break;
478#endif
479 default:
480 r = -EINVAL;
481 }
482 dev->host_irq_disabled = false;
483
484 if (!r)
485 dev->irq_requested_type |= host_irq_type;
486
487 return r;
488}
489
490static int assign_guest_irq(struct kvm *kvm,
491 struct kvm_assigned_dev_kernel *dev,
492 struct kvm_assigned_irq *irq,
493 unsigned long guest_irq_type)
494{
495 int id;
496 int r = -EEXIST;
497
498 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
499 return r;
500
501 id = kvm_request_irq_source_id(kvm);
502 if (id < 0)
503 return id;
504
505 dev->irq_source_id = id;
506
507 switch (guest_irq_type) {
508 case KVM_DEV_IRQ_GUEST_INTX:
509 r = assigned_device_enable_guest_intx(kvm, dev, irq);
510 break;
511#ifdef __KVM_HAVE_MSI
512 case KVM_DEV_IRQ_GUEST_MSI:
513 r = assigned_device_enable_guest_msi(kvm, dev, irq);
514 break;
515#endif
516#ifdef __KVM_HAVE_MSIX
517 case KVM_DEV_IRQ_GUEST_MSIX:
518 r = assigned_device_enable_guest_msix(kvm, dev, irq);
519 break;
520#endif
521 default:
522 r = -EINVAL;
523 }
524
525 if (!r) {
526 dev->irq_requested_type |= guest_irq_type;
527 if (dev->ack_notifier.gsi != -1)
528 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
529 } else
530 kvm_free_irq_source_id(kvm, dev->irq_source_id);
531
532 return r;
533}
534
535/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
536static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
537 struct kvm_assigned_irq *assigned_irq)
538{
539 int r = -EINVAL;
540 struct kvm_assigned_dev_kernel *match;
541 unsigned long host_irq_type, guest_irq_type;
542
543 if (!irqchip_in_kernel(kvm))
544 return r;
545
546 mutex_lock(&kvm->lock);
547 r = -ENODEV;
548 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
549 assigned_irq->assigned_dev_id);
550 if (!match)
551 goto out;
552
553 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
554 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
555
556 r = -EINVAL;
557 /* can only assign one type at a time */
558 if (hweight_long(host_irq_type) > 1)
559 goto out;
560 if (hweight_long(guest_irq_type) > 1)
561 goto out;
562 if (host_irq_type == 0 && guest_irq_type == 0)
563 goto out;
564
565 r = 0;
566 if (host_irq_type)
567 r = assign_host_irq(kvm, match, host_irq_type);
568 if (r)
569 goto out;
570
571 if (guest_irq_type)
572 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
573out:
574 mutex_unlock(&kvm->lock);
575 return r;
576}
577
578static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
579 struct kvm_assigned_irq
580 *assigned_irq)
581{
582 int r = -ENODEV;
583 struct kvm_assigned_dev_kernel *match;
584 unsigned long irq_type;
585
586 mutex_lock(&kvm->lock);
587
588 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
589 assigned_irq->assigned_dev_id);
590 if (!match)
591 goto out;
592
593 irq_type = assigned_irq->flags & (KVM_DEV_IRQ_HOST_MASK |
594 KVM_DEV_IRQ_GUEST_MASK);
595 r = kvm_deassign_irq(kvm, match, irq_type);
596out:
597 mutex_unlock(&kvm->lock);
598 return r;
599}
600
601/*
602 * We want to test whether the caller has been granted permissions to
603 * use this device. To be able to configure and control the device,
604 * the user needs access to PCI configuration space and BAR resources.
605 * These are accessed through PCI sysfs. PCI config space is often
606 * passed to the process calling this ioctl via file descriptor, so we
607 * can't rely on access to that file. We can check for permissions
608 * on each of the BAR resource files, which is a pretty clear
609 * indicator that the user has been granted access to the device.
610 */
611static int probe_sysfs_permissions(struct pci_dev *dev)
612{
613#ifdef CONFIG_SYSFS
614 int i;
615 bool bar_found = false;
616
617 for (i = PCI_STD_RESOURCES; i <= PCI_STD_RESOURCE_END; i++) {
618 char *kpath, *syspath;
619 struct path path;
620 struct inode *inode;
621 int r;
622
623 if (!pci_resource_len(dev, i))
624 continue;
625
626 kpath = kobject_get_path(&dev->dev.kobj, GFP_KERNEL);
627 if (!kpath)
628 return -ENOMEM;
629
630 /* Per sysfs-rules, sysfs is always at /sys */
631 syspath = kasprintf(GFP_KERNEL, "/sys%s/resource%d", kpath, i);
632 kfree(kpath);
633 if (!syspath)
634 return -ENOMEM;
635
636 r = kern_path(syspath, LOOKUP_FOLLOW, &path);
637 kfree(syspath);
638 if (r)
639 return r;
640
641 inode = path.dentry->d_inode;
642
643 r = inode_permission(inode, MAY_READ | MAY_WRITE | MAY_ACCESS);
644 path_put(&path);
645 if (r)
646 return r;
647
648 bar_found = true;
649 }
650
651 /* If no resources, probably something special */
652 if (!bar_found)
653 return -EPERM;
654
655 return 0;
656#else
657 return -EINVAL; /* No way to control the device without sysfs */
658#endif
659}
660
661static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
662 struct kvm_assigned_pci_dev *assigned_dev)
663{
664 int r = 0, idx;
665 struct kvm_assigned_dev_kernel *match;
666 struct pci_dev *dev;
667
668 if (!(assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU))
669 return -EINVAL;
670
671 mutex_lock(&kvm->lock);
672 idx = srcu_read_lock(&kvm->srcu);
673
674 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
675 assigned_dev->assigned_dev_id);
676 if (match) {
677 /* device already assigned */
678 r = -EEXIST;
679 goto out;
680 }
681
682 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
683 if (match == NULL) {
684 printk(KERN_INFO "%s: Couldn't allocate memory\n",
685 __func__);
686 r = -ENOMEM;
687 goto out;
688 }
689 dev = pci_get_domain_bus_and_slot(assigned_dev->segnr,
690 assigned_dev->busnr,
691 assigned_dev->devfn);
692 if (!dev) {
693 printk(KERN_INFO "%s: host device not found\n", __func__);
694 r = -EINVAL;
695 goto out_free;
696 }
697
698 /* Don't allow bridges to be assigned */
699 if (dev->hdr_type != PCI_HEADER_TYPE_NORMAL) {
700 r = -EPERM;
701 goto out_put;
702 }
703
704 r = probe_sysfs_permissions(dev);
705 if (r)
706 goto out_put;
707
708 if (pci_enable_device(dev)) {
709 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
710 r = -EBUSY;
711 goto out_put;
712 }
713 r = pci_request_regions(dev, "kvm_assigned_device");
714 if (r) {
715 printk(KERN_INFO "%s: Could not get access to device regions\n",
716 __func__);
717 goto out_disable;
718 }
719
720 pci_reset_function(dev);
721 pci_save_state(dev);
722 match->pci_saved_state = pci_store_saved_state(dev);
723 if (!match->pci_saved_state)
724 printk(KERN_DEBUG "%s: Couldn't store %s saved state\n",
725 __func__, dev_name(&dev->dev));
726
727 if (!pci_intx_mask_supported(dev))
728 assigned_dev->flags &= ~KVM_DEV_ASSIGN_PCI_2_3;
729
730 match->assigned_dev_id = assigned_dev->assigned_dev_id;
731 match->host_segnr = assigned_dev->segnr;
732 match->host_busnr = assigned_dev->busnr;
733 match->host_devfn = assigned_dev->devfn;
734 match->flags = assigned_dev->flags;
735 match->dev = dev;
736 spin_lock_init(&match->intx_lock);
737 spin_lock_init(&match->intx_mask_lock);
738 match->irq_source_id = -1;
739 match->kvm = kvm;
740 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
741
742 list_add(&match->list, &kvm->arch.assigned_dev_head);
743
744 if (!kvm->arch.iommu_domain) {
745 r = kvm_iommu_map_guest(kvm);
746 if (r)
747 goto out_list_del;
748 }
749 r = kvm_assign_device(kvm, match);
750 if (r)
751 goto out_list_del;
752
753out:
754 srcu_read_unlock(&kvm->srcu, idx);
755 mutex_unlock(&kvm->lock);
756 return r;
757out_list_del:
758 if (pci_load_and_free_saved_state(dev, &match->pci_saved_state))
759 printk(KERN_INFO "%s: Couldn't reload %s saved state\n",
760 __func__, dev_name(&dev->dev));
761 list_del(&match->list);
762 pci_release_regions(dev);
763out_disable:
764 pci_disable_device(dev);
765out_put:
766 pci_dev_put(dev);
767out_free:
768 kfree(match);
769 srcu_read_unlock(&kvm->srcu, idx);
770 mutex_unlock(&kvm->lock);
771 return r;
772}
773
774static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
775 struct kvm_assigned_pci_dev *assigned_dev)
776{
777 int r = 0;
778 struct kvm_assigned_dev_kernel *match;
779
780 mutex_lock(&kvm->lock);
781
782 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
783 assigned_dev->assigned_dev_id);
784 if (!match) {
785 printk(KERN_INFO "%s: device hasn't been assigned before, "
786 "so cannot be deassigned\n", __func__);
787 r = -EINVAL;
788 goto out;
789 }
790
791 kvm_deassign_device(kvm, match);
792
793 kvm_free_assigned_device(kvm, match);
794
795out:
796 mutex_unlock(&kvm->lock);
797 return r;
798}
799
800
801#ifdef __KVM_HAVE_MSIX
802static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
803 struct kvm_assigned_msix_nr *entry_nr)
804{
805 int r = 0;
806 struct kvm_assigned_dev_kernel *adev;
807
808 mutex_lock(&kvm->lock);
809
810 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
811 entry_nr->assigned_dev_id);
812 if (!adev) {
813 r = -EINVAL;
814 goto msix_nr_out;
815 }
816
817 if (adev->entries_nr == 0) {
818 adev->entries_nr = entry_nr->entry_nr;
819 if (adev->entries_nr == 0 ||
820 adev->entries_nr > KVM_MAX_MSIX_PER_DEV) {
821 r = -EINVAL;
822 goto msix_nr_out;
823 }
824
825 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
826 entry_nr->entry_nr,
827 GFP_KERNEL);
828 if (!adev->host_msix_entries) {
829 r = -ENOMEM;
830 goto msix_nr_out;
831 }
832 adev->guest_msix_entries =
833 kzalloc(sizeof(struct msix_entry) * entry_nr->entry_nr,
834 GFP_KERNEL);
835 if (!adev->guest_msix_entries) {
836 kfree(adev->host_msix_entries);
837 r = -ENOMEM;
838 goto msix_nr_out;
839 }
840 } else /* Not allowed set MSI-X number twice */
841 r = -EINVAL;
842msix_nr_out:
843 mutex_unlock(&kvm->lock);
844 return r;
845}
846
847static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
848 struct kvm_assigned_msix_entry *entry)
849{
850 int r = 0, i;
851 struct kvm_assigned_dev_kernel *adev;
852
853 mutex_lock(&kvm->lock);
854
855 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
856 entry->assigned_dev_id);
857
858 if (!adev) {
859 r = -EINVAL;
860 goto msix_entry_out;
861 }
862
863 for (i = 0; i < adev->entries_nr; i++)
864 if (adev->guest_msix_entries[i].vector == 0 ||
865 adev->guest_msix_entries[i].entry == entry->entry) {
866 adev->guest_msix_entries[i].entry = entry->entry;
867 adev->guest_msix_entries[i].vector = entry->gsi;
868 adev->host_msix_entries[i].entry = entry->entry;
869 break;
870 }
871 if (i == adev->entries_nr) {
872 r = -ENOSPC;
873 goto msix_entry_out;
874 }
875
876msix_entry_out:
877 mutex_unlock(&kvm->lock);
878
879 return r;
880}
881#endif
882
883static int kvm_vm_ioctl_set_pci_irq_mask(struct kvm *kvm,
884 struct kvm_assigned_pci_dev *assigned_dev)
885{
886 int r = 0;
887 struct kvm_assigned_dev_kernel *match;
888
889 mutex_lock(&kvm->lock);
890
891 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
892 assigned_dev->assigned_dev_id);
893 if (!match) {
894 r = -ENODEV;
895 goto out;
896 }
897
898 spin_lock(&match->intx_mask_lock);
899
900 match->flags &= ~KVM_DEV_ASSIGN_MASK_INTX;
901 match->flags |= assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX;
902
903 if (match->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
904 if (assigned_dev->flags & KVM_DEV_ASSIGN_MASK_INTX) {
905 kvm_set_irq(match->kvm, match->irq_source_id,
906 match->guest_irq, 0, false);
907 /*
908 * Masking at hardware-level is performed on demand,
909 * i.e. when an IRQ actually arrives at the host.
910 */
911 } else if (!(assigned_dev->flags & KVM_DEV_ASSIGN_PCI_2_3)) {
912 /*
913 * Unmask the IRQ line if required. Unmasking at
914 * device level will be performed by user space.
915 */
916 spin_lock_irq(&match->intx_lock);
917 if (match->host_irq_disabled) {
918 enable_irq(match->host_irq);
919 match->host_irq_disabled = false;
920 }
921 spin_unlock_irq(&match->intx_lock);
922 }
923 }
924
925 spin_unlock(&match->intx_mask_lock);
926
927out:
928 mutex_unlock(&kvm->lock);
929 return r;
930}
931
932long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
933 unsigned long arg)
934{
935 void __user *argp = (void __user *)arg;
936 int r;
937
938 switch (ioctl) {
939 case KVM_ASSIGN_PCI_DEVICE: {
940 struct kvm_assigned_pci_dev assigned_dev;
941
942 r = -EFAULT;
943 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
944 goto out;
945 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
946 if (r)
947 goto out;
948 break;
949 }
950 case KVM_ASSIGN_IRQ: {
951 r = -EOPNOTSUPP;
952 break;
953 }
954 case KVM_ASSIGN_DEV_IRQ: {
955 struct kvm_assigned_irq assigned_irq;
956
957 r = -EFAULT;
958 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
959 goto out;
960 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
961 if (r)
962 goto out;
963 break;
964 }
965 case KVM_DEASSIGN_DEV_IRQ: {
966 struct kvm_assigned_irq assigned_irq;
967
968 r = -EFAULT;
969 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
970 goto out;
971 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
972 if (r)
973 goto out;
974 break;
975 }
976 case KVM_DEASSIGN_PCI_DEVICE: {
977 struct kvm_assigned_pci_dev assigned_dev;
978
979 r = -EFAULT;
980 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
981 goto out;
982 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
983 if (r)
984 goto out;
985 break;
986 }
987#ifdef __KVM_HAVE_MSIX
988 case KVM_ASSIGN_SET_MSIX_NR: {
989 struct kvm_assigned_msix_nr entry_nr;
990 r = -EFAULT;
991 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
992 goto out;
993 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
994 if (r)
995 goto out;
996 break;
997 }
998 case KVM_ASSIGN_SET_MSIX_ENTRY: {
999 struct kvm_assigned_msix_entry entry;
1000 r = -EFAULT;
1001 if (copy_from_user(&entry, argp, sizeof entry))
1002 goto out;
1003 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
1004 if (r)
1005 goto out;
1006 break;
1007 }
1008#endif
1009 case KVM_ASSIGN_SET_INTX_MASK: {
1010 struct kvm_assigned_pci_dev assigned_dev;
1011
1012 r = -EFAULT;
1013 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1014 goto out;
1015 r = kvm_vm_ioctl_set_pci_irq_mask(kvm, &assigned_dev);
1016 break;
1017 }
1018 default:
1019 r = -ENOTTY;
1020 break;
1021 }
1022out:
1023 return r;
1024}