aboutsummaryrefslogtreecommitdiffstats
path: root/virt
diff options
context:
space:
mode:
Diffstat (limited to 'virt')
-rw-r--r--virt/kvm/assigned-dev.c818
-rw-r--r--virt/kvm/eventfd.c2
-rw-r--r--virt/kvm/ioapic.c80
-rw-r--r--virt/kvm/ioapic.h5
-rw-r--r--virt/kvm/irq_comm.c231
-rw-r--r--virt/kvm/kvm_main.c961
6 files changed, 1167 insertions, 930 deletions
diff --git a/virt/kvm/assigned-dev.c b/virt/kvm/assigned-dev.c
new file mode 100644
index 00000000000..fd9c097b760
--- /dev/null
+++ b/virt/kvm/assigned-dev.c
@@ -0,0 +1,818 @@
1/*
2 * Kernel-based Virtual Machine - device assignment support
3 *
4 * Copyright (C) 2006-9 Red Hat, Inc
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 "irq.h"
20
21static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
22 int assigned_dev_id)
23{
24 struct list_head *ptr;
25 struct kvm_assigned_dev_kernel *match;
26
27 list_for_each(ptr, head) {
28 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
29 if (match->assigned_dev_id == assigned_dev_id)
30 return match;
31 }
32 return NULL;
33}
34
35static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
36 *assigned_dev, int irq)
37{
38 int i, index;
39 struct msix_entry *host_msix_entries;
40
41 host_msix_entries = assigned_dev->host_msix_entries;
42
43 index = -1;
44 for (i = 0; i < assigned_dev->entries_nr; i++)
45 if (irq == host_msix_entries[i].vector) {
46 index = i;
47 break;
48 }
49 if (index < 0) {
50 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
51 return 0;
52 }
53
54 return index;
55}
56
57static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
58{
59 struct kvm_assigned_dev_kernel *assigned_dev;
60 struct kvm *kvm;
61 int i;
62
63 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
64 interrupt_work);
65 kvm = assigned_dev->kvm;
66
67 spin_lock_irq(&assigned_dev->assigned_dev_lock);
68 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
69 struct kvm_guest_msix_entry *guest_entries =
70 assigned_dev->guest_msix_entries;
71 for (i = 0; i < assigned_dev->entries_nr; i++) {
72 if (!(guest_entries[i].flags &
73 KVM_ASSIGNED_MSIX_PENDING))
74 continue;
75 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
76 kvm_set_irq(assigned_dev->kvm,
77 assigned_dev->irq_source_id,
78 guest_entries[i].vector, 1);
79 }
80 } else
81 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
82 assigned_dev->guest_irq, 1);
83
84 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
85}
86
87static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
88{
89 unsigned long flags;
90 struct kvm_assigned_dev_kernel *assigned_dev =
91 (struct kvm_assigned_dev_kernel *) dev_id;
92
93 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
94 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
95 int index = find_index_from_host_irq(assigned_dev, irq);
96 if (index < 0)
97 goto out;
98 assigned_dev->guest_msix_entries[index].flags |=
99 KVM_ASSIGNED_MSIX_PENDING;
100 }
101
102 schedule_work(&assigned_dev->interrupt_work);
103
104 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
105 disable_irq_nosync(irq);
106 assigned_dev->host_irq_disabled = true;
107 }
108
109out:
110 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
111 return IRQ_HANDLED;
112}
113
114/* Ack the irq line for an assigned device */
115static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
116{
117 struct kvm_assigned_dev_kernel *dev;
118 unsigned long flags;
119
120 if (kian->gsi == -1)
121 return;
122
123 dev = container_of(kian, struct kvm_assigned_dev_kernel,
124 ack_notifier);
125
126 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
127
128 /* The guest irq may be shared so this ack may be
129 * from another device.
130 */
131 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
132 if (dev->host_irq_disabled) {
133 enable_irq(dev->host_irq);
134 dev->host_irq_disabled = false;
135 }
136 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
137}
138
139static void deassign_guest_irq(struct kvm *kvm,
140 struct kvm_assigned_dev_kernel *assigned_dev)
141{
142 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
143 assigned_dev->ack_notifier.gsi = -1;
144
145 if (assigned_dev->irq_source_id != -1)
146 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
147 assigned_dev->irq_source_id = -1;
148 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
149}
150
151/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
152static void deassign_host_irq(struct kvm *kvm,
153 struct kvm_assigned_dev_kernel *assigned_dev)
154{
155 /*
156 * In kvm_free_device_irq, cancel_work_sync return true if:
157 * 1. work is scheduled, and then cancelled.
158 * 2. work callback is executed.
159 *
160 * The first one ensured that the irq is disabled and no more events
161 * would happen. But for the second one, the irq may be enabled (e.g.
162 * for MSI). So we disable irq here to prevent further events.
163 *
164 * Notice this maybe result in nested disable if the interrupt type is
165 * INTx, but it's OK for we are going to free it.
166 *
167 * If this function is a part of VM destroy, please ensure that till
168 * now, the kvm state is still legal for probably we also have to wait
169 * interrupt_work done.
170 */
171 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
172 int i;
173 for (i = 0; i < assigned_dev->entries_nr; i++)
174 disable_irq_nosync(assigned_dev->
175 host_msix_entries[i].vector);
176
177 cancel_work_sync(&assigned_dev->interrupt_work);
178
179 for (i = 0; i < assigned_dev->entries_nr; i++)
180 free_irq(assigned_dev->host_msix_entries[i].vector,
181 (void *)assigned_dev);
182
183 assigned_dev->entries_nr = 0;
184 kfree(assigned_dev->host_msix_entries);
185 kfree(assigned_dev->guest_msix_entries);
186 pci_disable_msix(assigned_dev->dev);
187 } else {
188 /* Deal with MSI and INTx */
189 disable_irq_nosync(assigned_dev->host_irq);
190 cancel_work_sync(&assigned_dev->interrupt_work);
191
192 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
193
194 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
195 pci_disable_msi(assigned_dev->dev);
196 }
197
198 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
199}
200
201static int kvm_deassign_irq(struct kvm *kvm,
202 struct kvm_assigned_dev_kernel *assigned_dev,
203 unsigned long irq_requested_type)
204{
205 unsigned long guest_irq_type, host_irq_type;
206
207 if (!irqchip_in_kernel(kvm))
208 return -EINVAL;
209 /* no irq assignment to deassign */
210 if (!assigned_dev->irq_requested_type)
211 return -ENXIO;
212
213 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
214 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
215
216 if (host_irq_type)
217 deassign_host_irq(kvm, assigned_dev);
218 if (guest_irq_type)
219 deassign_guest_irq(kvm, assigned_dev);
220
221 return 0;
222}
223
224static void kvm_free_assigned_irq(struct kvm *kvm,
225 struct kvm_assigned_dev_kernel *assigned_dev)
226{
227 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
228}
229
230static void kvm_free_assigned_device(struct kvm *kvm,
231 struct kvm_assigned_dev_kernel
232 *assigned_dev)
233{
234 kvm_free_assigned_irq(kvm, assigned_dev);
235
236 pci_reset_function(assigned_dev->dev);
237
238 pci_release_regions(assigned_dev->dev);
239 pci_disable_device(assigned_dev->dev);
240 pci_dev_put(assigned_dev->dev);
241
242 list_del(&assigned_dev->list);
243 kfree(assigned_dev);
244}
245
246void kvm_free_all_assigned_devices(struct kvm *kvm)
247{
248 struct list_head *ptr, *ptr2;
249 struct kvm_assigned_dev_kernel *assigned_dev;
250
251 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
252 assigned_dev = list_entry(ptr,
253 struct kvm_assigned_dev_kernel,
254 list);
255
256 kvm_free_assigned_device(kvm, assigned_dev);
257 }
258}
259
260static int assigned_device_enable_host_intx(struct kvm *kvm,
261 struct kvm_assigned_dev_kernel *dev)
262{
263 dev->host_irq = dev->dev->irq;
264 /* Even though this is PCI, we don't want to use shared
265 * interrupts. Sharing host devices with guest-assigned devices
266 * on the same interrupt line is not a happy situation: there
267 * are going to be long delays in accepting, acking, etc.
268 */
269 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
270 0, "kvm_assigned_intx_device", (void *)dev))
271 return -EIO;
272 return 0;
273}
274
275#ifdef __KVM_HAVE_MSI
276static int assigned_device_enable_host_msi(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *dev)
278{
279 int r;
280
281 if (!dev->dev->msi_enabled) {
282 r = pci_enable_msi(dev->dev);
283 if (r)
284 return r;
285 }
286
287 dev->host_irq = dev->dev->irq;
288 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
289 "kvm_assigned_msi_device", (void *)dev)) {
290 pci_disable_msi(dev->dev);
291 return -EIO;
292 }
293
294 return 0;
295}
296#endif
297
298#ifdef __KVM_HAVE_MSIX
299static int assigned_device_enable_host_msix(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *dev)
301{
302 int i, r = -EINVAL;
303
304 /* host_msix_entries and guest_msix_entries should have been
305 * initialized */
306 if (dev->entries_nr == 0)
307 return r;
308
309 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
310 if (r)
311 return r;
312
313 for (i = 0; i < dev->entries_nr; i++) {
314 r = request_irq(dev->host_msix_entries[i].vector,
315 kvm_assigned_dev_intr, 0,
316 "kvm_assigned_msix_device",
317 (void *)dev);
318 /* FIXME: free requested_irq's on failure */
319 if (r)
320 return r;
321 }
322
323 return 0;
324}
325
326#endif
327
328static int assigned_device_enable_guest_intx(struct kvm *kvm,
329 struct kvm_assigned_dev_kernel *dev,
330 struct kvm_assigned_irq *irq)
331{
332 dev->guest_irq = irq->guest_irq;
333 dev->ack_notifier.gsi = irq->guest_irq;
334 return 0;
335}
336
337#ifdef __KVM_HAVE_MSI
338static int assigned_device_enable_guest_msi(struct kvm *kvm,
339 struct kvm_assigned_dev_kernel *dev,
340 struct kvm_assigned_irq *irq)
341{
342 dev->guest_irq = irq->guest_irq;
343 dev->ack_notifier.gsi = -1;
344 dev->host_irq_disabled = false;
345 return 0;
346}
347#endif
348
349#ifdef __KVM_HAVE_MSIX
350static int assigned_device_enable_guest_msix(struct kvm *kvm,
351 struct kvm_assigned_dev_kernel *dev,
352 struct kvm_assigned_irq *irq)
353{
354 dev->guest_irq = irq->guest_irq;
355 dev->ack_notifier.gsi = -1;
356 dev->host_irq_disabled = false;
357 return 0;
358}
359#endif
360
361static int assign_host_irq(struct kvm *kvm,
362 struct kvm_assigned_dev_kernel *dev,
363 __u32 host_irq_type)
364{
365 int r = -EEXIST;
366
367 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
368 return r;
369
370 switch (host_irq_type) {
371 case KVM_DEV_IRQ_HOST_INTX:
372 r = assigned_device_enable_host_intx(kvm, dev);
373 break;
374#ifdef __KVM_HAVE_MSI
375 case KVM_DEV_IRQ_HOST_MSI:
376 r = assigned_device_enable_host_msi(kvm, dev);
377 break;
378#endif
379#ifdef __KVM_HAVE_MSIX
380 case KVM_DEV_IRQ_HOST_MSIX:
381 r = assigned_device_enable_host_msix(kvm, dev);
382 break;
383#endif
384 default:
385 r = -EINVAL;
386 }
387
388 if (!r)
389 dev->irq_requested_type |= host_irq_type;
390
391 return r;
392}
393
394static int assign_guest_irq(struct kvm *kvm,
395 struct kvm_assigned_dev_kernel *dev,
396 struct kvm_assigned_irq *irq,
397 unsigned long guest_irq_type)
398{
399 int id;
400 int r = -EEXIST;
401
402 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
403 return r;
404
405 id = kvm_request_irq_source_id(kvm);
406 if (id < 0)
407 return id;
408
409 dev->irq_source_id = id;
410
411 switch (guest_irq_type) {
412 case KVM_DEV_IRQ_GUEST_INTX:
413 r = assigned_device_enable_guest_intx(kvm, dev, irq);
414 break;
415#ifdef __KVM_HAVE_MSI
416 case KVM_DEV_IRQ_GUEST_MSI:
417 r = assigned_device_enable_guest_msi(kvm, dev, irq);
418 break;
419#endif
420#ifdef __KVM_HAVE_MSIX
421 case KVM_DEV_IRQ_GUEST_MSIX:
422 r = assigned_device_enable_guest_msix(kvm, dev, irq);
423 break;
424#endif
425 default:
426 r = -EINVAL;
427 }
428
429 if (!r) {
430 dev->irq_requested_type |= guest_irq_type;
431 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
432 } else
433 kvm_free_irq_source_id(kvm, dev->irq_source_id);
434
435 return r;
436}
437
438/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
439static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
440 struct kvm_assigned_irq *assigned_irq)
441{
442 int r = -EINVAL;
443 struct kvm_assigned_dev_kernel *match;
444 unsigned long host_irq_type, guest_irq_type;
445
446 if (!capable(CAP_SYS_RAWIO))
447 return -EPERM;
448
449 if (!irqchip_in_kernel(kvm))
450 return r;
451
452 mutex_lock(&kvm->lock);
453 r = -ENODEV;
454 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
455 assigned_irq->assigned_dev_id);
456 if (!match)
457 goto out;
458
459 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
460 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
461
462 r = -EINVAL;
463 /* can only assign one type at a time */
464 if (hweight_long(host_irq_type) > 1)
465 goto out;
466 if (hweight_long(guest_irq_type) > 1)
467 goto out;
468 if (host_irq_type == 0 && guest_irq_type == 0)
469 goto out;
470
471 r = 0;
472 if (host_irq_type)
473 r = assign_host_irq(kvm, match, host_irq_type);
474 if (r)
475 goto out;
476
477 if (guest_irq_type)
478 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
479out:
480 mutex_unlock(&kvm->lock);
481 return r;
482}
483
484static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
485 struct kvm_assigned_irq
486 *assigned_irq)
487{
488 int r = -ENODEV;
489 struct kvm_assigned_dev_kernel *match;
490
491 mutex_lock(&kvm->lock);
492
493 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
494 assigned_irq->assigned_dev_id);
495 if (!match)
496 goto out;
497
498 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
499out:
500 mutex_unlock(&kvm->lock);
501 return r;
502}
503
504static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
505 struct kvm_assigned_pci_dev *assigned_dev)
506{
507 int r = 0;
508 struct kvm_assigned_dev_kernel *match;
509 struct pci_dev *dev;
510
511 down_read(&kvm->slots_lock);
512 mutex_lock(&kvm->lock);
513
514 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
515 assigned_dev->assigned_dev_id);
516 if (match) {
517 /* device already assigned */
518 r = -EEXIST;
519 goto out;
520 }
521
522 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
523 if (match == NULL) {
524 printk(KERN_INFO "%s: Couldn't allocate memory\n",
525 __func__);
526 r = -ENOMEM;
527 goto out;
528 }
529 dev = pci_get_bus_and_slot(assigned_dev->busnr,
530 assigned_dev->devfn);
531 if (!dev) {
532 printk(KERN_INFO "%s: host device not found\n", __func__);
533 r = -EINVAL;
534 goto out_free;
535 }
536 if (pci_enable_device(dev)) {
537 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
538 r = -EBUSY;
539 goto out_put;
540 }
541 r = pci_request_regions(dev, "kvm_assigned_device");
542 if (r) {
543 printk(KERN_INFO "%s: Could not get access to device regions\n",
544 __func__);
545 goto out_disable;
546 }
547
548 pci_reset_function(dev);
549
550 match->assigned_dev_id = assigned_dev->assigned_dev_id;
551 match->host_busnr = assigned_dev->busnr;
552 match->host_devfn = assigned_dev->devfn;
553 match->flags = assigned_dev->flags;
554 match->dev = dev;
555 spin_lock_init(&match->assigned_dev_lock);
556 match->irq_source_id = -1;
557 match->kvm = kvm;
558 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
559 INIT_WORK(&match->interrupt_work,
560 kvm_assigned_dev_interrupt_work_handler);
561
562 list_add(&match->list, &kvm->arch.assigned_dev_head);
563
564 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
565 if (!kvm->arch.iommu_domain) {
566 r = kvm_iommu_map_guest(kvm);
567 if (r)
568 goto out_list_del;
569 }
570 r = kvm_assign_device(kvm, match);
571 if (r)
572 goto out_list_del;
573 }
574
575out:
576 mutex_unlock(&kvm->lock);
577 up_read(&kvm->slots_lock);
578 return r;
579out_list_del:
580 list_del(&match->list);
581 pci_release_regions(dev);
582out_disable:
583 pci_disable_device(dev);
584out_put:
585 pci_dev_put(dev);
586out_free:
587 kfree(match);
588 mutex_unlock(&kvm->lock);
589 up_read(&kvm->slots_lock);
590 return r;
591}
592
593static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
594 struct kvm_assigned_pci_dev *assigned_dev)
595{
596 int r = 0;
597 struct kvm_assigned_dev_kernel *match;
598
599 mutex_lock(&kvm->lock);
600
601 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
602 assigned_dev->assigned_dev_id);
603 if (!match) {
604 printk(KERN_INFO "%s: device hasn't been assigned before, "
605 "so cannot be deassigned\n", __func__);
606 r = -EINVAL;
607 goto out;
608 }
609
610 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
611 kvm_deassign_device(kvm, match);
612
613 kvm_free_assigned_device(kvm, match);
614
615out:
616 mutex_unlock(&kvm->lock);
617 return r;
618}
619
620
621#ifdef __KVM_HAVE_MSIX
622static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
623 struct kvm_assigned_msix_nr *entry_nr)
624{
625 int r = 0;
626 struct kvm_assigned_dev_kernel *adev;
627
628 mutex_lock(&kvm->lock);
629
630 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
631 entry_nr->assigned_dev_id);
632 if (!adev) {
633 r = -EINVAL;
634 goto msix_nr_out;
635 }
636
637 if (adev->entries_nr == 0) {
638 adev->entries_nr = entry_nr->entry_nr;
639 if (adev->entries_nr == 0 ||
640 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
641 r = -EINVAL;
642 goto msix_nr_out;
643 }
644
645 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
646 entry_nr->entry_nr,
647 GFP_KERNEL);
648 if (!adev->host_msix_entries) {
649 r = -ENOMEM;
650 goto msix_nr_out;
651 }
652 adev->guest_msix_entries = kzalloc(
653 sizeof(struct kvm_guest_msix_entry) *
654 entry_nr->entry_nr, GFP_KERNEL);
655 if (!adev->guest_msix_entries) {
656 kfree(adev->host_msix_entries);
657 r = -ENOMEM;
658 goto msix_nr_out;
659 }
660 } else /* Not allowed set MSI-X number twice */
661 r = -EINVAL;
662msix_nr_out:
663 mutex_unlock(&kvm->lock);
664 return r;
665}
666
667static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
668 struct kvm_assigned_msix_entry *entry)
669{
670 int r = 0, i;
671 struct kvm_assigned_dev_kernel *adev;
672
673 mutex_lock(&kvm->lock);
674
675 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
676 entry->assigned_dev_id);
677
678 if (!adev) {
679 r = -EINVAL;
680 goto msix_entry_out;
681 }
682
683 for (i = 0; i < adev->entries_nr; i++)
684 if (adev->guest_msix_entries[i].vector == 0 ||
685 adev->guest_msix_entries[i].entry == entry->entry) {
686 adev->guest_msix_entries[i].entry = entry->entry;
687 adev->guest_msix_entries[i].vector = entry->gsi;
688 adev->host_msix_entries[i].entry = entry->entry;
689 break;
690 }
691 if (i == adev->entries_nr) {
692 r = -ENOSPC;
693 goto msix_entry_out;
694 }
695
696msix_entry_out:
697 mutex_unlock(&kvm->lock);
698
699 return r;
700}
701#endif
702
703long kvm_vm_ioctl_assigned_device(struct kvm *kvm, unsigned ioctl,
704 unsigned long arg)
705{
706 void __user *argp = (void __user *)arg;
707 int r = -ENOTTY;
708
709 switch (ioctl) {
710 case KVM_ASSIGN_PCI_DEVICE: {
711 struct kvm_assigned_pci_dev assigned_dev;
712
713 r = -EFAULT;
714 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
715 goto out;
716 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
717 if (r)
718 goto out;
719 break;
720 }
721 case KVM_ASSIGN_IRQ: {
722 r = -EOPNOTSUPP;
723 break;
724 }
725#ifdef KVM_CAP_ASSIGN_DEV_IRQ
726 case KVM_ASSIGN_DEV_IRQ: {
727 struct kvm_assigned_irq assigned_irq;
728
729 r = -EFAULT;
730 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
731 goto out;
732 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
733 if (r)
734 goto out;
735 break;
736 }
737 case KVM_DEASSIGN_DEV_IRQ: {
738 struct kvm_assigned_irq assigned_irq;
739
740 r = -EFAULT;
741 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
742 goto out;
743 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
744 if (r)
745 goto out;
746 break;
747 }
748#endif
749#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
750 case KVM_DEASSIGN_PCI_DEVICE: {
751 struct kvm_assigned_pci_dev assigned_dev;
752
753 r = -EFAULT;
754 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
755 goto out;
756 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
757 if (r)
758 goto out;
759 break;
760 }
761#endif
762#ifdef KVM_CAP_IRQ_ROUTING
763 case KVM_SET_GSI_ROUTING: {
764 struct kvm_irq_routing routing;
765 struct kvm_irq_routing __user *urouting;
766 struct kvm_irq_routing_entry *entries;
767
768 r = -EFAULT;
769 if (copy_from_user(&routing, argp, sizeof(routing)))
770 goto out;
771 r = -EINVAL;
772 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
773 goto out;
774 if (routing.flags)
775 goto out;
776 r = -ENOMEM;
777 entries = vmalloc(routing.nr * sizeof(*entries));
778 if (!entries)
779 goto out;
780 r = -EFAULT;
781 urouting = argp;
782 if (copy_from_user(entries, urouting->entries,
783 routing.nr * sizeof(*entries)))
784 goto out_free_irq_routing;
785 r = kvm_set_irq_routing(kvm, entries, routing.nr,
786 routing.flags);
787 out_free_irq_routing:
788 vfree(entries);
789 break;
790 }
791#endif /* KVM_CAP_IRQ_ROUTING */
792#ifdef __KVM_HAVE_MSIX
793 case KVM_ASSIGN_SET_MSIX_NR: {
794 struct kvm_assigned_msix_nr entry_nr;
795 r = -EFAULT;
796 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
797 goto out;
798 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
799 if (r)
800 goto out;
801 break;
802 }
803 case KVM_ASSIGN_SET_MSIX_ENTRY: {
804 struct kvm_assigned_msix_entry entry;
805 r = -EFAULT;
806 if (copy_from_user(&entry, argp, sizeof entry))
807 goto out;
808 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
809 if (r)
810 goto out;
811 break;
812 }
813#endif
814 }
815out:
816 return r;
817}
818
diff --git a/virt/kvm/eventfd.c b/virt/kvm/eventfd.c
index bb4ebd89b9f..30f70fd511c 100644
--- a/virt/kvm/eventfd.c
+++ b/virt/kvm/eventfd.c
@@ -61,10 +61,8 @@ irqfd_inject(struct work_struct *work)
61 struct _irqfd *irqfd = container_of(work, struct _irqfd, inject); 61 struct _irqfd *irqfd = container_of(work, struct _irqfd, inject);
62 struct kvm *kvm = irqfd->kvm; 62 struct kvm *kvm = irqfd->kvm;
63 63
64 mutex_lock(&kvm->irq_lock);
65 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1); 64 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 1);
66 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0); 65 kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID, irqfd->gsi, 0);
67 mutex_unlock(&kvm->irq_lock);
68} 66}
69 67
70/* 68/*
diff --git a/virt/kvm/ioapic.c b/virt/kvm/ioapic.c
index 9fe140bb38e..38a2d20b89d 100644
--- a/virt/kvm/ioapic.c
+++ b/virt/kvm/ioapic.c
@@ -182,6 +182,7 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
182 union kvm_ioapic_redirect_entry entry; 182 union kvm_ioapic_redirect_entry entry;
183 int ret = 1; 183 int ret = 1;
184 184
185 mutex_lock(&ioapic->lock);
185 if (irq >= 0 && irq < IOAPIC_NUM_PINS) { 186 if (irq >= 0 && irq < IOAPIC_NUM_PINS) {
186 entry = ioapic->redirtbl[irq]; 187 entry = ioapic->redirtbl[irq];
187 level ^= entry.fields.polarity; 188 level ^= entry.fields.polarity;
@@ -198,34 +199,51 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level)
198 } 199 }
199 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0); 200 trace_kvm_ioapic_set_irq(entry.bits, irq, ret == 0);
200 } 201 }
202 mutex_unlock(&ioapic->lock);
203
201 return ret; 204 return ret;
202} 205}
203 206
204static void __kvm_ioapic_update_eoi(struct kvm_ioapic *ioapic, int pin, 207static void __kvm_ioapic_update_eoi(struct kvm_ioapic *ioapic, int vector,
205 int trigger_mode) 208 int trigger_mode)
206{ 209{
207 union kvm_ioapic_redirect_entry *ent; 210 int i;
211
212 for (i = 0; i < IOAPIC_NUM_PINS; i++) {
213 union kvm_ioapic_redirect_entry *ent = &ioapic->redirtbl[i];
208 214
209 ent = &ioapic->redirtbl[pin]; 215 if (ent->fields.vector != vector)
216 continue;
210 217
211 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, pin); 218 /*
219 * We are dropping lock while calling ack notifiers because ack
220 * notifier callbacks for assigned devices call into IOAPIC
221 * recursively. Since remote_irr is cleared only after call
222 * to notifiers if the same vector will be delivered while lock
223 * is dropped it will be put into irr and will be delivered
224 * after ack notifier returns.
225 */
226 mutex_unlock(&ioapic->lock);
227 kvm_notify_acked_irq(ioapic->kvm, KVM_IRQCHIP_IOAPIC, i);
228 mutex_lock(&ioapic->lock);
229
230 if (trigger_mode != IOAPIC_LEVEL_TRIG)
231 continue;
212 232
213 if (trigger_mode == IOAPIC_LEVEL_TRIG) {
214 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG); 233 ASSERT(ent->fields.trig_mode == IOAPIC_LEVEL_TRIG);
215 ent->fields.remote_irr = 0; 234 ent->fields.remote_irr = 0;
216 if (!ent->fields.mask && (ioapic->irr & (1 << pin))) 235 if (!ent->fields.mask && (ioapic->irr & (1 << i)))
217 ioapic_service(ioapic, pin); 236 ioapic_service(ioapic, i);
218 } 237 }
219} 238}
220 239
221void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode) 240void kvm_ioapic_update_eoi(struct kvm *kvm, int vector, int trigger_mode)
222{ 241{
223 struct kvm_ioapic *ioapic = kvm->arch.vioapic; 242 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
224 int i;
225 243
226 for (i = 0; i < IOAPIC_NUM_PINS; i++) 244 mutex_lock(&ioapic->lock);
227 if (ioapic->redirtbl[i].fields.vector == vector) 245 __kvm_ioapic_update_eoi(ioapic, vector, trigger_mode);
228 __kvm_ioapic_update_eoi(ioapic, i, trigger_mode); 246 mutex_unlock(&ioapic->lock);
229} 247}
230 248
231static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev) 249static inline struct kvm_ioapic *to_ioapic(struct kvm_io_device *dev)
@@ -250,8 +268,8 @@ static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
250 ioapic_debug("addr %lx\n", (unsigned long)addr); 268 ioapic_debug("addr %lx\n", (unsigned long)addr);
251 ASSERT(!(addr & 0xf)); /* check alignment */ 269 ASSERT(!(addr & 0xf)); /* check alignment */
252 270
253 mutex_lock(&ioapic->kvm->irq_lock);
254 addr &= 0xff; 271 addr &= 0xff;
272 mutex_lock(&ioapic->lock);
255 switch (addr) { 273 switch (addr) {
256 case IOAPIC_REG_SELECT: 274 case IOAPIC_REG_SELECT:
257 result = ioapic->ioregsel; 275 result = ioapic->ioregsel;
@@ -265,6 +283,8 @@ static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
265 result = 0; 283 result = 0;
266 break; 284 break;
267 } 285 }
286 mutex_unlock(&ioapic->lock);
287
268 switch (len) { 288 switch (len) {
269 case 8: 289 case 8:
270 *(u64 *) val = result; 290 *(u64 *) val = result;
@@ -277,7 +297,6 @@ static int ioapic_mmio_read(struct kvm_io_device *this, gpa_t addr, int len,
277 default: 297 default:
278 printk(KERN_WARNING "ioapic: wrong length %d\n", len); 298 printk(KERN_WARNING "ioapic: wrong length %d\n", len);
279 } 299 }
280 mutex_unlock(&ioapic->kvm->irq_lock);
281 return 0; 300 return 0;
282} 301}
283 302
@@ -293,15 +312,15 @@ static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
293 (void*)addr, len, val); 312 (void*)addr, len, val);
294 ASSERT(!(addr & 0xf)); /* check alignment */ 313 ASSERT(!(addr & 0xf)); /* check alignment */
295 314
296 mutex_lock(&ioapic->kvm->irq_lock);
297 if (len == 4 || len == 8) 315 if (len == 4 || len == 8)
298 data = *(u32 *) val; 316 data = *(u32 *) val;
299 else { 317 else {
300 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len); 318 printk(KERN_WARNING "ioapic: Unsupported size %d\n", len);
301 goto unlock; 319 return 0;
302 } 320 }
303 321
304 addr &= 0xff; 322 addr &= 0xff;
323 mutex_lock(&ioapic->lock);
305 switch (addr) { 324 switch (addr) {
306 case IOAPIC_REG_SELECT: 325 case IOAPIC_REG_SELECT:
307 ioapic->ioregsel = data; 326 ioapic->ioregsel = data;
@@ -312,15 +331,14 @@ static int ioapic_mmio_write(struct kvm_io_device *this, gpa_t addr, int len,
312 break; 331 break;
313#ifdef CONFIG_IA64 332#ifdef CONFIG_IA64
314 case IOAPIC_REG_EOI: 333 case IOAPIC_REG_EOI:
315 kvm_ioapic_update_eoi(ioapic->kvm, data, IOAPIC_LEVEL_TRIG); 334 __kvm_ioapic_update_eoi(ioapic, data, IOAPIC_LEVEL_TRIG);
316 break; 335 break;
317#endif 336#endif
318 337
319 default: 338 default:
320 break; 339 break;
321 } 340 }
322unlock: 341 mutex_unlock(&ioapic->lock);
323 mutex_unlock(&ioapic->kvm->irq_lock);
324 return 0; 342 return 0;
325} 343}
326 344
@@ -349,6 +367,7 @@ int kvm_ioapic_init(struct kvm *kvm)
349 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL); 367 ioapic = kzalloc(sizeof(struct kvm_ioapic), GFP_KERNEL);
350 if (!ioapic) 368 if (!ioapic)
351 return -ENOMEM; 369 return -ENOMEM;
370 mutex_init(&ioapic->lock);
352 kvm->arch.vioapic = ioapic; 371 kvm->arch.vioapic = ioapic;
353 kvm_ioapic_reset(ioapic); 372 kvm_ioapic_reset(ioapic);
354 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops); 373 kvm_iodevice_init(&ioapic->dev, &ioapic_mmio_ops);
@@ -360,3 +379,26 @@ int kvm_ioapic_init(struct kvm *kvm)
360 return ret; 379 return ret;
361} 380}
362 381
382int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
383{
384 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
385 if (!ioapic)
386 return -EINVAL;
387
388 mutex_lock(&ioapic->lock);
389 memcpy(state, ioapic, sizeof(struct kvm_ioapic_state));
390 mutex_unlock(&ioapic->lock);
391 return 0;
392}
393
394int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state)
395{
396 struct kvm_ioapic *ioapic = ioapic_irqchip(kvm);
397 if (!ioapic)
398 return -EINVAL;
399
400 mutex_lock(&ioapic->lock);
401 memcpy(ioapic, state, sizeof(struct kvm_ioapic_state));
402 mutex_unlock(&ioapic->lock);
403 return 0;
404}
diff --git a/virt/kvm/ioapic.h b/virt/kvm/ioapic.h
index 7080b713c16..419c43b667a 100644
--- a/virt/kvm/ioapic.h
+++ b/virt/kvm/ioapic.h
@@ -41,9 +41,11 @@ struct kvm_ioapic {
41 u32 irr; 41 u32 irr;
42 u32 pad; 42 u32 pad;
43 union kvm_ioapic_redirect_entry redirtbl[IOAPIC_NUM_PINS]; 43 union kvm_ioapic_redirect_entry redirtbl[IOAPIC_NUM_PINS];
44 unsigned long irq_states[IOAPIC_NUM_PINS];
44 struct kvm_io_device dev; 45 struct kvm_io_device dev;
45 struct kvm *kvm; 46 struct kvm *kvm;
46 void (*ack_notifier)(void *opaque, int irq); 47 void (*ack_notifier)(void *opaque, int irq);
48 struct mutex lock;
47}; 49};
48 50
49#ifdef DEBUG 51#ifdef DEBUG
@@ -73,4 +75,7 @@ int kvm_ioapic_set_irq(struct kvm_ioapic *ioapic, int irq, int level);
73void kvm_ioapic_reset(struct kvm_ioapic *ioapic); 75void kvm_ioapic_reset(struct kvm_ioapic *ioapic);
74int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src, 76int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
75 struct kvm_lapic_irq *irq); 77 struct kvm_lapic_irq *irq);
78int kvm_get_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state);
79int kvm_set_ioapic(struct kvm *kvm, struct kvm_ioapic_state *state);
80
76#endif 81#endif
diff --git a/virt/kvm/irq_comm.c b/virt/kvm/irq_comm.c
index 001663ff401..9b077342ab5 100644
--- a/virt/kvm/irq_comm.c
+++ b/virt/kvm/irq_comm.c
@@ -31,20 +31,39 @@
31 31
32#include "ioapic.h" 32#include "ioapic.h"
33 33
34static inline int kvm_irq_line_state(unsigned long *irq_state,
35 int irq_source_id, int level)
36{
37 /* Logical OR for level trig interrupt */
38 if (level)
39 set_bit(irq_source_id, irq_state);
40 else
41 clear_bit(irq_source_id, irq_state);
42
43 return !!(*irq_state);
44}
45
34static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e, 46static int kvm_set_pic_irq(struct kvm_kernel_irq_routing_entry *e,
35 struct kvm *kvm, int level) 47 struct kvm *kvm, int irq_source_id, int level)
36{ 48{
37#ifdef CONFIG_X86 49#ifdef CONFIG_X86
38 return kvm_pic_set_irq(pic_irqchip(kvm), e->irqchip.pin, level); 50 struct kvm_pic *pic = pic_irqchip(kvm);
51 level = kvm_irq_line_state(&pic->irq_states[e->irqchip.pin],
52 irq_source_id, level);
53 return kvm_pic_set_irq(pic, e->irqchip.pin, level);
39#else 54#else
40 return -1; 55 return -1;
41#endif 56#endif
42} 57}
43 58
44static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e, 59static int kvm_set_ioapic_irq(struct kvm_kernel_irq_routing_entry *e,
45 struct kvm *kvm, int level) 60 struct kvm *kvm, int irq_source_id, int level)
46{ 61{
47 return kvm_ioapic_set_irq(kvm->arch.vioapic, e->irqchip.pin, level); 62 struct kvm_ioapic *ioapic = kvm->arch.vioapic;
63 level = kvm_irq_line_state(&ioapic->irq_states[e->irqchip.pin],
64 irq_source_id, level);
65
66 return kvm_ioapic_set_irq(ioapic, e->irqchip.pin, level);
48} 67}
49 68
50inline static bool kvm_is_dm_lowest_prio(struct kvm_lapic_irq *irq) 69inline static bool kvm_is_dm_lowest_prio(struct kvm_lapic_irq *irq)
@@ -63,8 +82,6 @@ int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
63 int i, r = -1; 82 int i, r = -1;
64 struct kvm_vcpu *vcpu, *lowest = NULL; 83 struct kvm_vcpu *vcpu, *lowest = NULL;
65 84
66 WARN_ON(!mutex_is_locked(&kvm->irq_lock));
67
68 if (irq->dest_mode == 0 && irq->dest_id == 0xff && 85 if (irq->dest_mode == 0 && irq->dest_id == 0xff &&
69 kvm_is_dm_lowest_prio(irq)) 86 kvm_is_dm_lowest_prio(irq))
70 printk(KERN_INFO "kvm: apic: phys broadcast and lowest prio\n"); 87 printk(KERN_INFO "kvm: apic: phys broadcast and lowest prio\n");
@@ -96,10 +113,13 @@ int kvm_irq_delivery_to_apic(struct kvm *kvm, struct kvm_lapic *src,
96} 113}
97 114
98static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e, 115static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
99 struct kvm *kvm, int level) 116 struct kvm *kvm, int irq_source_id, int level)
100{ 117{
101 struct kvm_lapic_irq irq; 118 struct kvm_lapic_irq irq;
102 119
120 if (!level)
121 return -1;
122
103 trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data); 123 trace_kvm_msi_set_irq(e->msi.address_lo, e->msi.data);
104 124
105 irq.dest_id = (e->msi.address_lo & 125 irq.dest_id = (e->msi.address_lo &
@@ -116,78 +136,67 @@ static int kvm_set_msi(struct kvm_kernel_irq_routing_entry *e,
116 return kvm_irq_delivery_to_apic(kvm, NULL, &irq); 136 return kvm_irq_delivery_to_apic(kvm, NULL, &irq);
117} 137}
118 138
119/* This should be called with the kvm->irq_lock mutex held 139/*
120 * Return value: 140 * Return value:
121 * < 0 Interrupt was ignored (masked or not delivered for other reasons) 141 * < 0 Interrupt was ignored (masked or not delivered for other reasons)
122 * = 0 Interrupt was coalesced (previous irq is still pending) 142 * = 0 Interrupt was coalesced (previous irq is still pending)
123 * > 0 Number of CPUs interrupt was delivered to 143 * > 0 Number of CPUs interrupt was delivered to
124 */ 144 */
125int kvm_set_irq(struct kvm *kvm, int irq_source_id, int irq, int level) 145int kvm_set_irq(struct kvm *kvm, int irq_source_id, u32 irq, int level)
126{ 146{
127 struct kvm_kernel_irq_routing_entry *e; 147 struct kvm_kernel_irq_routing_entry *e, irq_set[KVM_NR_IRQCHIPS];
128 unsigned long *irq_state, sig_level; 148 int ret = -1, i = 0;
129 int ret = -1; 149 struct kvm_irq_routing_table *irq_rt;
150 struct hlist_node *n;
130 151
131 trace_kvm_set_irq(irq, level, irq_source_id); 152 trace_kvm_set_irq(irq, level, irq_source_id);
132 153
133 WARN_ON(!mutex_is_locked(&kvm->irq_lock));
134
135 if (irq < KVM_IOAPIC_NUM_PINS) {
136 irq_state = (unsigned long *)&kvm->arch.irq_states[irq];
137
138 /* Logical OR for level trig interrupt */
139 if (level)
140 set_bit(irq_source_id, irq_state);
141 else
142 clear_bit(irq_source_id, irq_state);
143 sig_level = !!(*irq_state);
144 } else if (!level)
145 return ret;
146 else /* Deal with MSI/MSI-X */
147 sig_level = 1;
148
149 /* Not possible to detect if the guest uses the PIC or the 154 /* Not possible to detect if the guest uses the PIC or the
150 * IOAPIC. So set the bit in both. The guest will ignore 155 * IOAPIC. So set the bit in both. The guest will ignore
151 * writes to the unused one. 156 * writes to the unused one.
152 */ 157 */
153 list_for_each_entry(e, &kvm->irq_routing, link) 158 rcu_read_lock();
154 if (e->gsi == irq) { 159 irq_rt = rcu_dereference(kvm->irq_routing);
155 int r = e->set(e, kvm, sig_level); 160 if (irq < irq_rt->nr_rt_entries)
156 if (r < 0) 161 hlist_for_each_entry(e, n, &irq_rt->map[irq], link)
157 continue; 162 irq_set[i++] = *e;
163 rcu_read_unlock();
164
165 while(i--) {
166 int r;
167 r = irq_set[i].set(&irq_set[i], kvm, irq_source_id, level);
168 if (r < 0)
169 continue;
170
171 ret = r + ((ret < 0) ? 0 : ret);
172 }
158 173
159 ret = r + ((ret < 0) ? 0 : ret);
160 }
161 return ret; 174 return ret;
162} 175}
163 176
164void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin) 177void kvm_notify_acked_irq(struct kvm *kvm, unsigned irqchip, unsigned pin)
165{ 178{
166 struct kvm_kernel_irq_routing_entry *e;
167 struct kvm_irq_ack_notifier *kian; 179 struct kvm_irq_ack_notifier *kian;
168 struct hlist_node *n; 180 struct hlist_node *n;
169 unsigned gsi = pin; 181 int gsi;
170 182
171 trace_kvm_ack_irq(irqchip, pin); 183 trace_kvm_ack_irq(irqchip, pin);
172 184
173 list_for_each_entry(e, &kvm->irq_routing, link) 185 rcu_read_lock();
174 if (e->type == KVM_IRQ_ROUTING_IRQCHIP && 186 gsi = rcu_dereference(kvm->irq_routing)->chip[irqchip][pin];
175 e->irqchip.irqchip == irqchip && 187 if (gsi != -1)
176 e->irqchip.pin == pin) { 188 hlist_for_each_entry_rcu(kian, n, &kvm->irq_ack_notifier_list,
177 gsi = e->gsi; 189 link)
178 break; 190 if (kian->gsi == gsi)
179 } 191 kian->irq_acked(kian);
180 192 rcu_read_unlock();
181 hlist_for_each_entry(kian, n, &kvm->arch.irq_ack_notifier_list, link)
182 if (kian->gsi == gsi)
183 kian->irq_acked(kian);
184} 193}
185 194
186void kvm_register_irq_ack_notifier(struct kvm *kvm, 195void kvm_register_irq_ack_notifier(struct kvm *kvm,
187 struct kvm_irq_ack_notifier *kian) 196 struct kvm_irq_ack_notifier *kian)
188{ 197{
189 mutex_lock(&kvm->irq_lock); 198 mutex_lock(&kvm->irq_lock);
190 hlist_add_head(&kian->link, &kvm->arch.irq_ack_notifier_list); 199 hlist_add_head_rcu(&kian->link, &kvm->irq_ack_notifier_list);
191 mutex_unlock(&kvm->irq_lock); 200 mutex_unlock(&kvm->irq_lock);
192} 201}
193 202
@@ -195,8 +204,9 @@ void kvm_unregister_irq_ack_notifier(struct kvm *kvm,
195 struct kvm_irq_ack_notifier *kian) 204 struct kvm_irq_ack_notifier *kian)
196{ 205{
197 mutex_lock(&kvm->irq_lock); 206 mutex_lock(&kvm->irq_lock);
198 hlist_del_init(&kian->link); 207 hlist_del_init_rcu(&kian->link);
199 mutex_unlock(&kvm->irq_lock); 208 mutex_unlock(&kvm->irq_lock);
209 synchronize_rcu();
200} 210}
201 211
202int kvm_request_irq_source_id(struct kvm *kvm) 212int kvm_request_irq_source_id(struct kvm *kvm)
@@ -205,16 +215,17 @@ int kvm_request_irq_source_id(struct kvm *kvm)
205 int irq_source_id; 215 int irq_source_id;
206 216
207 mutex_lock(&kvm->irq_lock); 217 mutex_lock(&kvm->irq_lock);
208 irq_source_id = find_first_zero_bit(bitmap, 218 irq_source_id = find_first_zero_bit(bitmap, BITS_PER_LONG);
209 sizeof(kvm->arch.irq_sources_bitmap));
210 219
211 if (irq_source_id >= sizeof(kvm->arch.irq_sources_bitmap)) { 220 if (irq_source_id >= BITS_PER_LONG) {
212 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n"); 221 printk(KERN_WARNING "kvm: exhaust allocatable IRQ sources!\n");
213 return -EFAULT; 222 irq_source_id = -EFAULT;
223 goto unlock;
214 } 224 }
215 225
216 ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID); 226 ASSERT(irq_source_id != KVM_USERSPACE_IRQ_SOURCE_ID);
217 set_bit(irq_source_id, bitmap); 227 set_bit(irq_source_id, bitmap);
228unlock:
218 mutex_unlock(&kvm->irq_lock); 229 mutex_unlock(&kvm->irq_lock);
219 230
220 return irq_source_id; 231 return irq_source_id;
@@ -228,13 +239,23 @@ void kvm_free_irq_source_id(struct kvm *kvm, int irq_source_id)
228 239
229 mutex_lock(&kvm->irq_lock); 240 mutex_lock(&kvm->irq_lock);
230 if (irq_source_id < 0 || 241 if (irq_source_id < 0 ||
231 irq_source_id >= sizeof(kvm->arch.irq_sources_bitmap)) { 242 irq_source_id >= BITS_PER_LONG) {
232 printk(KERN_ERR "kvm: IRQ source ID out of range!\n"); 243 printk(KERN_ERR "kvm: IRQ source ID out of range!\n");
233 return; 244 goto unlock;
234 } 245 }
235 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++)
236 clear_bit(irq_source_id, &kvm->arch.irq_states[i]);
237 clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap); 246 clear_bit(irq_source_id, &kvm->arch.irq_sources_bitmap);
247 if (!irqchip_in_kernel(kvm))
248 goto unlock;
249
250 for (i = 0; i < KVM_IOAPIC_NUM_PINS; i++) {
251 clear_bit(irq_source_id, &kvm->arch.vioapic->irq_states[i]);
252 if (i >= 16)
253 continue;
254#ifdef CONFIG_X86
255 clear_bit(irq_source_id, &pic_irqchip(kvm)->irq_states[i]);
256#endif
257 }
258unlock:
238 mutex_unlock(&kvm->irq_lock); 259 mutex_unlock(&kvm->irq_lock);
239} 260}
240 261
@@ -243,7 +264,7 @@ void kvm_register_irq_mask_notifier(struct kvm *kvm, int irq,
243{ 264{
244 mutex_lock(&kvm->irq_lock); 265 mutex_lock(&kvm->irq_lock);
245 kimn->irq = irq; 266 kimn->irq = irq;
246 hlist_add_head(&kimn->link, &kvm->mask_notifier_list); 267 hlist_add_head_rcu(&kimn->link, &kvm->mask_notifier_list);
247 mutex_unlock(&kvm->irq_lock); 268 mutex_unlock(&kvm->irq_lock);
248} 269}
249 270
@@ -251,8 +272,9 @@ void kvm_unregister_irq_mask_notifier(struct kvm *kvm, int irq,
251 struct kvm_irq_mask_notifier *kimn) 272 struct kvm_irq_mask_notifier *kimn)
252{ 273{
253 mutex_lock(&kvm->irq_lock); 274 mutex_lock(&kvm->irq_lock);
254 hlist_del(&kimn->link); 275 hlist_del_rcu(&kimn->link);
255 mutex_unlock(&kvm->irq_lock); 276 mutex_unlock(&kvm->irq_lock);
277 synchronize_rcu();
256} 278}
257 279
258void kvm_fire_mask_notifiers(struct kvm *kvm, int irq, bool mask) 280void kvm_fire_mask_notifiers(struct kvm *kvm, int irq, bool mask)
@@ -260,33 +282,37 @@ void kvm_fire_mask_notifiers(struct kvm *kvm, int irq, bool mask)
260 struct kvm_irq_mask_notifier *kimn; 282 struct kvm_irq_mask_notifier *kimn;
261 struct hlist_node *n; 283 struct hlist_node *n;
262 284
263 WARN_ON(!mutex_is_locked(&kvm->irq_lock)); 285 rcu_read_lock();
264 286 hlist_for_each_entry_rcu(kimn, n, &kvm->mask_notifier_list, link)
265 hlist_for_each_entry(kimn, n, &kvm->mask_notifier_list, link)
266 if (kimn->irq == irq) 287 if (kimn->irq == irq)
267 kimn->func(kimn, mask); 288 kimn->func(kimn, mask);
268} 289 rcu_read_unlock();
269
270static void __kvm_free_irq_routing(struct list_head *irq_routing)
271{
272 struct kvm_kernel_irq_routing_entry *e, *n;
273
274 list_for_each_entry_safe(e, n, irq_routing, link)
275 kfree(e);
276} 290}
277 291
278void kvm_free_irq_routing(struct kvm *kvm) 292void kvm_free_irq_routing(struct kvm *kvm)
279{ 293{
280 mutex_lock(&kvm->irq_lock); 294 /* Called only during vm destruction. Nobody can use the pointer
281 __kvm_free_irq_routing(&kvm->irq_routing); 295 at this stage */
282 mutex_unlock(&kvm->irq_lock); 296 kfree(kvm->irq_routing);
283} 297}
284 298
285static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e, 299static int setup_routing_entry(struct kvm_irq_routing_table *rt,
300 struct kvm_kernel_irq_routing_entry *e,
286 const struct kvm_irq_routing_entry *ue) 301 const struct kvm_irq_routing_entry *ue)
287{ 302{
288 int r = -EINVAL; 303 int r = -EINVAL;
289 int delta; 304 int delta;
305 struct kvm_kernel_irq_routing_entry *ei;
306 struct hlist_node *n;
307
308 /*
309 * Do not allow GSI to be mapped to the same irqchip more than once.
310 * Allow only one to one mapping between GSI and MSI.
311 */
312 hlist_for_each_entry(ei, n, &rt->map[ue->gsi], link)
313 if (ei->type == KVM_IRQ_ROUTING_MSI ||
314 ue->u.irqchip.irqchip == ei->irqchip.irqchip)
315 return r;
290 316
291 e->gsi = ue->gsi; 317 e->gsi = ue->gsi;
292 e->type = ue->type; 318 e->type = ue->type;
@@ -309,6 +335,9 @@ static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e,
309 } 335 }
310 e->irqchip.irqchip = ue->u.irqchip.irqchip; 336 e->irqchip.irqchip = ue->u.irqchip.irqchip;
311 e->irqchip.pin = ue->u.irqchip.pin + delta; 337 e->irqchip.pin = ue->u.irqchip.pin + delta;
338 if (e->irqchip.pin >= KVM_IOAPIC_NUM_PINS)
339 goto out;
340 rt->chip[ue->u.irqchip.irqchip][e->irqchip.pin] = ue->gsi;
312 break; 341 break;
313 case KVM_IRQ_ROUTING_MSI: 342 case KVM_IRQ_ROUTING_MSI:
314 e->set = kvm_set_msi; 343 e->set = kvm_set_msi;
@@ -319,6 +348,8 @@ static int setup_routing_entry(struct kvm_kernel_irq_routing_entry *e,
319 default: 348 default:
320 goto out; 349 goto out;
321 } 350 }
351
352 hlist_add_head(&e->link, &rt->map[e->gsi]);
322 r = 0; 353 r = 0;
323out: 354out:
324 return r; 355 return r;
@@ -330,43 +361,53 @@ int kvm_set_irq_routing(struct kvm *kvm,
330 unsigned nr, 361 unsigned nr,
331 unsigned flags) 362 unsigned flags)
332{ 363{
333 struct list_head irq_list = LIST_HEAD_INIT(irq_list); 364 struct kvm_irq_routing_table *new, *old;
334 struct list_head tmp = LIST_HEAD_INIT(tmp); 365 u32 i, j, nr_rt_entries = 0;
335 struct kvm_kernel_irq_routing_entry *e = NULL;
336 unsigned i;
337 int r; 366 int r;
338 367
339 for (i = 0; i < nr; ++i) { 368 for (i = 0; i < nr; ++i) {
369 if (ue[i].gsi >= KVM_MAX_IRQ_ROUTES)
370 return -EINVAL;
371 nr_rt_entries = max(nr_rt_entries, ue[i].gsi);
372 }
373
374 nr_rt_entries += 1;
375
376 new = kzalloc(sizeof(*new) + (nr_rt_entries * sizeof(struct hlist_head))
377 + (nr * sizeof(struct kvm_kernel_irq_routing_entry)),
378 GFP_KERNEL);
379
380 if (!new)
381 return -ENOMEM;
382
383 new->rt_entries = (void *)&new->map[nr_rt_entries];
384
385 new->nr_rt_entries = nr_rt_entries;
386 for (i = 0; i < 3; i++)
387 for (j = 0; j < KVM_IOAPIC_NUM_PINS; j++)
388 new->chip[i][j] = -1;
389
390 for (i = 0; i < nr; ++i) {
340 r = -EINVAL; 391 r = -EINVAL;
341 if (ue->gsi >= KVM_MAX_IRQ_ROUTES)
342 goto out;
343 if (ue->flags) 392 if (ue->flags)
344 goto out; 393 goto out;
345 r = -ENOMEM; 394 r = setup_routing_entry(new, &new->rt_entries[i], ue);
346 e = kzalloc(sizeof(*e), GFP_KERNEL);
347 if (!e)
348 goto out;
349 r = setup_routing_entry(e, ue);
350 if (r) 395 if (r)
351 goto out; 396 goto out;
352 ++ue; 397 ++ue;
353 list_add(&e->link, &irq_list);
354 e = NULL;
355 } 398 }
356 399
357 mutex_lock(&kvm->irq_lock); 400 mutex_lock(&kvm->irq_lock);
358 list_splice(&kvm->irq_routing, &tmp); 401 old = kvm->irq_routing;
359 INIT_LIST_HEAD(&kvm->irq_routing); 402 rcu_assign_pointer(kvm->irq_routing, new);
360 list_splice(&irq_list, &kvm->irq_routing);
361 INIT_LIST_HEAD(&irq_list);
362 list_splice(&tmp, &irq_list);
363 mutex_unlock(&kvm->irq_lock); 403 mutex_unlock(&kvm->irq_lock);
404 synchronize_rcu();
364 405
406 new = old;
365 r = 0; 407 r = 0;
366 408
367out: 409out:
368 kfree(e); 410 kfree(new);
369 __kvm_free_irq_routing(&irq_list);
370 return r; 411 return r;
371} 412}
372 413
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 7495ce34734..f92ba138007 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -43,6 +43,7 @@
43#include <linux/swap.h> 43#include <linux/swap.h>
44#include <linux/bitops.h> 44#include <linux/bitops.h>
45#include <linux/spinlock.h> 45#include <linux/spinlock.h>
46#include <linux/compat.h>
46 47
47#include <asm/processor.h> 48#include <asm/processor.h>
48#include <asm/io.h> 49#include <asm/io.h>
@@ -53,12 +54,6 @@
53#include "coalesced_mmio.h" 54#include "coalesced_mmio.h"
54#endif 55#endif
55 56
56#ifdef KVM_CAP_DEVICE_ASSIGNMENT
57#include <linux/pci.h>
58#include <linux/interrupt.h>
59#include "irq.h"
60#endif
61
62#define CREATE_TRACE_POINTS 57#define CREATE_TRACE_POINTS
63#include <trace/events/kvm.h> 58#include <trace/events/kvm.h>
64 59
@@ -75,6 +70,8 @@ DEFINE_SPINLOCK(kvm_lock);
75LIST_HEAD(vm_list); 70LIST_HEAD(vm_list);
76 71
77static cpumask_var_t cpus_hardware_enabled; 72static cpumask_var_t cpus_hardware_enabled;
73static int kvm_usage_count = 0;
74static atomic_t hardware_enable_failed;
78 75
79struct kmem_cache *kvm_vcpu_cache; 76struct kmem_cache *kvm_vcpu_cache;
80EXPORT_SYMBOL_GPL(kvm_vcpu_cache); 77EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
@@ -85,615 +82,13 @@ struct dentry *kvm_debugfs_dir;
85 82
86static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, 83static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
87 unsigned long arg); 84 unsigned long arg);
85static int hardware_enable_all(void);
86static void hardware_disable_all(void);
88 87
89static bool kvm_rebooting; 88static bool kvm_rebooting;
90 89
91static bool largepages_enabled = true; 90static bool largepages_enabled = true;
92 91
93#ifdef KVM_CAP_DEVICE_ASSIGNMENT
94static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
95 int assigned_dev_id)
96{
97 struct list_head *ptr;
98 struct kvm_assigned_dev_kernel *match;
99
100 list_for_each(ptr, head) {
101 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
102 if (match->assigned_dev_id == assigned_dev_id)
103 return match;
104 }
105 return NULL;
106}
107
108static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
109 *assigned_dev, int irq)
110{
111 int i, index;
112 struct msix_entry *host_msix_entries;
113
114 host_msix_entries = assigned_dev->host_msix_entries;
115
116 index = -1;
117 for (i = 0; i < assigned_dev->entries_nr; i++)
118 if (irq == host_msix_entries[i].vector) {
119 index = i;
120 break;
121 }
122 if (index < 0) {
123 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
124 return 0;
125 }
126
127 return index;
128}
129
130static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
131{
132 struct kvm_assigned_dev_kernel *assigned_dev;
133 struct kvm *kvm;
134 int i;
135
136 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
137 interrupt_work);
138 kvm = assigned_dev->kvm;
139
140 mutex_lock(&kvm->irq_lock);
141 spin_lock_irq(&assigned_dev->assigned_dev_lock);
142 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
143 struct kvm_guest_msix_entry *guest_entries =
144 assigned_dev->guest_msix_entries;
145 for (i = 0; i < assigned_dev->entries_nr; i++) {
146 if (!(guest_entries[i].flags &
147 KVM_ASSIGNED_MSIX_PENDING))
148 continue;
149 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
150 kvm_set_irq(assigned_dev->kvm,
151 assigned_dev->irq_source_id,
152 guest_entries[i].vector, 1);
153 }
154 } else
155 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
156 assigned_dev->guest_irq, 1);
157
158 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
159 mutex_unlock(&assigned_dev->kvm->irq_lock);
160}
161
162static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
163{
164 unsigned long flags;
165 struct kvm_assigned_dev_kernel *assigned_dev =
166 (struct kvm_assigned_dev_kernel *) dev_id;
167
168 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
169 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
170 int index = find_index_from_host_irq(assigned_dev, irq);
171 if (index < 0)
172 goto out;
173 assigned_dev->guest_msix_entries[index].flags |=
174 KVM_ASSIGNED_MSIX_PENDING;
175 }
176
177 schedule_work(&assigned_dev->interrupt_work);
178
179 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
180 disable_irq_nosync(irq);
181 assigned_dev->host_irq_disabled = true;
182 }
183
184out:
185 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
186 return IRQ_HANDLED;
187}
188
189/* Ack the irq line for an assigned device */
190static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
191{
192 struct kvm_assigned_dev_kernel *dev;
193 unsigned long flags;
194
195 if (kian->gsi == -1)
196 return;
197
198 dev = container_of(kian, struct kvm_assigned_dev_kernel,
199 ack_notifier);
200
201 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
202
203 /* The guest irq may be shared so this ack may be
204 * from another device.
205 */
206 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
207 if (dev->host_irq_disabled) {
208 enable_irq(dev->host_irq);
209 dev->host_irq_disabled = false;
210 }
211 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
212}
213
214static void deassign_guest_irq(struct kvm *kvm,
215 struct kvm_assigned_dev_kernel *assigned_dev)
216{
217 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
218 assigned_dev->ack_notifier.gsi = -1;
219
220 if (assigned_dev->irq_source_id != -1)
221 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
222 assigned_dev->irq_source_id = -1;
223 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
224}
225
226/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
227static void deassign_host_irq(struct kvm *kvm,
228 struct kvm_assigned_dev_kernel *assigned_dev)
229{
230 /*
231 * In kvm_free_device_irq, cancel_work_sync return true if:
232 * 1. work is scheduled, and then cancelled.
233 * 2. work callback is executed.
234 *
235 * The first one ensured that the irq is disabled and no more events
236 * would happen. But for the second one, the irq may be enabled (e.g.
237 * for MSI). So we disable irq here to prevent further events.
238 *
239 * Notice this maybe result in nested disable if the interrupt type is
240 * INTx, but it's OK for we are going to free it.
241 *
242 * If this function is a part of VM destroy, please ensure that till
243 * now, the kvm state is still legal for probably we also have to wait
244 * interrupt_work done.
245 */
246 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
247 int i;
248 for (i = 0; i < assigned_dev->entries_nr; i++)
249 disable_irq_nosync(assigned_dev->
250 host_msix_entries[i].vector);
251
252 cancel_work_sync(&assigned_dev->interrupt_work);
253
254 for (i = 0; i < assigned_dev->entries_nr; i++)
255 free_irq(assigned_dev->host_msix_entries[i].vector,
256 (void *)assigned_dev);
257
258 assigned_dev->entries_nr = 0;
259 kfree(assigned_dev->host_msix_entries);
260 kfree(assigned_dev->guest_msix_entries);
261 pci_disable_msix(assigned_dev->dev);
262 } else {
263 /* Deal with MSI and INTx */
264 disable_irq_nosync(assigned_dev->host_irq);
265 cancel_work_sync(&assigned_dev->interrupt_work);
266
267 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
268
269 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
270 pci_disable_msi(assigned_dev->dev);
271 }
272
273 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
274}
275
276static int kvm_deassign_irq(struct kvm *kvm,
277 struct kvm_assigned_dev_kernel *assigned_dev,
278 unsigned long irq_requested_type)
279{
280 unsigned long guest_irq_type, host_irq_type;
281
282 if (!irqchip_in_kernel(kvm))
283 return -EINVAL;
284 /* no irq assignment to deassign */
285 if (!assigned_dev->irq_requested_type)
286 return -ENXIO;
287
288 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
289 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
290
291 if (host_irq_type)
292 deassign_host_irq(kvm, assigned_dev);
293 if (guest_irq_type)
294 deassign_guest_irq(kvm, assigned_dev);
295
296 return 0;
297}
298
299static void kvm_free_assigned_irq(struct kvm *kvm,
300 struct kvm_assigned_dev_kernel *assigned_dev)
301{
302 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
303}
304
305static void kvm_free_assigned_device(struct kvm *kvm,
306 struct kvm_assigned_dev_kernel
307 *assigned_dev)
308{
309 kvm_free_assigned_irq(kvm, assigned_dev);
310
311 pci_reset_function(assigned_dev->dev);
312
313 pci_release_regions(assigned_dev->dev);
314 pci_disable_device(assigned_dev->dev);
315 pci_dev_put(assigned_dev->dev);
316
317 list_del(&assigned_dev->list);
318 kfree(assigned_dev);
319}
320
321void kvm_free_all_assigned_devices(struct kvm *kvm)
322{
323 struct list_head *ptr, *ptr2;
324 struct kvm_assigned_dev_kernel *assigned_dev;
325
326 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
327 assigned_dev = list_entry(ptr,
328 struct kvm_assigned_dev_kernel,
329 list);
330
331 kvm_free_assigned_device(kvm, assigned_dev);
332 }
333}
334
335static int assigned_device_enable_host_intx(struct kvm *kvm,
336 struct kvm_assigned_dev_kernel *dev)
337{
338 dev->host_irq = dev->dev->irq;
339 /* Even though this is PCI, we don't want to use shared
340 * interrupts. Sharing host devices with guest-assigned devices
341 * on the same interrupt line is not a happy situation: there
342 * are going to be long delays in accepting, acking, etc.
343 */
344 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
345 0, "kvm_assigned_intx_device", (void *)dev))
346 return -EIO;
347 return 0;
348}
349
350#ifdef __KVM_HAVE_MSI
351static int assigned_device_enable_host_msi(struct kvm *kvm,
352 struct kvm_assigned_dev_kernel *dev)
353{
354 int r;
355
356 if (!dev->dev->msi_enabled) {
357 r = pci_enable_msi(dev->dev);
358 if (r)
359 return r;
360 }
361
362 dev->host_irq = dev->dev->irq;
363 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
364 "kvm_assigned_msi_device", (void *)dev)) {
365 pci_disable_msi(dev->dev);
366 return -EIO;
367 }
368
369 return 0;
370}
371#endif
372
373#ifdef __KVM_HAVE_MSIX
374static int assigned_device_enable_host_msix(struct kvm *kvm,
375 struct kvm_assigned_dev_kernel *dev)
376{
377 int i, r = -EINVAL;
378
379 /* host_msix_entries and guest_msix_entries should have been
380 * initialized */
381 if (dev->entries_nr == 0)
382 return r;
383
384 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
385 if (r)
386 return r;
387
388 for (i = 0; i < dev->entries_nr; i++) {
389 r = request_irq(dev->host_msix_entries[i].vector,
390 kvm_assigned_dev_intr, 0,
391 "kvm_assigned_msix_device",
392 (void *)dev);
393 /* FIXME: free requested_irq's on failure */
394 if (r)
395 return r;
396 }
397
398 return 0;
399}
400
401#endif
402
403static int assigned_device_enable_guest_intx(struct kvm *kvm,
404 struct kvm_assigned_dev_kernel *dev,
405 struct kvm_assigned_irq *irq)
406{
407 dev->guest_irq = irq->guest_irq;
408 dev->ack_notifier.gsi = irq->guest_irq;
409 return 0;
410}
411
412#ifdef __KVM_HAVE_MSI
413static int assigned_device_enable_guest_msi(struct kvm *kvm,
414 struct kvm_assigned_dev_kernel *dev,
415 struct kvm_assigned_irq *irq)
416{
417 dev->guest_irq = irq->guest_irq;
418 dev->ack_notifier.gsi = -1;
419 dev->host_irq_disabled = false;
420 return 0;
421}
422#endif
423#ifdef __KVM_HAVE_MSIX
424static int assigned_device_enable_guest_msix(struct kvm *kvm,
425 struct kvm_assigned_dev_kernel *dev,
426 struct kvm_assigned_irq *irq)
427{
428 dev->guest_irq = irq->guest_irq;
429 dev->ack_notifier.gsi = -1;
430 dev->host_irq_disabled = false;
431 return 0;
432}
433#endif
434
435static int assign_host_irq(struct kvm *kvm,
436 struct kvm_assigned_dev_kernel *dev,
437 __u32 host_irq_type)
438{
439 int r = -EEXIST;
440
441 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
442 return r;
443
444 switch (host_irq_type) {
445 case KVM_DEV_IRQ_HOST_INTX:
446 r = assigned_device_enable_host_intx(kvm, dev);
447 break;
448#ifdef __KVM_HAVE_MSI
449 case KVM_DEV_IRQ_HOST_MSI:
450 r = assigned_device_enable_host_msi(kvm, dev);
451 break;
452#endif
453#ifdef __KVM_HAVE_MSIX
454 case KVM_DEV_IRQ_HOST_MSIX:
455 r = assigned_device_enable_host_msix(kvm, dev);
456 break;
457#endif
458 default:
459 r = -EINVAL;
460 }
461
462 if (!r)
463 dev->irq_requested_type |= host_irq_type;
464
465 return r;
466}
467
468static int assign_guest_irq(struct kvm *kvm,
469 struct kvm_assigned_dev_kernel *dev,
470 struct kvm_assigned_irq *irq,
471 unsigned long guest_irq_type)
472{
473 int id;
474 int r = -EEXIST;
475
476 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
477 return r;
478
479 id = kvm_request_irq_source_id(kvm);
480 if (id < 0)
481 return id;
482
483 dev->irq_source_id = id;
484
485 switch (guest_irq_type) {
486 case KVM_DEV_IRQ_GUEST_INTX:
487 r = assigned_device_enable_guest_intx(kvm, dev, irq);
488 break;
489#ifdef __KVM_HAVE_MSI
490 case KVM_DEV_IRQ_GUEST_MSI:
491 r = assigned_device_enable_guest_msi(kvm, dev, irq);
492 break;
493#endif
494#ifdef __KVM_HAVE_MSIX
495 case KVM_DEV_IRQ_GUEST_MSIX:
496 r = assigned_device_enable_guest_msix(kvm, dev, irq);
497 break;
498#endif
499 default:
500 r = -EINVAL;
501 }
502
503 if (!r) {
504 dev->irq_requested_type |= guest_irq_type;
505 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
506 } else
507 kvm_free_irq_source_id(kvm, dev->irq_source_id);
508
509 return r;
510}
511
512/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
513static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
514 struct kvm_assigned_irq *assigned_irq)
515{
516 int r = -EINVAL;
517 struct kvm_assigned_dev_kernel *match;
518 unsigned long host_irq_type, guest_irq_type;
519
520 if (!capable(CAP_SYS_RAWIO))
521 return -EPERM;
522
523 if (!irqchip_in_kernel(kvm))
524 return r;
525
526 mutex_lock(&kvm->lock);
527 r = -ENODEV;
528 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
529 assigned_irq->assigned_dev_id);
530 if (!match)
531 goto out;
532
533 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
534 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
535
536 r = -EINVAL;
537 /* can only assign one type at a time */
538 if (hweight_long(host_irq_type) > 1)
539 goto out;
540 if (hweight_long(guest_irq_type) > 1)
541 goto out;
542 if (host_irq_type == 0 && guest_irq_type == 0)
543 goto out;
544
545 r = 0;
546 if (host_irq_type)
547 r = assign_host_irq(kvm, match, host_irq_type);
548 if (r)
549 goto out;
550
551 if (guest_irq_type)
552 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
553out:
554 mutex_unlock(&kvm->lock);
555 return r;
556}
557
558static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
559 struct kvm_assigned_irq
560 *assigned_irq)
561{
562 int r = -ENODEV;
563 struct kvm_assigned_dev_kernel *match;
564
565 mutex_lock(&kvm->lock);
566
567 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
568 assigned_irq->assigned_dev_id);
569 if (!match)
570 goto out;
571
572 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
573out:
574 mutex_unlock(&kvm->lock);
575 return r;
576}
577
578static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
579 struct kvm_assigned_pci_dev *assigned_dev)
580{
581 int r = 0;
582 struct kvm_assigned_dev_kernel *match;
583 struct pci_dev *dev;
584
585 down_read(&kvm->slots_lock);
586 mutex_lock(&kvm->lock);
587
588 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
589 assigned_dev->assigned_dev_id);
590 if (match) {
591 /* device already assigned */
592 r = -EEXIST;
593 goto out;
594 }
595
596 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
597 if (match == NULL) {
598 printk(KERN_INFO "%s: Couldn't allocate memory\n",
599 __func__);
600 r = -ENOMEM;
601 goto out;
602 }
603 dev = pci_get_bus_and_slot(assigned_dev->busnr,
604 assigned_dev->devfn);
605 if (!dev) {
606 printk(KERN_INFO "%s: host device not found\n", __func__);
607 r = -EINVAL;
608 goto out_free;
609 }
610 if (pci_enable_device(dev)) {
611 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
612 r = -EBUSY;
613 goto out_put;
614 }
615 r = pci_request_regions(dev, "kvm_assigned_device");
616 if (r) {
617 printk(KERN_INFO "%s: Could not get access to device regions\n",
618 __func__);
619 goto out_disable;
620 }
621
622 pci_reset_function(dev);
623
624 match->assigned_dev_id = assigned_dev->assigned_dev_id;
625 match->host_busnr = assigned_dev->busnr;
626 match->host_devfn = assigned_dev->devfn;
627 match->flags = assigned_dev->flags;
628 match->dev = dev;
629 spin_lock_init(&match->assigned_dev_lock);
630 match->irq_source_id = -1;
631 match->kvm = kvm;
632 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
633 INIT_WORK(&match->interrupt_work,
634 kvm_assigned_dev_interrupt_work_handler);
635
636 list_add(&match->list, &kvm->arch.assigned_dev_head);
637
638 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
639 if (!kvm->arch.iommu_domain) {
640 r = kvm_iommu_map_guest(kvm);
641 if (r)
642 goto out_list_del;
643 }
644 r = kvm_assign_device(kvm, match);
645 if (r)
646 goto out_list_del;
647 }
648
649out:
650 mutex_unlock(&kvm->lock);
651 up_read(&kvm->slots_lock);
652 return r;
653out_list_del:
654 list_del(&match->list);
655 pci_release_regions(dev);
656out_disable:
657 pci_disable_device(dev);
658out_put:
659 pci_dev_put(dev);
660out_free:
661 kfree(match);
662 mutex_unlock(&kvm->lock);
663 up_read(&kvm->slots_lock);
664 return r;
665}
666#endif
667
668#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
669static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
670 struct kvm_assigned_pci_dev *assigned_dev)
671{
672 int r = 0;
673 struct kvm_assigned_dev_kernel *match;
674
675 mutex_lock(&kvm->lock);
676
677 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
678 assigned_dev->assigned_dev_id);
679 if (!match) {
680 printk(KERN_INFO "%s: device hasn't been assigned before, "
681 "so cannot be deassigned\n", __func__);
682 r = -EINVAL;
683 goto out;
684 }
685
686 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
687 kvm_deassign_device(kvm, match);
688
689 kvm_free_assigned_device(kvm, match);
690
691out:
692 mutex_unlock(&kvm->lock);
693 return r;
694}
695#endif
696
697inline int kvm_is_mmio_pfn(pfn_t pfn) 92inline int kvm_is_mmio_pfn(pfn_t pfn)
698{ 93{
699 if (pfn_valid(pfn)) { 94 if (pfn_valid(pfn)) {
@@ -949,6 +344,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
949 344
950static struct kvm *kvm_create_vm(void) 345static struct kvm *kvm_create_vm(void)
951{ 346{
347 int r = 0;
952 struct kvm *kvm = kvm_arch_create_vm(); 348 struct kvm *kvm = kvm_arch_create_vm();
953#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 349#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
954 struct page *page; 350 struct page *page;
@@ -956,16 +352,21 @@ static struct kvm *kvm_create_vm(void)
956 352
957 if (IS_ERR(kvm)) 353 if (IS_ERR(kvm))
958 goto out; 354 goto out;
355
356 r = hardware_enable_all();
357 if (r)
358 goto out_err_nodisable;
359
959#ifdef CONFIG_HAVE_KVM_IRQCHIP 360#ifdef CONFIG_HAVE_KVM_IRQCHIP
960 INIT_LIST_HEAD(&kvm->irq_routing);
961 INIT_HLIST_HEAD(&kvm->mask_notifier_list); 361 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
362 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
962#endif 363#endif
963 364
964#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 365#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
965 page = alloc_page(GFP_KERNEL | __GFP_ZERO); 366 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
966 if (!page) { 367 if (!page) {
967 kfree(kvm); 368 r = -ENOMEM;
968 return ERR_PTR(-ENOMEM); 369 goto out_err;
969 } 370 }
970 kvm->coalesced_mmio_ring = 371 kvm->coalesced_mmio_ring =
971 (struct kvm_coalesced_mmio_ring *)page_address(page); 372 (struct kvm_coalesced_mmio_ring *)page_address(page);
@@ -973,15 +374,13 @@ static struct kvm *kvm_create_vm(void)
973 374
974#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 375#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
975 { 376 {
976 int err;
977 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; 377 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
978 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm); 378 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
979 if (err) { 379 if (r) {
980#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 380#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
981 put_page(page); 381 put_page(page);
982#endif 382#endif
983 kfree(kvm); 383 goto out_err;
984 return ERR_PTR(err);
985 } 384 }
986 } 385 }
987#endif 386#endif
@@ -1005,6 +404,12 @@ static struct kvm *kvm_create_vm(void)
1005#endif 404#endif
1006out: 405out:
1007 return kvm; 406 return kvm;
407
408out_err:
409 hardware_disable_all();
410out_err_nodisable:
411 kfree(kvm);
412 return ERR_PTR(r);
1008} 413}
1009 414
1010/* 415/*
@@ -1063,6 +468,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
1063 kvm_arch_flush_shadow(kvm); 468 kvm_arch_flush_shadow(kvm);
1064#endif 469#endif
1065 kvm_arch_destroy_vm(kvm); 470 kvm_arch_destroy_vm(kvm);
471 hardware_disable_all();
1066 mmdrop(mm); 472 mmdrop(mm);
1067} 473}
1068 474
@@ -1689,9 +1095,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1689 if (signal_pending(current)) 1095 if (signal_pending(current))
1690 break; 1096 break;
1691 1097
1692 vcpu_put(vcpu);
1693 schedule(); 1098 schedule();
1694 vcpu_load(vcpu);
1695 } 1099 }
1696 1100
1697 finish_wait(&vcpu->wq, &wait); 1101 finish_wait(&vcpu->wq, &wait);
@@ -1705,6 +1109,21 @@ void kvm_resched(struct kvm_vcpu *vcpu)
1705} 1109}
1706EXPORT_SYMBOL_GPL(kvm_resched); 1110EXPORT_SYMBOL_GPL(kvm_resched);
1707 1111
1112void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1113{
1114 ktime_t expires;
1115 DEFINE_WAIT(wait);
1116
1117 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1118
1119 /* Sleep for 100 us, and hope lock-holder got scheduled */
1120 expires = ktime_add_ns(ktime_get(), 100000UL);
1121 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1122
1123 finish_wait(&vcpu->wq, &wait);
1124}
1125EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1126
1708static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1127static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1709{ 1128{
1710 struct kvm_vcpu *vcpu = vma->vm_file->private_data; 1129 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
@@ -1828,88 +1247,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1828 return 0; 1247 return 0;
1829} 1248}
1830 1249
1831#ifdef __KVM_HAVE_MSIX
1832static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1833 struct kvm_assigned_msix_nr *entry_nr)
1834{
1835 int r = 0;
1836 struct kvm_assigned_dev_kernel *adev;
1837
1838 mutex_lock(&kvm->lock);
1839
1840 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1841 entry_nr->assigned_dev_id);
1842 if (!adev) {
1843 r = -EINVAL;
1844 goto msix_nr_out;
1845 }
1846
1847 if (adev->entries_nr == 0) {
1848 adev->entries_nr = entry_nr->entry_nr;
1849 if (adev->entries_nr == 0 ||
1850 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1851 r = -EINVAL;
1852 goto msix_nr_out;
1853 }
1854
1855 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1856 entry_nr->entry_nr,
1857 GFP_KERNEL);
1858 if (!adev->host_msix_entries) {
1859 r = -ENOMEM;
1860 goto msix_nr_out;
1861 }
1862 adev->guest_msix_entries = kzalloc(
1863 sizeof(struct kvm_guest_msix_entry) *
1864 entry_nr->entry_nr, GFP_KERNEL);
1865 if (!adev->guest_msix_entries) {
1866 kfree(adev->host_msix_entries);
1867 r = -ENOMEM;
1868 goto msix_nr_out;
1869 }
1870 } else /* Not allowed set MSI-X number twice */
1871 r = -EINVAL;
1872msix_nr_out:
1873 mutex_unlock(&kvm->lock);
1874 return r;
1875}
1876
1877static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1878 struct kvm_assigned_msix_entry *entry)
1879{
1880 int r = 0, i;
1881 struct kvm_assigned_dev_kernel *adev;
1882
1883 mutex_lock(&kvm->lock);
1884
1885 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1886 entry->assigned_dev_id);
1887
1888 if (!adev) {
1889 r = -EINVAL;
1890 goto msix_entry_out;
1891 }
1892
1893 for (i = 0; i < adev->entries_nr; i++)
1894 if (adev->guest_msix_entries[i].vector == 0 ||
1895 adev->guest_msix_entries[i].entry == entry->entry) {
1896 adev->guest_msix_entries[i].entry = entry->entry;
1897 adev->guest_msix_entries[i].vector = entry->gsi;
1898 adev->host_msix_entries[i].entry = entry->entry;
1899 break;
1900 }
1901 if (i == adev->entries_nr) {
1902 r = -ENOSPC;
1903 goto msix_entry_out;
1904 }
1905
1906msix_entry_out:
1907 mutex_unlock(&kvm->lock);
1908
1909 return r;
1910}
1911#endif
1912
1913static long kvm_vcpu_ioctl(struct file *filp, 1250static long kvm_vcpu_ioctl(struct file *filp,
1914 unsigned int ioctl, unsigned long arg) 1251 unsigned int ioctl, unsigned long arg)
1915{ 1252{
@@ -2168,112 +1505,6 @@ static long kvm_vm_ioctl(struct file *filp,
2168 break; 1505 break;
2169 } 1506 }
2170#endif 1507#endif
2171#ifdef KVM_CAP_DEVICE_ASSIGNMENT
2172 case KVM_ASSIGN_PCI_DEVICE: {
2173 struct kvm_assigned_pci_dev assigned_dev;
2174
2175 r = -EFAULT;
2176 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2177 goto out;
2178 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2179 if (r)
2180 goto out;
2181 break;
2182 }
2183 case KVM_ASSIGN_IRQ: {
2184 r = -EOPNOTSUPP;
2185 break;
2186 }
2187#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2188 case KVM_ASSIGN_DEV_IRQ: {
2189 struct kvm_assigned_irq assigned_irq;
2190
2191 r = -EFAULT;
2192 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2193 goto out;
2194 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2195 if (r)
2196 goto out;
2197 break;
2198 }
2199 case KVM_DEASSIGN_DEV_IRQ: {
2200 struct kvm_assigned_irq assigned_irq;
2201
2202 r = -EFAULT;
2203 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2204 goto out;
2205 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2206 if (r)
2207 goto out;
2208 break;
2209 }
2210#endif
2211#endif
2212#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2213 case KVM_DEASSIGN_PCI_DEVICE: {
2214 struct kvm_assigned_pci_dev assigned_dev;
2215
2216 r = -EFAULT;
2217 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2218 goto out;
2219 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2220 if (r)
2221 goto out;
2222 break;
2223 }
2224#endif
2225#ifdef KVM_CAP_IRQ_ROUTING
2226 case KVM_SET_GSI_ROUTING: {
2227 struct kvm_irq_routing routing;
2228 struct kvm_irq_routing __user *urouting;
2229 struct kvm_irq_routing_entry *entries;
2230
2231 r = -EFAULT;
2232 if (copy_from_user(&routing, argp, sizeof(routing)))
2233 goto out;
2234 r = -EINVAL;
2235 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2236 goto out;
2237 if (routing.flags)
2238 goto out;
2239 r = -ENOMEM;
2240 entries = vmalloc(routing.nr * sizeof(*entries));
2241 if (!entries)
2242 goto out;
2243 r = -EFAULT;
2244 urouting = argp;
2245 if (copy_from_user(entries, urouting->entries,
2246 routing.nr * sizeof(*entries)))
2247 goto out_free_irq_routing;
2248 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2249 routing.flags);
2250 out_free_irq_routing:
2251 vfree(entries);
2252 break;
2253 }
2254#endif /* KVM_CAP_IRQ_ROUTING */
2255#ifdef __KVM_HAVE_MSIX
2256 case KVM_ASSIGN_SET_MSIX_NR: {
2257 struct kvm_assigned_msix_nr entry_nr;
2258 r = -EFAULT;
2259 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2260 goto out;
2261 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2262 if (r)
2263 goto out;
2264 break;
2265 }
2266 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2267 struct kvm_assigned_msix_entry entry;
2268 r = -EFAULT;
2269 if (copy_from_user(&entry, argp, sizeof entry))
2270 goto out;
2271 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2272 if (r)
2273 goto out;
2274 break;
2275 }
2276#endif
2277 case KVM_IRQFD: { 1508 case KVM_IRQFD: {
2278 struct kvm_irqfd data; 1509 struct kvm_irqfd data;
2279 1510
@@ -2305,11 +1536,59 @@ static long kvm_vm_ioctl(struct file *filp,
2305#endif 1536#endif
2306 default: 1537 default:
2307 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1538 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1539 if (r == -ENOTTY)
1540 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2308 } 1541 }
2309out: 1542out:
2310 return r; 1543 return r;
2311} 1544}
2312 1545
1546#ifdef CONFIG_COMPAT
1547struct compat_kvm_dirty_log {
1548 __u32 slot;
1549 __u32 padding1;
1550 union {
1551 compat_uptr_t dirty_bitmap; /* one bit per page */
1552 __u64 padding2;
1553 };
1554};
1555
1556static long kvm_vm_compat_ioctl(struct file *filp,
1557 unsigned int ioctl, unsigned long arg)
1558{
1559 struct kvm *kvm = filp->private_data;
1560 int r;
1561
1562 if (kvm->mm != current->mm)
1563 return -EIO;
1564 switch (ioctl) {
1565 case KVM_GET_DIRTY_LOG: {
1566 struct compat_kvm_dirty_log compat_log;
1567 struct kvm_dirty_log log;
1568
1569 r = -EFAULT;
1570 if (copy_from_user(&compat_log, (void __user *)arg,
1571 sizeof(compat_log)))
1572 goto out;
1573 log.slot = compat_log.slot;
1574 log.padding1 = compat_log.padding1;
1575 log.padding2 = compat_log.padding2;
1576 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1577
1578 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1579 if (r)
1580 goto out;
1581 break;
1582 }
1583 default:
1584 r = kvm_vm_ioctl(filp, ioctl, arg);
1585 }
1586
1587out:
1588 return r;
1589}
1590#endif
1591
2313static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1592static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2314{ 1593{
2315 struct page *page[1]; 1594 struct page *page[1];
@@ -2344,7 +1623,9 @@ static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2344static struct file_operations kvm_vm_fops = { 1623static struct file_operations kvm_vm_fops = {
2345 .release = kvm_vm_release, 1624 .release = kvm_vm_release,
2346 .unlocked_ioctl = kvm_vm_ioctl, 1625 .unlocked_ioctl = kvm_vm_ioctl,
2347 .compat_ioctl = kvm_vm_ioctl, 1626#ifdef CONFIG_COMPAT
1627 .compat_ioctl = kvm_vm_compat_ioctl,
1628#endif
2348 .mmap = kvm_vm_mmap, 1629 .mmap = kvm_vm_mmap,
2349}; 1630};
2350 1631
@@ -2372,6 +1653,7 @@ static long kvm_dev_ioctl_check_extension_generic(long arg)
2372#ifdef CONFIG_KVM_APIC_ARCHITECTURE 1653#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2373 case KVM_CAP_SET_BOOT_CPU_ID: 1654 case KVM_CAP_SET_BOOT_CPU_ID:
2374#endif 1655#endif
1656 case KVM_CAP_INTERNAL_ERROR_DATA:
2375 return 1; 1657 return 1;
2376#ifdef CONFIG_HAVE_KVM_IRQCHIP 1658#ifdef CONFIG_HAVE_KVM_IRQCHIP
2377 case KVM_CAP_IRQ_ROUTING: 1659 case KVM_CAP_IRQ_ROUTING:
@@ -2442,11 +1724,21 @@ static struct miscdevice kvm_dev = {
2442static void hardware_enable(void *junk) 1724static void hardware_enable(void *junk)
2443{ 1725{
2444 int cpu = raw_smp_processor_id(); 1726 int cpu = raw_smp_processor_id();
1727 int r;
2445 1728
2446 if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) 1729 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2447 return; 1730 return;
1731
2448 cpumask_set_cpu(cpu, cpus_hardware_enabled); 1732 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2449 kvm_arch_hardware_enable(NULL); 1733
1734 r = kvm_arch_hardware_enable(NULL);
1735
1736 if (r) {
1737 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1738 atomic_inc(&hardware_enable_failed);
1739 printk(KERN_INFO "kvm: enabling virtualization on "
1740 "CPU%d failed\n", cpu);
1741 }
2450} 1742}
2451 1743
2452static void hardware_disable(void *junk) 1744static void hardware_disable(void *junk)
@@ -2459,11 +1751,52 @@ static void hardware_disable(void *junk)
2459 kvm_arch_hardware_disable(NULL); 1751 kvm_arch_hardware_disable(NULL);
2460} 1752}
2461 1753
1754static void hardware_disable_all_nolock(void)
1755{
1756 BUG_ON(!kvm_usage_count);
1757
1758 kvm_usage_count--;
1759 if (!kvm_usage_count)
1760 on_each_cpu(hardware_disable, NULL, 1);
1761}
1762
1763static void hardware_disable_all(void)
1764{
1765 spin_lock(&kvm_lock);
1766 hardware_disable_all_nolock();
1767 spin_unlock(&kvm_lock);
1768}
1769
1770static int hardware_enable_all(void)
1771{
1772 int r = 0;
1773
1774 spin_lock(&kvm_lock);
1775
1776 kvm_usage_count++;
1777 if (kvm_usage_count == 1) {
1778 atomic_set(&hardware_enable_failed, 0);
1779 on_each_cpu(hardware_enable, NULL, 1);
1780
1781 if (atomic_read(&hardware_enable_failed)) {
1782 hardware_disable_all_nolock();
1783 r = -EBUSY;
1784 }
1785 }
1786
1787 spin_unlock(&kvm_lock);
1788
1789 return r;
1790}
1791
2462static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, 1792static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2463 void *v) 1793 void *v)
2464{ 1794{
2465 int cpu = (long)v; 1795 int cpu = (long)v;
2466 1796
1797 if (!kvm_usage_count)
1798 return NOTIFY_OK;
1799
2467 val &= ~CPU_TASKS_FROZEN; 1800 val &= ~CPU_TASKS_FROZEN;
2468 switch (val) { 1801 switch (val) {
2469 case CPU_DYING: 1802 case CPU_DYING:
@@ -2666,13 +1999,15 @@ static void kvm_exit_debug(void)
2666 1999
2667static int kvm_suspend(struct sys_device *dev, pm_message_t state) 2000static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2668{ 2001{
2669 hardware_disable(NULL); 2002 if (kvm_usage_count)
2003 hardware_disable(NULL);
2670 return 0; 2004 return 0;
2671} 2005}
2672 2006
2673static int kvm_resume(struct sys_device *dev) 2007static int kvm_resume(struct sys_device *dev)
2674{ 2008{
2675 hardware_enable(NULL); 2009 if (kvm_usage_count)
2010 hardware_enable(NULL);
2676 return 0; 2011 return 0;
2677} 2012}
2678 2013
@@ -2747,7 +2082,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2747 goto out_free_1; 2082 goto out_free_1;
2748 } 2083 }
2749 2084
2750 on_each_cpu(hardware_enable, NULL, 1);
2751 r = register_cpu_notifier(&kvm_cpu_notifier); 2085 r = register_cpu_notifier(&kvm_cpu_notifier);
2752 if (r) 2086 if (r)
2753 goto out_free_2; 2087 goto out_free_2;
@@ -2797,7 +2131,6 @@ out_free_3:
2797 unregister_reboot_notifier(&kvm_reboot_notifier); 2131 unregister_reboot_notifier(&kvm_reboot_notifier);
2798 unregister_cpu_notifier(&kvm_cpu_notifier); 2132 unregister_cpu_notifier(&kvm_cpu_notifier);
2799out_free_2: 2133out_free_2:
2800 on_each_cpu(hardware_disable, NULL, 1);
2801out_free_1: 2134out_free_1:
2802 kvm_arch_hardware_unsetup(); 2135 kvm_arch_hardware_unsetup();
2803out_free_0a: 2136out_free_0a:
generated by cgit v1.2.2 (git 2.25.0) at 2025-01-12 17:38:22 -0500