aboutsummaryrefslogtreecommitdiffstats
path: root/virt
diff options
context:
space:
mode:
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>2009-12-09 01:14:38 -0500
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>2009-12-09 01:14:38 -0500
commitbcd6acd51f3d4d1ada201e9bc5c40a31d6d80c71 (patch)
tree2f6dffd2d3e4dd67355a224de7e7a960335a92fd /virt
parent11c34c7deaeeebcee342cbc35e1bb2a6711b2431 (diff)
parent3ff6a468b45b5dfeb0e903e56f4eb27d34b2437c (diff)
Merge commit 'origin/master' into next
Conflicts: include/linux/kvm.h
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 cdca63917e7..e1f2bf8d7b1 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>
@@ -54,12 +55,6 @@
54#include "coalesced_mmio.h" 55#include "coalesced_mmio.h"
55#endif 56#endif
56 57
57#ifdef KVM_CAP_DEVICE_ASSIGNMENT
58#include <linux/pci.h>
59#include <linux/interrupt.h>
60#include "irq.h"
61#endif
62
63#define CREATE_TRACE_POINTS 58#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h> 59#include <trace/events/kvm.h>
65 60
@@ -76,6 +71,8 @@ DEFINE_SPINLOCK(kvm_lock);
76LIST_HEAD(vm_list); 71LIST_HEAD(vm_list);
77 72
78static cpumask_var_t cpus_hardware_enabled; 73static cpumask_var_t cpus_hardware_enabled;
74static int kvm_usage_count = 0;
75static atomic_t hardware_enable_failed;
79 76
80struct kmem_cache *kvm_vcpu_cache; 77struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache); 78EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
@@ -86,615 +83,13 @@ struct dentry *kvm_debugfs_dir;
86 83
87static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, 84static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg); 85 unsigned long arg);
86static int hardware_enable_all(void);
87static void hardware_disable_all(void);
89 88
90static bool kvm_rebooting; 89static bool kvm_rebooting;
91 90
92static bool largepages_enabled = true; 91static bool largepages_enabled = true;
93 92
94#ifdef KVM_CAP_DEVICE_ASSIGNMENT
95static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
96 int assigned_dev_id)
97{
98 struct list_head *ptr;
99 struct kvm_assigned_dev_kernel *match;
100
101 list_for_each(ptr, head) {
102 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
103 if (match->assigned_dev_id == assigned_dev_id)
104 return match;
105 }
106 return NULL;
107}
108
109static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
110 *assigned_dev, int irq)
111{
112 int i, index;
113 struct msix_entry *host_msix_entries;
114
115 host_msix_entries = assigned_dev->host_msix_entries;
116
117 index = -1;
118 for (i = 0; i < assigned_dev->entries_nr; i++)
119 if (irq == host_msix_entries[i].vector) {
120 index = i;
121 break;
122 }
123 if (index < 0) {
124 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
125 return 0;
126 }
127
128 return index;
129}
130
131static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
132{
133 struct kvm_assigned_dev_kernel *assigned_dev;
134 struct kvm *kvm;
135 int i;
136
137 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
138 interrupt_work);
139 kvm = assigned_dev->kvm;
140
141 mutex_lock(&kvm->irq_lock);
142 spin_lock_irq(&assigned_dev->assigned_dev_lock);
143 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
144 struct kvm_guest_msix_entry *guest_entries =
145 assigned_dev->guest_msix_entries;
146 for (i = 0; i < assigned_dev->entries_nr; i++) {
147 if (!(guest_entries[i].flags &
148 KVM_ASSIGNED_MSIX_PENDING))
149 continue;
150 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
151 kvm_set_irq(assigned_dev->kvm,
152 assigned_dev->irq_source_id,
153 guest_entries[i].vector, 1);
154 }
155 } else
156 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
157 assigned_dev->guest_irq, 1);
158
159 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
160 mutex_unlock(&assigned_dev->kvm->irq_lock);
161}
162
163static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
164{
165 unsigned long flags;
166 struct kvm_assigned_dev_kernel *assigned_dev =
167 (struct kvm_assigned_dev_kernel *) dev_id;
168
169 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
170 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
171 int index = find_index_from_host_irq(assigned_dev, irq);
172 if (index < 0)
173 goto out;
174 assigned_dev->guest_msix_entries[index].flags |=
175 KVM_ASSIGNED_MSIX_PENDING;
176 }
177
178 schedule_work(&assigned_dev->interrupt_work);
179
180 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
181 disable_irq_nosync(irq);
182 assigned_dev->host_irq_disabled = true;
183 }
184
185out:
186 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
187 return IRQ_HANDLED;
188}
189
190/* Ack the irq line for an assigned device */
191static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
192{
193 struct kvm_assigned_dev_kernel *dev;
194 unsigned long flags;
195
196 if (kian->gsi == -1)
197 return;
198
199 dev = container_of(kian, struct kvm_assigned_dev_kernel,
200 ack_notifier);
201
202 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
203
204 /* The guest irq may be shared so this ack may be
205 * from another device.
206 */
207 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
208 if (dev->host_irq_disabled) {
209 enable_irq(dev->host_irq);
210 dev->host_irq_disabled = false;
211 }
212 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
213}
214
215static void deassign_guest_irq(struct kvm *kvm,
216 struct kvm_assigned_dev_kernel *assigned_dev)
217{
218 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
219 assigned_dev->ack_notifier.gsi = -1;
220
221 if (assigned_dev->irq_source_id != -1)
222 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
223 assigned_dev->irq_source_id = -1;
224 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
225}
226
227/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
228static void deassign_host_irq(struct kvm *kvm,
229 struct kvm_assigned_dev_kernel *assigned_dev)
230{
231 /*
232 * In kvm_free_device_irq, cancel_work_sync return true if:
233 * 1. work is scheduled, and then cancelled.
234 * 2. work callback is executed.
235 *
236 * The first one ensured that the irq is disabled and no more events
237 * would happen. But for the second one, the irq may be enabled (e.g.
238 * for MSI). So we disable irq here to prevent further events.
239 *
240 * Notice this maybe result in nested disable if the interrupt type is
241 * INTx, but it's OK for we are going to free it.
242 *
243 * If this function is a part of VM destroy, please ensure that till
244 * now, the kvm state is still legal for probably we also have to wait
245 * interrupt_work done.
246 */
247 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
248 int i;
249 for (i = 0; i < assigned_dev->entries_nr; i++)
250 disable_irq_nosync(assigned_dev->
251 host_msix_entries[i].vector);
252
253 cancel_work_sync(&assigned_dev->interrupt_work);
254
255 for (i = 0; i < assigned_dev->entries_nr; i++)
256 free_irq(assigned_dev->host_msix_entries[i].vector,
257 (void *)assigned_dev);
258
259 assigned_dev->entries_nr = 0;
260 kfree(assigned_dev->host_msix_entries);
261 kfree(assigned_dev->guest_msix_entries);
262 pci_disable_msix(assigned_dev->dev);
263 } else {
264 /* Deal with MSI and INTx */
265 disable_irq_nosync(assigned_dev->host_irq);
266 cancel_work_sync(&assigned_dev->interrupt_work);
267
268 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
269
270 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
271 pci_disable_msi(assigned_dev->dev);
272 }
273
274 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
275}
276
277static int kvm_deassign_irq(struct kvm *kvm,
278 struct kvm_assigned_dev_kernel *assigned_dev,
279 unsigned long irq_requested_type)
280{
281 unsigned long guest_irq_type, host_irq_type;
282
283 if (!irqchip_in_kernel(kvm))
284 return -EINVAL;
285 /* no irq assignment to deassign */
286 if (!assigned_dev->irq_requested_type)
287 return -ENXIO;
288
289 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
290 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
291
292 if (host_irq_type)
293 deassign_host_irq(kvm, assigned_dev);
294 if (guest_irq_type)
295 deassign_guest_irq(kvm, assigned_dev);
296
297 return 0;
298}
299
300static void kvm_free_assigned_irq(struct kvm *kvm,
301 struct kvm_assigned_dev_kernel *assigned_dev)
302{
303 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
304}
305
306static void kvm_free_assigned_device(struct kvm *kvm,
307 struct kvm_assigned_dev_kernel
308 *assigned_dev)
309{
310 kvm_free_assigned_irq(kvm, assigned_dev);
311
312 pci_reset_function(assigned_dev->dev);
313
314 pci_release_regions(assigned_dev->dev);
315 pci_disable_device(assigned_dev->dev);
316 pci_dev_put(assigned_dev->dev);
317
318 list_del(&assigned_dev->list);
319 kfree(assigned_dev);
320}
321
322void kvm_free_all_assigned_devices(struct kvm *kvm)
323{
324 struct list_head *ptr, *ptr2;
325 struct kvm_assigned_dev_kernel *assigned_dev;
326
327 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
328 assigned_dev = list_entry(ptr,
329 struct kvm_assigned_dev_kernel,
330 list);
331
332 kvm_free_assigned_device(kvm, assigned_dev);
333 }
334}
335
336static int assigned_device_enable_host_intx(struct kvm *kvm,
337 struct kvm_assigned_dev_kernel *dev)
338{
339 dev->host_irq = dev->dev->irq;
340 /* Even though this is PCI, we don't want to use shared
341 * interrupts. Sharing host devices with guest-assigned devices
342 * on the same interrupt line is not a happy situation: there
343 * are going to be long delays in accepting, acking, etc.
344 */
345 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
346 0, "kvm_assigned_intx_device", (void *)dev))
347 return -EIO;
348 return 0;
349}
350
351#ifdef __KVM_HAVE_MSI
352static int assigned_device_enable_host_msi(struct kvm *kvm,
353 struct kvm_assigned_dev_kernel *dev)
354{
355 int r;
356
357 if (!dev->dev->msi_enabled) {
358 r = pci_enable_msi(dev->dev);
359 if (r)
360 return r;
361 }
362
363 dev->host_irq = dev->dev->irq;
364 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
365 "kvm_assigned_msi_device", (void *)dev)) {
366 pci_disable_msi(dev->dev);
367 return -EIO;
368 }
369
370 return 0;
371}
372#endif
373
374#ifdef __KVM_HAVE_MSIX
375static int assigned_device_enable_host_msix(struct kvm *kvm,
376 struct kvm_assigned_dev_kernel *dev)
377{
378 int i, r = -EINVAL;
379
380 /* host_msix_entries and guest_msix_entries should have been
381 * initialized */
382 if (dev->entries_nr == 0)
383 return r;
384
385 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
386 if (r)
387 return r;
388
389 for (i = 0; i < dev->entries_nr; i++) {
390 r = request_irq(dev->host_msix_entries[i].vector,
391 kvm_assigned_dev_intr, 0,
392 "kvm_assigned_msix_device",
393 (void *)dev);
394 /* FIXME: free requested_irq's on failure */
395 if (r)
396 return r;
397 }
398
399 return 0;
400}
401
402#endif
403
404static int assigned_device_enable_guest_intx(struct kvm *kvm,
405 struct kvm_assigned_dev_kernel *dev,
406 struct kvm_assigned_irq *irq)
407{
408 dev->guest_irq = irq->guest_irq;
409 dev->ack_notifier.gsi = irq->guest_irq;
410 return 0;
411}
412
413#ifdef __KVM_HAVE_MSI
414static int assigned_device_enable_guest_msi(struct kvm *kvm,
415 struct kvm_assigned_dev_kernel *dev,
416 struct kvm_assigned_irq *irq)
417{
418 dev->guest_irq = irq->guest_irq;
419 dev->ack_notifier.gsi = -1;
420 dev->host_irq_disabled = false;
421 return 0;
422}
423#endif
424#ifdef __KVM_HAVE_MSIX
425static int assigned_device_enable_guest_msix(struct kvm *kvm,
426 struct kvm_assigned_dev_kernel *dev,
427 struct kvm_assigned_irq *irq)
428{
429 dev->guest_irq = irq->guest_irq;
430 dev->ack_notifier.gsi = -1;
431 dev->host_irq_disabled = false;
432 return 0;
433}
434#endif
435
436static int assign_host_irq(struct kvm *kvm,
437 struct kvm_assigned_dev_kernel *dev,
438 __u32 host_irq_type)
439{
440 int r = -EEXIST;
441
442 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
443 return r;
444
445 switch (host_irq_type) {
446 case KVM_DEV_IRQ_HOST_INTX:
447 r = assigned_device_enable_host_intx(kvm, dev);
448 break;
449#ifdef __KVM_HAVE_MSI
450 case KVM_DEV_IRQ_HOST_MSI:
451 r = assigned_device_enable_host_msi(kvm, dev);
452 break;
453#endif
454#ifdef __KVM_HAVE_MSIX
455 case KVM_DEV_IRQ_HOST_MSIX:
456 r = assigned_device_enable_host_msix(kvm, dev);
457 break;
458#endif
459 default:
460 r = -EINVAL;
461 }
462
463 if (!r)
464 dev->irq_requested_type |= host_irq_type;
465
466 return r;
467}
468
469static int assign_guest_irq(struct kvm *kvm,
470 struct kvm_assigned_dev_kernel *dev,
471 struct kvm_assigned_irq *irq,
472 unsigned long guest_irq_type)
473{
474 int id;
475 int r = -EEXIST;
476
477 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
478 return r;
479
480 id = kvm_request_irq_source_id(kvm);
481 if (id < 0)
482 return id;
483
484 dev->irq_source_id = id;
485
486 switch (guest_irq_type) {
487 case KVM_DEV_IRQ_GUEST_INTX:
488 r = assigned_device_enable_guest_intx(kvm, dev, irq);
489 break;
490#ifdef __KVM_HAVE_MSI
491 case KVM_DEV_IRQ_GUEST_MSI:
492 r = assigned_device_enable_guest_msi(kvm, dev, irq);
493 break;
494#endif
495#ifdef __KVM_HAVE_MSIX
496 case KVM_DEV_IRQ_GUEST_MSIX:
497 r = assigned_device_enable_guest_msix(kvm, dev, irq);
498 break;
499#endif
500 default:
501 r = -EINVAL;
502 }
503
504 if (!r) {
505 dev->irq_requested_type |= guest_irq_type;
506 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
507 } else
508 kvm_free_irq_source_id(kvm, dev->irq_source_id);
509
510 return r;
511}
512
513/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
514static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
515 struct kvm_assigned_irq *assigned_irq)
516{
517 int r = -EINVAL;
518 struct kvm_assigned_dev_kernel *match;
519 unsigned long host_irq_type, guest_irq_type;
520
521 if (!capable(CAP_SYS_RAWIO))
522 return -EPERM;
523
524 if (!irqchip_in_kernel(kvm))
525 return r;
526
527 mutex_lock(&kvm->lock);
528 r = -ENODEV;
529 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
530 assigned_irq->assigned_dev_id);
531 if (!match)
532 goto out;
533
534 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
535 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
536
537 r = -EINVAL;
538 /* can only assign one type at a time */
539 if (hweight_long(host_irq_type) > 1)
540 goto out;
541 if (hweight_long(guest_irq_type) > 1)
542 goto out;
543 if (host_irq_type == 0 && guest_irq_type == 0)
544 goto out;
545
546 r = 0;
547 if (host_irq_type)
548 r = assign_host_irq(kvm, match, host_irq_type);
549 if (r)
550 goto out;
551
552 if (guest_irq_type)
553 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
554out:
555 mutex_unlock(&kvm->lock);
556 return r;
557}
558
559static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
560 struct kvm_assigned_irq
561 *assigned_irq)
562{
563 int r = -ENODEV;
564 struct kvm_assigned_dev_kernel *match;
565
566 mutex_lock(&kvm->lock);
567
568 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
569 assigned_irq->assigned_dev_id);
570 if (!match)
571 goto out;
572
573 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
574out:
575 mutex_unlock(&kvm->lock);
576 return r;
577}
578
579static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
580 struct kvm_assigned_pci_dev *assigned_dev)
581{
582 int r = 0;
583 struct kvm_assigned_dev_kernel *match;
584 struct pci_dev *dev;
585
586 down_read(&kvm->slots_lock);
587 mutex_lock(&kvm->lock);
588
589 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
590 assigned_dev->assigned_dev_id);
591 if (match) {
592 /* device already assigned */
593 r = -EEXIST;
594 goto out;
595 }
596
597 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
598 if (match == NULL) {
599 printk(KERN_INFO "%s: Couldn't allocate memory\n",
600 __func__);
601 r = -ENOMEM;
602 goto out;
603 }
604 dev = pci_get_bus_and_slot(assigned_dev->busnr,
605 assigned_dev->devfn);
606 if (!dev) {
607 printk(KERN_INFO "%s: host device not found\n", __func__);
608 r = -EINVAL;
609 goto out_free;
610 }
611 if (pci_enable_device(dev)) {
612 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
613 r = -EBUSY;
614 goto out_put;
615 }
616 r = pci_request_regions(dev, "kvm_assigned_device");
617 if (r) {
618 printk(KERN_INFO "%s: Could not get access to device regions\n",
619 __func__);
620 goto out_disable;
621 }
622
623 pci_reset_function(dev);
624
625 match->assigned_dev_id = assigned_dev->assigned_dev_id;
626 match->host_busnr = assigned_dev->busnr;
627 match->host_devfn = assigned_dev->devfn;
628 match->flags = assigned_dev->flags;
629 match->dev = dev;
630 spin_lock_init(&match->assigned_dev_lock);
631 match->irq_source_id = -1;
632 match->kvm = kvm;
633 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
634 INIT_WORK(&match->interrupt_work,
635 kvm_assigned_dev_interrupt_work_handler);
636
637 list_add(&match->list, &kvm->arch.assigned_dev_head);
638
639 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
640 if (!kvm->arch.iommu_domain) {
641 r = kvm_iommu_map_guest(kvm);
642 if (r)
643 goto out_list_del;
644 }
645 r = kvm_assign_device(kvm, match);
646 if (r)
647 goto out_list_del;
648 }
649
650out:
651 mutex_unlock(&kvm->lock);
652 up_read(&kvm->slots_lock);
653 return r;
654out_list_del:
655 list_del(&match->list);
656 pci_release_regions(dev);
657out_disable:
658 pci_disable_device(dev);
659out_put:
660 pci_dev_put(dev);
661out_free:
662 kfree(match);
663 mutex_unlock(&kvm->lock);
664 up_read(&kvm->slots_lock);
665 return r;
666}
667#endif
668
669#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
670static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
671 struct kvm_assigned_pci_dev *assigned_dev)
672{
673 int r = 0;
674 struct kvm_assigned_dev_kernel *match;
675
676 mutex_lock(&kvm->lock);
677
678 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
679 assigned_dev->assigned_dev_id);
680 if (!match) {
681 printk(KERN_INFO "%s: device hasn't been assigned before, "
682 "so cannot be deassigned\n", __func__);
683 r = -EINVAL;
684 goto out;
685 }
686
687 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
688 kvm_deassign_device(kvm, match);
689
690 kvm_free_assigned_device(kvm, match);
691
692out:
693 mutex_unlock(&kvm->lock);
694 return r;
695}
696#endif
697
698inline int kvm_is_mmio_pfn(pfn_t pfn) 93inline int kvm_is_mmio_pfn(pfn_t pfn)
699{ 94{
700 if (pfn_valid(pfn)) { 95 if (pfn_valid(pfn)) {
@@ -950,6 +345,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
950 345
951static struct kvm *kvm_create_vm(void) 346static struct kvm *kvm_create_vm(void)
952{ 347{
348 int r = 0;
953 struct kvm *kvm = kvm_arch_create_vm(); 349 struct kvm *kvm = kvm_arch_create_vm();
954#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 350#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
955 struct page *page; 351 struct page *page;
@@ -957,16 +353,21 @@ static struct kvm *kvm_create_vm(void)
957 353
958 if (IS_ERR(kvm)) 354 if (IS_ERR(kvm))
959 goto out; 355 goto out;
356
357 r = hardware_enable_all();
358 if (r)
359 goto out_err_nodisable;
360
960#ifdef CONFIG_HAVE_KVM_IRQCHIP 361#ifdef CONFIG_HAVE_KVM_IRQCHIP
961 INIT_LIST_HEAD(&kvm->irq_routing);
962 INIT_HLIST_HEAD(&kvm->mask_notifier_list); 362 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
363 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
963#endif 364#endif
964 365
965#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 366#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
966 page = alloc_page(GFP_KERNEL | __GFP_ZERO); 367 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
967 if (!page) { 368 if (!page) {
968 kfree(kvm); 369 r = -ENOMEM;
969 return ERR_PTR(-ENOMEM); 370 goto out_err;
970 } 371 }
971 kvm->coalesced_mmio_ring = 372 kvm->coalesced_mmio_ring =
972 (struct kvm_coalesced_mmio_ring *)page_address(page); 373 (struct kvm_coalesced_mmio_ring *)page_address(page);
@@ -974,15 +375,13 @@ static struct kvm *kvm_create_vm(void)
974 375
975#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 376#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
976 { 377 {
977 int err;
978 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; 378 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
979 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm); 379 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
980 if (err) { 380 if (r) {
981#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 381#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
982 put_page(page); 382 put_page(page);
983#endif 383#endif
984 kfree(kvm); 384 goto out_err;
985 return ERR_PTR(err);
986 } 385 }
987 } 386 }
988#endif 387#endif
@@ -1006,6 +405,12 @@ static struct kvm *kvm_create_vm(void)
1006#endif 405#endif
1007out: 406out:
1008 return kvm; 407 return kvm;
408
409out_err:
410 hardware_disable_all();
411out_err_nodisable:
412 kfree(kvm);
413 return ERR_PTR(r);
1009} 414}
1010 415
1011/* 416/*
@@ -1064,6 +469,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
1064 kvm_arch_flush_shadow(kvm); 469 kvm_arch_flush_shadow(kvm);
1065#endif 470#endif
1066 kvm_arch_destroy_vm(kvm); 471 kvm_arch_destroy_vm(kvm);
472 hardware_disable_all();
1067 mmdrop(mm); 473 mmdrop(mm);
1068} 474}
1069 475
@@ -1690,9 +1096,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1690 if (signal_pending(current)) 1096 if (signal_pending(current))
1691 break; 1097 break;
1692 1098
1693 vcpu_put(vcpu);
1694 schedule(); 1099 schedule();
1695 vcpu_load(vcpu);
1696 } 1100 }
1697 1101
1698 finish_wait(&vcpu->wq, &wait); 1102 finish_wait(&vcpu->wq, &wait);
@@ -1706,6 +1110,21 @@ void kvm_resched(struct kvm_vcpu *vcpu)
1706} 1110}
1707EXPORT_SYMBOL_GPL(kvm_resched); 1111EXPORT_SYMBOL_GPL(kvm_resched);
1708 1112
1113void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1114{
1115 ktime_t expires;
1116 DEFINE_WAIT(wait);
1117
1118 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1119
1120 /* Sleep for 100 us, and hope lock-holder got scheduled */
1121 expires = ktime_add_ns(ktime_get(), 100000UL);
1122 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1123
1124 finish_wait(&vcpu->wq, &wait);
1125}
1126EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1127
1709static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1128static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1710{ 1129{
1711 struct kvm_vcpu *vcpu = vma->vm_file->private_data; 1130 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
@@ -1829,88 +1248,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1829 return 0; 1248 return 0;
1830} 1249}
1831 1250
1832#ifdef __KVM_HAVE_MSIX
1833static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1834 struct kvm_assigned_msix_nr *entry_nr)
1835{
1836 int r = 0;
1837 struct kvm_assigned_dev_kernel *adev;
1838
1839 mutex_lock(&kvm->lock);
1840
1841 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1842 entry_nr->assigned_dev_id);
1843 if (!adev) {
1844 r = -EINVAL;
1845 goto msix_nr_out;
1846 }
1847
1848 if (adev->entries_nr == 0) {
1849 adev->entries_nr = entry_nr->entry_nr;
1850 if (adev->entries_nr == 0 ||
1851 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1852 r = -EINVAL;
1853 goto msix_nr_out;
1854 }
1855
1856 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1857 entry_nr->entry_nr,
1858 GFP_KERNEL);
1859 if (!adev->host_msix_entries) {
1860 r = -ENOMEM;
1861 goto msix_nr_out;
1862 }
1863 adev->guest_msix_entries = kzalloc(
1864 sizeof(struct kvm_guest_msix_entry) *
1865 entry_nr->entry_nr, GFP_KERNEL);
1866 if (!adev->guest_msix_entries) {
1867 kfree(adev->host_msix_entries);
1868 r = -ENOMEM;
1869 goto msix_nr_out;
1870 }
1871 } else /* Not allowed set MSI-X number twice */
1872 r = -EINVAL;
1873msix_nr_out:
1874 mutex_unlock(&kvm->lock);
1875 return r;
1876}
1877
1878static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1879 struct kvm_assigned_msix_entry *entry)
1880{
1881 int r = 0, i;
1882 struct kvm_assigned_dev_kernel *adev;
1883
1884 mutex_lock(&kvm->lock);
1885
1886 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1887 entry->assigned_dev_id);
1888
1889 if (!adev) {
1890 r = -EINVAL;
1891 goto msix_entry_out;
1892 }
1893
1894 for (i = 0; i < adev->entries_nr; i++)
1895 if (adev->guest_msix_entries[i].vector == 0 ||
1896 adev->guest_msix_entries[i].entry == entry->entry) {
1897 adev->guest_msix_entries[i].entry = entry->entry;
1898 adev->guest_msix_entries[i].vector = entry->gsi;
1899 adev->host_msix_entries[i].entry = entry->entry;
1900 break;
1901 }
1902 if (i == adev->entries_nr) {
1903 r = -ENOSPC;
1904 goto msix_entry_out;
1905 }
1906
1907msix_entry_out:
1908 mutex_unlock(&kvm->lock);
1909
1910 return r;
1911}
1912#endif
1913
1914static long kvm_vcpu_ioctl(struct file *filp, 1251static long kvm_vcpu_ioctl(struct file *filp,
1915 unsigned int ioctl, unsigned long arg) 1252 unsigned int ioctl, unsigned long arg)
1916{ 1253{
@@ -2169,112 +1506,6 @@ static long kvm_vm_ioctl(struct file *filp,
2169 break; 1506 break;
2170 } 1507 }
2171#endif 1508#endif
2172#ifdef KVM_CAP_DEVICE_ASSIGNMENT
2173 case KVM_ASSIGN_PCI_DEVICE: {
2174 struct kvm_assigned_pci_dev assigned_dev;
2175
2176 r = -EFAULT;
2177 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2178 goto out;
2179 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2180 if (r)
2181 goto out;
2182 break;
2183 }
2184 case KVM_ASSIGN_IRQ: {
2185 r = -EOPNOTSUPP;
2186 break;
2187 }
2188#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2189 case KVM_ASSIGN_DEV_IRQ: {
2190 struct kvm_assigned_irq assigned_irq;
2191
2192 r = -EFAULT;
2193 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2194 goto out;
2195 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2196 if (r)
2197 goto out;
2198 break;
2199 }
2200 case KVM_DEASSIGN_DEV_IRQ: {
2201 struct kvm_assigned_irq assigned_irq;
2202
2203 r = -EFAULT;
2204 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2205 goto out;
2206 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2207 if (r)
2208 goto out;
2209 break;
2210 }
2211#endif
2212#endif
2213#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2214 case KVM_DEASSIGN_PCI_DEVICE: {
2215 struct kvm_assigned_pci_dev assigned_dev;
2216
2217 r = -EFAULT;
2218 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2219 goto out;
2220 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2221 if (r)
2222 goto out;
2223 break;
2224 }
2225#endif
2226#ifdef KVM_CAP_IRQ_ROUTING
2227 case KVM_SET_GSI_ROUTING: {
2228 struct kvm_irq_routing routing;
2229 struct kvm_irq_routing __user *urouting;
2230 struct kvm_irq_routing_entry *entries;
2231
2232 r = -EFAULT;
2233 if (copy_from_user(&routing, argp, sizeof(routing)))
2234 goto out;
2235 r = -EINVAL;
2236 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2237 goto out;
2238 if (routing.flags)
2239 goto out;
2240 r = -ENOMEM;
2241 entries = vmalloc(routing.nr * sizeof(*entries));
2242 if (!entries)
2243 goto out;
2244 r = -EFAULT;
2245 urouting = argp;
2246 if (copy_from_user(entries, urouting->entries,
2247 routing.nr * sizeof(*entries)))
2248 goto out_free_irq_routing;
2249 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2250 routing.flags);
2251 out_free_irq_routing:
2252 vfree(entries);
2253 break;
2254 }
2255#endif /* KVM_CAP_IRQ_ROUTING */
2256#ifdef __KVM_HAVE_MSIX
2257 case KVM_ASSIGN_SET_MSIX_NR: {
2258 struct kvm_assigned_msix_nr entry_nr;
2259 r = -EFAULT;
2260 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2261 goto out;
2262 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2263 if (r)
2264 goto out;
2265 break;
2266 }
2267 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2268 struct kvm_assigned_msix_entry entry;
2269 r = -EFAULT;
2270 if (copy_from_user(&entry, argp, sizeof entry))
2271 goto out;
2272 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2273 if (r)
2274 goto out;
2275 break;
2276 }
2277#endif
2278 case KVM_IRQFD: { 1509 case KVM_IRQFD: {
2279 struct kvm_irqfd data; 1510 struct kvm_irqfd data;
2280 1511
@@ -2306,11 +1537,59 @@ static long kvm_vm_ioctl(struct file *filp,
2306#endif 1537#endif
2307 default: 1538 default:
2308 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1539 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1540 if (r == -ENOTTY)
1541 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2309 } 1542 }
2310out: 1543out:
2311 return r; 1544 return r;
2312} 1545}
2313 1546
1547#ifdef CONFIG_COMPAT
1548struct compat_kvm_dirty_log {
1549 __u32 slot;
1550 __u32 padding1;
1551 union {
1552 compat_uptr_t dirty_bitmap; /* one bit per page */
1553 __u64 padding2;
1554 };
1555};
1556
1557static long kvm_vm_compat_ioctl(struct file *filp,
1558 unsigned int ioctl, unsigned long arg)
1559{
1560 struct kvm *kvm = filp->private_data;
1561 int r;
1562
1563 if (kvm->mm != current->mm)
1564 return -EIO;
1565 switch (ioctl) {
1566 case KVM_GET_DIRTY_LOG: {
1567 struct compat_kvm_dirty_log compat_log;
1568 struct kvm_dirty_log log;
1569
1570 r = -EFAULT;
1571 if (copy_from_user(&compat_log, (void __user *)arg,
1572 sizeof(compat_log)))
1573 goto out;
1574 log.slot = compat_log.slot;
1575 log.padding1 = compat_log.padding1;
1576 log.padding2 = compat_log.padding2;
1577 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1578
1579 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1580 if (r)
1581 goto out;
1582 break;
1583 }
1584 default:
1585 r = kvm_vm_ioctl(filp, ioctl, arg);
1586 }
1587
1588out:
1589 return r;
1590}
1591#endif
1592
2314static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1593static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2315{ 1594{
2316 struct page *page[1]; 1595 struct page *page[1];
@@ -2345,7 +1624,9 @@ static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2345static struct file_operations kvm_vm_fops = { 1624static struct file_operations kvm_vm_fops = {
2346 .release = kvm_vm_release, 1625 .release = kvm_vm_release,
2347 .unlocked_ioctl = kvm_vm_ioctl, 1626 .unlocked_ioctl = kvm_vm_ioctl,
2348 .compat_ioctl = kvm_vm_ioctl, 1627#ifdef CONFIG_COMPAT
1628 .compat_ioctl = kvm_vm_compat_ioctl,
1629#endif
2349 .mmap = kvm_vm_mmap, 1630 .mmap = kvm_vm_mmap,
2350}; 1631};
2351 1632
@@ -2373,6 +1654,7 @@ static long kvm_dev_ioctl_check_extension_generic(long arg)
2373#ifdef CONFIG_KVM_APIC_ARCHITECTURE 1654#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2374 case KVM_CAP_SET_BOOT_CPU_ID: 1655 case KVM_CAP_SET_BOOT_CPU_ID:
2375#endif 1656#endif
1657 case KVM_CAP_INTERNAL_ERROR_DATA:
2376 return 1; 1658 return 1;
2377#ifdef CONFIG_HAVE_KVM_IRQCHIP 1659#ifdef CONFIG_HAVE_KVM_IRQCHIP
2378 case KVM_CAP_IRQ_ROUTING: 1660 case KVM_CAP_IRQ_ROUTING:
@@ -2443,11 +1725,21 @@ static struct miscdevice kvm_dev = {
2443static void hardware_enable(void *junk) 1725static void hardware_enable(void *junk)
2444{ 1726{
2445 int cpu = raw_smp_processor_id(); 1727 int cpu = raw_smp_processor_id();
1728 int r;
2446 1729
2447 if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) 1730 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2448 return; 1731 return;
1732
2449 cpumask_set_cpu(cpu, cpus_hardware_enabled); 1733 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2450 kvm_arch_hardware_enable(NULL); 1734
1735 r = kvm_arch_hardware_enable(NULL);
1736
1737 if (r) {
1738 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1739 atomic_inc(&hardware_enable_failed);
1740 printk(KERN_INFO "kvm: enabling virtualization on "
1741 "CPU%d failed\n", cpu);
1742 }
2451} 1743}
2452 1744
2453static void hardware_disable(void *junk) 1745static void hardware_disable(void *junk)
@@ -2460,11 +1752,52 @@ static void hardware_disable(void *junk)
2460 kvm_arch_hardware_disable(NULL); 1752 kvm_arch_hardware_disable(NULL);
2461} 1753}
2462 1754
1755static void hardware_disable_all_nolock(void)
1756{
1757 BUG_ON(!kvm_usage_count);
1758
1759 kvm_usage_count--;
1760 if (!kvm_usage_count)
1761 on_each_cpu(hardware_disable, NULL, 1);
1762}
1763
1764static void hardware_disable_all(void)
1765{
1766 spin_lock(&kvm_lock);
1767 hardware_disable_all_nolock();
1768 spin_unlock(&kvm_lock);
1769}
1770
1771static int hardware_enable_all(void)
1772{
1773 int r = 0;
1774
1775 spin_lock(&kvm_lock);
1776
1777 kvm_usage_count++;
1778 if (kvm_usage_count == 1) {
1779 atomic_set(&hardware_enable_failed, 0);
1780 on_each_cpu(hardware_enable, NULL, 1);
1781
1782 if (atomic_read(&hardware_enable_failed)) {
1783 hardware_disable_all_nolock();
1784 r = -EBUSY;
1785 }
1786 }
1787
1788 spin_unlock(&kvm_lock);
1789
1790 return r;
1791}
1792
2463static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, 1793static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2464 void *v) 1794 void *v)
2465{ 1795{
2466 int cpu = (long)v; 1796 int cpu = (long)v;
2467 1797
1798 if (!kvm_usage_count)
1799 return NOTIFY_OK;
1800
2468 val &= ~CPU_TASKS_FROZEN; 1801 val &= ~CPU_TASKS_FROZEN;
2469 switch (val) { 1802 switch (val) {
2470 case CPU_DYING: 1803 case CPU_DYING:
@@ -2667,13 +2000,15 @@ static void kvm_exit_debug(void)
2667 2000
2668static int kvm_suspend(struct sys_device *dev, pm_message_t state) 2001static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2669{ 2002{
2670 hardware_disable(NULL); 2003 if (kvm_usage_count)
2004 hardware_disable(NULL);
2671 return 0; 2005 return 0;
2672} 2006}
2673 2007
2674static int kvm_resume(struct sys_device *dev) 2008static int kvm_resume(struct sys_device *dev)
2675{ 2009{
2676 hardware_enable(NULL); 2010 if (kvm_usage_count)
2011 hardware_enable(NULL);
2677 return 0; 2012 return 0;
2678} 2013}
2679 2014
@@ -2748,7 +2083,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2748 goto out_free_1; 2083 goto out_free_1;
2749 } 2084 }
2750 2085
2751 on_each_cpu(hardware_enable, NULL, 1);
2752 r = register_cpu_notifier(&kvm_cpu_notifier); 2086 r = register_cpu_notifier(&kvm_cpu_notifier);
2753 if (r) 2087 if (r)
2754 goto out_free_2; 2088 goto out_free_2;
@@ -2798,7 +2132,6 @@ out_free_3:
2798 unregister_reboot_notifier(&kvm_reboot_notifier); 2132 unregister_reboot_notifier(&kvm_reboot_notifier);
2799 unregister_cpu_notifier(&kvm_cpu_notifier); 2133 unregister_cpu_notifier(&kvm_cpu_notifier);
2800out_free_2: 2134out_free_2:
2801 on_each_cpu(hardware_disable, NULL, 1);
2802out_free_1: 2135out_free_1:
2803 kvm_arch_hardware_unsetup(); 2136 kvm_arch_hardware_unsetup();
2804out_free_0a: 2137out_free_0a: