aboutsummaryrefslogtreecommitdiffstats
path: root/virt/kvm/kvm_main.c
diff options
context:
space:
mode:
authorThomas Gleixner <tglx@linutronix.de>2010-02-21 14:17:22 -0500
committerThomas Gleixner <tglx@linutronix.de>2010-02-21 14:17:22 -0500
commit5f854cfc024622e4aae14d7cf422f6ff86278688 (patch)
tree426e77c6f6e4939c80440bf1fabcb020e3ee145b /virt/kvm/kvm_main.c
parentcc24da0742870f152ddf1002aa39dfcd83f7cf9c (diff)
parent4ec62b2b2e6bd7ddef7b6cea6e5db7b5578a6532 (diff)
Forward to 2.6.33-rc8
Merge branch 'linus' into rt/head with a pile of conflicts. Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Diffstat (limited to 'virt/kvm/kvm_main.c')
-rw-r--r--virt/kvm/kvm_main.c1231
1 files changed, 347 insertions, 884 deletions
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index b24e96d5d40c..a51ba60a78b1 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -43,29 +43,36 @@
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>
49#include <asm/uaccess.h> 50#include <asm/uaccess.h>
50#include <asm/pgtable.h> 51#include <asm/pgtable.h>
52#include <asm-generic/bitops/le.h>
51 53
52#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 54#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
53#include "coalesced_mmio.h" 55#include "coalesced_mmio.h"
54#endif 56#endif
55 57
56#ifdef KVM_CAP_DEVICE_ASSIGNMENT 58#define CREATE_TRACE_POINTS
57#include <linux/pci.h> 59#include <trace/events/kvm.h>
58#include <linux/interrupt.h>
59#include "irq.h"
60#endif
61 60
62MODULE_AUTHOR("Qumranet"); 61MODULE_AUTHOR("Qumranet");
63MODULE_LICENSE("GPL"); 62MODULE_LICENSE("GPL");
64 63
64/*
65 * Ordering of locks:
66 *
67 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
68 */
69
65DEFINE_SPINLOCK(kvm_lock); 70DEFINE_SPINLOCK(kvm_lock);
66LIST_HEAD(vm_list); 71LIST_HEAD(vm_list);
67 72
68static 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;
69 76
70struct kmem_cache *kvm_vcpu_cache; 77struct kmem_cache *kvm_vcpu_cache;
71EXPORT_SYMBOL_GPL(kvm_vcpu_cache); 78EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
@@ -76,627 +83,12 @@ struct dentry *kvm_debugfs_dir;
76 83
77static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, 84static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
78 unsigned long arg); 85 unsigned long arg);
86static int hardware_enable_all(void);
87static void hardware_disable_all(void);
79 88
80static bool kvm_rebooting; 89static bool kvm_rebooting;
81 90
82#ifdef KVM_CAP_DEVICE_ASSIGNMENT 91static bool largepages_enabled = true;
83static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
84 int assigned_dev_id)
85{
86 struct list_head *ptr;
87 struct kvm_assigned_dev_kernel *match;
88
89 list_for_each(ptr, head) {
90 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
91 if (match->assigned_dev_id == assigned_dev_id)
92 return match;
93 }
94 return NULL;
95}
96
97static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
98 *assigned_dev, int irq)
99{
100 int i, index;
101 struct msix_entry *host_msix_entries;
102
103 host_msix_entries = assigned_dev->host_msix_entries;
104
105 index = -1;
106 for (i = 0; i < assigned_dev->entries_nr; i++)
107 if (irq == host_msix_entries[i].vector) {
108 index = i;
109 break;
110 }
111 if (index < 0) {
112 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
113 return 0;
114 }
115
116 return index;
117}
118
119static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
120{
121 struct kvm_assigned_dev_kernel *assigned_dev;
122 struct kvm *kvm;
123 int irq, i;
124
125 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
126 interrupt_work);
127 kvm = assigned_dev->kvm;
128
129 /* This is taken to safely inject irq inside the guest. When
130 * the interrupt injection (or the ioapic code) uses a
131 * finer-grained lock, update this
132 */
133 mutex_lock(&kvm->lock);
134 spin_lock_irq(&assigned_dev->assigned_dev_lock);
135 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
136 struct kvm_guest_msix_entry *guest_entries =
137 assigned_dev->guest_msix_entries;
138 for (i = 0; i < assigned_dev->entries_nr; i++) {
139 if (!(guest_entries[i].flags &
140 KVM_ASSIGNED_MSIX_PENDING))
141 continue;
142 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
143 kvm_set_irq(assigned_dev->kvm,
144 assigned_dev->irq_source_id,
145 guest_entries[i].vector, 1);
146 irq = assigned_dev->host_msix_entries[i].vector;
147 if (irq != 0)
148 enable_irq(irq);
149 assigned_dev->host_irq_disabled = false;
150 }
151 } else {
152 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
153 assigned_dev->guest_irq, 1);
154 if (assigned_dev->irq_requested_type &
155 KVM_DEV_IRQ_GUEST_MSI) {
156 enable_irq(assigned_dev->host_irq);
157 assigned_dev->host_irq_disabled = false;
158 }
159 }
160
161 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
162 mutex_unlock(&assigned_dev->kvm->lock);
163}
164
165static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
166{
167 unsigned long flags;
168 struct kvm_assigned_dev_kernel *assigned_dev =
169 (struct kvm_assigned_dev_kernel *) dev_id;
170
171 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
172 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
173 int index = find_index_from_host_irq(assigned_dev, irq);
174 if (index < 0)
175 goto out;
176 assigned_dev->guest_msix_entries[index].flags |=
177 KVM_ASSIGNED_MSIX_PENDING;
178 }
179
180 schedule_work(&assigned_dev->interrupt_work);
181
182 disable_irq_nosync(irq);
183 assigned_dev->host_irq_disabled = true;
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(&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 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 return 0;
431}
432#endif
433
434static int assign_host_irq(struct kvm *kvm,
435 struct kvm_assigned_dev_kernel *dev,
436 __u32 host_irq_type)
437{
438 int r = -EEXIST;
439
440 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
441 return r;
442
443 switch (host_irq_type) {
444 case KVM_DEV_IRQ_HOST_INTX:
445 r = assigned_device_enable_host_intx(kvm, dev);
446 break;
447#ifdef __KVM_HAVE_MSI
448 case KVM_DEV_IRQ_HOST_MSI:
449 r = assigned_device_enable_host_msi(kvm, dev);
450 break;
451#endif
452#ifdef __KVM_HAVE_MSIX
453 case KVM_DEV_IRQ_HOST_MSIX:
454 r = assigned_device_enable_host_msix(kvm, dev);
455 break;
456#endif
457 default:
458 r = -EINVAL;
459 }
460
461 if (!r)
462 dev->irq_requested_type |= host_irq_type;
463
464 return r;
465}
466
467static int assign_guest_irq(struct kvm *kvm,
468 struct kvm_assigned_dev_kernel *dev,
469 struct kvm_assigned_irq *irq,
470 unsigned long guest_irq_type)
471{
472 int id;
473 int r = -EEXIST;
474
475 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
476 return r;
477
478 id = kvm_request_irq_source_id(kvm);
479 if (id < 0)
480 return id;
481
482 dev->irq_source_id = id;
483
484 switch (guest_irq_type) {
485 case KVM_DEV_IRQ_GUEST_INTX:
486 r = assigned_device_enable_guest_intx(kvm, dev, irq);
487 break;
488#ifdef __KVM_HAVE_MSI
489 case KVM_DEV_IRQ_GUEST_MSI:
490 r = assigned_device_enable_guest_msi(kvm, dev, irq);
491 break;
492#endif
493#ifdef __KVM_HAVE_MSIX
494 case KVM_DEV_IRQ_GUEST_MSIX:
495 r = assigned_device_enable_guest_msix(kvm, dev, irq);
496 break;
497#endif
498 default:
499 r = -EINVAL;
500 }
501
502 if (!r) {
503 dev->irq_requested_type |= guest_irq_type;
504 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
505 } else
506 kvm_free_irq_source_id(kvm, dev->irq_source_id);
507
508 return r;
509}
510
511/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
512static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
513 struct kvm_assigned_irq *assigned_irq)
514{
515 int r = -EINVAL;
516 struct kvm_assigned_dev_kernel *match;
517 unsigned long host_irq_type, guest_irq_type;
518
519 if (!capable(CAP_SYS_RAWIO))
520 return -EPERM;
521
522 if (!irqchip_in_kernel(kvm))
523 return r;
524
525 mutex_lock(&kvm->lock);
526 r = -ENODEV;
527 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
528 assigned_irq->assigned_dev_id);
529 if (!match)
530 goto out;
531
532 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
533 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
534
535 r = -EINVAL;
536 /* can only assign one type at a time */
537 if (hweight_long(host_irq_type) > 1)
538 goto out;
539 if (hweight_long(guest_irq_type) > 1)
540 goto out;
541 if (host_irq_type == 0 && guest_irq_type == 0)
542 goto out;
543
544 r = 0;
545 if (host_irq_type)
546 r = assign_host_irq(kvm, match, host_irq_type);
547 if (r)
548 goto out;
549
550 if (guest_irq_type)
551 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
552out:
553 mutex_unlock(&kvm->lock);
554 return r;
555}
556
557static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
558 struct kvm_assigned_irq
559 *assigned_irq)
560{
561 int r = -ENODEV;
562 struct kvm_assigned_dev_kernel *match;
563
564 mutex_lock(&kvm->lock);
565
566 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
567 assigned_irq->assigned_dev_id);
568 if (!match)
569 goto out;
570
571 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
572out:
573 mutex_unlock(&kvm->lock);
574 return r;
575}
576
577static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
578 struct kvm_assigned_pci_dev *assigned_dev)
579{
580 int r = 0;
581 struct kvm_assigned_dev_kernel *match;
582 struct pci_dev *dev;
583
584 down_read(&kvm->slots_lock);
585 mutex_lock(&kvm->lock);
586
587 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
588 assigned_dev->assigned_dev_id);
589 if (match) {
590 /* device already assigned */
591 r = -EEXIST;
592 goto out;
593 }
594
595 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
596 if (match == NULL) {
597 printk(KERN_INFO "%s: Couldn't allocate memory\n",
598 __func__);
599 r = -ENOMEM;
600 goto out;
601 }
602 dev = pci_get_bus_and_slot(assigned_dev->busnr,
603 assigned_dev->devfn);
604 if (!dev) {
605 printk(KERN_INFO "%s: host device not found\n", __func__);
606 r = -EINVAL;
607 goto out_free;
608 }
609 if (pci_enable_device(dev)) {
610 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
611 r = -EBUSY;
612 goto out_put;
613 }
614 r = pci_request_regions(dev, "kvm_assigned_device");
615 if (r) {
616 printk(KERN_INFO "%s: Could not get access to device regions\n",
617 __func__);
618 goto out_disable;
619 }
620
621 pci_reset_function(dev);
622
623 match->assigned_dev_id = assigned_dev->assigned_dev_id;
624 match->host_busnr = assigned_dev->busnr;
625 match->host_devfn = assigned_dev->devfn;
626 match->flags = assigned_dev->flags;
627 match->dev = dev;
628 spin_lock_init(&match->assigned_dev_lock);
629 match->irq_source_id = -1;
630 match->kvm = kvm;
631 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
632 INIT_WORK(&match->interrupt_work,
633 kvm_assigned_dev_interrupt_work_handler);
634
635 list_add(&match->list, &kvm->arch.assigned_dev_head);
636
637 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
638 if (!kvm->arch.iommu_domain) {
639 r = kvm_iommu_map_guest(kvm);
640 if (r)
641 goto out_list_del;
642 }
643 r = kvm_assign_device(kvm, match);
644 if (r)
645 goto out_list_del;
646 }
647
648out:
649 mutex_unlock(&kvm->lock);
650 up_read(&kvm->slots_lock);
651 return r;
652out_list_del:
653 list_del(&match->list);
654 pci_release_regions(dev);
655out_disable:
656 pci_disable_device(dev);
657out_put:
658 pci_dev_put(dev);
659out_free:
660 kfree(match);
661 mutex_unlock(&kvm->lock);
662 up_read(&kvm->slots_lock);
663 return r;
664}
665#endif
666
667#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
668static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
669 struct kvm_assigned_pci_dev *assigned_dev)
670{
671 int r = 0;
672 struct kvm_assigned_dev_kernel *match;
673
674 mutex_lock(&kvm->lock);
675
676 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
677 assigned_dev->assigned_dev_id);
678 if (!match) {
679 printk(KERN_INFO "%s: device hasn't been assigned before, "
680 "so cannot be deassigned\n", __func__);
681 r = -EINVAL;
682 goto out;
683 }
684
685 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
686 kvm_deassign_device(kvm, match);
687
688 kvm_free_assigned_device(kvm, match);
689
690out:
691 mutex_unlock(&kvm->lock);
692 return r;
693}
694#endif
695
696static inline int valid_vcpu(int n)
697{
698 return likely(n >= 0 && n < KVM_MAX_VCPUS);
699}
700 92
701inline int kvm_is_mmio_pfn(pfn_t pfn) 93inline int kvm_is_mmio_pfn(pfn_t pfn)
702{ 94{
@@ -742,15 +134,11 @@ static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
742 bool called = true; 134 bool called = true;
743 struct kvm_vcpu *vcpu; 135 struct kvm_vcpu *vcpu;
744 136
745 if (alloc_cpumask_var(&cpus, GFP_ATOMIC)) 137 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
746 cpumask_clear(cpus);
747 138
748 spin_lock(&kvm->requests_lock); 139 spin_lock(&kvm->requests_lock);
749 me = get_cpu(); 140 me = raw_smp_processor_id();
750 for (i = 0; i < KVM_MAX_VCPUS; ++i) { 141 kvm_for_each_vcpu(i, vcpu, kvm) {
751 vcpu = kvm->vcpus[i];
752 if (!vcpu)
753 continue;
754 if (test_and_set_bit(req, &vcpu->requests)) 142 if (test_and_set_bit(req, &vcpu->requests))
755 continue; 143 continue;
756 cpu = vcpu->cpu; 144 cpu = vcpu->cpu;
@@ -763,7 +151,6 @@ static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
763 smp_call_function_many(cpus, ack_flush, NULL, 1); 151 smp_call_function_many(cpus, ack_flush, NULL, 1);
764 else 152 else
765 called = false; 153 called = false;
766 put_cpu();
767 spin_unlock(&kvm->requests_lock); 154 spin_unlock(&kvm->requests_lock);
768 free_cpumask_var(cpus); 155 free_cpumask_var(cpus);
769 return called; 156 return called;
@@ -859,6 +246,19 @@ static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
859 246
860} 247}
861 248
249static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
250 struct mm_struct *mm,
251 unsigned long address,
252 pte_t pte)
253{
254 struct kvm *kvm = mmu_notifier_to_kvm(mn);
255
256 spin_lock(&kvm->mmu_lock);
257 kvm->mmu_notifier_seq++;
258 kvm_set_spte_hva(kvm, address, pte);
259 spin_unlock(&kvm->mmu_lock);
260}
261
862static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn, 262static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
863 struct mm_struct *mm, 263 struct mm_struct *mm,
864 unsigned long start, 264 unsigned long start,
@@ -938,12 +338,14 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
938 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start, 338 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
939 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end, 339 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
940 .clear_flush_young = kvm_mmu_notifier_clear_flush_young, 340 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
341 .change_pte = kvm_mmu_notifier_change_pte,
941 .release = kvm_mmu_notifier_release, 342 .release = kvm_mmu_notifier_release,
942}; 343};
943#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */ 344#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
944 345
945static struct kvm *kvm_create_vm(void) 346static struct kvm *kvm_create_vm(void)
946{ 347{
348 int r = 0;
947 struct kvm *kvm = kvm_arch_create_vm(); 349 struct kvm *kvm = kvm_arch_create_vm();
948#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 350#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
949 struct page *page; 351 struct page *page;
@@ -951,16 +353,21 @@ static struct kvm *kvm_create_vm(void)
951 353
952 if (IS_ERR(kvm)) 354 if (IS_ERR(kvm))
953 goto out; 355 goto out;
356
357 r = hardware_enable_all();
358 if (r)
359 goto out_err_nodisable;
360
954#ifdef CONFIG_HAVE_KVM_IRQCHIP 361#ifdef CONFIG_HAVE_KVM_IRQCHIP
955 INIT_LIST_HEAD(&kvm->irq_routing);
956 INIT_HLIST_HEAD(&kvm->mask_notifier_list); 362 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
363 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
957#endif 364#endif
958 365
959#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 366#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
960 page = alloc_page(GFP_KERNEL | __GFP_ZERO); 367 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
961 if (!page) { 368 if (!page) {
962 kfree(kvm); 369 r = -ENOMEM;
963 return ERR_PTR(-ENOMEM); 370 goto out_err;
964 } 371 }
965 kvm->coalesced_mmio_ring = 372 kvm->coalesced_mmio_ring =
966 (struct kvm_coalesced_mmio_ring *)page_address(page); 373 (struct kvm_coalesced_mmio_ring *)page_address(page);
@@ -968,15 +375,13 @@ static struct kvm *kvm_create_vm(void)
968 375
969#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 376#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
970 { 377 {
971 int err;
972 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; 378 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
973 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm); 379 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
974 if (err) { 380 if (r) {
975#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 381#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
976 put_page(page); 382 put_page(page);
977#endif 383#endif
978 kfree(kvm); 384 goto out_err;
979 return ERR_PTR(err);
980 } 385 }
981 } 386 }
982#endif 387#endif
@@ -986,7 +391,9 @@ static struct kvm *kvm_create_vm(void)
986 spin_lock_init(&kvm->mmu_lock); 391 spin_lock_init(&kvm->mmu_lock);
987 spin_lock_init(&kvm->requests_lock); 392 spin_lock_init(&kvm->requests_lock);
988 kvm_io_bus_init(&kvm->pio_bus); 393 kvm_io_bus_init(&kvm->pio_bus);
394 kvm_eventfd_init(kvm);
989 mutex_init(&kvm->lock); 395 mutex_init(&kvm->lock);
396 mutex_init(&kvm->irq_lock);
990 kvm_io_bus_init(&kvm->mmio_bus); 397 kvm_io_bus_init(&kvm->mmio_bus);
991 init_rwsem(&kvm->slots_lock); 398 init_rwsem(&kvm->slots_lock);
992 atomic_set(&kvm->users_count, 1); 399 atomic_set(&kvm->users_count, 1);
@@ -998,6 +405,15 @@ static struct kvm *kvm_create_vm(void)
998#endif 405#endif
999out: 406out:
1000 return kvm; 407 return kvm;
408
409#if defined(KVM_COALESCED_MMIO_PAGE_OFFSET) || \
410 (defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER))
411out_err:
412 hardware_disable_all();
413#endif
414out_err_nodisable:
415 kfree(kvm);
416 return ERR_PTR(r);
1001} 417}
1002 418
1003/* 419/*
@@ -1006,19 +422,25 @@ out:
1006static void kvm_free_physmem_slot(struct kvm_memory_slot *free, 422static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
1007 struct kvm_memory_slot *dont) 423 struct kvm_memory_slot *dont)
1008{ 424{
425 int i;
426
1009 if (!dont || free->rmap != dont->rmap) 427 if (!dont || free->rmap != dont->rmap)
1010 vfree(free->rmap); 428 vfree(free->rmap);
1011 429
1012 if (!dont || free->dirty_bitmap != dont->dirty_bitmap) 430 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
1013 vfree(free->dirty_bitmap); 431 vfree(free->dirty_bitmap);
1014 432
1015 if (!dont || free->lpage_info != dont->lpage_info) 433
1016 vfree(free->lpage_info); 434 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
435 if (!dont || free->lpage_info[i] != dont->lpage_info[i]) {
436 vfree(free->lpage_info[i]);
437 free->lpage_info[i] = NULL;
438 }
439 }
1017 440
1018 free->npages = 0; 441 free->npages = 0;
1019 free->dirty_bitmap = NULL; 442 free->dirty_bitmap = NULL;
1020 free->rmap = NULL; 443 free->rmap = NULL;
1021 free->lpage_info = NULL;
1022} 444}
1023 445
1024void kvm_free_physmem(struct kvm *kvm) 446void kvm_free_physmem(struct kvm *kvm)
@@ -1050,6 +472,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
1050 kvm_arch_flush_shadow(kvm); 472 kvm_arch_flush_shadow(kvm);
1051#endif 473#endif
1052 kvm_arch_destroy_vm(kvm); 474 kvm_arch_destroy_vm(kvm);
475 hardware_disable_all();
1053 mmdrop(mm); 476 mmdrop(mm);
1054} 477}
1055 478
@@ -1071,6 +494,8 @@ static int kvm_vm_release(struct inode *inode, struct file *filp)
1071{ 494{
1072 struct kvm *kvm = filp->private_data; 495 struct kvm *kvm = filp->private_data;
1073 496
497 kvm_irqfd_release(kvm);
498
1074 kvm_put_kvm(kvm); 499 kvm_put_kvm(kvm);
1075 return 0; 500 return 0;
1076} 501}
@@ -1089,8 +514,8 @@ int __kvm_set_memory_region(struct kvm *kvm,
1089{ 514{
1090 int r; 515 int r;
1091 gfn_t base_gfn; 516 gfn_t base_gfn;
1092 unsigned long npages, ugfn; 517 unsigned long npages;
1093 unsigned long largepages, i; 518 unsigned long i;
1094 struct kvm_memory_slot *memslot; 519 struct kvm_memory_slot *memslot;
1095 struct kvm_memory_slot old, new; 520 struct kvm_memory_slot old, new;
1096 521
@@ -1164,31 +589,51 @@ int __kvm_set_memory_region(struct kvm *kvm,
1164 else 589 else
1165 new.userspace_addr = 0; 590 new.userspace_addr = 0;
1166 } 591 }
1167 if (npages && !new.lpage_info) { 592 if (!npages)
1168 largepages = 1 + (base_gfn + npages - 1) / KVM_PAGES_PER_HPAGE; 593 goto skip_lpage;
1169 largepages -= base_gfn / KVM_PAGES_PER_HPAGE; 594
595 for (i = 0; i < KVM_NR_PAGE_SIZES - 1; ++i) {
596 unsigned long ugfn;
597 unsigned long j;
598 int lpages;
599 int level = i + 2;
1170 600
1171 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info)); 601 /* Avoid unused variable warning if no large pages */
602 (void)level;
1172 603
1173 if (!new.lpage_info) 604 if (new.lpage_info[i])
605 continue;
606
607 lpages = 1 + (base_gfn + npages - 1) /
608 KVM_PAGES_PER_HPAGE(level);
609 lpages -= base_gfn / KVM_PAGES_PER_HPAGE(level);
610
611 new.lpage_info[i] = vmalloc(lpages * sizeof(*new.lpage_info[i]));
612
613 if (!new.lpage_info[i])
1174 goto out_free; 614 goto out_free;
1175 615
1176 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info)); 616 memset(new.lpage_info[i], 0,
617 lpages * sizeof(*new.lpage_info[i]));
1177 618
1178 if (base_gfn % KVM_PAGES_PER_HPAGE) 619 if (base_gfn % KVM_PAGES_PER_HPAGE(level))
1179 new.lpage_info[0].write_count = 1; 620 new.lpage_info[i][0].write_count = 1;
1180 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE) 621 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE(level))
1181 new.lpage_info[largepages-1].write_count = 1; 622 new.lpage_info[i][lpages - 1].write_count = 1;
1182 ugfn = new.userspace_addr >> PAGE_SHIFT; 623 ugfn = new.userspace_addr >> PAGE_SHIFT;
1183 /* 624 /*
1184 * If the gfn and userspace address are not aligned wrt each 625 * If the gfn and userspace address are not aligned wrt each
1185 * other, disable large page support for this slot 626 * other, or if explicitly asked to, disable large page
627 * support for this slot
1186 */ 628 */
1187 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE - 1)) 629 if ((base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
1188 for (i = 0; i < largepages; ++i) 630 !largepages_enabled)
1189 new.lpage_info[i].write_count = 1; 631 for (j = 0; j < lpages; ++j)
632 new.lpage_info[i][j].write_count = 1;
1190 } 633 }
1191 634
635skip_lpage:
636
1192 /* Allocate page dirty bitmap if needed */ 637 /* Allocate page dirty bitmap if needed */
1193 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) { 638 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1194 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8; 639 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
@@ -1200,6 +645,10 @@ int __kvm_set_memory_region(struct kvm *kvm,
1200 if (old.npages) 645 if (old.npages)
1201 kvm_arch_flush_shadow(kvm); 646 kvm_arch_flush_shadow(kvm);
1202 } 647 }
648#else /* not defined CONFIG_S390 */
649 new.user_alloc = user_alloc;
650 if (user_alloc)
651 new.userspace_addr = mem->userspace_addr;
1203#endif /* not defined CONFIG_S390 */ 652#endif /* not defined CONFIG_S390 */
1204 653
1205 if (!npages) 654 if (!npages)
@@ -1299,6 +748,12 @@ out:
1299 return r; 748 return r;
1300} 749}
1301 750
751void kvm_disable_largepages(void)
752{
753 largepages_enabled = false;
754}
755EXPORT_SYMBOL_GPL(kvm_disable_largepages);
756
1302int is_error_page(struct page *page) 757int is_error_page(struct page *page)
1303{ 758{
1304 return page == bad_page; 759 return page == bad_page;
@@ -1620,8 +1075,8 @@ void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1620 unsigned long rel_gfn = gfn - memslot->base_gfn; 1075 unsigned long rel_gfn = gfn - memslot->base_gfn;
1621 1076
1622 /* avoid RMW */ 1077 /* avoid RMW */
1623 if (!test_bit(rel_gfn, memslot->dirty_bitmap)) 1078 if (!generic_test_le_bit(rel_gfn, memslot->dirty_bitmap))
1624 set_bit(rel_gfn, memslot->dirty_bitmap); 1079 generic___set_le_bit(rel_gfn, memslot->dirty_bitmap);
1625 } 1080 }
1626} 1081}
1627 1082
@@ -1635,9 +1090,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1635 for (;;) { 1090 for (;;) {
1636 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE); 1091 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1637 1092
1638 if ((kvm_arch_interrupt_allowed(vcpu) && 1093 if (kvm_arch_vcpu_runnable(vcpu)) {
1639 kvm_cpu_has_interrupt(vcpu)) ||
1640 kvm_arch_vcpu_runnable(vcpu)) {
1641 set_bit(KVM_REQ_UNHALT, &vcpu->requests); 1094 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1642 break; 1095 break;
1643 } 1096 }
@@ -1646,9 +1099,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1646 if (signal_pending(current)) 1099 if (signal_pending(current))
1647 break; 1100 break;
1648 1101
1649 vcpu_put(vcpu);
1650 schedule(); 1102 schedule();
1651 vcpu_load(vcpu);
1652 } 1103 }
1653 1104
1654 finish_wait(&vcpu->wq, &wait); 1105 finish_wait(&vcpu->wq, &wait);
@@ -1662,6 +1113,21 @@ void kvm_resched(struct kvm_vcpu *vcpu)
1662} 1113}
1663EXPORT_SYMBOL_GPL(kvm_resched); 1114EXPORT_SYMBOL_GPL(kvm_resched);
1664 1115
1116void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1117{
1118 ktime_t expires;
1119 DEFINE_WAIT(wait);
1120
1121 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1122
1123 /* Sleep for 100 us, and hope lock-holder got scheduled */
1124 expires = ktime_add_ns(ktime_get(), 100000UL);
1125 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1126
1127 finish_wait(&vcpu->wq, &wait);
1128}
1129EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1130
1665static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1131static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1666{ 1132{
1667 struct kvm_vcpu *vcpu = vma->vm_file->private_data; 1133 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
@@ -1684,7 +1150,7 @@ static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1684 return 0; 1150 return 0;
1685} 1151}
1686 1152
1687static struct vm_operations_struct kvm_vcpu_vm_ops = { 1153static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1688 .fault = kvm_vcpu_fault, 1154 .fault = kvm_vcpu_fault,
1689}; 1155};
1690 1156
@@ -1714,24 +1180,18 @@ static struct file_operations kvm_vcpu_fops = {
1714 */ 1180 */
1715static int create_vcpu_fd(struct kvm_vcpu *vcpu) 1181static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1716{ 1182{
1717 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0); 1183 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1718 if (fd < 0)
1719 kvm_put_kvm(vcpu->kvm);
1720 return fd;
1721} 1184}
1722 1185
1723/* 1186/*
1724 * Creates some virtual cpus. Good luck creating more than one. 1187 * Creates some virtual cpus. Good luck creating more than one.
1725 */ 1188 */
1726static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n) 1189static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1727{ 1190{
1728 int r; 1191 int r;
1729 struct kvm_vcpu *vcpu; 1192 struct kvm_vcpu *vcpu, *v;
1730 1193
1731 if (!valid_vcpu(n)) 1194 vcpu = kvm_arch_vcpu_create(kvm, id);
1732 return -EINVAL;
1733
1734 vcpu = kvm_arch_vcpu_create(kvm, n);
1735 if (IS_ERR(vcpu)) 1195 if (IS_ERR(vcpu))
1736 return PTR_ERR(vcpu); 1196 return PTR_ERR(vcpu);
1737 1197
@@ -1742,23 +1202,38 @@ static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1742 return r; 1202 return r;
1743 1203
1744 mutex_lock(&kvm->lock); 1204 mutex_lock(&kvm->lock);
1745 if (kvm->vcpus[n]) { 1205 if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1746 r = -EEXIST; 1206 r = -EINVAL;
1747 goto vcpu_destroy; 1207 goto vcpu_destroy;
1748 } 1208 }
1749 kvm->vcpus[n] = vcpu; 1209
1750 mutex_unlock(&kvm->lock); 1210 kvm_for_each_vcpu(r, v, kvm)
1211 if (v->vcpu_id == id) {
1212 r = -EEXIST;
1213 goto vcpu_destroy;
1214 }
1215
1216 BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1751 1217
1752 /* Now it's all set up, let userspace reach it */ 1218 /* Now it's all set up, let userspace reach it */
1753 kvm_get_kvm(kvm); 1219 kvm_get_kvm(kvm);
1754 r = create_vcpu_fd(vcpu); 1220 r = create_vcpu_fd(vcpu);
1755 if (r < 0) 1221 if (r < 0) {
1756 goto unlink; 1222 kvm_put_kvm(kvm);
1223 goto vcpu_destroy;
1224 }
1225
1226 kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1227 smp_wmb();
1228 atomic_inc(&kvm->online_vcpus);
1229
1230#ifdef CONFIG_KVM_APIC_ARCHITECTURE
1231 if (kvm->bsp_vcpu_id == id)
1232 kvm->bsp_vcpu = vcpu;
1233#endif
1234 mutex_unlock(&kvm->lock);
1757 return r; 1235 return r;
1758 1236
1759unlink:
1760 mutex_lock(&kvm->lock);
1761 kvm->vcpus[n] = NULL;
1762vcpu_destroy: 1237vcpu_destroy:
1763 mutex_unlock(&kvm->lock); 1238 mutex_unlock(&kvm->lock);
1764 kvm_arch_vcpu_destroy(vcpu); 1239 kvm_arch_vcpu_destroy(vcpu);
@@ -1776,88 +1251,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1776 return 0; 1251 return 0;
1777} 1252}
1778 1253
1779#ifdef __KVM_HAVE_MSIX
1780static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1781 struct kvm_assigned_msix_nr *entry_nr)
1782{
1783 int r = 0;
1784 struct kvm_assigned_dev_kernel *adev;
1785
1786 mutex_lock(&kvm->lock);
1787
1788 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1789 entry_nr->assigned_dev_id);
1790 if (!adev) {
1791 r = -EINVAL;
1792 goto msix_nr_out;
1793 }
1794
1795 if (adev->entries_nr == 0) {
1796 adev->entries_nr = entry_nr->entry_nr;
1797 if (adev->entries_nr == 0 ||
1798 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1799 r = -EINVAL;
1800 goto msix_nr_out;
1801 }
1802
1803 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1804 entry_nr->entry_nr,
1805 GFP_KERNEL);
1806 if (!adev->host_msix_entries) {
1807 r = -ENOMEM;
1808 goto msix_nr_out;
1809 }
1810 adev->guest_msix_entries = kzalloc(
1811 sizeof(struct kvm_guest_msix_entry) *
1812 entry_nr->entry_nr, GFP_KERNEL);
1813 if (!adev->guest_msix_entries) {
1814 kfree(adev->host_msix_entries);
1815 r = -ENOMEM;
1816 goto msix_nr_out;
1817 }
1818 } else /* Not allowed set MSI-X number twice */
1819 r = -EINVAL;
1820msix_nr_out:
1821 mutex_unlock(&kvm->lock);
1822 return r;
1823}
1824
1825static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1826 struct kvm_assigned_msix_entry *entry)
1827{
1828 int r = 0, i;
1829 struct kvm_assigned_dev_kernel *adev;
1830
1831 mutex_lock(&kvm->lock);
1832
1833 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1834 entry->assigned_dev_id);
1835
1836 if (!adev) {
1837 r = -EINVAL;
1838 goto msix_entry_out;
1839 }
1840
1841 for (i = 0; i < adev->entries_nr; i++)
1842 if (adev->guest_msix_entries[i].vector == 0 ||
1843 adev->guest_msix_entries[i].entry == entry->entry) {
1844 adev->guest_msix_entries[i].entry = entry->entry;
1845 adev->guest_msix_entries[i].vector = entry->gsi;
1846 adev->host_msix_entries[i].entry = entry->entry;
1847 break;
1848 }
1849 if (i == adev->entries_nr) {
1850 r = -ENOSPC;
1851 goto msix_entry_out;
1852 }
1853
1854msix_entry_out:
1855 mutex_unlock(&kvm->lock);
1856
1857 return r;
1858}
1859#endif
1860
1861static long kvm_vcpu_ioctl(struct file *filp, 1254static long kvm_vcpu_ioctl(struct file *filp,
1862 unsigned int ioctl, unsigned long arg) 1255 unsigned int ioctl, unsigned long arg)
1863{ 1256{
@@ -2116,118 +1509,89 @@ static long kvm_vm_ioctl(struct file *filp,
2116 break; 1509 break;
2117 } 1510 }
2118#endif 1511#endif
2119#ifdef KVM_CAP_DEVICE_ASSIGNMENT 1512 case KVM_IRQFD: {
2120 case KVM_ASSIGN_PCI_DEVICE: { 1513 struct kvm_irqfd data;
2121 struct kvm_assigned_pci_dev assigned_dev;
2122 1514
2123 r = -EFAULT; 1515 r = -EFAULT;
2124 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev)) 1516 if (copy_from_user(&data, argp, sizeof data))
2125 goto out;
2126 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2127 if (r)
2128 goto out; 1517 goto out;
1518 r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2129 break; 1519 break;
2130 } 1520 }
2131 case KVM_ASSIGN_IRQ: { 1521 case KVM_IOEVENTFD: {
2132 r = -EOPNOTSUPP; 1522 struct kvm_ioeventfd data;
2133 break;
2134 }
2135#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2136 case KVM_ASSIGN_DEV_IRQ: {
2137 struct kvm_assigned_irq assigned_irq;
2138 1523
2139 r = -EFAULT; 1524 r = -EFAULT;
2140 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq)) 1525 if (copy_from_user(&data, argp, sizeof data))
2141 goto out;
2142 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2143 if (r)
2144 goto out; 1526 goto out;
1527 r = kvm_ioeventfd(kvm, &data);
2145 break; 1528 break;
2146 } 1529 }
2147 case KVM_DEASSIGN_DEV_IRQ: { 1530#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2148 struct kvm_assigned_irq assigned_irq; 1531 case KVM_SET_BOOT_CPU_ID:
2149 1532 r = 0;
2150 r = -EFAULT; 1533 mutex_lock(&kvm->lock);
2151 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq)) 1534 if (atomic_read(&kvm->online_vcpus) != 0)
2152 goto out; 1535 r = -EBUSY;
2153 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq); 1536 else
2154 if (r) 1537 kvm->bsp_vcpu_id = arg;
2155 goto out; 1538 mutex_unlock(&kvm->lock);
2156 break; 1539 break;
2157 }
2158#endif
2159#endif 1540#endif
2160#ifdef KVM_CAP_DEVICE_DEASSIGNMENT 1541 default:
2161 case KVM_DEASSIGN_PCI_DEVICE: { 1542 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2162 struct kvm_assigned_pci_dev assigned_dev; 1543 if (r == -ENOTTY)
2163 1544 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2164 r = -EFAULT;
2165 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2166 goto out;
2167 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2168 if (r)
2169 goto out;
2170 break;
2171 } 1545 }
2172#endif 1546out:
2173#ifdef KVM_CAP_IRQ_ROUTING 1547 return r;
2174 case KVM_SET_GSI_ROUTING: { 1548}
2175 struct kvm_irq_routing routing; 1549
2176 struct kvm_irq_routing __user *urouting; 1550#ifdef CONFIG_COMPAT
2177 struct kvm_irq_routing_entry *entries; 1551struct compat_kvm_dirty_log {
1552 __u32 slot;
1553 __u32 padding1;
1554 union {
1555 compat_uptr_t dirty_bitmap; /* one bit per page */
1556 __u64 padding2;
1557 };
1558};
1559
1560static long kvm_vm_compat_ioctl(struct file *filp,
1561 unsigned int ioctl, unsigned long arg)
1562{
1563 struct kvm *kvm = filp->private_data;
1564 int r;
1565
1566 if (kvm->mm != current->mm)
1567 return -EIO;
1568 switch (ioctl) {
1569 case KVM_GET_DIRTY_LOG: {
1570 struct compat_kvm_dirty_log compat_log;
1571 struct kvm_dirty_log log;
2178 1572
2179 r = -EFAULT; 1573 r = -EFAULT;
2180 if (copy_from_user(&routing, argp, sizeof(routing))) 1574 if (copy_from_user(&compat_log, (void __user *)arg,
2181 goto out; 1575 sizeof(compat_log)))
2182 r = -EINVAL;
2183 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2184 goto out;
2185 if (routing.flags)
2186 goto out;
2187 r = -ENOMEM;
2188 entries = vmalloc(routing.nr * sizeof(*entries));
2189 if (!entries)
2190 goto out;
2191 r = -EFAULT;
2192 urouting = argp;
2193 if (copy_from_user(entries, urouting->entries,
2194 routing.nr * sizeof(*entries)))
2195 goto out_free_irq_routing;
2196 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2197 routing.flags);
2198 out_free_irq_routing:
2199 vfree(entries);
2200 break;
2201 }
2202#ifdef __KVM_HAVE_MSIX
2203 case KVM_ASSIGN_SET_MSIX_NR: {
2204 struct kvm_assigned_msix_nr entry_nr;
2205 r = -EFAULT;
2206 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2207 goto out;
2208 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2209 if (r)
2210 goto out;
2211 break;
2212 }
2213 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2214 struct kvm_assigned_msix_entry entry;
2215 r = -EFAULT;
2216 if (copy_from_user(&entry, argp, sizeof entry))
2217 goto out; 1576 goto out;
2218 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry); 1577 log.slot = compat_log.slot;
1578 log.padding1 = compat_log.padding1;
1579 log.padding2 = compat_log.padding2;
1580 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1581
1582 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2219 if (r) 1583 if (r)
2220 goto out; 1584 goto out;
2221 break; 1585 break;
2222 } 1586 }
2223#endif
2224#endif /* KVM_CAP_IRQ_ROUTING */
2225 default: 1587 default:
2226 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1588 r = kvm_vm_ioctl(filp, ioctl, arg);
2227 } 1589 }
1590
2228out: 1591out:
2229 return r; 1592 return r;
2230} 1593}
1594#endif
2231 1595
2232static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1596static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2233{ 1597{
@@ -2250,7 +1614,7 @@ static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2250 return 0; 1614 return 0;
2251} 1615}
2252 1616
2253static struct vm_operations_struct kvm_vm_vm_ops = { 1617static const struct vm_operations_struct kvm_vm_vm_ops = {
2254 .fault = kvm_vm_fault, 1618 .fault = kvm_vm_fault,
2255}; 1619};
2256 1620
@@ -2263,7 +1627,9 @@ static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2263static struct file_operations kvm_vm_fops = { 1627static struct file_operations kvm_vm_fops = {
2264 .release = kvm_vm_release, 1628 .release = kvm_vm_release,
2265 .unlocked_ioctl = kvm_vm_ioctl, 1629 .unlocked_ioctl = kvm_vm_ioctl,
2266 .compat_ioctl = kvm_vm_ioctl, 1630#ifdef CONFIG_COMPAT
1631 .compat_ioctl = kvm_vm_compat_ioctl,
1632#endif
2267 .mmap = kvm_vm_mmap, 1633 .mmap = kvm_vm_mmap,
2268}; 1634};
2269 1635
@@ -2275,7 +1641,7 @@ static int kvm_dev_ioctl_create_vm(void)
2275 kvm = kvm_create_vm(); 1641 kvm = kvm_create_vm();
2276 if (IS_ERR(kvm)) 1642 if (IS_ERR(kvm))
2277 return PTR_ERR(kvm); 1643 return PTR_ERR(kvm);
2278 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0); 1644 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2279 if (fd < 0) 1645 if (fd < 0)
2280 kvm_put_kvm(kvm); 1646 kvm_put_kvm(kvm);
2281 1647
@@ -2288,6 +1654,10 @@ static long kvm_dev_ioctl_check_extension_generic(long arg)
2288 case KVM_CAP_USER_MEMORY: 1654 case KVM_CAP_USER_MEMORY:
2289 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: 1655 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2290 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS: 1656 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
1657#ifdef CONFIG_KVM_APIC_ARCHITECTURE
1658 case KVM_CAP_SET_BOOT_CPU_ID:
1659#endif
1660 case KVM_CAP_INTERNAL_ERROR_DATA:
2291 return 1; 1661 return 1;
2292#ifdef CONFIG_HAVE_KVM_IRQCHIP 1662#ifdef CONFIG_HAVE_KVM_IRQCHIP
2293 case KVM_CAP_IRQ_ROUTING: 1663 case KVM_CAP_IRQ_ROUTING:
@@ -2335,7 +1705,7 @@ static long kvm_dev_ioctl(struct file *filp,
2335 case KVM_TRACE_ENABLE: 1705 case KVM_TRACE_ENABLE:
2336 case KVM_TRACE_PAUSE: 1706 case KVM_TRACE_PAUSE:
2337 case KVM_TRACE_DISABLE: 1707 case KVM_TRACE_DISABLE:
2338 r = kvm_trace_ioctl(ioctl, arg); 1708 r = -EOPNOTSUPP;
2339 break; 1709 break;
2340 default: 1710 default:
2341 return kvm_arch_dev_ioctl(filp, ioctl, arg); 1711 return kvm_arch_dev_ioctl(filp, ioctl, arg);
@@ -2358,11 +1728,21 @@ static struct miscdevice kvm_dev = {
2358static void hardware_enable(void *junk) 1728static void hardware_enable(void *junk)
2359{ 1729{
2360 int cpu = raw_smp_processor_id(); 1730 int cpu = raw_smp_processor_id();
1731 int r;
2361 1732
2362 if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) 1733 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2363 return; 1734 return;
1735
2364 cpumask_set_cpu(cpu, cpus_hardware_enabled); 1736 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2365 kvm_arch_hardware_enable(NULL); 1737
1738 r = kvm_arch_hardware_enable(NULL);
1739
1740 if (r) {
1741 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1742 atomic_inc(&hardware_enable_failed);
1743 printk(KERN_INFO "kvm: enabling virtualization on "
1744 "CPU%d failed\n", cpu);
1745 }
2366} 1746}
2367 1747
2368static void hardware_disable(void *junk) 1748static void hardware_disable(void *junk)
@@ -2375,11 +1755,52 @@ static void hardware_disable(void *junk)
2375 kvm_arch_hardware_disable(NULL); 1755 kvm_arch_hardware_disable(NULL);
2376} 1756}
2377 1757
1758static void hardware_disable_all_nolock(void)
1759{
1760 BUG_ON(!kvm_usage_count);
1761
1762 kvm_usage_count--;
1763 if (!kvm_usage_count)
1764 on_each_cpu(hardware_disable, NULL, 1);
1765}
1766
1767static void hardware_disable_all(void)
1768{
1769 spin_lock(&kvm_lock);
1770 hardware_disable_all_nolock();
1771 spin_unlock(&kvm_lock);
1772}
1773
1774static int hardware_enable_all(void)
1775{
1776 int r = 0;
1777
1778 spin_lock(&kvm_lock);
1779
1780 kvm_usage_count++;
1781 if (kvm_usage_count == 1) {
1782 atomic_set(&hardware_enable_failed, 0);
1783 on_each_cpu(hardware_enable, NULL, 1);
1784
1785 if (atomic_read(&hardware_enable_failed)) {
1786 hardware_disable_all_nolock();
1787 r = -EBUSY;
1788 }
1789 }
1790
1791 spin_unlock(&kvm_lock);
1792
1793 return r;
1794}
1795
2378static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, 1796static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2379 void *v) 1797 void *v)
2380{ 1798{
2381 int cpu = (long)v; 1799 int cpu = (long)v;
2382 1800
1801 if (!kvm_usage_count)
1802 return NOTIFY_OK;
1803
2383 val &= ~CPU_TASKS_FROZEN; 1804 val &= ~CPU_TASKS_FROZEN;
2384 switch (val) { 1805 switch (val) {
2385 case CPU_DYING: 1806 case CPU_DYING:
@@ -2449,26 +1870,71 @@ void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2449 } 1870 }
2450} 1871}
2451 1872
2452struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus, 1873/* kvm_io_bus_write - called under kvm->slots_lock */
2453 gpa_t addr, int len, int is_write) 1874int kvm_io_bus_write(struct kvm_io_bus *bus, gpa_t addr,
1875 int len, const void *val)
2454{ 1876{
2455 int i; 1877 int i;
1878 for (i = 0; i < bus->dev_count; i++)
1879 if (!kvm_iodevice_write(bus->devs[i], addr, len, val))
1880 return 0;
1881 return -EOPNOTSUPP;
1882}
2456 1883
2457 for (i = 0; i < bus->dev_count; i++) { 1884/* kvm_io_bus_read - called under kvm->slots_lock */
2458 struct kvm_io_device *pos = bus->devs[i]; 1885int kvm_io_bus_read(struct kvm_io_bus *bus, gpa_t addr, int len, void *val)
1886{
1887 int i;
1888 for (i = 0; i < bus->dev_count; i++)
1889 if (!kvm_iodevice_read(bus->devs[i], addr, len, val))
1890 return 0;
1891 return -EOPNOTSUPP;
1892}
2459 1893
2460 if (pos->in_range(pos, addr, len, is_write)) 1894int kvm_io_bus_register_dev(struct kvm *kvm, struct kvm_io_bus *bus,
2461 return pos; 1895 struct kvm_io_device *dev)
2462 } 1896{
1897 int ret;
2463 1898
2464 return NULL; 1899 down_write(&kvm->slots_lock);
1900 ret = __kvm_io_bus_register_dev(bus, dev);
1901 up_write(&kvm->slots_lock);
1902
1903 return ret;
2465} 1904}
2466 1905
2467void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev) 1906/* An unlocked version. Caller must have write lock on slots_lock. */
1907int __kvm_io_bus_register_dev(struct kvm_io_bus *bus,
1908 struct kvm_io_device *dev)
2468{ 1909{
2469 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1)); 1910 if (bus->dev_count > NR_IOBUS_DEVS-1)
1911 return -ENOSPC;
2470 1912
2471 bus->devs[bus->dev_count++] = dev; 1913 bus->devs[bus->dev_count++] = dev;
1914
1915 return 0;
1916}
1917
1918void kvm_io_bus_unregister_dev(struct kvm *kvm,
1919 struct kvm_io_bus *bus,
1920 struct kvm_io_device *dev)
1921{
1922 down_write(&kvm->slots_lock);
1923 __kvm_io_bus_unregister_dev(bus, dev);
1924 up_write(&kvm->slots_lock);
1925}
1926
1927/* An unlocked version. Caller must have write lock on slots_lock. */
1928void __kvm_io_bus_unregister_dev(struct kvm_io_bus *bus,
1929 struct kvm_io_device *dev)
1930{
1931 int i;
1932
1933 for (i = 0; i < bus->dev_count; i++)
1934 if (bus->devs[i] == dev) {
1935 bus->devs[i] = bus->devs[--bus->dev_count];
1936 break;
1937 }
2472} 1938}
2473 1939
2474static struct notifier_block kvm_cpu_notifier = { 1940static struct notifier_block kvm_cpu_notifier = {
@@ -2501,18 +1967,16 @@ static int vcpu_stat_get(void *_offset, u64 *val)
2501 *val = 0; 1967 *val = 0;
2502 spin_lock(&kvm_lock); 1968 spin_lock(&kvm_lock);
2503 list_for_each_entry(kvm, &vm_list, vm_list) 1969 list_for_each_entry(kvm, &vm_list, vm_list)
2504 for (i = 0; i < KVM_MAX_VCPUS; ++i) { 1970 kvm_for_each_vcpu(i, vcpu, kvm)
2505 vcpu = kvm->vcpus[i]; 1971 *val += *(u32 *)((void *)vcpu + offset);
2506 if (vcpu) 1972
2507 *val += *(u32 *)((void *)vcpu + offset);
2508 }
2509 spin_unlock(&kvm_lock); 1973 spin_unlock(&kvm_lock);
2510 return 0; 1974 return 0;
2511} 1975}
2512 1976
2513DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n"); 1977DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2514 1978
2515static struct file_operations *stat_fops[] = { 1979static const struct file_operations *stat_fops[] = {
2516 [KVM_STAT_VCPU] = &vcpu_stat_fops, 1980 [KVM_STAT_VCPU] = &vcpu_stat_fops,
2517 [KVM_STAT_VM] = &vm_stat_fops, 1981 [KVM_STAT_VM] = &vm_stat_fops,
2518}; 1982};
@@ -2539,13 +2003,15 @@ static void kvm_exit_debug(void)
2539 2003
2540static int kvm_suspend(struct sys_device *dev, pm_message_t state) 2004static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2541{ 2005{
2542 hardware_disable(NULL); 2006 if (kvm_usage_count)
2007 hardware_disable(NULL);
2543 return 0; 2008 return 0;
2544} 2009}
2545 2010
2546static int kvm_resume(struct sys_device *dev) 2011static int kvm_resume(struct sys_device *dev)
2547{ 2012{
2548 hardware_enable(NULL); 2013 if (kvm_usage_count)
2014 hardware_enable(NULL);
2549 return 0; 2015 return 0;
2550} 2016}
2551 2017
@@ -2590,8 +2056,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2590 int r; 2056 int r;
2591 int cpu; 2057 int cpu;
2592 2058
2593 kvm_init_debug();
2594
2595 r = kvm_arch_init(opaque); 2059 r = kvm_arch_init(opaque);
2596 if (r) 2060 if (r)
2597 goto out_fail; 2061 goto out_fail;
@@ -2622,7 +2086,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2622 goto out_free_1; 2086 goto out_free_1;
2623 } 2087 }
2624 2088
2625 on_each_cpu(hardware_enable, NULL, 1);
2626 r = register_cpu_notifier(&kvm_cpu_notifier); 2089 r = register_cpu_notifier(&kvm_cpu_notifier);
2627 if (r) 2090 if (r)
2628 goto out_free_2; 2091 goto out_free_2;
@@ -2658,6 +2121,8 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2658 kvm_preempt_ops.sched_in = kvm_sched_in; 2121 kvm_preempt_ops.sched_in = kvm_sched_in;
2659 kvm_preempt_ops.sched_out = kvm_sched_out; 2122 kvm_preempt_ops.sched_out = kvm_sched_out;
2660 2123
2124 kvm_init_debug();
2125
2661 return 0; 2126 return 0;
2662 2127
2663out_free: 2128out_free:
@@ -2670,7 +2135,6 @@ out_free_3:
2670 unregister_reboot_notifier(&kvm_reboot_notifier); 2135 unregister_reboot_notifier(&kvm_reboot_notifier);
2671 unregister_cpu_notifier(&kvm_cpu_notifier); 2136 unregister_cpu_notifier(&kvm_cpu_notifier);
2672out_free_2: 2137out_free_2:
2673 on_each_cpu(hardware_disable, NULL, 1);
2674out_free_1: 2138out_free_1:
2675 kvm_arch_hardware_unsetup(); 2139 kvm_arch_hardware_unsetup();
2676out_free_0a: 2140out_free_0a:
@@ -2679,7 +2143,6 @@ out_free_0:
2679 __free_page(bad_page); 2143 __free_page(bad_page);
2680out: 2144out:
2681 kvm_arch_exit(); 2145 kvm_arch_exit();
2682 kvm_exit_debug();
2683out_fail: 2146out_fail:
2684 return r; 2147 return r;
2685} 2148}
@@ -2687,7 +2150,8 @@ EXPORT_SYMBOL_GPL(kvm_init);
2687 2150
2688void kvm_exit(void) 2151void kvm_exit(void)
2689{ 2152{
2690 kvm_trace_cleanup(); 2153 tracepoint_synchronize_unregister();
2154 kvm_exit_debug();
2691 misc_deregister(&kvm_dev); 2155 misc_deregister(&kvm_dev);
2692 kmem_cache_destroy(kvm_vcpu_cache); 2156 kmem_cache_destroy(kvm_vcpu_cache);
2693 sysdev_unregister(&kvm_sysdev); 2157 sysdev_unregister(&kvm_sysdev);
@@ -2697,7 +2161,6 @@ void kvm_exit(void)
2697 on_each_cpu(hardware_disable, NULL, 1); 2161 on_each_cpu(hardware_disable, NULL, 1);
2698 kvm_arch_hardware_unsetup(); 2162 kvm_arch_hardware_unsetup();
2699 kvm_arch_exit(); 2163 kvm_arch_exit();
2700 kvm_exit_debug();
2701 free_cpumask_var(cpus_hardware_enabled); 2164 free_cpumask_var(cpus_hardware_enabled);
2702 __free_page(bad_page); 2165 __free_page(bad_page);
2703} 2166}
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-29 12:44:15 -0500