?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: Enable Function Level Reset for assigned device
[linux-2.6/btrfs-unstable.git] / virt / kvm / kvm_main.c
blob1838052f3c9e15292dc51db567b2a5a98419f41f
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "iodev.h"
19
20 #include <linux/kvm_host.h>
21 #include <linux/kvm.h>
22 #include <linux/module.h>
23 #include <linux/errno.h>
24 #include <linux/percpu.h>
25 #include <linux/gfp.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/sysdev.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44
45 #include <asm/processor.h>
46 #include <asm/io.h>
47 #include <asm/uaccess.h>
48 #include <asm/pgtable.h>
49
50 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
51 #include "coalesced_mmio.h"
52 #endif
53
54 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
55 #include <linux/pci.h>
56 #include <linux/interrupt.h>
57 #include "irq.h"
58 #endif
59
60 MODULE_AUTHOR("Qumranet");
61 MODULE_LICENSE("GPL");
62
63 DEFINE_SPINLOCK(kvm_lock);
64 LIST_HEAD(vm_list);
65
66 static cpumask_t cpus_hardware_enabled;
67
68 struct kmem_cache *kvm_vcpu_cache;
69 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
70
71 static __read_mostly struct preempt_ops kvm_preempt_ops;
72
73 struct dentry *kvm_debugfs_dir;
74
75 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
76 unsigned long arg);
77
78 bool kvm_rebooting;
79
80 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
81 static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
82 int assigned_dev_id)
83 {
84 struct list_head *ptr;
85 struct kvm_assigned_dev_kernel *match;
86
87 list_for_each(ptr, head) {
88 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
89 if (match->assigned_dev_id == assigned_dev_id)
90 return match;
91 }
92 return NULL;
93 }
94
95 static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
96 {
97 struct kvm_assigned_dev_kernel *assigned_dev;
98
99 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
100 interrupt_work);
101
102 /* This is taken to safely inject irq inside the guest. When
103 * the interrupt injection (or the ioapic code) uses a
104 * finer-grained lock, update this
105 */
106 mutex_lock(&assigned_dev->kvm->lock);
107 kvm_set_irq(assigned_dev->kvm,
108 assigned_dev->irq_source_id,
109 assigned_dev->guest_irq, 1);
110 mutex_unlock(&assigned_dev->kvm->lock);
111 kvm_put_kvm(assigned_dev->kvm);
112 }
113
114 static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
115 {
116 struct kvm_assigned_dev_kernel *assigned_dev =
117 (struct kvm_assigned_dev_kernel *) dev_id;
118
119 kvm_get_kvm(assigned_dev->kvm);
120 schedule_work(&assigned_dev->interrupt_work);
121 disable_irq_nosync(irq);
122 return IRQ_HANDLED;
123 }
124
125 /* Ack the irq line for an assigned device */
126 static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
127 {
128 struct kvm_assigned_dev_kernel *dev;
129
130 if (kian->gsi == -1)
131 return;
132
133 dev = container_of(kian, struct kvm_assigned_dev_kernel,
134 ack_notifier);
135 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
136 enable_irq(dev->host_irq);
137 }
138
139 static void kvm_free_assigned_device(struct kvm *kvm,
140 struct kvm_assigned_dev_kernel
141 *assigned_dev)
142 {
143 if (irqchip_in_kernel(kvm) && assigned_dev->irq_requested)
144 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
145
146 kvm_unregister_irq_ack_notifier(&assigned_dev->ack_notifier);
147 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
148
149 if (cancel_work_sync(&assigned_dev->interrupt_work))
150 /* We had pending work. That means we will have to take
151 * care of kvm_put_kvm.
152 */
153 kvm_put_kvm(kvm);
154
155 pci_reset_function(assigned_dev->dev);
156
157 pci_release_regions(assigned_dev->dev);
158 pci_disable_device(assigned_dev->dev);
159 pci_dev_put(assigned_dev->dev);
160
161 list_del(&assigned_dev->list);
162 kfree(assigned_dev);
163 }
164
165 void kvm_free_all_assigned_devices(struct kvm *kvm)
166 {
167 struct list_head *ptr, *ptr2;
168 struct kvm_assigned_dev_kernel *assigned_dev;
169
170 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
171 assigned_dev = list_entry(ptr,
172 struct kvm_assigned_dev_kernel,
173 list);
174
175 kvm_free_assigned_device(kvm, assigned_dev);
176 }
177 }
178
179 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
180 struct kvm_assigned_irq
181 *assigned_irq)
182 {
183 int r = 0;
184 struct kvm_assigned_dev_kernel *match;
185
186 mutex_lock(&kvm->lock);
187
188 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
189 assigned_irq->assigned_dev_id);
190 if (!match) {
191 mutex_unlock(&kvm->lock);
192 return -EINVAL;
193 }
194
195 if (match->irq_requested) {
196 match->guest_irq = assigned_irq->guest_irq;
197 match->ack_notifier.gsi = assigned_irq->guest_irq;
198 mutex_unlock(&kvm->lock);
199 return 0;
200 }
201
202 INIT_WORK(&match->interrupt_work,
203 kvm_assigned_dev_interrupt_work_handler);
204
205 if (irqchip_in_kernel(kvm)) {
206 if (!capable(CAP_SYS_RAWIO)) {
207 r = -EPERM;
208 goto out_release;
209 }
210
211 if (assigned_irq->host_irq)
212 match->host_irq = assigned_irq->host_irq;
213 else
214 match->host_irq = match->dev->irq;
215 match->guest_irq = assigned_irq->guest_irq;
216 match->ack_notifier.gsi = assigned_irq->guest_irq;
217 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
218 kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
219 r = kvm_request_irq_source_id(kvm);
220 if (r < 0)
221 goto out_release;
222 else
223 match->irq_source_id = r;
224
225 /* Even though this is PCI, we don't want to use shared
226 * interrupts. Sharing host devices with guest-assigned devices
227 * on the same interrupt line is not a happy situation: there
228 * are going to be long delays in accepting, acking, etc.
229 */
230 if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
231 "kvm_assigned_device", (void *)match)) {
232 r = -EIO;
233 goto out_release;
234 }
235 }
236
237 match->irq_requested = true;
238 mutex_unlock(&kvm->lock);
239 return r;
240 out_release:
241 mutex_unlock(&kvm->lock);
242 kvm_free_assigned_device(kvm, match);
243 return r;
244 }
245
246 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
247 struct kvm_assigned_pci_dev *assigned_dev)
248 {
249 int r = 0;
250 struct kvm_assigned_dev_kernel *match;
251 struct pci_dev *dev;
252
253 mutex_lock(&kvm->lock);
254
255 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
256 assigned_dev->assigned_dev_id);
257 if (match) {
258 /* device already assigned */
259 r = -EINVAL;
260 goto out;
261 }
262
263 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
264 if (match == NULL) {
265 printk(KERN_INFO "%s: Couldn't allocate memory\n",
266 __func__);
267 r = -ENOMEM;
268 goto out;
269 }
270 dev = pci_get_bus_and_slot(assigned_dev->busnr,
271 assigned_dev->devfn);
272 if (!dev) {
273 printk(KERN_INFO "%s: host device not found\n", __func__);
274 r = -EINVAL;
275 goto out_free;
276 }
277 if (pci_enable_device(dev)) {
278 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
279 r = -EBUSY;
280 goto out_put;
281 }
282 r = pci_request_regions(dev, "kvm_assigned_device");
283 if (r) {
284 printk(KERN_INFO "%s: Could not get access to device regions\n",
285 __func__);
286 goto out_disable;
287 }
288
289 pci_reset_function(dev);
290
291 match->assigned_dev_id = assigned_dev->assigned_dev_id;
292 match->host_busnr = assigned_dev->busnr;
293 match->host_devfn = assigned_dev->devfn;
294 match->dev = dev;
295
296 match->kvm = kvm;
297
298 list_add(&match->list, &kvm->arch.assigned_dev_head);
299
300 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
301 r = kvm_iommu_map_guest(kvm, match);
302 if (r)
303 goto out_list_del;
304 }
305
306 out:
307 mutex_unlock(&kvm->lock);
308 return r;
309 out_list_del:
310 list_del(&match->list);
311 pci_release_regions(dev);
312 out_disable:
313 pci_disable_device(dev);
314 out_put:
315 pci_dev_put(dev);
316 out_free:
317 kfree(match);
318 mutex_unlock(&kvm->lock);
319 return r;
320 }
321 #endif
322
323 static inline int valid_vcpu(int n)
324 {
325 return likely(n >= 0 && n < KVM_MAX_VCPUS);
326 }
327
328 inline int kvm_is_mmio_pfn(pfn_t pfn)
329 {
330 if (pfn_valid(pfn))
331 return PageReserved(pfn_to_page(pfn));
332
333 return true;
334 }
335
336 /*
337 * Switches to specified vcpu, until a matching vcpu_put()
338 */
339 void vcpu_load(struct kvm_vcpu *vcpu)
340 {
341 int cpu;
342
343 mutex_lock(&vcpu->mutex);
344 cpu = get_cpu();
345 preempt_notifier_register(&vcpu->preempt_notifier);
346 kvm_arch_vcpu_load(vcpu, cpu);
347 put_cpu();
348 }
349
350 void vcpu_put(struct kvm_vcpu *vcpu)
351 {
352 preempt_disable();
353 kvm_arch_vcpu_put(vcpu);
354 preempt_notifier_unregister(&vcpu->preempt_notifier);
355 preempt_enable();
356 mutex_unlock(&vcpu->mutex);
357 }
358
359 static void ack_flush(void *_completed)
360 {
361 }
362
363 void kvm_flush_remote_tlbs(struct kvm *kvm)
364 {
365 int i, cpu, me;
366 cpumask_t cpus;
367 struct kvm_vcpu *vcpu;
368
369 me = get_cpu();
370 cpus_clear(cpus);
371 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
372 vcpu = kvm->vcpus[i];
373 if (!vcpu)
374 continue;
375 if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
376 continue;
377 cpu = vcpu->cpu;
378 if (cpu != -1 && cpu != me)
379 cpu_set(cpu, cpus);
380 }
381 if (cpus_empty(cpus))
382 goto out;
383 ++kvm->stat.remote_tlb_flush;
384 smp_call_function_mask(cpus, ack_flush, NULL, 1);
385 out:
386 put_cpu();
387 }
388
389 void kvm_reload_remote_mmus(struct kvm *kvm)
390 {
391 int i, cpu, me;
392 cpumask_t cpus;
393 struct kvm_vcpu *vcpu;
394
395 me = get_cpu();
396 cpus_clear(cpus);
397 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
398 vcpu = kvm->vcpus[i];
399 if (!vcpu)
400 continue;
401 if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
402 continue;
403 cpu = vcpu->cpu;
404 if (cpu != -1 && cpu != me)
405 cpu_set(cpu, cpus);
406 }
407 if (cpus_empty(cpus))
408 goto out;
409 smp_call_function_mask(cpus, ack_flush, NULL, 1);
410 out:
411 put_cpu();
412 }
413
414
415 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
416 {
417 struct page *page;
418 int r;
419
420 mutex_init(&vcpu->mutex);
421 vcpu->cpu = -1;
422 vcpu->kvm = kvm;
423 vcpu->vcpu_id = id;
424 init_waitqueue_head(&vcpu->wq);
425
426 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
427 if (!page) {
428 r = -ENOMEM;
429 goto fail;
430 }
431 vcpu->run = page_address(page);
432
433 r = kvm_arch_vcpu_init(vcpu);
434 if (r < 0)
435 goto fail_free_run;
436 return 0;
437
438 fail_free_run:
439 free_page((unsigned long)vcpu->run);
440 fail:
441 return r;
442 }
443 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
444
445 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
446 {
447 kvm_arch_vcpu_uninit(vcpu);
448 free_page((unsigned long)vcpu->run);
449 }
450 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
451
452 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
453 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
454 {
455 return container_of(mn, struct kvm, mmu_notifier);
456 }
457
458 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
459 struct mm_struct *mm,
460 unsigned long address)
461 {
462 struct kvm *kvm = mmu_notifier_to_kvm(mn);
463 int need_tlb_flush;
464
465 /*
466 * When ->invalidate_page runs, the linux pte has been zapped
467 * already but the page is still allocated until
468 * ->invalidate_page returns. So if we increase the sequence
469 * here the kvm page fault will notice if the spte can't be
470 * established because the page is going to be freed. If
471 * instead the kvm page fault establishes the spte before
472 * ->invalidate_page runs, kvm_unmap_hva will release it
473 * before returning.
474 *
475 * The sequence increase only need to be seen at spin_unlock
476 * time, and not at spin_lock time.
477 *
478 * Increasing the sequence after the spin_unlock would be
479 * unsafe because the kvm page fault could then establish the
480 * pte after kvm_unmap_hva returned, without noticing the page
481 * is going to be freed.
482 */
483 spin_lock(&kvm->mmu_lock);
484 kvm->mmu_notifier_seq++;
485 need_tlb_flush = kvm_unmap_hva(kvm, address);
486 spin_unlock(&kvm->mmu_lock);
487
488 /* we've to flush the tlb before the pages can be freed */
489 if (need_tlb_flush)
490 kvm_flush_remote_tlbs(kvm);
491
492 }
493
494 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
495 struct mm_struct *mm,
496 unsigned long start,
497 unsigned long end)
498 {
499 struct kvm *kvm = mmu_notifier_to_kvm(mn);
500 int need_tlb_flush = 0;
501
502 spin_lock(&kvm->mmu_lock);
503 /*
504 * The count increase must become visible at unlock time as no
505 * spte can be established without taking the mmu_lock and
506 * count is also read inside the mmu_lock critical section.
507 */
508 kvm->mmu_notifier_count++;
509 for (; start < end; start += PAGE_SIZE)
510 need_tlb_flush |= kvm_unmap_hva(kvm, start);
511 spin_unlock(&kvm->mmu_lock);
512
513 /* we've to flush the tlb before the pages can be freed */
514 if (need_tlb_flush)
515 kvm_flush_remote_tlbs(kvm);
516 }
517
518 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
519 struct mm_struct *mm,
520 unsigned long start,
521 unsigned long end)
522 {
523 struct kvm *kvm = mmu_notifier_to_kvm(mn);
524
525 spin_lock(&kvm->mmu_lock);
526 /*
527 * This sequence increase will notify the kvm page fault that
528 * the page that is going to be mapped in the spte could have
529 * been freed.
530 */
531 kvm->mmu_notifier_seq++;
532 /*
533 * The above sequence increase must be visible before the
534 * below count decrease but both values are read by the kvm
535 * page fault under mmu_lock spinlock so we don't need to add
536 * a smb_wmb() here in between the two.
537 */
538 kvm->mmu_notifier_count--;
539 spin_unlock(&kvm->mmu_lock);
540
541 BUG_ON(kvm->mmu_notifier_count < 0);
542 }
543
544 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
545 struct mm_struct *mm,
546 unsigned long address)
547 {
548 struct kvm *kvm = mmu_notifier_to_kvm(mn);
549 int young;
550
551 spin_lock(&kvm->mmu_lock);
552 young = kvm_age_hva(kvm, address);
553 spin_unlock(&kvm->mmu_lock);
554
555 if (young)
556 kvm_flush_remote_tlbs(kvm);
557
558 return young;
559 }
560
561 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
562 .invalidate_page = kvm_mmu_notifier_invalidate_page,
563 .invalidate_range_start = kvm_mmu_notifier_invalidate_range_start,
564 .invalidate_range_end = kvm_mmu_notifier_invalidate_range_end,
565 .clear_flush_young = kvm_mmu_notifier_clear_flush_young,
566 };
567 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
568
569 static struct kvm *kvm_create_vm(void)
570 {
571 struct kvm *kvm = kvm_arch_create_vm();
572 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
573 struct page *page;
574 #endif
575
576 if (IS_ERR(kvm))
577 goto out;
578
579 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
580 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
581 if (!page) {
582 kfree(kvm);
583 return ERR_PTR(-ENOMEM);
584 }
585 kvm->coalesced_mmio_ring =
586 (struct kvm_coalesced_mmio_ring *)page_address(page);
587 #endif
588
589 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
590 {
591 int err;
592 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
593 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
594 if (err) {
595 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
596 put_page(page);
597 #endif
598 kfree(kvm);
599 return ERR_PTR(err);
600 }
601 }
602 #endif
603
604 kvm->mm = current->mm;
605 atomic_inc(&kvm->mm->mm_count);
606 spin_lock_init(&kvm->mmu_lock);
607 kvm_io_bus_init(&kvm->pio_bus);
608 mutex_init(&kvm->lock);
609 kvm_io_bus_init(&kvm->mmio_bus);
610 init_rwsem(&kvm->slots_lock);
611 atomic_set(&kvm->users_count, 1);
612 spin_lock(&kvm_lock);
613 list_add(&kvm->vm_list, &vm_list);
614 spin_unlock(&kvm_lock);
615 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
616 kvm_coalesced_mmio_init(kvm);
617 #endif
618 out:
619 return kvm;
620 }
621
622 /*
623 * Free any memory in @free but not in @dont.
624 */
625 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
626 struct kvm_memory_slot *dont)
627 {
628 if (!dont || free->rmap != dont->rmap)
629 vfree(free->rmap);
630
631 if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
632 vfree(free->dirty_bitmap);
633
634 if (!dont || free->lpage_info != dont->lpage_info)
635 vfree(free->lpage_info);
636
637 free->npages = 0;
638 free->dirty_bitmap = NULL;
639 free->rmap = NULL;
640 free->lpage_info = NULL;
641 }
642
643 void kvm_free_physmem(struct kvm *kvm)
644 {
645 int i;
646
647 for (i = 0; i < kvm->nmemslots; ++i)
648 kvm_free_physmem_slot(&kvm->memslots[i], NULL);
649 }
650
651 static void kvm_destroy_vm(struct kvm *kvm)
652 {
653 struct mm_struct *mm = kvm->mm;
654
655 spin_lock(&kvm_lock);
656 list_del(&kvm->vm_list);
657 spin_unlock(&kvm_lock);
658 kvm_io_bus_destroy(&kvm->pio_bus);
659 kvm_io_bus_destroy(&kvm->mmio_bus);
660 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
661 if (kvm->coalesced_mmio_ring != NULL)
662 free_page((unsigned long)kvm->coalesced_mmio_ring);
663 #endif
664 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
665 mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
666 #endif
667 kvm_arch_destroy_vm(kvm);
668 mmdrop(mm);
669 }
670
671 void kvm_get_kvm(struct kvm *kvm)
672 {
673 atomic_inc(&kvm->users_count);
674 }
675 EXPORT_SYMBOL_GPL(kvm_get_kvm);
676
677 void kvm_put_kvm(struct kvm *kvm)
678 {
679 if (atomic_dec_and_test(&kvm->users_count))
680 kvm_destroy_vm(kvm);
681 }
682 EXPORT_SYMBOL_GPL(kvm_put_kvm);
683
684
685 static int kvm_vm_release(struct inode *inode, struct file *filp)
686 {
687 struct kvm *kvm = filp->private_data;
688
689 kvm_put_kvm(kvm);
690 return 0;
691 }
692
693 /*
694 * Allocate some memory and give it an address in the guest physical address
695 * space.
696 *
697 * Discontiguous memory is allowed, mostly for framebuffers.
698 *
699 * Must be called holding mmap_sem for write.
700 */
701 int __kvm_set_memory_region(struct kvm *kvm,
702 struct kvm_userspace_memory_region *mem,
703 int user_alloc)
704 {
705 int r;
706 gfn_t base_gfn;
707 unsigned long npages;
708 unsigned long i;
709 struct kvm_memory_slot *memslot;
710 struct kvm_memory_slot old, new;
711
712 r = -EINVAL;
713 /* General sanity checks */
714 if (mem->memory_size & (PAGE_SIZE - 1))
715 goto out;
716 if (mem->guest_phys_addr & (PAGE_SIZE - 1))
717 goto out;
718 if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
719 goto out;
720 if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
721 goto out;
722
723 memslot = &kvm->memslots[mem->slot];
724 base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
725 npages = mem->memory_size >> PAGE_SHIFT;
726
727 if (!npages)
728 mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
729
730 new = old = *memslot;
731
732 new.base_gfn = base_gfn;
733 new.npages = npages;
734 new.flags = mem->flags;
735
736 /* Disallow changing a memory slot's size. */
737 r = -EINVAL;
738 if (npages && old.npages && npages != old.npages)
739 goto out_free;
740
741 /* Check for overlaps */
742 r = -EEXIST;
743 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
744 struct kvm_memory_slot *s = &kvm->memslots[i];
745
746 if (s == memslot)
747 continue;
748 if (!((base_gfn + npages <= s->base_gfn) ||
749 (base_gfn >= s->base_gfn + s->npages)))
750 goto out_free;
751 }
752
753 /* Free page dirty bitmap if unneeded */
754 if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
755 new.dirty_bitmap = NULL;
756
757 r = -ENOMEM;
758
759 /* Allocate if a slot is being created */
760 #ifndef CONFIG_S390
761 if (npages && !new.rmap) {
762 new.rmap = vmalloc(npages * sizeof(struct page *));
763
764 if (!new.rmap)
765 goto out_free;
766
767 memset(new.rmap, 0, npages * sizeof(*new.rmap));
768
769 new.user_alloc = user_alloc;
770 /*
771 * hva_to_rmmap() serialzies with the mmu_lock and to be
772 * safe it has to ignore memslots with !user_alloc &&
773 * !userspace_addr.
774 */
775 if (user_alloc)
776 new.userspace_addr = mem->userspace_addr;
777 else
778 new.userspace_addr = 0;
779 }
780 if (npages && !new.lpage_info) {
781 int largepages = npages / KVM_PAGES_PER_HPAGE;
782 if (npages % KVM_PAGES_PER_HPAGE)
783 largepages++;
784 if (base_gfn % KVM_PAGES_PER_HPAGE)
785 largepages++;
786
787 new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
788
789 if (!new.lpage_info)
790 goto out_free;
791
792 memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
793
794 if (base_gfn % KVM_PAGES_PER_HPAGE)
795 new.lpage_info[0].write_count = 1;
796 if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
797 new.lpage_info[largepages-1].write_count = 1;
798 }
799
800 /* Allocate page dirty bitmap if needed */
801 if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
802 unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
803
804 new.dirty_bitmap = vmalloc(dirty_bytes);
805 if (!new.dirty_bitmap)
806 goto out_free;
807 memset(new.dirty_bitmap, 0, dirty_bytes);
808 }
809 #endif /* not defined CONFIG_S390 */
810
811 if (!npages)
812 kvm_arch_flush_shadow(kvm);
813
814 spin_lock(&kvm->mmu_lock);
815 if (mem->slot >= kvm->nmemslots)
816 kvm->nmemslots = mem->slot + 1;
817
818 *memslot = new;
819 spin_unlock(&kvm->mmu_lock);
820
821 r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
822 if (r) {
823 spin_lock(&kvm->mmu_lock);
824 *memslot = old;
825 spin_unlock(&kvm->mmu_lock);
826 goto out_free;
827 }
828
829 kvm_free_physmem_slot(&old, &new);
830 #ifdef CONFIG_DMAR
831 /* map the pages in iommu page table */
832 r = kvm_iommu_map_pages(kvm, base_gfn, npages);
833 if (r)
834 goto out;
835 #endif
836 return 0;
837
838 out_free:
839 kvm_free_physmem_slot(&new, &old);
840 out:
841 return r;
842
843 }
844 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
845
846 int kvm_set_memory_region(struct kvm *kvm,
847 struct kvm_userspace_memory_region *mem,
848 int user_alloc)
849 {
850 int r;
851
852 down_write(&kvm->slots_lock);
853 r = __kvm_set_memory_region(kvm, mem, user_alloc);
854 up_write(&kvm->slots_lock);
855 return r;
856 }
857 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
858
859 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
860 struct
861 kvm_userspace_memory_region *mem,
862 int user_alloc)
863 {
864 if (mem->slot >= KVM_MEMORY_SLOTS)
865 return -EINVAL;
866 return kvm_set_memory_region(kvm, mem, user_alloc);
867 }
868
869 int kvm_get_dirty_log(struct kvm *kvm,
870 struct kvm_dirty_log *log, int *is_dirty)
871 {
872 struct kvm_memory_slot *memslot;
873 int r, i;
874 int n;
875 unsigned long any = 0;
876
877 r = -EINVAL;
878 if (log->slot >= KVM_MEMORY_SLOTS)
879 goto out;
880
881 memslot = &kvm->memslots[log->slot];
882 r = -ENOENT;
883 if (!memslot->dirty_bitmap)
884 goto out;
885
886 n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
887
888 for (i = 0; !any && i < n/sizeof(long); ++i)
889 any = memslot->dirty_bitmap[i];
890
891 r = -EFAULT;
892 if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
893 goto out;
894
895 if (any)
896 *is_dirty = 1;
897
898 r = 0;
899 out:
900 return r;
901 }
902
903 int is_error_page(struct page *page)
904 {
905 return page == bad_page;
906 }
907 EXPORT_SYMBOL_GPL(is_error_page);
908
909 int is_error_pfn(pfn_t pfn)
910 {
911 return pfn == bad_pfn;
912 }
913 EXPORT_SYMBOL_GPL(is_error_pfn);
914
915 static inline unsigned long bad_hva(void)
916 {
917 return PAGE_OFFSET;
918 }
919
920 int kvm_is_error_hva(unsigned long addr)
921 {
922 return addr == bad_hva();
923 }
924 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
925
926 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
927 {
928 int i;
929
930 for (i = 0; i < kvm->nmemslots; ++i) {
931 struct kvm_memory_slot *memslot = &kvm->memslots[i];
932
933 if (gfn >= memslot->base_gfn
934 && gfn < memslot->base_gfn + memslot->npages)
935 return memslot;
936 }
937 return NULL;
938 }
939
940 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
941 {
942 gfn = unalias_gfn(kvm, gfn);
943 return __gfn_to_memslot(kvm, gfn);
944 }
945
946 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
947 {
948 int i;
949
950 gfn = unalias_gfn(kvm, gfn);
951 for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
952 struct kvm_memory_slot *memslot = &kvm->memslots[i];
953
954 if (gfn >= memslot->base_gfn
955 && gfn < memslot->base_gfn + memslot->npages)
956 return 1;
957 }
958 return 0;
959 }
960 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
961
962 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
963 {
964 struct kvm_memory_slot *slot;
965
966 gfn = unalias_gfn(kvm, gfn);
967 slot = __gfn_to_memslot(kvm, gfn);
968 if (!slot)
969 return bad_hva();
970 return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
971 }
972 EXPORT_SYMBOL_GPL(gfn_to_hva);
973
974 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
975 {
976 struct page *page[1];
977 unsigned long addr;
978 int npages;
979 pfn_t pfn;
980
981 might_sleep();
982
983 addr = gfn_to_hva(kvm, gfn);
984 if (kvm_is_error_hva(addr)) {
985 get_page(bad_page);
986 return page_to_pfn(bad_page);
987 }
988
989 npages = get_user_pages_fast(addr, 1, 1, page);
990
991 if (unlikely(npages != 1)) {
992 struct vm_area_struct *vma;
993
994 down_read(&current->mm->mmap_sem);
995 vma = find_vma(current->mm, addr);
996
997 if (vma == NULL || addr < vma->vm_start ||
998 !(vma->vm_flags & VM_PFNMAP)) {
999 up_read(&current->mm->mmap_sem);
1000 get_page(bad_page);
1001 return page_to_pfn(bad_page);
1002 }
1003
1004 pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1005 up_read(&current->mm->mmap_sem);
1006 BUG_ON(!kvm_is_mmio_pfn(pfn));
1007 } else
1008 pfn = page_to_pfn(page[0]);
1009
1010 return pfn;
1011 }
1012
1013 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1014
1015 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1016 {
1017 pfn_t pfn;
1018
1019 pfn = gfn_to_pfn(kvm, gfn);
1020 if (!kvm_is_mmio_pfn(pfn))
1021 return pfn_to_page(pfn);
1022
1023 WARN_ON(kvm_is_mmio_pfn(pfn));
1024
1025 get_page(bad_page);
1026 return bad_page;
1027 }
1028
1029 EXPORT_SYMBOL_GPL(gfn_to_page);
1030
1031 void kvm_release_page_clean(struct page *page)
1032 {
1033 kvm_release_pfn_clean(page_to_pfn(page));
1034 }
1035 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1036
1037 void kvm_release_pfn_clean(pfn_t pfn)
1038 {
1039 if (!kvm_is_mmio_pfn(pfn))
1040 put_page(pfn_to_page(pfn));
1041 }
1042 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1043
1044 void kvm_release_page_dirty(struct page *page)
1045 {
1046 kvm_release_pfn_dirty(page_to_pfn(page));
1047 }
1048 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1049
1050 void kvm_release_pfn_dirty(pfn_t pfn)
1051 {
1052 kvm_set_pfn_dirty(pfn);
1053 kvm_release_pfn_clean(pfn);
1054 }
1055 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1056
1057 void kvm_set_page_dirty(struct page *page)
1058 {
1059 kvm_set_pfn_dirty(page_to_pfn(page));
1060 }
1061 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1062
1063 void kvm_set_pfn_dirty(pfn_t pfn)
1064 {
1065 if (!kvm_is_mmio_pfn(pfn)) {
1066 struct page *page = pfn_to_page(pfn);
1067 if (!PageReserved(page))
1068 SetPageDirty(page);
1069 }
1070 }
1071 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1072
1073 void kvm_set_pfn_accessed(pfn_t pfn)
1074 {
1075 if (!kvm_is_mmio_pfn(pfn))
1076 mark_page_accessed(pfn_to_page(pfn));
1077 }
1078 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1079
1080 void kvm_get_pfn(pfn_t pfn)
1081 {
1082 if (!kvm_is_mmio_pfn(pfn))
1083 get_page(pfn_to_page(pfn));
1084 }
1085 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1086
1087 static int next_segment(unsigned long len, int offset)
1088 {
1089 if (len > PAGE_SIZE - offset)
1090 return PAGE_SIZE - offset;
1091 else
1092 return len;
1093 }
1094
1095 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1096 int len)
1097 {
1098 int r;
1099 unsigned long addr;
1100
1101 addr = gfn_to_hva(kvm, gfn);
1102 if (kvm_is_error_hva(addr))
1103 return -EFAULT;
1104 r = copy_from_user(data, (void __user *)addr + offset, len);
1105 if (r)
1106 return -EFAULT;
1107 return 0;
1108 }
1109 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1110
1111 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1112 {
1113 gfn_t gfn = gpa >> PAGE_SHIFT;
1114 int seg;
1115 int offset = offset_in_page(gpa);
1116 int ret;
1117
1118 while ((seg = next_segment(len, offset)) != 0) {
1119 ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1120 if (ret < 0)
1121 return ret;
1122 offset = 0;
1123 len -= seg;
1124 data += seg;
1125 ++gfn;
1126 }
1127 return 0;
1128 }
1129 EXPORT_SYMBOL_GPL(kvm_read_guest);
1130
1131 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1132 unsigned long len)
1133 {
1134 int r;
1135 unsigned long addr;
1136 gfn_t gfn = gpa >> PAGE_SHIFT;
1137 int offset = offset_in_page(gpa);
1138
1139 addr = gfn_to_hva(kvm, gfn);
1140 if (kvm_is_error_hva(addr))
1141 return -EFAULT;
1142 pagefault_disable();
1143 r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1144 pagefault_enable();
1145 if (r)
1146 return -EFAULT;
1147 return 0;
1148 }
1149 EXPORT_SYMBOL(kvm_read_guest_atomic);
1150
1151 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1152 int offset, int len)
1153 {
1154 int r;
1155 unsigned long addr;
1156
1157 addr = gfn_to_hva(kvm, gfn);
1158 if (kvm_is_error_hva(addr))
1159 return -EFAULT;
1160 r = copy_to_user((void __user *)addr + offset, data, len);
1161 if (r)
1162 return -EFAULT;
1163 mark_page_dirty(kvm, gfn);
1164 return 0;
1165 }
1166 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1167
1168 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1169 unsigned long len)
1170 {
1171 gfn_t gfn = gpa >> PAGE_SHIFT;
1172 int seg;
1173 int offset = offset_in_page(gpa);
1174 int ret;
1175
1176 while ((seg = next_segment(len, offset)) != 0) {
1177 ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1178 if (ret < 0)
1179 return ret;
1180 offset = 0;
1181 len -= seg;
1182 data += seg;
1183 ++gfn;
1184 }
1185 return 0;
1186 }
1187
1188 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1189 {
1190 return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1191 }
1192 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1193
1194 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1195 {
1196 gfn_t gfn = gpa >> PAGE_SHIFT;
1197 int seg;
1198 int offset = offset_in_page(gpa);
1199 int ret;
1200
1201 while ((seg = next_segment(len, offset)) != 0) {
1202 ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1203 if (ret < 0)
1204 return ret;
1205 offset = 0;
1206 len -= seg;
1207 ++gfn;
1208 }
1209 return 0;
1210 }
1211 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1212
1213 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1214 {
1215 struct kvm_memory_slot *memslot;
1216
1217 gfn = unalias_gfn(kvm, gfn);
1218 memslot = __gfn_to_memslot(kvm, gfn);
1219 if (memslot && memslot->dirty_bitmap) {
1220 unsigned long rel_gfn = gfn - memslot->base_gfn;
1221
1222 /* avoid RMW */
1223 if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1224 set_bit(rel_gfn, memslot->dirty_bitmap);
1225 }
1226 }
1227
1228 /*
1229 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1230 */
1231 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1232 {
1233 DEFINE_WAIT(wait);
1234
1235 for (;;) {
1236 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1237
1238 if (kvm_cpu_has_interrupt(vcpu) ||
1239 kvm_cpu_has_pending_timer(vcpu) ||
1240 kvm_arch_vcpu_runnable(vcpu)) {
1241 set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1242 break;
1243 }
1244 if (signal_pending(current))
1245 break;
1246
1247 vcpu_put(vcpu);
1248 schedule();
1249 vcpu_load(vcpu);
1250 }
1251
1252 finish_wait(&vcpu->wq, &wait);
1253 }
1254
1255 void kvm_resched(struct kvm_vcpu *vcpu)
1256 {
1257 if (!need_resched())
1258 return;
1259 cond_resched();
1260 }
1261 EXPORT_SYMBOL_GPL(kvm_resched);
1262
1263 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1264 {
1265 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1266 struct page *page;
1267
1268 if (vmf->pgoff == 0)
1269 page = virt_to_page(vcpu->run);
1270 #ifdef CONFIG_X86
1271 else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1272 page = virt_to_page(vcpu->arch.pio_data);
1273 #endif
1274 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1275 else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1276 page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1277 #endif
1278 else
1279 return VM_FAULT_SIGBUS;
1280 get_page(page);
1281 vmf->page = page;
1282 return 0;
1283 }
1284
1285 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1286 .fault = kvm_vcpu_fault,
1287 };
1288
1289 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1290 {
1291 vma->vm_ops = &kvm_vcpu_vm_ops;
1292 return 0;
1293 }
1294
1295 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1296 {
1297 struct kvm_vcpu *vcpu = filp->private_data;
1298
1299 kvm_put_kvm(vcpu->kvm);
1300 return 0;
1301 }
1302
1303 static const struct file_operations kvm_vcpu_fops = {
1304 .release = kvm_vcpu_release,
1305 .unlocked_ioctl = kvm_vcpu_ioctl,
1306 .compat_ioctl = kvm_vcpu_ioctl,
1307 .mmap = kvm_vcpu_mmap,
1308 };
1309
1310 /*
1311 * Allocates an inode for the vcpu.
1312 */
1313 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1314 {
1315 int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1316 if (fd < 0)
1317 kvm_put_kvm(vcpu->kvm);
1318 return fd;
1319 }
1320
1321 /*
1322 * Creates some virtual cpus. Good luck creating more than one.
1323 */
1324 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1325 {
1326 int r;
1327 struct kvm_vcpu *vcpu;
1328
1329 if (!valid_vcpu(n))
1330 return -EINVAL;
1331
1332 vcpu = kvm_arch_vcpu_create(kvm, n);
1333 if (IS_ERR(vcpu))
1334 return PTR_ERR(vcpu);
1335
1336 preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1337
1338 r = kvm_arch_vcpu_setup(vcpu);
1339 if (r)
1340 return r;
1341
1342 mutex_lock(&kvm->lock);
1343 if (kvm->vcpus[n]) {
1344 r = -EEXIST;
1345 goto vcpu_destroy;
1346 }
1347 kvm->vcpus[n] = vcpu;
1348 mutex_unlock(&kvm->lock);
1349
1350 /* Now it's all set up, let userspace reach it */
1351 kvm_get_kvm(kvm);
1352 r = create_vcpu_fd(vcpu);
1353 if (r < 0)
1354 goto unlink;
1355 return r;
1356
1357 unlink:
1358 mutex_lock(&kvm->lock);
1359 kvm->vcpus[n] = NULL;
1360 vcpu_destroy:
1361 mutex_unlock(&kvm->lock);
1362 kvm_arch_vcpu_destroy(vcpu);
1363 return r;
1364 }
1365
1366 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1367 {
1368 if (sigset) {
1369 sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1370 vcpu->sigset_active = 1;
1371 vcpu->sigset = *sigset;
1372 } else
1373 vcpu->sigset_active = 0;
1374 return 0;
1375 }
1376
1377 static long kvm_vcpu_ioctl(struct file *filp,
1378 unsigned int ioctl, unsigned long arg)
1379 {
1380 struct kvm_vcpu *vcpu = filp->private_data;
1381 void __user *argp = (void __user *)arg;
1382 int r;
1383 struct kvm_fpu *fpu = NULL;
1384 struct kvm_sregs *kvm_sregs = NULL;
1385
1386 if (vcpu->kvm->mm != current->mm)
1387 return -EIO;
1388 switch (ioctl) {
1389 case KVM_RUN:
1390 r = -EINVAL;
1391 if (arg)
1392 goto out;
1393 r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1394 break;
1395 case KVM_GET_REGS: {
1396 struct kvm_regs *kvm_regs;
1397
1398 r = -ENOMEM;
1399 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1400 if (!kvm_regs)
1401 goto out;
1402 r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1403 if (r)
1404 goto out_free1;
1405 r = -EFAULT;
1406 if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1407 goto out_free1;
1408 r = 0;
1409 out_free1:
1410 kfree(kvm_regs);
1411 break;
1412 }
1413 case KVM_SET_REGS: {
1414 struct kvm_regs *kvm_regs;
1415
1416 r = -ENOMEM;
1417 kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1418 if (!kvm_regs)
1419 goto out;
1420 r = -EFAULT;
1421 if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1422 goto out_free2;
1423 r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1424 if (r)
1425 goto out_free2;
1426 r = 0;
1427 out_free2:
1428 kfree(kvm_regs);
1429 break;
1430 }
1431 case KVM_GET_SREGS: {
1432 kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1433 r = -ENOMEM;
1434 if (!kvm_sregs)
1435 goto out;
1436 r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1437 if (r)
1438 goto out;
1439 r = -EFAULT;
1440 if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1441 goto out;
1442 r = 0;
1443 break;
1444 }
1445 case KVM_SET_SREGS: {
1446 kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1447 r = -ENOMEM;
1448 if (!kvm_sregs)
1449 goto out;
1450 r = -EFAULT;
1451 if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1452 goto out;
1453 r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1454 if (r)
1455 goto out;
1456 r = 0;
1457 break;
1458 }
1459 case KVM_GET_MP_STATE: {
1460 struct kvm_mp_state mp_state;
1461
1462 r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1463 if (r)
1464 goto out;
1465 r = -EFAULT;
1466 if (copy_to_user(argp, &mp_state, sizeof mp_state))
1467 goto out;
1468 r = 0;
1469 break;
1470 }
1471 case KVM_SET_MP_STATE: {
1472 struct kvm_mp_state mp_state;
1473
1474 r = -EFAULT;
1475 if (copy_from_user(&mp_state, argp, sizeof mp_state))
1476 goto out;
1477 r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1478 if (r)
1479 goto out;
1480 r = 0;
1481 break;
1482 }
1483 case KVM_TRANSLATE: {
1484 struct kvm_translation tr;
1485
1486 r = -EFAULT;
1487 if (copy_from_user(&tr, argp, sizeof tr))
1488 goto out;
1489 r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1490 if (r)
1491 goto out;
1492 r = -EFAULT;
1493 if (copy_to_user(argp, &tr, sizeof tr))
1494 goto out;
1495 r = 0;
1496 break;
1497 }
1498 case KVM_DEBUG_GUEST: {
1499 struct kvm_debug_guest dbg;
1500
1501 r = -EFAULT;
1502 if (copy_from_user(&dbg, argp, sizeof dbg))
1503 goto out;
1504 r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1505 if (r)
1506 goto out;
1507 r = 0;
1508 break;
1509 }
1510 case KVM_SET_SIGNAL_MASK: {
1511 struct kvm_signal_mask __user *sigmask_arg = argp;
1512 struct kvm_signal_mask kvm_sigmask;
1513 sigset_t sigset, *p;
1514
1515 p = NULL;
1516 if (argp) {
1517 r = -EFAULT;
1518 if (copy_from_user(&kvm_sigmask, argp,
1519 sizeof kvm_sigmask))
1520 goto out;
1521 r = -EINVAL;
1522 if (kvm_sigmask.len != sizeof sigset)
1523 goto out;
1524 r = -EFAULT;
1525 if (copy_from_user(&sigset, sigmask_arg->sigset,
1526 sizeof sigset))
1527 goto out;
1528 p = &sigset;
1529 }
1530 r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1531 break;
1532 }
1533 case KVM_GET_FPU: {
1534 fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1535 r = -ENOMEM;
1536 if (!fpu)
1537 goto out;
1538 r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1539 if (r)
1540 goto out;
1541 r = -EFAULT;
1542 if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1543 goto out;
1544 r = 0;
1545 break;
1546 }
1547 case KVM_SET_FPU: {
1548 fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1549 r = -ENOMEM;
1550 if (!fpu)
1551 goto out;
1552 r = -EFAULT;
1553 if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1554 goto out;
1555 r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1556 if (r)
1557 goto out;
1558 r = 0;
1559 break;
1560 }
1561 default:
1562 r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1563 }
1564 out:
1565 kfree(fpu);
1566 kfree(kvm_sregs);
1567 return r;
1568 }
1569
1570 static long kvm_vm_ioctl(struct file *filp,
1571 unsigned int ioctl, unsigned long arg)
1572 {
1573 struct kvm *kvm = filp->private_data;
1574 void __user *argp = (void __user *)arg;
1575 int r;
1576
1577 if (kvm->mm != current->mm)
1578 return -EIO;
1579 switch (ioctl) {
1580 case KVM_CREATE_VCPU:
1581 r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1582 if (r < 0)
1583 goto out;
1584 break;
1585 case KVM_SET_USER_MEMORY_REGION: {
1586 struct kvm_userspace_memory_region kvm_userspace_mem;
1587
1588 r = -EFAULT;
1589 if (copy_from_user(&kvm_userspace_mem, argp,
1590 sizeof kvm_userspace_mem))
1591 goto out;
1592
1593 r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1594 if (r)
1595 goto out;
1596 break;
1597 }
1598 case KVM_GET_DIRTY_LOG: {
1599 struct kvm_dirty_log log;
1600
1601 r = -EFAULT;
1602 if (copy_from_user(&log, argp, sizeof log))
1603 goto out;
1604 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1605 if (r)
1606 goto out;
1607 break;
1608 }
1609 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1610 case KVM_REGISTER_COALESCED_MMIO: {
1611 struct kvm_coalesced_mmio_zone zone;
1612 r = -EFAULT;
1613 if (copy_from_user(&zone, argp, sizeof zone))
1614 goto out;
1615 r = -ENXIO;
1616 r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1617 if (r)
1618 goto out;
1619 r = 0;
1620 break;
1621 }
1622 case KVM_UNREGISTER_COALESCED_MMIO: {
1623 struct kvm_coalesced_mmio_zone zone;
1624 r = -EFAULT;
1625 if (copy_from_user(&zone, argp, sizeof zone))
1626 goto out;
1627 r = -ENXIO;
1628 r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1629 if (r)
1630 goto out;
1631 r = 0;
1632 break;
1633 }
1634 #endif
1635 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1636 case KVM_ASSIGN_PCI_DEVICE: {
1637 struct kvm_assigned_pci_dev assigned_dev;
1638
1639 r = -EFAULT;
1640 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1641 goto out;
1642 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1643 if (r)
1644 goto out;
1645 break;
1646 }
1647 case KVM_ASSIGN_IRQ: {
1648 struct kvm_assigned_irq assigned_irq;
1649
1650 r = -EFAULT;
1651 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1652 goto out;
1653 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1654 if (r)
1655 goto out;
1656 break;
1657 }
1658 #endif
1659 default:
1660 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1661 }
1662 out:
1663 return r;
1664 }
1665
1666 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1667 {
1668 struct page *page[1];
1669 unsigned long addr;
1670 int npages;
1671 gfn_t gfn = vmf->pgoff;
1672 struct kvm *kvm = vma->vm_file->private_data;
1673
1674 addr = gfn_to_hva(kvm, gfn);
1675 if (kvm_is_error_hva(addr))
1676 return VM_FAULT_SIGBUS;
1677
1678 npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1679 NULL);
1680 if (unlikely(npages != 1))
1681 return VM_FAULT_SIGBUS;
1682
1683 vmf->page = page[0];
1684 return 0;
1685 }
1686
1687 static struct vm_operations_struct kvm_vm_vm_ops = {
1688 .fault = kvm_vm_fault,
1689 };
1690
1691 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1692 {
1693 vma->vm_ops = &kvm_vm_vm_ops;
1694 return 0;
1695 }
1696
1697 static const struct file_operations kvm_vm_fops = {
1698 .release = kvm_vm_release,
1699 .unlocked_ioctl = kvm_vm_ioctl,
1700 .compat_ioctl = kvm_vm_ioctl,
1701 .mmap = kvm_vm_mmap,
1702 };
1703
1704 static int kvm_dev_ioctl_create_vm(void)
1705 {
1706 int fd;
1707 struct kvm *kvm;
1708
1709 kvm = kvm_create_vm();
1710 if (IS_ERR(kvm))
1711 return PTR_ERR(kvm);
1712 fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1713 if (fd < 0)
1714 kvm_put_kvm(kvm);
1715
1716 return fd;
1717 }
1718
1719 static long kvm_dev_ioctl(struct file *filp,
1720 unsigned int ioctl, unsigned long arg)
1721 {
1722 long r = -EINVAL;
1723
1724 switch (ioctl) {
1725 case KVM_GET_API_VERSION:
1726 r = -EINVAL;
1727 if (arg)
1728 goto out;
1729 r = KVM_API_VERSION;
1730 break;
1731 case KVM_CREATE_VM:
1732 r = -EINVAL;
1733 if (arg)
1734 goto out;
1735 r = kvm_dev_ioctl_create_vm();
1736 break;
1737 case KVM_CHECK_EXTENSION:
1738 r = kvm_dev_ioctl_check_extension(arg);
1739 break;
1740 case KVM_GET_VCPU_MMAP_SIZE:
1741 r = -EINVAL;
1742 if (arg)
1743 goto out;
1744 r = PAGE_SIZE; /* struct kvm_run */
1745 #ifdef CONFIG_X86
1746 r += PAGE_SIZE; /* pio data page */
1747 #endif
1748 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1749 r += PAGE_SIZE; /* coalesced mmio ring page */
1750 #endif
1751 break;
1752 case KVM_TRACE_ENABLE:
1753 case KVM_TRACE_PAUSE:
1754 case KVM_TRACE_DISABLE:
1755 r = kvm_trace_ioctl(ioctl, arg);
1756 break;
1757 default:
1758 return kvm_arch_dev_ioctl(filp, ioctl, arg);
1759 }
1760 out:
1761 return r;
1762 }
1763
1764 static struct file_operations kvm_chardev_ops = {
1765 .unlocked_ioctl = kvm_dev_ioctl,
1766 .compat_ioctl = kvm_dev_ioctl,
1767 };
1768
1769 static struct miscdevice kvm_dev = {
1770 KVM_MINOR,
1771 "kvm",
1772 &kvm_chardev_ops,
1773 };
1774
1775 static void hardware_enable(void *junk)
1776 {
1777 int cpu = raw_smp_processor_id();
1778
1779 if (cpu_isset(cpu, cpus_hardware_enabled))
1780 return;
1781 cpu_set(cpu, cpus_hardware_enabled);
1782 kvm_arch_hardware_enable(NULL);
1783 }
1784
1785 static void hardware_disable(void *junk)
1786 {
1787 int cpu = raw_smp_processor_id();
1788
1789 if (!cpu_isset(cpu, cpus_hardware_enabled))
1790 return;
1791 cpu_clear(cpu, cpus_hardware_enabled);
1792 kvm_arch_hardware_disable(NULL);
1793 }
1794
1795 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1796 void *v)
1797 {
1798 int cpu = (long)v;
1799
1800 val &= ~CPU_TASKS_FROZEN;
1801 switch (val) {
1802 case CPU_DYING:
1803 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1804 cpu);
1805 hardware_disable(NULL);
1806 break;
1807 case CPU_UP_CANCELED:
1808 printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1809 cpu);
1810 smp_call_function_single(cpu, hardware_disable, NULL, 1);
1811 break;
1812 case CPU_ONLINE:
1813 printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1814 cpu);
1815 smp_call_function_single(cpu, hardware_enable, NULL, 1);
1816 break;
1817 }
1818 return NOTIFY_OK;
1819 }
1820
1821
1822 asmlinkage void kvm_handle_fault_on_reboot(void)
1823 {
1824 if (kvm_rebooting)
1825 /* spin while reset goes on */
1826 while (true)
1827 ;
1828 /* Fault while not rebooting. We want the trace. */
1829 BUG();
1830 }
1831 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1832
1833 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1834 void *v)
1835 {
1836 if (val == SYS_RESTART) {
1837 /*
1838 * Some (well, at least mine) BIOSes hang on reboot if
1839 * in vmx root mode.
1840 */
1841 printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1842 kvm_rebooting = true;
1843 on_each_cpu(hardware_disable, NULL, 1);
1844 }
1845 return NOTIFY_OK;
1846 }
1847
1848 static struct notifier_block kvm_reboot_notifier = {
1849 .notifier_call = kvm_reboot,
1850 .priority = 0,
1851 };
1852
1853 void kvm_io_bus_init(struct kvm_io_bus *bus)
1854 {
1855 memset(bus, 0, sizeof(*bus));
1856 }
1857
1858 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1859 {
1860 int i;
1861
1862 for (i = 0; i < bus->dev_count; i++) {
1863 struct kvm_io_device *pos = bus->devs[i];
1864
1865 kvm_iodevice_destructor(pos);
1866 }
1867 }
1868
1869 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1870 gpa_t addr, int len, int is_write)
1871 {
1872 int i;
1873
1874 for (i = 0; i < bus->dev_count; i++) {
1875 struct kvm_io_device *pos = bus->devs[i];
1876
1877 if (pos->in_range(pos, addr, len, is_write))
1878 return pos;
1879 }
1880
1881 return NULL;
1882 }
1883
1884 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1885 {
1886 BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1887
1888 bus->devs[bus->dev_count++] = dev;
1889 }
1890
1891 static struct notifier_block kvm_cpu_notifier = {
1892 .notifier_call = kvm_cpu_hotplug,
1893 .priority = 20, /* must be > scheduler priority */
1894 };
1895
1896 static int vm_stat_get(void *_offset, u64 *val)
1897 {
1898 unsigned offset = (long)_offset;
1899 struct kvm *kvm;
1900
1901 *val = 0;
1902 spin_lock(&kvm_lock);
1903 list_for_each_entry(kvm, &vm_list, vm_list)
1904 *val += *(u32 *)((void *)kvm + offset);
1905 spin_unlock(&kvm_lock);
1906 return 0;
1907 }
1908
1909 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1910
1911 static int vcpu_stat_get(void *_offset, u64 *val)
1912 {
1913 unsigned offset = (long)_offset;
1914 struct kvm *kvm;
1915 struct kvm_vcpu *vcpu;
1916 int i;
1917
1918 *val = 0;
1919 spin_lock(&kvm_lock);
1920 list_for_each_entry(kvm, &vm_list, vm_list)
1921 for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1922 vcpu = kvm->vcpus[i];
1923 if (vcpu)
1924 *val += *(u32 *)((void *)vcpu + offset);
1925 }
1926 spin_unlock(&kvm_lock);
1927 return 0;
1928 }
1929
1930 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1931
1932 static struct file_operations *stat_fops[] = {
1933 [KVM_STAT_VCPU] = &vcpu_stat_fops,
1934 [KVM_STAT_VM] = &vm_stat_fops,
1935 };
1936
1937 static void kvm_init_debug(void)
1938 {
1939 struct kvm_stats_debugfs_item *p;
1940
1941 kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1942 for (p = debugfs_entries; p->name; ++p)
1943 p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1944 (void *)(long)p->offset,
1945 stat_fops[p->kind]);
1946 }
1947
1948 static void kvm_exit_debug(void)
1949 {
1950 struct kvm_stats_debugfs_item *p;
1951
1952 for (p = debugfs_entries; p->name; ++p)
1953 debugfs_remove(p->dentry);
1954 debugfs_remove(kvm_debugfs_dir);
1955 }
1956
1957 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1958 {
1959 hardware_disable(NULL);
1960 return 0;
1961 }
1962
1963 static int kvm_resume(struct sys_device *dev)
1964 {
1965 hardware_enable(NULL);
1966 return 0;
1967 }
1968
1969 static struct sysdev_class kvm_sysdev_class = {
1970 .name = "kvm",
1971 .suspend = kvm_suspend,
1972 .resume = kvm_resume,
1973 };
1974
1975 static struct sys_device kvm_sysdev = {
1976 .id = 0,
1977 .cls = &kvm_sysdev_class,
1978 };
1979
1980 struct page *bad_page;
1981 pfn_t bad_pfn;
1982
1983 static inline
1984 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1985 {
1986 return container_of(pn, struct kvm_vcpu, preempt_notifier);
1987 }
1988
1989 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1990 {
1991 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1992
1993 kvm_arch_vcpu_load(vcpu, cpu);
1994 }
1995
1996 static void kvm_sched_out(struct preempt_notifier *pn,
1997 struct task_struct *next)
1998 {
1999 struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2000
2001 kvm_arch_vcpu_put(vcpu);
2002 }
2003
2004 int kvm_init(void *opaque, unsigned int vcpu_size,
2005 struct module *module)
2006 {
2007 int r;
2008 int cpu;
2009
2010 kvm_init_debug();
2011
2012 r = kvm_arch_init(opaque);
2013 if (r)
2014 goto out_fail;
2015
2016 bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2017
2018 if (bad_page == NULL) {
2019 r = -ENOMEM;
2020 goto out;
2021 }
2022
2023 bad_pfn = page_to_pfn(bad_page);
2024
2025 r = kvm_arch_hardware_setup();
2026 if (r < 0)
2027 goto out_free_0;
2028
2029 for_each_online_cpu(cpu) {
2030 smp_call_function_single(cpu,
2031 kvm_arch_check_processor_compat,
2032 &r, 1);
2033 if (r < 0)
2034 goto out_free_1;
2035 }
2036
2037 on_each_cpu(hardware_enable, NULL, 1);
2038 r = register_cpu_notifier(&kvm_cpu_notifier);
2039 if (r)
2040 goto out_free_2;
2041 register_reboot_notifier(&kvm_reboot_notifier);
2042
2043 r = sysdev_class_register(&kvm_sysdev_class);
2044 if (r)
2045 goto out_free_3;
2046
2047 r = sysdev_register(&kvm_sysdev);
2048 if (r)
2049 goto out_free_4;
2050
2051 /* A kmem cache lets us meet the alignment requirements of fx_save. */
2052 kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2053 __alignof__(struct kvm_vcpu),
2054 0, NULL);
2055 if (!kvm_vcpu_cache) {
2056 r = -ENOMEM;
2057 goto out_free_5;
2058 }
2059
2060 kvm_chardev_ops.owner = module;
2061
2062 r = misc_register(&kvm_dev);
2063 if (r) {
2064 printk(KERN_ERR "kvm: misc device register failed\n");
2065 goto out_free;
2066 }
2067
2068 kvm_preempt_ops.sched_in = kvm_sched_in;
2069 kvm_preempt_ops.sched_out = kvm_sched_out;
2070
2071 return 0;
2072
2073 out_free:
2074 kmem_cache_destroy(kvm_vcpu_cache);
2075 out_free_5:
2076 sysdev_unregister(&kvm_sysdev);
2077 out_free_4:
2078 sysdev_class_unregister(&kvm_sysdev_class);
2079 out_free_3:
2080 unregister_reboot_notifier(&kvm_reboot_notifier);
2081 unregister_cpu_notifier(&kvm_cpu_notifier);
2082 out_free_2:
2083 on_each_cpu(hardware_disable, NULL, 1);
2084 out_free_1:
2085 kvm_arch_hardware_unsetup();
2086 out_free_0:
2087 __free_page(bad_page);
2088 out:
2089 kvm_arch_exit();
2090 kvm_exit_debug();
2091 out_fail:
2092 return r;
2093 }
2094 EXPORT_SYMBOL_GPL(kvm_init);
2095
2096 void kvm_exit(void)
2097 {
2098 kvm_trace_cleanup();
2099 misc_deregister(&kvm_dev);
2100 kmem_cache_destroy(kvm_vcpu_cache);
2101 sysdev_unregister(&kvm_sysdev);
2102 sysdev_class_unregister(&kvm_sysdev_class);
2103 unregister_reboot_notifier(&kvm_reboot_notifier);
2104 unregister_cpu_notifier(&kvm_cpu_notifier);
2105 on_each_cpu(hardware_disable, NULL, 1);
2106 kvm_arch_hardware_unsetup();
2107 kvm_arch_exit();
2108 kvm_exit_debug();
2109 __free_page(bad_page);
2110 }
2111 EXPORT_SYMBOL_GPL(kvm_exit);
(deprecated) Btrfs devel tree