xref: /openbmc/linux/virt/kvm/kvm_main.c (revision 9ac8d3fb)
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(kvm, &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_release_regions(assigned_dev->dev);
156 	pci_disable_device(assigned_dev->dev);
157 	pci_dev_put(assigned_dev->dev);
158 
159 	list_del(&assigned_dev->list);
160 	kfree(assigned_dev);
161 }
162 
163 void kvm_free_all_assigned_devices(struct kvm *kvm)
164 {
165 	struct list_head *ptr, *ptr2;
166 	struct kvm_assigned_dev_kernel *assigned_dev;
167 
168 	list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
169 		assigned_dev = list_entry(ptr,
170 					  struct kvm_assigned_dev_kernel,
171 					  list);
172 
173 		kvm_free_assigned_device(kvm, assigned_dev);
174 	}
175 }
176 
177 static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
178 				   struct kvm_assigned_irq
179 				   *assigned_irq)
180 {
181 	int r = 0;
182 	struct kvm_assigned_dev_kernel *match;
183 
184 	mutex_lock(&kvm->lock);
185 
186 	match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
187 				      assigned_irq->assigned_dev_id);
188 	if (!match) {
189 		mutex_unlock(&kvm->lock);
190 		return -EINVAL;
191 	}
192 
193 	if (match->irq_requested) {
194 		match->guest_irq = assigned_irq->guest_irq;
195 		match->ack_notifier.gsi = assigned_irq->guest_irq;
196 		mutex_unlock(&kvm->lock);
197 		return 0;
198 	}
199 
200 	INIT_WORK(&match->interrupt_work,
201 		  kvm_assigned_dev_interrupt_work_handler);
202 
203 	if (irqchip_in_kernel(kvm)) {
204 		if (!capable(CAP_SYS_RAWIO)) {
205 			r = -EPERM;
206 			goto out_release;
207 		}
208 
209 		if (assigned_irq->host_irq)
210 			match->host_irq = assigned_irq->host_irq;
211 		else
212 			match->host_irq = match->dev->irq;
213 		match->guest_irq = assigned_irq->guest_irq;
214 		match->ack_notifier.gsi = assigned_irq->guest_irq;
215 		match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
216 		kvm_register_irq_ack_notifier(kvm, &match->ack_notifier);
217 		r = kvm_request_irq_source_id(kvm);
218 		if (r < 0)
219 			goto out_release;
220 		else
221 			match->irq_source_id = r;
222 
223 		/* Even though this is PCI, we don't want to use shared
224 		 * interrupts. Sharing host devices with guest-assigned devices
225 		 * on the same interrupt line is not a happy situation: there
226 		 * are going to be long delays in accepting, acking, etc.
227 		 */
228 		if (request_irq(match->host_irq, kvm_assigned_dev_intr, 0,
229 				"kvm_assigned_device", (void *)match)) {
230 			r = -EIO;
231 			goto out_release;
232 		}
233 	}
234 
235 	match->irq_requested = true;
236 	mutex_unlock(&kvm->lock);
237 	return r;
238 out_release:
239 	mutex_unlock(&kvm->lock);
240 	kvm_free_assigned_device(kvm, match);
241 	return r;
242 }
243 
244 static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
245 				      struct kvm_assigned_pci_dev *assigned_dev)
246 {
247 	int r = 0;
248 	struct kvm_assigned_dev_kernel *match;
249 	struct pci_dev *dev;
250 
251 	mutex_lock(&kvm->lock);
252 
253 	match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
254 				      assigned_dev->assigned_dev_id);
255 	if (match) {
256 		/* device already assigned */
257 		r = -EINVAL;
258 		goto out;
259 	}
260 
261 	match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
262 	if (match == NULL) {
263 		printk(KERN_INFO "%s: Couldn't allocate memory\n",
264 		       __func__);
265 		r = -ENOMEM;
266 		goto out;
267 	}
268 	dev = pci_get_bus_and_slot(assigned_dev->busnr,
269 				   assigned_dev->devfn);
270 	if (!dev) {
271 		printk(KERN_INFO "%s: host device not found\n", __func__);
272 		r = -EINVAL;
273 		goto out_free;
274 	}
275 	if (pci_enable_device(dev)) {
276 		printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
277 		r = -EBUSY;
278 		goto out_put;
279 	}
280 	r = pci_request_regions(dev, "kvm_assigned_device");
281 	if (r) {
282 		printk(KERN_INFO "%s: Could not get access to device regions\n",
283 		       __func__);
284 		goto out_disable;
285 	}
286 	match->assigned_dev_id = assigned_dev->assigned_dev_id;
287 	match->host_busnr = assigned_dev->busnr;
288 	match->host_devfn = assigned_dev->devfn;
289 	match->dev = dev;
290 
291 	match->kvm = kvm;
292 
293 	list_add(&match->list, &kvm->arch.assigned_dev_head);
294 
295 	if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
296 		r = kvm_iommu_map_guest(kvm, match);
297 		if (r)
298 			goto out_list_del;
299 	}
300 
301 out:
302 	mutex_unlock(&kvm->lock);
303 	return r;
304 out_list_del:
305 	list_del(&match->list);
306 	pci_release_regions(dev);
307 out_disable:
308 	pci_disable_device(dev);
309 out_put:
310 	pci_dev_put(dev);
311 out_free:
312 	kfree(match);
313 	mutex_unlock(&kvm->lock);
314 	return r;
315 }
316 #endif
317 
318 static inline int valid_vcpu(int n)
319 {
320 	return likely(n >= 0 && n < KVM_MAX_VCPUS);
321 }
322 
323 inline int kvm_is_mmio_pfn(pfn_t pfn)
324 {
325 	if (pfn_valid(pfn))
326 		return PageReserved(pfn_to_page(pfn));
327 
328 	return true;
329 }
330 
331 /*
332  * Switches to specified vcpu, until a matching vcpu_put()
333  */
334 void vcpu_load(struct kvm_vcpu *vcpu)
335 {
336 	int cpu;
337 
338 	mutex_lock(&vcpu->mutex);
339 	cpu = get_cpu();
340 	preempt_notifier_register(&vcpu->preempt_notifier);
341 	kvm_arch_vcpu_load(vcpu, cpu);
342 	put_cpu();
343 }
344 
345 void vcpu_put(struct kvm_vcpu *vcpu)
346 {
347 	preempt_disable();
348 	kvm_arch_vcpu_put(vcpu);
349 	preempt_notifier_unregister(&vcpu->preempt_notifier);
350 	preempt_enable();
351 	mutex_unlock(&vcpu->mutex);
352 }
353 
354 static void ack_flush(void *_completed)
355 {
356 }
357 
358 void kvm_flush_remote_tlbs(struct kvm *kvm)
359 {
360 	int i, cpu, me;
361 	cpumask_t cpus;
362 	struct kvm_vcpu *vcpu;
363 
364 	me = get_cpu();
365 	cpus_clear(cpus);
366 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
367 		vcpu = kvm->vcpus[i];
368 		if (!vcpu)
369 			continue;
370 		if (test_and_set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
371 			continue;
372 		cpu = vcpu->cpu;
373 		if (cpu != -1 && cpu != me)
374 			cpu_set(cpu, cpus);
375 	}
376 	if (cpus_empty(cpus))
377 		goto out;
378 	++kvm->stat.remote_tlb_flush;
379 	smp_call_function_mask(cpus, ack_flush, NULL, 1);
380 out:
381 	put_cpu();
382 }
383 
384 void kvm_reload_remote_mmus(struct kvm *kvm)
385 {
386 	int i, cpu, me;
387 	cpumask_t cpus;
388 	struct kvm_vcpu *vcpu;
389 
390 	me = get_cpu();
391 	cpus_clear(cpus);
392 	for (i = 0; i < KVM_MAX_VCPUS; ++i) {
393 		vcpu = kvm->vcpus[i];
394 		if (!vcpu)
395 			continue;
396 		if (test_and_set_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests))
397 			continue;
398 		cpu = vcpu->cpu;
399 		if (cpu != -1 && cpu != me)
400 			cpu_set(cpu, cpus);
401 	}
402 	if (cpus_empty(cpus))
403 		goto out;
404 	smp_call_function_mask(cpus, ack_flush, NULL, 1);
405 out:
406 	put_cpu();
407 }
408 
409 
410 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
411 {
412 	struct page *page;
413 	int r;
414 
415 	mutex_init(&vcpu->mutex);
416 	vcpu->cpu = -1;
417 	vcpu->kvm = kvm;
418 	vcpu->vcpu_id = id;
419 	init_waitqueue_head(&vcpu->wq);
420 
421 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
422 	if (!page) {
423 		r = -ENOMEM;
424 		goto fail;
425 	}
426 	vcpu->run = page_address(page);
427 
428 	r = kvm_arch_vcpu_init(vcpu);
429 	if (r < 0)
430 		goto fail_free_run;
431 	return 0;
432 
433 fail_free_run:
434 	free_page((unsigned long)vcpu->run);
435 fail:
436 	return r;
437 }
438 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
439 
440 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
441 {
442 	kvm_arch_vcpu_uninit(vcpu);
443 	free_page((unsigned long)vcpu->run);
444 }
445 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
446 
447 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
448 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
449 {
450 	return container_of(mn, struct kvm, mmu_notifier);
451 }
452 
453 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
454 					     struct mm_struct *mm,
455 					     unsigned long address)
456 {
457 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
458 	int need_tlb_flush;
459 
460 	/*
461 	 * When ->invalidate_page runs, the linux pte has been zapped
462 	 * already but the page is still allocated until
463 	 * ->invalidate_page returns. So if we increase the sequence
464 	 * here the kvm page fault will notice if the spte can't be
465 	 * established because the page is going to be freed. If
466 	 * instead the kvm page fault establishes the spte before
467 	 * ->invalidate_page runs, kvm_unmap_hva will release it
468 	 * before returning.
469 	 *
470 	 * The sequence increase only need to be seen at spin_unlock
471 	 * time, and not at spin_lock time.
472 	 *
473 	 * Increasing the sequence after the spin_unlock would be
474 	 * unsafe because the kvm page fault could then establish the
475 	 * pte after kvm_unmap_hva returned, without noticing the page
476 	 * is going to be freed.
477 	 */
478 	spin_lock(&kvm->mmu_lock);
479 	kvm->mmu_notifier_seq++;
480 	need_tlb_flush = kvm_unmap_hva(kvm, address);
481 	spin_unlock(&kvm->mmu_lock);
482 
483 	/* we've to flush the tlb before the pages can be freed */
484 	if (need_tlb_flush)
485 		kvm_flush_remote_tlbs(kvm);
486 
487 }
488 
489 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
490 						    struct mm_struct *mm,
491 						    unsigned long start,
492 						    unsigned long end)
493 {
494 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
495 	int need_tlb_flush = 0;
496 
497 	spin_lock(&kvm->mmu_lock);
498 	/*
499 	 * The count increase must become visible at unlock time as no
500 	 * spte can be established without taking the mmu_lock and
501 	 * count is also read inside the mmu_lock critical section.
502 	 */
503 	kvm->mmu_notifier_count++;
504 	for (; start < end; start += PAGE_SIZE)
505 		need_tlb_flush |= kvm_unmap_hva(kvm, start);
506 	spin_unlock(&kvm->mmu_lock);
507 
508 	/* we've to flush the tlb before the pages can be freed */
509 	if (need_tlb_flush)
510 		kvm_flush_remote_tlbs(kvm);
511 }
512 
513 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
514 						  struct mm_struct *mm,
515 						  unsigned long start,
516 						  unsigned long end)
517 {
518 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
519 
520 	spin_lock(&kvm->mmu_lock);
521 	/*
522 	 * This sequence increase will notify the kvm page fault that
523 	 * the page that is going to be mapped in the spte could have
524 	 * been freed.
525 	 */
526 	kvm->mmu_notifier_seq++;
527 	/*
528 	 * The above sequence increase must be visible before the
529 	 * below count decrease but both values are read by the kvm
530 	 * page fault under mmu_lock spinlock so we don't need to add
531 	 * a smb_wmb() here in between the two.
532 	 */
533 	kvm->mmu_notifier_count--;
534 	spin_unlock(&kvm->mmu_lock);
535 
536 	BUG_ON(kvm->mmu_notifier_count < 0);
537 }
538 
539 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
540 					      struct mm_struct *mm,
541 					      unsigned long address)
542 {
543 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
544 	int young;
545 
546 	spin_lock(&kvm->mmu_lock);
547 	young = kvm_age_hva(kvm, address);
548 	spin_unlock(&kvm->mmu_lock);
549 
550 	if (young)
551 		kvm_flush_remote_tlbs(kvm);
552 
553 	return young;
554 }
555 
556 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
557 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
558 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
559 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
560 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
561 };
562 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
563 
564 static struct kvm *kvm_create_vm(void)
565 {
566 	struct kvm *kvm = kvm_arch_create_vm();
567 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
568 	struct page *page;
569 #endif
570 
571 	if (IS_ERR(kvm))
572 		goto out;
573 
574 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
575 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
576 	if (!page) {
577 		kfree(kvm);
578 		return ERR_PTR(-ENOMEM);
579 	}
580 	kvm->coalesced_mmio_ring =
581 			(struct kvm_coalesced_mmio_ring *)page_address(page);
582 #endif
583 
584 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
585 	{
586 		int err;
587 		kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
588 		err = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
589 		if (err) {
590 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
591 			put_page(page);
592 #endif
593 			kfree(kvm);
594 			return ERR_PTR(err);
595 		}
596 	}
597 #endif
598 
599 	kvm->mm = current->mm;
600 	atomic_inc(&kvm->mm->mm_count);
601 	spin_lock_init(&kvm->mmu_lock);
602 	kvm_io_bus_init(&kvm->pio_bus);
603 	mutex_init(&kvm->lock);
604 	kvm_io_bus_init(&kvm->mmio_bus);
605 	init_rwsem(&kvm->slots_lock);
606 	atomic_set(&kvm->users_count, 1);
607 	spin_lock(&kvm_lock);
608 	list_add(&kvm->vm_list, &vm_list);
609 	spin_unlock(&kvm_lock);
610 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
611 	kvm_coalesced_mmio_init(kvm);
612 #endif
613 out:
614 	return kvm;
615 }
616 
617 /*
618  * Free any memory in @free but not in @dont.
619  */
620 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
621 				  struct kvm_memory_slot *dont)
622 {
623 	if (!dont || free->rmap != dont->rmap)
624 		vfree(free->rmap);
625 
626 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
627 		vfree(free->dirty_bitmap);
628 
629 	if (!dont || free->lpage_info != dont->lpage_info)
630 		vfree(free->lpage_info);
631 
632 	free->npages = 0;
633 	free->dirty_bitmap = NULL;
634 	free->rmap = NULL;
635 	free->lpage_info = NULL;
636 }
637 
638 void kvm_free_physmem(struct kvm *kvm)
639 {
640 	int i;
641 
642 	for (i = 0; i < kvm->nmemslots; ++i)
643 		kvm_free_physmem_slot(&kvm->memslots[i], NULL);
644 }
645 
646 static void kvm_destroy_vm(struct kvm *kvm)
647 {
648 	struct mm_struct *mm = kvm->mm;
649 
650 	spin_lock(&kvm_lock);
651 	list_del(&kvm->vm_list);
652 	spin_unlock(&kvm_lock);
653 	kvm_io_bus_destroy(&kvm->pio_bus);
654 	kvm_io_bus_destroy(&kvm->mmio_bus);
655 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
656 	if (kvm->coalesced_mmio_ring != NULL)
657 		free_page((unsigned long)kvm->coalesced_mmio_ring);
658 #endif
659 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
660 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
661 #endif
662 	kvm_arch_destroy_vm(kvm);
663 	mmdrop(mm);
664 }
665 
666 void kvm_get_kvm(struct kvm *kvm)
667 {
668 	atomic_inc(&kvm->users_count);
669 }
670 EXPORT_SYMBOL_GPL(kvm_get_kvm);
671 
672 void kvm_put_kvm(struct kvm *kvm)
673 {
674 	if (atomic_dec_and_test(&kvm->users_count))
675 		kvm_destroy_vm(kvm);
676 }
677 EXPORT_SYMBOL_GPL(kvm_put_kvm);
678 
679 
680 static int kvm_vm_release(struct inode *inode, struct file *filp)
681 {
682 	struct kvm *kvm = filp->private_data;
683 
684 	kvm_put_kvm(kvm);
685 	return 0;
686 }
687 
688 /*
689  * Allocate some memory and give it an address in the guest physical address
690  * space.
691  *
692  * Discontiguous memory is allowed, mostly for framebuffers.
693  *
694  * Must be called holding mmap_sem for write.
695  */
696 int __kvm_set_memory_region(struct kvm *kvm,
697 			    struct kvm_userspace_memory_region *mem,
698 			    int user_alloc)
699 {
700 	int r;
701 	gfn_t base_gfn;
702 	unsigned long npages;
703 	unsigned long i;
704 	struct kvm_memory_slot *memslot;
705 	struct kvm_memory_slot old, new;
706 
707 	r = -EINVAL;
708 	/* General sanity checks */
709 	if (mem->memory_size & (PAGE_SIZE - 1))
710 		goto out;
711 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
712 		goto out;
713 	if (mem->slot >= KVM_MEMORY_SLOTS + KVM_PRIVATE_MEM_SLOTS)
714 		goto out;
715 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
716 		goto out;
717 
718 	memslot = &kvm->memslots[mem->slot];
719 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
720 	npages = mem->memory_size >> PAGE_SHIFT;
721 
722 	if (!npages)
723 		mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
724 
725 	new = old = *memslot;
726 
727 	new.base_gfn = base_gfn;
728 	new.npages = npages;
729 	new.flags = mem->flags;
730 
731 	/* Disallow changing a memory slot's size. */
732 	r = -EINVAL;
733 	if (npages && old.npages && npages != old.npages)
734 		goto out_free;
735 
736 	/* Check for overlaps */
737 	r = -EEXIST;
738 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
739 		struct kvm_memory_slot *s = &kvm->memslots[i];
740 
741 		if (s == memslot)
742 			continue;
743 		if (!((base_gfn + npages <= s->base_gfn) ||
744 		      (base_gfn >= s->base_gfn + s->npages)))
745 			goto out_free;
746 	}
747 
748 	/* Free page dirty bitmap if unneeded */
749 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
750 		new.dirty_bitmap = NULL;
751 
752 	r = -ENOMEM;
753 
754 	/* Allocate if a slot is being created */
755 #ifndef CONFIG_S390
756 	if (npages && !new.rmap) {
757 		new.rmap = vmalloc(npages * sizeof(struct page *));
758 
759 		if (!new.rmap)
760 			goto out_free;
761 
762 		memset(new.rmap, 0, npages * sizeof(*new.rmap));
763 
764 		new.user_alloc = user_alloc;
765 		/*
766 		 * hva_to_rmmap() serialzies with the mmu_lock and to be
767 		 * safe it has to ignore memslots with !user_alloc &&
768 		 * !userspace_addr.
769 		 */
770 		if (user_alloc)
771 			new.userspace_addr = mem->userspace_addr;
772 		else
773 			new.userspace_addr = 0;
774 	}
775 	if (npages && !new.lpage_info) {
776 		int largepages = npages / KVM_PAGES_PER_HPAGE;
777 		if (npages % KVM_PAGES_PER_HPAGE)
778 			largepages++;
779 		if (base_gfn % KVM_PAGES_PER_HPAGE)
780 			largepages++;
781 
782 		new.lpage_info = vmalloc(largepages * sizeof(*new.lpage_info));
783 
784 		if (!new.lpage_info)
785 			goto out_free;
786 
787 		memset(new.lpage_info, 0, largepages * sizeof(*new.lpage_info));
788 
789 		if (base_gfn % KVM_PAGES_PER_HPAGE)
790 			new.lpage_info[0].write_count = 1;
791 		if ((base_gfn+npages) % KVM_PAGES_PER_HPAGE)
792 			new.lpage_info[largepages-1].write_count = 1;
793 	}
794 
795 	/* Allocate page dirty bitmap if needed */
796 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
797 		unsigned dirty_bytes = ALIGN(npages, BITS_PER_LONG) / 8;
798 
799 		new.dirty_bitmap = vmalloc(dirty_bytes);
800 		if (!new.dirty_bitmap)
801 			goto out_free;
802 		memset(new.dirty_bitmap, 0, dirty_bytes);
803 	}
804 #endif /* not defined CONFIG_S390 */
805 
806 	if (!npages)
807 		kvm_arch_flush_shadow(kvm);
808 
809 	spin_lock(&kvm->mmu_lock);
810 	if (mem->slot >= kvm->nmemslots)
811 		kvm->nmemslots = mem->slot + 1;
812 
813 	*memslot = new;
814 	spin_unlock(&kvm->mmu_lock);
815 
816 	r = kvm_arch_set_memory_region(kvm, mem, old, user_alloc);
817 	if (r) {
818 		spin_lock(&kvm->mmu_lock);
819 		*memslot = old;
820 		spin_unlock(&kvm->mmu_lock);
821 		goto out_free;
822 	}
823 
824 	kvm_free_physmem_slot(&old, &new);
825 #ifdef CONFIG_DMAR
826 	/* map the pages in iommu page table */
827 	r = kvm_iommu_map_pages(kvm, base_gfn, npages);
828 	if (r)
829 		goto out;
830 #endif
831 	return 0;
832 
833 out_free:
834 	kvm_free_physmem_slot(&new, &old);
835 out:
836 	return r;
837 
838 }
839 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
840 
841 int kvm_set_memory_region(struct kvm *kvm,
842 			  struct kvm_userspace_memory_region *mem,
843 			  int user_alloc)
844 {
845 	int r;
846 
847 	down_write(&kvm->slots_lock);
848 	r = __kvm_set_memory_region(kvm, mem, user_alloc);
849 	up_write(&kvm->slots_lock);
850 	return r;
851 }
852 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
853 
854 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
855 				   struct
856 				   kvm_userspace_memory_region *mem,
857 				   int user_alloc)
858 {
859 	if (mem->slot >= KVM_MEMORY_SLOTS)
860 		return -EINVAL;
861 	return kvm_set_memory_region(kvm, mem, user_alloc);
862 }
863 
864 int kvm_get_dirty_log(struct kvm *kvm,
865 			struct kvm_dirty_log *log, int *is_dirty)
866 {
867 	struct kvm_memory_slot *memslot;
868 	int r, i;
869 	int n;
870 	unsigned long any = 0;
871 
872 	r = -EINVAL;
873 	if (log->slot >= KVM_MEMORY_SLOTS)
874 		goto out;
875 
876 	memslot = &kvm->memslots[log->slot];
877 	r = -ENOENT;
878 	if (!memslot->dirty_bitmap)
879 		goto out;
880 
881 	n = ALIGN(memslot->npages, BITS_PER_LONG) / 8;
882 
883 	for (i = 0; !any && i < n/sizeof(long); ++i)
884 		any = memslot->dirty_bitmap[i];
885 
886 	r = -EFAULT;
887 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
888 		goto out;
889 
890 	if (any)
891 		*is_dirty = 1;
892 
893 	r = 0;
894 out:
895 	return r;
896 }
897 
898 int is_error_page(struct page *page)
899 {
900 	return page == bad_page;
901 }
902 EXPORT_SYMBOL_GPL(is_error_page);
903 
904 int is_error_pfn(pfn_t pfn)
905 {
906 	return pfn == bad_pfn;
907 }
908 EXPORT_SYMBOL_GPL(is_error_pfn);
909 
910 static inline unsigned long bad_hva(void)
911 {
912 	return PAGE_OFFSET;
913 }
914 
915 int kvm_is_error_hva(unsigned long addr)
916 {
917 	return addr == bad_hva();
918 }
919 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
920 
921 static struct kvm_memory_slot *__gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
922 {
923 	int i;
924 
925 	for (i = 0; i < kvm->nmemslots; ++i) {
926 		struct kvm_memory_slot *memslot = &kvm->memslots[i];
927 
928 		if (gfn >= memslot->base_gfn
929 		    && gfn < memslot->base_gfn + memslot->npages)
930 			return memslot;
931 	}
932 	return NULL;
933 }
934 
935 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
936 {
937 	gfn = unalias_gfn(kvm, gfn);
938 	return __gfn_to_memslot(kvm, gfn);
939 }
940 
941 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
942 {
943 	int i;
944 
945 	gfn = unalias_gfn(kvm, gfn);
946 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
947 		struct kvm_memory_slot *memslot = &kvm->memslots[i];
948 
949 		if (gfn >= memslot->base_gfn
950 		    && gfn < memslot->base_gfn + memslot->npages)
951 			return 1;
952 	}
953 	return 0;
954 }
955 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
956 
957 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
958 {
959 	struct kvm_memory_slot *slot;
960 
961 	gfn = unalias_gfn(kvm, gfn);
962 	slot = __gfn_to_memslot(kvm, gfn);
963 	if (!slot)
964 		return bad_hva();
965 	return (slot->userspace_addr + (gfn - slot->base_gfn) * PAGE_SIZE);
966 }
967 EXPORT_SYMBOL_GPL(gfn_to_hva);
968 
969 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
970 {
971 	struct page *page[1];
972 	unsigned long addr;
973 	int npages;
974 	pfn_t pfn;
975 
976 	might_sleep();
977 
978 	addr = gfn_to_hva(kvm, gfn);
979 	if (kvm_is_error_hva(addr)) {
980 		get_page(bad_page);
981 		return page_to_pfn(bad_page);
982 	}
983 
984 	npages = get_user_pages_fast(addr, 1, 1, page);
985 
986 	if (unlikely(npages != 1)) {
987 		struct vm_area_struct *vma;
988 
989 		down_read(&current->mm->mmap_sem);
990 		vma = find_vma(current->mm, addr);
991 
992 		if (vma == NULL || addr < vma->vm_start ||
993 		    !(vma->vm_flags & VM_PFNMAP)) {
994 			up_read(&current->mm->mmap_sem);
995 			get_page(bad_page);
996 			return page_to_pfn(bad_page);
997 		}
998 
999 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1000 		up_read(&current->mm->mmap_sem);
1001 		BUG_ON(!kvm_is_mmio_pfn(pfn));
1002 	} else
1003 		pfn = page_to_pfn(page[0]);
1004 
1005 	return pfn;
1006 }
1007 
1008 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1009 
1010 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1011 {
1012 	pfn_t pfn;
1013 
1014 	pfn = gfn_to_pfn(kvm, gfn);
1015 	if (!kvm_is_mmio_pfn(pfn))
1016 		return pfn_to_page(pfn);
1017 
1018 	WARN_ON(kvm_is_mmio_pfn(pfn));
1019 
1020 	get_page(bad_page);
1021 	return bad_page;
1022 }
1023 
1024 EXPORT_SYMBOL_GPL(gfn_to_page);
1025 
1026 void kvm_release_page_clean(struct page *page)
1027 {
1028 	kvm_release_pfn_clean(page_to_pfn(page));
1029 }
1030 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1031 
1032 void kvm_release_pfn_clean(pfn_t pfn)
1033 {
1034 	if (!kvm_is_mmio_pfn(pfn))
1035 		put_page(pfn_to_page(pfn));
1036 }
1037 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1038 
1039 void kvm_release_page_dirty(struct page *page)
1040 {
1041 	kvm_release_pfn_dirty(page_to_pfn(page));
1042 }
1043 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1044 
1045 void kvm_release_pfn_dirty(pfn_t pfn)
1046 {
1047 	kvm_set_pfn_dirty(pfn);
1048 	kvm_release_pfn_clean(pfn);
1049 }
1050 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1051 
1052 void kvm_set_page_dirty(struct page *page)
1053 {
1054 	kvm_set_pfn_dirty(page_to_pfn(page));
1055 }
1056 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1057 
1058 void kvm_set_pfn_dirty(pfn_t pfn)
1059 {
1060 	if (!kvm_is_mmio_pfn(pfn)) {
1061 		struct page *page = pfn_to_page(pfn);
1062 		if (!PageReserved(page))
1063 			SetPageDirty(page);
1064 	}
1065 }
1066 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1067 
1068 void kvm_set_pfn_accessed(pfn_t pfn)
1069 {
1070 	if (!kvm_is_mmio_pfn(pfn))
1071 		mark_page_accessed(pfn_to_page(pfn));
1072 }
1073 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1074 
1075 void kvm_get_pfn(pfn_t pfn)
1076 {
1077 	if (!kvm_is_mmio_pfn(pfn))
1078 		get_page(pfn_to_page(pfn));
1079 }
1080 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1081 
1082 static int next_segment(unsigned long len, int offset)
1083 {
1084 	if (len > PAGE_SIZE - offset)
1085 		return PAGE_SIZE - offset;
1086 	else
1087 		return len;
1088 }
1089 
1090 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1091 			int len)
1092 {
1093 	int r;
1094 	unsigned long addr;
1095 
1096 	addr = gfn_to_hva(kvm, gfn);
1097 	if (kvm_is_error_hva(addr))
1098 		return -EFAULT;
1099 	r = copy_from_user(data, (void __user *)addr + offset, len);
1100 	if (r)
1101 		return -EFAULT;
1102 	return 0;
1103 }
1104 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1105 
1106 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1107 {
1108 	gfn_t gfn = gpa >> PAGE_SHIFT;
1109 	int seg;
1110 	int offset = offset_in_page(gpa);
1111 	int ret;
1112 
1113 	while ((seg = next_segment(len, offset)) != 0) {
1114 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1115 		if (ret < 0)
1116 			return ret;
1117 		offset = 0;
1118 		len -= seg;
1119 		data += seg;
1120 		++gfn;
1121 	}
1122 	return 0;
1123 }
1124 EXPORT_SYMBOL_GPL(kvm_read_guest);
1125 
1126 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1127 			  unsigned long len)
1128 {
1129 	int r;
1130 	unsigned long addr;
1131 	gfn_t gfn = gpa >> PAGE_SHIFT;
1132 	int offset = offset_in_page(gpa);
1133 
1134 	addr = gfn_to_hva(kvm, gfn);
1135 	if (kvm_is_error_hva(addr))
1136 		return -EFAULT;
1137 	pagefault_disable();
1138 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1139 	pagefault_enable();
1140 	if (r)
1141 		return -EFAULT;
1142 	return 0;
1143 }
1144 EXPORT_SYMBOL(kvm_read_guest_atomic);
1145 
1146 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1147 			 int offset, int len)
1148 {
1149 	int r;
1150 	unsigned long addr;
1151 
1152 	addr = gfn_to_hva(kvm, gfn);
1153 	if (kvm_is_error_hva(addr))
1154 		return -EFAULT;
1155 	r = copy_to_user((void __user *)addr + offset, data, len);
1156 	if (r)
1157 		return -EFAULT;
1158 	mark_page_dirty(kvm, gfn);
1159 	return 0;
1160 }
1161 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1162 
1163 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1164 		    unsigned long len)
1165 {
1166 	gfn_t gfn = gpa >> PAGE_SHIFT;
1167 	int seg;
1168 	int offset = offset_in_page(gpa);
1169 	int ret;
1170 
1171 	while ((seg = next_segment(len, offset)) != 0) {
1172 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1173 		if (ret < 0)
1174 			return ret;
1175 		offset = 0;
1176 		len -= seg;
1177 		data += seg;
1178 		++gfn;
1179 	}
1180 	return 0;
1181 }
1182 
1183 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1184 {
1185 	return kvm_write_guest_page(kvm, gfn, empty_zero_page, offset, len);
1186 }
1187 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1188 
1189 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1190 {
1191 	gfn_t gfn = gpa >> PAGE_SHIFT;
1192 	int seg;
1193 	int offset = offset_in_page(gpa);
1194 	int ret;
1195 
1196         while ((seg = next_segment(len, offset)) != 0) {
1197 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1198 		if (ret < 0)
1199 			return ret;
1200 		offset = 0;
1201 		len -= seg;
1202 		++gfn;
1203 	}
1204 	return 0;
1205 }
1206 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1207 
1208 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1209 {
1210 	struct kvm_memory_slot *memslot;
1211 
1212 	gfn = unalias_gfn(kvm, gfn);
1213 	memslot = __gfn_to_memslot(kvm, gfn);
1214 	if (memslot && memslot->dirty_bitmap) {
1215 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1216 
1217 		/* avoid RMW */
1218 		if (!test_bit(rel_gfn, memslot->dirty_bitmap))
1219 			set_bit(rel_gfn, memslot->dirty_bitmap);
1220 	}
1221 }
1222 
1223 /*
1224  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1225  */
1226 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1227 {
1228 	DEFINE_WAIT(wait);
1229 
1230 	for (;;) {
1231 		prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1232 
1233 		if (kvm_cpu_has_interrupt(vcpu) ||
1234 		    kvm_cpu_has_pending_timer(vcpu) ||
1235 		    kvm_arch_vcpu_runnable(vcpu)) {
1236 			set_bit(KVM_REQ_UNHALT, &vcpu->requests);
1237 			break;
1238 		}
1239 		if (signal_pending(current))
1240 			break;
1241 
1242 		vcpu_put(vcpu);
1243 		schedule();
1244 		vcpu_load(vcpu);
1245 	}
1246 
1247 	finish_wait(&vcpu->wq, &wait);
1248 }
1249 
1250 void kvm_resched(struct kvm_vcpu *vcpu)
1251 {
1252 	if (!need_resched())
1253 		return;
1254 	cond_resched();
1255 }
1256 EXPORT_SYMBOL_GPL(kvm_resched);
1257 
1258 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1259 {
1260 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1261 	struct page *page;
1262 
1263 	if (vmf->pgoff == 0)
1264 		page = virt_to_page(vcpu->run);
1265 #ifdef CONFIG_X86
1266 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1267 		page = virt_to_page(vcpu->arch.pio_data);
1268 #endif
1269 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1270 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1271 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1272 #endif
1273 	else
1274 		return VM_FAULT_SIGBUS;
1275 	get_page(page);
1276 	vmf->page = page;
1277 	return 0;
1278 }
1279 
1280 static struct vm_operations_struct kvm_vcpu_vm_ops = {
1281 	.fault = kvm_vcpu_fault,
1282 };
1283 
1284 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1285 {
1286 	vma->vm_ops = &kvm_vcpu_vm_ops;
1287 	return 0;
1288 }
1289 
1290 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1291 {
1292 	struct kvm_vcpu *vcpu = filp->private_data;
1293 
1294 	kvm_put_kvm(vcpu->kvm);
1295 	return 0;
1296 }
1297 
1298 static const struct file_operations kvm_vcpu_fops = {
1299 	.release        = kvm_vcpu_release,
1300 	.unlocked_ioctl = kvm_vcpu_ioctl,
1301 	.compat_ioctl   = kvm_vcpu_ioctl,
1302 	.mmap           = kvm_vcpu_mmap,
1303 };
1304 
1305 /*
1306  * Allocates an inode for the vcpu.
1307  */
1308 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1309 {
1310 	int fd = anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, 0);
1311 	if (fd < 0)
1312 		kvm_put_kvm(vcpu->kvm);
1313 	return fd;
1314 }
1315 
1316 /*
1317  * Creates some virtual cpus.  Good luck creating more than one.
1318  */
1319 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, int n)
1320 {
1321 	int r;
1322 	struct kvm_vcpu *vcpu;
1323 
1324 	if (!valid_vcpu(n))
1325 		return -EINVAL;
1326 
1327 	vcpu = kvm_arch_vcpu_create(kvm, n);
1328 	if (IS_ERR(vcpu))
1329 		return PTR_ERR(vcpu);
1330 
1331 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1332 
1333 	r = kvm_arch_vcpu_setup(vcpu);
1334 	if (r)
1335 		return r;
1336 
1337 	mutex_lock(&kvm->lock);
1338 	if (kvm->vcpus[n]) {
1339 		r = -EEXIST;
1340 		goto vcpu_destroy;
1341 	}
1342 	kvm->vcpus[n] = vcpu;
1343 	mutex_unlock(&kvm->lock);
1344 
1345 	/* Now it's all set up, let userspace reach it */
1346 	kvm_get_kvm(kvm);
1347 	r = create_vcpu_fd(vcpu);
1348 	if (r < 0)
1349 		goto unlink;
1350 	return r;
1351 
1352 unlink:
1353 	mutex_lock(&kvm->lock);
1354 	kvm->vcpus[n] = NULL;
1355 vcpu_destroy:
1356 	mutex_unlock(&kvm->lock);
1357 	kvm_arch_vcpu_destroy(vcpu);
1358 	return r;
1359 }
1360 
1361 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1362 {
1363 	if (sigset) {
1364 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1365 		vcpu->sigset_active = 1;
1366 		vcpu->sigset = *sigset;
1367 	} else
1368 		vcpu->sigset_active = 0;
1369 	return 0;
1370 }
1371 
1372 static long kvm_vcpu_ioctl(struct file *filp,
1373 			   unsigned int ioctl, unsigned long arg)
1374 {
1375 	struct kvm_vcpu *vcpu = filp->private_data;
1376 	void __user *argp = (void __user *)arg;
1377 	int r;
1378 	struct kvm_fpu *fpu = NULL;
1379 	struct kvm_sregs *kvm_sregs = NULL;
1380 
1381 	if (vcpu->kvm->mm != current->mm)
1382 		return -EIO;
1383 	switch (ioctl) {
1384 	case KVM_RUN:
1385 		r = -EINVAL;
1386 		if (arg)
1387 			goto out;
1388 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1389 		break;
1390 	case KVM_GET_REGS: {
1391 		struct kvm_regs *kvm_regs;
1392 
1393 		r = -ENOMEM;
1394 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1395 		if (!kvm_regs)
1396 			goto out;
1397 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1398 		if (r)
1399 			goto out_free1;
1400 		r = -EFAULT;
1401 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1402 			goto out_free1;
1403 		r = 0;
1404 out_free1:
1405 		kfree(kvm_regs);
1406 		break;
1407 	}
1408 	case KVM_SET_REGS: {
1409 		struct kvm_regs *kvm_regs;
1410 
1411 		r = -ENOMEM;
1412 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1413 		if (!kvm_regs)
1414 			goto out;
1415 		r = -EFAULT;
1416 		if (copy_from_user(kvm_regs, argp, sizeof(struct kvm_regs)))
1417 			goto out_free2;
1418 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1419 		if (r)
1420 			goto out_free2;
1421 		r = 0;
1422 out_free2:
1423 		kfree(kvm_regs);
1424 		break;
1425 	}
1426 	case KVM_GET_SREGS: {
1427 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1428 		r = -ENOMEM;
1429 		if (!kvm_sregs)
1430 			goto out;
1431 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1432 		if (r)
1433 			goto out;
1434 		r = -EFAULT;
1435 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1436 			goto out;
1437 		r = 0;
1438 		break;
1439 	}
1440 	case KVM_SET_SREGS: {
1441 		kvm_sregs = kmalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1442 		r = -ENOMEM;
1443 		if (!kvm_sregs)
1444 			goto out;
1445 		r = -EFAULT;
1446 		if (copy_from_user(kvm_sregs, argp, sizeof(struct kvm_sregs)))
1447 			goto out;
1448 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1449 		if (r)
1450 			goto out;
1451 		r = 0;
1452 		break;
1453 	}
1454 	case KVM_GET_MP_STATE: {
1455 		struct kvm_mp_state mp_state;
1456 
1457 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1458 		if (r)
1459 			goto out;
1460 		r = -EFAULT;
1461 		if (copy_to_user(argp, &mp_state, sizeof mp_state))
1462 			goto out;
1463 		r = 0;
1464 		break;
1465 	}
1466 	case KVM_SET_MP_STATE: {
1467 		struct kvm_mp_state mp_state;
1468 
1469 		r = -EFAULT;
1470 		if (copy_from_user(&mp_state, argp, sizeof mp_state))
1471 			goto out;
1472 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1473 		if (r)
1474 			goto out;
1475 		r = 0;
1476 		break;
1477 	}
1478 	case KVM_TRANSLATE: {
1479 		struct kvm_translation tr;
1480 
1481 		r = -EFAULT;
1482 		if (copy_from_user(&tr, argp, sizeof tr))
1483 			goto out;
1484 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1485 		if (r)
1486 			goto out;
1487 		r = -EFAULT;
1488 		if (copy_to_user(argp, &tr, sizeof tr))
1489 			goto out;
1490 		r = 0;
1491 		break;
1492 	}
1493 	case KVM_DEBUG_GUEST: {
1494 		struct kvm_debug_guest dbg;
1495 
1496 		r = -EFAULT;
1497 		if (copy_from_user(&dbg, argp, sizeof dbg))
1498 			goto out;
1499 		r = kvm_arch_vcpu_ioctl_debug_guest(vcpu, &dbg);
1500 		if (r)
1501 			goto out;
1502 		r = 0;
1503 		break;
1504 	}
1505 	case KVM_SET_SIGNAL_MASK: {
1506 		struct kvm_signal_mask __user *sigmask_arg = argp;
1507 		struct kvm_signal_mask kvm_sigmask;
1508 		sigset_t sigset, *p;
1509 
1510 		p = NULL;
1511 		if (argp) {
1512 			r = -EFAULT;
1513 			if (copy_from_user(&kvm_sigmask, argp,
1514 					   sizeof kvm_sigmask))
1515 				goto out;
1516 			r = -EINVAL;
1517 			if (kvm_sigmask.len != sizeof sigset)
1518 				goto out;
1519 			r = -EFAULT;
1520 			if (copy_from_user(&sigset, sigmask_arg->sigset,
1521 					   sizeof sigset))
1522 				goto out;
1523 			p = &sigset;
1524 		}
1525 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1526 		break;
1527 	}
1528 	case KVM_GET_FPU: {
1529 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1530 		r = -ENOMEM;
1531 		if (!fpu)
1532 			goto out;
1533 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1534 		if (r)
1535 			goto out;
1536 		r = -EFAULT;
1537 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1538 			goto out;
1539 		r = 0;
1540 		break;
1541 	}
1542 	case KVM_SET_FPU: {
1543 		fpu = kmalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1544 		r = -ENOMEM;
1545 		if (!fpu)
1546 			goto out;
1547 		r = -EFAULT;
1548 		if (copy_from_user(fpu, argp, sizeof(struct kvm_fpu)))
1549 			goto out;
1550 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1551 		if (r)
1552 			goto out;
1553 		r = 0;
1554 		break;
1555 	}
1556 	default:
1557 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1558 	}
1559 out:
1560 	kfree(fpu);
1561 	kfree(kvm_sregs);
1562 	return r;
1563 }
1564 
1565 static long kvm_vm_ioctl(struct file *filp,
1566 			   unsigned int ioctl, unsigned long arg)
1567 {
1568 	struct kvm *kvm = filp->private_data;
1569 	void __user *argp = (void __user *)arg;
1570 	int r;
1571 
1572 	if (kvm->mm != current->mm)
1573 		return -EIO;
1574 	switch (ioctl) {
1575 	case KVM_CREATE_VCPU:
1576 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1577 		if (r < 0)
1578 			goto out;
1579 		break;
1580 	case KVM_SET_USER_MEMORY_REGION: {
1581 		struct kvm_userspace_memory_region kvm_userspace_mem;
1582 
1583 		r = -EFAULT;
1584 		if (copy_from_user(&kvm_userspace_mem, argp,
1585 						sizeof kvm_userspace_mem))
1586 			goto out;
1587 
1588 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1589 		if (r)
1590 			goto out;
1591 		break;
1592 	}
1593 	case KVM_GET_DIRTY_LOG: {
1594 		struct kvm_dirty_log log;
1595 
1596 		r = -EFAULT;
1597 		if (copy_from_user(&log, argp, sizeof log))
1598 			goto out;
1599 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1600 		if (r)
1601 			goto out;
1602 		break;
1603 	}
1604 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1605 	case KVM_REGISTER_COALESCED_MMIO: {
1606 		struct kvm_coalesced_mmio_zone zone;
1607 		r = -EFAULT;
1608 		if (copy_from_user(&zone, argp, sizeof zone))
1609 			goto out;
1610 		r = -ENXIO;
1611 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1612 		if (r)
1613 			goto out;
1614 		r = 0;
1615 		break;
1616 	}
1617 	case KVM_UNREGISTER_COALESCED_MMIO: {
1618 		struct kvm_coalesced_mmio_zone zone;
1619 		r = -EFAULT;
1620 		if (copy_from_user(&zone, argp, sizeof zone))
1621 			goto out;
1622 		r = -ENXIO;
1623 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
1624 		if (r)
1625 			goto out;
1626 		r = 0;
1627 		break;
1628 	}
1629 #endif
1630 #ifdef KVM_CAP_DEVICE_ASSIGNMENT
1631 	case KVM_ASSIGN_PCI_DEVICE: {
1632 		struct kvm_assigned_pci_dev assigned_dev;
1633 
1634 		r = -EFAULT;
1635 		if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
1636 			goto out;
1637 		r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
1638 		if (r)
1639 			goto out;
1640 		break;
1641 	}
1642 	case KVM_ASSIGN_IRQ: {
1643 		struct kvm_assigned_irq assigned_irq;
1644 
1645 		r = -EFAULT;
1646 		if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
1647 			goto out;
1648 		r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
1649 		if (r)
1650 			goto out;
1651 		break;
1652 	}
1653 #endif
1654 	default:
1655 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1656 	}
1657 out:
1658 	return r;
1659 }
1660 
1661 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1662 {
1663 	struct page *page[1];
1664 	unsigned long addr;
1665 	int npages;
1666 	gfn_t gfn = vmf->pgoff;
1667 	struct kvm *kvm = vma->vm_file->private_data;
1668 
1669 	addr = gfn_to_hva(kvm, gfn);
1670 	if (kvm_is_error_hva(addr))
1671 		return VM_FAULT_SIGBUS;
1672 
1673 	npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
1674 				NULL);
1675 	if (unlikely(npages != 1))
1676 		return VM_FAULT_SIGBUS;
1677 
1678 	vmf->page = page[0];
1679 	return 0;
1680 }
1681 
1682 static struct vm_operations_struct kvm_vm_vm_ops = {
1683 	.fault = kvm_vm_fault,
1684 };
1685 
1686 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
1687 {
1688 	vma->vm_ops = &kvm_vm_vm_ops;
1689 	return 0;
1690 }
1691 
1692 static const struct file_operations kvm_vm_fops = {
1693 	.release        = kvm_vm_release,
1694 	.unlocked_ioctl = kvm_vm_ioctl,
1695 	.compat_ioctl   = kvm_vm_ioctl,
1696 	.mmap           = kvm_vm_mmap,
1697 };
1698 
1699 static int kvm_dev_ioctl_create_vm(void)
1700 {
1701 	int fd;
1702 	struct kvm *kvm;
1703 
1704 	kvm = kvm_create_vm();
1705 	if (IS_ERR(kvm))
1706 		return PTR_ERR(kvm);
1707 	fd = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, 0);
1708 	if (fd < 0)
1709 		kvm_put_kvm(kvm);
1710 
1711 	return fd;
1712 }
1713 
1714 static long kvm_dev_ioctl(struct file *filp,
1715 			  unsigned int ioctl, unsigned long arg)
1716 {
1717 	long r = -EINVAL;
1718 
1719 	switch (ioctl) {
1720 	case KVM_GET_API_VERSION:
1721 		r = -EINVAL;
1722 		if (arg)
1723 			goto out;
1724 		r = KVM_API_VERSION;
1725 		break;
1726 	case KVM_CREATE_VM:
1727 		r = -EINVAL;
1728 		if (arg)
1729 			goto out;
1730 		r = kvm_dev_ioctl_create_vm();
1731 		break;
1732 	case KVM_CHECK_EXTENSION:
1733 		r = kvm_dev_ioctl_check_extension(arg);
1734 		break;
1735 	case KVM_GET_VCPU_MMAP_SIZE:
1736 		r = -EINVAL;
1737 		if (arg)
1738 			goto out;
1739 		r = PAGE_SIZE;     /* struct kvm_run */
1740 #ifdef CONFIG_X86
1741 		r += PAGE_SIZE;    /* pio data page */
1742 #endif
1743 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1744 		r += PAGE_SIZE;    /* coalesced mmio ring page */
1745 #endif
1746 		break;
1747 	case KVM_TRACE_ENABLE:
1748 	case KVM_TRACE_PAUSE:
1749 	case KVM_TRACE_DISABLE:
1750 		r = kvm_trace_ioctl(ioctl, arg);
1751 		break;
1752 	default:
1753 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
1754 	}
1755 out:
1756 	return r;
1757 }
1758 
1759 static struct file_operations kvm_chardev_ops = {
1760 	.unlocked_ioctl = kvm_dev_ioctl,
1761 	.compat_ioctl   = kvm_dev_ioctl,
1762 };
1763 
1764 static struct miscdevice kvm_dev = {
1765 	KVM_MINOR,
1766 	"kvm",
1767 	&kvm_chardev_ops,
1768 };
1769 
1770 static void hardware_enable(void *junk)
1771 {
1772 	int cpu = raw_smp_processor_id();
1773 
1774 	if (cpu_isset(cpu, cpus_hardware_enabled))
1775 		return;
1776 	cpu_set(cpu, cpus_hardware_enabled);
1777 	kvm_arch_hardware_enable(NULL);
1778 }
1779 
1780 static void hardware_disable(void *junk)
1781 {
1782 	int cpu = raw_smp_processor_id();
1783 
1784 	if (!cpu_isset(cpu, cpus_hardware_enabled))
1785 		return;
1786 	cpu_clear(cpu, cpus_hardware_enabled);
1787 	kvm_arch_hardware_disable(NULL);
1788 }
1789 
1790 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
1791 			   void *v)
1792 {
1793 	int cpu = (long)v;
1794 
1795 	val &= ~CPU_TASKS_FROZEN;
1796 	switch (val) {
1797 	case CPU_DYING:
1798 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1799 		       cpu);
1800 		hardware_disable(NULL);
1801 		break;
1802 	case CPU_UP_CANCELED:
1803 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
1804 		       cpu);
1805 		smp_call_function_single(cpu, hardware_disable, NULL, 1);
1806 		break;
1807 	case CPU_ONLINE:
1808 		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
1809 		       cpu);
1810 		smp_call_function_single(cpu, hardware_enable, NULL, 1);
1811 		break;
1812 	}
1813 	return NOTIFY_OK;
1814 }
1815 
1816 
1817 asmlinkage void kvm_handle_fault_on_reboot(void)
1818 {
1819 	if (kvm_rebooting)
1820 		/* spin while reset goes on */
1821 		while (true)
1822 			;
1823 	/* Fault while not rebooting.  We want the trace. */
1824 	BUG();
1825 }
1826 EXPORT_SYMBOL_GPL(kvm_handle_fault_on_reboot);
1827 
1828 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
1829 		      void *v)
1830 {
1831 	if (val == SYS_RESTART) {
1832 		/*
1833 		 * Some (well, at least mine) BIOSes hang on reboot if
1834 		 * in vmx root mode.
1835 		 */
1836 		printk(KERN_INFO "kvm: exiting hardware virtualization\n");
1837 		kvm_rebooting = true;
1838 		on_each_cpu(hardware_disable, NULL, 1);
1839 	}
1840 	return NOTIFY_OK;
1841 }
1842 
1843 static struct notifier_block kvm_reboot_notifier = {
1844 	.notifier_call = kvm_reboot,
1845 	.priority = 0,
1846 };
1847 
1848 void kvm_io_bus_init(struct kvm_io_bus *bus)
1849 {
1850 	memset(bus, 0, sizeof(*bus));
1851 }
1852 
1853 void kvm_io_bus_destroy(struct kvm_io_bus *bus)
1854 {
1855 	int i;
1856 
1857 	for (i = 0; i < bus->dev_count; i++) {
1858 		struct kvm_io_device *pos = bus->devs[i];
1859 
1860 		kvm_iodevice_destructor(pos);
1861 	}
1862 }
1863 
1864 struct kvm_io_device *kvm_io_bus_find_dev(struct kvm_io_bus *bus,
1865 					  gpa_t addr, int len, int is_write)
1866 {
1867 	int i;
1868 
1869 	for (i = 0; i < bus->dev_count; i++) {
1870 		struct kvm_io_device *pos = bus->devs[i];
1871 
1872 		if (pos->in_range(pos, addr, len, is_write))
1873 			return pos;
1874 	}
1875 
1876 	return NULL;
1877 }
1878 
1879 void kvm_io_bus_register_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev)
1880 {
1881 	BUG_ON(bus->dev_count > (NR_IOBUS_DEVS-1));
1882 
1883 	bus->devs[bus->dev_count++] = dev;
1884 }
1885 
1886 static struct notifier_block kvm_cpu_notifier = {
1887 	.notifier_call = kvm_cpu_hotplug,
1888 	.priority = 20, /* must be > scheduler priority */
1889 };
1890 
1891 static int vm_stat_get(void *_offset, u64 *val)
1892 {
1893 	unsigned offset = (long)_offset;
1894 	struct kvm *kvm;
1895 
1896 	*val = 0;
1897 	spin_lock(&kvm_lock);
1898 	list_for_each_entry(kvm, &vm_list, vm_list)
1899 		*val += *(u32 *)((void *)kvm + offset);
1900 	spin_unlock(&kvm_lock);
1901 	return 0;
1902 }
1903 
1904 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
1905 
1906 static int vcpu_stat_get(void *_offset, u64 *val)
1907 {
1908 	unsigned offset = (long)_offset;
1909 	struct kvm *kvm;
1910 	struct kvm_vcpu *vcpu;
1911 	int i;
1912 
1913 	*val = 0;
1914 	spin_lock(&kvm_lock);
1915 	list_for_each_entry(kvm, &vm_list, vm_list)
1916 		for (i = 0; i < KVM_MAX_VCPUS; ++i) {
1917 			vcpu = kvm->vcpus[i];
1918 			if (vcpu)
1919 				*val += *(u32 *)((void *)vcpu + offset);
1920 		}
1921 	spin_unlock(&kvm_lock);
1922 	return 0;
1923 }
1924 
1925 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
1926 
1927 static struct file_operations *stat_fops[] = {
1928 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
1929 	[KVM_STAT_VM]   = &vm_stat_fops,
1930 };
1931 
1932 static void kvm_init_debug(void)
1933 {
1934 	struct kvm_stats_debugfs_item *p;
1935 
1936 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
1937 	for (p = debugfs_entries; p->name; ++p)
1938 		p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
1939 						(void *)(long)p->offset,
1940 						stat_fops[p->kind]);
1941 }
1942 
1943 static void kvm_exit_debug(void)
1944 {
1945 	struct kvm_stats_debugfs_item *p;
1946 
1947 	for (p = debugfs_entries; p->name; ++p)
1948 		debugfs_remove(p->dentry);
1949 	debugfs_remove(kvm_debugfs_dir);
1950 }
1951 
1952 static int kvm_suspend(struct sys_device *dev, pm_message_t state)
1953 {
1954 	hardware_disable(NULL);
1955 	return 0;
1956 }
1957 
1958 static int kvm_resume(struct sys_device *dev)
1959 {
1960 	hardware_enable(NULL);
1961 	return 0;
1962 }
1963 
1964 static struct sysdev_class kvm_sysdev_class = {
1965 	.name = "kvm",
1966 	.suspend = kvm_suspend,
1967 	.resume = kvm_resume,
1968 };
1969 
1970 static struct sys_device kvm_sysdev = {
1971 	.id = 0,
1972 	.cls = &kvm_sysdev_class,
1973 };
1974 
1975 struct page *bad_page;
1976 pfn_t bad_pfn;
1977 
1978 static inline
1979 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
1980 {
1981 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
1982 }
1983 
1984 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
1985 {
1986 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1987 
1988 	kvm_arch_vcpu_load(vcpu, cpu);
1989 }
1990 
1991 static void kvm_sched_out(struct preempt_notifier *pn,
1992 			  struct task_struct *next)
1993 {
1994 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
1995 
1996 	kvm_arch_vcpu_put(vcpu);
1997 }
1998 
1999 int kvm_init(void *opaque, unsigned int vcpu_size,
2000 		  struct module *module)
2001 {
2002 	int r;
2003 	int cpu;
2004 
2005 	kvm_init_debug();
2006 
2007 	r = kvm_arch_init(opaque);
2008 	if (r)
2009 		goto out_fail;
2010 
2011 	bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2012 
2013 	if (bad_page == NULL) {
2014 		r = -ENOMEM;
2015 		goto out;
2016 	}
2017 
2018 	bad_pfn = page_to_pfn(bad_page);
2019 
2020 	r = kvm_arch_hardware_setup();
2021 	if (r < 0)
2022 		goto out_free_0;
2023 
2024 	for_each_online_cpu(cpu) {
2025 		smp_call_function_single(cpu,
2026 				kvm_arch_check_processor_compat,
2027 				&r, 1);
2028 		if (r < 0)
2029 			goto out_free_1;
2030 	}
2031 
2032 	on_each_cpu(hardware_enable, NULL, 1);
2033 	r = register_cpu_notifier(&kvm_cpu_notifier);
2034 	if (r)
2035 		goto out_free_2;
2036 	register_reboot_notifier(&kvm_reboot_notifier);
2037 
2038 	r = sysdev_class_register(&kvm_sysdev_class);
2039 	if (r)
2040 		goto out_free_3;
2041 
2042 	r = sysdev_register(&kvm_sysdev);
2043 	if (r)
2044 		goto out_free_4;
2045 
2046 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
2047 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size,
2048 					   __alignof__(struct kvm_vcpu),
2049 					   0, NULL);
2050 	if (!kvm_vcpu_cache) {
2051 		r = -ENOMEM;
2052 		goto out_free_5;
2053 	}
2054 
2055 	kvm_chardev_ops.owner = module;
2056 
2057 	r = misc_register(&kvm_dev);
2058 	if (r) {
2059 		printk(KERN_ERR "kvm: misc device register failed\n");
2060 		goto out_free;
2061 	}
2062 
2063 	kvm_preempt_ops.sched_in = kvm_sched_in;
2064 	kvm_preempt_ops.sched_out = kvm_sched_out;
2065 
2066 	return 0;
2067 
2068 out_free:
2069 	kmem_cache_destroy(kvm_vcpu_cache);
2070 out_free_5:
2071 	sysdev_unregister(&kvm_sysdev);
2072 out_free_4:
2073 	sysdev_class_unregister(&kvm_sysdev_class);
2074 out_free_3:
2075 	unregister_reboot_notifier(&kvm_reboot_notifier);
2076 	unregister_cpu_notifier(&kvm_cpu_notifier);
2077 out_free_2:
2078 	on_each_cpu(hardware_disable, NULL, 1);
2079 out_free_1:
2080 	kvm_arch_hardware_unsetup();
2081 out_free_0:
2082 	__free_page(bad_page);
2083 out:
2084 	kvm_arch_exit();
2085 	kvm_exit_debug();
2086 out_fail:
2087 	return r;
2088 }
2089 EXPORT_SYMBOL_GPL(kvm_init);
2090 
2091 void kvm_exit(void)
2092 {
2093 	kvm_trace_cleanup();
2094 	misc_deregister(&kvm_dev);
2095 	kmem_cache_destroy(kvm_vcpu_cache);
2096 	sysdev_unregister(&kvm_sysdev);
2097 	sysdev_class_unregister(&kvm_sysdev_class);
2098 	unregister_reboot_notifier(&kvm_reboot_notifier);
2099 	unregister_cpu_notifier(&kvm_cpu_notifier);
2100 	on_each_cpu(hardware_disable, NULL, 1);
2101 	kvm_arch_hardware_unsetup();
2102 	kvm_arch_exit();
2103 	kvm_exit_debug();
2104 	__free_page(bad_page);
2105 }
2106 EXPORT_SYMBOL_GPL(kvm_exit);
2107