xref: /openbmc/linux/virt/kvm/kvm_main.c (revision 63dc02bd)
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  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18 
19 #include "iodev.h"
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.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/syscore_ops.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 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52 
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/uaccess.h>
56 #include <asm/pgtable.h>
57 
58 #include "coalesced_mmio.h"
59 #include "async_pf.h"
60 
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/kvm.h>
63 
64 MODULE_AUTHOR("Qumranet");
65 MODULE_LICENSE("GPL");
66 
67 /*
68  * Ordering of locks:
69  *
70  * 		kvm->lock --> kvm->slots_lock --> kvm->irq_lock
71  */
72 
73 DEFINE_RAW_SPINLOCK(kvm_lock);
74 LIST_HEAD(vm_list);
75 
76 static cpumask_var_t cpus_hardware_enabled;
77 static int kvm_usage_count = 0;
78 static atomic_t hardware_enable_failed;
79 
80 struct kmem_cache *kvm_vcpu_cache;
81 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
82 
83 static __read_mostly struct preempt_ops kvm_preempt_ops;
84 
85 struct dentry *kvm_debugfs_dir;
86 
87 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 			   unsigned long arg);
89 #ifdef CONFIG_COMPAT
90 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
91 				  unsigned long arg);
92 #endif
93 static int hardware_enable_all(void);
94 static void hardware_disable_all(void);
95 
96 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
97 
98 bool kvm_rebooting;
99 EXPORT_SYMBOL_GPL(kvm_rebooting);
100 
101 static bool largepages_enabled = true;
102 
103 static struct page *hwpoison_page;
104 static pfn_t hwpoison_pfn;
105 
106 struct page *fault_page;
107 pfn_t fault_pfn;
108 
109 inline int kvm_is_mmio_pfn(pfn_t pfn)
110 {
111 	if (pfn_valid(pfn)) {
112 		int reserved;
113 		struct page *tail = pfn_to_page(pfn);
114 		struct page *head = compound_trans_head(tail);
115 		reserved = PageReserved(head);
116 		if (head != tail) {
117 			/*
118 			 * "head" is not a dangling pointer
119 			 * (compound_trans_head takes care of that)
120 			 * but the hugepage may have been splitted
121 			 * from under us (and we may not hold a
122 			 * reference count on the head page so it can
123 			 * be reused before we run PageReferenced), so
124 			 * we've to check PageTail before returning
125 			 * what we just read.
126 			 */
127 			smp_rmb();
128 			if (PageTail(tail))
129 				return reserved;
130 		}
131 		return PageReserved(tail);
132 	}
133 
134 	return true;
135 }
136 
137 /*
138  * Switches to specified vcpu, until a matching vcpu_put()
139  */
140 void vcpu_load(struct kvm_vcpu *vcpu)
141 {
142 	int cpu;
143 
144 	mutex_lock(&vcpu->mutex);
145 	if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
146 		/* The thread running this VCPU changed. */
147 		struct pid *oldpid = vcpu->pid;
148 		struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
149 		rcu_assign_pointer(vcpu->pid, newpid);
150 		synchronize_rcu();
151 		put_pid(oldpid);
152 	}
153 	cpu = get_cpu();
154 	preempt_notifier_register(&vcpu->preempt_notifier);
155 	kvm_arch_vcpu_load(vcpu, cpu);
156 	put_cpu();
157 }
158 
159 void vcpu_put(struct kvm_vcpu *vcpu)
160 {
161 	preempt_disable();
162 	kvm_arch_vcpu_put(vcpu);
163 	preempt_notifier_unregister(&vcpu->preempt_notifier);
164 	preempt_enable();
165 	mutex_unlock(&vcpu->mutex);
166 }
167 
168 static void ack_flush(void *_completed)
169 {
170 }
171 
172 static bool make_all_cpus_request(struct kvm *kvm, unsigned int req)
173 {
174 	int i, cpu, me;
175 	cpumask_var_t cpus;
176 	bool called = true;
177 	struct kvm_vcpu *vcpu;
178 
179 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
180 
181 	me = get_cpu();
182 	kvm_for_each_vcpu(i, vcpu, kvm) {
183 		kvm_make_request(req, vcpu);
184 		cpu = vcpu->cpu;
185 
186 		/* Set ->requests bit before we read ->mode */
187 		smp_mb();
188 
189 		if (cpus != NULL && cpu != -1 && cpu != me &&
190 		      kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
191 			cpumask_set_cpu(cpu, cpus);
192 	}
193 	if (unlikely(cpus == NULL))
194 		smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
195 	else if (!cpumask_empty(cpus))
196 		smp_call_function_many(cpus, ack_flush, NULL, 1);
197 	else
198 		called = false;
199 	put_cpu();
200 	free_cpumask_var(cpus);
201 	return called;
202 }
203 
204 void kvm_flush_remote_tlbs(struct kvm *kvm)
205 {
206 	long dirty_count = kvm->tlbs_dirty;
207 
208 	smp_mb();
209 	if (make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
210 		++kvm->stat.remote_tlb_flush;
211 	cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
212 }
213 
214 void kvm_reload_remote_mmus(struct kvm *kvm)
215 {
216 	make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
217 }
218 
219 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
220 {
221 	struct page *page;
222 	int r;
223 
224 	mutex_init(&vcpu->mutex);
225 	vcpu->cpu = -1;
226 	vcpu->kvm = kvm;
227 	vcpu->vcpu_id = id;
228 	vcpu->pid = NULL;
229 	init_waitqueue_head(&vcpu->wq);
230 	kvm_async_pf_vcpu_init(vcpu);
231 
232 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
233 	if (!page) {
234 		r = -ENOMEM;
235 		goto fail;
236 	}
237 	vcpu->run = page_address(page);
238 
239 	r = kvm_arch_vcpu_init(vcpu);
240 	if (r < 0)
241 		goto fail_free_run;
242 	return 0;
243 
244 fail_free_run:
245 	free_page((unsigned long)vcpu->run);
246 fail:
247 	return r;
248 }
249 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
250 
251 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
252 {
253 	put_pid(vcpu->pid);
254 	kvm_arch_vcpu_uninit(vcpu);
255 	free_page((unsigned long)vcpu->run);
256 }
257 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
258 
259 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
260 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
261 {
262 	return container_of(mn, struct kvm, mmu_notifier);
263 }
264 
265 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
266 					     struct mm_struct *mm,
267 					     unsigned long address)
268 {
269 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
270 	int need_tlb_flush, idx;
271 
272 	/*
273 	 * When ->invalidate_page runs, the linux pte has been zapped
274 	 * already but the page is still allocated until
275 	 * ->invalidate_page returns. So if we increase the sequence
276 	 * here the kvm page fault will notice if the spte can't be
277 	 * established because the page is going to be freed. If
278 	 * instead the kvm page fault establishes the spte before
279 	 * ->invalidate_page runs, kvm_unmap_hva will release it
280 	 * before returning.
281 	 *
282 	 * The sequence increase only need to be seen at spin_unlock
283 	 * time, and not at spin_lock time.
284 	 *
285 	 * Increasing the sequence after the spin_unlock would be
286 	 * unsafe because the kvm page fault could then establish the
287 	 * pte after kvm_unmap_hva returned, without noticing the page
288 	 * is going to be freed.
289 	 */
290 	idx = srcu_read_lock(&kvm->srcu);
291 	spin_lock(&kvm->mmu_lock);
292 
293 	kvm->mmu_notifier_seq++;
294 	need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
295 	/* we've to flush the tlb before the pages can be freed */
296 	if (need_tlb_flush)
297 		kvm_flush_remote_tlbs(kvm);
298 
299 	spin_unlock(&kvm->mmu_lock);
300 	srcu_read_unlock(&kvm->srcu, idx);
301 }
302 
303 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
304 					struct mm_struct *mm,
305 					unsigned long address,
306 					pte_t pte)
307 {
308 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
309 	int idx;
310 
311 	idx = srcu_read_lock(&kvm->srcu);
312 	spin_lock(&kvm->mmu_lock);
313 	kvm->mmu_notifier_seq++;
314 	kvm_set_spte_hva(kvm, address, pte);
315 	spin_unlock(&kvm->mmu_lock);
316 	srcu_read_unlock(&kvm->srcu, idx);
317 }
318 
319 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
320 						    struct mm_struct *mm,
321 						    unsigned long start,
322 						    unsigned long end)
323 {
324 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
325 	int need_tlb_flush = 0, idx;
326 
327 	idx = srcu_read_lock(&kvm->srcu);
328 	spin_lock(&kvm->mmu_lock);
329 	/*
330 	 * The count increase must become visible at unlock time as no
331 	 * spte can be established without taking the mmu_lock and
332 	 * count is also read inside the mmu_lock critical section.
333 	 */
334 	kvm->mmu_notifier_count++;
335 	for (; start < end; start += PAGE_SIZE)
336 		need_tlb_flush |= kvm_unmap_hva(kvm, start);
337 	need_tlb_flush |= kvm->tlbs_dirty;
338 	/* we've to flush the tlb before the pages can be freed */
339 	if (need_tlb_flush)
340 		kvm_flush_remote_tlbs(kvm);
341 
342 	spin_unlock(&kvm->mmu_lock);
343 	srcu_read_unlock(&kvm->srcu, idx);
344 }
345 
346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
347 						  struct mm_struct *mm,
348 						  unsigned long start,
349 						  unsigned long end)
350 {
351 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
352 
353 	spin_lock(&kvm->mmu_lock);
354 	/*
355 	 * This sequence increase will notify the kvm page fault that
356 	 * the page that is going to be mapped in the spte could have
357 	 * been freed.
358 	 */
359 	kvm->mmu_notifier_seq++;
360 	smp_wmb();
361 	/*
362 	 * The above sequence increase must be visible before the
363 	 * below count decrease, which is ensured by the smp_wmb above
364 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
365 	 */
366 	kvm->mmu_notifier_count--;
367 	spin_unlock(&kvm->mmu_lock);
368 
369 	BUG_ON(kvm->mmu_notifier_count < 0);
370 }
371 
372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
373 					      struct mm_struct *mm,
374 					      unsigned long address)
375 {
376 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
377 	int young, idx;
378 
379 	idx = srcu_read_lock(&kvm->srcu);
380 	spin_lock(&kvm->mmu_lock);
381 
382 	young = kvm_age_hva(kvm, address);
383 	if (young)
384 		kvm_flush_remote_tlbs(kvm);
385 
386 	spin_unlock(&kvm->mmu_lock);
387 	srcu_read_unlock(&kvm->srcu, idx);
388 
389 	return young;
390 }
391 
392 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
393 				       struct mm_struct *mm,
394 				       unsigned long address)
395 {
396 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
397 	int young, idx;
398 
399 	idx = srcu_read_lock(&kvm->srcu);
400 	spin_lock(&kvm->mmu_lock);
401 	young = kvm_test_age_hva(kvm, address);
402 	spin_unlock(&kvm->mmu_lock);
403 	srcu_read_unlock(&kvm->srcu, idx);
404 
405 	return young;
406 }
407 
408 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
409 				     struct mm_struct *mm)
410 {
411 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 	int idx;
413 
414 	idx = srcu_read_lock(&kvm->srcu);
415 	kvm_arch_flush_shadow(kvm);
416 	srcu_read_unlock(&kvm->srcu, idx);
417 }
418 
419 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
420 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
421 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
422 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
423 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
424 	.test_young		= kvm_mmu_notifier_test_young,
425 	.change_pte		= kvm_mmu_notifier_change_pte,
426 	.release		= kvm_mmu_notifier_release,
427 };
428 
429 static int kvm_init_mmu_notifier(struct kvm *kvm)
430 {
431 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
432 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
433 }
434 
435 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
436 
437 static int kvm_init_mmu_notifier(struct kvm *kvm)
438 {
439 	return 0;
440 }
441 
442 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
443 
444 static void kvm_init_memslots_id(struct kvm *kvm)
445 {
446 	int i;
447 	struct kvm_memslots *slots = kvm->memslots;
448 
449 	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
450 		slots->id_to_index[i] = slots->memslots[i].id = i;
451 }
452 
453 static struct kvm *kvm_create_vm(unsigned long type)
454 {
455 	int r, i;
456 	struct kvm *kvm = kvm_arch_alloc_vm();
457 
458 	if (!kvm)
459 		return ERR_PTR(-ENOMEM);
460 
461 	r = kvm_arch_init_vm(kvm, type);
462 	if (r)
463 		goto out_err_nodisable;
464 
465 	r = hardware_enable_all();
466 	if (r)
467 		goto out_err_nodisable;
468 
469 #ifdef CONFIG_HAVE_KVM_IRQCHIP
470 	INIT_HLIST_HEAD(&kvm->mask_notifier_list);
471 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
472 #endif
473 
474 	r = -ENOMEM;
475 	kvm->memslots = kzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
476 	if (!kvm->memslots)
477 		goto out_err_nosrcu;
478 	kvm_init_memslots_id(kvm);
479 	if (init_srcu_struct(&kvm->srcu))
480 		goto out_err_nosrcu;
481 	for (i = 0; i < KVM_NR_BUSES; i++) {
482 		kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
483 					GFP_KERNEL);
484 		if (!kvm->buses[i])
485 			goto out_err;
486 	}
487 
488 	spin_lock_init(&kvm->mmu_lock);
489 	kvm->mm = current->mm;
490 	atomic_inc(&kvm->mm->mm_count);
491 	kvm_eventfd_init(kvm);
492 	mutex_init(&kvm->lock);
493 	mutex_init(&kvm->irq_lock);
494 	mutex_init(&kvm->slots_lock);
495 	atomic_set(&kvm->users_count, 1);
496 
497 	r = kvm_init_mmu_notifier(kvm);
498 	if (r)
499 		goto out_err;
500 
501 	raw_spin_lock(&kvm_lock);
502 	list_add(&kvm->vm_list, &vm_list);
503 	raw_spin_unlock(&kvm_lock);
504 
505 	return kvm;
506 
507 out_err:
508 	cleanup_srcu_struct(&kvm->srcu);
509 out_err_nosrcu:
510 	hardware_disable_all();
511 out_err_nodisable:
512 	for (i = 0; i < KVM_NR_BUSES; i++)
513 		kfree(kvm->buses[i]);
514 	kfree(kvm->memslots);
515 	kvm_arch_free_vm(kvm);
516 	return ERR_PTR(r);
517 }
518 
519 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
520 {
521 	if (!memslot->dirty_bitmap)
522 		return;
523 
524 	if (2 * kvm_dirty_bitmap_bytes(memslot) > PAGE_SIZE)
525 		vfree(memslot->dirty_bitmap_head);
526 	else
527 		kfree(memslot->dirty_bitmap_head);
528 
529 	memslot->dirty_bitmap = NULL;
530 	memslot->dirty_bitmap_head = NULL;
531 }
532 
533 /*
534  * Free any memory in @free but not in @dont.
535  */
536 static void kvm_free_physmem_slot(struct kvm_memory_slot *free,
537 				  struct kvm_memory_slot *dont)
538 {
539 	if (!dont || free->rmap != dont->rmap)
540 		vfree(free->rmap);
541 
542 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
543 		kvm_destroy_dirty_bitmap(free);
544 
545 	kvm_arch_free_memslot(free, dont);
546 
547 	free->npages = 0;
548 	free->rmap = NULL;
549 }
550 
551 void kvm_free_physmem(struct kvm *kvm)
552 {
553 	struct kvm_memslots *slots = kvm->memslots;
554 	struct kvm_memory_slot *memslot;
555 
556 	kvm_for_each_memslot(memslot, slots)
557 		kvm_free_physmem_slot(memslot, NULL);
558 
559 	kfree(kvm->memslots);
560 }
561 
562 static void kvm_destroy_vm(struct kvm *kvm)
563 {
564 	int i;
565 	struct mm_struct *mm = kvm->mm;
566 
567 	kvm_arch_sync_events(kvm);
568 	raw_spin_lock(&kvm_lock);
569 	list_del(&kvm->vm_list);
570 	raw_spin_unlock(&kvm_lock);
571 	kvm_free_irq_routing(kvm);
572 	for (i = 0; i < KVM_NR_BUSES; i++)
573 		kvm_io_bus_destroy(kvm->buses[i]);
574 	kvm_coalesced_mmio_free(kvm);
575 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
576 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
577 #else
578 	kvm_arch_flush_shadow(kvm);
579 #endif
580 	kvm_arch_destroy_vm(kvm);
581 	kvm_free_physmem(kvm);
582 	cleanup_srcu_struct(&kvm->srcu);
583 	kvm_arch_free_vm(kvm);
584 	hardware_disable_all();
585 	mmdrop(mm);
586 }
587 
588 void kvm_get_kvm(struct kvm *kvm)
589 {
590 	atomic_inc(&kvm->users_count);
591 }
592 EXPORT_SYMBOL_GPL(kvm_get_kvm);
593 
594 void kvm_put_kvm(struct kvm *kvm)
595 {
596 	if (atomic_dec_and_test(&kvm->users_count))
597 		kvm_destroy_vm(kvm);
598 }
599 EXPORT_SYMBOL_GPL(kvm_put_kvm);
600 
601 
602 static int kvm_vm_release(struct inode *inode, struct file *filp)
603 {
604 	struct kvm *kvm = filp->private_data;
605 
606 	kvm_irqfd_release(kvm);
607 
608 	kvm_put_kvm(kvm);
609 	return 0;
610 }
611 
612 /*
613  * Allocation size is twice as large as the actual dirty bitmap size.
614  * This makes it possible to do double buffering: see x86's
615  * kvm_vm_ioctl_get_dirty_log().
616  */
617 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
618 {
619 #ifndef CONFIG_S390
620 	unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
621 
622 	if (dirty_bytes > PAGE_SIZE)
623 		memslot->dirty_bitmap = vzalloc(dirty_bytes);
624 	else
625 		memslot->dirty_bitmap = kzalloc(dirty_bytes, GFP_KERNEL);
626 
627 	if (!memslot->dirty_bitmap)
628 		return -ENOMEM;
629 
630 	memslot->dirty_bitmap_head = memslot->dirty_bitmap;
631 	memslot->nr_dirty_pages = 0;
632 #endif /* !CONFIG_S390 */
633 	return 0;
634 }
635 
636 static int cmp_memslot(const void *slot1, const void *slot2)
637 {
638 	struct kvm_memory_slot *s1, *s2;
639 
640 	s1 = (struct kvm_memory_slot *)slot1;
641 	s2 = (struct kvm_memory_slot *)slot2;
642 
643 	if (s1->npages < s2->npages)
644 		return 1;
645 	if (s1->npages > s2->npages)
646 		return -1;
647 
648 	return 0;
649 }
650 
651 /*
652  * Sort the memslots base on its size, so the larger slots
653  * will get better fit.
654  */
655 static void sort_memslots(struct kvm_memslots *slots)
656 {
657 	int i;
658 
659 	sort(slots->memslots, KVM_MEM_SLOTS_NUM,
660 	      sizeof(struct kvm_memory_slot), cmp_memslot, NULL);
661 
662 	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
663 		slots->id_to_index[slots->memslots[i].id] = i;
664 }
665 
666 void update_memslots(struct kvm_memslots *slots, struct kvm_memory_slot *new)
667 {
668 	if (new) {
669 		int id = new->id;
670 		struct kvm_memory_slot *old = id_to_memslot(slots, id);
671 		unsigned long npages = old->npages;
672 
673 		*old = *new;
674 		if (new->npages != npages)
675 			sort_memslots(slots);
676 	}
677 
678 	slots->generation++;
679 }
680 
681 /*
682  * Allocate some memory and give it an address in the guest physical address
683  * space.
684  *
685  * Discontiguous memory is allowed, mostly for framebuffers.
686  *
687  * Must be called holding mmap_sem for write.
688  */
689 int __kvm_set_memory_region(struct kvm *kvm,
690 			    struct kvm_userspace_memory_region *mem,
691 			    int user_alloc)
692 {
693 	int r;
694 	gfn_t base_gfn;
695 	unsigned long npages;
696 	unsigned long i;
697 	struct kvm_memory_slot *memslot;
698 	struct kvm_memory_slot old, new;
699 	struct kvm_memslots *slots, *old_memslots;
700 
701 	r = -EINVAL;
702 	/* General sanity checks */
703 	if (mem->memory_size & (PAGE_SIZE - 1))
704 		goto out;
705 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
706 		goto out;
707 	/* We can read the guest memory with __xxx_user() later on. */
708 	if (user_alloc &&
709 	    ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
710 	     !access_ok(VERIFY_WRITE,
711 			(void __user *)(unsigned long)mem->userspace_addr,
712 			mem->memory_size)))
713 		goto out;
714 	if (mem->slot >= KVM_MEM_SLOTS_NUM)
715 		goto out;
716 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
717 		goto out;
718 
719 	memslot = id_to_memslot(kvm->memslots, mem->slot);
720 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
721 	npages = mem->memory_size >> PAGE_SHIFT;
722 
723 	r = -EINVAL;
724 	if (npages > KVM_MEM_MAX_NR_PAGES)
725 		goto out;
726 
727 	if (!npages)
728 		mem->flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
729 
730 	new = old = *memslot;
731 
732 	new.id = mem->slot;
733 	new.base_gfn = base_gfn;
734 	new.npages = npages;
735 	new.flags = mem->flags;
736 
737 	/* Disallow changing a memory slot's size. */
738 	r = -EINVAL;
739 	if (npages && old.npages && npages != old.npages)
740 		goto out_free;
741 
742 	/* Check for overlaps */
743 	r = -EEXIST;
744 	for (i = 0; i < KVM_MEMORY_SLOTS; ++i) {
745 		struct kvm_memory_slot *s = &kvm->memslots->memslots[i];
746 
747 		if (s == memslot || !s->npages)
748 			continue;
749 		if (!((base_gfn + npages <= s->base_gfn) ||
750 		      (base_gfn >= s->base_gfn + s->npages)))
751 			goto out_free;
752 	}
753 
754 	/* Free page dirty bitmap if unneeded */
755 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
756 		new.dirty_bitmap = NULL;
757 
758 	r = -ENOMEM;
759 
760 	/* Allocate if a slot is being created */
761 	if (npages && !old.npages) {
762 		new.user_alloc = user_alloc;
763 		new.userspace_addr = mem->userspace_addr;
764 #ifndef CONFIG_S390
765 		new.rmap = vzalloc(npages * sizeof(*new.rmap));
766 		if (!new.rmap)
767 			goto out_free;
768 #endif /* not defined CONFIG_S390 */
769 		if (kvm_arch_create_memslot(&new, npages))
770 			goto out_free;
771 	}
772 
773 	/* Allocate page dirty bitmap if needed */
774 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
775 		if (kvm_create_dirty_bitmap(&new) < 0)
776 			goto out_free;
777 		/* destroy any largepage mappings for dirty tracking */
778 	}
779 
780 	if (!npages) {
781 		struct kvm_memory_slot *slot;
782 
783 		r = -ENOMEM;
784 		slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
785 				GFP_KERNEL);
786 		if (!slots)
787 			goto out_free;
788 		slot = id_to_memslot(slots, mem->slot);
789 		slot->flags |= KVM_MEMSLOT_INVALID;
790 
791 		update_memslots(slots, NULL);
792 
793 		old_memslots = kvm->memslots;
794 		rcu_assign_pointer(kvm->memslots, slots);
795 		synchronize_srcu_expedited(&kvm->srcu);
796 		/* From this point no new shadow pages pointing to a deleted
797 		 * memslot will be created.
798 		 *
799 		 * validation of sp->gfn happens in:
800 		 * 	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
801 		 * 	- kvm_is_visible_gfn (mmu_check_roots)
802 		 */
803 		kvm_arch_flush_shadow(kvm);
804 		kfree(old_memslots);
805 	}
806 
807 	r = kvm_arch_prepare_memory_region(kvm, &new, old, mem, user_alloc);
808 	if (r)
809 		goto out_free;
810 
811 	/* map/unmap the pages in iommu page table */
812 	if (npages) {
813 		r = kvm_iommu_map_pages(kvm, &new);
814 		if (r)
815 			goto out_free;
816 	} else
817 		kvm_iommu_unmap_pages(kvm, &old);
818 
819 	r = -ENOMEM;
820 	slots = kmemdup(kvm->memslots, sizeof(struct kvm_memslots),
821 			GFP_KERNEL);
822 	if (!slots)
823 		goto out_free;
824 
825 	/* actual memory is freed via old in kvm_free_physmem_slot below */
826 	if (!npages) {
827 		new.rmap = NULL;
828 		new.dirty_bitmap = NULL;
829 		memset(&new.arch, 0, sizeof(new.arch));
830 	}
831 
832 	update_memslots(slots, &new);
833 	old_memslots = kvm->memslots;
834 	rcu_assign_pointer(kvm->memslots, slots);
835 	synchronize_srcu_expedited(&kvm->srcu);
836 
837 	kvm_arch_commit_memory_region(kvm, mem, old, user_alloc);
838 
839 	/*
840 	 * If the new memory slot is created, we need to clear all
841 	 * mmio sptes.
842 	 */
843 	if (npages && old.base_gfn != mem->guest_phys_addr >> PAGE_SHIFT)
844 		kvm_arch_flush_shadow(kvm);
845 
846 	kvm_free_physmem_slot(&old, &new);
847 	kfree(old_memslots);
848 
849 	return 0;
850 
851 out_free:
852 	kvm_free_physmem_slot(&new, &old);
853 out:
854 	return r;
855 
856 }
857 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
858 
859 int kvm_set_memory_region(struct kvm *kvm,
860 			  struct kvm_userspace_memory_region *mem,
861 			  int user_alloc)
862 {
863 	int r;
864 
865 	mutex_lock(&kvm->slots_lock);
866 	r = __kvm_set_memory_region(kvm, mem, user_alloc);
867 	mutex_unlock(&kvm->slots_lock);
868 	return r;
869 }
870 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
871 
872 int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
873 				   struct
874 				   kvm_userspace_memory_region *mem,
875 				   int user_alloc)
876 {
877 	if (mem->slot >= KVM_MEMORY_SLOTS)
878 		return -EINVAL;
879 	return kvm_set_memory_region(kvm, mem, user_alloc);
880 }
881 
882 int kvm_get_dirty_log(struct kvm *kvm,
883 			struct kvm_dirty_log *log, int *is_dirty)
884 {
885 	struct kvm_memory_slot *memslot;
886 	int r, i;
887 	unsigned long n;
888 	unsigned long any = 0;
889 
890 	r = -EINVAL;
891 	if (log->slot >= KVM_MEMORY_SLOTS)
892 		goto out;
893 
894 	memslot = id_to_memslot(kvm->memslots, log->slot);
895 	r = -ENOENT;
896 	if (!memslot->dirty_bitmap)
897 		goto out;
898 
899 	n = kvm_dirty_bitmap_bytes(memslot);
900 
901 	for (i = 0; !any && i < n/sizeof(long); ++i)
902 		any = memslot->dirty_bitmap[i];
903 
904 	r = -EFAULT;
905 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
906 		goto out;
907 
908 	if (any)
909 		*is_dirty = 1;
910 
911 	r = 0;
912 out:
913 	return r;
914 }
915 
916 bool kvm_largepages_enabled(void)
917 {
918 	return largepages_enabled;
919 }
920 
921 void kvm_disable_largepages(void)
922 {
923 	largepages_enabled = false;
924 }
925 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
926 
927 int is_error_page(struct page *page)
928 {
929 	return page == bad_page || page == hwpoison_page || page == fault_page;
930 }
931 EXPORT_SYMBOL_GPL(is_error_page);
932 
933 int is_error_pfn(pfn_t pfn)
934 {
935 	return pfn == bad_pfn || pfn == hwpoison_pfn || pfn == fault_pfn;
936 }
937 EXPORT_SYMBOL_GPL(is_error_pfn);
938 
939 int is_hwpoison_pfn(pfn_t pfn)
940 {
941 	return pfn == hwpoison_pfn;
942 }
943 EXPORT_SYMBOL_GPL(is_hwpoison_pfn);
944 
945 int is_fault_pfn(pfn_t pfn)
946 {
947 	return pfn == fault_pfn;
948 }
949 EXPORT_SYMBOL_GPL(is_fault_pfn);
950 
951 int is_noslot_pfn(pfn_t pfn)
952 {
953 	return pfn == bad_pfn;
954 }
955 EXPORT_SYMBOL_GPL(is_noslot_pfn);
956 
957 int is_invalid_pfn(pfn_t pfn)
958 {
959 	return pfn == hwpoison_pfn || pfn == fault_pfn;
960 }
961 EXPORT_SYMBOL_GPL(is_invalid_pfn);
962 
963 static inline unsigned long bad_hva(void)
964 {
965 	return PAGE_OFFSET;
966 }
967 
968 int kvm_is_error_hva(unsigned long addr)
969 {
970 	return addr == bad_hva();
971 }
972 EXPORT_SYMBOL_GPL(kvm_is_error_hva);
973 
974 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
975 {
976 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
977 }
978 EXPORT_SYMBOL_GPL(gfn_to_memslot);
979 
980 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
981 {
982 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
983 
984 	if (!memslot || memslot->id >= KVM_MEMORY_SLOTS ||
985 	      memslot->flags & KVM_MEMSLOT_INVALID)
986 		return 0;
987 
988 	return 1;
989 }
990 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
991 
992 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
993 {
994 	struct vm_area_struct *vma;
995 	unsigned long addr, size;
996 
997 	size = PAGE_SIZE;
998 
999 	addr = gfn_to_hva(kvm, gfn);
1000 	if (kvm_is_error_hva(addr))
1001 		return PAGE_SIZE;
1002 
1003 	down_read(&current->mm->mmap_sem);
1004 	vma = find_vma(current->mm, addr);
1005 	if (!vma)
1006 		goto out;
1007 
1008 	size = vma_kernel_pagesize(vma);
1009 
1010 out:
1011 	up_read(&current->mm->mmap_sem);
1012 
1013 	return size;
1014 }
1015 
1016 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1017 				     gfn_t *nr_pages)
1018 {
1019 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1020 		return bad_hva();
1021 
1022 	if (nr_pages)
1023 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
1024 
1025 	return gfn_to_hva_memslot(slot, gfn);
1026 }
1027 
1028 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1029 {
1030 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1031 }
1032 EXPORT_SYMBOL_GPL(gfn_to_hva);
1033 
1034 static pfn_t get_fault_pfn(void)
1035 {
1036 	get_page(fault_page);
1037 	return fault_pfn;
1038 }
1039 
1040 int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1041 	unsigned long start, int write, struct page **page)
1042 {
1043 	int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1044 
1045 	if (write)
1046 		flags |= FOLL_WRITE;
1047 
1048 	return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1049 }
1050 
1051 static inline int check_user_page_hwpoison(unsigned long addr)
1052 {
1053 	int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1054 
1055 	rc = __get_user_pages(current, current->mm, addr, 1,
1056 			      flags, NULL, NULL, NULL);
1057 	return rc == -EHWPOISON;
1058 }
1059 
1060 static pfn_t hva_to_pfn(struct kvm *kvm, unsigned long addr, bool atomic,
1061 			bool *async, bool write_fault, bool *writable)
1062 {
1063 	struct page *page[1];
1064 	int npages = 0;
1065 	pfn_t pfn;
1066 
1067 	/* we can do it either atomically or asynchronously, not both */
1068 	BUG_ON(atomic && async);
1069 
1070 	BUG_ON(!write_fault && !writable);
1071 
1072 	if (writable)
1073 		*writable = true;
1074 
1075 	if (atomic || async)
1076 		npages = __get_user_pages_fast(addr, 1, 1, page);
1077 
1078 	if (unlikely(npages != 1) && !atomic) {
1079 		might_sleep();
1080 
1081 		if (writable)
1082 			*writable = write_fault;
1083 
1084 		if (async) {
1085 			down_read(&current->mm->mmap_sem);
1086 			npages = get_user_page_nowait(current, current->mm,
1087 						     addr, write_fault, page);
1088 			up_read(&current->mm->mmap_sem);
1089 		} else
1090 			npages = get_user_pages_fast(addr, 1, write_fault,
1091 						     page);
1092 
1093 		/* map read fault as writable if possible */
1094 		if (unlikely(!write_fault) && npages == 1) {
1095 			struct page *wpage[1];
1096 
1097 			npages = __get_user_pages_fast(addr, 1, 1, wpage);
1098 			if (npages == 1) {
1099 				*writable = true;
1100 				put_page(page[0]);
1101 				page[0] = wpage[0];
1102 			}
1103 			npages = 1;
1104 		}
1105 	}
1106 
1107 	if (unlikely(npages != 1)) {
1108 		struct vm_area_struct *vma;
1109 
1110 		if (atomic)
1111 			return get_fault_pfn();
1112 
1113 		down_read(&current->mm->mmap_sem);
1114 		if (npages == -EHWPOISON ||
1115 			(!async && check_user_page_hwpoison(addr))) {
1116 			up_read(&current->mm->mmap_sem);
1117 			get_page(hwpoison_page);
1118 			return page_to_pfn(hwpoison_page);
1119 		}
1120 
1121 		vma = find_vma_intersection(current->mm, addr, addr+1);
1122 
1123 		if (vma == NULL)
1124 			pfn = get_fault_pfn();
1125 		else if ((vma->vm_flags & VM_PFNMAP)) {
1126 			pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1127 				vma->vm_pgoff;
1128 			BUG_ON(!kvm_is_mmio_pfn(pfn));
1129 		} else {
1130 			if (async && (vma->vm_flags & VM_WRITE))
1131 				*async = true;
1132 			pfn = get_fault_pfn();
1133 		}
1134 		up_read(&current->mm->mmap_sem);
1135 	} else
1136 		pfn = page_to_pfn(page[0]);
1137 
1138 	return pfn;
1139 }
1140 
1141 pfn_t hva_to_pfn_atomic(struct kvm *kvm, unsigned long addr)
1142 {
1143 	return hva_to_pfn(kvm, addr, true, NULL, true, NULL);
1144 }
1145 EXPORT_SYMBOL_GPL(hva_to_pfn_atomic);
1146 
1147 static pfn_t __gfn_to_pfn(struct kvm *kvm, gfn_t gfn, bool atomic, bool *async,
1148 			  bool write_fault, bool *writable)
1149 {
1150 	unsigned long addr;
1151 
1152 	if (async)
1153 		*async = false;
1154 
1155 	addr = gfn_to_hva(kvm, gfn);
1156 	if (kvm_is_error_hva(addr)) {
1157 		get_page(bad_page);
1158 		return page_to_pfn(bad_page);
1159 	}
1160 
1161 	return hva_to_pfn(kvm, addr, atomic, async, write_fault, writable);
1162 }
1163 
1164 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1165 {
1166 	return __gfn_to_pfn(kvm, gfn, true, NULL, true, NULL);
1167 }
1168 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1169 
1170 pfn_t gfn_to_pfn_async(struct kvm *kvm, gfn_t gfn, bool *async,
1171 		       bool write_fault, bool *writable)
1172 {
1173 	return __gfn_to_pfn(kvm, gfn, false, async, write_fault, writable);
1174 }
1175 EXPORT_SYMBOL_GPL(gfn_to_pfn_async);
1176 
1177 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1178 {
1179 	return __gfn_to_pfn(kvm, gfn, false, NULL, true, NULL);
1180 }
1181 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1182 
1183 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1184 		      bool *writable)
1185 {
1186 	return __gfn_to_pfn(kvm, gfn, false, NULL, write_fault, writable);
1187 }
1188 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1189 
1190 pfn_t gfn_to_pfn_memslot(struct kvm *kvm,
1191 			 struct kvm_memory_slot *slot, gfn_t gfn)
1192 {
1193 	unsigned long addr = gfn_to_hva_memslot(slot, gfn);
1194 	return hva_to_pfn(kvm, addr, false, NULL, true, NULL);
1195 }
1196 
1197 int gfn_to_page_many_atomic(struct kvm *kvm, gfn_t gfn, struct page **pages,
1198 								  int nr_pages)
1199 {
1200 	unsigned long addr;
1201 	gfn_t entry;
1202 
1203 	addr = gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, &entry);
1204 	if (kvm_is_error_hva(addr))
1205 		return -1;
1206 
1207 	if (entry < nr_pages)
1208 		return 0;
1209 
1210 	return __get_user_pages_fast(addr, nr_pages, 1, pages);
1211 }
1212 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1213 
1214 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1215 {
1216 	pfn_t pfn;
1217 
1218 	pfn = gfn_to_pfn(kvm, gfn);
1219 	if (!kvm_is_mmio_pfn(pfn))
1220 		return pfn_to_page(pfn);
1221 
1222 	WARN_ON(kvm_is_mmio_pfn(pfn));
1223 
1224 	get_page(bad_page);
1225 	return bad_page;
1226 }
1227 
1228 EXPORT_SYMBOL_GPL(gfn_to_page);
1229 
1230 void kvm_release_page_clean(struct page *page)
1231 {
1232 	kvm_release_pfn_clean(page_to_pfn(page));
1233 }
1234 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1235 
1236 void kvm_release_pfn_clean(pfn_t pfn)
1237 {
1238 	if (!kvm_is_mmio_pfn(pfn))
1239 		put_page(pfn_to_page(pfn));
1240 }
1241 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1242 
1243 void kvm_release_page_dirty(struct page *page)
1244 {
1245 	kvm_release_pfn_dirty(page_to_pfn(page));
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1248 
1249 void kvm_release_pfn_dirty(pfn_t pfn)
1250 {
1251 	kvm_set_pfn_dirty(pfn);
1252 	kvm_release_pfn_clean(pfn);
1253 }
1254 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
1255 
1256 void kvm_set_page_dirty(struct page *page)
1257 {
1258 	kvm_set_pfn_dirty(page_to_pfn(page));
1259 }
1260 EXPORT_SYMBOL_GPL(kvm_set_page_dirty);
1261 
1262 void kvm_set_pfn_dirty(pfn_t pfn)
1263 {
1264 	if (!kvm_is_mmio_pfn(pfn)) {
1265 		struct page *page = pfn_to_page(pfn);
1266 		if (!PageReserved(page))
1267 			SetPageDirty(page);
1268 	}
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1271 
1272 void kvm_set_pfn_accessed(pfn_t pfn)
1273 {
1274 	if (!kvm_is_mmio_pfn(pfn))
1275 		mark_page_accessed(pfn_to_page(pfn));
1276 }
1277 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1278 
1279 void kvm_get_pfn(pfn_t pfn)
1280 {
1281 	if (!kvm_is_mmio_pfn(pfn))
1282 		get_page(pfn_to_page(pfn));
1283 }
1284 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1285 
1286 static int next_segment(unsigned long len, int offset)
1287 {
1288 	if (len > PAGE_SIZE - offset)
1289 		return PAGE_SIZE - offset;
1290 	else
1291 		return len;
1292 }
1293 
1294 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1295 			int len)
1296 {
1297 	int r;
1298 	unsigned long addr;
1299 
1300 	addr = gfn_to_hva(kvm, gfn);
1301 	if (kvm_is_error_hva(addr))
1302 		return -EFAULT;
1303 	r = __copy_from_user(data, (void __user *)addr + offset, len);
1304 	if (r)
1305 		return -EFAULT;
1306 	return 0;
1307 }
1308 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1309 
1310 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1311 {
1312 	gfn_t gfn = gpa >> PAGE_SHIFT;
1313 	int seg;
1314 	int offset = offset_in_page(gpa);
1315 	int ret;
1316 
1317 	while ((seg = next_segment(len, offset)) != 0) {
1318 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1319 		if (ret < 0)
1320 			return ret;
1321 		offset = 0;
1322 		len -= seg;
1323 		data += seg;
1324 		++gfn;
1325 	}
1326 	return 0;
1327 }
1328 EXPORT_SYMBOL_GPL(kvm_read_guest);
1329 
1330 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1331 			  unsigned long len)
1332 {
1333 	int r;
1334 	unsigned long addr;
1335 	gfn_t gfn = gpa >> PAGE_SHIFT;
1336 	int offset = offset_in_page(gpa);
1337 
1338 	addr = gfn_to_hva(kvm, gfn);
1339 	if (kvm_is_error_hva(addr))
1340 		return -EFAULT;
1341 	pagefault_disable();
1342 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1343 	pagefault_enable();
1344 	if (r)
1345 		return -EFAULT;
1346 	return 0;
1347 }
1348 EXPORT_SYMBOL(kvm_read_guest_atomic);
1349 
1350 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn, const void *data,
1351 			 int offset, int len)
1352 {
1353 	int r;
1354 	unsigned long addr;
1355 
1356 	addr = gfn_to_hva(kvm, gfn);
1357 	if (kvm_is_error_hva(addr))
1358 		return -EFAULT;
1359 	r = __copy_to_user((void __user *)addr + offset, data, len);
1360 	if (r)
1361 		return -EFAULT;
1362 	mark_page_dirty(kvm, gfn);
1363 	return 0;
1364 }
1365 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1366 
1367 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1368 		    unsigned long len)
1369 {
1370 	gfn_t gfn = gpa >> PAGE_SHIFT;
1371 	int seg;
1372 	int offset = offset_in_page(gpa);
1373 	int ret;
1374 
1375 	while ((seg = next_segment(len, offset)) != 0) {
1376 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1377 		if (ret < 0)
1378 			return ret;
1379 		offset = 0;
1380 		len -= seg;
1381 		data += seg;
1382 		++gfn;
1383 	}
1384 	return 0;
1385 }
1386 
1387 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1388 			      gpa_t gpa)
1389 {
1390 	struct kvm_memslots *slots = kvm_memslots(kvm);
1391 	int offset = offset_in_page(gpa);
1392 	gfn_t gfn = gpa >> PAGE_SHIFT;
1393 
1394 	ghc->gpa = gpa;
1395 	ghc->generation = slots->generation;
1396 	ghc->memslot = gfn_to_memslot(kvm, gfn);
1397 	ghc->hva = gfn_to_hva_many(ghc->memslot, gfn, NULL);
1398 	if (!kvm_is_error_hva(ghc->hva))
1399 		ghc->hva += offset;
1400 	else
1401 		return -EFAULT;
1402 
1403 	return 0;
1404 }
1405 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1406 
1407 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1408 			   void *data, unsigned long len)
1409 {
1410 	struct kvm_memslots *slots = kvm_memslots(kvm);
1411 	int r;
1412 
1413 	if (slots->generation != ghc->generation)
1414 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1415 
1416 	if (kvm_is_error_hva(ghc->hva))
1417 		return -EFAULT;
1418 
1419 	r = __copy_to_user((void __user *)ghc->hva, data, len);
1420 	if (r)
1421 		return -EFAULT;
1422 	mark_page_dirty_in_slot(kvm, ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1423 
1424 	return 0;
1425 }
1426 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1427 
1428 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1429 			   void *data, unsigned long len)
1430 {
1431 	struct kvm_memslots *slots = kvm_memslots(kvm);
1432 	int r;
1433 
1434 	if (slots->generation != ghc->generation)
1435 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa);
1436 
1437 	if (kvm_is_error_hva(ghc->hva))
1438 		return -EFAULT;
1439 
1440 	r = __copy_from_user(data, (void __user *)ghc->hva, len);
1441 	if (r)
1442 		return -EFAULT;
1443 
1444 	return 0;
1445 }
1446 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1447 
1448 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1449 {
1450 	return kvm_write_guest_page(kvm, gfn, (const void *) empty_zero_page,
1451 				    offset, len);
1452 }
1453 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1454 
1455 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1456 {
1457 	gfn_t gfn = gpa >> PAGE_SHIFT;
1458 	int seg;
1459 	int offset = offset_in_page(gpa);
1460 	int ret;
1461 
1462         while ((seg = next_segment(len, offset)) != 0) {
1463 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1464 		if (ret < 0)
1465 			return ret;
1466 		offset = 0;
1467 		len -= seg;
1468 		++gfn;
1469 	}
1470 	return 0;
1471 }
1472 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1473 
1474 void mark_page_dirty_in_slot(struct kvm *kvm, struct kvm_memory_slot *memslot,
1475 			     gfn_t gfn)
1476 {
1477 	if (memslot && memslot->dirty_bitmap) {
1478 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1479 
1480 		if (!test_and_set_bit_le(rel_gfn, memslot->dirty_bitmap))
1481 			memslot->nr_dirty_pages++;
1482 	}
1483 }
1484 
1485 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1486 {
1487 	struct kvm_memory_slot *memslot;
1488 
1489 	memslot = gfn_to_memslot(kvm, gfn);
1490 	mark_page_dirty_in_slot(kvm, memslot, gfn);
1491 }
1492 
1493 /*
1494  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1495  */
1496 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1497 {
1498 	DEFINE_WAIT(wait);
1499 
1500 	for (;;) {
1501 		prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1502 
1503 		if (kvm_arch_vcpu_runnable(vcpu)) {
1504 			kvm_make_request(KVM_REQ_UNHALT, vcpu);
1505 			break;
1506 		}
1507 		if (kvm_cpu_has_pending_timer(vcpu))
1508 			break;
1509 		if (signal_pending(current))
1510 			break;
1511 
1512 		schedule();
1513 	}
1514 
1515 	finish_wait(&vcpu->wq, &wait);
1516 }
1517 
1518 void kvm_resched(struct kvm_vcpu *vcpu)
1519 {
1520 	if (!need_resched())
1521 		return;
1522 	cond_resched();
1523 }
1524 EXPORT_SYMBOL_GPL(kvm_resched);
1525 
1526 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
1527 {
1528 	struct kvm *kvm = me->kvm;
1529 	struct kvm_vcpu *vcpu;
1530 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
1531 	int yielded = 0;
1532 	int pass;
1533 	int i;
1534 
1535 	/*
1536 	 * We boost the priority of a VCPU that is runnable but not
1537 	 * currently running, because it got preempted by something
1538 	 * else and called schedule in __vcpu_run.  Hopefully that
1539 	 * VCPU is holding the lock that we need and will release it.
1540 	 * We approximate round-robin by starting at the last boosted VCPU.
1541 	 */
1542 	for (pass = 0; pass < 2 && !yielded; pass++) {
1543 		kvm_for_each_vcpu(i, vcpu, kvm) {
1544 			struct task_struct *task = NULL;
1545 			struct pid *pid;
1546 			if (!pass && i < last_boosted_vcpu) {
1547 				i = last_boosted_vcpu;
1548 				continue;
1549 			} else if (pass && i > last_boosted_vcpu)
1550 				break;
1551 			if (vcpu == me)
1552 				continue;
1553 			if (waitqueue_active(&vcpu->wq))
1554 				continue;
1555 			rcu_read_lock();
1556 			pid = rcu_dereference(vcpu->pid);
1557 			if (pid)
1558 				task = get_pid_task(vcpu->pid, PIDTYPE_PID);
1559 			rcu_read_unlock();
1560 			if (!task)
1561 				continue;
1562 			if (task->flags & PF_VCPU) {
1563 				put_task_struct(task);
1564 				continue;
1565 			}
1566 			if (yield_to(task, 1)) {
1567 				put_task_struct(task);
1568 				kvm->last_boosted_vcpu = i;
1569 				yielded = 1;
1570 				break;
1571 			}
1572 			put_task_struct(task);
1573 		}
1574 	}
1575 }
1576 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1577 
1578 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1579 {
1580 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
1581 	struct page *page;
1582 
1583 	if (vmf->pgoff == 0)
1584 		page = virt_to_page(vcpu->run);
1585 #ifdef CONFIG_X86
1586 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
1587 		page = virt_to_page(vcpu->arch.pio_data);
1588 #endif
1589 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1590 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
1591 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
1592 #endif
1593 	else
1594 		return kvm_arch_vcpu_fault(vcpu, vmf);
1595 	get_page(page);
1596 	vmf->page = page;
1597 	return 0;
1598 }
1599 
1600 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
1601 	.fault = kvm_vcpu_fault,
1602 };
1603 
1604 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
1605 {
1606 	vma->vm_ops = &kvm_vcpu_vm_ops;
1607 	return 0;
1608 }
1609 
1610 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
1611 {
1612 	struct kvm_vcpu *vcpu = filp->private_data;
1613 
1614 	kvm_put_kvm(vcpu->kvm);
1615 	return 0;
1616 }
1617 
1618 static struct file_operations kvm_vcpu_fops = {
1619 	.release        = kvm_vcpu_release,
1620 	.unlocked_ioctl = kvm_vcpu_ioctl,
1621 #ifdef CONFIG_COMPAT
1622 	.compat_ioctl   = kvm_vcpu_compat_ioctl,
1623 #endif
1624 	.mmap           = kvm_vcpu_mmap,
1625 	.llseek		= noop_llseek,
1626 };
1627 
1628 /*
1629  * Allocates an inode for the vcpu.
1630  */
1631 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
1632 {
1633 	return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR);
1634 }
1635 
1636 /*
1637  * Creates some virtual cpus.  Good luck creating more than one.
1638  */
1639 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
1640 {
1641 	int r;
1642 	struct kvm_vcpu *vcpu, *v;
1643 
1644 	vcpu = kvm_arch_vcpu_create(kvm, id);
1645 	if (IS_ERR(vcpu))
1646 		return PTR_ERR(vcpu);
1647 
1648 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
1649 
1650 	r = kvm_arch_vcpu_setup(vcpu);
1651 	if (r)
1652 		goto vcpu_destroy;
1653 
1654 	mutex_lock(&kvm->lock);
1655 	if (!kvm_vcpu_compatible(vcpu)) {
1656 		r = -EINVAL;
1657 		goto unlock_vcpu_destroy;
1658 	}
1659 	if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
1660 		r = -EINVAL;
1661 		goto unlock_vcpu_destroy;
1662 	}
1663 
1664 	kvm_for_each_vcpu(r, v, kvm)
1665 		if (v->vcpu_id == id) {
1666 			r = -EEXIST;
1667 			goto unlock_vcpu_destroy;
1668 		}
1669 
1670 	BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
1671 
1672 	/* Now it's all set up, let userspace reach it */
1673 	kvm_get_kvm(kvm);
1674 	r = create_vcpu_fd(vcpu);
1675 	if (r < 0) {
1676 		kvm_put_kvm(kvm);
1677 		goto unlock_vcpu_destroy;
1678 	}
1679 
1680 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
1681 	smp_wmb();
1682 	atomic_inc(&kvm->online_vcpus);
1683 
1684 	mutex_unlock(&kvm->lock);
1685 	return r;
1686 
1687 unlock_vcpu_destroy:
1688 	mutex_unlock(&kvm->lock);
1689 vcpu_destroy:
1690 	kvm_arch_vcpu_destroy(vcpu);
1691 	return r;
1692 }
1693 
1694 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1695 {
1696 	if (sigset) {
1697 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
1698 		vcpu->sigset_active = 1;
1699 		vcpu->sigset = *sigset;
1700 	} else
1701 		vcpu->sigset_active = 0;
1702 	return 0;
1703 }
1704 
1705 static long kvm_vcpu_ioctl(struct file *filp,
1706 			   unsigned int ioctl, unsigned long arg)
1707 {
1708 	struct kvm_vcpu *vcpu = filp->private_data;
1709 	void __user *argp = (void __user *)arg;
1710 	int r;
1711 	struct kvm_fpu *fpu = NULL;
1712 	struct kvm_sregs *kvm_sregs = NULL;
1713 
1714 	if (vcpu->kvm->mm != current->mm)
1715 		return -EIO;
1716 
1717 #if defined(CONFIG_S390) || defined(CONFIG_PPC)
1718 	/*
1719 	 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
1720 	 * so vcpu_load() would break it.
1721 	 */
1722 	if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_INTERRUPT)
1723 		return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1724 #endif
1725 
1726 
1727 	vcpu_load(vcpu);
1728 	switch (ioctl) {
1729 	case KVM_RUN:
1730 		r = -EINVAL;
1731 		if (arg)
1732 			goto out;
1733 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
1734 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
1735 		break;
1736 	case KVM_GET_REGS: {
1737 		struct kvm_regs *kvm_regs;
1738 
1739 		r = -ENOMEM;
1740 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
1741 		if (!kvm_regs)
1742 			goto out;
1743 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
1744 		if (r)
1745 			goto out_free1;
1746 		r = -EFAULT;
1747 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
1748 			goto out_free1;
1749 		r = 0;
1750 out_free1:
1751 		kfree(kvm_regs);
1752 		break;
1753 	}
1754 	case KVM_SET_REGS: {
1755 		struct kvm_regs *kvm_regs;
1756 
1757 		r = -ENOMEM;
1758 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
1759 		if (IS_ERR(kvm_regs)) {
1760 			r = PTR_ERR(kvm_regs);
1761 			goto out;
1762 		}
1763 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
1764 		if (r)
1765 			goto out_free2;
1766 		r = 0;
1767 out_free2:
1768 		kfree(kvm_regs);
1769 		break;
1770 	}
1771 	case KVM_GET_SREGS: {
1772 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
1773 		r = -ENOMEM;
1774 		if (!kvm_sregs)
1775 			goto out;
1776 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
1777 		if (r)
1778 			goto out;
1779 		r = -EFAULT;
1780 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
1781 			goto out;
1782 		r = 0;
1783 		break;
1784 	}
1785 	case KVM_SET_SREGS: {
1786 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
1787 		if (IS_ERR(kvm_sregs)) {
1788 			r = PTR_ERR(kvm_sregs);
1789 			goto out;
1790 		}
1791 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
1792 		if (r)
1793 			goto out;
1794 		r = 0;
1795 		break;
1796 	}
1797 	case KVM_GET_MP_STATE: {
1798 		struct kvm_mp_state mp_state;
1799 
1800 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
1801 		if (r)
1802 			goto out;
1803 		r = -EFAULT;
1804 		if (copy_to_user(argp, &mp_state, sizeof mp_state))
1805 			goto out;
1806 		r = 0;
1807 		break;
1808 	}
1809 	case KVM_SET_MP_STATE: {
1810 		struct kvm_mp_state mp_state;
1811 
1812 		r = -EFAULT;
1813 		if (copy_from_user(&mp_state, argp, sizeof mp_state))
1814 			goto out;
1815 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
1816 		if (r)
1817 			goto out;
1818 		r = 0;
1819 		break;
1820 	}
1821 	case KVM_TRANSLATE: {
1822 		struct kvm_translation tr;
1823 
1824 		r = -EFAULT;
1825 		if (copy_from_user(&tr, argp, sizeof tr))
1826 			goto out;
1827 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
1828 		if (r)
1829 			goto out;
1830 		r = -EFAULT;
1831 		if (copy_to_user(argp, &tr, sizeof tr))
1832 			goto out;
1833 		r = 0;
1834 		break;
1835 	}
1836 	case KVM_SET_GUEST_DEBUG: {
1837 		struct kvm_guest_debug dbg;
1838 
1839 		r = -EFAULT;
1840 		if (copy_from_user(&dbg, argp, sizeof dbg))
1841 			goto out;
1842 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
1843 		if (r)
1844 			goto out;
1845 		r = 0;
1846 		break;
1847 	}
1848 	case KVM_SET_SIGNAL_MASK: {
1849 		struct kvm_signal_mask __user *sigmask_arg = argp;
1850 		struct kvm_signal_mask kvm_sigmask;
1851 		sigset_t sigset, *p;
1852 
1853 		p = NULL;
1854 		if (argp) {
1855 			r = -EFAULT;
1856 			if (copy_from_user(&kvm_sigmask, argp,
1857 					   sizeof kvm_sigmask))
1858 				goto out;
1859 			r = -EINVAL;
1860 			if (kvm_sigmask.len != sizeof sigset)
1861 				goto out;
1862 			r = -EFAULT;
1863 			if (copy_from_user(&sigset, sigmask_arg->sigset,
1864 					   sizeof sigset))
1865 				goto out;
1866 			p = &sigset;
1867 		}
1868 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
1869 		break;
1870 	}
1871 	case KVM_GET_FPU: {
1872 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
1873 		r = -ENOMEM;
1874 		if (!fpu)
1875 			goto out;
1876 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
1877 		if (r)
1878 			goto out;
1879 		r = -EFAULT;
1880 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
1881 			goto out;
1882 		r = 0;
1883 		break;
1884 	}
1885 	case KVM_SET_FPU: {
1886 		fpu = memdup_user(argp, sizeof(*fpu));
1887 		if (IS_ERR(fpu)) {
1888 			r = PTR_ERR(fpu);
1889 			goto out;
1890 		}
1891 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
1892 		if (r)
1893 			goto out;
1894 		r = 0;
1895 		break;
1896 	}
1897 	default:
1898 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
1899 	}
1900 out:
1901 	vcpu_put(vcpu);
1902 	kfree(fpu);
1903 	kfree(kvm_sregs);
1904 	return r;
1905 }
1906 
1907 #ifdef CONFIG_COMPAT
1908 static long kvm_vcpu_compat_ioctl(struct file *filp,
1909 				  unsigned int ioctl, unsigned long arg)
1910 {
1911 	struct kvm_vcpu *vcpu = filp->private_data;
1912 	void __user *argp = compat_ptr(arg);
1913 	int r;
1914 
1915 	if (vcpu->kvm->mm != current->mm)
1916 		return -EIO;
1917 
1918 	switch (ioctl) {
1919 	case KVM_SET_SIGNAL_MASK: {
1920 		struct kvm_signal_mask __user *sigmask_arg = argp;
1921 		struct kvm_signal_mask kvm_sigmask;
1922 		compat_sigset_t csigset;
1923 		sigset_t sigset;
1924 
1925 		if (argp) {
1926 			r = -EFAULT;
1927 			if (copy_from_user(&kvm_sigmask, argp,
1928 					   sizeof kvm_sigmask))
1929 				goto out;
1930 			r = -EINVAL;
1931 			if (kvm_sigmask.len != sizeof csigset)
1932 				goto out;
1933 			r = -EFAULT;
1934 			if (copy_from_user(&csigset, sigmask_arg->sigset,
1935 					   sizeof csigset))
1936 				goto out;
1937 		}
1938 		sigset_from_compat(&sigset, &csigset);
1939 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
1940 		break;
1941 	}
1942 	default:
1943 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
1944 	}
1945 
1946 out:
1947 	return r;
1948 }
1949 #endif
1950 
1951 static long kvm_vm_ioctl(struct file *filp,
1952 			   unsigned int ioctl, unsigned long arg)
1953 {
1954 	struct kvm *kvm = filp->private_data;
1955 	void __user *argp = (void __user *)arg;
1956 	int r;
1957 
1958 	if (kvm->mm != current->mm)
1959 		return -EIO;
1960 	switch (ioctl) {
1961 	case KVM_CREATE_VCPU:
1962 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
1963 		if (r < 0)
1964 			goto out;
1965 		break;
1966 	case KVM_SET_USER_MEMORY_REGION: {
1967 		struct kvm_userspace_memory_region kvm_userspace_mem;
1968 
1969 		r = -EFAULT;
1970 		if (copy_from_user(&kvm_userspace_mem, argp,
1971 						sizeof kvm_userspace_mem))
1972 			goto out;
1973 
1974 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem, 1);
1975 		if (r)
1976 			goto out;
1977 		break;
1978 	}
1979 	case KVM_GET_DIRTY_LOG: {
1980 		struct kvm_dirty_log log;
1981 
1982 		r = -EFAULT;
1983 		if (copy_from_user(&log, argp, sizeof log))
1984 			goto out;
1985 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1986 		if (r)
1987 			goto out;
1988 		break;
1989 	}
1990 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
1991 	case KVM_REGISTER_COALESCED_MMIO: {
1992 		struct kvm_coalesced_mmio_zone zone;
1993 		r = -EFAULT;
1994 		if (copy_from_user(&zone, argp, sizeof zone))
1995 			goto out;
1996 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
1997 		if (r)
1998 			goto out;
1999 		r = 0;
2000 		break;
2001 	}
2002 	case KVM_UNREGISTER_COALESCED_MMIO: {
2003 		struct kvm_coalesced_mmio_zone zone;
2004 		r = -EFAULT;
2005 		if (copy_from_user(&zone, argp, sizeof zone))
2006 			goto out;
2007 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2008 		if (r)
2009 			goto out;
2010 		r = 0;
2011 		break;
2012 	}
2013 #endif
2014 	case KVM_IRQFD: {
2015 		struct kvm_irqfd data;
2016 
2017 		r = -EFAULT;
2018 		if (copy_from_user(&data, argp, sizeof data))
2019 			goto out;
2020 		r = kvm_irqfd(kvm, data.fd, data.gsi, data.flags);
2021 		break;
2022 	}
2023 	case KVM_IOEVENTFD: {
2024 		struct kvm_ioeventfd data;
2025 
2026 		r = -EFAULT;
2027 		if (copy_from_user(&data, argp, sizeof data))
2028 			goto out;
2029 		r = kvm_ioeventfd(kvm, &data);
2030 		break;
2031 	}
2032 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2033 	case KVM_SET_BOOT_CPU_ID:
2034 		r = 0;
2035 		mutex_lock(&kvm->lock);
2036 		if (atomic_read(&kvm->online_vcpus) != 0)
2037 			r = -EBUSY;
2038 		else
2039 			kvm->bsp_vcpu_id = arg;
2040 		mutex_unlock(&kvm->lock);
2041 		break;
2042 #endif
2043 	default:
2044 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2045 		if (r == -ENOTTY)
2046 			r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2047 	}
2048 out:
2049 	return r;
2050 }
2051 
2052 #ifdef CONFIG_COMPAT
2053 struct compat_kvm_dirty_log {
2054 	__u32 slot;
2055 	__u32 padding1;
2056 	union {
2057 		compat_uptr_t dirty_bitmap; /* one bit per page */
2058 		__u64 padding2;
2059 	};
2060 };
2061 
2062 static long kvm_vm_compat_ioctl(struct file *filp,
2063 			   unsigned int ioctl, unsigned long arg)
2064 {
2065 	struct kvm *kvm = filp->private_data;
2066 	int r;
2067 
2068 	if (kvm->mm != current->mm)
2069 		return -EIO;
2070 	switch (ioctl) {
2071 	case KVM_GET_DIRTY_LOG: {
2072 		struct compat_kvm_dirty_log compat_log;
2073 		struct kvm_dirty_log log;
2074 
2075 		r = -EFAULT;
2076 		if (copy_from_user(&compat_log, (void __user *)arg,
2077 				   sizeof(compat_log)))
2078 			goto out;
2079 		log.slot	 = compat_log.slot;
2080 		log.padding1	 = compat_log.padding1;
2081 		log.padding2	 = compat_log.padding2;
2082 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2083 
2084 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2085 		if (r)
2086 			goto out;
2087 		break;
2088 	}
2089 	default:
2090 		r = kvm_vm_ioctl(filp, ioctl, arg);
2091 	}
2092 
2093 out:
2094 	return r;
2095 }
2096 #endif
2097 
2098 static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2099 {
2100 	struct page *page[1];
2101 	unsigned long addr;
2102 	int npages;
2103 	gfn_t gfn = vmf->pgoff;
2104 	struct kvm *kvm = vma->vm_file->private_data;
2105 
2106 	addr = gfn_to_hva(kvm, gfn);
2107 	if (kvm_is_error_hva(addr))
2108 		return VM_FAULT_SIGBUS;
2109 
2110 	npages = get_user_pages(current, current->mm, addr, 1, 1, 0, page,
2111 				NULL);
2112 	if (unlikely(npages != 1))
2113 		return VM_FAULT_SIGBUS;
2114 
2115 	vmf->page = page[0];
2116 	return 0;
2117 }
2118 
2119 static const struct vm_operations_struct kvm_vm_vm_ops = {
2120 	.fault = kvm_vm_fault,
2121 };
2122 
2123 static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2124 {
2125 	vma->vm_ops = &kvm_vm_vm_ops;
2126 	return 0;
2127 }
2128 
2129 static struct file_operations kvm_vm_fops = {
2130 	.release        = kvm_vm_release,
2131 	.unlocked_ioctl = kvm_vm_ioctl,
2132 #ifdef CONFIG_COMPAT
2133 	.compat_ioctl   = kvm_vm_compat_ioctl,
2134 #endif
2135 	.mmap           = kvm_vm_mmap,
2136 	.llseek		= noop_llseek,
2137 };
2138 
2139 static int kvm_dev_ioctl_create_vm(unsigned long type)
2140 {
2141 	int r;
2142 	struct kvm *kvm;
2143 
2144 	kvm = kvm_create_vm(type);
2145 	if (IS_ERR(kvm))
2146 		return PTR_ERR(kvm);
2147 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2148 	r = kvm_coalesced_mmio_init(kvm);
2149 	if (r < 0) {
2150 		kvm_put_kvm(kvm);
2151 		return r;
2152 	}
2153 #endif
2154 	r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
2155 	if (r < 0)
2156 		kvm_put_kvm(kvm);
2157 
2158 	return r;
2159 }
2160 
2161 static long kvm_dev_ioctl_check_extension_generic(long arg)
2162 {
2163 	switch (arg) {
2164 	case KVM_CAP_USER_MEMORY:
2165 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2166 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2167 #ifdef CONFIG_KVM_APIC_ARCHITECTURE
2168 	case KVM_CAP_SET_BOOT_CPU_ID:
2169 #endif
2170 	case KVM_CAP_INTERNAL_ERROR_DATA:
2171 		return 1;
2172 #ifdef CONFIG_HAVE_KVM_IRQCHIP
2173 	case KVM_CAP_IRQ_ROUTING:
2174 		return KVM_MAX_IRQ_ROUTES;
2175 #endif
2176 	default:
2177 		break;
2178 	}
2179 	return kvm_dev_ioctl_check_extension(arg);
2180 }
2181 
2182 static long kvm_dev_ioctl(struct file *filp,
2183 			  unsigned int ioctl, unsigned long arg)
2184 {
2185 	long r = -EINVAL;
2186 
2187 	switch (ioctl) {
2188 	case KVM_GET_API_VERSION:
2189 		r = -EINVAL;
2190 		if (arg)
2191 			goto out;
2192 		r = KVM_API_VERSION;
2193 		break;
2194 	case KVM_CREATE_VM:
2195 		r = kvm_dev_ioctl_create_vm(arg);
2196 		break;
2197 	case KVM_CHECK_EXTENSION:
2198 		r = kvm_dev_ioctl_check_extension_generic(arg);
2199 		break;
2200 	case KVM_GET_VCPU_MMAP_SIZE:
2201 		r = -EINVAL;
2202 		if (arg)
2203 			goto out;
2204 		r = PAGE_SIZE;     /* struct kvm_run */
2205 #ifdef CONFIG_X86
2206 		r += PAGE_SIZE;    /* pio data page */
2207 #endif
2208 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2209 		r += PAGE_SIZE;    /* coalesced mmio ring page */
2210 #endif
2211 		break;
2212 	case KVM_TRACE_ENABLE:
2213 	case KVM_TRACE_PAUSE:
2214 	case KVM_TRACE_DISABLE:
2215 		r = -EOPNOTSUPP;
2216 		break;
2217 	default:
2218 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
2219 	}
2220 out:
2221 	return r;
2222 }
2223 
2224 static struct file_operations kvm_chardev_ops = {
2225 	.unlocked_ioctl = kvm_dev_ioctl,
2226 	.compat_ioctl   = kvm_dev_ioctl,
2227 	.llseek		= noop_llseek,
2228 };
2229 
2230 static struct miscdevice kvm_dev = {
2231 	KVM_MINOR,
2232 	"kvm",
2233 	&kvm_chardev_ops,
2234 };
2235 
2236 static void hardware_enable_nolock(void *junk)
2237 {
2238 	int cpu = raw_smp_processor_id();
2239 	int r;
2240 
2241 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2242 		return;
2243 
2244 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
2245 
2246 	r = kvm_arch_hardware_enable(NULL);
2247 
2248 	if (r) {
2249 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2250 		atomic_inc(&hardware_enable_failed);
2251 		printk(KERN_INFO "kvm: enabling virtualization on "
2252 				 "CPU%d failed\n", cpu);
2253 	}
2254 }
2255 
2256 static void hardware_enable(void *junk)
2257 {
2258 	raw_spin_lock(&kvm_lock);
2259 	hardware_enable_nolock(junk);
2260 	raw_spin_unlock(&kvm_lock);
2261 }
2262 
2263 static void hardware_disable_nolock(void *junk)
2264 {
2265 	int cpu = raw_smp_processor_id();
2266 
2267 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
2268 		return;
2269 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
2270 	kvm_arch_hardware_disable(NULL);
2271 }
2272 
2273 static void hardware_disable(void *junk)
2274 {
2275 	raw_spin_lock(&kvm_lock);
2276 	hardware_disable_nolock(junk);
2277 	raw_spin_unlock(&kvm_lock);
2278 }
2279 
2280 static void hardware_disable_all_nolock(void)
2281 {
2282 	BUG_ON(!kvm_usage_count);
2283 
2284 	kvm_usage_count--;
2285 	if (!kvm_usage_count)
2286 		on_each_cpu(hardware_disable_nolock, NULL, 1);
2287 }
2288 
2289 static void hardware_disable_all(void)
2290 {
2291 	raw_spin_lock(&kvm_lock);
2292 	hardware_disable_all_nolock();
2293 	raw_spin_unlock(&kvm_lock);
2294 }
2295 
2296 static int hardware_enable_all(void)
2297 {
2298 	int r = 0;
2299 
2300 	raw_spin_lock(&kvm_lock);
2301 
2302 	kvm_usage_count++;
2303 	if (kvm_usage_count == 1) {
2304 		atomic_set(&hardware_enable_failed, 0);
2305 		on_each_cpu(hardware_enable_nolock, NULL, 1);
2306 
2307 		if (atomic_read(&hardware_enable_failed)) {
2308 			hardware_disable_all_nolock();
2309 			r = -EBUSY;
2310 		}
2311 	}
2312 
2313 	raw_spin_unlock(&kvm_lock);
2314 
2315 	return r;
2316 }
2317 
2318 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2319 			   void *v)
2320 {
2321 	int cpu = (long)v;
2322 
2323 	if (!kvm_usage_count)
2324 		return NOTIFY_OK;
2325 
2326 	val &= ~CPU_TASKS_FROZEN;
2327 	switch (val) {
2328 	case CPU_DYING:
2329 		printk(KERN_INFO "kvm: disabling virtualization on CPU%d\n",
2330 		       cpu);
2331 		hardware_disable(NULL);
2332 		break;
2333 	case CPU_STARTING:
2334 		printk(KERN_INFO "kvm: enabling virtualization on CPU%d\n",
2335 		       cpu);
2336 		hardware_enable(NULL);
2337 		break;
2338 	}
2339 	return NOTIFY_OK;
2340 }
2341 
2342 
2343 asmlinkage void kvm_spurious_fault(void)
2344 {
2345 	/* Fault while not rebooting.  We want the trace. */
2346 	BUG();
2347 }
2348 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
2349 
2350 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
2351 		      void *v)
2352 {
2353 	/*
2354 	 * Some (well, at least mine) BIOSes hang on reboot if
2355 	 * in vmx root mode.
2356 	 *
2357 	 * And Intel TXT required VMX off for all cpu when system shutdown.
2358 	 */
2359 	printk(KERN_INFO "kvm: exiting hardware virtualization\n");
2360 	kvm_rebooting = true;
2361 	on_each_cpu(hardware_disable_nolock, NULL, 1);
2362 	return NOTIFY_OK;
2363 }
2364 
2365 static struct notifier_block kvm_reboot_notifier = {
2366 	.notifier_call = kvm_reboot,
2367 	.priority = 0,
2368 };
2369 
2370 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
2371 {
2372 	int i;
2373 
2374 	for (i = 0; i < bus->dev_count; i++) {
2375 		struct kvm_io_device *pos = bus->range[i].dev;
2376 
2377 		kvm_iodevice_destructor(pos);
2378 	}
2379 	kfree(bus);
2380 }
2381 
2382 int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
2383 {
2384 	const struct kvm_io_range *r1 = p1;
2385 	const struct kvm_io_range *r2 = p2;
2386 
2387 	if (r1->addr < r2->addr)
2388 		return -1;
2389 	if (r1->addr + r1->len > r2->addr + r2->len)
2390 		return 1;
2391 	return 0;
2392 }
2393 
2394 int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
2395 			  gpa_t addr, int len)
2396 {
2397 	if (bus->dev_count == NR_IOBUS_DEVS)
2398 		return -ENOSPC;
2399 
2400 	bus->range[bus->dev_count++] = (struct kvm_io_range) {
2401 		.addr = addr,
2402 		.len = len,
2403 		.dev = dev,
2404 	};
2405 
2406 	sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
2407 		kvm_io_bus_sort_cmp, NULL);
2408 
2409 	return 0;
2410 }
2411 
2412 int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
2413 			     gpa_t addr, int len)
2414 {
2415 	struct kvm_io_range *range, key;
2416 	int off;
2417 
2418 	key = (struct kvm_io_range) {
2419 		.addr = addr,
2420 		.len = len,
2421 	};
2422 
2423 	range = bsearch(&key, bus->range, bus->dev_count,
2424 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
2425 	if (range == NULL)
2426 		return -ENOENT;
2427 
2428 	off = range - bus->range;
2429 
2430 	while (off > 0 && kvm_io_bus_sort_cmp(&key, &bus->range[off-1]) == 0)
2431 		off--;
2432 
2433 	return off;
2434 }
2435 
2436 /* kvm_io_bus_write - called under kvm->slots_lock */
2437 int kvm_io_bus_write(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2438 		     int len, const void *val)
2439 {
2440 	int idx;
2441 	struct kvm_io_bus *bus;
2442 	struct kvm_io_range range;
2443 
2444 	range = (struct kvm_io_range) {
2445 		.addr = addr,
2446 		.len = len,
2447 	};
2448 
2449 	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2450 	idx = kvm_io_bus_get_first_dev(bus, addr, len);
2451 	if (idx < 0)
2452 		return -EOPNOTSUPP;
2453 
2454 	while (idx < bus->dev_count &&
2455 		kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2456 		if (!kvm_iodevice_write(bus->range[idx].dev, addr, len, val))
2457 			return 0;
2458 		idx++;
2459 	}
2460 
2461 	return -EOPNOTSUPP;
2462 }
2463 
2464 /* kvm_io_bus_read - called under kvm->slots_lock */
2465 int kvm_io_bus_read(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2466 		    int len, void *val)
2467 {
2468 	int idx;
2469 	struct kvm_io_bus *bus;
2470 	struct kvm_io_range range;
2471 
2472 	range = (struct kvm_io_range) {
2473 		.addr = addr,
2474 		.len = len,
2475 	};
2476 
2477 	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
2478 	idx = kvm_io_bus_get_first_dev(bus, addr, len);
2479 	if (idx < 0)
2480 		return -EOPNOTSUPP;
2481 
2482 	while (idx < bus->dev_count &&
2483 		kvm_io_bus_sort_cmp(&range, &bus->range[idx]) == 0) {
2484 		if (!kvm_iodevice_read(bus->range[idx].dev, addr, len, val))
2485 			return 0;
2486 		idx++;
2487 	}
2488 
2489 	return -EOPNOTSUPP;
2490 }
2491 
2492 /* Caller must hold slots_lock. */
2493 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
2494 			    int len, struct kvm_io_device *dev)
2495 {
2496 	struct kvm_io_bus *new_bus, *bus;
2497 
2498 	bus = kvm->buses[bus_idx];
2499 	if (bus->dev_count > NR_IOBUS_DEVS-1)
2500 		return -ENOSPC;
2501 
2502 	new_bus = kmemdup(bus, sizeof(struct kvm_io_bus), GFP_KERNEL);
2503 	if (!new_bus)
2504 		return -ENOMEM;
2505 	kvm_io_bus_insert_dev(new_bus, dev, addr, len);
2506 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2507 	synchronize_srcu_expedited(&kvm->srcu);
2508 	kfree(bus);
2509 
2510 	return 0;
2511 }
2512 
2513 /* Caller must hold slots_lock. */
2514 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
2515 			      struct kvm_io_device *dev)
2516 {
2517 	int i, r;
2518 	struct kvm_io_bus *new_bus, *bus;
2519 
2520 	bus = kvm->buses[bus_idx];
2521 
2522 	new_bus = kmemdup(bus, sizeof(*bus), GFP_KERNEL);
2523 	if (!new_bus)
2524 		return -ENOMEM;
2525 
2526 	r = -ENOENT;
2527 	for (i = 0; i < new_bus->dev_count; i++)
2528 		if (new_bus->range[i].dev == dev) {
2529 			r = 0;
2530 			new_bus->dev_count--;
2531 			new_bus->range[i] = new_bus->range[new_bus->dev_count];
2532 			sort(new_bus->range, new_bus->dev_count,
2533 			     sizeof(struct kvm_io_range),
2534 			     kvm_io_bus_sort_cmp, NULL);
2535 			break;
2536 		}
2537 
2538 	if (r) {
2539 		kfree(new_bus);
2540 		return r;
2541 	}
2542 
2543 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
2544 	synchronize_srcu_expedited(&kvm->srcu);
2545 	kfree(bus);
2546 	return r;
2547 }
2548 
2549 static struct notifier_block kvm_cpu_notifier = {
2550 	.notifier_call = kvm_cpu_hotplug,
2551 };
2552 
2553 static int vm_stat_get(void *_offset, u64 *val)
2554 {
2555 	unsigned offset = (long)_offset;
2556 	struct kvm *kvm;
2557 
2558 	*val = 0;
2559 	raw_spin_lock(&kvm_lock);
2560 	list_for_each_entry(kvm, &vm_list, vm_list)
2561 		*val += *(u32 *)((void *)kvm + offset);
2562 	raw_spin_unlock(&kvm_lock);
2563 	return 0;
2564 }
2565 
2566 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
2567 
2568 static int vcpu_stat_get(void *_offset, u64 *val)
2569 {
2570 	unsigned offset = (long)_offset;
2571 	struct kvm *kvm;
2572 	struct kvm_vcpu *vcpu;
2573 	int i;
2574 
2575 	*val = 0;
2576 	raw_spin_lock(&kvm_lock);
2577 	list_for_each_entry(kvm, &vm_list, vm_list)
2578 		kvm_for_each_vcpu(i, vcpu, kvm)
2579 			*val += *(u32 *)((void *)vcpu + offset);
2580 
2581 	raw_spin_unlock(&kvm_lock);
2582 	return 0;
2583 }
2584 
2585 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
2586 
2587 static const struct file_operations *stat_fops[] = {
2588 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
2589 	[KVM_STAT_VM]   = &vm_stat_fops,
2590 };
2591 
2592 static int kvm_init_debug(void)
2593 {
2594 	int r = -EFAULT;
2595 	struct kvm_stats_debugfs_item *p;
2596 
2597 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
2598 	if (kvm_debugfs_dir == NULL)
2599 		goto out;
2600 
2601 	for (p = debugfs_entries; p->name; ++p) {
2602 		p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
2603 						(void *)(long)p->offset,
2604 						stat_fops[p->kind]);
2605 		if (p->dentry == NULL)
2606 			goto out_dir;
2607 	}
2608 
2609 	return 0;
2610 
2611 out_dir:
2612 	debugfs_remove_recursive(kvm_debugfs_dir);
2613 out:
2614 	return r;
2615 }
2616 
2617 static void kvm_exit_debug(void)
2618 {
2619 	struct kvm_stats_debugfs_item *p;
2620 
2621 	for (p = debugfs_entries; p->name; ++p)
2622 		debugfs_remove(p->dentry);
2623 	debugfs_remove(kvm_debugfs_dir);
2624 }
2625 
2626 static int kvm_suspend(void)
2627 {
2628 	if (kvm_usage_count)
2629 		hardware_disable_nolock(NULL);
2630 	return 0;
2631 }
2632 
2633 static void kvm_resume(void)
2634 {
2635 	if (kvm_usage_count) {
2636 		WARN_ON(raw_spin_is_locked(&kvm_lock));
2637 		hardware_enable_nolock(NULL);
2638 	}
2639 }
2640 
2641 static struct syscore_ops kvm_syscore_ops = {
2642 	.suspend = kvm_suspend,
2643 	.resume = kvm_resume,
2644 };
2645 
2646 struct page *bad_page;
2647 pfn_t bad_pfn;
2648 
2649 static inline
2650 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
2651 {
2652 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
2653 }
2654 
2655 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
2656 {
2657 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2658 
2659 	kvm_arch_vcpu_load(vcpu, cpu);
2660 }
2661 
2662 static void kvm_sched_out(struct preempt_notifier *pn,
2663 			  struct task_struct *next)
2664 {
2665 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
2666 
2667 	kvm_arch_vcpu_put(vcpu);
2668 }
2669 
2670 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
2671 		  struct module *module)
2672 {
2673 	int r;
2674 	int cpu;
2675 
2676 	r = kvm_arch_init(opaque);
2677 	if (r)
2678 		goto out_fail;
2679 
2680 	bad_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2681 
2682 	if (bad_page == NULL) {
2683 		r = -ENOMEM;
2684 		goto out;
2685 	}
2686 
2687 	bad_pfn = page_to_pfn(bad_page);
2688 
2689 	hwpoison_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2690 
2691 	if (hwpoison_page == NULL) {
2692 		r = -ENOMEM;
2693 		goto out_free_0;
2694 	}
2695 
2696 	hwpoison_pfn = page_to_pfn(hwpoison_page);
2697 
2698 	fault_page = alloc_page(GFP_KERNEL | __GFP_ZERO);
2699 
2700 	if (fault_page == NULL) {
2701 		r = -ENOMEM;
2702 		goto out_free_0;
2703 	}
2704 
2705 	fault_pfn = page_to_pfn(fault_page);
2706 
2707 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
2708 		r = -ENOMEM;
2709 		goto out_free_0;
2710 	}
2711 
2712 	r = kvm_arch_hardware_setup();
2713 	if (r < 0)
2714 		goto out_free_0a;
2715 
2716 	for_each_online_cpu(cpu) {
2717 		smp_call_function_single(cpu,
2718 				kvm_arch_check_processor_compat,
2719 				&r, 1);
2720 		if (r < 0)
2721 			goto out_free_1;
2722 	}
2723 
2724 	r = register_cpu_notifier(&kvm_cpu_notifier);
2725 	if (r)
2726 		goto out_free_2;
2727 	register_reboot_notifier(&kvm_reboot_notifier);
2728 
2729 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
2730 	if (!vcpu_align)
2731 		vcpu_align = __alignof__(struct kvm_vcpu);
2732 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
2733 					   0, NULL);
2734 	if (!kvm_vcpu_cache) {
2735 		r = -ENOMEM;
2736 		goto out_free_3;
2737 	}
2738 
2739 	r = kvm_async_pf_init();
2740 	if (r)
2741 		goto out_free;
2742 
2743 	kvm_chardev_ops.owner = module;
2744 	kvm_vm_fops.owner = module;
2745 	kvm_vcpu_fops.owner = module;
2746 
2747 	r = misc_register(&kvm_dev);
2748 	if (r) {
2749 		printk(KERN_ERR "kvm: misc device register failed\n");
2750 		goto out_unreg;
2751 	}
2752 
2753 	register_syscore_ops(&kvm_syscore_ops);
2754 
2755 	kvm_preempt_ops.sched_in = kvm_sched_in;
2756 	kvm_preempt_ops.sched_out = kvm_sched_out;
2757 
2758 	r = kvm_init_debug();
2759 	if (r) {
2760 		printk(KERN_ERR "kvm: create debugfs files failed\n");
2761 		goto out_undebugfs;
2762 	}
2763 
2764 	return 0;
2765 
2766 out_undebugfs:
2767 	unregister_syscore_ops(&kvm_syscore_ops);
2768 out_unreg:
2769 	kvm_async_pf_deinit();
2770 out_free:
2771 	kmem_cache_destroy(kvm_vcpu_cache);
2772 out_free_3:
2773 	unregister_reboot_notifier(&kvm_reboot_notifier);
2774 	unregister_cpu_notifier(&kvm_cpu_notifier);
2775 out_free_2:
2776 out_free_1:
2777 	kvm_arch_hardware_unsetup();
2778 out_free_0a:
2779 	free_cpumask_var(cpus_hardware_enabled);
2780 out_free_0:
2781 	if (fault_page)
2782 		__free_page(fault_page);
2783 	if (hwpoison_page)
2784 		__free_page(hwpoison_page);
2785 	__free_page(bad_page);
2786 out:
2787 	kvm_arch_exit();
2788 out_fail:
2789 	return r;
2790 }
2791 EXPORT_SYMBOL_GPL(kvm_init);
2792 
2793 void kvm_exit(void)
2794 {
2795 	kvm_exit_debug();
2796 	misc_deregister(&kvm_dev);
2797 	kmem_cache_destroy(kvm_vcpu_cache);
2798 	kvm_async_pf_deinit();
2799 	unregister_syscore_ops(&kvm_syscore_ops);
2800 	unregister_reboot_notifier(&kvm_reboot_notifier);
2801 	unregister_cpu_notifier(&kvm_cpu_notifier);
2802 	on_each_cpu(hardware_disable_nolock, NULL, 1);
2803 	kvm_arch_hardware_unsetup();
2804 	kvm_arch_exit();
2805 	free_cpumask_var(cpus_hardware_enabled);
2806 	__free_page(hwpoison_page);
2807 	__free_page(bad_page);
2808 }
2809 EXPORT_SYMBOL_GPL(kvm_exit);
2810