xref: /linux/virt/kvm/kvm_main.c (revision cc04a46f11ea046ed53e2c832ae29e4790f7e35f)
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 <kvm/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/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
58 
59 #include "coalesced_mmio.h"
60 #include "async_pf.h"
61 #include "vfio.h"
62 
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
65 
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
68 
69 /* Architectures should define their poll value according to the halt latency */
70 static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
71 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
72 
73 /* Default doubles per-vcpu halt_poll_ns. */
74 static unsigned int halt_poll_ns_grow = 2;
75 module_param(halt_poll_ns_grow, int, S_IRUGO);
76 
77 /* Default resets per-vcpu halt_poll_ns . */
78 static unsigned int halt_poll_ns_shrink;
79 module_param(halt_poll_ns_shrink, int, S_IRUGO);
80 
81 /*
82  * Ordering of locks:
83  *
84  *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
85  */
86 
87 DEFINE_SPINLOCK(kvm_lock);
88 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
89 LIST_HEAD(vm_list);
90 
91 static cpumask_var_t cpus_hardware_enabled;
92 static int kvm_usage_count;
93 static atomic_t hardware_enable_failed;
94 
95 struct kmem_cache *kvm_vcpu_cache;
96 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
97 
98 static __read_mostly struct preempt_ops kvm_preempt_ops;
99 
100 struct dentry *kvm_debugfs_dir;
101 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
102 
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
104 			   unsigned long arg);
105 #ifdef CONFIG_KVM_COMPAT
106 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
107 				  unsigned long arg);
108 #endif
109 static int hardware_enable_all(void);
110 static void hardware_disable_all(void);
111 
112 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
113 
114 static void kvm_release_pfn_dirty(pfn_t pfn);
115 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
116 
117 __visible bool kvm_rebooting;
118 EXPORT_SYMBOL_GPL(kvm_rebooting);
119 
120 static bool largepages_enabled = true;
121 
122 bool kvm_is_reserved_pfn(pfn_t pfn)
123 {
124 	if (pfn_valid(pfn))
125 		return PageReserved(pfn_to_page(pfn));
126 
127 	return true;
128 }
129 
130 /*
131  * Switches to specified vcpu, until a matching vcpu_put()
132  */
133 int vcpu_load(struct kvm_vcpu *vcpu)
134 {
135 	int cpu;
136 
137 	if (mutex_lock_killable(&vcpu->mutex))
138 		return -EINTR;
139 	cpu = get_cpu();
140 	preempt_notifier_register(&vcpu->preempt_notifier);
141 	kvm_arch_vcpu_load(vcpu, cpu);
142 	put_cpu();
143 	return 0;
144 }
145 
146 void vcpu_put(struct kvm_vcpu *vcpu)
147 {
148 	preempt_disable();
149 	kvm_arch_vcpu_put(vcpu);
150 	preempt_notifier_unregister(&vcpu->preempt_notifier);
151 	preempt_enable();
152 	mutex_unlock(&vcpu->mutex);
153 }
154 
155 static void ack_flush(void *_completed)
156 {
157 }
158 
159 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
160 {
161 	int i, cpu, me;
162 	cpumask_var_t cpus;
163 	bool called = true;
164 	struct kvm_vcpu *vcpu;
165 
166 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
167 
168 	me = get_cpu();
169 	kvm_for_each_vcpu(i, vcpu, kvm) {
170 		kvm_make_request(req, vcpu);
171 		cpu = vcpu->cpu;
172 
173 		/* Set ->requests bit before we read ->mode */
174 		smp_mb();
175 
176 		if (cpus != NULL && cpu != -1 && cpu != me &&
177 		      kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
178 			cpumask_set_cpu(cpu, cpus);
179 	}
180 	if (unlikely(cpus == NULL))
181 		smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
182 	else if (!cpumask_empty(cpus))
183 		smp_call_function_many(cpus, ack_flush, NULL, 1);
184 	else
185 		called = false;
186 	put_cpu();
187 	free_cpumask_var(cpus);
188 	return called;
189 }
190 
191 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
192 void kvm_flush_remote_tlbs(struct kvm *kvm)
193 {
194 	long dirty_count = kvm->tlbs_dirty;
195 
196 	smp_mb();
197 	if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
198 		++kvm->stat.remote_tlb_flush;
199 	cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
200 }
201 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
202 #endif
203 
204 void kvm_reload_remote_mmus(struct kvm *kvm)
205 {
206 	kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
207 }
208 
209 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
210 {
211 	kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
212 }
213 
214 void kvm_make_scan_ioapic_request(struct kvm *kvm)
215 {
216 	kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
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 	vcpu->halt_poll_ns = 0;
230 	init_waitqueue_head(&vcpu->wq);
231 	kvm_async_pf_vcpu_init(vcpu);
232 
233 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
234 	if (!page) {
235 		r = -ENOMEM;
236 		goto fail;
237 	}
238 	vcpu->run = page_address(page);
239 
240 	kvm_vcpu_set_in_spin_loop(vcpu, false);
241 	kvm_vcpu_set_dy_eligible(vcpu, false);
242 	vcpu->preempted = false;
243 
244 	r = kvm_arch_vcpu_init(vcpu);
245 	if (r < 0)
246 		goto fail_free_run;
247 	return 0;
248 
249 fail_free_run:
250 	free_page((unsigned long)vcpu->run);
251 fail:
252 	return r;
253 }
254 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
255 
256 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
257 {
258 	put_pid(vcpu->pid);
259 	kvm_arch_vcpu_uninit(vcpu);
260 	free_page((unsigned long)vcpu->run);
261 }
262 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
263 
264 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
265 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
266 {
267 	return container_of(mn, struct kvm, mmu_notifier);
268 }
269 
270 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
271 					     struct mm_struct *mm,
272 					     unsigned long address)
273 {
274 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
275 	int need_tlb_flush, idx;
276 
277 	/*
278 	 * When ->invalidate_page runs, the linux pte has been zapped
279 	 * already but the page is still allocated until
280 	 * ->invalidate_page returns. So if we increase the sequence
281 	 * here the kvm page fault will notice if the spte can't be
282 	 * established because the page is going to be freed. If
283 	 * instead the kvm page fault establishes the spte before
284 	 * ->invalidate_page runs, kvm_unmap_hva will release it
285 	 * before returning.
286 	 *
287 	 * The sequence increase only need to be seen at spin_unlock
288 	 * time, and not at spin_lock time.
289 	 *
290 	 * Increasing the sequence after the spin_unlock would be
291 	 * unsafe because the kvm page fault could then establish the
292 	 * pte after kvm_unmap_hva returned, without noticing the page
293 	 * is going to be freed.
294 	 */
295 	idx = srcu_read_lock(&kvm->srcu);
296 	spin_lock(&kvm->mmu_lock);
297 
298 	kvm->mmu_notifier_seq++;
299 	need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
300 	/* we've to flush the tlb before the pages can be freed */
301 	if (need_tlb_flush)
302 		kvm_flush_remote_tlbs(kvm);
303 
304 	spin_unlock(&kvm->mmu_lock);
305 
306 	kvm_arch_mmu_notifier_invalidate_page(kvm, address);
307 
308 	srcu_read_unlock(&kvm->srcu, idx);
309 }
310 
311 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
312 					struct mm_struct *mm,
313 					unsigned long address,
314 					pte_t pte)
315 {
316 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
317 	int idx;
318 
319 	idx = srcu_read_lock(&kvm->srcu);
320 	spin_lock(&kvm->mmu_lock);
321 	kvm->mmu_notifier_seq++;
322 	kvm_set_spte_hva(kvm, address, pte);
323 	spin_unlock(&kvm->mmu_lock);
324 	srcu_read_unlock(&kvm->srcu, idx);
325 }
326 
327 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
328 						    struct mm_struct *mm,
329 						    unsigned long start,
330 						    unsigned long end)
331 {
332 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
333 	int need_tlb_flush = 0, idx;
334 
335 	idx = srcu_read_lock(&kvm->srcu);
336 	spin_lock(&kvm->mmu_lock);
337 	/*
338 	 * The count increase must become visible at unlock time as no
339 	 * spte can be established without taking the mmu_lock and
340 	 * count is also read inside the mmu_lock critical section.
341 	 */
342 	kvm->mmu_notifier_count++;
343 	need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
344 	need_tlb_flush |= kvm->tlbs_dirty;
345 	/* we've to flush the tlb before the pages can be freed */
346 	if (need_tlb_flush)
347 		kvm_flush_remote_tlbs(kvm);
348 
349 	spin_unlock(&kvm->mmu_lock);
350 	srcu_read_unlock(&kvm->srcu, idx);
351 }
352 
353 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
354 						  struct mm_struct *mm,
355 						  unsigned long start,
356 						  unsigned long end)
357 {
358 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
359 
360 	spin_lock(&kvm->mmu_lock);
361 	/*
362 	 * This sequence increase will notify the kvm page fault that
363 	 * the page that is going to be mapped in the spte could have
364 	 * been freed.
365 	 */
366 	kvm->mmu_notifier_seq++;
367 	smp_wmb();
368 	/*
369 	 * The above sequence increase must be visible before the
370 	 * below count decrease, which is ensured by the smp_wmb above
371 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
372 	 */
373 	kvm->mmu_notifier_count--;
374 	spin_unlock(&kvm->mmu_lock);
375 
376 	BUG_ON(kvm->mmu_notifier_count < 0);
377 }
378 
379 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
380 					      struct mm_struct *mm,
381 					      unsigned long start,
382 					      unsigned long end)
383 {
384 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
385 	int young, idx;
386 
387 	idx = srcu_read_lock(&kvm->srcu);
388 	spin_lock(&kvm->mmu_lock);
389 
390 	young = kvm_age_hva(kvm, start, end);
391 	if (young)
392 		kvm_flush_remote_tlbs(kvm);
393 
394 	spin_unlock(&kvm->mmu_lock);
395 	srcu_read_unlock(&kvm->srcu, idx);
396 
397 	return young;
398 }
399 
400 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
401 					struct mm_struct *mm,
402 					unsigned long start,
403 					unsigned long end)
404 {
405 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
406 	int young, idx;
407 
408 	idx = srcu_read_lock(&kvm->srcu);
409 	spin_lock(&kvm->mmu_lock);
410 	/*
411 	 * Even though we do not flush TLB, this will still adversely
412 	 * affect performance on pre-Haswell Intel EPT, where there is
413 	 * no EPT Access Bit to clear so that we have to tear down EPT
414 	 * tables instead. If we find this unacceptable, we can always
415 	 * add a parameter to kvm_age_hva so that it effectively doesn't
416 	 * do anything on clear_young.
417 	 *
418 	 * Also note that currently we never issue secondary TLB flushes
419 	 * from clear_young, leaving this job up to the regular system
420 	 * cadence. If we find this inaccurate, we might come up with a
421 	 * more sophisticated heuristic later.
422 	 */
423 	young = kvm_age_hva(kvm, start, end);
424 	spin_unlock(&kvm->mmu_lock);
425 	srcu_read_unlock(&kvm->srcu, idx);
426 
427 	return young;
428 }
429 
430 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
431 				       struct mm_struct *mm,
432 				       unsigned long address)
433 {
434 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
435 	int young, idx;
436 
437 	idx = srcu_read_lock(&kvm->srcu);
438 	spin_lock(&kvm->mmu_lock);
439 	young = kvm_test_age_hva(kvm, address);
440 	spin_unlock(&kvm->mmu_lock);
441 	srcu_read_unlock(&kvm->srcu, idx);
442 
443 	return young;
444 }
445 
446 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
447 				     struct mm_struct *mm)
448 {
449 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
450 	int idx;
451 
452 	idx = srcu_read_lock(&kvm->srcu);
453 	kvm_arch_flush_shadow_all(kvm);
454 	srcu_read_unlock(&kvm->srcu, idx);
455 }
456 
457 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
458 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
459 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
460 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
461 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
462 	.clear_young		= kvm_mmu_notifier_clear_young,
463 	.test_young		= kvm_mmu_notifier_test_young,
464 	.change_pte		= kvm_mmu_notifier_change_pte,
465 	.release		= kvm_mmu_notifier_release,
466 };
467 
468 static int kvm_init_mmu_notifier(struct kvm *kvm)
469 {
470 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
471 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
472 }
473 
474 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
475 
476 static int kvm_init_mmu_notifier(struct kvm *kvm)
477 {
478 	return 0;
479 }
480 
481 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
482 
483 static struct kvm_memslots *kvm_alloc_memslots(void)
484 {
485 	int i;
486 	struct kvm_memslots *slots;
487 
488 	slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
489 	if (!slots)
490 		return NULL;
491 
492 	/*
493 	 * Init kvm generation close to the maximum to easily test the
494 	 * code of handling generation number wrap-around.
495 	 */
496 	slots->generation = -150;
497 	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
498 		slots->id_to_index[i] = slots->memslots[i].id = i;
499 
500 	return slots;
501 }
502 
503 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
504 {
505 	if (!memslot->dirty_bitmap)
506 		return;
507 
508 	kvfree(memslot->dirty_bitmap);
509 	memslot->dirty_bitmap = NULL;
510 }
511 
512 /*
513  * Free any memory in @free but not in @dont.
514  */
515 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
516 			      struct kvm_memory_slot *dont)
517 {
518 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
519 		kvm_destroy_dirty_bitmap(free);
520 
521 	kvm_arch_free_memslot(kvm, free, dont);
522 
523 	free->npages = 0;
524 }
525 
526 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
527 {
528 	struct kvm_memory_slot *memslot;
529 
530 	if (!slots)
531 		return;
532 
533 	kvm_for_each_memslot(memslot, slots)
534 		kvm_free_memslot(kvm, memslot, NULL);
535 
536 	kvfree(slots);
537 }
538 
539 static struct kvm *kvm_create_vm(unsigned long type)
540 {
541 	int r, i;
542 	struct kvm *kvm = kvm_arch_alloc_vm();
543 
544 	if (!kvm)
545 		return ERR_PTR(-ENOMEM);
546 
547 	r = kvm_arch_init_vm(kvm, type);
548 	if (r)
549 		goto out_err_no_disable;
550 
551 	r = hardware_enable_all();
552 	if (r)
553 		goto out_err_no_disable;
554 
555 #ifdef CONFIG_HAVE_KVM_IRQFD
556 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
557 #endif
558 
559 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
560 
561 	r = -ENOMEM;
562 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
563 		kvm->memslots[i] = kvm_alloc_memslots();
564 		if (!kvm->memslots[i])
565 			goto out_err_no_srcu;
566 	}
567 
568 	if (init_srcu_struct(&kvm->srcu))
569 		goto out_err_no_srcu;
570 	if (init_srcu_struct(&kvm->irq_srcu))
571 		goto out_err_no_irq_srcu;
572 	for (i = 0; i < KVM_NR_BUSES; i++) {
573 		kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
574 					GFP_KERNEL);
575 		if (!kvm->buses[i])
576 			goto out_err;
577 	}
578 
579 	spin_lock_init(&kvm->mmu_lock);
580 	kvm->mm = current->mm;
581 	atomic_inc(&kvm->mm->mm_count);
582 	kvm_eventfd_init(kvm);
583 	mutex_init(&kvm->lock);
584 	mutex_init(&kvm->irq_lock);
585 	mutex_init(&kvm->slots_lock);
586 	atomic_set(&kvm->users_count, 1);
587 	INIT_LIST_HEAD(&kvm->devices);
588 
589 	r = kvm_init_mmu_notifier(kvm);
590 	if (r)
591 		goto out_err;
592 
593 	spin_lock(&kvm_lock);
594 	list_add(&kvm->vm_list, &vm_list);
595 	spin_unlock(&kvm_lock);
596 
597 	preempt_notifier_inc();
598 
599 	return kvm;
600 
601 out_err:
602 	cleanup_srcu_struct(&kvm->irq_srcu);
603 out_err_no_irq_srcu:
604 	cleanup_srcu_struct(&kvm->srcu);
605 out_err_no_srcu:
606 	hardware_disable_all();
607 out_err_no_disable:
608 	for (i = 0; i < KVM_NR_BUSES; i++)
609 		kfree(kvm->buses[i]);
610 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
611 		kvm_free_memslots(kvm, kvm->memslots[i]);
612 	kvm_arch_free_vm(kvm);
613 	return ERR_PTR(r);
614 }
615 
616 /*
617  * Avoid using vmalloc for a small buffer.
618  * Should not be used when the size is statically known.
619  */
620 void *kvm_kvzalloc(unsigned long size)
621 {
622 	if (size > PAGE_SIZE)
623 		return vzalloc(size);
624 	else
625 		return kzalloc(size, GFP_KERNEL);
626 }
627 
628 static void kvm_destroy_devices(struct kvm *kvm)
629 {
630 	struct list_head *node, *tmp;
631 
632 	list_for_each_safe(node, tmp, &kvm->devices) {
633 		struct kvm_device *dev =
634 			list_entry(node, struct kvm_device, vm_node);
635 
636 		list_del(node);
637 		dev->ops->destroy(dev);
638 	}
639 }
640 
641 static void kvm_destroy_vm(struct kvm *kvm)
642 {
643 	int i;
644 	struct mm_struct *mm = kvm->mm;
645 
646 	kvm_arch_sync_events(kvm);
647 	spin_lock(&kvm_lock);
648 	list_del(&kvm->vm_list);
649 	spin_unlock(&kvm_lock);
650 	kvm_free_irq_routing(kvm);
651 	for (i = 0; i < KVM_NR_BUSES; i++)
652 		kvm_io_bus_destroy(kvm->buses[i]);
653 	kvm_coalesced_mmio_free(kvm);
654 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
655 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
656 #else
657 	kvm_arch_flush_shadow_all(kvm);
658 #endif
659 	kvm_arch_destroy_vm(kvm);
660 	kvm_destroy_devices(kvm);
661 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
662 		kvm_free_memslots(kvm, kvm->memslots[i]);
663 	cleanup_srcu_struct(&kvm->irq_srcu);
664 	cleanup_srcu_struct(&kvm->srcu);
665 	kvm_arch_free_vm(kvm);
666 	preempt_notifier_dec();
667 	hardware_disable_all();
668 	mmdrop(mm);
669 }
670 
671 void kvm_get_kvm(struct kvm *kvm)
672 {
673 	atomic_inc(&kvm->users_count);
674 }
675 EXPORT_SYMBOL_GPL(kvm_get_kvm);
676 
677 void kvm_put_kvm(struct kvm *kvm)
678 {
679 	if (atomic_dec_and_test(&kvm->users_count))
680 		kvm_destroy_vm(kvm);
681 }
682 EXPORT_SYMBOL_GPL(kvm_put_kvm);
683 
684 
685 static int kvm_vm_release(struct inode *inode, struct file *filp)
686 {
687 	struct kvm *kvm = filp->private_data;
688 
689 	kvm_irqfd_release(kvm);
690 
691 	kvm_put_kvm(kvm);
692 	return 0;
693 }
694 
695 /*
696  * Allocation size is twice as large as the actual dirty bitmap size.
697  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
698  */
699 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
700 {
701 	unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
702 
703 	memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
704 	if (!memslot->dirty_bitmap)
705 		return -ENOMEM;
706 
707 	return 0;
708 }
709 
710 /*
711  * Insert memslot and re-sort memslots based on their GFN,
712  * so binary search could be used to lookup GFN.
713  * Sorting algorithm takes advantage of having initially
714  * sorted array and known changed memslot position.
715  */
716 static void update_memslots(struct kvm_memslots *slots,
717 			    struct kvm_memory_slot *new)
718 {
719 	int id = new->id;
720 	int i = slots->id_to_index[id];
721 	struct kvm_memory_slot *mslots = slots->memslots;
722 
723 	WARN_ON(mslots[i].id != id);
724 	if (!new->npages) {
725 		WARN_ON(!mslots[i].npages);
726 		if (mslots[i].npages)
727 			slots->used_slots--;
728 	} else {
729 		if (!mslots[i].npages)
730 			slots->used_slots++;
731 	}
732 
733 	while (i < KVM_MEM_SLOTS_NUM - 1 &&
734 	       new->base_gfn <= mslots[i + 1].base_gfn) {
735 		if (!mslots[i + 1].npages)
736 			break;
737 		mslots[i] = mslots[i + 1];
738 		slots->id_to_index[mslots[i].id] = i;
739 		i++;
740 	}
741 
742 	/*
743 	 * The ">=" is needed when creating a slot with base_gfn == 0,
744 	 * so that it moves before all those with base_gfn == npages == 0.
745 	 *
746 	 * On the other hand, if new->npages is zero, the above loop has
747 	 * already left i pointing to the beginning of the empty part of
748 	 * mslots, and the ">=" would move the hole backwards in this
749 	 * case---which is wrong.  So skip the loop when deleting a slot.
750 	 */
751 	if (new->npages) {
752 		while (i > 0 &&
753 		       new->base_gfn >= mslots[i - 1].base_gfn) {
754 			mslots[i] = mslots[i - 1];
755 			slots->id_to_index[mslots[i].id] = i;
756 			i--;
757 		}
758 	} else
759 		WARN_ON_ONCE(i != slots->used_slots);
760 
761 	mslots[i] = *new;
762 	slots->id_to_index[mslots[i].id] = i;
763 }
764 
765 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
766 {
767 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
768 
769 #ifdef __KVM_HAVE_READONLY_MEM
770 	valid_flags |= KVM_MEM_READONLY;
771 #endif
772 
773 	if (mem->flags & ~valid_flags)
774 		return -EINVAL;
775 
776 	return 0;
777 }
778 
779 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
780 		int as_id, struct kvm_memslots *slots)
781 {
782 	struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
783 
784 	/*
785 	 * Set the low bit in the generation, which disables SPTE caching
786 	 * until the end of synchronize_srcu_expedited.
787 	 */
788 	WARN_ON(old_memslots->generation & 1);
789 	slots->generation = old_memslots->generation + 1;
790 
791 	rcu_assign_pointer(kvm->memslots[as_id], slots);
792 	synchronize_srcu_expedited(&kvm->srcu);
793 
794 	/*
795 	 * Increment the new memslot generation a second time. This prevents
796 	 * vm exits that race with memslot updates from caching a memslot
797 	 * generation that will (potentially) be valid forever.
798 	 */
799 	slots->generation++;
800 
801 	kvm_arch_memslots_updated(kvm, slots);
802 
803 	return old_memslots;
804 }
805 
806 /*
807  * Allocate some memory and give it an address in the guest physical address
808  * space.
809  *
810  * Discontiguous memory is allowed, mostly for framebuffers.
811  *
812  * Must be called holding kvm->slots_lock for write.
813  */
814 int __kvm_set_memory_region(struct kvm *kvm,
815 			    const struct kvm_userspace_memory_region *mem)
816 {
817 	int r;
818 	gfn_t base_gfn;
819 	unsigned long npages;
820 	struct kvm_memory_slot *slot;
821 	struct kvm_memory_slot old, new;
822 	struct kvm_memslots *slots = NULL, *old_memslots;
823 	int as_id, id;
824 	enum kvm_mr_change change;
825 
826 	r = check_memory_region_flags(mem);
827 	if (r)
828 		goto out;
829 
830 	r = -EINVAL;
831 	as_id = mem->slot >> 16;
832 	id = (u16)mem->slot;
833 
834 	/* General sanity checks */
835 	if (mem->memory_size & (PAGE_SIZE - 1))
836 		goto out;
837 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
838 		goto out;
839 	/* We can read the guest memory with __xxx_user() later on. */
840 	if ((id < KVM_USER_MEM_SLOTS) &&
841 	    ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
842 	     !access_ok(VERIFY_WRITE,
843 			(void __user *)(unsigned long)mem->userspace_addr,
844 			mem->memory_size)))
845 		goto out;
846 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
847 		goto out;
848 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
849 		goto out;
850 
851 	slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
852 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
853 	npages = mem->memory_size >> PAGE_SHIFT;
854 
855 	if (npages > KVM_MEM_MAX_NR_PAGES)
856 		goto out;
857 
858 	new = old = *slot;
859 
860 	new.id = id;
861 	new.base_gfn = base_gfn;
862 	new.npages = npages;
863 	new.flags = mem->flags;
864 
865 	if (npages) {
866 		if (!old.npages)
867 			change = KVM_MR_CREATE;
868 		else { /* Modify an existing slot. */
869 			if ((mem->userspace_addr != old.userspace_addr) ||
870 			    (npages != old.npages) ||
871 			    ((new.flags ^ old.flags) & KVM_MEM_READONLY))
872 				goto out;
873 
874 			if (base_gfn != old.base_gfn)
875 				change = KVM_MR_MOVE;
876 			else if (new.flags != old.flags)
877 				change = KVM_MR_FLAGS_ONLY;
878 			else { /* Nothing to change. */
879 				r = 0;
880 				goto out;
881 			}
882 		}
883 	} else {
884 		if (!old.npages)
885 			goto out;
886 
887 		change = KVM_MR_DELETE;
888 		new.base_gfn = 0;
889 		new.flags = 0;
890 	}
891 
892 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
893 		/* Check for overlaps */
894 		r = -EEXIST;
895 		kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
896 			if ((slot->id >= KVM_USER_MEM_SLOTS) ||
897 			    (slot->id == id))
898 				continue;
899 			if (!((base_gfn + npages <= slot->base_gfn) ||
900 			      (base_gfn >= slot->base_gfn + slot->npages)))
901 				goto out;
902 		}
903 	}
904 
905 	/* Free page dirty bitmap if unneeded */
906 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
907 		new.dirty_bitmap = NULL;
908 
909 	r = -ENOMEM;
910 	if (change == KVM_MR_CREATE) {
911 		new.userspace_addr = mem->userspace_addr;
912 
913 		if (kvm_arch_create_memslot(kvm, &new, npages))
914 			goto out_free;
915 	}
916 
917 	/* Allocate page dirty bitmap if needed */
918 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
919 		if (kvm_create_dirty_bitmap(&new) < 0)
920 			goto out_free;
921 	}
922 
923 	slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
924 	if (!slots)
925 		goto out_free;
926 	memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
927 
928 	if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
929 		slot = id_to_memslot(slots, id);
930 		slot->flags |= KVM_MEMSLOT_INVALID;
931 
932 		old_memslots = install_new_memslots(kvm, as_id, slots);
933 
934 		/* slot was deleted or moved, clear iommu mapping */
935 		kvm_iommu_unmap_pages(kvm, &old);
936 		/* From this point no new shadow pages pointing to a deleted,
937 		 * or moved, memslot will be created.
938 		 *
939 		 * validation of sp->gfn happens in:
940 		 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
941 		 *	- kvm_is_visible_gfn (mmu_check_roots)
942 		 */
943 		kvm_arch_flush_shadow_memslot(kvm, slot);
944 
945 		/*
946 		 * We can re-use the old_memslots from above, the only difference
947 		 * from the currently installed memslots is the invalid flag.  This
948 		 * will get overwritten by update_memslots anyway.
949 		 */
950 		slots = old_memslots;
951 	}
952 
953 	r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
954 	if (r)
955 		goto out_slots;
956 
957 	/* actual memory is freed via old in kvm_free_memslot below */
958 	if (change == KVM_MR_DELETE) {
959 		new.dirty_bitmap = NULL;
960 		memset(&new.arch, 0, sizeof(new.arch));
961 	}
962 
963 	update_memslots(slots, &new);
964 	old_memslots = install_new_memslots(kvm, as_id, slots);
965 
966 	kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
967 
968 	kvm_free_memslot(kvm, &old, &new);
969 	kvfree(old_memslots);
970 
971 	/*
972 	 * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
973 	 * un-mapped and re-mapped if their base changes.  Since base change
974 	 * unmapping is handled above with slot deletion, mapping alone is
975 	 * needed here.  Anything else the iommu might care about for existing
976 	 * slots (size changes, userspace addr changes and read-only flag
977 	 * changes) is disallowed above, so any other attribute changes getting
978 	 * here can be skipped.
979 	 */
980 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
981 		r = kvm_iommu_map_pages(kvm, &new);
982 		return r;
983 	}
984 
985 	return 0;
986 
987 out_slots:
988 	kvfree(slots);
989 out_free:
990 	kvm_free_memslot(kvm, &new, &old);
991 out:
992 	return r;
993 }
994 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
995 
996 int kvm_set_memory_region(struct kvm *kvm,
997 			  const struct kvm_userspace_memory_region *mem)
998 {
999 	int r;
1000 
1001 	mutex_lock(&kvm->slots_lock);
1002 	r = __kvm_set_memory_region(kvm, mem);
1003 	mutex_unlock(&kvm->slots_lock);
1004 	return r;
1005 }
1006 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1007 
1008 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1009 					  struct kvm_userspace_memory_region *mem)
1010 {
1011 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1012 		return -EINVAL;
1013 
1014 	return kvm_set_memory_region(kvm, mem);
1015 }
1016 
1017 int kvm_get_dirty_log(struct kvm *kvm,
1018 			struct kvm_dirty_log *log, int *is_dirty)
1019 {
1020 	struct kvm_memslots *slots;
1021 	struct kvm_memory_slot *memslot;
1022 	int r, i, as_id, id;
1023 	unsigned long n;
1024 	unsigned long any = 0;
1025 
1026 	r = -EINVAL;
1027 	as_id = log->slot >> 16;
1028 	id = (u16)log->slot;
1029 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1030 		goto out;
1031 
1032 	slots = __kvm_memslots(kvm, as_id);
1033 	memslot = id_to_memslot(slots, id);
1034 	r = -ENOENT;
1035 	if (!memslot->dirty_bitmap)
1036 		goto out;
1037 
1038 	n = kvm_dirty_bitmap_bytes(memslot);
1039 
1040 	for (i = 0; !any && i < n/sizeof(long); ++i)
1041 		any = memslot->dirty_bitmap[i];
1042 
1043 	r = -EFAULT;
1044 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1045 		goto out;
1046 
1047 	if (any)
1048 		*is_dirty = 1;
1049 
1050 	r = 0;
1051 out:
1052 	return r;
1053 }
1054 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1055 
1056 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1057 /**
1058  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1059  *	are dirty write protect them for next write.
1060  * @kvm:	pointer to kvm instance
1061  * @log:	slot id and address to which we copy the log
1062  * @is_dirty:	flag set if any page is dirty
1063  *
1064  * We need to keep it in mind that VCPU threads can write to the bitmap
1065  * concurrently. So, to avoid losing track of dirty pages we keep the
1066  * following order:
1067  *
1068  *    1. Take a snapshot of the bit and clear it if needed.
1069  *    2. Write protect the corresponding page.
1070  *    3. Copy the snapshot to the userspace.
1071  *    4. Upon return caller flushes TLB's if needed.
1072  *
1073  * Between 2 and 4, the guest may write to the page using the remaining TLB
1074  * entry.  This is not a problem because the page is reported dirty using
1075  * the snapshot taken before and step 4 ensures that writes done after
1076  * exiting to userspace will be logged for the next call.
1077  *
1078  */
1079 int kvm_get_dirty_log_protect(struct kvm *kvm,
1080 			struct kvm_dirty_log *log, bool *is_dirty)
1081 {
1082 	struct kvm_memslots *slots;
1083 	struct kvm_memory_slot *memslot;
1084 	int r, i, as_id, id;
1085 	unsigned long n;
1086 	unsigned long *dirty_bitmap;
1087 	unsigned long *dirty_bitmap_buffer;
1088 
1089 	r = -EINVAL;
1090 	as_id = log->slot >> 16;
1091 	id = (u16)log->slot;
1092 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1093 		goto out;
1094 
1095 	slots = __kvm_memslots(kvm, as_id);
1096 	memslot = id_to_memslot(slots, id);
1097 
1098 	dirty_bitmap = memslot->dirty_bitmap;
1099 	r = -ENOENT;
1100 	if (!dirty_bitmap)
1101 		goto out;
1102 
1103 	n = kvm_dirty_bitmap_bytes(memslot);
1104 
1105 	dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1106 	memset(dirty_bitmap_buffer, 0, n);
1107 
1108 	spin_lock(&kvm->mmu_lock);
1109 	*is_dirty = false;
1110 	for (i = 0; i < n / sizeof(long); i++) {
1111 		unsigned long mask;
1112 		gfn_t offset;
1113 
1114 		if (!dirty_bitmap[i])
1115 			continue;
1116 
1117 		*is_dirty = true;
1118 
1119 		mask = xchg(&dirty_bitmap[i], 0);
1120 		dirty_bitmap_buffer[i] = mask;
1121 
1122 		if (mask) {
1123 			offset = i * BITS_PER_LONG;
1124 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1125 								offset, mask);
1126 		}
1127 	}
1128 
1129 	spin_unlock(&kvm->mmu_lock);
1130 
1131 	r = -EFAULT;
1132 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1133 		goto out;
1134 
1135 	r = 0;
1136 out:
1137 	return r;
1138 }
1139 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1140 #endif
1141 
1142 bool kvm_largepages_enabled(void)
1143 {
1144 	return largepages_enabled;
1145 }
1146 
1147 void kvm_disable_largepages(void)
1148 {
1149 	largepages_enabled = false;
1150 }
1151 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1152 
1153 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1154 {
1155 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1156 }
1157 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1158 
1159 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1160 {
1161 	return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1162 }
1163 
1164 int kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1165 {
1166 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1167 
1168 	if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1169 	      memslot->flags & KVM_MEMSLOT_INVALID)
1170 		return 0;
1171 
1172 	return 1;
1173 }
1174 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1175 
1176 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1177 {
1178 	struct vm_area_struct *vma;
1179 	unsigned long addr, size;
1180 
1181 	size = PAGE_SIZE;
1182 
1183 	addr = gfn_to_hva(kvm, gfn);
1184 	if (kvm_is_error_hva(addr))
1185 		return PAGE_SIZE;
1186 
1187 	down_read(&current->mm->mmap_sem);
1188 	vma = find_vma(current->mm, addr);
1189 	if (!vma)
1190 		goto out;
1191 
1192 	size = vma_kernel_pagesize(vma);
1193 
1194 out:
1195 	up_read(&current->mm->mmap_sem);
1196 
1197 	return size;
1198 }
1199 
1200 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1201 {
1202 	return slot->flags & KVM_MEM_READONLY;
1203 }
1204 
1205 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1206 				       gfn_t *nr_pages, bool write)
1207 {
1208 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1209 		return KVM_HVA_ERR_BAD;
1210 
1211 	if (memslot_is_readonly(slot) && write)
1212 		return KVM_HVA_ERR_RO_BAD;
1213 
1214 	if (nr_pages)
1215 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
1216 
1217 	return __gfn_to_hva_memslot(slot, gfn);
1218 }
1219 
1220 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1221 				     gfn_t *nr_pages)
1222 {
1223 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1224 }
1225 
1226 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1227 					gfn_t gfn)
1228 {
1229 	return gfn_to_hva_many(slot, gfn, NULL);
1230 }
1231 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1232 
1233 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1234 {
1235 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1236 }
1237 EXPORT_SYMBOL_GPL(gfn_to_hva);
1238 
1239 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1240 {
1241 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1242 }
1243 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1244 
1245 /*
1246  * If writable is set to false, the hva returned by this function is only
1247  * allowed to be read.
1248  */
1249 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1250 				      gfn_t gfn, bool *writable)
1251 {
1252 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1253 
1254 	if (!kvm_is_error_hva(hva) && writable)
1255 		*writable = !memslot_is_readonly(slot);
1256 
1257 	return hva;
1258 }
1259 
1260 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1261 {
1262 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1263 
1264 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1265 }
1266 
1267 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1268 {
1269 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1270 
1271 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1272 }
1273 
1274 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1275 	unsigned long start, int write, struct page **page)
1276 {
1277 	int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1278 
1279 	if (write)
1280 		flags |= FOLL_WRITE;
1281 
1282 	return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1283 }
1284 
1285 static inline int check_user_page_hwpoison(unsigned long addr)
1286 {
1287 	int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1288 
1289 	rc = __get_user_pages(current, current->mm, addr, 1,
1290 			      flags, NULL, NULL, NULL);
1291 	return rc == -EHWPOISON;
1292 }
1293 
1294 /*
1295  * The atomic path to get the writable pfn which will be stored in @pfn,
1296  * true indicates success, otherwise false is returned.
1297  */
1298 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1299 			    bool write_fault, bool *writable, pfn_t *pfn)
1300 {
1301 	struct page *page[1];
1302 	int npages;
1303 
1304 	if (!(async || atomic))
1305 		return false;
1306 
1307 	/*
1308 	 * Fast pin a writable pfn only if it is a write fault request
1309 	 * or the caller allows to map a writable pfn for a read fault
1310 	 * request.
1311 	 */
1312 	if (!(write_fault || writable))
1313 		return false;
1314 
1315 	npages = __get_user_pages_fast(addr, 1, 1, page);
1316 	if (npages == 1) {
1317 		*pfn = page_to_pfn(page[0]);
1318 
1319 		if (writable)
1320 			*writable = true;
1321 		return true;
1322 	}
1323 
1324 	return false;
1325 }
1326 
1327 /*
1328  * The slow path to get the pfn of the specified host virtual address,
1329  * 1 indicates success, -errno is returned if error is detected.
1330  */
1331 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1332 			   bool *writable, pfn_t *pfn)
1333 {
1334 	struct page *page[1];
1335 	int npages = 0;
1336 
1337 	might_sleep();
1338 
1339 	if (writable)
1340 		*writable = write_fault;
1341 
1342 	if (async) {
1343 		down_read(&current->mm->mmap_sem);
1344 		npages = get_user_page_nowait(current, current->mm,
1345 					      addr, write_fault, page);
1346 		up_read(&current->mm->mmap_sem);
1347 	} else
1348 		npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1349 						   write_fault, 0, page,
1350 						   FOLL_TOUCH|FOLL_HWPOISON);
1351 	if (npages != 1)
1352 		return npages;
1353 
1354 	/* map read fault as writable if possible */
1355 	if (unlikely(!write_fault) && writable) {
1356 		struct page *wpage[1];
1357 
1358 		npages = __get_user_pages_fast(addr, 1, 1, wpage);
1359 		if (npages == 1) {
1360 			*writable = true;
1361 			put_page(page[0]);
1362 			page[0] = wpage[0];
1363 		}
1364 
1365 		npages = 1;
1366 	}
1367 	*pfn = page_to_pfn(page[0]);
1368 	return npages;
1369 }
1370 
1371 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1372 {
1373 	if (unlikely(!(vma->vm_flags & VM_READ)))
1374 		return false;
1375 
1376 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1377 		return false;
1378 
1379 	return true;
1380 }
1381 
1382 /*
1383  * Pin guest page in memory and return its pfn.
1384  * @addr: host virtual address which maps memory to the guest
1385  * @atomic: whether this function can sleep
1386  * @async: whether this function need to wait IO complete if the
1387  *         host page is not in the memory
1388  * @write_fault: whether we should get a writable host page
1389  * @writable: whether it allows to map a writable host page for !@write_fault
1390  *
1391  * The function will map a writable host page for these two cases:
1392  * 1): @write_fault = true
1393  * 2): @write_fault = false && @writable, @writable will tell the caller
1394  *     whether the mapping is writable.
1395  */
1396 static pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1397 			bool write_fault, bool *writable)
1398 {
1399 	struct vm_area_struct *vma;
1400 	pfn_t pfn = 0;
1401 	int npages;
1402 
1403 	/* we can do it either atomically or asynchronously, not both */
1404 	BUG_ON(atomic && async);
1405 
1406 	if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1407 		return pfn;
1408 
1409 	if (atomic)
1410 		return KVM_PFN_ERR_FAULT;
1411 
1412 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1413 	if (npages == 1)
1414 		return pfn;
1415 
1416 	down_read(&current->mm->mmap_sem);
1417 	if (npages == -EHWPOISON ||
1418 	      (!async && check_user_page_hwpoison(addr))) {
1419 		pfn = KVM_PFN_ERR_HWPOISON;
1420 		goto exit;
1421 	}
1422 
1423 	vma = find_vma_intersection(current->mm, addr, addr + 1);
1424 
1425 	if (vma == NULL)
1426 		pfn = KVM_PFN_ERR_FAULT;
1427 	else if ((vma->vm_flags & VM_PFNMAP)) {
1428 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1429 			vma->vm_pgoff;
1430 		BUG_ON(!kvm_is_reserved_pfn(pfn));
1431 	} else {
1432 		if (async && vma_is_valid(vma, write_fault))
1433 			*async = true;
1434 		pfn = KVM_PFN_ERR_FAULT;
1435 	}
1436 exit:
1437 	up_read(&current->mm->mmap_sem);
1438 	return pfn;
1439 }
1440 
1441 pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn, bool atomic,
1442 			   bool *async, bool write_fault, bool *writable)
1443 {
1444 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1445 
1446 	if (addr == KVM_HVA_ERR_RO_BAD)
1447 		return KVM_PFN_ERR_RO_FAULT;
1448 
1449 	if (kvm_is_error_hva(addr))
1450 		return KVM_PFN_NOSLOT;
1451 
1452 	/* Do not map writable pfn in the readonly memslot. */
1453 	if (writable && memslot_is_readonly(slot)) {
1454 		*writable = false;
1455 		writable = NULL;
1456 	}
1457 
1458 	return hva_to_pfn(addr, atomic, async, write_fault,
1459 			  writable);
1460 }
1461 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1462 
1463 pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1464 		      bool *writable)
1465 {
1466 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1467 				    write_fault, writable);
1468 }
1469 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1470 
1471 pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1472 {
1473 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1474 }
1475 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1476 
1477 pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1478 {
1479 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1480 }
1481 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1482 
1483 pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1484 {
1485 	return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1486 }
1487 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1488 
1489 pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1490 {
1491 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1492 }
1493 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1494 
1495 pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1496 {
1497 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1498 }
1499 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1500 
1501 pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1502 {
1503 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1504 }
1505 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1506 
1507 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1508 			    struct page **pages, int nr_pages)
1509 {
1510 	unsigned long addr;
1511 	gfn_t entry;
1512 
1513 	addr = gfn_to_hva_many(slot, gfn, &entry);
1514 	if (kvm_is_error_hva(addr))
1515 		return -1;
1516 
1517 	if (entry < nr_pages)
1518 		return 0;
1519 
1520 	return __get_user_pages_fast(addr, nr_pages, 1, pages);
1521 }
1522 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1523 
1524 static struct page *kvm_pfn_to_page(pfn_t pfn)
1525 {
1526 	if (is_error_noslot_pfn(pfn))
1527 		return KVM_ERR_PTR_BAD_PAGE;
1528 
1529 	if (kvm_is_reserved_pfn(pfn)) {
1530 		WARN_ON(1);
1531 		return KVM_ERR_PTR_BAD_PAGE;
1532 	}
1533 
1534 	return pfn_to_page(pfn);
1535 }
1536 
1537 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1538 {
1539 	pfn_t pfn;
1540 
1541 	pfn = gfn_to_pfn(kvm, gfn);
1542 
1543 	return kvm_pfn_to_page(pfn);
1544 }
1545 EXPORT_SYMBOL_GPL(gfn_to_page);
1546 
1547 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1548 {
1549 	pfn_t pfn;
1550 
1551 	pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1552 
1553 	return kvm_pfn_to_page(pfn);
1554 }
1555 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1556 
1557 void kvm_release_page_clean(struct page *page)
1558 {
1559 	WARN_ON(is_error_page(page));
1560 
1561 	kvm_release_pfn_clean(page_to_pfn(page));
1562 }
1563 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1564 
1565 void kvm_release_pfn_clean(pfn_t pfn)
1566 {
1567 	if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1568 		put_page(pfn_to_page(pfn));
1569 }
1570 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1571 
1572 void kvm_release_page_dirty(struct page *page)
1573 {
1574 	WARN_ON(is_error_page(page));
1575 
1576 	kvm_release_pfn_dirty(page_to_pfn(page));
1577 }
1578 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1579 
1580 static void kvm_release_pfn_dirty(pfn_t pfn)
1581 {
1582 	kvm_set_pfn_dirty(pfn);
1583 	kvm_release_pfn_clean(pfn);
1584 }
1585 
1586 void kvm_set_pfn_dirty(pfn_t pfn)
1587 {
1588 	if (!kvm_is_reserved_pfn(pfn)) {
1589 		struct page *page = pfn_to_page(pfn);
1590 
1591 		if (!PageReserved(page))
1592 			SetPageDirty(page);
1593 	}
1594 }
1595 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1596 
1597 void kvm_set_pfn_accessed(pfn_t pfn)
1598 {
1599 	if (!kvm_is_reserved_pfn(pfn))
1600 		mark_page_accessed(pfn_to_page(pfn));
1601 }
1602 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1603 
1604 void kvm_get_pfn(pfn_t pfn)
1605 {
1606 	if (!kvm_is_reserved_pfn(pfn))
1607 		get_page(pfn_to_page(pfn));
1608 }
1609 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1610 
1611 static int next_segment(unsigned long len, int offset)
1612 {
1613 	if (len > PAGE_SIZE - offset)
1614 		return PAGE_SIZE - offset;
1615 	else
1616 		return len;
1617 }
1618 
1619 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1620 				 void *data, int offset, int len)
1621 {
1622 	int r;
1623 	unsigned long addr;
1624 
1625 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1626 	if (kvm_is_error_hva(addr))
1627 		return -EFAULT;
1628 	r = __copy_from_user(data, (void __user *)addr + offset, len);
1629 	if (r)
1630 		return -EFAULT;
1631 	return 0;
1632 }
1633 
1634 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1635 			int len)
1636 {
1637 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1638 
1639 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1640 }
1641 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1642 
1643 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1644 			     int offset, int len)
1645 {
1646 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1647 
1648 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1649 }
1650 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1651 
1652 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1653 {
1654 	gfn_t gfn = gpa >> PAGE_SHIFT;
1655 	int seg;
1656 	int offset = offset_in_page(gpa);
1657 	int ret;
1658 
1659 	while ((seg = next_segment(len, offset)) != 0) {
1660 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1661 		if (ret < 0)
1662 			return ret;
1663 		offset = 0;
1664 		len -= seg;
1665 		data += seg;
1666 		++gfn;
1667 	}
1668 	return 0;
1669 }
1670 EXPORT_SYMBOL_GPL(kvm_read_guest);
1671 
1672 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1673 {
1674 	gfn_t gfn = gpa >> PAGE_SHIFT;
1675 	int seg;
1676 	int offset = offset_in_page(gpa);
1677 	int ret;
1678 
1679 	while ((seg = next_segment(len, offset)) != 0) {
1680 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1681 		if (ret < 0)
1682 			return ret;
1683 		offset = 0;
1684 		len -= seg;
1685 		data += seg;
1686 		++gfn;
1687 	}
1688 	return 0;
1689 }
1690 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1691 
1692 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1693 			           void *data, int offset, unsigned long len)
1694 {
1695 	int r;
1696 	unsigned long addr;
1697 
1698 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1699 	if (kvm_is_error_hva(addr))
1700 		return -EFAULT;
1701 	pagefault_disable();
1702 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1703 	pagefault_enable();
1704 	if (r)
1705 		return -EFAULT;
1706 	return 0;
1707 }
1708 
1709 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1710 			  unsigned long len)
1711 {
1712 	gfn_t gfn = gpa >> PAGE_SHIFT;
1713 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1714 	int offset = offset_in_page(gpa);
1715 
1716 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1717 }
1718 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1719 
1720 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1721 			       void *data, unsigned long len)
1722 {
1723 	gfn_t gfn = gpa >> PAGE_SHIFT;
1724 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1725 	int offset = offset_in_page(gpa);
1726 
1727 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1728 }
1729 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1730 
1731 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1732 			          const void *data, int offset, int len)
1733 {
1734 	int r;
1735 	unsigned long addr;
1736 
1737 	addr = gfn_to_hva_memslot(memslot, gfn);
1738 	if (kvm_is_error_hva(addr))
1739 		return -EFAULT;
1740 	r = __copy_to_user((void __user *)addr + offset, data, len);
1741 	if (r)
1742 		return -EFAULT;
1743 	mark_page_dirty_in_slot(memslot, gfn);
1744 	return 0;
1745 }
1746 
1747 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1748 			 const void *data, int offset, int len)
1749 {
1750 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1751 
1752 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1753 }
1754 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1755 
1756 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1757 			      const void *data, int offset, int len)
1758 {
1759 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1760 
1761 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1762 }
1763 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1764 
1765 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1766 		    unsigned long len)
1767 {
1768 	gfn_t gfn = gpa >> PAGE_SHIFT;
1769 	int seg;
1770 	int offset = offset_in_page(gpa);
1771 	int ret;
1772 
1773 	while ((seg = next_segment(len, offset)) != 0) {
1774 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1775 		if (ret < 0)
1776 			return ret;
1777 		offset = 0;
1778 		len -= seg;
1779 		data += seg;
1780 		++gfn;
1781 	}
1782 	return 0;
1783 }
1784 EXPORT_SYMBOL_GPL(kvm_write_guest);
1785 
1786 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1787 		         unsigned long len)
1788 {
1789 	gfn_t gfn = gpa >> PAGE_SHIFT;
1790 	int seg;
1791 	int offset = offset_in_page(gpa);
1792 	int ret;
1793 
1794 	while ((seg = next_segment(len, offset)) != 0) {
1795 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1796 		if (ret < 0)
1797 			return ret;
1798 		offset = 0;
1799 		len -= seg;
1800 		data += seg;
1801 		++gfn;
1802 	}
1803 	return 0;
1804 }
1805 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1806 
1807 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1808 			      gpa_t gpa, unsigned long len)
1809 {
1810 	struct kvm_memslots *slots = kvm_memslots(kvm);
1811 	int offset = offset_in_page(gpa);
1812 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
1813 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1814 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1815 	gfn_t nr_pages_avail;
1816 
1817 	ghc->gpa = gpa;
1818 	ghc->generation = slots->generation;
1819 	ghc->len = len;
1820 	ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1821 	ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1822 	if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1823 		ghc->hva += offset;
1824 	} else {
1825 		/*
1826 		 * If the requested region crosses two memslots, we still
1827 		 * verify that the entire region is valid here.
1828 		 */
1829 		while (start_gfn <= end_gfn) {
1830 			ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1831 			ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1832 						   &nr_pages_avail);
1833 			if (kvm_is_error_hva(ghc->hva))
1834 				return -EFAULT;
1835 			start_gfn += nr_pages_avail;
1836 		}
1837 		/* Use the slow path for cross page reads and writes. */
1838 		ghc->memslot = NULL;
1839 	}
1840 	return 0;
1841 }
1842 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1843 
1844 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1845 			   void *data, unsigned long len)
1846 {
1847 	struct kvm_memslots *slots = kvm_memslots(kvm);
1848 	int r;
1849 
1850 	BUG_ON(len > ghc->len);
1851 
1852 	if (slots->generation != ghc->generation)
1853 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1854 
1855 	if (unlikely(!ghc->memslot))
1856 		return kvm_write_guest(kvm, ghc->gpa, data, len);
1857 
1858 	if (kvm_is_error_hva(ghc->hva))
1859 		return -EFAULT;
1860 
1861 	r = __copy_to_user((void __user *)ghc->hva, data, len);
1862 	if (r)
1863 		return -EFAULT;
1864 	mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1865 
1866 	return 0;
1867 }
1868 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1869 
1870 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1871 			   void *data, unsigned long len)
1872 {
1873 	struct kvm_memslots *slots = kvm_memslots(kvm);
1874 	int r;
1875 
1876 	BUG_ON(len > ghc->len);
1877 
1878 	if (slots->generation != ghc->generation)
1879 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1880 
1881 	if (unlikely(!ghc->memslot))
1882 		return kvm_read_guest(kvm, ghc->gpa, data, len);
1883 
1884 	if (kvm_is_error_hva(ghc->hva))
1885 		return -EFAULT;
1886 
1887 	r = __copy_from_user(data, (void __user *)ghc->hva, len);
1888 	if (r)
1889 		return -EFAULT;
1890 
1891 	return 0;
1892 }
1893 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1894 
1895 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1896 {
1897 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1898 
1899 	return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1900 }
1901 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1902 
1903 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1904 {
1905 	gfn_t gfn = gpa >> PAGE_SHIFT;
1906 	int seg;
1907 	int offset = offset_in_page(gpa);
1908 	int ret;
1909 
1910 	while ((seg = next_segment(len, offset)) != 0) {
1911 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1912 		if (ret < 0)
1913 			return ret;
1914 		offset = 0;
1915 		len -= seg;
1916 		++gfn;
1917 	}
1918 	return 0;
1919 }
1920 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1921 
1922 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1923 				    gfn_t gfn)
1924 {
1925 	if (memslot && memslot->dirty_bitmap) {
1926 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1927 
1928 		set_bit_le(rel_gfn, memslot->dirty_bitmap);
1929 	}
1930 }
1931 
1932 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1933 {
1934 	struct kvm_memory_slot *memslot;
1935 
1936 	memslot = gfn_to_memslot(kvm, gfn);
1937 	mark_page_dirty_in_slot(memslot, gfn);
1938 }
1939 EXPORT_SYMBOL_GPL(mark_page_dirty);
1940 
1941 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1942 {
1943 	struct kvm_memory_slot *memslot;
1944 
1945 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1946 	mark_page_dirty_in_slot(memslot, gfn);
1947 }
1948 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1949 
1950 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1951 {
1952 	int old, val;
1953 
1954 	old = val = vcpu->halt_poll_ns;
1955 	/* 10us base */
1956 	if (val == 0 && halt_poll_ns_grow)
1957 		val = 10000;
1958 	else
1959 		val *= halt_poll_ns_grow;
1960 
1961 	vcpu->halt_poll_ns = val;
1962 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1963 }
1964 
1965 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1966 {
1967 	int old, val;
1968 
1969 	old = val = vcpu->halt_poll_ns;
1970 	if (halt_poll_ns_shrink == 0)
1971 		val = 0;
1972 	else
1973 		val /= halt_poll_ns_shrink;
1974 
1975 	vcpu->halt_poll_ns = val;
1976 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1977 }
1978 
1979 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1980 {
1981 	if (kvm_arch_vcpu_runnable(vcpu)) {
1982 		kvm_make_request(KVM_REQ_UNHALT, vcpu);
1983 		return -EINTR;
1984 	}
1985 	if (kvm_cpu_has_pending_timer(vcpu))
1986 		return -EINTR;
1987 	if (signal_pending(current))
1988 		return -EINTR;
1989 
1990 	return 0;
1991 }
1992 
1993 /*
1994  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1995  */
1996 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1997 {
1998 	ktime_t start, cur;
1999 	DEFINE_WAIT(wait);
2000 	bool waited = false;
2001 	u64 block_ns;
2002 
2003 	start = cur = ktime_get();
2004 	if (vcpu->halt_poll_ns) {
2005 		ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2006 
2007 		++vcpu->stat.halt_attempted_poll;
2008 		do {
2009 			/*
2010 			 * This sets KVM_REQ_UNHALT if an interrupt
2011 			 * arrives.
2012 			 */
2013 			if (kvm_vcpu_check_block(vcpu) < 0) {
2014 				++vcpu->stat.halt_successful_poll;
2015 				goto out;
2016 			}
2017 			cur = ktime_get();
2018 		} while (single_task_running() && ktime_before(cur, stop));
2019 	}
2020 
2021 	for (;;) {
2022 		prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2023 
2024 		if (kvm_vcpu_check_block(vcpu) < 0)
2025 			break;
2026 
2027 		waited = true;
2028 		schedule();
2029 	}
2030 
2031 	finish_wait(&vcpu->wq, &wait);
2032 	cur = ktime_get();
2033 
2034 out:
2035 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2036 
2037 	if (halt_poll_ns) {
2038 		if (block_ns <= vcpu->halt_poll_ns)
2039 			;
2040 		/* we had a long block, shrink polling */
2041 		else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2042 			shrink_halt_poll_ns(vcpu);
2043 		/* we had a short halt and our poll time is too small */
2044 		else if (vcpu->halt_poll_ns < halt_poll_ns &&
2045 			block_ns < halt_poll_ns)
2046 			grow_halt_poll_ns(vcpu);
2047 	} else
2048 		vcpu->halt_poll_ns = 0;
2049 
2050 	trace_kvm_vcpu_wakeup(block_ns, waited);
2051 }
2052 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2053 
2054 #ifndef CONFIG_S390
2055 /*
2056  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2057  */
2058 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2059 {
2060 	int me;
2061 	int cpu = vcpu->cpu;
2062 	wait_queue_head_t *wqp;
2063 
2064 	wqp = kvm_arch_vcpu_wq(vcpu);
2065 	if (waitqueue_active(wqp)) {
2066 		wake_up_interruptible(wqp);
2067 		++vcpu->stat.halt_wakeup;
2068 	}
2069 
2070 	me = get_cpu();
2071 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2072 		if (kvm_arch_vcpu_should_kick(vcpu))
2073 			smp_send_reschedule(cpu);
2074 	put_cpu();
2075 }
2076 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2077 #endif /* !CONFIG_S390 */
2078 
2079 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2080 {
2081 	struct pid *pid;
2082 	struct task_struct *task = NULL;
2083 	int ret = 0;
2084 
2085 	rcu_read_lock();
2086 	pid = rcu_dereference(target->pid);
2087 	if (pid)
2088 		task = get_pid_task(pid, PIDTYPE_PID);
2089 	rcu_read_unlock();
2090 	if (!task)
2091 		return ret;
2092 	ret = yield_to(task, 1);
2093 	put_task_struct(task);
2094 
2095 	return ret;
2096 }
2097 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2098 
2099 /*
2100  * Helper that checks whether a VCPU is eligible for directed yield.
2101  * Most eligible candidate to yield is decided by following heuristics:
2102  *
2103  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2104  *  (preempted lock holder), indicated by @in_spin_loop.
2105  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2106  *
2107  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2108  *  chance last time (mostly it has become eligible now since we have probably
2109  *  yielded to lockholder in last iteration. This is done by toggling
2110  *  @dy_eligible each time a VCPU checked for eligibility.)
2111  *
2112  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2113  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2114  *  burning. Giving priority for a potential lock-holder increases lock
2115  *  progress.
2116  *
2117  *  Since algorithm is based on heuristics, accessing another VCPU data without
2118  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2119  *  and continue with next VCPU and so on.
2120  */
2121 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2122 {
2123 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2124 	bool eligible;
2125 
2126 	eligible = !vcpu->spin_loop.in_spin_loop ||
2127 		    vcpu->spin_loop.dy_eligible;
2128 
2129 	if (vcpu->spin_loop.in_spin_loop)
2130 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2131 
2132 	return eligible;
2133 #else
2134 	return true;
2135 #endif
2136 }
2137 
2138 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2139 {
2140 	struct kvm *kvm = me->kvm;
2141 	struct kvm_vcpu *vcpu;
2142 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2143 	int yielded = 0;
2144 	int try = 3;
2145 	int pass;
2146 	int i;
2147 
2148 	kvm_vcpu_set_in_spin_loop(me, true);
2149 	/*
2150 	 * We boost the priority of a VCPU that is runnable but not
2151 	 * currently running, because it got preempted by something
2152 	 * else and called schedule in __vcpu_run.  Hopefully that
2153 	 * VCPU is holding the lock that we need and will release it.
2154 	 * We approximate round-robin by starting at the last boosted VCPU.
2155 	 */
2156 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
2157 		kvm_for_each_vcpu(i, vcpu, kvm) {
2158 			if (!pass && i <= last_boosted_vcpu) {
2159 				i = last_boosted_vcpu;
2160 				continue;
2161 			} else if (pass && i > last_boosted_vcpu)
2162 				break;
2163 			if (!ACCESS_ONCE(vcpu->preempted))
2164 				continue;
2165 			if (vcpu == me)
2166 				continue;
2167 			if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2168 				continue;
2169 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2170 				continue;
2171 
2172 			yielded = kvm_vcpu_yield_to(vcpu);
2173 			if (yielded > 0) {
2174 				kvm->last_boosted_vcpu = i;
2175 				break;
2176 			} else if (yielded < 0) {
2177 				try--;
2178 				if (!try)
2179 					break;
2180 			}
2181 		}
2182 	}
2183 	kvm_vcpu_set_in_spin_loop(me, false);
2184 
2185 	/* Ensure vcpu is not eligible during next spinloop */
2186 	kvm_vcpu_set_dy_eligible(me, false);
2187 }
2188 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2189 
2190 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2191 {
2192 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2193 	struct page *page;
2194 
2195 	if (vmf->pgoff == 0)
2196 		page = virt_to_page(vcpu->run);
2197 #ifdef CONFIG_X86
2198 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2199 		page = virt_to_page(vcpu->arch.pio_data);
2200 #endif
2201 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2202 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2203 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2204 #endif
2205 	else
2206 		return kvm_arch_vcpu_fault(vcpu, vmf);
2207 	get_page(page);
2208 	vmf->page = page;
2209 	return 0;
2210 }
2211 
2212 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2213 	.fault = kvm_vcpu_fault,
2214 };
2215 
2216 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2217 {
2218 	vma->vm_ops = &kvm_vcpu_vm_ops;
2219 	return 0;
2220 }
2221 
2222 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2223 {
2224 	struct kvm_vcpu *vcpu = filp->private_data;
2225 
2226 	kvm_put_kvm(vcpu->kvm);
2227 	return 0;
2228 }
2229 
2230 static struct file_operations kvm_vcpu_fops = {
2231 	.release        = kvm_vcpu_release,
2232 	.unlocked_ioctl = kvm_vcpu_ioctl,
2233 #ifdef CONFIG_KVM_COMPAT
2234 	.compat_ioctl   = kvm_vcpu_compat_ioctl,
2235 #endif
2236 	.mmap           = kvm_vcpu_mmap,
2237 	.llseek		= noop_llseek,
2238 };
2239 
2240 /*
2241  * Allocates an inode for the vcpu.
2242  */
2243 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2244 {
2245 	return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2246 }
2247 
2248 /*
2249  * Creates some virtual cpus.  Good luck creating more than one.
2250  */
2251 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2252 {
2253 	int r;
2254 	struct kvm_vcpu *vcpu, *v;
2255 
2256 	if (id >= KVM_MAX_VCPUS)
2257 		return -EINVAL;
2258 
2259 	vcpu = kvm_arch_vcpu_create(kvm, id);
2260 	if (IS_ERR(vcpu))
2261 		return PTR_ERR(vcpu);
2262 
2263 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2264 
2265 	r = kvm_arch_vcpu_setup(vcpu);
2266 	if (r)
2267 		goto vcpu_destroy;
2268 
2269 	mutex_lock(&kvm->lock);
2270 	if (!kvm_vcpu_compatible(vcpu)) {
2271 		r = -EINVAL;
2272 		goto unlock_vcpu_destroy;
2273 	}
2274 	if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2275 		r = -EINVAL;
2276 		goto unlock_vcpu_destroy;
2277 	}
2278 
2279 	kvm_for_each_vcpu(r, v, kvm)
2280 		if (v->vcpu_id == id) {
2281 			r = -EEXIST;
2282 			goto unlock_vcpu_destroy;
2283 		}
2284 
2285 	BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2286 
2287 	/* Now it's all set up, let userspace reach it */
2288 	kvm_get_kvm(kvm);
2289 	r = create_vcpu_fd(vcpu);
2290 	if (r < 0) {
2291 		kvm_put_kvm(kvm);
2292 		goto unlock_vcpu_destroy;
2293 	}
2294 
2295 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2296 
2297 	/*
2298 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2299 	 * before kvm->online_vcpu's incremented value.
2300 	 */
2301 	smp_wmb();
2302 	atomic_inc(&kvm->online_vcpus);
2303 
2304 	mutex_unlock(&kvm->lock);
2305 	kvm_arch_vcpu_postcreate(vcpu);
2306 	return r;
2307 
2308 unlock_vcpu_destroy:
2309 	mutex_unlock(&kvm->lock);
2310 vcpu_destroy:
2311 	kvm_arch_vcpu_destroy(vcpu);
2312 	return r;
2313 }
2314 
2315 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2316 {
2317 	if (sigset) {
2318 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2319 		vcpu->sigset_active = 1;
2320 		vcpu->sigset = *sigset;
2321 	} else
2322 		vcpu->sigset_active = 0;
2323 	return 0;
2324 }
2325 
2326 static long kvm_vcpu_ioctl(struct file *filp,
2327 			   unsigned int ioctl, unsigned long arg)
2328 {
2329 	struct kvm_vcpu *vcpu = filp->private_data;
2330 	void __user *argp = (void __user *)arg;
2331 	int r;
2332 	struct kvm_fpu *fpu = NULL;
2333 	struct kvm_sregs *kvm_sregs = NULL;
2334 
2335 	if (vcpu->kvm->mm != current->mm)
2336 		return -EIO;
2337 
2338 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2339 		return -EINVAL;
2340 
2341 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2342 	/*
2343 	 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2344 	 * so vcpu_load() would break it.
2345 	 */
2346 	if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2347 		return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2348 #endif
2349 
2350 
2351 	r = vcpu_load(vcpu);
2352 	if (r)
2353 		return r;
2354 	switch (ioctl) {
2355 	case KVM_RUN:
2356 		r = -EINVAL;
2357 		if (arg)
2358 			goto out;
2359 		if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2360 			/* The thread running this VCPU changed. */
2361 			struct pid *oldpid = vcpu->pid;
2362 			struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2363 
2364 			rcu_assign_pointer(vcpu->pid, newpid);
2365 			if (oldpid)
2366 				synchronize_rcu();
2367 			put_pid(oldpid);
2368 		}
2369 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2370 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2371 		break;
2372 	case KVM_GET_REGS: {
2373 		struct kvm_regs *kvm_regs;
2374 
2375 		r = -ENOMEM;
2376 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2377 		if (!kvm_regs)
2378 			goto out;
2379 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2380 		if (r)
2381 			goto out_free1;
2382 		r = -EFAULT;
2383 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2384 			goto out_free1;
2385 		r = 0;
2386 out_free1:
2387 		kfree(kvm_regs);
2388 		break;
2389 	}
2390 	case KVM_SET_REGS: {
2391 		struct kvm_regs *kvm_regs;
2392 
2393 		r = -ENOMEM;
2394 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2395 		if (IS_ERR(kvm_regs)) {
2396 			r = PTR_ERR(kvm_regs);
2397 			goto out;
2398 		}
2399 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2400 		kfree(kvm_regs);
2401 		break;
2402 	}
2403 	case KVM_GET_SREGS: {
2404 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2405 		r = -ENOMEM;
2406 		if (!kvm_sregs)
2407 			goto out;
2408 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2409 		if (r)
2410 			goto out;
2411 		r = -EFAULT;
2412 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2413 			goto out;
2414 		r = 0;
2415 		break;
2416 	}
2417 	case KVM_SET_SREGS: {
2418 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2419 		if (IS_ERR(kvm_sregs)) {
2420 			r = PTR_ERR(kvm_sregs);
2421 			kvm_sregs = NULL;
2422 			goto out;
2423 		}
2424 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2425 		break;
2426 	}
2427 	case KVM_GET_MP_STATE: {
2428 		struct kvm_mp_state mp_state;
2429 
2430 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2431 		if (r)
2432 			goto out;
2433 		r = -EFAULT;
2434 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2435 			goto out;
2436 		r = 0;
2437 		break;
2438 	}
2439 	case KVM_SET_MP_STATE: {
2440 		struct kvm_mp_state mp_state;
2441 
2442 		r = -EFAULT;
2443 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2444 			goto out;
2445 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2446 		break;
2447 	}
2448 	case KVM_TRANSLATE: {
2449 		struct kvm_translation tr;
2450 
2451 		r = -EFAULT;
2452 		if (copy_from_user(&tr, argp, sizeof(tr)))
2453 			goto out;
2454 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2455 		if (r)
2456 			goto out;
2457 		r = -EFAULT;
2458 		if (copy_to_user(argp, &tr, sizeof(tr)))
2459 			goto out;
2460 		r = 0;
2461 		break;
2462 	}
2463 	case KVM_SET_GUEST_DEBUG: {
2464 		struct kvm_guest_debug dbg;
2465 
2466 		r = -EFAULT;
2467 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
2468 			goto out;
2469 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2470 		break;
2471 	}
2472 	case KVM_SET_SIGNAL_MASK: {
2473 		struct kvm_signal_mask __user *sigmask_arg = argp;
2474 		struct kvm_signal_mask kvm_sigmask;
2475 		sigset_t sigset, *p;
2476 
2477 		p = NULL;
2478 		if (argp) {
2479 			r = -EFAULT;
2480 			if (copy_from_user(&kvm_sigmask, argp,
2481 					   sizeof(kvm_sigmask)))
2482 				goto out;
2483 			r = -EINVAL;
2484 			if (kvm_sigmask.len != sizeof(sigset))
2485 				goto out;
2486 			r = -EFAULT;
2487 			if (copy_from_user(&sigset, sigmask_arg->sigset,
2488 					   sizeof(sigset)))
2489 				goto out;
2490 			p = &sigset;
2491 		}
2492 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2493 		break;
2494 	}
2495 	case KVM_GET_FPU: {
2496 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2497 		r = -ENOMEM;
2498 		if (!fpu)
2499 			goto out;
2500 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2501 		if (r)
2502 			goto out;
2503 		r = -EFAULT;
2504 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2505 			goto out;
2506 		r = 0;
2507 		break;
2508 	}
2509 	case KVM_SET_FPU: {
2510 		fpu = memdup_user(argp, sizeof(*fpu));
2511 		if (IS_ERR(fpu)) {
2512 			r = PTR_ERR(fpu);
2513 			fpu = NULL;
2514 			goto out;
2515 		}
2516 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2517 		break;
2518 	}
2519 	default:
2520 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2521 	}
2522 out:
2523 	vcpu_put(vcpu);
2524 	kfree(fpu);
2525 	kfree(kvm_sregs);
2526 	return r;
2527 }
2528 
2529 #ifdef CONFIG_KVM_COMPAT
2530 static long kvm_vcpu_compat_ioctl(struct file *filp,
2531 				  unsigned int ioctl, unsigned long arg)
2532 {
2533 	struct kvm_vcpu *vcpu = filp->private_data;
2534 	void __user *argp = compat_ptr(arg);
2535 	int r;
2536 
2537 	if (vcpu->kvm->mm != current->mm)
2538 		return -EIO;
2539 
2540 	switch (ioctl) {
2541 	case KVM_SET_SIGNAL_MASK: {
2542 		struct kvm_signal_mask __user *sigmask_arg = argp;
2543 		struct kvm_signal_mask kvm_sigmask;
2544 		compat_sigset_t csigset;
2545 		sigset_t sigset;
2546 
2547 		if (argp) {
2548 			r = -EFAULT;
2549 			if (copy_from_user(&kvm_sigmask, argp,
2550 					   sizeof(kvm_sigmask)))
2551 				goto out;
2552 			r = -EINVAL;
2553 			if (kvm_sigmask.len != sizeof(csigset))
2554 				goto out;
2555 			r = -EFAULT;
2556 			if (copy_from_user(&csigset, sigmask_arg->sigset,
2557 					   sizeof(csigset)))
2558 				goto out;
2559 			sigset_from_compat(&sigset, &csigset);
2560 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2561 		} else
2562 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2563 		break;
2564 	}
2565 	default:
2566 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
2567 	}
2568 
2569 out:
2570 	return r;
2571 }
2572 #endif
2573 
2574 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2575 				 int (*accessor)(struct kvm_device *dev,
2576 						 struct kvm_device_attr *attr),
2577 				 unsigned long arg)
2578 {
2579 	struct kvm_device_attr attr;
2580 
2581 	if (!accessor)
2582 		return -EPERM;
2583 
2584 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2585 		return -EFAULT;
2586 
2587 	return accessor(dev, &attr);
2588 }
2589 
2590 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2591 			     unsigned long arg)
2592 {
2593 	struct kvm_device *dev = filp->private_data;
2594 
2595 	switch (ioctl) {
2596 	case KVM_SET_DEVICE_ATTR:
2597 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2598 	case KVM_GET_DEVICE_ATTR:
2599 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2600 	case KVM_HAS_DEVICE_ATTR:
2601 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2602 	default:
2603 		if (dev->ops->ioctl)
2604 			return dev->ops->ioctl(dev, ioctl, arg);
2605 
2606 		return -ENOTTY;
2607 	}
2608 }
2609 
2610 static int kvm_device_release(struct inode *inode, struct file *filp)
2611 {
2612 	struct kvm_device *dev = filp->private_data;
2613 	struct kvm *kvm = dev->kvm;
2614 
2615 	kvm_put_kvm(kvm);
2616 	return 0;
2617 }
2618 
2619 static const struct file_operations kvm_device_fops = {
2620 	.unlocked_ioctl = kvm_device_ioctl,
2621 #ifdef CONFIG_KVM_COMPAT
2622 	.compat_ioctl = kvm_device_ioctl,
2623 #endif
2624 	.release = kvm_device_release,
2625 };
2626 
2627 struct kvm_device *kvm_device_from_filp(struct file *filp)
2628 {
2629 	if (filp->f_op != &kvm_device_fops)
2630 		return NULL;
2631 
2632 	return filp->private_data;
2633 }
2634 
2635 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2636 #ifdef CONFIG_KVM_MPIC
2637 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
2638 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
2639 #endif
2640 
2641 #ifdef CONFIG_KVM_XICS
2642 	[KVM_DEV_TYPE_XICS]		= &kvm_xics_ops,
2643 #endif
2644 };
2645 
2646 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2647 {
2648 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
2649 		return -ENOSPC;
2650 
2651 	if (kvm_device_ops_table[type] != NULL)
2652 		return -EEXIST;
2653 
2654 	kvm_device_ops_table[type] = ops;
2655 	return 0;
2656 }
2657 
2658 void kvm_unregister_device_ops(u32 type)
2659 {
2660 	if (kvm_device_ops_table[type] != NULL)
2661 		kvm_device_ops_table[type] = NULL;
2662 }
2663 
2664 static int kvm_ioctl_create_device(struct kvm *kvm,
2665 				   struct kvm_create_device *cd)
2666 {
2667 	struct kvm_device_ops *ops = NULL;
2668 	struct kvm_device *dev;
2669 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2670 	int ret;
2671 
2672 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2673 		return -ENODEV;
2674 
2675 	ops = kvm_device_ops_table[cd->type];
2676 	if (ops == NULL)
2677 		return -ENODEV;
2678 
2679 	if (test)
2680 		return 0;
2681 
2682 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2683 	if (!dev)
2684 		return -ENOMEM;
2685 
2686 	dev->ops = ops;
2687 	dev->kvm = kvm;
2688 
2689 	ret = ops->create(dev, cd->type);
2690 	if (ret < 0) {
2691 		kfree(dev);
2692 		return ret;
2693 	}
2694 
2695 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2696 	if (ret < 0) {
2697 		ops->destroy(dev);
2698 		return ret;
2699 	}
2700 
2701 	list_add(&dev->vm_node, &kvm->devices);
2702 	kvm_get_kvm(kvm);
2703 	cd->fd = ret;
2704 	return 0;
2705 }
2706 
2707 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2708 {
2709 	switch (arg) {
2710 	case KVM_CAP_USER_MEMORY:
2711 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2712 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2713 	case KVM_CAP_INTERNAL_ERROR_DATA:
2714 #ifdef CONFIG_HAVE_KVM_MSI
2715 	case KVM_CAP_SIGNAL_MSI:
2716 #endif
2717 #ifdef CONFIG_HAVE_KVM_IRQFD
2718 	case KVM_CAP_IRQFD:
2719 	case KVM_CAP_IRQFD_RESAMPLE:
2720 #endif
2721 	case KVM_CAP_CHECK_EXTENSION_VM:
2722 		return 1;
2723 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2724 	case KVM_CAP_IRQ_ROUTING:
2725 		return KVM_MAX_IRQ_ROUTES;
2726 #endif
2727 #if KVM_ADDRESS_SPACE_NUM > 1
2728 	case KVM_CAP_MULTI_ADDRESS_SPACE:
2729 		return KVM_ADDRESS_SPACE_NUM;
2730 #endif
2731 	default:
2732 		break;
2733 	}
2734 	return kvm_vm_ioctl_check_extension(kvm, arg);
2735 }
2736 
2737 static long kvm_vm_ioctl(struct file *filp,
2738 			   unsigned int ioctl, unsigned long arg)
2739 {
2740 	struct kvm *kvm = filp->private_data;
2741 	void __user *argp = (void __user *)arg;
2742 	int r;
2743 
2744 	if (kvm->mm != current->mm)
2745 		return -EIO;
2746 	switch (ioctl) {
2747 	case KVM_CREATE_VCPU:
2748 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2749 		break;
2750 	case KVM_SET_USER_MEMORY_REGION: {
2751 		struct kvm_userspace_memory_region kvm_userspace_mem;
2752 
2753 		r = -EFAULT;
2754 		if (copy_from_user(&kvm_userspace_mem, argp,
2755 						sizeof(kvm_userspace_mem)))
2756 			goto out;
2757 
2758 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2759 		break;
2760 	}
2761 	case KVM_GET_DIRTY_LOG: {
2762 		struct kvm_dirty_log log;
2763 
2764 		r = -EFAULT;
2765 		if (copy_from_user(&log, argp, sizeof(log)))
2766 			goto out;
2767 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2768 		break;
2769 	}
2770 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2771 	case KVM_REGISTER_COALESCED_MMIO: {
2772 		struct kvm_coalesced_mmio_zone zone;
2773 
2774 		r = -EFAULT;
2775 		if (copy_from_user(&zone, argp, sizeof(zone)))
2776 			goto out;
2777 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2778 		break;
2779 	}
2780 	case KVM_UNREGISTER_COALESCED_MMIO: {
2781 		struct kvm_coalesced_mmio_zone zone;
2782 
2783 		r = -EFAULT;
2784 		if (copy_from_user(&zone, argp, sizeof(zone)))
2785 			goto out;
2786 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2787 		break;
2788 	}
2789 #endif
2790 	case KVM_IRQFD: {
2791 		struct kvm_irqfd data;
2792 
2793 		r = -EFAULT;
2794 		if (copy_from_user(&data, argp, sizeof(data)))
2795 			goto out;
2796 		r = kvm_irqfd(kvm, &data);
2797 		break;
2798 	}
2799 	case KVM_IOEVENTFD: {
2800 		struct kvm_ioeventfd data;
2801 
2802 		r = -EFAULT;
2803 		if (copy_from_user(&data, argp, sizeof(data)))
2804 			goto out;
2805 		r = kvm_ioeventfd(kvm, &data);
2806 		break;
2807 	}
2808 #ifdef CONFIG_HAVE_KVM_MSI
2809 	case KVM_SIGNAL_MSI: {
2810 		struct kvm_msi msi;
2811 
2812 		r = -EFAULT;
2813 		if (copy_from_user(&msi, argp, sizeof(msi)))
2814 			goto out;
2815 		r = kvm_send_userspace_msi(kvm, &msi);
2816 		break;
2817 	}
2818 #endif
2819 #ifdef __KVM_HAVE_IRQ_LINE
2820 	case KVM_IRQ_LINE_STATUS:
2821 	case KVM_IRQ_LINE: {
2822 		struct kvm_irq_level irq_event;
2823 
2824 		r = -EFAULT;
2825 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2826 			goto out;
2827 
2828 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2829 					ioctl == KVM_IRQ_LINE_STATUS);
2830 		if (r)
2831 			goto out;
2832 
2833 		r = -EFAULT;
2834 		if (ioctl == KVM_IRQ_LINE_STATUS) {
2835 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2836 				goto out;
2837 		}
2838 
2839 		r = 0;
2840 		break;
2841 	}
2842 #endif
2843 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2844 	case KVM_SET_GSI_ROUTING: {
2845 		struct kvm_irq_routing routing;
2846 		struct kvm_irq_routing __user *urouting;
2847 		struct kvm_irq_routing_entry *entries;
2848 
2849 		r = -EFAULT;
2850 		if (copy_from_user(&routing, argp, sizeof(routing)))
2851 			goto out;
2852 		r = -EINVAL;
2853 		if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2854 			goto out;
2855 		if (routing.flags)
2856 			goto out;
2857 		r = -ENOMEM;
2858 		entries = vmalloc(routing.nr * sizeof(*entries));
2859 		if (!entries)
2860 			goto out;
2861 		r = -EFAULT;
2862 		urouting = argp;
2863 		if (copy_from_user(entries, urouting->entries,
2864 				   routing.nr * sizeof(*entries)))
2865 			goto out_free_irq_routing;
2866 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
2867 					routing.flags);
2868 out_free_irq_routing:
2869 		vfree(entries);
2870 		break;
2871 	}
2872 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2873 	case KVM_CREATE_DEVICE: {
2874 		struct kvm_create_device cd;
2875 
2876 		r = -EFAULT;
2877 		if (copy_from_user(&cd, argp, sizeof(cd)))
2878 			goto out;
2879 
2880 		r = kvm_ioctl_create_device(kvm, &cd);
2881 		if (r)
2882 			goto out;
2883 
2884 		r = -EFAULT;
2885 		if (copy_to_user(argp, &cd, sizeof(cd)))
2886 			goto out;
2887 
2888 		r = 0;
2889 		break;
2890 	}
2891 	case KVM_CHECK_EXTENSION:
2892 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2893 		break;
2894 	default:
2895 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2896 	}
2897 out:
2898 	return r;
2899 }
2900 
2901 #ifdef CONFIG_KVM_COMPAT
2902 struct compat_kvm_dirty_log {
2903 	__u32 slot;
2904 	__u32 padding1;
2905 	union {
2906 		compat_uptr_t dirty_bitmap; /* one bit per page */
2907 		__u64 padding2;
2908 	};
2909 };
2910 
2911 static long kvm_vm_compat_ioctl(struct file *filp,
2912 			   unsigned int ioctl, unsigned long arg)
2913 {
2914 	struct kvm *kvm = filp->private_data;
2915 	int r;
2916 
2917 	if (kvm->mm != current->mm)
2918 		return -EIO;
2919 	switch (ioctl) {
2920 	case KVM_GET_DIRTY_LOG: {
2921 		struct compat_kvm_dirty_log compat_log;
2922 		struct kvm_dirty_log log;
2923 
2924 		r = -EFAULT;
2925 		if (copy_from_user(&compat_log, (void __user *)arg,
2926 				   sizeof(compat_log)))
2927 			goto out;
2928 		log.slot	 = compat_log.slot;
2929 		log.padding1	 = compat_log.padding1;
2930 		log.padding2	 = compat_log.padding2;
2931 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2932 
2933 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2934 		break;
2935 	}
2936 	default:
2937 		r = kvm_vm_ioctl(filp, ioctl, arg);
2938 	}
2939 
2940 out:
2941 	return r;
2942 }
2943 #endif
2944 
2945 static struct file_operations kvm_vm_fops = {
2946 	.release        = kvm_vm_release,
2947 	.unlocked_ioctl = kvm_vm_ioctl,
2948 #ifdef CONFIG_KVM_COMPAT
2949 	.compat_ioctl   = kvm_vm_compat_ioctl,
2950 #endif
2951 	.llseek		= noop_llseek,
2952 };
2953 
2954 static int kvm_dev_ioctl_create_vm(unsigned long type)
2955 {
2956 	int r;
2957 	struct kvm *kvm;
2958 
2959 	kvm = kvm_create_vm(type);
2960 	if (IS_ERR(kvm))
2961 		return PTR_ERR(kvm);
2962 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2963 	r = kvm_coalesced_mmio_init(kvm);
2964 	if (r < 0) {
2965 		kvm_put_kvm(kvm);
2966 		return r;
2967 	}
2968 #endif
2969 	r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2970 	if (r < 0)
2971 		kvm_put_kvm(kvm);
2972 
2973 	return r;
2974 }
2975 
2976 static long kvm_dev_ioctl(struct file *filp,
2977 			  unsigned int ioctl, unsigned long arg)
2978 {
2979 	long r = -EINVAL;
2980 
2981 	switch (ioctl) {
2982 	case KVM_GET_API_VERSION:
2983 		if (arg)
2984 			goto out;
2985 		r = KVM_API_VERSION;
2986 		break;
2987 	case KVM_CREATE_VM:
2988 		r = kvm_dev_ioctl_create_vm(arg);
2989 		break;
2990 	case KVM_CHECK_EXTENSION:
2991 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2992 		break;
2993 	case KVM_GET_VCPU_MMAP_SIZE:
2994 		if (arg)
2995 			goto out;
2996 		r = PAGE_SIZE;     /* struct kvm_run */
2997 #ifdef CONFIG_X86
2998 		r += PAGE_SIZE;    /* pio data page */
2999 #endif
3000 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3001 		r += PAGE_SIZE;    /* coalesced mmio ring page */
3002 #endif
3003 		break;
3004 	case KVM_TRACE_ENABLE:
3005 	case KVM_TRACE_PAUSE:
3006 	case KVM_TRACE_DISABLE:
3007 		r = -EOPNOTSUPP;
3008 		break;
3009 	default:
3010 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
3011 	}
3012 out:
3013 	return r;
3014 }
3015 
3016 static struct file_operations kvm_chardev_ops = {
3017 	.unlocked_ioctl = kvm_dev_ioctl,
3018 	.compat_ioctl   = kvm_dev_ioctl,
3019 	.llseek		= noop_llseek,
3020 };
3021 
3022 static struct miscdevice kvm_dev = {
3023 	KVM_MINOR,
3024 	"kvm",
3025 	&kvm_chardev_ops,
3026 };
3027 
3028 static void hardware_enable_nolock(void *junk)
3029 {
3030 	int cpu = raw_smp_processor_id();
3031 	int r;
3032 
3033 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3034 		return;
3035 
3036 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
3037 
3038 	r = kvm_arch_hardware_enable();
3039 
3040 	if (r) {
3041 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3042 		atomic_inc(&hardware_enable_failed);
3043 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3044 	}
3045 }
3046 
3047 static void hardware_enable(void)
3048 {
3049 	raw_spin_lock(&kvm_count_lock);
3050 	if (kvm_usage_count)
3051 		hardware_enable_nolock(NULL);
3052 	raw_spin_unlock(&kvm_count_lock);
3053 }
3054 
3055 static void hardware_disable_nolock(void *junk)
3056 {
3057 	int cpu = raw_smp_processor_id();
3058 
3059 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3060 		return;
3061 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3062 	kvm_arch_hardware_disable();
3063 }
3064 
3065 static void hardware_disable(void)
3066 {
3067 	raw_spin_lock(&kvm_count_lock);
3068 	if (kvm_usage_count)
3069 		hardware_disable_nolock(NULL);
3070 	raw_spin_unlock(&kvm_count_lock);
3071 }
3072 
3073 static void hardware_disable_all_nolock(void)
3074 {
3075 	BUG_ON(!kvm_usage_count);
3076 
3077 	kvm_usage_count--;
3078 	if (!kvm_usage_count)
3079 		on_each_cpu(hardware_disable_nolock, NULL, 1);
3080 }
3081 
3082 static void hardware_disable_all(void)
3083 {
3084 	raw_spin_lock(&kvm_count_lock);
3085 	hardware_disable_all_nolock();
3086 	raw_spin_unlock(&kvm_count_lock);
3087 }
3088 
3089 static int hardware_enable_all(void)
3090 {
3091 	int r = 0;
3092 
3093 	raw_spin_lock(&kvm_count_lock);
3094 
3095 	kvm_usage_count++;
3096 	if (kvm_usage_count == 1) {
3097 		atomic_set(&hardware_enable_failed, 0);
3098 		on_each_cpu(hardware_enable_nolock, NULL, 1);
3099 
3100 		if (atomic_read(&hardware_enable_failed)) {
3101 			hardware_disable_all_nolock();
3102 			r = -EBUSY;
3103 		}
3104 	}
3105 
3106 	raw_spin_unlock(&kvm_count_lock);
3107 
3108 	return r;
3109 }
3110 
3111 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3112 			   void *v)
3113 {
3114 	val &= ~CPU_TASKS_FROZEN;
3115 	switch (val) {
3116 	case CPU_DYING:
3117 		hardware_disable();
3118 		break;
3119 	case CPU_STARTING:
3120 		hardware_enable();
3121 		break;
3122 	}
3123 	return NOTIFY_OK;
3124 }
3125 
3126 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3127 		      void *v)
3128 {
3129 	/*
3130 	 * Some (well, at least mine) BIOSes hang on reboot if
3131 	 * in vmx root mode.
3132 	 *
3133 	 * And Intel TXT required VMX off for all cpu when system shutdown.
3134 	 */
3135 	pr_info("kvm: exiting hardware virtualization\n");
3136 	kvm_rebooting = true;
3137 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3138 	return NOTIFY_OK;
3139 }
3140 
3141 static struct notifier_block kvm_reboot_notifier = {
3142 	.notifier_call = kvm_reboot,
3143 	.priority = 0,
3144 };
3145 
3146 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3147 {
3148 	int i;
3149 
3150 	for (i = 0; i < bus->dev_count; i++) {
3151 		struct kvm_io_device *pos = bus->range[i].dev;
3152 
3153 		kvm_iodevice_destructor(pos);
3154 	}
3155 	kfree(bus);
3156 }
3157 
3158 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3159 				 const struct kvm_io_range *r2)
3160 {
3161 	gpa_t addr1 = r1->addr;
3162 	gpa_t addr2 = r2->addr;
3163 
3164 	if (addr1 < addr2)
3165 		return -1;
3166 
3167 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
3168 	 * accept any overlapping write.  Any order is acceptable for
3169 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3170 	 * we process all of them.
3171 	 */
3172 	if (r2->len) {
3173 		addr1 += r1->len;
3174 		addr2 += r2->len;
3175 	}
3176 
3177 	if (addr1 > addr2)
3178 		return 1;
3179 
3180 	return 0;
3181 }
3182 
3183 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3184 {
3185 	return kvm_io_bus_cmp(p1, p2);
3186 }
3187 
3188 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3189 			  gpa_t addr, int len)
3190 {
3191 	bus->range[bus->dev_count++] = (struct kvm_io_range) {
3192 		.addr = addr,
3193 		.len = len,
3194 		.dev = dev,
3195 	};
3196 
3197 	sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3198 		kvm_io_bus_sort_cmp, NULL);
3199 
3200 	return 0;
3201 }
3202 
3203 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3204 			     gpa_t addr, int len)
3205 {
3206 	struct kvm_io_range *range, key;
3207 	int off;
3208 
3209 	key = (struct kvm_io_range) {
3210 		.addr = addr,
3211 		.len = len,
3212 	};
3213 
3214 	range = bsearch(&key, bus->range, bus->dev_count,
3215 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3216 	if (range == NULL)
3217 		return -ENOENT;
3218 
3219 	off = range - bus->range;
3220 
3221 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3222 		off--;
3223 
3224 	return off;
3225 }
3226 
3227 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3228 			      struct kvm_io_range *range, const void *val)
3229 {
3230 	int idx;
3231 
3232 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3233 	if (idx < 0)
3234 		return -EOPNOTSUPP;
3235 
3236 	while (idx < bus->dev_count &&
3237 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3238 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3239 					range->len, val))
3240 			return idx;
3241 		idx++;
3242 	}
3243 
3244 	return -EOPNOTSUPP;
3245 }
3246 
3247 /* kvm_io_bus_write - called under kvm->slots_lock */
3248 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3249 		     int len, const void *val)
3250 {
3251 	struct kvm_io_bus *bus;
3252 	struct kvm_io_range range;
3253 	int r;
3254 
3255 	range = (struct kvm_io_range) {
3256 		.addr = addr,
3257 		.len = len,
3258 	};
3259 
3260 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3261 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
3262 	return r < 0 ? r : 0;
3263 }
3264 
3265 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3266 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3267 			    gpa_t addr, int len, const void *val, long cookie)
3268 {
3269 	struct kvm_io_bus *bus;
3270 	struct kvm_io_range range;
3271 
3272 	range = (struct kvm_io_range) {
3273 		.addr = addr,
3274 		.len = len,
3275 	};
3276 
3277 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3278 
3279 	/* First try the device referenced by cookie. */
3280 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
3281 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3282 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3283 					val))
3284 			return cookie;
3285 
3286 	/*
3287 	 * cookie contained garbage; fall back to search and return the
3288 	 * correct cookie value.
3289 	 */
3290 	return __kvm_io_bus_write(vcpu, bus, &range, val);
3291 }
3292 
3293 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3294 			     struct kvm_io_range *range, void *val)
3295 {
3296 	int idx;
3297 
3298 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3299 	if (idx < 0)
3300 		return -EOPNOTSUPP;
3301 
3302 	while (idx < bus->dev_count &&
3303 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3304 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3305 				       range->len, val))
3306 			return idx;
3307 		idx++;
3308 	}
3309 
3310 	return -EOPNOTSUPP;
3311 }
3312 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3313 
3314 /* kvm_io_bus_read - called under kvm->slots_lock */
3315 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3316 		    int len, void *val)
3317 {
3318 	struct kvm_io_bus *bus;
3319 	struct kvm_io_range range;
3320 	int r;
3321 
3322 	range = (struct kvm_io_range) {
3323 		.addr = addr,
3324 		.len = len,
3325 	};
3326 
3327 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3328 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
3329 	return r < 0 ? r : 0;
3330 }
3331 
3332 
3333 /* Caller must hold slots_lock. */
3334 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3335 			    int len, struct kvm_io_device *dev)
3336 {
3337 	struct kvm_io_bus *new_bus, *bus;
3338 
3339 	bus = kvm->buses[bus_idx];
3340 	/* exclude ioeventfd which is limited by maximum fd */
3341 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3342 		return -ENOSPC;
3343 
3344 	new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3345 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3346 	if (!new_bus)
3347 		return -ENOMEM;
3348 	memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3349 	       sizeof(struct kvm_io_range)));
3350 	kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3351 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3352 	synchronize_srcu_expedited(&kvm->srcu);
3353 	kfree(bus);
3354 
3355 	return 0;
3356 }
3357 
3358 /* Caller must hold slots_lock. */
3359 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3360 			      struct kvm_io_device *dev)
3361 {
3362 	int i, r;
3363 	struct kvm_io_bus *new_bus, *bus;
3364 
3365 	bus = kvm->buses[bus_idx];
3366 	r = -ENOENT;
3367 	for (i = 0; i < bus->dev_count; i++)
3368 		if (bus->range[i].dev == dev) {
3369 			r = 0;
3370 			break;
3371 		}
3372 
3373 	if (r)
3374 		return r;
3375 
3376 	new_bus = kzalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3377 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3378 	if (!new_bus)
3379 		return -ENOMEM;
3380 
3381 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3382 	new_bus->dev_count--;
3383 	memcpy(new_bus->range + i, bus->range + i + 1,
3384 	       (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3385 
3386 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3387 	synchronize_srcu_expedited(&kvm->srcu);
3388 	kfree(bus);
3389 	return r;
3390 }
3391 
3392 static struct notifier_block kvm_cpu_notifier = {
3393 	.notifier_call = kvm_cpu_hotplug,
3394 };
3395 
3396 static int vm_stat_get(void *_offset, u64 *val)
3397 {
3398 	unsigned offset = (long)_offset;
3399 	struct kvm *kvm;
3400 
3401 	*val = 0;
3402 	spin_lock(&kvm_lock);
3403 	list_for_each_entry(kvm, &vm_list, vm_list)
3404 		*val += *(u32 *)((void *)kvm + offset);
3405 	spin_unlock(&kvm_lock);
3406 	return 0;
3407 }
3408 
3409 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3410 
3411 static int vcpu_stat_get(void *_offset, u64 *val)
3412 {
3413 	unsigned offset = (long)_offset;
3414 	struct kvm *kvm;
3415 	struct kvm_vcpu *vcpu;
3416 	int i;
3417 
3418 	*val = 0;
3419 	spin_lock(&kvm_lock);
3420 	list_for_each_entry(kvm, &vm_list, vm_list)
3421 		kvm_for_each_vcpu(i, vcpu, kvm)
3422 			*val += *(u32 *)((void *)vcpu + offset);
3423 
3424 	spin_unlock(&kvm_lock);
3425 	return 0;
3426 }
3427 
3428 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3429 
3430 static const struct file_operations *stat_fops[] = {
3431 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
3432 	[KVM_STAT_VM]   = &vm_stat_fops,
3433 };
3434 
3435 static int kvm_init_debug(void)
3436 {
3437 	int r = -EEXIST;
3438 	struct kvm_stats_debugfs_item *p;
3439 
3440 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3441 	if (kvm_debugfs_dir == NULL)
3442 		goto out;
3443 
3444 	for (p = debugfs_entries; p->name; ++p) {
3445 		p->dentry = debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3446 						(void *)(long)p->offset,
3447 						stat_fops[p->kind]);
3448 		if (p->dentry == NULL)
3449 			goto out_dir;
3450 	}
3451 
3452 	return 0;
3453 
3454 out_dir:
3455 	debugfs_remove_recursive(kvm_debugfs_dir);
3456 out:
3457 	return r;
3458 }
3459 
3460 static void kvm_exit_debug(void)
3461 {
3462 	struct kvm_stats_debugfs_item *p;
3463 
3464 	for (p = debugfs_entries; p->name; ++p)
3465 		debugfs_remove(p->dentry);
3466 	debugfs_remove(kvm_debugfs_dir);
3467 }
3468 
3469 static int kvm_suspend(void)
3470 {
3471 	if (kvm_usage_count)
3472 		hardware_disable_nolock(NULL);
3473 	return 0;
3474 }
3475 
3476 static void kvm_resume(void)
3477 {
3478 	if (kvm_usage_count) {
3479 		WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3480 		hardware_enable_nolock(NULL);
3481 	}
3482 }
3483 
3484 static struct syscore_ops kvm_syscore_ops = {
3485 	.suspend = kvm_suspend,
3486 	.resume = kvm_resume,
3487 };
3488 
3489 static inline
3490 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3491 {
3492 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
3493 }
3494 
3495 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3496 {
3497 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3498 
3499 	if (vcpu->preempted)
3500 		vcpu->preempted = false;
3501 
3502 	kvm_arch_sched_in(vcpu, cpu);
3503 
3504 	kvm_arch_vcpu_load(vcpu, cpu);
3505 }
3506 
3507 static void kvm_sched_out(struct preempt_notifier *pn,
3508 			  struct task_struct *next)
3509 {
3510 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3511 
3512 	if (current->state == TASK_RUNNING)
3513 		vcpu->preempted = true;
3514 	kvm_arch_vcpu_put(vcpu);
3515 }
3516 
3517 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3518 		  struct module *module)
3519 {
3520 	int r;
3521 	int cpu;
3522 
3523 	r = kvm_arch_init(opaque);
3524 	if (r)
3525 		goto out_fail;
3526 
3527 	/*
3528 	 * kvm_arch_init makes sure there's at most one caller
3529 	 * for architectures that support multiple implementations,
3530 	 * like intel and amd on x86.
3531 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3532 	 * conflicts in case kvm is already setup for another implementation.
3533 	 */
3534 	r = kvm_irqfd_init();
3535 	if (r)
3536 		goto out_irqfd;
3537 
3538 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3539 		r = -ENOMEM;
3540 		goto out_free_0;
3541 	}
3542 
3543 	r = kvm_arch_hardware_setup();
3544 	if (r < 0)
3545 		goto out_free_0a;
3546 
3547 	for_each_online_cpu(cpu) {
3548 		smp_call_function_single(cpu,
3549 				kvm_arch_check_processor_compat,
3550 				&r, 1);
3551 		if (r < 0)
3552 			goto out_free_1;
3553 	}
3554 
3555 	r = register_cpu_notifier(&kvm_cpu_notifier);
3556 	if (r)
3557 		goto out_free_2;
3558 	register_reboot_notifier(&kvm_reboot_notifier);
3559 
3560 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
3561 	if (!vcpu_align)
3562 		vcpu_align = __alignof__(struct kvm_vcpu);
3563 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3564 					   0, NULL);
3565 	if (!kvm_vcpu_cache) {
3566 		r = -ENOMEM;
3567 		goto out_free_3;
3568 	}
3569 
3570 	r = kvm_async_pf_init();
3571 	if (r)
3572 		goto out_free;
3573 
3574 	kvm_chardev_ops.owner = module;
3575 	kvm_vm_fops.owner = module;
3576 	kvm_vcpu_fops.owner = module;
3577 
3578 	r = misc_register(&kvm_dev);
3579 	if (r) {
3580 		pr_err("kvm: misc device register failed\n");
3581 		goto out_unreg;
3582 	}
3583 
3584 	register_syscore_ops(&kvm_syscore_ops);
3585 
3586 	kvm_preempt_ops.sched_in = kvm_sched_in;
3587 	kvm_preempt_ops.sched_out = kvm_sched_out;
3588 
3589 	r = kvm_init_debug();
3590 	if (r) {
3591 		pr_err("kvm: create debugfs files failed\n");
3592 		goto out_undebugfs;
3593 	}
3594 
3595 	r = kvm_vfio_ops_init();
3596 	WARN_ON(r);
3597 
3598 	return 0;
3599 
3600 out_undebugfs:
3601 	unregister_syscore_ops(&kvm_syscore_ops);
3602 	misc_deregister(&kvm_dev);
3603 out_unreg:
3604 	kvm_async_pf_deinit();
3605 out_free:
3606 	kmem_cache_destroy(kvm_vcpu_cache);
3607 out_free_3:
3608 	unregister_reboot_notifier(&kvm_reboot_notifier);
3609 	unregister_cpu_notifier(&kvm_cpu_notifier);
3610 out_free_2:
3611 out_free_1:
3612 	kvm_arch_hardware_unsetup();
3613 out_free_0a:
3614 	free_cpumask_var(cpus_hardware_enabled);
3615 out_free_0:
3616 	kvm_irqfd_exit();
3617 out_irqfd:
3618 	kvm_arch_exit();
3619 out_fail:
3620 	return r;
3621 }
3622 EXPORT_SYMBOL_GPL(kvm_init);
3623 
3624 void kvm_exit(void)
3625 {
3626 	kvm_exit_debug();
3627 	misc_deregister(&kvm_dev);
3628 	kmem_cache_destroy(kvm_vcpu_cache);
3629 	kvm_async_pf_deinit();
3630 	unregister_syscore_ops(&kvm_syscore_ops);
3631 	unregister_reboot_notifier(&kvm_reboot_notifier);
3632 	unregister_cpu_notifier(&kvm_cpu_notifier);
3633 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3634 	kvm_arch_hardware_unsetup();
3635 	kvm_arch_exit();
3636 	kvm_irqfd_exit();
3637 	free_cpumask_var(cpus_hardware_enabled);
3638 	kvm_vfio_ops_exit();
3639 }
3640 EXPORT_SYMBOL_GPL(kvm_exit);
3641