xref: /openbmc/linux/virt/kvm/kvm_main.c (revision a06c488d)
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(kvm_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(kvm_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 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
210 {
211 	struct page *page;
212 	int r;
213 
214 	mutex_init(&vcpu->mutex);
215 	vcpu->cpu = -1;
216 	vcpu->kvm = kvm;
217 	vcpu->vcpu_id = id;
218 	vcpu->pid = NULL;
219 	vcpu->halt_poll_ns = 0;
220 	init_waitqueue_head(&vcpu->wq);
221 	kvm_async_pf_vcpu_init(vcpu);
222 
223 	vcpu->pre_pcpu = -1;
224 	INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
225 
226 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
227 	if (!page) {
228 		r = -ENOMEM;
229 		goto fail;
230 	}
231 	vcpu->run = page_address(page);
232 
233 	kvm_vcpu_set_in_spin_loop(vcpu, false);
234 	kvm_vcpu_set_dy_eligible(vcpu, false);
235 	vcpu->preempted = false;
236 
237 	r = kvm_arch_vcpu_init(vcpu);
238 	if (r < 0)
239 		goto fail_free_run;
240 	return 0;
241 
242 fail_free_run:
243 	free_page((unsigned long)vcpu->run);
244 fail:
245 	return r;
246 }
247 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
248 
249 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
250 {
251 	put_pid(vcpu->pid);
252 	kvm_arch_vcpu_uninit(vcpu);
253 	free_page((unsigned long)vcpu->run);
254 }
255 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
256 
257 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
258 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
259 {
260 	return container_of(mn, struct kvm, mmu_notifier);
261 }
262 
263 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
264 					     struct mm_struct *mm,
265 					     unsigned long address)
266 {
267 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
268 	int need_tlb_flush, idx;
269 
270 	/*
271 	 * When ->invalidate_page runs, the linux pte has been zapped
272 	 * already but the page is still allocated until
273 	 * ->invalidate_page returns. So if we increase the sequence
274 	 * here the kvm page fault will notice if the spte can't be
275 	 * established because the page is going to be freed. If
276 	 * instead the kvm page fault establishes the spte before
277 	 * ->invalidate_page runs, kvm_unmap_hva will release it
278 	 * before returning.
279 	 *
280 	 * The sequence increase only need to be seen at spin_unlock
281 	 * time, and not at spin_lock time.
282 	 *
283 	 * Increasing the sequence after the spin_unlock would be
284 	 * unsafe because the kvm page fault could then establish the
285 	 * pte after kvm_unmap_hva returned, without noticing the page
286 	 * is going to be freed.
287 	 */
288 	idx = srcu_read_lock(&kvm->srcu);
289 	spin_lock(&kvm->mmu_lock);
290 
291 	kvm->mmu_notifier_seq++;
292 	need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
293 	/* we've to flush the tlb before the pages can be freed */
294 	if (need_tlb_flush)
295 		kvm_flush_remote_tlbs(kvm);
296 
297 	spin_unlock(&kvm->mmu_lock);
298 
299 	kvm_arch_mmu_notifier_invalidate_page(kvm, address);
300 
301 	srcu_read_unlock(&kvm->srcu, idx);
302 }
303 
304 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
305 					struct mm_struct *mm,
306 					unsigned long address,
307 					pte_t pte)
308 {
309 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
310 	int idx;
311 
312 	idx = srcu_read_lock(&kvm->srcu);
313 	spin_lock(&kvm->mmu_lock);
314 	kvm->mmu_notifier_seq++;
315 	kvm_set_spte_hva(kvm, address, pte);
316 	spin_unlock(&kvm->mmu_lock);
317 	srcu_read_unlock(&kvm->srcu, idx);
318 }
319 
320 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
321 						    struct mm_struct *mm,
322 						    unsigned long start,
323 						    unsigned long end)
324 {
325 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
326 	int need_tlb_flush = 0, idx;
327 
328 	idx = srcu_read_lock(&kvm->srcu);
329 	spin_lock(&kvm->mmu_lock);
330 	/*
331 	 * The count increase must become visible at unlock time as no
332 	 * spte can be established without taking the mmu_lock and
333 	 * count is also read inside the mmu_lock critical section.
334 	 */
335 	kvm->mmu_notifier_count++;
336 	need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
337 	need_tlb_flush |= kvm->tlbs_dirty;
338 	/* we've to flush the tlb before the pages can be freed */
339 	if (need_tlb_flush)
340 		kvm_flush_remote_tlbs(kvm);
341 
342 	spin_unlock(&kvm->mmu_lock);
343 	srcu_read_unlock(&kvm->srcu, idx);
344 }
345 
346 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
347 						  struct mm_struct *mm,
348 						  unsigned long start,
349 						  unsigned long end)
350 {
351 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
352 
353 	spin_lock(&kvm->mmu_lock);
354 	/*
355 	 * This sequence increase will notify the kvm page fault that
356 	 * the page that is going to be mapped in the spte could have
357 	 * been freed.
358 	 */
359 	kvm->mmu_notifier_seq++;
360 	smp_wmb();
361 	/*
362 	 * The above sequence increase must be visible before the
363 	 * below count decrease, which is ensured by the smp_wmb above
364 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
365 	 */
366 	kvm->mmu_notifier_count--;
367 	spin_unlock(&kvm->mmu_lock);
368 
369 	BUG_ON(kvm->mmu_notifier_count < 0);
370 }
371 
372 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
373 					      struct mm_struct *mm,
374 					      unsigned long start,
375 					      unsigned long end)
376 {
377 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
378 	int young, idx;
379 
380 	idx = srcu_read_lock(&kvm->srcu);
381 	spin_lock(&kvm->mmu_lock);
382 
383 	young = kvm_age_hva(kvm, start, end);
384 	if (young)
385 		kvm_flush_remote_tlbs(kvm);
386 
387 	spin_unlock(&kvm->mmu_lock);
388 	srcu_read_unlock(&kvm->srcu, idx);
389 
390 	return young;
391 }
392 
393 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
394 					struct mm_struct *mm,
395 					unsigned long start,
396 					unsigned long end)
397 {
398 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
399 	int young, idx;
400 
401 	idx = srcu_read_lock(&kvm->srcu);
402 	spin_lock(&kvm->mmu_lock);
403 	/*
404 	 * Even though we do not flush TLB, this will still adversely
405 	 * affect performance on pre-Haswell Intel EPT, where there is
406 	 * no EPT Access Bit to clear so that we have to tear down EPT
407 	 * tables instead. If we find this unacceptable, we can always
408 	 * add a parameter to kvm_age_hva so that it effectively doesn't
409 	 * do anything on clear_young.
410 	 *
411 	 * Also note that currently we never issue secondary TLB flushes
412 	 * from clear_young, leaving this job up to the regular system
413 	 * cadence. If we find this inaccurate, we might come up with a
414 	 * more sophisticated heuristic later.
415 	 */
416 	young = kvm_age_hva(kvm, start, end);
417 	spin_unlock(&kvm->mmu_lock);
418 	srcu_read_unlock(&kvm->srcu, idx);
419 
420 	return young;
421 }
422 
423 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
424 				       struct mm_struct *mm,
425 				       unsigned long address)
426 {
427 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
428 	int young, idx;
429 
430 	idx = srcu_read_lock(&kvm->srcu);
431 	spin_lock(&kvm->mmu_lock);
432 	young = kvm_test_age_hva(kvm, address);
433 	spin_unlock(&kvm->mmu_lock);
434 	srcu_read_unlock(&kvm->srcu, idx);
435 
436 	return young;
437 }
438 
439 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
440 				     struct mm_struct *mm)
441 {
442 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
443 	int idx;
444 
445 	idx = srcu_read_lock(&kvm->srcu);
446 	kvm_arch_flush_shadow_all(kvm);
447 	srcu_read_unlock(&kvm->srcu, idx);
448 }
449 
450 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
451 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
452 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
453 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
454 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
455 	.clear_young		= kvm_mmu_notifier_clear_young,
456 	.test_young		= kvm_mmu_notifier_test_young,
457 	.change_pte		= kvm_mmu_notifier_change_pte,
458 	.release		= kvm_mmu_notifier_release,
459 };
460 
461 static int kvm_init_mmu_notifier(struct kvm *kvm)
462 {
463 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
464 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
465 }
466 
467 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
468 
469 static int kvm_init_mmu_notifier(struct kvm *kvm)
470 {
471 	return 0;
472 }
473 
474 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
475 
476 static struct kvm_memslots *kvm_alloc_memslots(void)
477 {
478 	int i;
479 	struct kvm_memslots *slots;
480 
481 	slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
482 	if (!slots)
483 		return NULL;
484 
485 	/*
486 	 * Init kvm generation close to the maximum to easily test the
487 	 * code of handling generation number wrap-around.
488 	 */
489 	slots->generation = -150;
490 	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
491 		slots->id_to_index[i] = slots->memslots[i].id = i;
492 
493 	return slots;
494 }
495 
496 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
497 {
498 	if (!memslot->dirty_bitmap)
499 		return;
500 
501 	kvfree(memslot->dirty_bitmap);
502 	memslot->dirty_bitmap = NULL;
503 }
504 
505 /*
506  * Free any memory in @free but not in @dont.
507  */
508 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
509 			      struct kvm_memory_slot *dont)
510 {
511 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
512 		kvm_destroy_dirty_bitmap(free);
513 
514 	kvm_arch_free_memslot(kvm, free, dont);
515 
516 	free->npages = 0;
517 }
518 
519 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
520 {
521 	struct kvm_memory_slot *memslot;
522 
523 	if (!slots)
524 		return;
525 
526 	kvm_for_each_memslot(memslot, slots)
527 		kvm_free_memslot(kvm, memslot, NULL);
528 
529 	kvfree(slots);
530 }
531 
532 static struct kvm *kvm_create_vm(unsigned long type)
533 {
534 	int r, i;
535 	struct kvm *kvm = kvm_arch_alloc_vm();
536 
537 	if (!kvm)
538 		return ERR_PTR(-ENOMEM);
539 
540 	r = kvm_arch_init_vm(kvm, type);
541 	if (r)
542 		goto out_err_no_disable;
543 
544 	r = hardware_enable_all();
545 	if (r)
546 		goto out_err_no_disable;
547 
548 #ifdef CONFIG_HAVE_KVM_IRQFD
549 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
550 #endif
551 
552 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
553 
554 	r = -ENOMEM;
555 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
556 		kvm->memslots[i] = kvm_alloc_memslots();
557 		if (!kvm->memslots[i])
558 			goto out_err_no_srcu;
559 	}
560 
561 	if (init_srcu_struct(&kvm->srcu))
562 		goto out_err_no_srcu;
563 	if (init_srcu_struct(&kvm->irq_srcu))
564 		goto out_err_no_irq_srcu;
565 	for (i = 0; i < KVM_NR_BUSES; i++) {
566 		kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
567 					GFP_KERNEL);
568 		if (!kvm->buses[i])
569 			goto out_err;
570 	}
571 
572 	spin_lock_init(&kvm->mmu_lock);
573 	kvm->mm = current->mm;
574 	atomic_inc(&kvm->mm->mm_count);
575 	kvm_eventfd_init(kvm);
576 	mutex_init(&kvm->lock);
577 	mutex_init(&kvm->irq_lock);
578 	mutex_init(&kvm->slots_lock);
579 	atomic_set(&kvm->users_count, 1);
580 	INIT_LIST_HEAD(&kvm->devices);
581 
582 	r = kvm_init_mmu_notifier(kvm);
583 	if (r)
584 		goto out_err;
585 
586 	spin_lock(&kvm_lock);
587 	list_add(&kvm->vm_list, &vm_list);
588 	spin_unlock(&kvm_lock);
589 
590 	preempt_notifier_inc();
591 
592 	return kvm;
593 
594 out_err:
595 	cleanup_srcu_struct(&kvm->irq_srcu);
596 out_err_no_irq_srcu:
597 	cleanup_srcu_struct(&kvm->srcu);
598 out_err_no_srcu:
599 	hardware_disable_all();
600 out_err_no_disable:
601 	for (i = 0; i < KVM_NR_BUSES; i++)
602 		kfree(kvm->buses[i]);
603 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
604 		kvm_free_memslots(kvm, kvm->memslots[i]);
605 	kvm_arch_free_vm(kvm);
606 	return ERR_PTR(r);
607 }
608 
609 /*
610  * Avoid using vmalloc for a small buffer.
611  * Should not be used when the size is statically known.
612  */
613 void *kvm_kvzalloc(unsigned long size)
614 {
615 	if (size > PAGE_SIZE)
616 		return vzalloc(size);
617 	else
618 		return kzalloc(size, GFP_KERNEL);
619 }
620 
621 static void kvm_destroy_devices(struct kvm *kvm)
622 {
623 	struct list_head *node, *tmp;
624 
625 	list_for_each_safe(node, tmp, &kvm->devices) {
626 		struct kvm_device *dev =
627 			list_entry(node, struct kvm_device, vm_node);
628 
629 		list_del(node);
630 		dev->ops->destroy(dev);
631 	}
632 }
633 
634 static void kvm_destroy_vm(struct kvm *kvm)
635 {
636 	int i;
637 	struct mm_struct *mm = kvm->mm;
638 
639 	kvm_arch_sync_events(kvm);
640 	spin_lock(&kvm_lock);
641 	list_del(&kvm->vm_list);
642 	spin_unlock(&kvm_lock);
643 	kvm_free_irq_routing(kvm);
644 	for (i = 0; i < KVM_NR_BUSES; i++)
645 		kvm_io_bus_destroy(kvm->buses[i]);
646 	kvm_coalesced_mmio_free(kvm);
647 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
648 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
649 #else
650 	kvm_arch_flush_shadow_all(kvm);
651 #endif
652 	kvm_arch_destroy_vm(kvm);
653 	kvm_destroy_devices(kvm);
654 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
655 		kvm_free_memslots(kvm, kvm->memslots[i]);
656 	cleanup_srcu_struct(&kvm->irq_srcu);
657 	cleanup_srcu_struct(&kvm->srcu);
658 	kvm_arch_free_vm(kvm);
659 	preempt_notifier_dec();
660 	hardware_disable_all();
661 	mmdrop(mm);
662 }
663 
664 void kvm_get_kvm(struct kvm *kvm)
665 {
666 	atomic_inc(&kvm->users_count);
667 }
668 EXPORT_SYMBOL_GPL(kvm_get_kvm);
669 
670 void kvm_put_kvm(struct kvm *kvm)
671 {
672 	if (atomic_dec_and_test(&kvm->users_count))
673 		kvm_destroy_vm(kvm);
674 }
675 EXPORT_SYMBOL_GPL(kvm_put_kvm);
676 
677 
678 static int kvm_vm_release(struct inode *inode, struct file *filp)
679 {
680 	struct kvm *kvm = filp->private_data;
681 
682 	kvm_irqfd_release(kvm);
683 
684 	kvm_put_kvm(kvm);
685 	return 0;
686 }
687 
688 /*
689  * Allocation size is twice as large as the actual dirty bitmap size.
690  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
691  */
692 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
693 {
694 	unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
695 
696 	memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
697 	if (!memslot->dirty_bitmap)
698 		return -ENOMEM;
699 
700 	return 0;
701 }
702 
703 /*
704  * Insert memslot and re-sort memslots based on their GFN,
705  * so binary search could be used to lookup GFN.
706  * Sorting algorithm takes advantage of having initially
707  * sorted array and known changed memslot position.
708  */
709 static void update_memslots(struct kvm_memslots *slots,
710 			    struct kvm_memory_slot *new)
711 {
712 	int id = new->id;
713 	int i = slots->id_to_index[id];
714 	struct kvm_memory_slot *mslots = slots->memslots;
715 
716 	WARN_ON(mslots[i].id != id);
717 	if (!new->npages) {
718 		WARN_ON(!mslots[i].npages);
719 		if (mslots[i].npages)
720 			slots->used_slots--;
721 	} else {
722 		if (!mslots[i].npages)
723 			slots->used_slots++;
724 	}
725 
726 	while (i < KVM_MEM_SLOTS_NUM - 1 &&
727 	       new->base_gfn <= mslots[i + 1].base_gfn) {
728 		if (!mslots[i + 1].npages)
729 			break;
730 		mslots[i] = mslots[i + 1];
731 		slots->id_to_index[mslots[i].id] = i;
732 		i++;
733 	}
734 
735 	/*
736 	 * The ">=" is needed when creating a slot with base_gfn == 0,
737 	 * so that it moves before all those with base_gfn == npages == 0.
738 	 *
739 	 * On the other hand, if new->npages is zero, the above loop has
740 	 * already left i pointing to the beginning of the empty part of
741 	 * mslots, and the ">=" would move the hole backwards in this
742 	 * case---which is wrong.  So skip the loop when deleting a slot.
743 	 */
744 	if (new->npages) {
745 		while (i > 0 &&
746 		       new->base_gfn >= mslots[i - 1].base_gfn) {
747 			mslots[i] = mslots[i - 1];
748 			slots->id_to_index[mslots[i].id] = i;
749 			i--;
750 		}
751 	} else
752 		WARN_ON_ONCE(i != slots->used_slots);
753 
754 	mslots[i] = *new;
755 	slots->id_to_index[mslots[i].id] = i;
756 }
757 
758 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
759 {
760 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
761 
762 #ifdef __KVM_HAVE_READONLY_MEM
763 	valid_flags |= KVM_MEM_READONLY;
764 #endif
765 
766 	if (mem->flags & ~valid_flags)
767 		return -EINVAL;
768 
769 	return 0;
770 }
771 
772 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
773 		int as_id, struct kvm_memslots *slots)
774 {
775 	struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
776 
777 	/*
778 	 * Set the low bit in the generation, which disables SPTE caching
779 	 * until the end of synchronize_srcu_expedited.
780 	 */
781 	WARN_ON(old_memslots->generation & 1);
782 	slots->generation = old_memslots->generation + 1;
783 
784 	rcu_assign_pointer(kvm->memslots[as_id], slots);
785 	synchronize_srcu_expedited(&kvm->srcu);
786 
787 	/*
788 	 * Increment the new memslot generation a second time. This prevents
789 	 * vm exits that race with memslot updates from caching a memslot
790 	 * generation that will (potentially) be valid forever.
791 	 */
792 	slots->generation++;
793 
794 	kvm_arch_memslots_updated(kvm, slots);
795 
796 	return old_memslots;
797 }
798 
799 /*
800  * Allocate some memory and give it an address in the guest physical address
801  * space.
802  *
803  * Discontiguous memory is allowed, mostly for framebuffers.
804  *
805  * Must be called holding kvm->slots_lock for write.
806  */
807 int __kvm_set_memory_region(struct kvm *kvm,
808 			    const struct kvm_userspace_memory_region *mem)
809 {
810 	int r;
811 	gfn_t base_gfn;
812 	unsigned long npages;
813 	struct kvm_memory_slot *slot;
814 	struct kvm_memory_slot old, new;
815 	struct kvm_memslots *slots = NULL, *old_memslots;
816 	int as_id, id;
817 	enum kvm_mr_change change;
818 
819 	r = check_memory_region_flags(mem);
820 	if (r)
821 		goto out;
822 
823 	r = -EINVAL;
824 	as_id = mem->slot >> 16;
825 	id = (u16)mem->slot;
826 
827 	/* General sanity checks */
828 	if (mem->memory_size & (PAGE_SIZE - 1))
829 		goto out;
830 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
831 		goto out;
832 	/* We can read the guest memory with __xxx_user() later on. */
833 	if ((id < KVM_USER_MEM_SLOTS) &&
834 	    ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
835 	     !access_ok(VERIFY_WRITE,
836 			(void __user *)(unsigned long)mem->userspace_addr,
837 			mem->memory_size)))
838 		goto out;
839 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
840 		goto out;
841 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
842 		goto out;
843 
844 	slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
845 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
846 	npages = mem->memory_size >> PAGE_SHIFT;
847 
848 	if (npages > KVM_MEM_MAX_NR_PAGES)
849 		goto out;
850 
851 	new = old = *slot;
852 
853 	new.id = id;
854 	new.base_gfn = base_gfn;
855 	new.npages = npages;
856 	new.flags = mem->flags;
857 
858 	if (npages) {
859 		if (!old.npages)
860 			change = KVM_MR_CREATE;
861 		else { /* Modify an existing slot. */
862 			if ((mem->userspace_addr != old.userspace_addr) ||
863 			    (npages != old.npages) ||
864 			    ((new.flags ^ old.flags) & KVM_MEM_READONLY))
865 				goto out;
866 
867 			if (base_gfn != old.base_gfn)
868 				change = KVM_MR_MOVE;
869 			else if (new.flags != old.flags)
870 				change = KVM_MR_FLAGS_ONLY;
871 			else { /* Nothing to change. */
872 				r = 0;
873 				goto out;
874 			}
875 		}
876 	} else {
877 		if (!old.npages)
878 			goto out;
879 
880 		change = KVM_MR_DELETE;
881 		new.base_gfn = 0;
882 		new.flags = 0;
883 	}
884 
885 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
886 		/* Check for overlaps */
887 		r = -EEXIST;
888 		kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
889 			if ((slot->id >= KVM_USER_MEM_SLOTS) ||
890 			    (slot->id == id))
891 				continue;
892 			if (!((base_gfn + npages <= slot->base_gfn) ||
893 			      (base_gfn >= slot->base_gfn + slot->npages)))
894 				goto out;
895 		}
896 	}
897 
898 	/* Free page dirty bitmap if unneeded */
899 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
900 		new.dirty_bitmap = NULL;
901 
902 	r = -ENOMEM;
903 	if (change == KVM_MR_CREATE) {
904 		new.userspace_addr = mem->userspace_addr;
905 
906 		if (kvm_arch_create_memslot(kvm, &new, npages))
907 			goto out_free;
908 	}
909 
910 	/* Allocate page dirty bitmap if needed */
911 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
912 		if (kvm_create_dirty_bitmap(&new) < 0)
913 			goto out_free;
914 	}
915 
916 	slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
917 	if (!slots)
918 		goto out_free;
919 	memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
920 
921 	if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
922 		slot = id_to_memslot(slots, id);
923 		slot->flags |= KVM_MEMSLOT_INVALID;
924 
925 		old_memslots = install_new_memslots(kvm, as_id, slots);
926 
927 		/* slot was deleted or moved, clear iommu mapping */
928 		kvm_iommu_unmap_pages(kvm, &old);
929 		/* From this point no new shadow pages pointing to a deleted,
930 		 * or moved, memslot will be created.
931 		 *
932 		 * validation of sp->gfn happens in:
933 		 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
934 		 *	- kvm_is_visible_gfn (mmu_check_roots)
935 		 */
936 		kvm_arch_flush_shadow_memslot(kvm, slot);
937 
938 		/*
939 		 * We can re-use the old_memslots from above, the only difference
940 		 * from the currently installed memslots is the invalid flag.  This
941 		 * will get overwritten by update_memslots anyway.
942 		 */
943 		slots = old_memslots;
944 	}
945 
946 	r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
947 	if (r)
948 		goto out_slots;
949 
950 	/* actual memory is freed via old in kvm_free_memslot below */
951 	if (change == KVM_MR_DELETE) {
952 		new.dirty_bitmap = NULL;
953 		memset(&new.arch, 0, sizeof(new.arch));
954 	}
955 
956 	update_memslots(slots, &new);
957 	old_memslots = install_new_memslots(kvm, as_id, slots);
958 
959 	kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
960 
961 	kvm_free_memslot(kvm, &old, &new);
962 	kvfree(old_memslots);
963 
964 	/*
965 	 * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
966 	 * un-mapped and re-mapped if their base changes.  Since base change
967 	 * unmapping is handled above with slot deletion, mapping alone is
968 	 * needed here.  Anything else the iommu might care about for existing
969 	 * slots (size changes, userspace addr changes and read-only flag
970 	 * changes) is disallowed above, so any other attribute changes getting
971 	 * here can be skipped.
972 	 */
973 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
974 		r = kvm_iommu_map_pages(kvm, &new);
975 		return r;
976 	}
977 
978 	return 0;
979 
980 out_slots:
981 	kvfree(slots);
982 out_free:
983 	kvm_free_memslot(kvm, &new, &old);
984 out:
985 	return r;
986 }
987 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
988 
989 int kvm_set_memory_region(struct kvm *kvm,
990 			  const struct kvm_userspace_memory_region *mem)
991 {
992 	int r;
993 
994 	mutex_lock(&kvm->slots_lock);
995 	r = __kvm_set_memory_region(kvm, mem);
996 	mutex_unlock(&kvm->slots_lock);
997 	return r;
998 }
999 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1000 
1001 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1002 					  struct kvm_userspace_memory_region *mem)
1003 {
1004 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1005 		return -EINVAL;
1006 
1007 	return kvm_set_memory_region(kvm, mem);
1008 }
1009 
1010 int kvm_get_dirty_log(struct kvm *kvm,
1011 			struct kvm_dirty_log *log, int *is_dirty)
1012 {
1013 	struct kvm_memslots *slots;
1014 	struct kvm_memory_slot *memslot;
1015 	int r, i, as_id, id;
1016 	unsigned long n;
1017 	unsigned long any = 0;
1018 
1019 	r = -EINVAL;
1020 	as_id = log->slot >> 16;
1021 	id = (u16)log->slot;
1022 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1023 		goto out;
1024 
1025 	slots = __kvm_memslots(kvm, as_id);
1026 	memslot = id_to_memslot(slots, id);
1027 	r = -ENOENT;
1028 	if (!memslot->dirty_bitmap)
1029 		goto out;
1030 
1031 	n = kvm_dirty_bitmap_bytes(memslot);
1032 
1033 	for (i = 0; !any && i < n/sizeof(long); ++i)
1034 		any = memslot->dirty_bitmap[i];
1035 
1036 	r = -EFAULT;
1037 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1038 		goto out;
1039 
1040 	if (any)
1041 		*is_dirty = 1;
1042 
1043 	r = 0;
1044 out:
1045 	return r;
1046 }
1047 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1048 
1049 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1050 /**
1051  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1052  *	are dirty write protect them for next write.
1053  * @kvm:	pointer to kvm instance
1054  * @log:	slot id and address to which we copy the log
1055  * @is_dirty:	flag set if any page is dirty
1056  *
1057  * We need to keep it in mind that VCPU threads can write to the bitmap
1058  * concurrently. So, to avoid losing track of dirty pages we keep the
1059  * following order:
1060  *
1061  *    1. Take a snapshot of the bit and clear it if needed.
1062  *    2. Write protect the corresponding page.
1063  *    3. Copy the snapshot to the userspace.
1064  *    4. Upon return caller flushes TLB's if needed.
1065  *
1066  * Between 2 and 4, the guest may write to the page using the remaining TLB
1067  * entry.  This is not a problem because the page is reported dirty using
1068  * the snapshot taken before and step 4 ensures that writes done after
1069  * exiting to userspace will be logged for the next call.
1070  *
1071  */
1072 int kvm_get_dirty_log_protect(struct kvm *kvm,
1073 			struct kvm_dirty_log *log, bool *is_dirty)
1074 {
1075 	struct kvm_memslots *slots;
1076 	struct kvm_memory_slot *memslot;
1077 	int r, i, as_id, id;
1078 	unsigned long n;
1079 	unsigned long *dirty_bitmap;
1080 	unsigned long *dirty_bitmap_buffer;
1081 
1082 	r = -EINVAL;
1083 	as_id = log->slot >> 16;
1084 	id = (u16)log->slot;
1085 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1086 		goto out;
1087 
1088 	slots = __kvm_memslots(kvm, as_id);
1089 	memslot = id_to_memslot(slots, id);
1090 
1091 	dirty_bitmap = memslot->dirty_bitmap;
1092 	r = -ENOENT;
1093 	if (!dirty_bitmap)
1094 		goto out;
1095 
1096 	n = kvm_dirty_bitmap_bytes(memslot);
1097 
1098 	dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1099 	memset(dirty_bitmap_buffer, 0, n);
1100 
1101 	spin_lock(&kvm->mmu_lock);
1102 	*is_dirty = false;
1103 	for (i = 0; i < n / sizeof(long); i++) {
1104 		unsigned long mask;
1105 		gfn_t offset;
1106 
1107 		if (!dirty_bitmap[i])
1108 			continue;
1109 
1110 		*is_dirty = true;
1111 
1112 		mask = xchg(&dirty_bitmap[i], 0);
1113 		dirty_bitmap_buffer[i] = mask;
1114 
1115 		if (mask) {
1116 			offset = i * BITS_PER_LONG;
1117 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1118 								offset, mask);
1119 		}
1120 	}
1121 
1122 	spin_unlock(&kvm->mmu_lock);
1123 
1124 	r = -EFAULT;
1125 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1126 		goto out;
1127 
1128 	r = 0;
1129 out:
1130 	return r;
1131 }
1132 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1133 #endif
1134 
1135 bool kvm_largepages_enabled(void)
1136 {
1137 	return largepages_enabled;
1138 }
1139 
1140 void kvm_disable_largepages(void)
1141 {
1142 	largepages_enabled = false;
1143 }
1144 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1145 
1146 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1147 {
1148 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1149 }
1150 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1151 
1152 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1153 {
1154 	return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1155 }
1156 
1157 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1158 {
1159 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1160 
1161 	if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1162 	      memslot->flags & KVM_MEMSLOT_INVALID)
1163 		return false;
1164 
1165 	return true;
1166 }
1167 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1168 
1169 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1170 {
1171 	struct vm_area_struct *vma;
1172 	unsigned long addr, size;
1173 
1174 	size = PAGE_SIZE;
1175 
1176 	addr = gfn_to_hva(kvm, gfn);
1177 	if (kvm_is_error_hva(addr))
1178 		return PAGE_SIZE;
1179 
1180 	down_read(&current->mm->mmap_sem);
1181 	vma = find_vma(current->mm, addr);
1182 	if (!vma)
1183 		goto out;
1184 
1185 	size = vma_kernel_pagesize(vma);
1186 
1187 out:
1188 	up_read(&current->mm->mmap_sem);
1189 
1190 	return size;
1191 }
1192 
1193 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1194 {
1195 	return slot->flags & KVM_MEM_READONLY;
1196 }
1197 
1198 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1199 				       gfn_t *nr_pages, bool write)
1200 {
1201 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1202 		return KVM_HVA_ERR_BAD;
1203 
1204 	if (memslot_is_readonly(slot) && write)
1205 		return KVM_HVA_ERR_RO_BAD;
1206 
1207 	if (nr_pages)
1208 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
1209 
1210 	return __gfn_to_hva_memslot(slot, gfn);
1211 }
1212 
1213 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1214 				     gfn_t *nr_pages)
1215 {
1216 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1217 }
1218 
1219 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1220 					gfn_t gfn)
1221 {
1222 	return gfn_to_hva_many(slot, gfn, NULL);
1223 }
1224 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1225 
1226 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1227 {
1228 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1229 }
1230 EXPORT_SYMBOL_GPL(gfn_to_hva);
1231 
1232 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1233 {
1234 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1235 }
1236 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1237 
1238 /*
1239  * If writable is set to false, the hva returned by this function is only
1240  * allowed to be read.
1241  */
1242 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1243 				      gfn_t gfn, bool *writable)
1244 {
1245 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1246 
1247 	if (!kvm_is_error_hva(hva) && writable)
1248 		*writable = !memslot_is_readonly(slot);
1249 
1250 	return hva;
1251 }
1252 
1253 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1254 {
1255 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1256 
1257 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1258 }
1259 
1260 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1261 {
1262 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1263 
1264 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1265 }
1266 
1267 static int get_user_page_nowait(struct task_struct *tsk, struct mm_struct *mm,
1268 	unsigned long start, int write, struct page **page)
1269 {
1270 	int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1271 
1272 	if (write)
1273 		flags |= FOLL_WRITE;
1274 
1275 	return __get_user_pages(tsk, mm, start, 1, flags, page, NULL, NULL);
1276 }
1277 
1278 static inline int check_user_page_hwpoison(unsigned long addr)
1279 {
1280 	int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1281 
1282 	rc = __get_user_pages(current, current->mm, addr, 1,
1283 			      flags, NULL, NULL, NULL);
1284 	return rc == -EHWPOISON;
1285 }
1286 
1287 /*
1288  * The atomic path to get the writable pfn which will be stored in @pfn,
1289  * true indicates success, otherwise false is returned.
1290  */
1291 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1292 			    bool write_fault, bool *writable, kvm_pfn_t *pfn)
1293 {
1294 	struct page *page[1];
1295 	int npages;
1296 
1297 	if (!(async || atomic))
1298 		return false;
1299 
1300 	/*
1301 	 * Fast pin a writable pfn only if it is a write fault request
1302 	 * or the caller allows to map a writable pfn for a read fault
1303 	 * request.
1304 	 */
1305 	if (!(write_fault || writable))
1306 		return false;
1307 
1308 	npages = __get_user_pages_fast(addr, 1, 1, page);
1309 	if (npages == 1) {
1310 		*pfn = page_to_pfn(page[0]);
1311 
1312 		if (writable)
1313 			*writable = true;
1314 		return true;
1315 	}
1316 
1317 	return false;
1318 }
1319 
1320 /*
1321  * The slow path to get the pfn of the specified host virtual address,
1322  * 1 indicates success, -errno is returned if error is detected.
1323  */
1324 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1325 			   bool *writable, kvm_pfn_t *pfn)
1326 {
1327 	struct page *page[1];
1328 	int npages = 0;
1329 
1330 	might_sleep();
1331 
1332 	if (writable)
1333 		*writable = write_fault;
1334 
1335 	if (async) {
1336 		down_read(&current->mm->mmap_sem);
1337 		npages = get_user_page_nowait(current, current->mm,
1338 					      addr, write_fault, page);
1339 		up_read(&current->mm->mmap_sem);
1340 	} else
1341 		npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1342 						   write_fault, 0, page,
1343 						   FOLL_TOUCH|FOLL_HWPOISON);
1344 	if (npages != 1)
1345 		return npages;
1346 
1347 	/* map read fault as writable if possible */
1348 	if (unlikely(!write_fault) && writable) {
1349 		struct page *wpage[1];
1350 
1351 		npages = __get_user_pages_fast(addr, 1, 1, wpage);
1352 		if (npages == 1) {
1353 			*writable = true;
1354 			put_page(page[0]);
1355 			page[0] = wpage[0];
1356 		}
1357 
1358 		npages = 1;
1359 	}
1360 	*pfn = page_to_pfn(page[0]);
1361 	return npages;
1362 }
1363 
1364 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1365 {
1366 	if (unlikely(!(vma->vm_flags & VM_READ)))
1367 		return false;
1368 
1369 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1370 		return false;
1371 
1372 	return true;
1373 }
1374 
1375 /*
1376  * Pin guest page in memory and return its pfn.
1377  * @addr: host virtual address which maps memory to the guest
1378  * @atomic: whether this function can sleep
1379  * @async: whether this function need to wait IO complete if the
1380  *         host page is not in the memory
1381  * @write_fault: whether we should get a writable host page
1382  * @writable: whether it allows to map a writable host page for !@write_fault
1383  *
1384  * The function will map a writable host page for these two cases:
1385  * 1): @write_fault = true
1386  * 2): @write_fault = false && @writable, @writable will tell the caller
1387  *     whether the mapping is writable.
1388  */
1389 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1390 			bool write_fault, bool *writable)
1391 {
1392 	struct vm_area_struct *vma;
1393 	kvm_pfn_t pfn = 0;
1394 	int npages;
1395 
1396 	/* we can do it either atomically or asynchronously, not both */
1397 	BUG_ON(atomic && async);
1398 
1399 	if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1400 		return pfn;
1401 
1402 	if (atomic)
1403 		return KVM_PFN_ERR_FAULT;
1404 
1405 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1406 	if (npages == 1)
1407 		return pfn;
1408 
1409 	down_read(&current->mm->mmap_sem);
1410 	if (npages == -EHWPOISON ||
1411 	      (!async && check_user_page_hwpoison(addr))) {
1412 		pfn = KVM_PFN_ERR_HWPOISON;
1413 		goto exit;
1414 	}
1415 
1416 	vma = find_vma_intersection(current->mm, addr, addr + 1);
1417 
1418 	if (vma == NULL)
1419 		pfn = KVM_PFN_ERR_FAULT;
1420 	else if ((vma->vm_flags & VM_PFNMAP)) {
1421 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1422 			vma->vm_pgoff;
1423 		BUG_ON(!kvm_is_reserved_pfn(pfn));
1424 	} else {
1425 		if (async && vma_is_valid(vma, write_fault))
1426 			*async = true;
1427 		pfn = KVM_PFN_ERR_FAULT;
1428 	}
1429 exit:
1430 	up_read(&current->mm->mmap_sem);
1431 	return pfn;
1432 }
1433 
1434 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1435 			       bool atomic, bool *async, bool write_fault,
1436 			       bool *writable)
1437 {
1438 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1439 
1440 	if (addr == KVM_HVA_ERR_RO_BAD)
1441 		return KVM_PFN_ERR_RO_FAULT;
1442 
1443 	if (kvm_is_error_hva(addr))
1444 		return KVM_PFN_NOSLOT;
1445 
1446 	/* Do not map writable pfn in the readonly memslot. */
1447 	if (writable && memslot_is_readonly(slot)) {
1448 		*writable = false;
1449 		writable = NULL;
1450 	}
1451 
1452 	return hva_to_pfn(addr, atomic, async, write_fault,
1453 			  writable);
1454 }
1455 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1456 
1457 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1458 		      bool *writable)
1459 {
1460 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1461 				    write_fault, writable);
1462 }
1463 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1464 
1465 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1466 {
1467 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1468 }
1469 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1470 
1471 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1472 {
1473 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1474 }
1475 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1476 
1477 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1478 {
1479 	return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1480 }
1481 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1482 
1483 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1484 {
1485 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1486 }
1487 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1488 
1489 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1490 {
1491 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1492 }
1493 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1494 
1495 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1496 {
1497 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1498 }
1499 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1500 
1501 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1502 			    struct page **pages, int nr_pages)
1503 {
1504 	unsigned long addr;
1505 	gfn_t entry;
1506 
1507 	addr = gfn_to_hva_many(slot, gfn, &entry);
1508 	if (kvm_is_error_hva(addr))
1509 		return -1;
1510 
1511 	if (entry < nr_pages)
1512 		return 0;
1513 
1514 	return __get_user_pages_fast(addr, nr_pages, 1, pages);
1515 }
1516 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1517 
1518 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1519 {
1520 	if (is_error_noslot_pfn(pfn))
1521 		return KVM_ERR_PTR_BAD_PAGE;
1522 
1523 	if (kvm_is_reserved_pfn(pfn)) {
1524 		WARN_ON(1);
1525 		return KVM_ERR_PTR_BAD_PAGE;
1526 	}
1527 
1528 	return pfn_to_page(pfn);
1529 }
1530 
1531 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1532 {
1533 	kvm_pfn_t pfn;
1534 
1535 	pfn = gfn_to_pfn(kvm, gfn);
1536 
1537 	return kvm_pfn_to_page(pfn);
1538 }
1539 EXPORT_SYMBOL_GPL(gfn_to_page);
1540 
1541 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1542 {
1543 	kvm_pfn_t pfn;
1544 
1545 	pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1546 
1547 	return kvm_pfn_to_page(pfn);
1548 }
1549 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1550 
1551 void kvm_release_page_clean(struct page *page)
1552 {
1553 	WARN_ON(is_error_page(page));
1554 
1555 	kvm_release_pfn_clean(page_to_pfn(page));
1556 }
1557 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1558 
1559 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1560 {
1561 	if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1562 		put_page(pfn_to_page(pfn));
1563 }
1564 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1565 
1566 void kvm_release_page_dirty(struct page *page)
1567 {
1568 	WARN_ON(is_error_page(page));
1569 
1570 	kvm_release_pfn_dirty(page_to_pfn(page));
1571 }
1572 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1573 
1574 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1575 {
1576 	kvm_set_pfn_dirty(pfn);
1577 	kvm_release_pfn_clean(pfn);
1578 }
1579 
1580 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1581 {
1582 	if (!kvm_is_reserved_pfn(pfn)) {
1583 		struct page *page = pfn_to_page(pfn);
1584 
1585 		if (!PageReserved(page))
1586 			SetPageDirty(page);
1587 	}
1588 }
1589 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1590 
1591 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1592 {
1593 	if (!kvm_is_reserved_pfn(pfn))
1594 		mark_page_accessed(pfn_to_page(pfn));
1595 }
1596 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1597 
1598 void kvm_get_pfn(kvm_pfn_t pfn)
1599 {
1600 	if (!kvm_is_reserved_pfn(pfn))
1601 		get_page(pfn_to_page(pfn));
1602 }
1603 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1604 
1605 static int next_segment(unsigned long len, int offset)
1606 {
1607 	if (len > PAGE_SIZE - offset)
1608 		return PAGE_SIZE - offset;
1609 	else
1610 		return len;
1611 }
1612 
1613 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1614 				 void *data, int offset, int len)
1615 {
1616 	int r;
1617 	unsigned long addr;
1618 
1619 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1620 	if (kvm_is_error_hva(addr))
1621 		return -EFAULT;
1622 	r = __copy_from_user(data, (void __user *)addr + offset, len);
1623 	if (r)
1624 		return -EFAULT;
1625 	return 0;
1626 }
1627 
1628 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1629 			int len)
1630 {
1631 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1632 
1633 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1634 }
1635 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1636 
1637 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1638 			     int offset, int len)
1639 {
1640 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1641 
1642 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1643 }
1644 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1645 
1646 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1647 {
1648 	gfn_t gfn = gpa >> PAGE_SHIFT;
1649 	int seg;
1650 	int offset = offset_in_page(gpa);
1651 	int ret;
1652 
1653 	while ((seg = next_segment(len, offset)) != 0) {
1654 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1655 		if (ret < 0)
1656 			return ret;
1657 		offset = 0;
1658 		len -= seg;
1659 		data += seg;
1660 		++gfn;
1661 	}
1662 	return 0;
1663 }
1664 EXPORT_SYMBOL_GPL(kvm_read_guest);
1665 
1666 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1667 {
1668 	gfn_t gfn = gpa >> PAGE_SHIFT;
1669 	int seg;
1670 	int offset = offset_in_page(gpa);
1671 	int ret;
1672 
1673 	while ((seg = next_segment(len, offset)) != 0) {
1674 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1675 		if (ret < 0)
1676 			return ret;
1677 		offset = 0;
1678 		len -= seg;
1679 		data += seg;
1680 		++gfn;
1681 	}
1682 	return 0;
1683 }
1684 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1685 
1686 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1687 			           void *data, int offset, unsigned long len)
1688 {
1689 	int r;
1690 	unsigned long addr;
1691 
1692 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1693 	if (kvm_is_error_hva(addr))
1694 		return -EFAULT;
1695 	pagefault_disable();
1696 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1697 	pagefault_enable();
1698 	if (r)
1699 		return -EFAULT;
1700 	return 0;
1701 }
1702 
1703 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1704 			  unsigned long len)
1705 {
1706 	gfn_t gfn = gpa >> PAGE_SHIFT;
1707 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1708 	int offset = offset_in_page(gpa);
1709 
1710 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1711 }
1712 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1713 
1714 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1715 			       void *data, unsigned long len)
1716 {
1717 	gfn_t gfn = gpa >> PAGE_SHIFT;
1718 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1719 	int offset = offset_in_page(gpa);
1720 
1721 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1722 }
1723 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1724 
1725 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1726 			          const void *data, int offset, int len)
1727 {
1728 	int r;
1729 	unsigned long addr;
1730 
1731 	addr = gfn_to_hva_memslot(memslot, gfn);
1732 	if (kvm_is_error_hva(addr))
1733 		return -EFAULT;
1734 	r = __copy_to_user((void __user *)addr + offset, data, len);
1735 	if (r)
1736 		return -EFAULT;
1737 	mark_page_dirty_in_slot(memslot, gfn);
1738 	return 0;
1739 }
1740 
1741 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1742 			 const void *data, int offset, int len)
1743 {
1744 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1745 
1746 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1747 }
1748 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1749 
1750 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1751 			      const void *data, int offset, int len)
1752 {
1753 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1754 
1755 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1756 }
1757 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1758 
1759 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1760 		    unsigned long len)
1761 {
1762 	gfn_t gfn = gpa >> PAGE_SHIFT;
1763 	int seg;
1764 	int offset = offset_in_page(gpa);
1765 	int ret;
1766 
1767 	while ((seg = next_segment(len, offset)) != 0) {
1768 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1769 		if (ret < 0)
1770 			return ret;
1771 		offset = 0;
1772 		len -= seg;
1773 		data += seg;
1774 		++gfn;
1775 	}
1776 	return 0;
1777 }
1778 EXPORT_SYMBOL_GPL(kvm_write_guest);
1779 
1780 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1781 		         unsigned long len)
1782 {
1783 	gfn_t gfn = gpa >> PAGE_SHIFT;
1784 	int seg;
1785 	int offset = offset_in_page(gpa);
1786 	int ret;
1787 
1788 	while ((seg = next_segment(len, offset)) != 0) {
1789 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1790 		if (ret < 0)
1791 			return ret;
1792 		offset = 0;
1793 		len -= seg;
1794 		data += seg;
1795 		++gfn;
1796 	}
1797 	return 0;
1798 }
1799 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1800 
1801 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1802 			      gpa_t gpa, unsigned long len)
1803 {
1804 	struct kvm_memslots *slots = kvm_memslots(kvm);
1805 	int offset = offset_in_page(gpa);
1806 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
1807 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1808 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1809 	gfn_t nr_pages_avail;
1810 
1811 	ghc->gpa = gpa;
1812 	ghc->generation = slots->generation;
1813 	ghc->len = len;
1814 	ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1815 	ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1816 	if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1817 		ghc->hva += offset;
1818 	} else {
1819 		/*
1820 		 * If the requested region crosses two memslots, we still
1821 		 * verify that the entire region is valid here.
1822 		 */
1823 		while (start_gfn <= end_gfn) {
1824 			ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1825 			ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1826 						   &nr_pages_avail);
1827 			if (kvm_is_error_hva(ghc->hva))
1828 				return -EFAULT;
1829 			start_gfn += nr_pages_avail;
1830 		}
1831 		/* Use the slow path for cross page reads and writes. */
1832 		ghc->memslot = NULL;
1833 	}
1834 	return 0;
1835 }
1836 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1837 
1838 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1839 			   void *data, unsigned long len)
1840 {
1841 	struct kvm_memslots *slots = kvm_memslots(kvm);
1842 	int r;
1843 
1844 	BUG_ON(len > ghc->len);
1845 
1846 	if (slots->generation != ghc->generation)
1847 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1848 
1849 	if (unlikely(!ghc->memslot))
1850 		return kvm_write_guest(kvm, ghc->gpa, data, len);
1851 
1852 	if (kvm_is_error_hva(ghc->hva))
1853 		return -EFAULT;
1854 
1855 	r = __copy_to_user((void __user *)ghc->hva, data, len);
1856 	if (r)
1857 		return -EFAULT;
1858 	mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1859 
1860 	return 0;
1861 }
1862 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1863 
1864 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1865 			   void *data, unsigned long len)
1866 {
1867 	struct kvm_memslots *slots = kvm_memslots(kvm);
1868 	int r;
1869 
1870 	BUG_ON(len > ghc->len);
1871 
1872 	if (slots->generation != ghc->generation)
1873 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1874 
1875 	if (unlikely(!ghc->memslot))
1876 		return kvm_read_guest(kvm, ghc->gpa, data, len);
1877 
1878 	if (kvm_is_error_hva(ghc->hva))
1879 		return -EFAULT;
1880 
1881 	r = __copy_from_user(data, (void __user *)ghc->hva, len);
1882 	if (r)
1883 		return -EFAULT;
1884 
1885 	return 0;
1886 }
1887 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1888 
1889 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1890 {
1891 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1892 
1893 	return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1894 }
1895 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1896 
1897 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1898 {
1899 	gfn_t gfn = gpa >> PAGE_SHIFT;
1900 	int seg;
1901 	int offset = offset_in_page(gpa);
1902 	int ret;
1903 
1904 	while ((seg = next_segment(len, offset)) != 0) {
1905 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1906 		if (ret < 0)
1907 			return ret;
1908 		offset = 0;
1909 		len -= seg;
1910 		++gfn;
1911 	}
1912 	return 0;
1913 }
1914 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1915 
1916 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1917 				    gfn_t gfn)
1918 {
1919 	if (memslot && memslot->dirty_bitmap) {
1920 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1921 
1922 		set_bit_le(rel_gfn, memslot->dirty_bitmap);
1923 	}
1924 }
1925 
1926 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1927 {
1928 	struct kvm_memory_slot *memslot;
1929 
1930 	memslot = gfn_to_memslot(kvm, gfn);
1931 	mark_page_dirty_in_slot(memslot, gfn);
1932 }
1933 EXPORT_SYMBOL_GPL(mark_page_dirty);
1934 
1935 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1936 {
1937 	struct kvm_memory_slot *memslot;
1938 
1939 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1940 	mark_page_dirty_in_slot(memslot, gfn);
1941 }
1942 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1943 
1944 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1945 {
1946 	int old, val;
1947 
1948 	old = val = vcpu->halt_poll_ns;
1949 	/* 10us base */
1950 	if (val == 0 && halt_poll_ns_grow)
1951 		val = 10000;
1952 	else
1953 		val *= halt_poll_ns_grow;
1954 
1955 	vcpu->halt_poll_ns = val;
1956 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1957 }
1958 
1959 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1960 {
1961 	int old, val;
1962 
1963 	old = val = vcpu->halt_poll_ns;
1964 	if (halt_poll_ns_shrink == 0)
1965 		val = 0;
1966 	else
1967 		val /= halt_poll_ns_shrink;
1968 
1969 	vcpu->halt_poll_ns = val;
1970 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1971 }
1972 
1973 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1974 {
1975 	if (kvm_arch_vcpu_runnable(vcpu)) {
1976 		kvm_make_request(KVM_REQ_UNHALT, vcpu);
1977 		return -EINTR;
1978 	}
1979 	if (kvm_cpu_has_pending_timer(vcpu))
1980 		return -EINTR;
1981 	if (signal_pending(current))
1982 		return -EINTR;
1983 
1984 	return 0;
1985 }
1986 
1987 /*
1988  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
1989  */
1990 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1991 {
1992 	ktime_t start, cur;
1993 	DEFINE_WAIT(wait);
1994 	bool waited = false;
1995 	u64 block_ns;
1996 
1997 	start = cur = ktime_get();
1998 	if (vcpu->halt_poll_ns) {
1999 		ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2000 
2001 		++vcpu->stat.halt_attempted_poll;
2002 		do {
2003 			/*
2004 			 * This sets KVM_REQ_UNHALT if an interrupt
2005 			 * arrives.
2006 			 */
2007 			if (kvm_vcpu_check_block(vcpu) < 0) {
2008 				++vcpu->stat.halt_successful_poll;
2009 				goto out;
2010 			}
2011 			cur = ktime_get();
2012 		} while (single_task_running() && ktime_before(cur, stop));
2013 	}
2014 
2015 	kvm_arch_vcpu_blocking(vcpu);
2016 
2017 	for (;;) {
2018 		prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2019 
2020 		if (kvm_vcpu_check_block(vcpu) < 0)
2021 			break;
2022 
2023 		waited = true;
2024 		schedule();
2025 	}
2026 
2027 	finish_wait(&vcpu->wq, &wait);
2028 	cur = ktime_get();
2029 
2030 	kvm_arch_vcpu_unblocking(vcpu);
2031 out:
2032 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2033 
2034 	if (halt_poll_ns) {
2035 		if (block_ns <= vcpu->halt_poll_ns)
2036 			;
2037 		/* we had a long block, shrink polling */
2038 		else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2039 			shrink_halt_poll_ns(vcpu);
2040 		/* we had a short halt and our poll time is too small */
2041 		else if (vcpu->halt_poll_ns < halt_poll_ns &&
2042 			block_ns < halt_poll_ns)
2043 			grow_halt_poll_ns(vcpu);
2044 	} else
2045 		vcpu->halt_poll_ns = 0;
2046 
2047 	trace_kvm_vcpu_wakeup(block_ns, waited);
2048 }
2049 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2050 
2051 #ifndef CONFIG_S390
2052 /*
2053  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2054  */
2055 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2056 {
2057 	int me;
2058 	int cpu = vcpu->cpu;
2059 	wait_queue_head_t *wqp;
2060 
2061 	wqp = kvm_arch_vcpu_wq(vcpu);
2062 	if (waitqueue_active(wqp)) {
2063 		wake_up_interruptible(wqp);
2064 		++vcpu->stat.halt_wakeup;
2065 	}
2066 
2067 	me = get_cpu();
2068 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2069 		if (kvm_arch_vcpu_should_kick(vcpu))
2070 			smp_send_reschedule(cpu);
2071 	put_cpu();
2072 }
2073 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2074 #endif /* !CONFIG_S390 */
2075 
2076 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2077 {
2078 	struct pid *pid;
2079 	struct task_struct *task = NULL;
2080 	int ret = 0;
2081 
2082 	rcu_read_lock();
2083 	pid = rcu_dereference(target->pid);
2084 	if (pid)
2085 		task = get_pid_task(pid, PIDTYPE_PID);
2086 	rcu_read_unlock();
2087 	if (!task)
2088 		return ret;
2089 	ret = yield_to(task, 1);
2090 	put_task_struct(task);
2091 
2092 	return ret;
2093 }
2094 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2095 
2096 /*
2097  * Helper that checks whether a VCPU is eligible for directed yield.
2098  * Most eligible candidate to yield is decided by following heuristics:
2099  *
2100  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2101  *  (preempted lock holder), indicated by @in_spin_loop.
2102  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2103  *
2104  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2105  *  chance last time (mostly it has become eligible now since we have probably
2106  *  yielded to lockholder in last iteration. This is done by toggling
2107  *  @dy_eligible each time a VCPU checked for eligibility.)
2108  *
2109  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2110  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2111  *  burning. Giving priority for a potential lock-holder increases lock
2112  *  progress.
2113  *
2114  *  Since algorithm is based on heuristics, accessing another VCPU data without
2115  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2116  *  and continue with next VCPU and so on.
2117  */
2118 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2119 {
2120 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2121 	bool eligible;
2122 
2123 	eligible = !vcpu->spin_loop.in_spin_loop ||
2124 		    vcpu->spin_loop.dy_eligible;
2125 
2126 	if (vcpu->spin_loop.in_spin_loop)
2127 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2128 
2129 	return eligible;
2130 #else
2131 	return true;
2132 #endif
2133 }
2134 
2135 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2136 {
2137 	struct kvm *kvm = me->kvm;
2138 	struct kvm_vcpu *vcpu;
2139 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2140 	int yielded = 0;
2141 	int try = 3;
2142 	int pass;
2143 	int i;
2144 
2145 	kvm_vcpu_set_in_spin_loop(me, true);
2146 	/*
2147 	 * We boost the priority of a VCPU that is runnable but not
2148 	 * currently running, because it got preempted by something
2149 	 * else and called schedule in __vcpu_run.  Hopefully that
2150 	 * VCPU is holding the lock that we need and will release it.
2151 	 * We approximate round-robin by starting at the last boosted VCPU.
2152 	 */
2153 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
2154 		kvm_for_each_vcpu(i, vcpu, kvm) {
2155 			if (!pass && i <= last_boosted_vcpu) {
2156 				i = last_boosted_vcpu;
2157 				continue;
2158 			} else if (pass && i > last_boosted_vcpu)
2159 				break;
2160 			if (!ACCESS_ONCE(vcpu->preempted))
2161 				continue;
2162 			if (vcpu == me)
2163 				continue;
2164 			if (waitqueue_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2165 				continue;
2166 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2167 				continue;
2168 
2169 			yielded = kvm_vcpu_yield_to(vcpu);
2170 			if (yielded > 0) {
2171 				kvm->last_boosted_vcpu = i;
2172 				break;
2173 			} else if (yielded < 0) {
2174 				try--;
2175 				if (!try)
2176 					break;
2177 			}
2178 		}
2179 	}
2180 	kvm_vcpu_set_in_spin_loop(me, false);
2181 
2182 	/* Ensure vcpu is not eligible during next spinloop */
2183 	kvm_vcpu_set_dy_eligible(me, false);
2184 }
2185 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2186 
2187 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2188 {
2189 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2190 	struct page *page;
2191 
2192 	if (vmf->pgoff == 0)
2193 		page = virt_to_page(vcpu->run);
2194 #ifdef CONFIG_X86
2195 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2196 		page = virt_to_page(vcpu->arch.pio_data);
2197 #endif
2198 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2199 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2200 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2201 #endif
2202 	else
2203 		return kvm_arch_vcpu_fault(vcpu, vmf);
2204 	get_page(page);
2205 	vmf->page = page;
2206 	return 0;
2207 }
2208 
2209 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2210 	.fault = kvm_vcpu_fault,
2211 };
2212 
2213 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2214 {
2215 	vma->vm_ops = &kvm_vcpu_vm_ops;
2216 	return 0;
2217 }
2218 
2219 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2220 {
2221 	struct kvm_vcpu *vcpu = filp->private_data;
2222 
2223 	kvm_put_kvm(vcpu->kvm);
2224 	return 0;
2225 }
2226 
2227 static struct file_operations kvm_vcpu_fops = {
2228 	.release        = kvm_vcpu_release,
2229 	.unlocked_ioctl = kvm_vcpu_ioctl,
2230 #ifdef CONFIG_KVM_COMPAT
2231 	.compat_ioctl   = kvm_vcpu_compat_ioctl,
2232 #endif
2233 	.mmap           = kvm_vcpu_mmap,
2234 	.llseek		= noop_llseek,
2235 };
2236 
2237 /*
2238  * Allocates an inode for the vcpu.
2239  */
2240 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2241 {
2242 	return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2243 }
2244 
2245 /*
2246  * Creates some virtual cpus.  Good luck creating more than one.
2247  */
2248 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2249 {
2250 	int r;
2251 	struct kvm_vcpu *vcpu;
2252 
2253 	if (id >= KVM_MAX_VCPUS)
2254 		return -EINVAL;
2255 
2256 	vcpu = kvm_arch_vcpu_create(kvm, id);
2257 	if (IS_ERR(vcpu))
2258 		return PTR_ERR(vcpu);
2259 
2260 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2261 
2262 	r = kvm_arch_vcpu_setup(vcpu);
2263 	if (r)
2264 		goto vcpu_destroy;
2265 
2266 	mutex_lock(&kvm->lock);
2267 	if (!kvm_vcpu_compatible(vcpu)) {
2268 		r = -EINVAL;
2269 		goto unlock_vcpu_destroy;
2270 	}
2271 	if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2272 		r = -EINVAL;
2273 		goto unlock_vcpu_destroy;
2274 	}
2275 	if (kvm_get_vcpu_by_id(kvm, id)) {
2276 		r = -EEXIST;
2277 		goto unlock_vcpu_destroy;
2278 	}
2279 
2280 	BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2281 
2282 	/* Now it's all set up, let userspace reach it */
2283 	kvm_get_kvm(kvm);
2284 	r = create_vcpu_fd(vcpu);
2285 	if (r < 0) {
2286 		kvm_put_kvm(kvm);
2287 		goto unlock_vcpu_destroy;
2288 	}
2289 
2290 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2291 
2292 	/*
2293 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2294 	 * before kvm->online_vcpu's incremented value.
2295 	 */
2296 	smp_wmb();
2297 	atomic_inc(&kvm->online_vcpus);
2298 
2299 	mutex_unlock(&kvm->lock);
2300 	kvm_arch_vcpu_postcreate(vcpu);
2301 	return r;
2302 
2303 unlock_vcpu_destroy:
2304 	mutex_unlock(&kvm->lock);
2305 vcpu_destroy:
2306 	kvm_arch_vcpu_destroy(vcpu);
2307 	return r;
2308 }
2309 
2310 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2311 {
2312 	if (sigset) {
2313 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2314 		vcpu->sigset_active = 1;
2315 		vcpu->sigset = *sigset;
2316 	} else
2317 		vcpu->sigset_active = 0;
2318 	return 0;
2319 }
2320 
2321 static long kvm_vcpu_ioctl(struct file *filp,
2322 			   unsigned int ioctl, unsigned long arg)
2323 {
2324 	struct kvm_vcpu *vcpu = filp->private_data;
2325 	void __user *argp = (void __user *)arg;
2326 	int r;
2327 	struct kvm_fpu *fpu = NULL;
2328 	struct kvm_sregs *kvm_sregs = NULL;
2329 
2330 	if (vcpu->kvm->mm != current->mm)
2331 		return -EIO;
2332 
2333 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2334 		return -EINVAL;
2335 
2336 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2337 	/*
2338 	 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2339 	 * so vcpu_load() would break it.
2340 	 */
2341 	if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2342 		return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2343 #endif
2344 
2345 
2346 	r = vcpu_load(vcpu);
2347 	if (r)
2348 		return r;
2349 	switch (ioctl) {
2350 	case KVM_RUN:
2351 		r = -EINVAL;
2352 		if (arg)
2353 			goto out;
2354 		if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2355 			/* The thread running this VCPU changed. */
2356 			struct pid *oldpid = vcpu->pid;
2357 			struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2358 
2359 			rcu_assign_pointer(vcpu->pid, newpid);
2360 			if (oldpid)
2361 				synchronize_rcu();
2362 			put_pid(oldpid);
2363 		}
2364 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2365 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2366 		break;
2367 	case KVM_GET_REGS: {
2368 		struct kvm_regs *kvm_regs;
2369 
2370 		r = -ENOMEM;
2371 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2372 		if (!kvm_regs)
2373 			goto out;
2374 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2375 		if (r)
2376 			goto out_free1;
2377 		r = -EFAULT;
2378 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2379 			goto out_free1;
2380 		r = 0;
2381 out_free1:
2382 		kfree(kvm_regs);
2383 		break;
2384 	}
2385 	case KVM_SET_REGS: {
2386 		struct kvm_regs *kvm_regs;
2387 
2388 		r = -ENOMEM;
2389 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2390 		if (IS_ERR(kvm_regs)) {
2391 			r = PTR_ERR(kvm_regs);
2392 			goto out;
2393 		}
2394 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2395 		kfree(kvm_regs);
2396 		break;
2397 	}
2398 	case KVM_GET_SREGS: {
2399 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2400 		r = -ENOMEM;
2401 		if (!kvm_sregs)
2402 			goto out;
2403 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2404 		if (r)
2405 			goto out;
2406 		r = -EFAULT;
2407 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2408 			goto out;
2409 		r = 0;
2410 		break;
2411 	}
2412 	case KVM_SET_SREGS: {
2413 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2414 		if (IS_ERR(kvm_sregs)) {
2415 			r = PTR_ERR(kvm_sregs);
2416 			kvm_sregs = NULL;
2417 			goto out;
2418 		}
2419 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2420 		break;
2421 	}
2422 	case KVM_GET_MP_STATE: {
2423 		struct kvm_mp_state mp_state;
2424 
2425 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2426 		if (r)
2427 			goto out;
2428 		r = -EFAULT;
2429 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2430 			goto out;
2431 		r = 0;
2432 		break;
2433 	}
2434 	case KVM_SET_MP_STATE: {
2435 		struct kvm_mp_state mp_state;
2436 
2437 		r = -EFAULT;
2438 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2439 			goto out;
2440 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2441 		break;
2442 	}
2443 	case KVM_TRANSLATE: {
2444 		struct kvm_translation tr;
2445 
2446 		r = -EFAULT;
2447 		if (copy_from_user(&tr, argp, sizeof(tr)))
2448 			goto out;
2449 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2450 		if (r)
2451 			goto out;
2452 		r = -EFAULT;
2453 		if (copy_to_user(argp, &tr, sizeof(tr)))
2454 			goto out;
2455 		r = 0;
2456 		break;
2457 	}
2458 	case KVM_SET_GUEST_DEBUG: {
2459 		struct kvm_guest_debug dbg;
2460 
2461 		r = -EFAULT;
2462 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
2463 			goto out;
2464 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2465 		break;
2466 	}
2467 	case KVM_SET_SIGNAL_MASK: {
2468 		struct kvm_signal_mask __user *sigmask_arg = argp;
2469 		struct kvm_signal_mask kvm_sigmask;
2470 		sigset_t sigset, *p;
2471 
2472 		p = NULL;
2473 		if (argp) {
2474 			r = -EFAULT;
2475 			if (copy_from_user(&kvm_sigmask, argp,
2476 					   sizeof(kvm_sigmask)))
2477 				goto out;
2478 			r = -EINVAL;
2479 			if (kvm_sigmask.len != sizeof(sigset))
2480 				goto out;
2481 			r = -EFAULT;
2482 			if (copy_from_user(&sigset, sigmask_arg->sigset,
2483 					   sizeof(sigset)))
2484 				goto out;
2485 			p = &sigset;
2486 		}
2487 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2488 		break;
2489 	}
2490 	case KVM_GET_FPU: {
2491 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2492 		r = -ENOMEM;
2493 		if (!fpu)
2494 			goto out;
2495 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2496 		if (r)
2497 			goto out;
2498 		r = -EFAULT;
2499 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2500 			goto out;
2501 		r = 0;
2502 		break;
2503 	}
2504 	case KVM_SET_FPU: {
2505 		fpu = memdup_user(argp, sizeof(*fpu));
2506 		if (IS_ERR(fpu)) {
2507 			r = PTR_ERR(fpu);
2508 			fpu = NULL;
2509 			goto out;
2510 		}
2511 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2512 		break;
2513 	}
2514 	default:
2515 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2516 	}
2517 out:
2518 	vcpu_put(vcpu);
2519 	kfree(fpu);
2520 	kfree(kvm_sregs);
2521 	return r;
2522 }
2523 
2524 #ifdef CONFIG_KVM_COMPAT
2525 static long kvm_vcpu_compat_ioctl(struct file *filp,
2526 				  unsigned int ioctl, unsigned long arg)
2527 {
2528 	struct kvm_vcpu *vcpu = filp->private_data;
2529 	void __user *argp = compat_ptr(arg);
2530 	int r;
2531 
2532 	if (vcpu->kvm->mm != current->mm)
2533 		return -EIO;
2534 
2535 	switch (ioctl) {
2536 	case KVM_SET_SIGNAL_MASK: {
2537 		struct kvm_signal_mask __user *sigmask_arg = argp;
2538 		struct kvm_signal_mask kvm_sigmask;
2539 		compat_sigset_t csigset;
2540 		sigset_t sigset;
2541 
2542 		if (argp) {
2543 			r = -EFAULT;
2544 			if (copy_from_user(&kvm_sigmask, argp,
2545 					   sizeof(kvm_sigmask)))
2546 				goto out;
2547 			r = -EINVAL;
2548 			if (kvm_sigmask.len != sizeof(csigset))
2549 				goto out;
2550 			r = -EFAULT;
2551 			if (copy_from_user(&csigset, sigmask_arg->sigset,
2552 					   sizeof(csigset)))
2553 				goto out;
2554 			sigset_from_compat(&sigset, &csigset);
2555 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2556 		} else
2557 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2558 		break;
2559 	}
2560 	default:
2561 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
2562 	}
2563 
2564 out:
2565 	return r;
2566 }
2567 #endif
2568 
2569 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2570 				 int (*accessor)(struct kvm_device *dev,
2571 						 struct kvm_device_attr *attr),
2572 				 unsigned long arg)
2573 {
2574 	struct kvm_device_attr attr;
2575 
2576 	if (!accessor)
2577 		return -EPERM;
2578 
2579 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2580 		return -EFAULT;
2581 
2582 	return accessor(dev, &attr);
2583 }
2584 
2585 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2586 			     unsigned long arg)
2587 {
2588 	struct kvm_device *dev = filp->private_data;
2589 
2590 	switch (ioctl) {
2591 	case KVM_SET_DEVICE_ATTR:
2592 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2593 	case KVM_GET_DEVICE_ATTR:
2594 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2595 	case KVM_HAS_DEVICE_ATTR:
2596 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2597 	default:
2598 		if (dev->ops->ioctl)
2599 			return dev->ops->ioctl(dev, ioctl, arg);
2600 
2601 		return -ENOTTY;
2602 	}
2603 }
2604 
2605 static int kvm_device_release(struct inode *inode, struct file *filp)
2606 {
2607 	struct kvm_device *dev = filp->private_data;
2608 	struct kvm *kvm = dev->kvm;
2609 
2610 	kvm_put_kvm(kvm);
2611 	return 0;
2612 }
2613 
2614 static const struct file_operations kvm_device_fops = {
2615 	.unlocked_ioctl = kvm_device_ioctl,
2616 #ifdef CONFIG_KVM_COMPAT
2617 	.compat_ioctl = kvm_device_ioctl,
2618 #endif
2619 	.release = kvm_device_release,
2620 };
2621 
2622 struct kvm_device *kvm_device_from_filp(struct file *filp)
2623 {
2624 	if (filp->f_op != &kvm_device_fops)
2625 		return NULL;
2626 
2627 	return filp->private_data;
2628 }
2629 
2630 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2631 #ifdef CONFIG_KVM_MPIC
2632 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
2633 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
2634 #endif
2635 
2636 #ifdef CONFIG_KVM_XICS
2637 	[KVM_DEV_TYPE_XICS]		= &kvm_xics_ops,
2638 #endif
2639 };
2640 
2641 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2642 {
2643 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
2644 		return -ENOSPC;
2645 
2646 	if (kvm_device_ops_table[type] != NULL)
2647 		return -EEXIST;
2648 
2649 	kvm_device_ops_table[type] = ops;
2650 	return 0;
2651 }
2652 
2653 void kvm_unregister_device_ops(u32 type)
2654 {
2655 	if (kvm_device_ops_table[type] != NULL)
2656 		kvm_device_ops_table[type] = NULL;
2657 }
2658 
2659 static int kvm_ioctl_create_device(struct kvm *kvm,
2660 				   struct kvm_create_device *cd)
2661 {
2662 	struct kvm_device_ops *ops = NULL;
2663 	struct kvm_device *dev;
2664 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2665 	int ret;
2666 
2667 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2668 		return -ENODEV;
2669 
2670 	ops = kvm_device_ops_table[cd->type];
2671 	if (ops == NULL)
2672 		return -ENODEV;
2673 
2674 	if (test)
2675 		return 0;
2676 
2677 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2678 	if (!dev)
2679 		return -ENOMEM;
2680 
2681 	dev->ops = ops;
2682 	dev->kvm = kvm;
2683 
2684 	ret = ops->create(dev, cd->type);
2685 	if (ret < 0) {
2686 		kfree(dev);
2687 		return ret;
2688 	}
2689 
2690 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2691 	if (ret < 0) {
2692 		ops->destroy(dev);
2693 		return ret;
2694 	}
2695 
2696 	list_add(&dev->vm_node, &kvm->devices);
2697 	kvm_get_kvm(kvm);
2698 	cd->fd = ret;
2699 	return 0;
2700 }
2701 
2702 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2703 {
2704 	switch (arg) {
2705 	case KVM_CAP_USER_MEMORY:
2706 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2707 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2708 	case KVM_CAP_INTERNAL_ERROR_DATA:
2709 #ifdef CONFIG_HAVE_KVM_MSI
2710 	case KVM_CAP_SIGNAL_MSI:
2711 #endif
2712 #ifdef CONFIG_HAVE_KVM_IRQFD
2713 	case KVM_CAP_IRQFD:
2714 	case KVM_CAP_IRQFD_RESAMPLE:
2715 #endif
2716 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2717 	case KVM_CAP_CHECK_EXTENSION_VM:
2718 		return 1;
2719 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2720 	case KVM_CAP_IRQ_ROUTING:
2721 		return KVM_MAX_IRQ_ROUTES;
2722 #endif
2723 #if KVM_ADDRESS_SPACE_NUM > 1
2724 	case KVM_CAP_MULTI_ADDRESS_SPACE:
2725 		return KVM_ADDRESS_SPACE_NUM;
2726 #endif
2727 	default:
2728 		break;
2729 	}
2730 	return kvm_vm_ioctl_check_extension(kvm, arg);
2731 }
2732 
2733 static long kvm_vm_ioctl(struct file *filp,
2734 			   unsigned int ioctl, unsigned long arg)
2735 {
2736 	struct kvm *kvm = filp->private_data;
2737 	void __user *argp = (void __user *)arg;
2738 	int r;
2739 
2740 	if (kvm->mm != current->mm)
2741 		return -EIO;
2742 	switch (ioctl) {
2743 	case KVM_CREATE_VCPU:
2744 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2745 		break;
2746 	case KVM_SET_USER_MEMORY_REGION: {
2747 		struct kvm_userspace_memory_region kvm_userspace_mem;
2748 
2749 		r = -EFAULT;
2750 		if (copy_from_user(&kvm_userspace_mem, argp,
2751 						sizeof(kvm_userspace_mem)))
2752 			goto out;
2753 
2754 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2755 		break;
2756 	}
2757 	case KVM_GET_DIRTY_LOG: {
2758 		struct kvm_dirty_log log;
2759 
2760 		r = -EFAULT;
2761 		if (copy_from_user(&log, argp, sizeof(log)))
2762 			goto out;
2763 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2764 		break;
2765 	}
2766 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2767 	case KVM_REGISTER_COALESCED_MMIO: {
2768 		struct kvm_coalesced_mmio_zone zone;
2769 
2770 		r = -EFAULT;
2771 		if (copy_from_user(&zone, argp, sizeof(zone)))
2772 			goto out;
2773 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2774 		break;
2775 	}
2776 	case KVM_UNREGISTER_COALESCED_MMIO: {
2777 		struct kvm_coalesced_mmio_zone zone;
2778 
2779 		r = -EFAULT;
2780 		if (copy_from_user(&zone, argp, sizeof(zone)))
2781 			goto out;
2782 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2783 		break;
2784 	}
2785 #endif
2786 	case KVM_IRQFD: {
2787 		struct kvm_irqfd data;
2788 
2789 		r = -EFAULT;
2790 		if (copy_from_user(&data, argp, sizeof(data)))
2791 			goto out;
2792 		r = kvm_irqfd(kvm, &data);
2793 		break;
2794 	}
2795 	case KVM_IOEVENTFD: {
2796 		struct kvm_ioeventfd data;
2797 
2798 		r = -EFAULT;
2799 		if (copy_from_user(&data, argp, sizeof(data)))
2800 			goto out;
2801 		r = kvm_ioeventfd(kvm, &data);
2802 		break;
2803 	}
2804 #ifdef CONFIG_HAVE_KVM_MSI
2805 	case KVM_SIGNAL_MSI: {
2806 		struct kvm_msi msi;
2807 
2808 		r = -EFAULT;
2809 		if (copy_from_user(&msi, argp, sizeof(msi)))
2810 			goto out;
2811 		r = kvm_send_userspace_msi(kvm, &msi);
2812 		break;
2813 	}
2814 #endif
2815 #ifdef __KVM_HAVE_IRQ_LINE
2816 	case KVM_IRQ_LINE_STATUS:
2817 	case KVM_IRQ_LINE: {
2818 		struct kvm_irq_level irq_event;
2819 
2820 		r = -EFAULT;
2821 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2822 			goto out;
2823 
2824 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2825 					ioctl == KVM_IRQ_LINE_STATUS);
2826 		if (r)
2827 			goto out;
2828 
2829 		r = -EFAULT;
2830 		if (ioctl == KVM_IRQ_LINE_STATUS) {
2831 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2832 				goto out;
2833 		}
2834 
2835 		r = 0;
2836 		break;
2837 	}
2838 #endif
2839 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2840 	case KVM_SET_GSI_ROUTING: {
2841 		struct kvm_irq_routing routing;
2842 		struct kvm_irq_routing __user *urouting;
2843 		struct kvm_irq_routing_entry *entries;
2844 
2845 		r = -EFAULT;
2846 		if (copy_from_user(&routing, argp, sizeof(routing)))
2847 			goto out;
2848 		r = -EINVAL;
2849 		if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2850 			goto out;
2851 		if (routing.flags)
2852 			goto out;
2853 		r = -ENOMEM;
2854 		entries = vmalloc(routing.nr * sizeof(*entries));
2855 		if (!entries)
2856 			goto out;
2857 		r = -EFAULT;
2858 		urouting = argp;
2859 		if (copy_from_user(entries, urouting->entries,
2860 				   routing.nr * sizeof(*entries)))
2861 			goto out_free_irq_routing;
2862 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
2863 					routing.flags);
2864 out_free_irq_routing:
2865 		vfree(entries);
2866 		break;
2867 	}
2868 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2869 	case KVM_CREATE_DEVICE: {
2870 		struct kvm_create_device cd;
2871 
2872 		r = -EFAULT;
2873 		if (copy_from_user(&cd, argp, sizeof(cd)))
2874 			goto out;
2875 
2876 		r = kvm_ioctl_create_device(kvm, &cd);
2877 		if (r)
2878 			goto out;
2879 
2880 		r = -EFAULT;
2881 		if (copy_to_user(argp, &cd, sizeof(cd)))
2882 			goto out;
2883 
2884 		r = 0;
2885 		break;
2886 	}
2887 	case KVM_CHECK_EXTENSION:
2888 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2889 		break;
2890 	default:
2891 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2892 	}
2893 out:
2894 	return r;
2895 }
2896 
2897 #ifdef CONFIG_KVM_COMPAT
2898 struct compat_kvm_dirty_log {
2899 	__u32 slot;
2900 	__u32 padding1;
2901 	union {
2902 		compat_uptr_t dirty_bitmap; /* one bit per page */
2903 		__u64 padding2;
2904 	};
2905 };
2906 
2907 static long kvm_vm_compat_ioctl(struct file *filp,
2908 			   unsigned int ioctl, unsigned long arg)
2909 {
2910 	struct kvm *kvm = filp->private_data;
2911 	int r;
2912 
2913 	if (kvm->mm != current->mm)
2914 		return -EIO;
2915 	switch (ioctl) {
2916 	case KVM_GET_DIRTY_LOG: {
2917 		struct compat_kvm_dirty_log compat_log;
2918 		struct kvm_dirty_log log;
2919 
2920 		r = -EFAULT;
2921 		if (copy_from_user(&compat_log, (void __user *)arg,
2922 				   sizeof(compat_log)))
2923 			goto out;
2924 		log.slot	 = compat_log.slot;
2925 		log.padding1	 = compat_log.padding1;
2926 		log.padding2	 = compat_log.padding2;
2927 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2928 
2929 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2930 		break;
2931 	}
2932 	default:
2933 		r = kvm_vm_ioctl(filp, ioctl, arg);
2934 	}
2935 
2936 out:
2937 	return r;
2938 }
2939 #endif
2940 
2941 static struct file_operations kvm_vm_fops = {
2942 	.release        = kvm_vm_release,
2943 	.unlocked_ioctl = kvm_vm_ioctl,
2944 #ifdef CONFIG_KVM_COMPAT
2945 	.compat_ioctl   = kvm_vm_compat_ioctl,
2946 #endif
2947 	.llseek		= noop_llseek,
2948 };
2949 
2950 static int kvm_dev_ioctl_create_vm(unsigned long type)
2951 {
2952 	int r;
2953 	struct kvm *kvm;
2954 
2955 	kvm = kvm_create_vm(type);
2956 	if (IS_ERR(kvm))
2957 		return PTR_ERR(kvm);
2958 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2959 	r = kvm_coalesced_mmio_init(kvm);
2960 	if (r < 0) {
2961 		kvm_put_kvm(kvm);
2962 		return r;
2963 	}
2964 #endif
2965 	r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2966 	if (r < 0)
2967 		kvm_put_kvm(kvm);
2968 
2969 	return r;
2970 }
2971 
2972 static long kvm_dev_ioctl(struct file *filp,
2973 			  unsigned int ioctl, unsigned long arg)
2974 {
2975 	long r = -EINVAL;
2976 
2977 	switch (ioctl) {
2978 	case KVM_GET_API_VERSION:
2979 		if (arg)
2980 			goto out;
2981 		r = KVM_API_VERSION;
2982 		break;
2983 	case KVM_CREATE_VM:
2984 		r = kvm_dev_ioctl_create_vm(arg);
2985 		break;
2986 	case KVM_CHECK_EXTENSION:
2987 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
2988 		break;
2989 	case KVM_GET_VCPU_MMAP_SIZE:
2990 		if (arg)
2991 			goto out;
2992 		r = PAGE_SIZE;     /* struct kvm_run */
2993 #ifdef CONFIG_X86
2994 		r += PAGE_SIZE;    /* pio data page */
2995 #endif
2996 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2997 		r += PAGE_SIZE;    /* coalesced mmio ring page */
2998 #endif
2999 		break;
3000 	case KVM_TRACE_ENABLE:
3001 	case KVM_TRACE_PAUSE:
3002 	case KVM_TRACE_DISABLE:
3003 		r = -EOPNOTSUPP;
3004 		break;
3005 	default:
3006 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
3007 	}
3008 out:
3009 	return r;
3010 }
3011 
3012 static struct file_operations kvm_chardev_ops = {
3013 	.unlocked_ioctl = kvm_dev_ioctl,
3014 	.compat_ioctl   = kvm_dev_ioctl,
3015 	.llseek		= noop_llseek,
3016 };
3017 
3018 static struct miscdevice kvm_dev = {
3019 	KVM_MINOR,
3020 	"kvm",
3021 	&kvm_chardev_ops,
3022 };
3023 
3024 static void hardware_enable_nolock(void *junk)
3025 {
3026 	int cpu = raw_smp_processor_id();
3027 	int r;
3028 
3029 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3030 		return;
3031 
3032 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
3033 
3034 	r = kvm_arch_hardware_enable();
3035 
3036 	if (r) {
3037 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3038 		atomic_inc(&hardware_enable_failed);
3039 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3040 	}
3041 }
3042 
3043 static void hardware_enable(void)
3044 {
3045 	raw_spin_lock(&kvm_count_lock);
3046 	if (kvm_usage_count)
3047 		hardware_enable_nolock(NULL);
3048 	raw_spin_unlock(&kvm_count_lock);
3049 }
3050 
3051 static void hardware_disable_nolock(void *junk)
3052 {
3053 	int cpu = raw_smp_processor_id();
3054 
3055 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3056 		return;
3057 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3058 	kvm_arch_hardware_disable();
3059 }
3060 
3061 static void hardware_disable(void)
3062 {
3063 	raw_spin_lock(&kvm_count_lock);
3064 	if (kvm_usage_count)
3065 		hardware_disable_nolock(NULL);
3066 	raw_spin_unlock(&kvm_count_lock);
3067 }
3068 
3069 static void hardware_disable_all_nolock(void)
3070 {
3071 	BUG_ON(!kvm_usage_count);
3072 
3073 	kvm_usage_count--;
3074 	if (!kvm_usage_count)
3075 		on_each_cpu(hardware_disable_nolock, NULL, 1);
3076 }
3077 
3078 static void hardware_disable_all(void)
3079 {
3080 	raw_spin_lock(&kvm_count_lock);
3081 	hardware_disable_all_nolock();
3082 	raw_spin_unlock(&kvm_count_lock);
3083 }
3084 
3085 static int hardware_enable_all(void)
3086 {
3087 	int r = 0;
3088 
3089 	raw_spin_lock(&kvm_count_lock);
3090 
3091 	kvm_usage_count++;
3092 	if (kvm_usage_count == 1) {
3093 		atomic_set(&hardware_enable_failed, 0);
3094 		on_each_cpu(hardware_enable_nolock, NULL, 1);
3095 
3096 		if (atomic_read(&hardware_enable_failed)) {
3097 			hardware_disable_all_nolock();
3098 			r = -EBUSY;
3099 		}
3100 	}
3101 
3102 	raw_spin_unlock(&kvm_count_lock);
3103 
3104 	return r;
3105 }
3106 
3107 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3108 			   void *v)
3109 {
3110 	val &= ~CPU_TASKS_FROZEN;
3111 	switch (val) {
3112 	case CPU_DYING:
3113 		hardware_disable();
3114 		break;
3115 	case CPU_STARTING:
3116 		hardware_enable();
3117 		break;
3118 	}
3119 	return NOTIFY_OK;
3120 }
3121 
3122 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3123 		      void *v)
3124 {
3125 	/*
3126 	 * Some (well, at least mine) BIOSes hang on reboot if
3127 	 * in vmx root mode.
3128 	 *
3129 	 * And Intel TXT required VMX off for all cpu when system shutdown.
3130 	 */
3131 	pr_info("kvm: exiting hardware virtualization\n");
3132 	kvm_rebooting = true;
3133 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3134 	return NOTIFY_OK;
3135 }
3136 
3137 static struct notifier_block kvm_reboot_notifier = {
3138 	.notifier_call = kvm_reboot,
3139 	.priority = 0,
3140 };
3141 
3142 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3143 {
3144 	int i;
3145 
3146 	for (i = 0; i < bus->dev_count; i++) {
3147 		struct kvm_io_device *pos = bus->range[i].dev;
3148 
3149 		kvm_iodevice_destructor(pos);
3150 	}
3151 	kfree(bus);
3152 }
3153 
3154 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3155 				 const struct kvm_io_range *r2)
3156 {
3157 	gpa_t addr1 = r1->addr;
3158 	gpa_t addr2 = r2->addr;
3159 
3160 	if (addr1 < addr2)
3161 		return -1;
3162 
3163 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
3164 	 * accept any overlapping write.  Any order is acceptable for
3165 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3166 	 * we process all of them.
3167 	 */
3168 	if (r2->len) {
3169 		addr1 += r1->len;
3170 		addr2 += r2->len;
3171 	}
3172 
3173 	if (addr1 > addr2)
3174 		return 1;
3175 
3176 	return 0;
3177 }
3178 
3179 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3180 {
3181 	return kvm_io_bus_cmp(p1, p2);
3182 }
3183 
3184 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3185 			  gpa_t addr, int len)
3186 {
3187 	bus->range[bus->dev_count++] = (struct kvm_io_range) {
3188 		.addr = addr,
3189 		.len = len,
3190 		.dev = dev,
3191 	};
3192 
3193 	sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3194 		kvm_io_bus_sort_cmp, NULL);
3195 
3196 	return 0;
3197 }
3198 
3199 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3200 			     gpa_t addr, int len)
3201 {
3202 	struct kvm_io_range *range, key;
3203 	int off;
3204 
3205 	key = (struct kvm_io_range) {
3206 		.addr = addr,
3207 		.len = len,
3208 	};
3209 
3210 	range = bsearch(&key, bus->range, bus->dev_count,
3211 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3212 	if (range == NULL)
3213 		return -ENOENT;
3214 
3215 	off = range - bus->range;
3216 
3217 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3218 		off--;
3219 
3220 	return off;
3221 }
3222 
3223 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3224 			      struct kvm_io_range *range, const void *val)
3225 {
3226 	int idx;
3227 
3228 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3229 	if (idx < 0)
3230 		return -EOPNOTSUPP;
3231 
3232 	while (idx < bus->dev_count &&
3233 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3234 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3235 					range->len, val))
3236 			return idx;
3237 		idx++;
3238 	}
3239 
3240 	return -EOPNOTSUPP;
3241 }
3242 
3243 /* kvm_io_bus_write - called under kvm->slots_lock */
3244 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3245 		     int len, const void *val)
3246 {
3247 	struct kvm_io_bus *bus;
3248 	struct kvm_io_range range;
3249 	int r;
3250 
3251 	range = (struct kvm_io_range) {
3252 		.addr = addr,
3253 		.len = len,
3254 	};
3255 
3256 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3257 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
3258 	return r < 0 ? r : 0;
3259 }
3260 
3261 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3262 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3263 			    gpa_t addr, int len, const void *val, long cookie)
3264 {
3265 	struct kvm_io_bus *bus;
3266 	struct kvm_io_range range;
3267 
3268 	range = (struct kvm_io_range) {
3269 		.addr = addr,
3270 		.len = len,
3271 	};
3272 
3273 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3274 
3275 	/* First try the device referenced by cookie. */
3276 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
3277 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3278 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3279 					val))
3280 			return cookie;
3281 
3282 	/*
3283 	 * cookie contained garbage; fall back to search and return the
3284 	 * correct cookie value.
3285 	 */
3286 	return __kvm_io_bus_write(vcpu, bus, &range, val);
3287 }
3288 
3289 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3290 			     struct kvm_io_range *range, void *val)
3291 {
3292 	int idx;
3293 
3294 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3295 	if (idx < 0)
3296 		return -EOPNOTSUPP;
3297 
3298 	while (idx < bus->dev_count &&
3299 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3300 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3301 				       range->len, val))
3302 			return idx;
3303 		idx++;
3304 	}
3305 
3306 	return -EOPNOTSUPP;
3307 }
3308 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3309 
3310 /* kvm_io_bus_read - called under kvm->slots_lock */
3311 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3312 		    int len, void *val)
3313 {
3314 	struct kvm_io_bus *bus;
3315 	struct kvm_io_range range;
3316 	int r;
3317 
3318 	range = (struct kvm_io_range) {
3319 		.addr = addr,
3320 		.len = len,
3321 	};
3322 
3323 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3324 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
3325 	return r < 0 ? r : 0;
3326 }
3327 
3328 
3329 /* Caller must hold slots_lock. */
3330 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3331 			    int len, struct kvm_io_device *dev)
3332 {
3333 	struct kvm_io_bus *new_bus, *bus;
3334 
3335 	bus = kvm->buses[bus_idx];
3336 	/* exclude ioeventfd which is limited by maximum fd */
3337 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3338 		return -ENOSPC;
3339 
3340 	new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3341 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3342 	if (!new_bus)
3343 		return -ENOMEM;
3344 	memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3345 	       sizeof(struct kvm_io_range)));
3346 	kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3347 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3348 	synchronize_srcu_expedited(&kvm->srcu);
3349 	kfree(bus);
3350 
3351 	return 0;
3352 }
3353 
3354 /* Caller must hold slots_lock. */
3355 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3356 			      struct kvm_io_device *dev)
3357 {
3358 	int i, r;
3359 	struct kvm_io_bus *new_bus, *bus;
3360 
3361 	bus = kvm->buses[bus_idx];
3362 	r = -ENOENT;
3363 	for (i = 0; i < bus->dev_count; i++)
3364 		if (bus->range[i].dev == dev) {
3365 			r = 0;
3366 			break;
3367 		}
3368 
3369 	if (r)
3370 		return r;
3371 
3372 	new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3373 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3374 	if (!new_bus)
3375 		return -ENOMEM;
3376 
3377 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3378 	new_bus->dev_count--;
3379 	memcpy(new_bus->range + i, bus->range + i + 1,
3380 	       (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3381 
3382 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3383 	synchronize_srcu_expedited(&kvm->srcu);
3384 	kfree(bus);
3385 	return r;
3386 }
3387 
3388 static struct notifier_block kvm_cpu_notifier = {
3389 	.notifier_call = kvm_cpu_hotplug,
3390 };
3391 
3392 static int vm_stat_get(void *_offset, u64 *val)
3393 {
3394 	unsigned offset = (long)_offset;
3395 	struct kvm *kvm;
3396 
3397 	*val = 0;
3398 	spin_lock(&kvm_lock);
3399 	list_for_each_entry(kvm, &vm_list, vm_list)
3400 		*val += *(u32 *)((void *)kvm + offset);
3401 	spin_unlock(&kvm_lock);
3402 	return 0;
3403 }
3404 
3405 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3406 
3407 static int vcpu_stat_get(void *_offset, u64 *val)
3408 {
3409 	unsigned offset = (long)_offset;
3410 	struct kvm *kvm;
3411 	struct kvm_vcpu *vcpu;
3412 	int i;
3413 
3414 	*val = 0;
3415 	spin_lock(&kvm_lock);
3416 	list_for_each_entry(kvm, &vm_list, vm_list)
3417 		kvm_for_each_vcpu(i, vcpu, kvm)
3418 			*val += *(u32 *)((void *)vcpu + offset);
3419 
3420 	spin_unlock(&kvm_lock);
3421 	return 0;
3422 }
3423 
3424 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3425 
3426 static const struct file_operations *stat_fops[] = {
3427 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
3428 	[KVM_STAT_VM]   = &vm_stat_fops,
3429 };
3430 
3431 static int kvm_init_debug(void)
3432 {
3433 	int r = -EEXIST;
3434 	struct kvm_stats_debugfs_item *p;
3435 
3436 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3437 	if (kvm_debugfs_dir == NULL)
3438 		goto out;
3439 
3440 	for (p = debugfs_entries; p->name; ++p) {
3441 		if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3442 					 (void *)(long)p->offset,
3443 					 stat_fops[p->kind]))
3444 			goto out_dir;
3445 	}
3446 
3447 	return 0;
3448 
3449 out_dir:
3450 	debugfs_remove_recursive(kvm_debugfs_dir);
3451 out:
3452 	return r;
3453 }
3454 
3455 static int kvm_suspend(void)
3456 {
3457 	if (kvm_usage_count)
3458 		hardware_disable_nolock(NULL);
3459 	return 0;
3460 }
3461 
3462 static void kvm_resume(void)
3463 {
3464 	if (kvm_usage_count) {
3465 		WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3466 		hardware_enable_nolock(NULL);
3467 	}
3468 }
3469 
3470 static struct syscore_ops kvm_syscore_ops = {
3471 	.suspend = kvm_suspend,
3472 	.resume = kvm_resume,
3473 };
3474 
3475 static inline
3476 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3477 {
3478 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
3479 }
3480 
3481 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3482 {
3483 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3484 
3485 	if (vcpu->preempted)
3486 		vcpu->preempted = false;
3487 
3488 	kvm_arch_sched_in(vcpu, cpu);
3489 
3490 	kvm_arch_vcpu_load(vcpu, cpu);
3491 }
3492 
3493 static void kvm_sched_out(struct preempt_notifier *pn,
3494 			  struct task_struct *next)
3495 {
3496 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3497 
3498 	if (current->state == TASK_RUNNING)
3499 		vcpu->preempted = true;
3500 	kvm_arch_vcpu_put(vcpu);
3501 }
3502 
3503 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3504 		  struct module *module)
3505 {
3506 	int r;
3507 	int cpu;
3508 
3509 	r = kvm_arch_init(opaque);
3510 	if (r)
3511 		goto out_fail;
3512 
3513 	/*
3514 	 * kvm_arch_init makes sure there's at most one caller
3515 	 * for architectures that support multiple implementations,
3516 	 * like intel and amd on x86.
3517 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3518 	 * conflicts in case kvm is already setup for another implementation.
3519 	 */
3520 	r = kvm_irqfd_init();
3521 	if (r)
3522 		goto out_irqfd;
3523 
3524 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3525 		r = -ENOMEM;
3526 		goto out_free_0;
3527 	}
3528 
3529 	r = kvm_arch_hardware_setup();
3530 	if (r < 0)
3531 		goto out_free_0a;
3532 
3533 	for_each_online_cpu(cpu) {
3534 		smp_call_function_single(cpu,
3535 				kvm_arch_check_processor_compat,
3536 				&r, 1);
3537 		if (r < 0)
3538 			goto out_free_1;
3539 	}
3540 
3541 	r = register_cpu_notifier(&kvm_cpu_notifier);
3542 	if (r)
3543 		goto out_free_2;
3544 	register_reboot_notifier(&kvm_reboot_notifier);
3545 
3546 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
3547 	if (!vcpu_align)
3548 		vcpu_align = __alignof__(struct kvm_vcpu);
3549 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3550 					   0, NULL);
3551 	if (!kvm_vcpu_cache) {
3552 		r = -ENOMEM;
3553 		goto out_free_3;
3554 	}
3555 
3556 	r = kvm_async_pf_init();
3557 	if (r)
3558 		goto out_free;
3559 
3560 	kvm_chardev_ops.owner = module;
3561 	kvm_vm_fops.owner = module;
3562 	kvm_vcpu_fops.owner = module;
3563 
3564 	r = misc_register(&kvm_dev);
3565 	if (r) {
3566 		pr_err("kvm: misc device register failed\n");
3567 		goto out_unreg;
3568 	}
3569 
3570 	register_syscore_ops(&kvm_syscore_ops);
3571 
3572 	kvm_preempt_ops.sched_in = kvm_sched_in;
3573 	kvm_preempt_ops.sched_out = kvm_sched_out;
3574 
3575 	r = kvm_init_debug();
3576 	if (r) {
3577 		pr_err("kvm: create debugfs files failed\n");
3578 		goto out_undebugfs;
3579 	}
3580 
3581 	r = kvm_vfio_ops_init();
3582 	WARN_ON(r);
3583 
3584 	return 0;
3585 
3586 out_undebugfs:
3587 	unregister_syscore_ops(&kvm_syscore_ops);
3588 	misc_deregister(&kvm_dev);
3589 out_unreg:
3590 	kvm_async_pf_deinit();
3591 out_free:
3592 	kmem_cache_destroy(kvm_vcpu_cache);
3593 out_free_3:
3594 	unregister_reboot_notifier(&kvm_reboot_notifier);
3595 	unregister_cpu_notifier(&kvm_cpu_notifier);
3596 out_free_2:
3597 out_free_1:
3598 	kvm_arch_hardware_unsetup();
3599 out_free_0a:
3600 	free_cpumask_var(cpus_hardware_enabled);
3601 out_free_0:
3602 	kvm_irqfd_exit();
3603 out_irqfd:
3604 	kvm_arch_exit();
3605 out_fail:
3606 	return r;
3607 }
3608 EXPORT_SYMBOL_GPL(kvm_init);
3609 
3610 void kvm_exit(void)
3611 {
3612 	debugfs_remove_recursive(kvm_debugfs_dir);
3613 	misc_deregister(&kvm_dev);
3614 	kmem_cache_destroy(kvm_vcpu_cache);
3615 	kvm_async_pf_deinit();
3616 	unregister_syscore_ops(&kvm_syscore_ops);
3617 	unregister_reboot_notifier(&kvm_reboot_notifier);
3618 	unregister_cpu_notifier(&kvm_cpu_notifier);
3619 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3620 	kvm_arch_hardware_unsetup();
3621 	kvm_arch_exit();
3622 	kvm_irqfd_exit();
3623 	free_cpumask_var(cpus_hardware_enabled);
3624 	kvm_vfio_ops_exit();
3625 }
3626 EXPORT_SYMBOL_GPL(kvm_exit);
3627