xref: /openbmc/linux/virt/kvm/kvm_main.c (revision 4da722ca)
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/signal.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/stat.h>
38 #include <linux/cpumask.h>
39 #include <linux/smp.h>
40 #include <linux/anon_inodes.h>
41 #include <linux/profile.h>
42 #include <linux/kvm_para.h>
43 #include <linux/pagemap.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/bitops.h>
47 #include <linux/spinlock.h>
48 #include <linux/compat.h>
49 #include <linux/srcu.h>
50 #include <linux/hugetlb.h>
51 #include <linux/slab.h>
52 #include <linux/sort.h>
53 #include <linux/bsearch.h>
54 
55 #include <asm/processor.h>
56 #include <asm/io.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 #include <asm/pgtable.h>
60 
61 #include "coalesced_mmio.h"
62 #include "async_pf.h"
63 #include "vfio.h"
64 
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
67 
68 /* Worst case buffer size needed for holding an integer. */
69 #define ITOA_MAX_LEN 12
70 
71 MODULE_AUTHOR("Qumranet");
72 MODULE_LICENSE("GPL");
73 
74 /* Architectures should define their poll value according to the halt latency */
75 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
76 module_param(halt_poll_ns, uint, 0644);
77 EXPORT_SYMBOL_GPL(halt_poll_ns);
78 
79 /* Default doubles per-vcpu halt_poll_ns. */
80 unsigned int halt_poll_ns_grow = 2;
81 module_param(halt_poll_ns_grow, uint, 0644);
82 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
83 
84 /* Default resets per-vcpu halt_poll_ns . */
85 unsigned int halt_poll_ns_shrink;
86 module_param(halt_poll_ns_shrink, uint, 0644);
87 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
88 
89 /*
90  * Ordering of locks:
91  *
92  *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
93  */
94 
95 DEFINE_SPINLOCK(kvm_lock);
96 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
97 LIST_HEAD(vm_list);
98 
99 static cpumask_var_t cpus_hardware_enabled;
100 static int kvm_usage_count;
101 static atomic_t hardware_enable_failed;
102 
103 struct kmem_cache *kvm_vcpu_cache;
104 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
105 
106 static __read_mostly struct preempt_ops kvm_preempt_ops;
107 
108 struct dentry *kvm_debugfs_dir;
109 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
110 
111 static int kvm_debugfs_num_entries;
112 static const struct file_operations *stat_fops_per_vm[];
113 
114 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
115 			   unsigned long arg);
116 #ifdef CONFIG_KVM_COMPAT
117 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
118 				  unsigned long arg);
119 #endif
120 static int hardware_enable_all(void);
121 static void hardware_disable_all(void);
122 
123 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
124 
125 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
126 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
127 
128 __visible bool kvm_rebooting;
129 EXPORT_SYMBOL_GPL(kvm_rebooting);
130 
131 static bool largepages_enabled = true;
132 
133 #define KVM_EVENT_CREATE_VM 0
134 #define KVM_EVENT_DESTROY_VM 1
135 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
136 static unsigned long long kvm_createvm_count;
137 static unsigned long long kvm_active_vms;
138 
139 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
140 {
141 	if (pfn_valid(pfn))
142 		return PageReserved(pfn_to_page(pfn));
143 
144 	return true;
145 }
146 
147 /*
148  * Switches to specified vcpu, until a matching vcpu_put()
149  */
150 int vcpu_load(struct kvm_vcpu *vcpu)
151 {
152 	int cpu;
153 
154 	if (mutex_lock_killable(&vcpu->mutex))
155 		return -EINTR;
156 	cpu = get_cpu();
157 	preempt_notifier_register(&vcpu->preempt_notifier);
158 	kvm_arch_vcpu_load(vcpu, cpu);
159 	put_cpu();
160 	return 0;
161 }
162 EXPORT_SYMBOL_GPL(vcpu_load);
163 
164 void vcpu_put(struct kvm_vcpu *vcpu)
165 {
166 	preempt_disable();
167 	kvm_arch_vcpu_put(vcpu);
168 	preempt_notifier_unregister(&vcpu->preempt_notifier);
169 	preempt_enable();
170 	mutex_unlock(&vcpu->mutex);
171 }
172 EXPORT_SYMBOL_GPL(vcpu_put);
173 
174 /* TODO: merge with kvm_arch_vcpu_should_kick */
175 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
176 {
177 	int mode = kvm_vcpu_exiting_guest_mode(vcpu);
178 
179 	/*
180 	 * We need to wait for the VCPU to reenable interrupts and get out of
181 	 * READING_SHADOW_PAGE_TABLES mode.
182 	 */
183 	if (req & KVM_REQUEST_WAIT)
184 		return mode != OUTSIDE_GUEST_MODE;
185 
186 	/*
187 	 * Need to kick a running VCPU, but otherwise there is nothing to do.
188 	 */
189 	return mode == IN_GUEST_MODE;
190 }
191 
192 static void ack_flush(void *_completed)
193 {
194 }
195 
196 static inline bool kvm_kick_many_cpus(const struct cpumask *cpus, bool wait)
197 {
198 	if (unlikely(!cpus))
199 		cpus = cpu_online_mask;
200 
201 	if (cpumask_empty(cpus))
202 		return false;
203 
204 	smp_call_function_many(cpus, ack_flush, NULL, wait);
205 	return true;
206 }
207 
208 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
209 {
210 	int i, cpu, me;
211 	cpumask_var_t cpus;
212 	bool called;
213 	struct kvm_vcpu *vcpu;
214 
215 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
216 
217 	me = get_cpu();
218 	kvm_for_each_vcpu(i, vcpu, kvm) {
219 		kvm_make_request(req, vcpu);
220 		cpu = vcpu->cpu;
221 
222 		if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
223 			continue;
224 
225 		if (cpus != NULL && cpu != -1 && cpu != me &&
226 		    kvm_request_needs_ipi(vcpu, req))
227 			__cpumask_set_cpu(cpu, cpus);
228 	}
229 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
230 	put_cpu();
231 	free_cpumask_var(cpus);
232 	return called;
233 }
234 
235 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
236 void kvm_flush_remote_tlbs(struct kvm *kvm)
237 {
238 	/*
239 	 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
240 	 * kvm_make_all_cpus_request.
241 	 */
242 	long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
243 
244 	/*
245 	 * We want to publish modifications to the page tables before reading
246 	 * mode. Pairs with a memory barrier in arch-specific code.
247 	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
248 	 * and smp_mb in walk_shadow_page_lockless_begin/end.
249 	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
250 	 *
251 	 * There is already an smp_mb__after_atomic() before
252 	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
253 	 * barrier here.
254 	 */
255 	if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
256 		++kvm->stat.remote_tlb_flush;
257 	cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
258 }
259 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
260 #endif
261 
262 void kvm_reload_remote_mmus(struct kvm *kvm)
263 {
264 	kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
265 }
266 
267 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
268 {
269 	struct page *page;
270 	int r;
271 
272 	mutex_init(&vcpu->mutex);
273 	vcpu->cpu = -1;
274 	vcpu->kvm = kvm;
275 	vcpu->vcpu_id = id;
276 	vcpu->pid = NULL;
277 	init_swait_queue_head(&vcpu->wq);
278 	kvm_async_pf_vcpu_init(vcpu);
279 
280 	vcpu->pre_pcpu = -1;
281 	INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
282 
283 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
284 	if (!page) {
285 		r = -ENOMEM;
286 		goto fail;
287 	}
288 	vcpu->run = page_address(page);
289 
290 	kvm_vcpu_set_in_spin_loop(vcpu, false);
291 	kvm_vcpu_set_dy_eligible(vcpu, false);
292 	vcpu->preempted = false;
293 
294 	r = kvm_arch_vcpu_init(vcpu);
295 	if (r < 0)
296 		goto fail_free_run;
297 	return 0;
298 
299 fail_free_run:
300 	free_page((unsigned long)vcpu->run);
301 fail:
302 	return r;
303 }
304 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
305 
306 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
307 {
308 	/*
309 	 * no need for rcu_read_lock as VCPU_RUN is the only place that
310 	 * will change the vcpu->pid pointer and on uninit all file
311 	 * descriptors are already gone.
312 	 */
313 	put_pid(rcu_dereference_protected(vcpu->pid, 1));
314 	kvm_arch_vcpu_uninit(vcpu);
315 	free_page((unsigned long)vcpu->run);
316 }
317 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
318 
319 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
320 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
321 {
322 	return container_of(mn, struct kvm, mmu_notifier);
323 }
324 
325 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
326 					     struct mm_struct *mm,
327 					     unsigned long address)
328 {
329 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
330 	int need_tlb_flush, idx;
331 
332 	/*
333 	 * When ->invalidate_page runs, the linux pte has been zapped
334 	 * already but the page is still allocated until
335 	 * ->invalidate_page returns. So if we increase the sequence
336 	 * here the kvm page fault will notice if the spte can't be
337 	 * established because the page is going to be freed. If
338 	 * instead the kvm page fault establishes the spte before
339 	 * ->invalidate_page runs, kvm_unmap_hva will release it
340 	 * before returning.
341 	 *
342 	 * The sequence increase only need to be seen at spin_unlock
343 	 * time, and not at spin_lock time.
344 	 *
345 	 * Increasing the sequence after the spin_unlock would be
346 	 * unsafe because the kvm page fault could then establish the
347 	 * pte after kvm_unmap_hva returned, without noticing the page
348 	 * is going to be freed.
349 	 */
350 	idx = srcu_read_lock(&kvm->srcu);
351 	spin_lock(&kvm->mmu_lock);
352 
353 	kvm->mmu_notifier_seq++;
354 	need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
355 	/* we've to flush the tlb before the pages can be freed */
356 	if (need_tlb_flush)
357 		kvm_flush_remote_tlbs(kvm);
358 
359 	spin_unlock(&kvm->mmu_lock);
360 
361 	kvm_arch_mmu_notifier_invalidate_page(kvm, address);
362 
363 	srcu_read_unlock(&kvm->srcu, idx);
364 }
365 
366 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
367 					struct mm_struct *mm,
368 					unsigned long address,
369 					pte_t pte)
370 {
371 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
372 	int idx;
373 
374 	idx = srcu_read_lock(&kvm->srcu);
375 	spin_lock(&kvm->mmu_lock);
376 	kvm->mmu_notifier_seq++;
377 	kvm_set_spte_hva(kvm, address, pte);
378 	spin_unlock(&kvm->mmu_lock);
379 	srcu_read_unlock(&kvm->srcu, idx);
380 }
381 
382 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
383 						    struct mm_struct *mm,
384 						    unsigned long start,
385 						    unsigned long end)
386 {
387 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
388 	int need_tlb_flush = 0, idx;
389 
390 	idx = srcu_read_lock(&kvm->srcu);
391 	spin_lock(&kvm->mmu_lock);
392 	/*
393 	 * The count increase must become visible at unlock time as no
394 	 * spte can be established without taking the mmu_lock and
395 	 * count is also read inside the mmu_lock critical section.
396 	 */
397 	kvm->mmu_notifier_count++;
398 	need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
399 	need_tlb_flush |= kvm->tlbs_dirty;
400 	/* we've to flush the tlb before the pages can be freed */
401 	if (need_tlb_flush)
402 		kvm_flush_remote_tlbs(kvm);
403 
404 	spin_unlock(&kvm->mmu_lock);
405 	srcu_read_unlock(&kvm->srcu, idx);
406 }
407 
408 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
409 						  struct mm_struct *mm,
410 						  unsigned long start,
411 						  unsigned long end)
412 {
413 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
414 
415 	spin_lock(&kvm->mmu_lock);
416 	/*
417 	 * This sequence increase will notify the kvm page fault that
418 	 * the page that is going to be mapped in the spte could have
419 	 * been freed.
420 	 */
421 	kvm->mmu_notifier_seq++;
422 	smp_wmb();
423 	/*
424 	 * The above sequence increase must be visible before the
425 	 * below count decrease, which is ensured by the smp_wmb above
426 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
427 	 */
428 	kvm->mmu_notifier_count--;
429 	spin_unlock(&kvm->mmu_lock);
430 
431 	BUG_ON(kvm->mmu_notifier_count < 0);
432 }
433 
434 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
435 					      struct mm_struct *mm,
436 					      unsigned long start,
437 					      unsigned long end)
438 {
439 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
440 	int young, idx;
441 
442 	idx = srcu_read_lock(&kvm->srcu);
443 	spin_lock(&kvm->mmu_lock);
444 
445 	young = kvm_age_hva(kvm, start, end);
446 	if (young)
447 		kvm_flush_remote_tlbs(kvm);
448 
449 	spin_unlock(&kvm->mmu_lock);
450 	srcu_read_unlock(&kvm->srcu, idx);
451 
452 	return young;
453 }
454 
455 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
456 					struct mm_struct *mm,
457 					unsigned long start,
458 					unsigned long end)
459 {
460 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
461 	int young, idx;
462 
463 	idx = srcu_read_lock(&kvm->srcu);
464 	spin_lock(&kvm->mmu_lock);
465 	/*
466 	 * Even though we do not flush TLB, this will still adversely
467 	 * affect performance on pre-Haswell Intel EPT, where there is
468 	 * no EPT Access Bit to clear so that we have to tear down EPT
469 	 * tables instead. If we find this unacceptable, we can always
470 	 * add a parameter to kvm_age_hva so that it effectively doesn't
471 	 * do anything on clear_young.
472 	 *
473 	 * Also note that currently we never issue secondary TLB flushes
474 	 * from clear_young, leaving this job up to the regular system
475 	 * cadence. If we find this inaccurate, we might come up with a
476 	 * more sophisticated heuristic later.
477 	 */
478 	young = kvm_age_hva(kvm, start, end);
479 	spin_unlock(&kvm->mmu_lock);
480 	srcu_read_unlock(&kvm->srcu, idx);
481 
482 	return young;
483 }
484 
485 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
486 				       struct mm_struct *mm,
487 				       unsigned long address)
488 {
489 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
490 	int young, idx;
491 
492 	idx = srcu_read_lock(&kvm->srcu);
493 	spin_lock(&kvm->mmu_lock);
494 	young = kvm_test_age_hva(kvm, address);
495 	spin_unlock(&kvm->mmu_lock);
496 	srcu_read_unlock(&kvm->srcu, idx);
497 
498 	return young;
499 }
500 
501 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
502 				     struct mm_struct *mm)
503 {
504 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
505 	int idx;
506 
507 	idx = srcu_read_lock(&kvm->srcu);
508 	kvm_arch_flush_shadow_all(kvm);
509 	srcu_read_unlock(&kvm->srcu, idx);
510 }
511 
512 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
513 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
514 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
515 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
516 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
517 	.clear_young		= kvm_mmu_notifier_clear_young,
518 	.test_young		= kvm_mmu_notifier_test_young,
519 	.change_pte		= kvm_mmu_notifier_change_pte,
520 	.release		= kvm_mmu_notifier_release,
521 };
522 
523 static int kvm_init_mmu_notifier(struct kvm *kvm)
524 {
525 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
526 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
527 }
528 
529 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
530 
531 static int kvm_init_mmu_notifier(struct kvm *kvm)
532 {
533 	return 0;
534 }
535 
536 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
537 
538 static struct kvm_memslots *kvm_alloc_memslots(void)
539 {
540 	int i;
541 	struct kvm_memslots *slots;
542 
543 	slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
544 	if (!slots)
545 		return NULL;
546 
547 	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
548 		slots->id_to_index[i] = slots->memslots[i].id = i;
549 
550 	return slots;
551 }
552 
553 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
554 {
555 	if (!memslot->dirty_bitmap)
556 		return;
557 
558 	kvfree(memslot->dirty_bitmap);
559 	memslot->dirty_bitmap = NULL;
560 }
561 
562 /*
563  * Free any memory in @free but not in @dont.
564  */
565 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
566 			      struct kvm_memory_slot *dont)
567 {
568 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
569 		kvm_destroy_dirty_bitmap(free);
570 
571 	kvm_arch_free_memslot(kvm, free, dont);
572 
573 	free->npages = 0;
574 }
575 
576 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
577 {
578 	struct kvm_memory_slot *memslot;
579 
580 	if (!slots)
581 		return;
582 
583 	kvm_for_each_memslot(memslot, slots)
584 		kvm_free_memslot(kvm, memslot, NULL);
585 
586 	kvfree(slots);
587 }
588 
589 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
590 {
591 	int i;
592 
593 	if (!kvm->debugfs_dentry)
594 		return;
595 
596 	debugfs_remove_recursive(kvm->debugfs_dentry);
597 
598 	if (kvm->debugfs_stat_data) {
599 		for (i = 0; i < kvm_debugfs_num_entries; i++)
600 			kfree(kvm->debugfs_stat_data[i]);
601 		kfree(kvm->debugfs_stat_data);
602 	}
603 }
604 
605 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
606 {
607 	char dir_name[ITOA_MAX_LEN * 2];
608 	struct kvm_stat_data *stat_data;
609 	struct kvm_stats_debugfs_item *p;
610 
611 	if (!debugfs_initialized())
612 		return 0;
613 
614 	snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
615 	kvm->debugfs_dentry = debugfs_create_dir(dir_name,
616 						 kvm_debugfs_dir);
617 	if (!kvm->debugfs_dentry)
618 		return -ENOMEM;
619 
620 	kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
621 					 sizeof(*kvm->debugfs_stat_data),
622 					 GFP_KERNEL);
623 	if (!kvm->debugfs_stat_data)
624 		return -ENOMEM;
625 
626 	for (p = debugfs_entries; p->name; p++) {
627 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL);
628 		if (!stat_data)
629 			return -ENOMEM;
630 
631 		stat_data->kvm = kvm;
632 		stat_data->offset = p->offset;
633 		kvm->debugfs_stat_data[p - debugfs_entries] = stat_data;
634 		if (!debugfs_create_file(p->name, 0644,
635 					 kvm->debugfs_dentry,
636 					 stat_data,
637 					 stat_fops_per_vm[p->kind]))
638 			return -ENOMEM;
639 	}
640 	return 0;
641 }
642 
643 static struct kvm *kvm_create_vm(unsigned long type)
644 {
645 	int r, i;
646 	struct kvm *kvm = kvm_arch_alloc_vm();
647 
648 	if (!kvm)
649 		return ERR_PTR(-ENOMEM);
650 
651 	spin_lock_init(&kvm->mmu_lock);
652 	mmgrab(current->mm);
653 	kvm->mm = current->mm;
654 	kvm_eventfd_init(kvm);
655 	mutex_init(&kvm->lock);
656 	mutex_init(&kvm->irq_lock);
657 	mutex_init(&kvm->slots_lock);
658 	refcount_set(&kvm->users_count, 1);
659 	INIT_LIST_HEAD(&kvm->devices);
660 
661 	r = kvm_arch_init_vm(kvm, type);
662 	if (r)
663 		goto out_err_no_disable;
664 
665 	r = hardware_enable_all();
666 	if (r)
667 		goto out_err_no_disable;
668 
669 #ifdef CONFIG_HAVE_KVM_IRQFD
670 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
671 #endif
672 
673 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
674 
675 	r = -ENOMEM;
676 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
677 		struct kvm_memslots *slots = kvm_alloc_memslots();
678 		if (!slots)
679 			goto out_err_no_srcu;
680 		/*
681 		 * Generations must be different for each address space.
682 		 * Init kvm generation close to the maximum to easily test the
683 		 * code of handling generation number wrap-around.
684 		 */
685 		slots->generation = i * 2 - 150;
686 		rcu_assign_pointer(kvm->memslots[i], slots);
687 	}
688 
689 	if (init_srcu_struct(&kvm->srcu))
690 		goto out_err_no_srcu;
691 	if (init_srcu_struct(&kvm->irq_srcu))
692 		goto out_err_no_irq_srcu;
693 	for (i = 0; i < KVM_NR_BUSES; i++) {
694 		rcu_assign_pointer(kvm->buses[i],
695 			kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL));
696 		if (!kvm->buses[i])
697 			goto out_err;
698 	}
699 
700 	r = kvm_init_mmu_notifier(kvm);
701 	if (r)
702 		goto out_err;
703 
704 	spin_lock(&kvm_lock);
705 	list_add(&kvm->vm_list, &vm_list);
706 	spin_unlock(&kvm_lock);
707 
708 	preempt_notifier_inc();
709 
710 	return kvm;
711 
712 out_err:
713 	cleanup_srcu_struct(&kvm->irq_srcu);
714 out_err_no_irq_srcu:
715 	cleanup_srcu_struct(&kvm->srcu);
716 out_err_no_srcu:
717 	hardware_disable_all();
718 out_err_no_disable:
719 	for (i = 0; i < KVM_NR_BUSES; i++)
720 		kfree(rcu_access_pointer(kvm->buses[i]));
721 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
722 		kvm_free_memslots(kvm,
723 			rcu_dereference_protected(kvm->memslots[i], 1));
724 	kvm_arch_free_vm(kvm);
725 	mmdrop(current->mm);
726 	return ERR_PTR(r);
727 }
728 
729 static void kvm_destroy_devices(struct kvm *kvm)
730 {
731 	struct kvm_device *dev, *tmp;
732 
733 	/*
734 	 * We do not need to take the kvm->lock here, because nobody else
735 	 * has a reference to the struct kvm at this point and therefore
736 	 * cannot access the devices list anyhow.
737 	 */
738 	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
739 		list_del(&dev->vm_node);
740 		dev->ops->destroy(dev);
741 	}
742 }
743 
744 static void kvm_destroy_vm(struct kvm *kvm)
745 {
746 	int i;
747 	struct mm_struct *mm = kvm->mm;
748 
749 	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
750 	kvm_destroy_vm_debugfs(kvm);
751 	kvm_arch_sync_events(kvm);
752 	spin_lock(&kvm_lock);
753 	list_del(&kvm->vm_list);
754 	spin_unlock(&kvm_lock);
755 	kvm_free_irq_routing(kvm);
756 	for (i = 0; i < KVM_NR_BUSES; i++) {
757 		struct kvm_io_bus *bus;
758 
759 		bus = rcu_dereference_protected(kvm->buses[i], 1);
760 		if (bus)
761 			kvm_io_bus_destroy(bus);
762 		kvm->buses[i] = NULL;
763 	}
764 	kvm_coalesced_mmio_free(kvm);
765 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
766 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
767 #else
768 	kvm_arch_flush_shadow_all(kvm);
769 #endif
770 	kvm_arch_destroy_vm(kvm);
771 	kvm_destroy_devices(kvm);
772 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
773 		kvm_free_memslots(kvm,
774 			rcu_dereference_protected(kvm->memslots[i], 1));
775 	cleanup_srcu_struct(&kvm->irq_srcu);
776 	cleanup_srcu_struct(&kvm->srcu);
777 	kvm_arch_free_vm(kvm);
778 	preempt_notifier_dec();
779 	hardware_disable_all();
780 	mmdrop(mm);
781 }
782 
783 void kvm_get_kvm(struct kvm *kvm)
784 {
785 	refcount_inc(&kvm->users_count);
786 }
787 EXPORT_SYMBOL_GPL(kvm_get_kvm);
788 
789 void kvm_put_kvm(struct kvm *kvm)
790 {
791 	if (refcount_dec_and_test(&kvm->users_count))
792 		kvm_destroy_vm(kvm);
793 }
794 EXPORT_SYMBOL_GPL(kvm_put_kvm);
795 
796 
797 static int kvm_vm_release(struct inode *inode, struct file *filp)
798 {
799 	struct kvm *kvm = filp->private_data;
800 
801 	kvm_irqfd_release(kvm);
802 
803 	kvm_put_kvm(kvm);
804 	return 0;
805 }
806 
807 /*
808  * Allocation size is twice as large as the actual dirty bitmap size.
809  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
810  */
811 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
812 {
813 	unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
814 
815 	memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL);
816 	if (!memslot->dirty_bitmap)
817 		return -ENOMEM;
818 
819 	return 0;
820 }
821 
822 /*
823  * Insert memslot and re-sort memslots based on their GFN,
824  * so binary search could be used to lookup GFN.
825  * Sorting algorithm takes advantage of having initially
826  * sorted array and known changed memslot position.
827  */
828 static void update_memslots(struct kvm_memslots *slots,
829 			    struct kvm_memory_slot *new)
830 {
831 	int id = new->id;
832 	int i = slots->id_to_index[id];
833 	struct kvm_memory_slot *mslots = slots->memslots;
834 
835 	WARN_ON(mslots[i].id != id);
836 	if (!new->npages) {
837 		WARN_ON(!mslots[i].npages);
838 		if (mslots[i].npages)
839 			slots->used_slots--;
840 	} else {
841 		if (!mslots[i].npages)
842 			slots->used_slots++;
843 	}
844 
845 	while (i < KVM_MEM_SLOTS_NUM - 1 &&
846 	       new->base_gfn <= mslots[i + 1].base_gfn) {
847 		if (!mslots[i + 1].npages)
848 			break;
849 		mslots[i] = mslots[i + 1];
850 		slots->id_to_index[mslots[i].id] = i;
851 		i++;
852 	}
853 
854 	/*
855 	 * The ">=" is needed when creating a slot with base_gfn == 0,
856 	 * so that it moves before all those with base_gfn == npages == 0.
857 	 *
858 	 * On the other hand, if new->npages is zero, the above loop has
859 	 * already left i pointing to the beginning of the empty part of
860 	 * mslots, and the ">=" would move the hole backwards in this
861 	 * case---which is wrong.  So skip the loop when deleting a slot.
862 	 */
863 	if (new->npages) {
864 		while (i > 0 &&
865 		       new->base_gfn >= mslots[i - 1].base_gfn) {
866 			mslots[i] = mslots[i - 1];
867 			slots->id_to_index[mslots[i].id] = i;
868 			i--;
869 		}
870 	} else
871 		WARN_ON_ONCE(i != slots->used_slots);
872 
873 	mslots[i] = *new;
874 	slots->id_to_index[mslots[i].id] = i;
875 }
876 
877 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
878 {
879 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
880 
881 #ifdef __KVM_HAVE_READONLY_MEM
882 	valid_flags |= KVM_MEM_READONLY;
883 #endif
884 
885 	if (mem->flags & ~valid_flags)
886 		return -EINVAL;
887 
888 	return 0;
889 }
890 
891 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
892 		int as_id, struct kvm_memslots *slots)
893 {
894 	struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
895 
896 	/*
897 	 * Set the low bit in the generation, which disables SPTE caching
898 	 * until the end of synchronize_srcu_expedited.
899 	 */
900 	WARN_ON(old_memslots->generation & 1);
901 	slots->generation = old_memslots->generation + 1;
902 
903 	rcu_assign_pointer(kvm->memslots[as_id], slots);
904 	synchronize_srcu_expedited(&kvm->srcu);
905 
906 	/*
907 	 * Increment the new memslot generation a second time. This prevents
908 	 * vm exits that race with memslot updates from caching a memslot
909 	 * generation that will (potentially) be valid forever.
910 	 *
911 	 * Generations must be unique even across address spaces.  We do not need
912 	 * a global counter for that, instead the generation space is evenly split
913 	 * across address spaces.  For example, with two address spaces, address
914 	 * space 0 will use generations 0, 4, 8, ... while * address space 1 will
915 	 * use generations 2, 6, 10, 14, ...
916 	 */
917 	slots->generation += KVM_ADDRESS_SPACE_NUM * 2 - 1;
918 
919 	kvm_arch_memslots_updated(kvm, slots);
920 
921 	return old_memslots;
922 }
923 
924 /*
925  * Allocate some memory and give it an address in the guest physical address
926  * space.
927  *
928  * Discontiguous memory is allowed, mostly for framebuffers.
929  *
930  * Must be called holding kvm->slots_lock for write.
931  */
932 int __kvm_set_memory_region(struct kvm *kvm,
933 			    const struct kvm_userspace_memory_region *mem)
934 {
935 	int r;
936 	gfn_t base_gfn;
937 	unsigned long npages;
938 	struct kvm_memory_slot *slot;
939 	struct kvm_memory_slot old, new;
940 	struct kvm_memslots *slots = NULL, *old_memslots;
941 	int as_id, id;
942 	enum kvm_mr_change change;
943 
944 	r = check_memory_region_flags(mem);
945 	if (r)
946 		goto out;
947 
948 	r = -EINVAL;
949 	as_id = mem->slot >> 16;
950 	id = (u16)mem->slot;
951 
952 	/* General sanity checks */
953 	if (mem->memory_size & (PAGE_SIZE - 1))
954 		goto out;
955 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
956 		goto out;
957 	/* We can read the guest memory with __xxx_user() later on. */
958 	if ((id < KVM_USER_MEM_SLOTS) &&
959 	    ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
960 	     !access_ok(VERIFY_WRITE,
961 			(void __user *)(unsigned long)mem->userspace_addr,
962 			mem->memory_size)))
963 		goto out;
964 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
965 		goto out;
966 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
967 		goto out;
968 
969 	slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
970 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
971 	npages = mem->memory_size >> PAGE_SHIFT;
972 
973 	if (npages > KVM_MEM_MAX_NR_PAGES)
974 		goto out;
975 
976 	new = old = *slot;
977 
978 	new.id = id;
979 	new.base_gfn = base_gfn;
980 	new.npages = npages;
981 	new.flags = mem->flags;
982 
983 	if (npages) {
984 		if (!old.npages)
985 			change = KVM_MR_CREATE;
986 		else { /* Modify an existing slot. */
987 			if ((mem->userspace_addr != old.userspace_addr) ||
988 			    (npages != old.npages) ||
989 			    ((new.flags ^ old.flags) & KVM_MEM_READONLY))
990 				goto out;
991 
992 			if (base_gfn != old.base_gfn)
993 				change = KVM_MR_MOVE;
994 			else if (new.flags != old.flags)
995 				change = KVM_MR_FLAGS_ONLY;
996 			else { /* Nothing to change. */
997 				r = 0;
998 				goto out;
999 			}
1000 		}
1001 	} else {
1002 		if (!old.npages)
1003 			goto out;
1004 
1005 		change = KVM_MR_DELETE;
1006 		new.base_gfn = 0;
1007 		new.flags = 0;
1008 	}
1009 
1010 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
1011 		/* Check for overlaps */
1012 		r = -EEXIST;
1013 		kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
1014 			if ((slot->id >= KVM_USER_MEM_SLOTS) ||
1015 			    (slot->id == id))
1016 				continue;
1017 			if (!((base_gfn + npages <= slot->base_gfn) ||
1018 			      (base_gfn >= slot->base_gfn + slot->npages)))
1019 				goto out;
1020 		}
1021 	}
1022 
1023 	/* Free page dirty bitmap if unneeded */
1024 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
1025 		new.dirty_bitmap = NULL;
1026 
1027 	r = -ENOMEM;
1028 	if (change == KVM_MR_CREATE) {
1029 		new.userspace_addr = mem->userspace_addr;
1030 
1031 		if (kvm_arch_create_memslot(kvm, &new, npages))
1032 			goto out_free;
1033 	}
1034 
1035 	/* Allocate page dirty bitmap if needed */
1036 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
1037 		if (kvm_create_dirty_bitmap(&new) < 0)
1038 			goto out_free;
1039 	}
1040 
1041 	slots = kvzalloc(sizeof(struct kvm_memslots), GFP_KERNEL);
1042 	if (!slots)
1043 		goto out_free;
1044 	memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
1045 
1046 	if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
1047 		slot = id_to_memslot(slots, id);
1048 		slot->flags |= KVM_MEMSLOT_INVALID;
1049 
1050 		old_memslots = install_new_memslots(kvm, as_id, slots);
1051 
1052 		/* From this point no new shadow pages pointing to a deleted,
1053 		 * or moved, memslot will be created.
1054 		 *
1055 		 * validation of sp->gfn happens in:
1056 		 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1057 		 *	- kvm_is_visible_gfn (mmu_check_roots)
1058 		 */
1059 		kvm_arch_flush_shadow_memslot(kvm, slot);
1060 
1061 		/*
1062 		 * We can re-use the old_memslots from above, the only difference
1063 		 * from the currently installed memslots is the invalid flag.  This
1064 		 * will get overwritten by update_memslots anyway.
1065 		 */
1066 		slots = old_memslots;
1067 	}
1068 
1069 	r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
1070 	if (r)
1071 		goto out_slots;
1072 
1073 	/* actual memory is freed via old in kvm_free_memslot below */
1074 	if (change == KVM_MR_DELETE) {
1075 		new.dirty_bitmap = NULL;
1076 		memset(&new.arch, 0, sizeof(new.arch));
1077 	}
1078 
1079 	update_memslots(slots, &new);
1080 	old_memslots = install_new_memslots(kvm, as_id, slots);
1081 
1082 	kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
1083 
1084 	kvm_free_memslot(kvm, &old, &new);
1085 	kvfree(old_memslots);
1086 	return 0;
1087 
1088 out_slots:
1089 	kvfree(slots);
1090 out_free:
1091 	kvm_free_memslot(kvm, &new, &old);
1092 out:
1093 	return r;
1094 }
1095 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1096 
1097 int kvm_set_memory_region(struct kvm *kvm,
1098 			  const struct kvm_userspace_memory_region *mem)
1099 {
1100 	int r;
1101 
1102 	mutex_lock(&kvm->slots_lock);
1103 	r = __kvm_set_memory_region(kvm, mem);
1104 	mutex_unlock(&kvm->slots_lock);
1105 	return r;
1106 }
1107 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1108 
1109 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1110 					  struct kvm_userspace_memory_region *mem)
1111 {
1112 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1113 		return -EINVAL;
1114 
1115 	return kvm_set_memory_region(kvm, mem);
1116 }
1117 
1118 int kvm_get_dirty_log(struct kvm *kvm,
1119 			struct kvm_dirty_log *log, int *is_dirty)
1120 {
1121 	struct kvm_memslots *slots;
1122 	struct kvm_memory_slot *memslot;
1123 	int i, as_id, id;
1124 	unsigned long n;
1125 	unsigned long any = 0;
1126 
1127 	as_id = log->slot >> 16;
1128 	id = (u16)log->slot;
1129 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1130 		return -EINVAL;
1131 
1132 	slots = __kvm_memslots(kvm, as_id);
1133 	memslot = id_to_memslot(slots, id);
1134 	if (!memslot->dirty_bitmap)
1135 		return -ENOENT;
1136 
1137 	n = kvm_dirty_bitmap_bytes(memslot);
1138 
1139 	for (i = 0; !any && i < n/sizeof(long); ++i)
1140 		any = memslot->dirty_bitmap[i];
1141 
1142 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1143 		return -EFAULT;
1144 
1145 	if (any)
1146 		*is_dirty = 1;
1147 	return 0;
1148 }
1149 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1150 
1151 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1152 /**
1153  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1154  *	are dirty write protect them for next write.
1155  * @kvm:	pointer to kvm instance
1156  * @log:	slot id and address to which we copy the log
1157  * @is_dirty:	flag set if any page is dirty
1158  *
1159  * We need to keep it in mind that VCPU threads can write to the bitmap
1160  * concurrently. So, to avoid losing track of dirty pages we keep the
1161  * following order:
1162  *
1163  *    1. Take a snapshot of the bit and clear it if needed.
1164  *    2. Write protect the corresponding page.
1165  *    3. Copy the snapshot to the userspace.
1166  *    4. Upon return caller flushes TLB's if needed.
1167  *
1168  * Between 2 and 4, the guest may write to the page using the remaining TLB
1169  * entry.  This is not a problem because the page is reported dirty using
1170  * the snapshot taken before and step 4 ensures that writes done after
1171  * exiting to userspace will be logged for the next call.
1172  *
1173  */
1174 int kvm_get_dirty_log_protect(struct kvm *kvm,
1175 			struct kvm_dirty_log *log, bool *is_dirty)
1176 {
1177 	struct kvm_memslots *slots;
1178 	struct kvm_memory_slot *memslot;
1179 	int i, as_id, id;
1180 	unsigned long n;
1181 	unsigned long *dirty_bitmap;
1182 	unsigned long *dirty_bitmap_buffer;
1183 
1184 	as_id = log->slot >> 16;
1185 	id = (u16)log->slot;
1186 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1187 		return -EINVAL;
1188 
1189 	slots = __kvm_memslots(kvm, as_id);
1190 	memslot = id_to_memslot(slots, id);
1191 
1192 	dirty_bitmap = memslot->dirty_bitmap;
1193 	if (!dirty_bitmap)
1194 		return -ENOENT;
1195 
1196 	n = kvm_dirty_bitmap_bytes(memslot);
1197 
1198 	dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1199 	memset(dirty_bitmap_buffer, 0, n);
1200 
1201 	spin_lock(&kvm->mmu_lock);
1202 	*is_dirty = false;
1203 	for (i = 0; i < n / sizeof(long); i++) {
1204 		unsigned long mask;
1205 		gfn_t offset;
1206 
1207 		if (!dirty_bitmap[i])
1208 			continue;
1209 
1210 		*is_dirty = true;
1211 
1212 		mask = xchg(&dirty_bitmap[i], 0);
1213 		dirty_bitmap_buffer[i] = mask;
1214 
1215 		if (mask) {
1216 			offset = i * BITS_PER_LONG;
1217 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1218 								offset, mask);
1219 		}
1220 	}
1221 
1222 	spin_unlock(&kvm->mmu_lock);
1223 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1224 		return -EFAULT;
1225 	return 0;
1226 }
1227 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1228 #endif
1229 
1230 bool kvm_largepages_enabled(void)
1231 {
1232 	return largepages_enabled;
1233 }
1234 
1235 void kvm_disable_largepages(void)
1236 {
1237 	largepages_enabled = false;
1238 }
1239 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1240 
1241 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1242 {
1243 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1244 }
1245 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1246 
1247 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1248 {
1249 	return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1250 }
1251 
1252 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1253 {
1254 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1255 
1256 	if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1257 	      memslot->flags & KVM_MEMSLOT_INVALID)
1258 		return false;
1259 
1260 	return true;
1261 }
1262 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1263 
1264 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1265 {
1266 	struct vm_area_struct *vma;
1267 	unsigned long addr, size;
1268 
1269 	size = PAGE_SIZE;
1270 
1271 	addr = gfn_to_hva(kvm, gfn);
1272 	if (kvm_is_error_hva(addr))
1273 		return PAGE_SIZE;
1274 
1275 	down_read(&current->mm->mmap_sem);
1276 	vma = find_vma(current->mm, addr);
1277 	if (!vma)
1278 		goto out;
1279 
1280 	size = vma_kernel_pagesize(vma);
1281 
1282 out:
1283 	up_read(&current->mm->mmap_sem);
1284 
1285 	return size;
1286 }
1287 
1288 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1289 {
1290 	return slot->flags & KVM_MEM_READONLY;
1291 }
1292 
1293 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1294 				       gfn_t *nr_pages, bool write)
1295 {
1296 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1297 		return KVM_HVA_ERR_BAD;
1298 
1299 	if (memslot_is_readonly(slot) && write)
1300 		return KVM_HVA_ERR_RO_BAD;
1301 
1302 	if (nr_pages)
1303 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
1304 
1305 	return __gfn_to_hva_memslot(slot, gfn);
1306 }
1307 
1308 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1309 				     gfn_t *nr_pages)
1310 {
1311 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1312 }
1313 
1314 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1315 					gfn_t gfn)
1316 {
1317 	return gfn_to_hva_many(slot, gfn, NULL);
1318 }
1319 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1320 
1321 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1322 {
1323 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1324 }
1325 EXPORT_SYMBOL_GPL(gfn_to_hva);
1326 
1327 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1328 {
1329 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1330 }
1331 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1332 
1333 /*
1334  * If writable is set to false, the hva returned by this function is only
1335  * allowed to be read.
1336  */
1337 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1338 				      gfn_t gfn, bool *writable)
1339 {
1340 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1341 
1342 	if (!kvm_is_error_hva(hva) && writable)
1343 		*writable = !memslot_is_readonly(slot);
1344 
1345 	return hva;
1346 }
1347 
1348 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1349 {
1350 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1351 
1352 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1353 }
1354 
1355 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1356 {
1357 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1358 
1359 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1360 }
1361 
1362 static int get_user_page_nowait(unsigned long start, int write,
1363 		struct page **page)
1364 {
1365 	int flags = FOLL_NOWAIT | FOLL_HWPOISON;
1366 
1367 	if (write)
1368 		flags |= FOLL_WRITE;
1369 
1370 	return get_user_pages(start, 1, flags, page, NULL);
1371 }
1372 
1373 static inline int check_user_page_hwpoison(unsigned long addr)
1374 {
1375 	int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
1376 
1377 	rc = get_user_pages(addr, 1, flags, NULL, NULL);
1378 	return rc == -EHWPOISON;
1379 }
1380 
1381 /*
1382  * The atomic path to get the writable pfn which will be stored in @pfn,
1383  * true indicates success, otherwise false is returned.
1384  */
1385 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1386 			    bool write_fault, bool *writable, kvm_pfn_t *pfn)
1387 {
1388 	struct page *page[1];
1389 	int npages;
1390 
1391 	if (!(async || atomic))
1392 		return false;
1393 
1394 	/*
1395 	 * Fast pin a writable pfn only if it is a write fault request
1396 	 * or the caller allows to map a writable pfn for a read fault
1397 	 * request.
1398 	 */
1399 	if (!(write_fault || writable))
1400 		return false;
1401 
1402 	npages = __get_user_pages_fast(addr, 1, 1, page);
1403 	if (npages == 1) {
1404 		*pfn = page_to_pfn(page[0]);
1405 
1406 		if (writable)
1407 			*writable = true;
1408 		return true;
1409 	}
1410 
1411 	return false;
1412 }
1413 
1414 /*
1415  * The slow path to get the pfn of the specified host virtual address,
1416  * 1 indicates success, -errno is returned if error is detected.
1417  */
1418 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1419 			   bool *writable, kvm_pfn_t *pfn)
1420 {
1421 	struct page *page[1];
1422 	int npages = 0;
1423 
1424 	might_sleep();
1425 
1426 	if (writable)
1427 		*writable = write_fault;
1428 
1429 	if (async) {
1430 		down_read(&current->mm->mmap_sem);
1431 		npages = get_user_page_nowait(addr, write_fault, page);
1432 		up_read(&current->mm->mmap_sem);
1433 	} else {
1434 		unsigned int flags = FOLL_HWPOISON;
1435 
1436 		if (write_fault)
1437 			flags |= FOLL_WRITE;
1438 
1439 		npages = get_user_pages_unlocked(addr, 1, page, flags);
1440 	}
1441 	if (npages != 1)
1442 		return npages;
1443 
1444 	/* map read fault as writable if possible */
1445 	if (unlikely(!write_fault) && writable) {
1446 		struct page *wpage[1];
1447 
1448 		npages = __get_user_pages_fast(addr, 1, 1, wpage);
1449 		if (npages == 1) {
1450 			*writable = true;
1451 			put_page(page[0]);
1452 			page[0] = wpage[0];
1453 		}
1454 
1455 		npages = 1;
1456 	}
1457 	*pfn = page_to_pfn(page[0]);
1458 	return npages;
1459 }
1460 
1461 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1462 {
1463 	if (unlikely(!(vma->vm_flags & VM_READ)))
1464 		return false;
1465 
1466 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1467 		return false;
1468 
1469 	return true;
1470 }
1471 
1472 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
1473 			       unsigned long addr, bool *async,
1474 			       bool write_fault, kvm_pfn_t *p_pfn)
1475 {
1476 	unsigned long pfn;
1477 	int r;
1478 
1479 	r = follow_pfn(vma, addr, &pfn);
1480 	if (r) {
1481 		/*
1482 		 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1483 		 * not call the fault handler, so do it here.
1484 		 */
1485 		bool unlocked = false;
1486 		r = fixup_user_fault(current, current->mm, addr,
1487 				     (write_fault ? FAULT_FLAG_WRITE : 0),
1488 				     &unlocked);
1489 		if (unlocked)
1490 			return -EAGAIN;
1491 		if (r)
1492 			return r;
1493 
1494 		r = follow_pfn(vma, addr, &pfn);
1495 		if (r)
1496 			return r;
1497 
1498 	}
1499 
1500 
1501 	/*
1502 	 * Get a reference here because callers of *hva_to_pfn* and
1503 	 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1504 	 * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
1505 	 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1506 	 * simply do nothing for reserved pfns.
1507 	 *
1508 	 * Whoever called remap_pfn_range is also going to call e.g.
1509 	 * unmap_mapping_range before the underlying pages are freed,
1510 	 * causing a call to our MMU notifier.
1511 	 */
1512 	kvm_get_pfn(pfn);
1513 
1514 	*p_pfn = pfn;
1515 	return 0;
1516 }
1517 
1518 /*
1519  * Pin guest page in memory and return its pfn.
1520  * @addr: host virtual address which maps memory to the guest
1521  * @atomic: whether this function can sleep
1522  * @async: whether this function need to wait IO complete if the
1523  *         host page is not in the memory
1524  * @write_fault: whether we should get a writable host page
1525  * @writable: whether it allows to map a writable host page for !@write_fault
1526  *
1527  * The function will map a writable host page for these two cases:
1528  * 1): @write_fault = true
1529  * 2): @write_fault = false && @writable, @writable will tell the caller
1530  *     whether the mapping is writable.
1531  */
1532 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1533 			bool write_fault, bool *writable)
1534 {
1535 	struct vm_area_struct *vma;
1536 	kvm_pfn_t pfn = 0;
1537 	int npages, r;
1538 
1539 	/* we can do it either atomically or asynchronously, not both */
1540 	BUG_ON(atomic && async);
1541 
1542 	if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1543 		return pfn;
1544 
1545 	if (atomic)
1546 		return KVM_PFN_ERR_FAULT;
1547 
1548 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1549 	if (npages == 1)
1550 		return pfn;
1551 
1552 	down_read(&current->mm->mmap_sem);
1553 	if (npages == -EHWPOISON ||
1554 	      (!async && check_user_page_hwpoison(addr))) {
1555 		pfn = KVM_PFN_ERR_HWPOISON;
1556 		goto exit;
1557 	}
1558 
1559 retry:
1560 	vma = find_vma_intersection(current->mm, addr, addr + 1);
1561 
1562 	if (vma == NULL)
1563 		pfn = KVM_PFN_ERR_FAULT;
1564 	else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
1565 		r = hva_to_pfn_remapped(vma, addr, async, write_fault, &pfn);
1566 		if (r == -EAGAIN)
1567 			goto retry;
1568 		if (r < 0)
1569 			pfn = KVM_PFN_ERR_FAULT;
1570 	} else {
1571 		if (async && vma_is_valid(vma, write_fault))
1572 			*async = true;
1573 		pfn = KVM_PFN_ERR_FAULT;
1574 	}
1575 exit:
1576 	up_read(&current->mm->mmap_sem);
1577 	return pfn;
1578 }
1579 
1580 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1581 			       bool atomic, bool *async, bool write_fault,
1582 			       bool *writable)
1583 {
1584 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1585 
1586 	if (addr == KVM_HVA_ERR_RO_BAD) {
1587 		if (writable)
1588 			*writable = false;
1589 		return KVM_PFN_ERR_RO_FAULT;
1590 	}
1591 
1592 	if (kvm_is_error_hva(addr)) {
1593 		if (writable)
1594 			*writable = false;
1595 		return KVM_PFN_NOSLOT;
1596 	}
1597 
1598 	/* Do not map writable pfn in the readonly memslot. */
1599 	if (writable && memslot_is_readonly(slot)) {
1600 		*writable = false;
1601 		writable = NULL;
1602 	}
1603 
1604 	return hva_to_pfn(addr, atomic, async, write_fault,
1605 			  writable);
1606 }
1607 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1608 
1609 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1610 		      bool *writable)
1611 {
1612 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1613 				    write_fault, writable);
1614 }
1615 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1616 
1617 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1618 {
1619 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1620 }
1621 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1622 
1623 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1624 {
1625 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1626 }
1627 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1628 
1629 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1630 {
1631 	return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1632 }
1633 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1634 
1635 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1636 {
1637 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1638 }
1639 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1640 
1641 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1642 {
1643 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1644 }
1645 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1646 
1647 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1648 {
1649 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1650 }
1651 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1652 
1653 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1654 			    struct page **pages, int nr_pages)
1655 {
1656 	unsigned long addr;
1657 	gfn_t entry;
1658 
1659 	addr = gfn_to_hva_many(slot, gfn, &entry);
1660 	if (kvm_is_error_hva(addr))
1661 		return -1;
1662 
1663 	if (entry < nr_pages)
1664 		return 0;
1665 
1666 	return __get_user_pages_fast(addr, nr_pages, 1, pages);
1667 }
1668 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1669 
1670 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1671 {
1672 	if (is_error_noslot_pfn(pfn))
1673 		return KVM_ERR_PTR_BAD_PAGE;
1674 
1675 	if (kvm_is_reserved_pfn(pfn)) {
1676 		WARN_ON(1);
1677 		return KVM_ERR_PTR_BAD_PAGE;
1678 	}
1679 
1680 	return pfn_to_page(pfn);
1681 }
1682 
1683 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1684 {
1685 	kvm_pfn_t pfn;
1686 
1687 	pfn = gfn_to_pfn(kvm, gfn);
1688 
1689 	return kvm_pfn_to_page(pfn);
1690 }
1691 EXPORT_SYMBOL_GPL(gfn_to_page);
1692 
1693 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1694 {
1695 	kvm_pfn_t pfn;
1696 
1697 	pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1698 
1699 	return kvm_pfn_to_page(pfn);
1700 }
1701 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1702 
1703 void kvm_release_page_clean(struct page *page)
1704 {
1705 	WARN_ON(is_error_page(page));
1706 
1707 	kvm_release_pfn_clean(page_to_pfn(page));
1708 }
1709 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1710 
1711 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1712 {
1713 	if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1714 		put_page(pfn_to_page(pfn));
1715 }
1716 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1717 
1718 void kvm_release_page_dirty(struct page *page)
1719 {
1720 	WARN_ON(is_error_page(page));
1721 
1722 	kvm_release_pfn_dirty(page_to_pfn(page));
1723 }
1724 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1725 
1726 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1727 {
1728 	kvm_set_pfn_dirty(pfn);
1729 	kvm_release_pfn_clean(pfn);
1730 }
1731 
1732 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1733 {
1734 	if (!kvm_is_reserved_pfn(pfn)) {
1735 		struct page *page = pfn_to_page(pfn);
1736 
1737 		if (!PageReserved(page))
1738 			SetPageDirty(page);
1739 	}
1740 }
1741 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1742 
1743 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1744 {
1745 	if (!kvm_is_reserved_pfn(pfn))
1746 		mark_page_accessed(pfn_to_page(pfn));
1747 }
1748 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1749 
1750 void kvm_get_pfn(kvm_pfn_t pfn)
1751 {
1752 	if (!kvm_is_reserved_pfn(pfn))
1753 		get_page(pfn_to_page(pfn));
1754 }
1755 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1756 
1757 static int next_segment(unsigned long len, int offset)
1758 {
1759 	if (len > PAGE_SIZE - offset)
1760 		return PAGE_SIZE - offset;
1761 	else
1762 		return len;
1763 }
1764 
1765 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1766 				 void *data, int offset, int len)
1767 {
1768 	int r;
1769 	unsigned long addr;
1770 
1771 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1772 	if (kvm_is_error_hva(addr))
1773 		return -EFAULT;
1774 	r = __copy_from_user(data, (void __user *)addr + offset, len);
1775 	if (r)
1776 		return -EFAULT;
1777 	return 0;
1778 }
1779 
1780 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1781 			int len)
1782 {
1783 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1784 
1785 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1786 }
1787 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1788 
1789 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1790 			     int offset, int len)
1791 {
1792 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1793 
1794 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1795 }
1796 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1797 
1798 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1799 {
1800 	gfn_t gfn = gpa >> PAGE_SHIFT;
1801 	int seg;
1802 	int offset = offset_in_page(gpa);
1803 	int ret;
1804 
1805 	while ((seg = next_segment(len, offset)) != 0) {
1806 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1807 		if (ret < 0)
1808 			return ret;
1809 		offset = 0;
1810 		len -= seg;
1811 		data += seg;
1812 		++gfn;
1813 	}
1814 	return 0;
1815 }
1816 EXPORT_SYMBOL_GPL(kvm_read_guest);
1817 
1818 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1819 {
1820 	gfn_t gfn = gpa >> PAGE_SHIFT;
1821 	int seg;
1822 	int offset = offset_in_page(gpa);
1823 	int ret;
1824 
1825 	while ((seg = next_segment(len, offset)) != 0) {
1826 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1827 		if (ret < 0)
1828 			return ret;
1829 		offset = 0;
1830 		len -= seg;
1831 		data += seg;
1832 		++gfn;
1833 	}
1834 	return 0;
1835 }
1836 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1837 
1838 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1839 			           void *data, int offset, unsigned long len)
1840 {
1841 	int r;
1842 	unsigned long addr;
1843 
1844 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1845 	if (kvm_is_error_hva(addr))
1846 		return -EFAULT;
1847 	pagefault_disable();
1848 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1849 	pagefault_enable();
1850 	if (r)
1851 		return -EFAULT;
1852 	return 0;
1853 }
1854 
1855 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1856 			  unsigned long len)
1857 {
1858 	gfn_t gfn = gpa >> PAGE_SHIFT;
1859 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1860 	int offset = offset_in_page(gpa);
1861 
1862 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1863 }
1864 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1865 
1866 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1867 			       void *data, unsigned long len)
1868 {
1869 	gfn_t gfn = gpa >> PAGE_SHIFT;
1870 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1871 	int offset = offset_in_page(gpa);
1872 
1873 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1874 }
1875 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1876 
1877 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1878 			          const void *data, int offset, int len)
1879 {
1880 	int r;
1881 	unsigned long addr;
1882 
1883 	addr = gfn_to_hva_memslot(memslot, gfn);
1884 	if (kvm_is_error_hva(addr))
1885 		return -EFAULT;
1886 	r = __copy_to_user((void __user *)addr + offset, data, len);
1887 	if (r)
1888 		return -EFAULT;
1889 	mark_page_dirty_in_slot(memslot, gfn);
1890 	return 0;
1891 }
1892 
1893 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1894 			 const void *data, int offset, int len)
1895 {
1896 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1897 
1898 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1899 }
1900 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1901 
1902 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1903 			      const void *data, int offset, int len)
1904 {
1905 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1906 
1907 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1908 }
1909 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1910 
1911 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1912 		    unsigned long len)
1913 {
1914 	gfn_t gfn = gpa >> PAGE_SHIFT;
1915 	int seg;
1916 	int offset = offset_in_page(gpa);
1917 	int ret;
1918 
1919 	while ((seg = next_segment(len, offset)) != 0) {
1920 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1921 		if (ret < 0)
1922 			return ret;
1923 		offset = 0;
1924 		len -= seg;
1925 		data += seg;
1926 		++gfn;
1927 	}
1928 	return 0;
1929 }
1930 EXPORT_SYMBOL_GPL(kvm_write_guest);
1931 
1932 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1933 		         unsigned long len)
1934 {
1935 	gfn_t gfn = gpa >> PAGE_SHIFT;
1936 	int seg;
1937 	int offset = offset_in_page(gpa);
1938 	int ret;
1939 
1940 	while ((seg = next_segment(len, offset)) != 0) {
1941 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1942 		if (ret < 0)
1943 			return ret;
1944 		offset = 0;
1945 		len -= seg;
1946 		data += seg;
1947 		++gfn;
1948 	}
1949 	return 0;
1950 }
1951 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1952 
1953 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
1954 				       struct gfn_to_hva_cache *ghc,
1955 				       gpa_t gpa, unsigned long len)
1956 {
1957 	int offset = offset_in_page(gpa);
1958 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
1959 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1960 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1961 	gfn_t nr_pages_avail;
1962 
1963 	ghc->gpa = gpa;
1964 	ghc->generation = slots->generation;
1965 	ghc->len = len;
1966 	ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1967 	ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1968 	if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1969 		ghc->hva += offset;
1970 	} else {
1971 		/*
1972 		 * If the requested region crosses two memslots, we still
1973 		 * verify that the entire region is valid here.
1974 		 */
1975 		while (start_gfn <= end_gfn) {
1976 			ghc->memslot = __gfn_to_memslot(slots, start_gfn);
1977 			ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1978 						   &nr_pages_avail);
1979 			if (kvm_is_error_hva(ghc->hva))
1980 				return -EFAULT;
1981 			start_gfn += nr_pages_avail;
1982 		}
1983 		/* Use the slow path for cross page reads and writes. */
1984 		ghc->memslot = NULL;
1985 	}
1986 	return 0;
1987 }
1988 
1989 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1990 			      gpa_t gpa, unsigned long len)
1991 {
1992 	struct kvm_memslots *slots = kvm_memslots(kvm);
1993 	return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
1994 }
1995 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1996 
1997 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1998 			   void *data, int offset, unsigned long len)
1999 {
2000 	struct kvm_memslots *slots = kvm_memslots(kvm);
2001 	int r;
2002 	gpa_t gpa = ghc->gpa + offset;
2003 
2004 	BUG_ON(len + offset > ghc->len);
2005 
2006 	if (slots->generation != ghc->generation)
2007 		__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2008 
2009 	if (unlikely(!ghc->memslot))
2010 		return kvm_write_guest(kvm, gpa, data, len);
2011 
2012 	if (kvm_is_error_hva(ghc->hva))
2013 		return -EFAULT;
2014 
2015 	r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
2016 	if (r)
2017 		return -EFAULT;
2018 	mark_page_dirty_in_slot(ghc->memslot, gpa >> PAGE_SHIFT);
2019 
2020 	return 0;
2021 }
2022 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
2023 
2024 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2025 			   void *data, unsigned long len)
2026 {
2027 	return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
2028 }
2029 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
2030 
2031 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
2032 			   void *data, unsigned long len)
2033 {
2034 	struct kvm_memslots *slots = kvm_memslots(kvm);
2035 	int r;
2036 
2037 	BUG_ON(len > ghc->len);
2038 
2039 	if (slots->generation != ghc->generation)
2040 		__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len);
2041 
2042 	if (unlikely(!ghc->memslot))
2043 		return kvm_read_guest(kvm, ghc->gpa, data, len);
2044 
2045 	if (kvm_is_error_hva(ghc->hva))
2046 		return -EFAULT;
2047 
2048 	r = __copy_from_user(data, (void __user *)ghc->hva, len);
2049 	if (r)
2050 		return -EFAULT;
2051 
2052 	return 0;
2053 }
2054 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
2055 
2056 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
2057 {
2058 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2059 
2060 	return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
2061 }
2062 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
2063 
2064 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
2065 {
2066 	gfn_t gfn = gpa >> PAGE_SHIFT;
2067 	int seg;
2068 	int offset = offset_in_page(gpa);
2069 	int ret;
2070 
2071 	while ((seg = next_segment(len, offset)) != 0) {
2072 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
2073 		if (ret < 0)
2074 			return ret;
2075 		offset = 0;
2076 		len -= seg;
2077 		++gfn;
2078 	}
2079 	return 0;
2080 }
2081 EXPORT_SYMBOL_GPL(kvm_clear_guest);
2082 
2083 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
2084 				    gfn_t gfn)
2085 {
2086 	if (memslot && memslot->dirty_bitmap) {
2087 		unsigned long rel_gfn = gfn - memslot->base_gfn;
2088 
2089 		set_bit_le(rel_gfn, memslot->dirty_bitmap);
2090 	}
2091 }
2092 
2093 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
2094 {
2095 	struct kvm_memory_slot *memslot;
2096 
2097 	memslot = gfn_to_memslot(kvm, gfn);
2098 	mark_page_dirty_in_slot(memslot, gfn);
2099 }
2100 EXPORT_SYMBOL_GPL(mark_page_dirty);
2101 
2102 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
2103 {
2104 	struct kvm_memory_slot *memslot;
2105 
2106 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2107 	mark_page_dirty_in_slot(memslot, gfn);
2108 }
2109 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
2110 
2111 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
2112 {
2113 	unsigned int old, val, grow;
2114 
2115 	old = val = vcpu->halt_poll_ns;
2116 	grow = READ_ONCE(halt_poll_ns_grow);
2117 	/* 10us base */
2118 	if (val == 0 && grow)
2119 		val = 10000;
2120 	else
2121 		val *= grow;
2122 
2123 	if (val > halt_poll_ns)
2124 		val = halt_poll_ns;
2125 
2126 	vcpu->halt_poll_ns = val;
2127 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
2128 }
2129 
2130 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
2131 {
2132 	unsigned int old, val, shrink;
2133 
2134 	old = val = vcpu->halt_poll_ns;
2135 	shrink = READ_ONCE(halt_poll_ns_shrink);
2136 	if (shrink == 0)
2137 		val = 0;
2138 	else
2139 		val /= shrink;
2140 
2141 	vcpu->halt_poll_ns = val;
2142 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
2143 }
2144 
2145 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
2146 {
2147 	if (kvm_arch_vcpu_runnable(vcpu)) {
2148 		kvm_make_request(KVM_REQ_UNHALT, vcpu);
2149 		return -EINTR;
2150 	}
2151 	if (kvm_cpu_has_pending_timer(vcpu))
2152 		return -EINTR;
2153 	if (signal_pending(current))
2154 		return -EINTR;
2155 
2156 	return 0;
2157 }
2158 
2159 /*
2160  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2161  */
2162 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2163 {
2164 	ktime_t start, cur;
2165 	DECLARE_SWAITQUEUE(wait);
2166 	bool waited = false;
2167 	u64 block_ns;
2168 
2169 	start = cur = ktime_get();
2170 	if (vcpu->halt_poll_ns) {
2171 		ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2172 
2173 		++vcpu->stat.halt_attempted_poll;
2174 		do {
2175 			/*
2176 			 * This sets KVM_REQ_UNHALT if an interrupt
2177 			 * arrives.
2178 			 */
2179 			if (kvm_vcpu_check_block(vcpu) < 0) {
2180 				++vcpu->stat.halt_successful_poll;
2181 				if (!vcpu_valid_wakeup(vcpu))
2182 					++vcpu->stat.halt_poll_invalid;
2183 				goto out;
2184 			}
2185 			cur = ktime_get();
2186 		} while (single_task_running() && ktime_before(cur, stop));
2187 	}
2188 
2189 	kvm_arch_vcpu_blocking(vcpu);
2190 
2191 	for (;;) {
2192 		prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2193 
2194 		if (kvm_vcpu_check_block(vcpu) < 0)
2195 			break;
2196 
2197 		waited = true;
2198 		schedule();
2199 	}
2200 
2201 	finish_swait(&vcpu->wq, &wait);
2202 	cur = ktime_get();
2203 
2204 	kvm_arch_vcpu_unblocking(vcpu);
2205 out:
2206 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2207 
2208 	if (!vcpu_valid_wakeup(vcpu))
2209 		shrink_halt_poll_ns(vcpu);
2210 	else if (halt_poll_ns) {
2211 		if (block_ns <= vcpu->halt_poll_ns)
2212 			;
2213 		/* we had a long block, shrink polling */
2214 		else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2215 			shrink_halt_poll_ns(vcpu);
2216 		/* we had a short halt and our poll time is too small */
2217 		else if (vcpu->halt_poll_ns < halt_poll_ns &&
2218 			block_ns < halt_poll_ns)
2219 			grow_halt_poll_ns(vcpu);
2220 	} else
2221 		vcpu->halt_poll_ns = 0;
2222 
2223 	trace_kvm_vcpu_wakeup(block_ns, waited, vcpu_valid_wakeup(vcpu));
2224 	kvm_arch_vcpu_block_finish(vcpu);
2225 }
2226 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2227 
2228 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
2229 {
2230 	struct swait_queue_head *wqp;
2231 
2232 	wqp = kvm_arch_vcpu_wq(vcpu);
2233 	if (swait_active(wqp)) {
2234 		swake_up(wqp);
2235 		++vcpu->stat.halt_wakeup;
2236 		return true;
2237 	}
2238 
2239 	return false;
2240 }
2241 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
2242 
2243 #ifndef CONFIG_S390
2244 /*
2245  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2246  */
2247 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2248 {
2249 	int me;
2250 	int cpu = vcpu->cpu;
2251 
2252 	if (kvm_vcpu_wake_up(vcpu))
2253 		return;
2254 
2255 	me = get_cpu();
2256 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2257 		if (kvm_arch_vcpu_should_kick(vcpu))
2258 			smp_send_reschedule(cpu);
2259 	put_cpu();
2260 }
2261 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2262 #endif /* !CONFIG_S390 */
2263 
2264 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2265 {
2266 	struct pid *pid;
2267 	struct task_struct *task = NULL;
2268 	int ret = 0;
2269 
2270 	rcu_read_lock();
2271 	pid = rcu_dereference(target->pid);
2272 	if (pid)
2273 		task = get_pid_task(pid, PIDTYPE_PID);
2274 	rcu_read_unlock();
2275 	if (!task)
2276 		return ret;
2277 	ret = yield_to(task, 1);
2278 	put_task_struct(task);
2279 
2280 	return ret;
2281 }
2282 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2283 
2284 /*
2285  * Helper that checks whether a VCPU is eligible for directed yield.
2286  * Most eligible candidate to yield is decided by following heuristics:
2287  *
2288  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2289  *  (preempted lock holder), indicated by @in_spin_loop.
2290  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2291  *
2292  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2293  *  chance last time (mostly it has become eligible now since we have probably
2294  *  yielded to lockholder in last iteration. This is done by toggling
2295  *  @dy_eligible each time a VCPU checked for eligibility.)
2296  *
2297  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2298  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2299  *  burning. Giving priority for a potential lock-holder increases lock
2300  *  progress.
2301  *
2302  *  Since algorithm is based on heuristics, accessing another VCPU data without
2303  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2304  *  and continue with next VCPU and so on.
2305  */
2306 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2307 {
2308 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2309 	bool eligible;
2310 
2311 	eligible = !vcpu->spin_loop.in_spin_loop ||
2312 		    vcpu->spin_loop.dy_eligible;
2313 
2314 	if (vcpu->spin_loop.in_spin_loop)
2315 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2316 
2317 	return eligible;
2318 #else
2319 	return true;
2320 #endif
2321 }
2322 
2323 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2324 {
2325 	struct kvm *kvm = me->kvm;
2326 	struct kvm_vcpu *vcpu;
2327 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2328 	int yielded = 0;
2329 	int try = 3;
2330 	int pass;
2331 	int i;
2332 
2333 	kvm_vcpu_set_in_spin_loop(me, true);
2334 	/*
2335 	 * We boost the priority of a VCPU that is runnable but not
2336 	 * currently running, because it got preempted by something
2337 	 * else and called schedule in __vcpu_run.  Hopefully that
2338 	 * VCPU is holding the lock that we need and will release it.
2339 	 * We approximate round-robin by starting at the last boosted VCPU.
2340 	 */
2341 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
2342 		kvm_for_each_vcpu(i, vcpu, kvm) {
2343 			if (!pass && i <= last_boosted_vcpu) {
2344 				i = last_boosted_vcpu;
2345 				continue;
2346 			} else if (pass && i > last_boosted_vcpu)
2347 				break;
2348 			if (!ACCESS_ONCE(vcpu->preempted))
2349 				continue;
2350 			if (vcpu == me)
2351 				continue;
2352 			if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2353 				continue;
2354 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2355 				continue;
2356 
2357 			yielded = kvm_vcpu_yield_to(vcpu);
2358 			if (yielded > 0) {
2359 				kvm->last_boosted_vcpu = i;
2360 				break;
2361 			} else if (yielded < 0) {
2362 				try--;
2363 				if (!try)
2364 					break;
2365 			}
2366 		}
2367 	}
2368 	kvm_vcpu_set_in_spin_loop(me, false);
2369 
2370 	/* Ensure vcpu is not eligible during next spinloop */
2371 	kvm_vcpu_set_dy_eligible(me, false);
2372 }
2373 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2374 
2375 static int kvm_vcpu_fault(struct vm_fault *vmf)
2376 {
2377 	struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
2378 	struct page *page;
2379 
2380 	if (vmf->pgoff == 0)
2381 		page = virt_to_page(vcpu->run);
2382 #ifdef CONFIG_X86
2383 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2384 		page = virt_to_page(vcpu->arch.pio_data);
2385 #endif
2386 #ifdef CONFIG_KVM_MMIO
2387 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2388 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2389 #endif
2390 	else
2391 		return kvm_arch_vcpu_fault(vcpu, vmf);
2392 	get_page(page);
2393 	vmf->page = page;
2394 	return 0;
2395 }
2396 
2397 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2398 	.fault = kvm_vcpu_fault,
2399 };
2400 
2401 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2402 {
2403 	vma->vm_ops = &kvm_vcpu_vm_ops;
2404 	return 0;
2405 }
2406 
2407 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2408 {
2409 	struct kvm_vcpu *vcpu = filp->private_data;
2410 
2411 	debugfs_remove_recursive(vcpu->debugfs_dentry);
2412 	kvm_put_kvm(vcpu->kvm);
2413 	return 0;
2414 }
2415 
2416 static struct file_operations kvm_vcpu_fops = {
2417 	.release        = kvm_vcpu_release,
2418 	.unlocked_ioctl = kvm_vcpu_ioctl,
2419 #ifdef CONFIG_KVM_COMPAT
2420 	.compat_ioctl   = kvm_vcpu_compat_ioctl,
2421 #endif
2422 	.mmap           = kvm_vcpu_mmap,
2423 	.llseek		= noop_llseek,
2424 };
2425 
2426 /*
2427  * Allocates an inode for the vcpu.
2428  */
2429 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2430 {
2431 	return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2432 }
2433 
2434 static int kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
2435 {
2436 	char dir_name[ITOA_MAX_LEN * 2];
2437 	int ret;
2438 
2439 	if (!kvm_arch_has_vcpu_debugfs())
2440 		return 0;
2441 
2442 	if (!debugfs_initialized())
2443 		return 0;
2444 
2445 	snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
2446 	vcpu->debugfs_dentry = debugfs_create_dir(dir_name,
2447 								vcpu->kvm->debugfs_dentry);
2448 	if (!vcpu->debugfs_dentry)
2449 		return -ENOMEM;
2450 
2451 	ret = kvm_arch_create_vcpu_debugfs(vcpu);
2452 	if (ret < 0) {
2453 		debugfs_remove_recursive(vcpu->debugfs_dentry);
2454 		return ret;
2455 	}
2456 
2457 	return 0;
2458 }
2459 
2460 /*
2461  * Creates some virtual cpus.  Good luck creating more than one.
2462  */
2463 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2464 {
2465 	int r;
2466 	struct kvm_vcpu *vcpu;
2467 
2468 	if (id >= KVM_MAX_VCPU_ID)
2469 		return -EINVAL;
2470 
2471 	mutex_lock(&kvm->lock);
2472 	if (kvm->created_vcpus == KVM_MAX_VCPUS) {
2473 		mutex_unlock(&kvm->lock);
2474 		return -EINVAL;
2475 	}
2476 
2477 	kvm->created_vcpus++;
2478 	mutex_unlock(&kvm->lock);
2479 
2480 	vcpu = kvm_arch_vcpu_create(kvm, id);
2481 	if (IS_ERR(vcpu)) {
2482 		r = PTR_ERR(vcpu);
2483 		goto vcpu_decrement;
2484 	}
2485 
2486 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2487 
2488 	r = kvm_arch_vcpu_setup(vcpu);
2489 	if (r)
2490 		goto vcpu_destroy;
2491 
2492 	r = kvm_create_vcpu_debugfs(vcpu);
2493 	if (r)
2494 		goto vcpu_destroy;
2495 
2496 	mutex_lock(&kvm->lock);
2497 	if (kvm_get_vcpu_by_id(kvm, id)) {
2498 		r = -EEXIST;
2499 		goto unlock_vcpu_destroy;
2500 	}
2501 
2502 	BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2503 
2504 	/* Now it's all set up, let userspace reach it */
2505 	kvm_get_kvm(kvm);
2506 	r = create_vcpu_fd(vcpu);
2507 	if (r < 0) {
2508 		kvm_put_kvm(kvm);
2509 		goto unlock_vcpu_destroy;
2510 	}
2511 
2512 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2513 
2514 	/*
2515 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2516 	 * before kvm->online_vcpu's incremented value.
2517 	 */
2518 	smp_wmb();
2519 	atomic_inc(&kvm->online_vcpus);
2520 
2521 	mutex_unlock(&kvm->lock);
2522 	kvm_arch_vcpu_postcreate(vcpu);
2523 	return r;
2524 
2525 unlock_vcpu_destroy:
2526 	mutex_unlock(&kvm->lock);
2527 	debugfs_remove_recursive(vcpu->debugfs_dentry);
2528 vcpu_destroy:
2529 	kvm_arch_vcpu_destroy(vcpu);
2530 vcpu_decrement:
2531 	mutex_lock(&kvm->lock);
2532 	kvm->created_vcpus--;
2533 	mutex_unlock(&kvm->lock);
2534 	return r;
2535 }
2536 
2537 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2538 {
2539 	if (sigset) {
2540 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2541 		vcpu->sigset_active = 1;
2542 		vcpu->sigset = *sigset;
2543 	} else
2544 		vcpu->sigset_active = 0;
2545 	return 0;
2546 }
2547 
2548 static long kvm_vcpu_ioctl(struct file *filp,
2549 			   unsigned int ioctl, unsigned long arg)
2550 {
2551 	struct kvm_vcpu *vcpu = filp->private_data;
2552 	void __user *argp = (void __user *)arg;
2553 	int r;
2554 	struct kvm_fpu *fpu = NULL;
2555 	struct kvm_sregs *kvm_sregs = NULL;
2556 
2557 	if (vcpu->kvm->mm != current->mm)
2558 		return -EIO;
2559 
2560 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2561 		return -EINVAL;
2562 
2563 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2564 	/*
2565 	 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2566 	 * so vcpu_load() would break it.
2567 	 */
2568 	if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2569 		return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2570 #endif
2571 
2572 
2573 	r = vcpu_load(vcpu);
2574 	if (r)
2575 		return r;
2576 	switch (ioctl) {
2577 	case KVM_RUN: {
2578 		struct pid *oldpid;
2579 		r = -EINVAL;
2580 		if (arg)
2581 			goto out;
2582 		oldpid = rcu_access_pointer(vcpu->pid);
2583 		if (unlikely(oldpid != current->pids[PIDTYPE_PID].pid)) {
2584 			/* The thread running this VCPU changed. */
2585 			struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2586 
2587 			rcu_assign_pointer(vcpu->pid, newpid);
2588 			if (oldpid)
2589 				synchronize_rcu();
2590 			put_pid(oldpid);
2591 		}
2592 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2593 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2594 		break;
2595 	}
2596 	case KVM_GET_REGS: {
2597 		struct kvm_regs *kvm_regs;
2598 
2599 		r = -ENOMEM;
2600 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2601 		if (!kvm_regs)
2602 			goto out;
2603 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2604 		if (r)
2605 			goto out_free1;
2606 		r = -EFAULT;
2607 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2608 			goto out_free1;
2609 		r = 0;
2610 out_free1:
2611 		kfree(kvm_regs);
2612 		break;
2613 	}
2614 	case KVM_SET_REGS: {
2615 		struct kvm_regs *kvm_regs;
2616 
2617 		r = -ENOMEM;
2618 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2619 		if (IS_ERR(kvm_regs)) {
2620 			r = PTR_ERR(kvm_regs);
2621 			goto out;
2622 		}
2623 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2624 		kfree(kvm_regs);
2625 		break;
2626 	}
2627 	case KVM_GET_SREGS: {
2628 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2629 		r = -ENOMEM;
2630 		if (!kvm_sregs)
2631 			goto out;
2632 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2633 		if (r)
2634 			goto out;
2635 		r = -EFAULT;
2636 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2637 			goto out;
2638 		r = 0;
2639 		break;
2640 	}
2641 	case KVM_SET_SREGS: {
2642 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2643 		if (IS_ERR(kvm_sregs)) {
2644 			r = PTR_ERR(kvm_sregs);
2645 			kvm_sregs = NULL;
2646 			goto out;
2647 		}
2648 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2649 		break;
2650 	}
2651 	case KVM_GET_MP_STATE: {
2652 		struct kvm_mp_state mp_state;
2653 
2654 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2655 		if (r)
2656 			goto out;
2657 		r = -EFAULT;
2658 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2659 			goto out;
2660 		r = 0;
2661 		break;
2662 	}
2663 	case KVM_SET_MP_STATE: {
2664 		struct kvm_mp_state mp_state;
2665 
2666 		r = -EFAULT;
2667 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2668 			goto out;
2669 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2670 		break;
2671 	}
2672 	case KVM_TRANSLATE: {
2673 		struct kvm_translation tr;
2674 
2675 		r = -EFAULT;
2676 		if (copy_from_user(&tr, argp, sizeof(tr)))
2677 			goto out;
2678 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2679 		if (r)
2680 			goto out;
2681 		r = -EFAULT;
2682 		if (copy_to_user(argp, &tr, sizeof(tr)))
2683 			goto out;
2684 		r = 0;
2685 		break;
2686 	}
2687 	case KVM_SET_GUEST_DEBUG: {
2688 		struct kvm_guest_debug dbg;
2689 
2690 		r = -EFAULT;
2691 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
2692 			goto out;
2693 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2694 		break;
2695 	}
2696 	case KVM_SET_SIGNAL_MASK: {
2697 		struct kvm_signal_mask __user *sigmask_arg = argp;
2698 		struct kvm_signal_mask kvm_sigmask;
2699 		sigset_t sigset, *p;
2700 
2701 		p = NULL;
2702 		if (argp) {
2703 			r = -EFAULT;
2704 			if (copy_from_user(&kvm_sigmask, argp,
2705 					   sizeof(kvm_sigmask)))
2706 				goto out;
2707 			r = -EINVAL;
2708 			if (kvm_sigmask.len != sizeof(sigset))
2709 				goto out;
2710 			r = -EFAULT;
2711 			if (copy_from_user(&sigset, sigmask_arg->sigset,
2712 					   sizeof(sigset)))
2713 				goto out;
2714 			p = &sigset;
2715 		}
2716 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2717 		break;
2718 	}
2719 	case KVM_GET_FPU: {
2720 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2721 		r = -ENOMEM;
2722 		if (!fpu)
2723 			goto out;
2724 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2725 		if (r)
2726 			goto out;
2727 		r = -EFAULT;
2728 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2729 			goto out;
2730 		r = 0;
2731 		break;
2732 	}
2733 	case KVM_SET_FPU: {
2734 		fpu = memdup_user(argp, sizeof(*fpu));
2735 		if (IS_ERR(fpu)) {
2736 			r = PTR_ERR(fpu);
2737 			fpu = NULL;
2738 			goto out;
2739 		}
2740 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2741 		break;
2742 	}
2743 	default:
2744 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2745 	}
2746 out:
2747 	vcpu_put(vcpu);
2748 	kfree(fpu);
2749 	kfree(kvm_sregs);
2750 	return r;
2751 }
2752 
2753 #ifdef CONFIG_KVM_COMPAT
2754 static long kvm_vcpu_compat_ioctl(struct file *filp,
2755 				  unsigned int ioctl, unsigned long arg)
2756 {
2757 	struct kvm_vcpu *vcpu = filp->private_data;
2758 	void __user *argp = compat_ptr(arg);
2759 	int r;
2760 
2761 	if (vcpu->kvm->mm != current->mm)
2762 		return -EIO;
2763 
2764 	switch (ioctl) {
2765 	case KVM_SET_SIGNAL_MASK: {
2766 		struct kvm_signal_mask __user *sigmask_arg = argp;
2767 		struct kvm_signal_mask kvm_sigmask;
2768 		compat_sigset_t csigset;
2769 		sigset_t sigset;
2770 
2771 		if (argp) {
2772 			r = -EFAULT;
2773 			if (copy_from_user(&kvm_sigmask, argp,
2774 					   sizeof(kvm_sigmask)))
2775 				goto out;
2776 			r = -EINVAL;
2777 			if (kvm_sigmask.len != sizeof(csigset))
2778 				goto out;
2779 			r = -EFAULT;
2780 			if (copy_from_user(&csigset, sigmask_arg->sigset,
2781 					   sizeof(csigset)))
2782 				goto out;
2783 			sigset_from_compat(&sigset, &csigset);
2784 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2785 		} else
2786 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2787 		break;
2788 	}
2789 	default:
2790 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
2791 	}
2792 
2793 out:
2794 	return r;
2795 }
2796 #endif
2797 
2798 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2799 				 int (*accessor)(struct kvm_device *dev,
2800 						 struct kvm_device_attr *attr),
2801 				 unsigned long arg)
2802 {
2803 	struct kvm_device_attr attr;
2804 
2805 	if (!accessor)
2806 		return -EPERM;
2807 
2808 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2809 		return -EFAULT;
2810 
2811 	return accessor(dev, &attr);
2812 }
2813 
2814 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2815 			     unsigned long arg)
2816 {
2817 	struct kvm_device *dev = filp->private_data;
2818 
2819 	switch (ioctl) {
2820 	case KVM_SET_DEVICE_ATTR:
2821 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2822 	case KVM_GET_DEVICE_ATTR:
2823 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2824 	case KVM_HAS_DEVICE_ATTR:
2825 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2826 	default:
2827 		if (dev->ops->ioctl)
2828 			return dev->ops->ioctl(dev, ioctl, arg);
2829 
2830 		return -ENOTTY;
2831 	}
2832 }
2833 
2834 static int kvm_device_release(struct inode *inode, struct file *filp)
2835 {
2836 	struct kvm_device *dev = filp->private_data;
2837 	struct kvm *kvm = dev->kvm;
2838 
2839 	kvm_put_kvm(kvm);
2840 	return 0;
2841 }
2842 
2843 static const struct file_operations kvm_device_fops = {
2844 	.unlocked_ioctl = kvm_device_ioctl,
2845 #ifdef CONFIG_KVM_COMPAT
2846 	.compat_ioctl = kvm_device_ioctl,
2847 #endif
2848 	.release = kvm_device_release,
2849 };
2850 
2851 struct kvm_device *kvm_device_from_filp(struct file *filp)
2852 {
2853 	if (filp->f_op != &kvm_device_fops)
2854 		return NULL;
2855 
2856 	return filp->private_data;
2857 }
2858 
2859 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2860 #ifdef CONFIG_KVM_MPIC
2861 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
2862 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
2863 #endif
2864 };
2865 
2866 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2867 {
2868 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
2869 		return -ENOSPC;
2870 
2871 	if (kvm_device_ops_table[type] != NULL)
2872 		return -EEXIST;
2873 
2874 	kvm_device_ops_table[type] = ops;
2875 	return 0;
2876 }
2877 
2878 void kvm_unregister_device_ops(u32 type)
2879 {
2880 	if (kvm_device_ops_table[type] != NULL)
2881 		kvm_device_ops_table[type] = NULL;
2882 }
2883 
2884 static int kvm_ioctl_create_device(struct kvm *kvm,
2885 				   struct kvm_create_device *cd)
2886 {
2887 	struct kvm_device_ops *ops = NULL;
2888 	struct kvm_device *dev;
2889 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2890 	int ret;
2891 
2892 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2893 		return -ENODEV;
2894 
2895 	ops = kvm_device_ops_table[cd->type];
2896 	if (ops == NULL)
2897 		return -ENODEV;
2898 
2899 	if (test)
2900 		return 0;
2901 
2902 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2903 	if (!dev)
2904 		return -ENOMEM;
2905 
2906 	dev->ops = ops;
2907 	dev->kvm = kvm;
2908 
2909 	mutex_lock(&kvm->lock);
2910 	ret = ops->create(dev, cd->type);
2911 	if (ret < 0) {
2912 		mutex_unlock(&kvm->lock);
2913 		kfree(dev);
2914 		return ret;
2915 	}
2916 	list_add(&dev->vm_node, &kvm->devices);
2917 	mutex_unlock(&kvm->lock);
2918 
2919 	if (ops->init)
2920 		ops->init(dev);
2921 
2922 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2923 	if (ret < 0) {
2924 		mutex_lock(&kvm->lock);
2925 		list_del(&dev->vm_node);
2926 		mutex_unlock(&kvm->lock);
2927 		ops->destroy(dev);
2928 		return ret;
2929 	}
2930 
2931 	kvm_get_kvm(kvm);
2932 	cd->fd = ret;
2933 	return 0;
2934 }
2935 
2936 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2937 {
2938 	switch (arg) {
2939 	case KVM_CAP_USER_MEMORY:
2940 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2941 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2942 	case KVM_CAP_INTERNAL_ERROR_DATA:
2943 #ifdef CONFIG_HAVE_KVM_MSI
2944 	case KVM_CAP_SIGNAL_MSI:
2945 #endif
2946 #ifdef CONFIG_HAVE_KVM_IRQFD
2947 	case KVM_CAP_IRQFD:
2948 	case KVM_CAP_IRQFD_RESAMPLE:
2949 #endif
2950 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2951 	case KVM_CAP_CHECK_EXTENSION_VM:
2952 		return 1;
2953 #ifdef CONFIG_KVM_MMIO
2954 	case KVM_CAP_COALESCED_MMIO:
2955 		return KVM_COALESCED_MMIO_PAGE_OFFSET;
2956 #endif
2957 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2958 	case KVM_CAP_IRQ_ROUTING:
2959 		return KVM_MAX_IRQ_ROUTES;
2960 #endif
2961 #if KVM_ADDRESS_SPACE_NUM > 1
2962 	case KVM_CAP_MULTI_ADDRESS_SPACE:
2963 		return KVM_ADDRESS_SPACE_NUM;
2964 #endif
2965 	case KVM_CAP_MAX_VCPU_ID:
2966 		return KVM_MAX_VCPU_ID;
2967 	default:
2968 		break;
2969 	}
2970 	return kvm_vm_ioctl_check_extension(kvm, arg);
2971 }
2972 
2973 static long kvm_vm_ioctl(struct file *filp,
2974 			   unsigned int ioctl, unsigned long arg)
2975 {
2976 	struct kvm *kvm = filp->private_data;
2977 	void __user *argp = (void __user *)arg;
2978 	int r;
2979 
2980 	if (kvm->mm != current->mm)
2981 		return -EIO;
2982 	switch (ioctl) {
2983 	case KVM_CREATE_VCPU:
2984 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2985 		break;
2986 	case KVM_SET_USER_MEMORY_REGION: {
2987 		struct kvm_userspace_memory_region kvm_userspace_mem;
2988 
2989 		r = -EFAULT;
2990 		if (copy_from_user(&kvm_userspace_mem, argp,
2991 						sizeof(kvm_userspace_mem)))
2992 			goto out;
2993 
2994 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2995 		break;
2996 	}
2997 	case KVM_GET_DIRTY_LOG: {
2998 		struct kvm_dirty_log log;
2999 
3000 		r = -EFAULT;
3001 		if (copy_from_user(&log, argp, sizeof(log)))
3002 			goto out;
3003 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3004 		break;
3005 	}
3006 #ifdef CONFIG_KVM_MMIO
3007 	case KVM_REGISTER_COALESCED_MMIO: {
3008 		struct kvm_coalesced_mmio_zone zone;
3009 
3010 		r = -EFAULT;
3011 		if (copy_from_user(&zone, argp, sizeof(zone)))
3012 			goto out;
3013 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
3014 		break;
3015 	}
3016 	case KVM_UNREGISTER_COALESCED_MMIO: {
3017 		struct kvm_coalesced_mmio_zone zone;
3018 
3019 		r = -EFAULT;
3020 		if (copy_from_user(&zone, argp, sizeof(zone)))
3021 			goto out;
3022 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
3023 		break;
3024 	}
3025 #endif
3026 	case KVM_IRQFD: {
3027 		struct kvm_irqfd data;
3028 
3029 		r = -EFAULT;
3030 		if (copy_from_user(&data, argp, sizeof(data)))
3031 			goto out;
3032 		r = kvm_irqfd(kvm, &data);
3033 		break;
3034 	}
3035 	case KVM_IOEVENTFD: {
3036 		struct kvm_ioeventfd data;
3037 
3038 		r = -EFAULT;
3039 		if (copy_from_user(&data, argp, sizeof(data)))
3040 			goto out;
3041 		r = kvm_ioeventfd(kvm, &data);
3042 		break;
3043 	}
3044 #ifdef CONFIG_HAVE_KVM_MSI
3045 	case KVM_SIGNAL_MSI: {
3046 		struct kvm_msi msi;
3047 
3048 		r = -EFAULT;
3049 		if (copy_from_user(&msi, argp, sizeof(msi)))
3050 			goto out;
3051 		r = kvm_send_userspace_msi(kvm, &msi);
3052 		break;
3053 	}
3054 #endif
3055 #ifdef __KVM_HAVE_IRQ_LINE
3056 	case KVM_IRQ_LINE_STATUS:
3057 	case KVM_IRQ_LINE: {
3058 		struct kvm_irq_level irq_event;
3059 
3060 		r = -EFAULT;
3061 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
3062 			goto out;
3063 
3064 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
3065 					ioctl == KVM_IRQ_LINE_STATUS);
3066 		if (r)
3067 			goto out;
3068 
3069 		r = -EFAULT;
3070 		if (ioctl == KVM_IRQ_LINE_STATUS) {
3071 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
3072 				goto out;
3073 		}
3074 
3075 		r = 0;
3076 		break;
3077 	}
3078 #endif
3079 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
3080 	case KVM_SET_GSI_ROUTING: {
3081 		struct kvm_irq_routing routing;
3082 		struct kvm_irq_routing __user *urouting;
3083 		struct kvm_irq_routing_entry *entries = NULL;
3084 
3085 		r = -EFAULT;
3086 		if (copy_from_user(&routing, argp, sizeof(routing)))
3087 			goto out;
3088 		r = -EINVAL;
3089 		if (!kvm_arch_can_set_irq_routing(kvm))
3090 			goto out;
3091 		if (routing.nr > KVM_MAX_IRQ_ROUTES)
3092 			goto out;
3093 		if (routing.flags)
3094 			goto out;
3095 		if (routing.nr) {
3096 			r = -ENOMEM;
3097 			entries = vmalloc(routing.nr * sizeof(*entries));
3098 			if (!entries)
3099 				goto out;
3100 			r = -EFAULT;
3101 			urouting = argp;
3102 			if (copy_from_user(entries, urouting->entries,
3103 					   routing.nr * sizeof(*entries)))
3104 				goto out_free_irq_routing;
3105 		}
3106 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
3107 					routing.flags);
3108 out_free_irq_routing:
3109 		vfree(entries);
3110 		break;
3111 	}
3112 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3113 	case KVM_CREATE_DEVICE: {
3114 		struct kvm_create_device cd;
3115 
3116 		r = -EFAULT;
3117 		if (copy_from_user(&cd, argp, sizeof(cd)))
3118 			goto out;
3119 
3120 		r = kvm_ioctl_create_device(kvm, &cd);
3121 		if (r)
3122 			goto out;
3123 
3124 		r = -EFAULT;
3125 		if (copy_to_user(argp, &cd, sizeof(cd)))
3126 			goto out;
3127 
3128 		r = 0;
3129 		break;
3130 	}
3131 	case KVM_CHECK_EXTENSION:
3132 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
3133 		break;
3134 	default:
3135 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
3136 	}
3137 out:
3138 	return r;
3139 }
3140 
3141 #ifdef CONFIG_KVM_COMPAT
3142 struct compat_kvm_dirty_log {
3143 	__u32 slot;
3144 	__u32 padding1;
3145 	union {
3146 		compat_uptr_t dirty_bitmap; /* one bit per page */
3147 		__u64 padding2;
3148 	};
3149 };
3150 
3151 static long kvm_vm_compat_ioctl(struct file *filp,
3152 			   unsigned int ioctl, unsigned long arg)
3153 {
3154 	struct kvm *kvm = filp->private_data;
3155 	int r;
3156 
3157 	if (kvm->mm != current->mm)
3158 		return -EIO;
3159 	switch (ioctl) {
3160 	case KVM_GET_DIRTY_LOG: {
3161 		struct compat_kvm_dirty_log compat_log;
3162 		struct kvm_dirty_log log;
3163 
3164 		if (copy_from_user(&compat_log, (void __user *)arg,
3165 				   sizeof(compat_log)))
3166 			return -EFAULT;
3167 		log.slot	 = compat_log.slot;
3168 		log.padding1	 = compat_log.padding1;
3169 		log.padding2	 = compat_log.padding2;
3170 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
3171 
3172 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
3173 		break;
3174 	}
3175 	default:
3176 		r = kvm_vm_ioctl(filp, ioctl, arg);
3177 	}
3178 	return r;
3179 }
3180 #endif
3181 
3182 static struct file_operations kvm_vm_fops = {
3183 	.release        = kvm_vm_release,
3184 	.unlocked_ioctl = kvm_vm_ioctl,
3185 #ifdef CONFIG_KVM_COMPAT
3186 	.compat_ioctl   = kvm_vm_compat_ioctl,
3187 #endif
3188 	.llseek		= noop_llseek,
3189 };
3190 
3191 static int kvm_dev_ioctl_create_vm(unsigned long type)
3192 {
3193 	int r;
3194 	struct kvm *kvm;
3195 	struct file *file;
3196 
3197 	kvm = kvm_create_vm(type);
3198 	if (IS_ERR(kvm))
3199 		return PTR_ERR(kvm);
3200 #ifdef CONFIG_KVM_MMIO
3201 	r = kvm_coalesced_mmio_init(kvm);
3202 	if (r < 0) {
3203 		kvm_put_kvm(kvm);
3204 		return r;
3205 	}
3206 #endif
3207 	r = get_unused_fd_flags(O_CLOEXEC);
3208 	if (r < 0) {
3209 		kvm_put_kvm(kvm);
3210 		return r;
3211 	}
3212 	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
3213 	if (IS_ERR(file)) {
3214 		put_unused_fd(r);
3215 		kvm_put_kvm(kvm);
3216 		return PTR_ERR(file);
3217 	}
3218 
3219 	/*
3220 	 * Don't call kvm_put_kvm anymore at this point; file->f_op is
3221 	 * already set, with ->release() being kvm_vm_release().  In error
3222 	 * cases it will be called by the final fput(file) and will take
3223 	 * care of doing kvm_put_kvm(kvm).
3224 	 */
3225 	if (kvm_create_vm_debugfs(kvm, r) < 0) {
3226 		put_unused_fd(r);
3227 		fput(file);
3228 		return -ENOMEM;
3229 	}
3230 	kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
3231 
3232 	fd_install(r, file);
3233 	return r;
3234 }
3235 
3236 static long kvm_dev_ioctl(struct file *filp,
3237 			  unsigned int ioctl, unsigned long arg)
3238 {
3239 	long r = -EINVAL;
3240 
3241 	switch (ioctl) {
3242 	case KVM_GET_API_VERSION:
3243 		if (arg)
3244 			goto out;
3245 		r = KVM_API_VERSION;
3246 		break;
3247 	case KVM_CREATE_VM:
3248 		r = kvm_dev_ioctl_create_vm(arg);
3249 		break;
3250 	case KVM_CHECK_EXTENSION:
3251 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3252 		break;
3253 	case KVM_GET_VCPU_MMAP_SIZE:
3254 		if (arg)
3255 			goto out;
3256 		r = PAGE_SIZE;     /* struct kvm_run */
3257 #ifdef CONFIG_X86
3258 		r += PAGE_SIZE;    /* pio data page */
3259 #endif
3260 #ifdef CONFIG_KVM_MMIO
3261 		r += PAGE_SIZE;    /* coalesced mmio ring page */
3262 #endif
3263 		break;
3264 	case KVM_TRACE_ENABLE:
3265 	case KVM_TRACE_PAUSE:
3266 	case KVM_TRACE_DISABLE:
3267 		r = -EOPNOTSUPP;
3268 		break;
3269 	default:
3270 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
3271 	}
3272 out:
3273 	return r;
3274 }
3275 
3276 static struct file_operations kvm_chardev_ops = {
3277 	.unlocked_ioctl = kvm_dev_ioctl,
3278 	.compat_ioctl   = kvm_dev_ioctl,
3279 	.llseek		= noop_llseek,
3280 };
3281 
3282 static struct miscdevice kvm_dev = {
3283 	KVM_MINOR,
3284 	"kvm",
3285 	&kvm_chardev_ops,
3286 };
3287 
3288 static void hardware_enable_nolock(void *junk)
3289 {
3290 	int cpu = raw_smp_processor_id();
3291 	int r;
3292 
3293 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3294 		return;
3295 
3296 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
3297 
3298 	r = kvm_arch_hardware_enable();
3299 
3300 	if (r) {
3301 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3302 		atomic_inc(&hardware_enable_failed);
3303 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3304 	}
3305 }
3306 
3307 static int kvm_starting_cpu(unsigned int cpu)
3308 {
3309 	raw_spin_lock(&kvm_count_lock);
3310 	if (kvm_usage_count)
3311 		hardware_enable_nolock(NULL);
3312 	raw_spin_unlock(&kvm_count_lock);
3313 	return 0;
3314 }
3315 
3316 static void hardware_disable_nolock(void *junk)
3317 {
3318 	int cpu = raw_smp_processor_id();
3319 
3320 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3321 		return;
3322 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3323 	kvm_arch_hardware_disable();
3324 }
3325 
3326 static int kvm_dying_cpu(unsigned int cpu)
3327 {
3328 	raw_spin_lock(&kvm_count_lock);
3329 	if (kvm_usage_count)
3330 		hardware_disable_nolock(NULL);
3331 	raw_spin_unlock(&kvm_count_lock);
3332 	return 0;
3333 }
3334 
3335 static void hardware_disable_all_nolock(void)
3336 {
3337 	BUG_ON(!kvm_usage_count);
3338 
3339 	kvm_usage_count--;
3340 	if (!kvm_usage_count)
3341 		on_each_cpu(hardware_disable_nolock, NULL, 1);
3342 }
3343 
3344 static void hardware_disable_all(void)
3345 {
3346 	raw_spin_lock(&kvm_count_lock);
3347 	hardware_disable_all_nolock();
3348 	raw_spin_unlock(&kvm_count_lock);
3349 }
3350 
3351 static int hardware_enable_all(void)
3352 {
3353 	int r = 0;
3354 
3355 	raw_spin_lock(&kvm_count_lock);
3356 
3357 	kvm_usage_count++;
3358 	if (kvm_usage_count == 1) {
3359 		atomic_set(&hardware_enable_failed, 0);
3360 		on_each_cpu(hardware_enable_nolock, NULL, 1);
3361 
3362 		if (atomic_read(&hardware_enable_failed)) {
3363 			hardware_disable_all_nolock();
3364 			r = -EBUSY;
3365 		}
3366 	}
3367 
3368 	raw_spin_unlock(&kvm_count_lock);
3369 
3370 	return r;
3371 }
3372 
3373 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3374 		      void *v)
3375 {
3376 	/*
3377 	 * Some (well, at least mine) BIOSes hang on reboot if
3378 	 * in vmx root mode.
3379 	 *
3380 	 * And Intel TXT required VMX off for all cpu when system shutdown.
3381 	 */
3382 	pr_info("kvm: exiting hardware virtualization\n");
3383 	kvm_rebooting = true;
3384 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3385 	return NOTIFY_OK;
3386 }
3387 
3388 static struct notifier_block kvm_reboot_notifier = {
3389 	.notifier_call = kvm_reboot,
3390 	.priority = 0,
3391 };
3392 
3393 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3394 {
3395 	int i;
3396 
3397 	for (i = 0; i < bus->dev_count; i++) {
3398 		struct kvm_io_device *pos = bus->range[i].dev;
3399 
3400 		kvm_iodevice_destructor(pos);
3401 	}
3402 	kfree(bus);
3403 }
3404 
3405 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3406 				 const struct kvm_io_range *r2)
3407 {
3408 	gpa_t addr1 = r1->addr;
3409 	gpa_t addr2 = r2->addr;
3410 
3411 	if (addr1 < addr2)
3412 		return -1;
3413 
3414 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
3415 	 * accept any overlapping write.  Any order is acceptable for
3416 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3417 	 * we process all of them.
3418 	 */
3419 	if (r2->len) {
3420 		addr1 += r1->len;
3421 		addr2 += r2->len;
3422 	}
3423 
3424 	if (addr1 > addr2)
3425 		return 1;
3426 
3427 	return 0;
3428 }
3429 
3430 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3431 {
3432 	return kvm_io_bus_cmp(p1, p2);
3433 }
3434 
3435 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3436 			  gpa_t addr, int len)
3437 {
3438 	bus->range[bus->dev_count++] = (struct kvm_io_range) {
3439 		.addr = addr,
3440 		.len = len,
3441 		.dev = dev,
3442 	};
3443 
3444 	sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3445 		kvm_io_bus_sort_cmp, NULL);
3446 
3447 	return 0;
3448 }
3449 
3450 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3451 			     gpa_t addr, int len)
3452 {
3453 	struct kvm_io_range *range, key;
3454 	int off;
3455 
3456 	key = (struct kvm_io_range) {
3457 		.addr = addr,
3458 		.len = len,
3459 	};
3460 
3461 	range = bsearch(&key, bus->range, bus->dev_count,
3462 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3463 	if (range == NULL)
3464 		return -ENOENT;
3465 
3466 	off = range - bus->range;
3467 
3468 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3469 		off--;
3470 
3471 	return off;
3472 }
3473 
3474 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3475 			      struct kvm_io_range *range, const void *val)
3476 {
3477 	int idx;
3478 
3479 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3480 	if (idx < 0)
3481 		return -EOPNOTSUPP;
3482 
3483 	while (idx < bus->dev_count &&
3484 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3485 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3486 					range->len, val))
3487 			return idx;
3488 		idx++;
3489 	}
3490 
3491 	return -EOPNOTSUPP;
3492 }
3493 
3494 /* kvm_io_bus_write - called under kvm->slots_lock */
3495 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3496 		     int len, const void *val)
3497 {
3498 	struct kvm_io_bus *bus;
3499 	struct kvm_io_range range;
3500 	int r;
3501 
3502 	range = (struct kvm_io_range) {
3503 		.addr = addr,
3504 		.len = len,
3505 	};
3506 
3507 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3508 	if (!bus)
3509 		return -ENOMEM;
3510 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
3511 	return r < 0 ? r : 0;
3512 }
3513 
3514 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3515 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3516 			    gpa_t addr, int len, const void *val, long cookie)
3517 {
3518 	struct kvm_io_bus *bus;
3519 	struct kvm_io_range range;
3520 
3521 	range = (struct kvm_io_range) {
3522 		.addr = addr,
3523 		.len = len,
3524 	};
3525 
3526 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3527 	if (!bus)
3528 		return -ENOMEM;
3529 
3530 	/* First try the device referenced by cookie. */
3531 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
3532 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3533 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3534 					val))
3535 			return cookie;
3536 
3537 	/*
3538 	 * cookie contained garbage; fall back to search and return the
3539 	 * correct cookie value.
3540 	 */
3541 	return __kvm_io_bus_write(vcpu, bus, &range, val);
3542 }
3543 
3544 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3545 			     struct kvm_io_range *range, void *val)
3546 {
3547 	int idx;
3548 
3549 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3550 	if (idx < 0)
3551 		return -EOPNOTSUPP;
3552 
3553 	while (idx < bus->dev_count &&
3554 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3555 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3556 				       range->len, val))
3557 			return idx;
3558 		idx++;
3559 	}
3560 
3561 	return -EOPNOTSUPP;
3562 }
3563 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3564 
3565 /* kvm_io_bus_read - called under kvm->slots_lock */
3566 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3567 		    int len, void *val)
3568 {
3569 	struct kvm_io_bus *bus;
3570 	struct kvm_io_range range;
3571 	int r;
3572 
3573 	range = (struct kvm_io_range) {
3574 		.addr = addr,
3575 		.len = len,
3576 	};
3577 
3578 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3579 	if (!bus)
3580 		return -ENOMEM;
3581 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
3582 	return r < 0 ? r : 0;
3583 }
3584 
3585 
3586 /* Caller must hold slots_lock. */
3587 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3588 			    int len, struct kvm_io_device *dev)
3589 {
3590 	struct kvm_io_bus *new_bus, *bus;
3591 
3592 	bus = kvm_get_bus(kvm, bus_idx);
3593 	if (!bus)
3594 		return -ENOMEM;
3595 
3596 	/* exclude ioeventfd which is limited by maximum fd */
3597 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3598 		return -ENOSPC;
3599 
3600 	new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3601 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3602 	if (!new_bus)
3603 		return -ENOMEM;
3604 	memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3605 	       sizeof(struct kvm_io_range)));
3606 	kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3607 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3608 	synchronize_srcu_expedited(&kvm->srcu);
3609 	kfree(bus);
3610 
3611 	return 0;
3612 }
3613 
3614 /* Caller must hold slots_lock. */
3615 void kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3616 			       struct kvm_io_device *dev)
3617 {
3618 	int i;
3619 	struct kvm_io_bus *new_bus, *bus;
3620 
3621 	bus = kvm_get_bus(kvm, bus_idx);
3622 	if (!bus)
3623 		return;
3624 
3625 	for (i = 0; i < bus->dev_count; i++)
3626 		if (bus->range[i].dev == dev) {
3627 			break;
3628 		}
3629 
3630 	if (i == bus->dev_count)
3631 		return;
3632 
3633 	new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3634 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3635 	if (!new_bus)  {
3636 		pr_err("kvm: failed to shrink bus, removing it completely\n");
3637 		goto broken;
3638 	}
3639 
3640 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3641 	new_bus->dev_count--;
3642 	memcpy(new_bus->range + i, bus->range + i + 1,
3643 	       (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3644 
3645 broken:
3646 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3647 	synchronize_srcu_expedited(&kvm->srcu);
3648 	kfree(bus);
3649 	return;
3650 }
3651 
3652 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3653 					 gpa_t addr)
3654 {
3655 	struct kvm_io_bus *bus;
3656 	int dev_idx, srcu_idx;
3657 	struct kvm_io_device *iodev = NULL;
3658 
3659 	srcu_idx = srcu_read_lock(&kvm->srcu);
3660 
3661 	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
3662 	if (!bus)
3663 		goto out_unlock;
3664 
3665 	dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
3666 	if (dev_idx < 0)
3667 		goto out_unlock;
3668 
3669 	iodev = bus->range[dev_idx].dev;
3670 
3671 out_unlock:
3672 	srcu_read_unlock(&kvm->srcu, srcu_idx);
3673 
3674 	return iodev;
3675 }
3676 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
3677 
3678 static int kvm_debugfs_open(struct inode *inode, struct file *file,
3679 			   int (*get)(void *, u64 *), int (*set)(void *, u64),
3680 			   const char *fmt)
3681 {
3682 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3683 					  inode->i_private;
3684 
3685 	/* The debugfs files are a reference to the kvm struct which
3686 	 * is still valid when kvm_destroy_vm is called.
3687 	 * To avoid the race between open and the removal of the debugfs
3688 	 * directory we test against the users count.
3689 	 */
3690 	if (!refcount_inc_not_zero(&stat_data->kvm->users_count))
3691 		return -ENOENT;
3692 
3693 	if (simple_attr_open(inode, file, get, set, fmt)) {
3694 		kvm_put_kvm(stat_data->kvm);
3695 		return -ENOMEM;
3696 	}
3697 
3698 	return 0;
3699 }
3700 
3701 static int kvm_debugfs_release(struct inode *inode, struct file *file)
3702 {
3703 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
3704 					  inode->i_private;
3705 
3706 	simple_attr_release(inode, file);
3707 	kvm_put_kvm(stat_data->kvm);
3708 
3709 	return 0;
3710 }
3711 
3712 static int vm_stat_get_per_vm(void *data, u64 *val)
3713 {
3714 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3715 
3716 	*val = *(ulong *)((void *)stat_data->kvm + stat_data->offset);
3717 
3718 	return 0;
3719 }
3720 
3721 static int vm_stat_clear_per_vm(void *data, u64 val)
3722 {
3723 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3724 
3725 	if (val)
3726 		return -EINVAL;
3727 
3728 	*(ulong *)((void *)stat_data->kvm + stat_data->offset) = 0;
3729 
3730 	return 0;
3731 }
3732 
3733 static int vm_stat_get_per_vm_open(struct inode *inode, struct file *file)
3734 {
3735 	__simple_attr_check_format("%llu\n", 0ull);
3736 	return kvm_debugfs_open(inode, file, vm_stat_get_per_vm,
3737 				vm_stat_clear_per_vm, "%llu\n");
3738 }
3739 
3740 static const struct file_operations vm_stat_get_per_vm_fops = {
3741 	.owner   = THIS_MODULE,
3742 	.open    = vm_stat_get_per_vm_open,
3743 	.release = kvm_debugfs_release,
3744 	.read    = simple_attr_read,
3745 	.write   = simple_attr_write,
3746 	.llseek  = no_llseek,
3747 };
3748 
3749 static int vcpu_stat_get_per_vm(void *data, u64 *val)
3750 {
3751 	int i;
3752 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3753 	struct kvm_vcpu *vcpu;
3754 
3755 	*val = 0;
3756 
3757 	kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3758 		*val += *(u64 *)((void *)vcpu + stat_data->offset);
3759 
3760 	return 0;
3761 }
3762 
3763 static int vcpu_stat_clear_per_vm(void *data, u64 val)
3764 {
3765 	int i;
3766 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
3767 	struct kvm_vcpu *vcpu;
3768 
3769 	if (val)
3770 		return -EINVAL;
3771 
3772 	kvm_for_each_vcpu(i, vcpu, stat_data->kvm)
3773 		*(u64 *)((void *)vcpu + stat_data->offset) = 0;
3774 
3775 	return 0;
3776 }
3777 
3778 static int vcpu_stat_get_per_vm_open(struct inode *inode, struct file *file)
3779 {
3780 	__simple_attr_check_format("%llu\n", 0ull);
3781 	return kvm_debugfs_open(inode, file, vcpu_stat_get_per_vm,
3782 				 vcpu_stat_clear_per_vm, "%llu\n");
3783 }
3784 
3785 static const struct file_operations vcpu_stat_get_per_vm_fops = {
3786 	.owner   = THIS_MODULE,
3787 	.open    = vcpu_stat_get_per_vm_open,
3788 	.release = kvm_debugfs_release,
3789 	.read    = simple_attr_read,
3790 	.write   = simple_attr_write,
3791 	.llseek  = no_llseek,
3792 };
3793 
3794 static const struct file_operations *stat_fops_per_vm[] = {
3795 	[KVM_STAT_VCPU] = &vcpu_stat_get_per_vm_fops,
3796 	[KVM_STAT_VM]   = &vm_stat_get_per_vm_fops,
3797 };
3798 
3799 static int vm_stat_get(void *_offset, u64 *val)
3800 {
3801 	unsigned offset = (long)_offset;
3802 	struct kvm *kvm;
3803 	struct kvm_stat_data stat_tmp = {.offset = offset};
3804 	u64 tmp_val;
3805 
3806 	*val = 0;
3807 	spin_lock(&kvm_lock);
3808 	list_for_each_entry(kvm, &vm_list, vm_list) {
3809 		stat_tmp.kvm = kvm;
3810 		vm_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3811 		*val += tmp_val;
3812 	}
3813 	spin_unlock(&kvm_lock);
3814 	return 0;
3815 }
3816 
3817 static int vm_stat_clear(void *_offset, u64 val)
3818 {
3819 	unsigned offset = (long)_offset;
3820 	struct kvm *kvm;
3821 	struct kvm_stat_data stat_tmp = {.offset = offset};
3822 
3823 	if (val)
3824 		return -EINVAL;
3825 
3826 	spin_lock(&kvm_lock);
3827 	list_for_each_entry(kvm, &vm_list, vm_list) {
3828 		stat_tmp.kvm = kvm;
3829 		vm_stat_clear_per_vm((void *)&stat_tmp, 0);
3830 	}
3831 	spin_unlock(&kvm_lock);
3832 
3833 	return 0;
3834 }
3835 
3836 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
3837 
3838 static int vcpu_stat_get(void *_offset, u64 *val)
3839 {
3840 	unsigned offset = (long)_offset;
3841 	struct kvm *kvm;
3842 	struct kvm_stat_data stat_tmp = {.offset = offset};
3843 	u64 tmp_val;
3844 
3845 	*val = 0;
3846 	spin_lock(&kvm_lock);
3847 	list_for_each_entry(kvm, &vm_list, vm_list) {
3848 		stat_tmp.kvm = kvm;
3849 		vcpu_stat_get_per_vm((void *)&stat_tmp, &tmp_val);
3850 		*val += tmp_val;
3851 	}
3852 	spin_unlock(&kvm_lock);
3853 	return 0;
3854 }
3855 
3856 static int vcpu_stat_clear(void *_offset, u64 val)
3857 {
3858 	unsigned offset = (long)_offset;
3859 	struct kvm *kvm;
3860 	struct kvm_stat_data stat_tmp = {.offset = offset};
3861 
3862 	if (val)
3863 		return -EINVAL;
3864 
3865 	spin_lock(&kvm_lock);
3866 	list_for_each_entry(kvm, &vm_list, vm_list) {
3867 		stat_tmp.kvm = kvm;
3868 		vcpu_stat_clear_per_vm((void *)&stat_tmp, 0);
3869 	}
3870 	spin_unlock(&kvm_lock);
3871 
3872 	return 0;
3873 }
3874 
3875 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
3876 			"%llu\n");
3877 
3878 static const struct file_operations *stat_fops[] = {
3879 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
3880 	[KVM_STAT_VM]   = &vm_stat_fops,
3881 };
3882 
3883 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
3884 {
3885 	struct kobj_uevent_env *env;
3886 	char *tmp, *pathbuf = NULL;
3887 	unsigned long long created, active;
3888 
3889 	if (!kvm_dev.this_device || !kvm)
3890 		return;
3891 
3892 	spin_lock(&kvm_lock);
3893 	if (type == KVM_EVENT_CREATE_VM) {
3894 		kvm_createvm_count++;
3895 		kvm_active_vms++;
3896 	} else if (type == KVM_EVENT_DESTROY_VM) {
3897 		kvm_active_vms--;
3898 	}
3899 	created = kvm_createvm_count;
3900 	active = kvm_active_vms;
3901 	spin_unlock(&kvm_lock);
3902 
3903 	env = kzalloc(sizeof(*env), GFP_KERNEL);
3904 	if (!env)
3905 		return;
3906 
3907 	add_uevent_var(env, "CREATED=%llu", created);
3908 	add_uevent_var(env, "COUNT=%llu", active);
3909 
3910 	if (type == KVM_EVENT_CREATE_VM)
3911 		add_uevent_var(env, "EVENT=create");
3912 	else if (type == KVM_EVENT_DESTROY_VM)
3913 		add_uevent_var(env, "EVENT=destroy");
3914 
3915 	if (kvm->debugfs_dentry) {
3916 		char p[ITOA_MAX_LEN];
3917 
3918 		snprintf(p, sizeof(p), "%s", kvm->debugfs_dentry->d_name.name);
3919 		tmp = strchrnul(p + 1, '-');
3920 		*tmp = '\0';
3921 		add_uevent_var(env, "PID=%s", p);
3922 		pathbuf = kmalloc(PATH_MAX, GFP_KERNEL);
3923 		if (pathbuf) {
3924 			/* sizeof counts the final '\0' */
3925 			int len = sizeof("STATS_PATH=") - 1;
3926 			const char *pvar = "STATS_PATH=";
3927 
3928 			tmp = dentry_path_raw(kvm->debugfs_dentry,
3929 					      pathbuf + len,
3930 					      PATH_MAX - len);
3931 			if (!IS_ERR(tmp)) {
3932 				memcpy(tmp - len, pvar, len);
3933 				env->envp[env->envp_idx++] = tmp - len;
3934 			}
3935 		}
3936 	}
3937 	/* no need for checks, since we are adding at most only 5 keys */
3938 	env->envp[env->envp_idx++] = NULL;
3939 	kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
3940 	kfree(env);
3941 	kfree(pathbuf);
3942 }
3943 
3944 static int kvm_init_debug(void)
3945 {
3946 	int r = -EEXIST;
3947 	struct kvm_stats_debugfs_item *p;
3948 
3949 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3950 	if (kvm_debugfs_dir == NULL)
3951 		goto out;
3952 
3953 	kvm_debugfs_num_entries = 0;
3954 	for (p = debugfs_entries; p->name; ++p, kvm_debugfs_num_entries++) {
3955 		if (!debugfs_create_file(p->name, 0644, kvm_debugfs_dir,
3956 					 (void *)(long)p->offset,
3957 					 stat_fops[p->kind]))
3958 			goto out_dir;
3959 	}
3960 
3961 	return 0;
3962 
3963 out_dir:
3964 	debugfs_remove_recursive(kvm_debugfs_dir);
3965 out:
3966 	return r;
3967 }
3968 
3969 static int kvm_suspend(void)
3970 {
3971 	if (kvm_usage_count)
3972 		hardware_disable_nolock(NULL);
3973 	return 0;
3974 }
3975 
3976 static void kvm_resume(void)
3977 {
3978 	if (kvm_usage_count) {
3979 		WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3980 		hardware_enable_nolock(NULL);
3981 	}
3982 }
3983 
3984 static struct syscore_ops kvm_syscore_ops = {
3985 	.suspend = kvm_suspend,
3986 	.resume = kvm_resume,
3987 };
3988 
3989 static inline
3990 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3991 {
3992 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
3993 }
3994 
3995 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3996 {
3997 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3998 
3999 	if (vcpu->preempted)
4000 		vcpu->preempted = false;
4001 
4002 	kvm_arch_sched_in(vcpu, cpu);
4003 
4004 	kvm_arch_vcpu_load(vcpu, cpu);
4005 }
4006 
4007 static void kvm_sched_out(struct preempt_notifier *pn,
4008 			  struct task_struct *next)
4009 {
4010 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
4011 
4012 	if (current->state == TASK_RUNNING)
4013 		vcpu->preempted = true;
4014 	kvm_arch_vcpu_put(vcpu);
4015 }
4016 
4017 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
4018 		  struct module *module)
4019 {
4020 	int r;
4021 	int cpu;
4022 
4023 	r = kvm_arch_init(opaque);
4024 	if (r)
4025 		goto out_fail;
4026 
4027 	/*
4028 	 * kvm_arch_init makes sure there's at most one caller
4029 	 * for architectures that support multiple implementations,
4030 	 * like intel and amd on x86.
4031 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
4032 	 * conflicts in case kvm is already setup for another implementation.
4033 	 */
4034 	r = kvm_irqfd_init();
4035 	if (r)
4036 		goto out_irqfd;
4037 
4038 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
4039 		r = -ENOMEM;
4040 		goto out_free_0;
4041 	}
4042 
4043 	r = kvm_arch_hardware_setup();
4044 	if (r < 0)
4045 		goto out_free_0a;
4046 
4047 	for_each_online_cpu(cpu) {
4048 		smp_call_function_single(cpu,
4049 				kvm_arch_check_processor_compat,
4050 				&r, 1);
4051 		if (r < 0)
4052 			goto out_free_1;
4053 	}
4054 
4055 	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
4056 				      kvm_starting_cpu, kvm_dying_cpu);
4057 	if (r)
4058 		goto out_free_2;
4059 	register_reboot_notifier(&kvm_reboot_notifier);
4060 
4061 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
4062 	if (!vcpu_align)
4063 		vcpu_align = __alignof__(struct kvm_vcpu);
4064 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
4065 					   0, NULL);
4066 	if (!kvm_vcpu_cache) {
4067 		r = -ENOMEM;
4068 		goto out_free_3;
4069 	}
4070 
4071 	r = kvm_async_pf_init();
4072 	if (r)
4073 		goto out_free;
4074 
4075 	kvm_chardev_ops.owner = module;
4076 	kvm_vm_fops.owner = module;
4077 	kvm_vcpu_fops.owner = module;
4078 
4079 	r = misc_register(&kvm_dev);
4080 	if (r) {
4081 		pr_err("kvm: misc device register failed\n");
4082 		goto out_unreg;
4083 	}
4084 
4085 	register_syscore_ops(&kvm_syscore_ops);
4086 
4087 	kvm_preempt_ops.sched_in = kvm_sched_in;
4088 	kvm_preempt_ops.sched_out = kvm_sched_out;
4089 
4090 	r = kvm_init_debug();
4091 	if (r) {
4092 		pr_err("kvm: create debugfs files failed\n");
4093 		goto out_undebugfs;
4094 	}
4095 
4096 	r = kvm_vfio_ops_init();
4097 	WARN_ON(r);
4098 
4099 	return 0;
4100 
4101 out_undebugfs:
4102 	unregister_syscore_ops(&kvm_syscore_ops);
4103 	misc_deregister(&kvm_dev);
4104 out_unreg:
4105 	kvm_async_pf_deinit();
4106 out_free:
4107 	kmem_cache_destroy(kvm_vcpu_cache);
4108 out_free_3:
4109 	unregister_reboot_notifier(&kvm_reboot_notifier);
4110 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4111 out_free_2:
4112 out_free_1:
4113 	kvm_arch_hardware_unsetup();
4114 out_free_0a:
4115 	free_cpumask_var(cpus_hardware_enabled);
4116 out_free_0:
4117 	kvm_irqfd_exit();
4118 out_irqfd:
4119 	kvm_arch_exit();
4120 out_fail:
4121 	return r;
4122 }
4123 EXPORT_SYMBOL_GPL(kvm_init);
4124 
4125 void kvm_exit(void)
4126 {
4127 	debugfs_remove_recursive(kvm_debugfs_dir);
4128 	misc_deregister(&kvm_dev);
4129 	kmem_cache_destroy(kvm_vcpu_cache);
4130 	kvm_async_pf_deinit();
4131 	unregister_syscore_ops(&kvm_syscore_ops);
4132 	unregister_reboot_notifier(&kvm_reboot_notifier);
4133 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
4134 	on_each_cpu(hardware_disable_nolock, NULL, 1);
4135 	kvm_arch_hardware_unsetup();
4136 	kvm_arch_exit();
4137 	kvm_irqfd_exit();
4138 	free_cpumask_var(cpus_hardware_enabled);
4139 	kvm_vfio_ops_exit();
4140 }
4141 EXPORT_SYMBOL_GPL(kvm_exit);
4142