xref: /openbmc/linux/virt/kvm/kvm_main.c (revision d9565bf4)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Kernel-based Virtual Machine driver for Linux
4  *
5  * This module enables machines with Intel VT-x extensions to run virtual
6  * machines without emulation or binary translation.
7  *
8  * Copyright (C) 2006 Qumranet, Inc.
9  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10  *
11  * Authors:
12  *   Avi Kivity   <avi@qumranet.com>
13  *   Yaniv Kamay  <yaniv@qumranet.com>
14  */
15 
16 #include <kvm/iodev.h>
17 
18 #include <linux/kvm_host.h>
19 #include <linux/kvm.h>
20 #include <linux/module.h>
21 #include <linux/errno.h>
22 #include <linux/percpu.h>
23 #include <linux/mm.h>
24 #include <linux/miscdevice.h>
25 #include <linux/vmalloc.h>
26 #include <linux/reboot.h>
27 #include <linux/debugfs.h>
28 #include <linux/highmem.h>
29 #include <linux/file.h>
30 #include <linux/syscore_ops.h>
31 #include <linux/cpu.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/sched/stat.h>
35 #include <linux/cpumask.h>
36 #include <linux/smp.h>
37 #include <linux/anon_inodes.h>
38 #include <linux/profile.h>
39 #include <linux/kvm_para.h>
40 #include <linux/pagemap.h>
41 #include <linux/mman.h>
42 #include <linux/swap.h>
43 #include <linux/bitops.h>
44 #include <linux/spinlock.h>
45 #include <linux/compat.h>
46 #include <linux/srcu.h>
47 #include <linux/hugetlb.h>
48 #include <linux/slab.h>
49 #include <linux/sort.h>
50 #include <linux/bsearch.h>
51 #include <linux/io.h>
52 #include <linux/lockdep.h>
53 #include <linux/kthread.h>
54 #include <linux/suspend.h>
55 
56 #include <asm/processor.h>
57 #include <asm/ioctl.h>
58 #include <linux/uaccess.h>
59 
60 #include "coalesced_mmio.h"
61 #include "async_pf.h"
62 #include "kvm_mm.h"
63 #include "vfio.h"
64 
65 #define CREATE_TRACE_POINTS
66 #include <trace/events/kvm.h>
67 
68 #include <linux/kvm_dirty_ring.h>
69 
70 /* Worst case buffer size needed for holding an integer. */
71 #define ITOA_MAX_LEN 12
72 
73 MODULE_AUTHOR("Qumranet");
74 MODULE_LICENSE("GPL");
75 
76 /* Architectures should define their poll value according to the halt latency */
77 unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
78 module_param(halt_poll_ns, uint, 0644);
79 EXPORT_SYMBOL_GPL(halt_poll_ns);
80 
81 /* Default doubles per-vcpu halt_poll_ns. */
82 unsigned int halt_poll_ns_grow = 2;
83 module_param(halt_poll_ns_grow, uint, 0644);
84 EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
85 
86 /* The start value to grow halt_poll_ns from */
87 unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
88 module_param(halt_poll_ns_grow_start, uint, 0644);
89 EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
90 
91 /* Default resets per-vcpu halt_poll_ns . */
92 unsigned int halt_poll_ns_shrink;
93 module_param(halt_poll_ns_shrink, uint, 0644);
94 EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
95 
96 /*
97  * Ordering of locks:
98  *
99  *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
100  */
101 
102 DEFINE_MUTEX(kvm_lock);
103 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
104 LIST_HEAD(vm_list);
105 
106 static cpumask_var_t cpus_hardware_enabled;
107 static int kvm_usage_count;
108 static atomic_t hardware_enable_failed;
109 
110 static struct kmem_cache *kvm_vcpu_cache;
111 
112 static __read_mostly struct preempt_ops kvm_preempt_ops;
113 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
114 
115 struct dentry *kvm_debugfs_dir;
116 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
117 
118 static const struct file_operations stat_fops_per_vm;
119 
120 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
121 			   unsigned long arg);
122 #ifdef CONFIG_KVM_COMPAT
123 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
124 				  unsigned long arg);
125 #define KVM_COMPAT(c)	.compat_ioctl	= (c)
126 #else
127 /*
128  * For architectures that don't implement a compat infrastructure,
129  * adopt a double line of defense:
130  * - Prevent a compat task from opening /dev/kvm
131  * - If the open has been done by a 64bit task, and the KVM fd
132  *   passed to a compat task, let the ioctls fail.
133  */
134 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
135 				unsigned long arg) { return -EINVAL; }
136 
137 static int kvm_no_compat_open(struct inode *inode, struct file *file)
138 {
139 	return is_compat_task() ? -ENODEV : 0;
140 }
141 #define KVM_COMPAT(c)	.compat_ioctl	= kvm_no_compat_ioctl,	\
142 			.open		= kvm_no_compat_open
143 #endif
144 static int hardware_enable_all(void);
145 static void hardware_disable_all(void);
146 
147 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
148 
149 __visible bool kvm_rebooting;
150 EXPORT_SYMBOL_GPL(kvm_rebooting);
151 
152 #define KVM_EVENT_CREATE_VM 0
153 #define KVM_EVENT_DESTROY_VM 1
154 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
155 static unsigned long long kvm_createvm_count;
156 static unsigned long long kvm_active_vms;
157 
158 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
159 
160 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
161 						   unsigned long start, unsigned long end)
162 {
163 }
164 
165 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
166 {
167 	/*
168 	 * The metadata used by is_zone_device_page() to determine whether or
169 	 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
170 	 * the device has been pinned, e.g. by get_user_pages().  WARN if the
171 	 * page_count() is zero to help detect bad usage of this helper.
172 	 */
173 	if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
174 		return false;
175 
176 	return is_zone_device_page(pfn_to_page(pfn));
177 }
178 
179 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
180 {
181 	/*
182 	 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
183 	 * perspective they are "normal" pages, albeit with slightly different
184 	 * usage rules.
185 	 */
186 	if (pfn_valid(pfn))
187 		return PageReserved(pfn_to_page(pfn)) &&
188 		       !is_zero_pfn(pfn) &&
189 		       !kvm_is_zone_device_pfn(pfn);
190 
191 	return true;
192 }
193 
194 /*
195  * Switches to specified vcpu, until a matching vcpu_put()
196  */
197 void vcpu_load(struct kvm_vcpu *vcpu)
198 {
199 	int cpu = get_cpu();
200 
201 	__this_cpu_write(kvm_running_vcpu, vcpu);
202 	preempt_notifier_register(&vcpu->preempt_notifier);
203 	kvm_arch_vcpu_load(vcpu, cpu);
204 	put_cpu();
205 }
206 EXPORT_SYMBOL_GPL(vcpu_load);
207 
208 void vcpu_put(struct kvm_vcpu *vcpu)
209 {
210 	preempt_disable();
211 	kvm_arch_vcpu_put(vcpu);
212 	preempt_notifier_unregister(&vcpu->preempt_notifier);
213 	__this_cpu_write(kvm_running_vcpu, NULL);
214 	preempt_enable();
215 }
216 EXPORT_SYMBOL_GPL(vcpu_put);
217 
218 /* TODO: merge with kvm_arch_vcpu_should_kick */
219 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
220 {
221 	int mode = kvm_vcpu_exiting_guest_mode(vcpu);
222 
223 	/*
224 	 * We need to wait for the VCPU to reenable interrupts and get out of
225 	 * READING_SHADOW_PAGE_TABLES mode.
226 	 */
227 	if (req & KVM_REQUEST_WAIT)
228 		return mode != OUTSIDE_GUEST_MODE;
229 
230 	/*
231 	 * Need to kick a running VCPU, but otherwise there is nothing to do.
232 	 */
233 	return mode == IN_GUEST_MODE;
234 }
235 
236 static void ack_flush(void *_completed)
237 {
238 }
239 
240 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
241 {
242 	if (cpumask_empty(cpus))
243 		return false;
244 
245 	smp_call_function_many(cpus, ack_flush, NULL, wait);
246 	return true;
247 }
248 
249 static void kvm_make_vcpu_request(struct kvm *kvm, struct kvm_vcpu *vcpu,
250 				  unsigned int req, struct cpumask *tmp,
251 				  int current_cpu)
252 {
253 	int cpu;
254 
255 	kvm_make_request(req, vcpu);
256 
257 	if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
258 		return;
259 
260 	/*
261 	 * Note, the vCPU could get migrated to a different pCPU at any point
262 	 * after kvm_request_needs_ipi(), which could result in sending an IPI
263 	 * to the previous pCPU.  But, that's OK because the purpose of the IPI
264 	 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
265 	 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
266 	 * after this point is also OK, as the requirement is only that KVM wait
267 	 * for vCPUs that were reading SPTEs _before_ any changes were
268 	 * finalized. See kvm_vcpu_kick() for more details on handling requests.
269 	 */
270 	if (kvm_request_needs_ipi(vcpu, req)) {
271 		cpu = READ_ONCE(vcpu->cpu);
272 		if (cpu != -1 && cpu != current_cpu)
273 			__cpumask_set_cpu(cpu, tmp);
274 	}
275 }
276 
277 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
278 				 unsigned long *vcpu_bitmap)
279 {
280 	struct kvm_vcpu *vcpu;
281 	struct cpumask *cpus;
282 	int i, me;
283 	bool called;
284 
285 	me = get_cpu();
286 
287 	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
288 	cpumask_clear(cpus);
289 
290 	for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
291 		vcpu = kvm_get_vcpu(kvm, i);
292 		if (!vcpu)
293 			continue;
294 		kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
295 	}
296 
297 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
298 	put_cpu();
299 
300 	return called;
301 }
302 
303 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
304 				      struct kvm_vcpu *except)
305 {
306 	struct kvm_vcpu *vcpu;
307 	struct cpumask *cpus;
308 	unsigned long i;
309 	bool called;
310 	int me;
311 
312 	me = get_cpu();
313 
314 	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
315 	cpumask_clear(cpus);
316 
317 	kvm_for_each_vcpu(i, vcpu, kvm) {
318 		if (vcpu == except)
319 			continue;
320 		kvm_make_vcpu_request(kvm, vcpu, req, cpus, me);
321 	}
322 
323 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
324 	put_cpu();
325 
326 	return called;
327 }
328 
329 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
330 {
331 	return kvm_make_all_cpus_request_except(kvm, req, NULL);
332 }
333 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
334 
335 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
336 void kvm_flush_remote_tlbs(struct kvm *kvm)
337 {
338 	++kvm->stat.generic.remote_tlb_flush_requests;
339 
340 	/*
341 	 * We want to publish modifications to the page tables before reading
342 	 * mode. Pairs with a memory barrier in arch-specific code.
343 	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
344 	 * and smp_mb in walk_shadow_page_lockless_begin/end.
345 	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
346 	 *
347 	 * There is already an smp_mb__after_atomic() before
348 	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
349 	 * barrier here.
350 	 */
351 	if (!kvm_arch_flush_remote_tlb(kvm)
352 	    || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
353 		++kvm->stat.generic.remote_tlb_flush;
354 }
355 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
356 #endif
357 
358 void kvm_reload_remote_mmus(struct kvm *kvm)
359 {
360 	kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
361 }
362 
363 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
364 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
365 					       gfp_t gfp_flags)
366 {
367 	gfp_flags |= mc->gfp_zero;
368 
369 	if (mc->kmem_cache)
370 		return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
371 	else
372 		return (void *)__get_free_page(gfp_flags);
373 }
374 
375 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
376 {
377 	void *obj;
378 
379 	if (mc->nobjs >= min)
380 		return 0;
381 	while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
382 		obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
383 		if (!obj)
384 			return mc->nobjs >= min ? 0 : -ENOMEM;
385 		mc->objects[mc->nobjs++] = obj;
386 	}
387 	return 0;
388 }
389 
390 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
391 {
392 	return mc->nobjs;
393 }
394 
395 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
396 {
397 	while (mc->nobjs) {
398 		if (mc->kmem_cache)
399 			kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
400 		else
401 			free_page((unsigned long)mc->objects[--mc->nobjs]);
402 	}
403 }
404 
405 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
406 {
407 	void *p;
408 
409 	if (WARN_ON(!mc->nobjs))
410 		p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
411 	else
412 		p = mc->objects[--mc->nobjs];
413 	BUG_ON(!p);
414 	return p;
415 }
416 #endif
417 
418 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
419 {
420 	mutex_init(&vcpu->mutex);
421 	vcpu->cpu = -1;
422 	vcpu->kvm = kvm;
423 	vcpu->vcpu_id = id;
424 	vcpu->pid = NULL;
425 #ifndef __KVM_HAVE_ARCH_WQP
426 	rcuwait_init(&vcpu->wait);
427 #endif
428 	kvm_async_pf_vcpu_init(vcpu);
429 
430 	kvm_vcpu_set_in_spin_loop(vcpu, false);
431 	kvm_vcpu_set_dy_eligible(vcpu, false);
432 	vcpu->preempted = false;
433 	vcpu->ready = false;
434 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
435 	vcpu->last_used_slot = NULL;
436 }
437 
438 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
439 {
440 	kvm_dirty_ring_free(&vcpu->dirty_ring);
441 	kvm_arch_vcpu_destroy(vcpu);
442 
443 	/*
444 	 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
445 	 * the vcpu->pid pointer, and at destruction time all file descriptors
446 	 * are already gone.
447 	 */
448 	put_pid(rcu_dereference_protected(vcpu->pid, 1));
449 
450 	free_page((unsigned long)vcpu->run);
451 	kmem_cache_free(kvm_vcpu_cache, vcpu);
452 }
453 
454 void kvm_destroy_vcpus(struct kvm *kvm)
455 {
456 	unsigned long i;
457 	struct kvm_vcpu *vcpu;
458 
459 	kvm_for_each_vcpu(i, vcpu, kvm) {
460 		kvm_vcpu_destroy(vcpu);
461 		xa_erase(&kvm->vcpu_array, i);
462 	}
463 
464 	atomic_set(&kvm->online_vcpus, 0);
465 }
466 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
467 
468 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
469 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
470 {
471 	return container_of(mn, struct kvm, mmu_notifier);
472 }
473 
474 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
475 					      struct mm_struct *mm,
476 					      unsigned long start, unsigned long end)
477 {
478 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
479 	int idx;
480 
481 	idx = srcu_read_lock(&kvm->srcu);
482 	kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
483 	srcu_read_unlock(&kvm->srcu, idx);
484 }
485 
486 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
487 
488 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
489 			     unsigned long end);
490 
491 struct kvm_hva_range {
492 	unsigned long start;
493 	unsigned long end;
494 	pte_t pte;
495 	hva_handler_t handler;
496 	on_lock_fn_t on_lock;
497 	bool flush_on_ret;
498 	bool may_block;
499 };
500 
501 /*
502  * Use a dedicated stub instead of NULL to indicate that there is no callback
503  * function/handler.  The compiler technically can't guarantee that a real
504  * function will have a non-zero address, and so it will generate code to
505  * check for !NULL, whereas comparing against a stub will be elided at compile
506  * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
507  */
508 static void kvm_null_fn(void)
509 {
510 
511 }
512 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
513 
514 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
515 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last)	     \
516 	for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
517 	     node;							     \
518 	     node = interval_tree_iter_next(node, start, last))	     \
519 
520 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
521 						  const struct kvm_hva_range *range)
522 {
523 	bool ret = false, locked = false;
524 	struct kvm_gfn_range gfn_range;
525 	struct kvm_memory_slot *slot;
526 	struct kvm_memslots *slots;
527 	int i, idx;
528 
529 	if (WARN_ON_ONCE(range->end <= range->start))
530 		return 0;
531 
532 	/* A null handler is allowed if and only if on_lock() is provided. */
533 	if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
534 			 IS_KVM_NULL_FN(range->handler)))
535 		return 0;
536 
537 	idx = srcu_read_lock(&kvm->srcu);
538 
539 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
540 		struct interval_tree_node *node;
541 
542 		slots = __kvm_memslots(kvm, i);
543 		kvm_for_each_memslot_in_hva_range(node, slots,
544 						  range->start, range->end - 1) {
545 			unsigned long hva_start, hva_end;
546 
547 			slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
548 			hva_start = max(range->start, slot->userspace_addr);
549 			hva_end = min(range->end, slot->userspace_addr +
550 						  (slot->npages << PAGE_SHIFT));
551 
552 			/*
553 			 * To optimize for the likely case where the address
554 			 * range is covered by zero or one memslots, don't
555 			 * bother making these conditional (to avoid writes on
556 			 * the second or later invocation of the handler).
557 			 */
558 			gfn_range.pte = range->pte;
559 			gfn_range.may_block = range->may_block;
560 
561 			/*
562 			 * {gfn(page) | page intersects with [hva_start, hva_end)} =
563 			 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
564 			 */
565 			gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
566 			gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
567 			gfn_range.slot = slot;
568 
569 			if (!locked) {
570 				locked = true;
571 				KVM_MMU_LOCK(kvm);
572 				if (!IS_KVM_NULL_FN(range->on_lock))
573 					range->on_lock(kvm, range->start, range->end);
574 				if (IS_KVM_NULL_FN(range->handler))
575 					break;
576 			}
577 			ret |= range->handler(kvm, &gfn_range);
578 		}
579 	}
580 
581 	if (range->flush_on_ret && ret)
582 		kvm_flush_remote_tlbs(kvm);
583 
584 	if (locked)
585 		KVM_MMU_UNLOCK(kvm);
586 
587 	srcu_read_unlock(&kvm->srcu, idx);
588 
589 	/* The notifiers are averse to booleans. :-( */
590 	return (int)ret;
591 }
592 
593 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
594 						unsigned long start,
595 						unsigned long end,
596 						pte_t pte,
597 						hva_handler_t handler)
598 {
599 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
600 	const struct kvm_hva_range range = {
601 		.start		= start,
602 		.end		= end,
603 		.pte		= pte,
604 		.handler	= handler,
605 		.on_lock	= (void *)kvm_null_fn,
606 		.flush_on_ret	= true,
607 		.may_block	= false,
608 	};
609 
610 	return __kvm_handle_hva_range(kvm, &range);
611 }
612 
613 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
614 							 unsigned long start,
615 							 unsigned long end,
616 							 hva_handler_t handler)
617 {
618 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
619 	const struct kvm_hva_range range = {
620 		.start		= start,
621 		.end		= end,
622 		.pte		= __pte(0),
623 		.handler	= handler,
624 		.on_lock	= (void *)kvm_null_fn,
625 		.flush_on_ret	= false,
626 		.may_block	= false,
627 	};
628 
629 	return __kvm_handle_hva_range(kvm, &range);
630 }
631 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
632 					struct mm_struct *mm,
633 					unsigned long address,
634 					pte_t pte)
635 {
636 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
637 
638 	trace_kvm_set_spte_hva(address);
639 
640 	/*
641 	 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
642 	 * If mmu_notifier_count is zero, then no in-progress invalidations,
643 	 * including this one, found a relevant memslot at start(); rechecking
644 	 * memslots here is unnecessary.  Note, a false positive (count elevated
645 	 * by a different invalidation) is sub-optimal but functionally ok.
646 	 */
647 	WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
648 	if (!READ_ONCE(kvm->mmu_notifier_count))
649 		return;
650 
651 	kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
652 }
653 
654 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
655 				   unsigned long end)
656 {
657 	/*
658 	 * The count increase must become visible at unlock time as no
659 	 * spte can be established without taking the mmu_lock and
660 	 * count is also read inside the mmu_lock critical section.
661 	 */
662 	kvm->mmu_notifier_count++;
663 	if (likely(kvm->mmu_notifier_count == 1)) {
664 		kvm->mmu_notifier_range_start = start;
665 		kvm->mmu_notifier_range_end = end;
666 	} else {
667 		/*
668 		 * Fully tracking multiple concurrent ranges has dimishing
669 		 * returns. Keep things simple and just find the minimal range
670 		 * which includes the current and new ranges. As there won't be
671 		 * enough information to subtract a range after its invalidate
672 		 * completes, any ranges invalidated concurrently will
673 		 * accumulate and persist until all outstanding invalidates
674 		 * complete.
675 		 */
676 		kvm->mmu_notifier_range_start =
677 			min(kvm->mmu_notifier_range_start, start);
678 		kvm->mmu_notifier_range_end =
679 			max(kvm->mmu_notifier_range_end, end);
680 	}
681 }
682 
683 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
684 					const struct mmu_notifier_range *range)
685 {
686 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
687 	const struct kvm_hva_range hva_range = {
688 		.start		= range->start,
689 		.end		= range->end,
690 		.pte		= __pte(0),
691 		.handler	= kvm_unmap_gfn_range,
692 		.on_lock	= kvm_inc_notifier_count,
693 		.flush_on_ret	= true,
694 		.may_block	= mmu_notifier_range_blockable(range),
695 	};
696 
697 	trace_kvm_unmap_hva_range(range->start, range->end);
698 
699 	/*
700 	 * Prevent memslot modification between range_start() and range_end()
701 	 * so that conditionally locking provides the same result in both
702 	 * functions.  Without that guarantee, the mmu_notifier_count
703 	 * adjustments will be imbalanced.
704 	 *
705 	 * Pairs with the decrement in range_end().
706 	 */
707 	spin_lock(&kvm->mn_invalidate_lock);
708 	kvm->mn_active_invalidate_count++;
709 	spin_unlock(&kvm->mn_invalidate_lock);
710 
711 	gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
712 					  hva_range.may_block);
713 
714 	__kvm_handle_hva_range(kvm, &hva_range);
715 
716 	return 0;
717 }
718 
719 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
720 				   unsigned long end)
721 {
722 	/*
723 	 * This sequence increase will notify the kvm page fault that
724 	 * the page that is going to be mapped in the spte could have
725 	 * been freed.
726 	 */
727 	kvm->mmu_notifier_seq++;
728 	smp_wmb();
729 	/*
730 	 * The above sequence increase must be visible before the
731 	 * below count decrease, which is ensured by the smp_wmb above
732 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
733 	 */
734 	kvm->mmu_notifier_count--;
735 }
736 
737 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
738 					const struct mmu_notifier_range *range)
739 {
740 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
741 	const struct kvm_hva_range hva_range = {
742 		.start		= range->start,
743 		.end		= range->end,
744 		.pte		= __pte(0),
745 		.handler	= (void *)kvm_null_fn,
746 		.on_lock	= kvm_dec_notifier_count,
747 		.flush_on_ret	= false,
748 		.may_block	= mmu_notifier_range_blockable(range),
749 	};
750 	bool wake;
751 
752 	__kvm_handle_hva_range(kvm, &hva_range);
753 
754 	/* Pairs with the increment in range_start(). */
755 	spin_lock(&kvm->mn_invalidate_lock);
756 	wake = (--kvm->mn_active_invalidate_count == 0);
757 	spin_unlock(&kvm->mn_invalidate_lock);
758 
759 	/*
760 	 * There can only be one waiter, since the wait happens under
761 	 * slots_lock.
762 	 */
763 	if (wake)
764 		rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
765 
766 	BUG_ON(kvm->mmu_notifier_count < 0);
767 }
768 
769 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
770 					      struct mm_struct *mm,
771 					      unsigned long start,
772 					      unsigned long end)
773 {
774 	trace_kvm_age_hva(start, end);
775 
776 	return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
777 }
778 
779 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
780 					struct mm_struct *mm,
781 					unsigned long start,
782 					unsigned long end)
783 {
784 	trace_kvm_age_hva(start, end);
785 
786 	/*
787 	 * Even though we do not flush TLB, this will still adversely
788 	 * affect performance on pre-Haswell Intel EPT, where there is
789 	 * no EPT Access Bit to clear so that we have to tear down EPT
790 	 * tables instead. If we find this unacceptable, we can always
791 	 * add a parameter to kvm_age_hva so that it effectively doesn't
792 	 * do anything on clear_young.
793 	 *
794 	 * Also note that currently we never issue secondary TLB flushes
795 	 * from clear_young, leaving this job up to the regular system
796 	 * cadence. If we find this inaccurate, we might come up with a
797 	 * more sophisticated heuristic later.
798 	 */
799 	return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
800 }
801 
802 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
803 				       struct mm_struct *mm,
804 				       unsigned long address)
805 {
806 	trace_kvm_test_age_hva(address);
807 
808 	return kvm_handle_hva_range_no_flush(mn, address, address + 1,
809 					     kvm_test_age_gfn);
810 }
811 
812 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
813 				     struct mm_struct *mm)
814 {
815 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
816 	int idx;
817 
818 	idx = srcu_read_lock(&kvm->srcu);
819 	kvm_arch_flush_shadow_all(kvm);
820 	srcu_read_unlock(&kvm->srcu, idx);
821 }
822 
823 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
824 	.invalidate_range	= kvm_mmu_notifier_invalidate_range,
825 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
826 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
827 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
828 	.clear_young		= kvm_mmu_notifier_clear_young,
829 	.test_young		= kvm_mmu_notifier_test_young,
830 	.change_pte		= kvm_mmu_notifier_change_pte,
831 	.release		= kvm_mmu_notifier_release,
832 };
833 
834 static int kvm_init_mmu_notifier(struct kvm *kvm)
835 {
836 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
837 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
838 }
839 
840 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
841 
842 static int kvm_init_mmu_notifier(struct kvm *kvm)
843 {
844 	return 0;
845 }
846 
847 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
848 
849 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
850 static int kvm_pm_notifier_call(struct notifier_block *bl,
851 				unsigned long state,
852 				void *unused)
853 {
854 	struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
855 
856 	return kvm_arch_pm_notifier(kvm, state);
857 }
858 
859 static void kvm_init_pm_notifier(struct kvm *kvm)
860 {
861 	kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
862 	/* Suspend KVM before we suspend ftrace, RCU, etc. */
863 	kvm->pm_notifier.priority = INT_MAX;
864 	register_pm_notifier(&kvm->pm_notifier);
865 }
866 
867 static void kvm_destroy_pm_notifier(struct kvm *kvm)
868 {
869 	unregister_pm_notifier(&kvm->pm_notifier);
870 }
871 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
872 static void kvm_init_pm_notifier(struct kvm *kvm)
873 {
874 }
875 
876 static void kvm_destroy_pm_notifier(struct kvm *kvm)
877 {
878 }
879 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
880 
881 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
882 {
883 	if (!memslot->dirty_bitmap)
884 		return;
885 
886 	kvfree(memslot->dirty_bitmap);
887 	memslot->dirty_bitmap = NULL;
888 }
889 
890 /* This does not remove the slot from struct kvm_memslots data structures */
891 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
892 {
893 	kvm_destroy_dirty_bitmap(slot);
894 
895 	kvm_arch_free_memslot(kvm, slot);
896 
897 	kfree(slot);
898 }
899 
900 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
901 {
902 	struct hlist_node *idnode;
903 	struct kvm_memory_slot *memslot;
904 	int bkt;
905 
906 	/*
907 	 * The same memslot objects live in both active and inactive sets,
908 	 * arbitrarily free using index '1' so the second invocation of this
909 	 * function isn't operating over a structure with dangling pointers
910 	 * (even though this function isn't actually touching them).
911 	 */
912 	if (!slots->node_idx)
913 		return;
914 
915 	hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
916 		kvm_free_memslot(kvm, memslot);
917 }
918 
919 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
920 {
921 	switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
922 	case KVM_STATS_TYPE_INSTANT:
923 		return 0444;
924 	case KVM_STATS_TYPE_CUMULATIVE:
925 	case KVM_STATS_TYPE_PEAK:
926 	default:
927 		return 0644;
928 	}
929 }
930 
931 
932 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
933 {
934 	int i;
935 	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
936 				      kvm_vcpu_stats_header.num_desc;
937 
938 	if (!kvm->debugfs_dentry)
939 		return;
940 
941 	debugfs_remove_recursive(kvm->debugfs_dentry);
942 
943 	if (kvm->debugfs_stat_data) {
944 		for (i = 0; i < kvm_debugfs_num_entries; i++)
945 			kfree(kvm->debugfs_stat_data[i]);
946 		kfree(kvm->debugfs_stat_data);
947 	}
948 }
949 
950 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
951 {
952 	static DEFINE_MUTEX(kvm_debugfs_lock);
953 	struct dentry *dent;
954 	char dir_name[ITOA_MAX_LEN * 2];
955 	struct kvm_stat_data *stat_data;
956 	const struct _kvm_stats_desc *pdesc;
957 	int i, ret;
958 	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
959 				      kvm_vcpu_stats_header.num_desc;
960 
961 	if (!debugfs_initialized())
962 		return 0;
963 
964 	snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
965 	mutex_lock(&kvm_debugfs_lock);
966 	dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
967 	if (dent) {
968 		pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
969 		dput(dent);
970 		mutex_unlock(&kvm_debugfs_lock);
971 		return 0;
972 	}
973 	dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
974 	mutex_unlock(&kvm_debugfs_lock);
975 	if (IS_ERR(dent))
976 		return 0;
977 
978 	kvm->debugfs_dentry = dent;
979 	kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
980 					 sizeof(*kvm->debugfs_stat_data),
981 					 GFP_KERNEL_ACCOUNT);
982 	if (!kvm->debugfs_stat_data)
983 		return -ENOMEM;
984 
985 	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
986 		pdesc = &kvm_vm_stats_desc[i];
987 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
988 		if (!stat_data)
989 			return -ENOMEM;
990 
991 		stat_data->kvm = kvm;
992 		stat_data->desc = pdesc;
993 		stat_data->kind = KVM_STAT_VM;
994 		kvm->debugfs_stat_data[i] = stat_data;
995 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
996 				    kvm->debugfs_dentry, stat_data,
997 				    &stat_fops_per_vm);
998 	}
999 
1000 	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1001 		pdesc = &kvm_vcpu_stats_desc[i];
1002 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1003 		if (!stat_data)
1004 			return -ENOMEM;
1005 
1006 		stat_data->kvm = kvm;
1007 		stat_data->desc = pdesc;
1008 		stat_data->kind = KVM_STAT_VCPU;
1009 		kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1010 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1011 				    kvm->debugfs_dentry, stat_data,
1012 				    &stat_fops_per_vm);
1013 	}
1014 
1015 	ret = kvm_arch_create_vm_debugfs(kvm);
1016 	if (ret) {
1017 		kvm_destroy_vm_debugfs(kvm);
1018 		return i;
1019 	}
1020 
1021 	return 0;
1022 }
1023 
1024 /*
1025  * Called after the VM is otherwise initialized, but just before adding it to
1026  * the vm_list.
1027  */
1028 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1029 {
1030 	return 0;
1031 }
1032 
1033 /*
1034  * Called just after removing the VM from the vm_list, but before doing any
1035  * other destruction.
1036  */
1037 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1038 {
1039 }
1040 
1041 /*
1042  * Called after per-vm debugfs created.  When called kvm->debugfs_dentry should
1043  * be setup already, so we can create arch-specific debugfs entries under it.
1044  * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1045  * a per-arch destroy interface is not needed.
1046  */
1047 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1048 {
1049 	return 0;
1050 }
1051 
1052 static struct kvm *kvm_create_vm(unsigned long type)
1053 {
1054 	struct kvm *kvm = kvm_arch_alloc_vm();
1055 	struct kvm_memslots *slots;
1056 	int r = -ENOMEM;
1057 	int i, j;
1058 
1059 	if (!kvm)
1060 		return ERR_PTR(-ENOMEM);
1061 
1062 	KVM_MMU_LOCK_INIT(kvm);
1063 	mmgrab(current->mm);
1064 	kvm->mm = current->mm;
1065 	kvm_eventfd_init(kvm);
1066 	mutex_init(&kvm->lock);
1067 	mutex_init(&kvm->irq_lock);
1068 	mutex_init(&kvm->slots_lock);
1069 	mutex_init(&kvm->slots_arch_lock);
1070 	spin_lock_init(&kvm->mn_invalidate_lock);
1071 	rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1072 	xa_init(&kvm->vcpu_array);
1073 
1074 	INIT_LIST_HEAD(&kvm->gpc_list);
1075 	spin_lock_init(&kvm->gpc_lock);
1076 
1077 	INIT_LIST_HEAD(&kvm->devices);
1078 
1079 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1080 
1081 	if (init_srcu_struct(&kvm->srcu))
1082 		goto out_err_no_srcu;
1083 	if (init_srcu_struct(&kvm->irq_srcu))
1084 		goto out_err_no_irq_srcu;
1085 
1086 	refcount_set(&kvm->users_count, 1);
1087 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1088 		for (j = 0; j < 2; j++) {
1089 			slots = &kvm->__memslots[i][j];
1090 
1091 			atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1092 			slots->hva_tree = RB_ROOT_CACHED;
1093 			slots->gfn_tree = RB_ROOT;
1094 			hash_init(slots->id_hash);
1095 			slots->node_idx = j;
1096 
1097 			/* Generations must be different for each address space. */
1098 			slots->generation = i;
1099 		}
1100 
1101 		rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1102 	}
1103 
1104 	for (i = 0; i < KVM_NR_BUSES; i++) {
1105 		rcu_assign_pointer(kvm->buses[i],
1106 			kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1107 		if (!kvm->buses[i])
1108 			goto out_err_no_arch_destroy_vm;
1109 	}
1110 
1111 	kvm->max_halt_poll_ns = halt_poll_ns;
1112 
1113 	r = kvm_arch_init_vm(kvm, type);
1114 	if (r)
1115 		goto out_err_no_arch_destroy_vm;
1116 
1117 	r = hardware_enable_all();
1118 	if (r)
1119 		goto out_err_no_disable;
1120 
1121 #ifdef CONFIG_HAVE_KVM_IRQFD
1122 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1123 #endif
1124 
1125 	r = kvm_init_mmu_notifier(kvm);
1126 	if (r)
1127 		goto out_err_no_mmu_notifier;
1128 
1129 	r = kvm_arch_post_init_vm(kvm);
1130 	if (r)
1131 		goto out_err;
1132 
1133 	mutex_lock(&kvm_lock);
1134 	list_add(&kvm->vm_list, &vm_list);
1135 	mutex_unlock(&kvm_lock);
1136 
1137 	preempt_notifier_inc();
1138 	kvm_init_pm_notifier(kvm);
1139 
1140 	return kvm;
1141 
1142 out_err:
1143 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1144 	if (kvm->mmu_notifier.ops)
1145 		mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1146 #endif
1147 out_err_no_mmu_notifier:
1148 	hardware_disable_all();
1149 out_err_no_disable:
1150 	kvm_arch_destroy_vm(kvm);
1151 out_err_no_arch_destroy_vm:
1152 	WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1153 	for (i = 0; i < KVM_NR_BUSES; i++)
1154 		kfree(kvm_get_bus(kvm, i));
1155 	cleanup_srcu_struct(&kvm->irq_srcu);
1156 out_err_no_irq_srcu:
1157 	cleanup_srcu_struct(&kvm->srcu);
1158 out_err_no_srcu:
1159 	kvm_arch_free_vm(kvm);
1160 	mmdrop(current->mm);
1161 	return ERR_PTR(r);
1162 }
1163 
1164 static void kvm_destroy_devices(struct kvm *kvm)
1165 {
1166 	struct kvm_device *dev, *tmp;
1167 
1168 	/*
1169 	 * We do not need to take the kvm->lock here, because nobody else
1170 	 * has a reference to the struct kvm at this point and therefore
1171 	 * cannot access the devices list anyhow.
1172 	 */
1173 	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1174 		list_del(&dev->vm_node);
1175 		dev->ops->destroy(dev);
1176 	}
1177 }
1178 
1179 static void kvm_destroy_vm(struct kvm *kvm)
1180 {
1181 	int i;
1182 	struct mm_struct *mm = kvm->mm;
1183 
1184 	kvm_destroy_pm_notifier(kvm);
1185 	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1186 	kvm_destroy_vm_debugfs(kvm);
1187 	kvm_arch_sync_events(kvm);
1188 	mutex_lock(&kvm_lock);
1189 	list_del(&kvm->vm_list);
1190 	mutex_unlock(&kvm_lock);
1191 	kvm_arch_pre_destroy_vm(kvm);
1192 
1193 	kvm_free_irq_routing(kvm);
1194 	for (i = 0; i < KVM_NR_BUSES; i++) {
1195 		struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1196 
1197 		if (bus)
1198 			kvm_io_bus_destroy(bus);
1199 		kvm->buses[i] = NULL;
1200 	}
1201 	kvm_coalesced_mmio_free(kvm);
1202 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1203 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1204 	/*
1205 	 * At this point, pending calls to invalidate_range_start()
1206 	 * have completed but no more MMU notifiers will run, so
1207 	 * mn_active_invalidate_count may remain unbalanced.
1208 	 * No threads can be waiting in install_new_memslots as the
1209 	 * last reference on KVM has been dropped, but freeing
1210 	 * memslots would deadlock without this manual intervention.
1211 	 */
1212 	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1213 	kvm->mn_active_invalidate_count = 0;
1214 #else
1215 	kvm_arch_flush_shadow_all(kvm);
1216 #endif
1217 	kvm_arch_destroy_vm(kvm);
1218 	kvm_destroy_devices(kvm);
1219 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1220 		kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1221 		kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1222 	}
1223 	cleanup_srcu_struct(&kvm->irq_srcu);
1224 	cleanup_srcu_struct(&kvm->srcu);
1225 	kvm_arch_free_vm(kvm);
1226 	preempt_notifier_dec();
1227 	hardware_disable_all();
1228 	mmdrop(mm);
1229 }
1230 
1231 void kvm_get_kvm(struct kvm *kvm)
1232 {
1233 	refcount_inc(&kvm->users_count);
1234 }
1235 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1236 
1237 /*
1238  * Make sure the vm is not during destruction, which is a safe version of
1239  * kvm_get_kvm().  Return true if kvm referenced successfully, false otherwise.
1240  */
1241 bool kvm_get_kvm_safe(struct kvm *kvm)
1242 {
1243 	return refcount_inc_not_zero(&kvm->users_count);
1244 }
1245 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1246 
1247 void kvm_put_kvm(struct kvm *kvm)
1248 {
1249 	if (refcount_dec_and_test(&kvm->users_count))
1250 		kvm_destroy_vm(kvm);
1251 }
1252 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1253 
1254 /*
1255  * Used to put a reference that was taken on behalf of an object associated
1256  * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1257  * of the new file descriptor fails and the reference cannot be transferred to
1258  * its final owner.  In such cases, the caller is still actively using @kvm and
1259  * will fail miserably if the refcount unexpectedly hits zero.
1260  */
1261 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1262 {
1263 	WARN_ON(refcount_dec_and_test(&kvm->users_count));
1264 }
1265 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1266 
1267 static int kvm_vm_release(struct inode *inode, struct file *filp)
1268 {
1269 	struct kvm *kvm = filp->private_data;
1270 
1271 	kvm_irqfd_release(kvm);
1272 
1273 	kvm_put_kvm(kvm);
1274 	return 0;
1275 }
1276 
1277 /*
1278  * Allocation size is twice as large as the actual dirty bitmap size.
1279  * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1280  */
1281 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1282 {
1283 	unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
1284 
1285 	memslot->dirty_bitmap = kvzalloc(dirty_bytes, GFP_KERNEL_ACCOUNT);
1286 	if (!memslot->dirty_bitmap)
1287 		return -ENOMEM;
1288 
1289 	return 0;
1290 }
1291 
1292 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1293 {
1294 	struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1295 	int node_idx_inactive = active->node_idx ^ 1;
1296 
1297 	return &kvm->__memslots[as_id][node_idx_inactive];
1298 }
1299 
1300 /*
1301  * Helper to get the address space ID when one of memslot pointers may be NULL.
1302  * This also serves as a sanity that at least one of the pointers is non-NULL,
1303  * and that their address space IDs don't diverge.
1304  */
1305 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1306 				  struct kvm_memory_slot *b)
1307 {
1308 	if (WARN_ON_ONCE(!a && !b))
1309 		return 0;
1310 
1311 	if (!a)
1312 		return b->as_id;
1313 	if (!b)
1314 		return a->as_id;
1315 
1316 	WARN_ON_ONCE(a->as_id != b->as_id);
1317 	return a->as_id;
1318 }
1319 
1320 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1321 				struct kvm_memory_slot *slot)
1322 {
1323 	struct rb_root *gfn_tree = &slots->gfn_tree;
1324 	struct rb_node **node, *parent;
1325 	int idx = slots->node_idx;
1326 
1327 	parent = NULL;
1328 	for (node = &gfn_tree->rb_node; *node; ) {
1329 		struct kvm_memory_slot *tmp;
1330 
1331 		tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1332 		parent = *node;
1333 		if (slot->base_gfn < tmp->base_gfn)
1334 			node = &(*node)->rb_left;
1335 		else if (slot->base_gfn > tmp->base_gfn)
1336 			node = &(*node)->rb_right;
1337 		else
1338 			BUG();
1339 	}
1340 
1341 	rb_link_node(&slot->gfn_node[idx], parent, node);
1342 	rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1343 }
1344 
1345 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1346 			       struct kvm_memory_slot *slot)
1347 {
1348 	rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1349 }
1350 
1351 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1352 				 struct kvm_memory_slot *old,
1353 				 struct kvm_memory_slot *new)
1354 {
1355 	int idx = slots->node_idx;
1356 
1357 	WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1358 
1359 	rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1360 			&slots->gfn_tree);
1361 }
1362 
1363 /*
1364  * Replace @old with @new in the inactive memslots.
1365  *
1366  * With NULL @old this simply adds @new.
1367  * With NULL @new this simply removes @old.
1368  *
1369  * If @new is non-NULL its hva_node[slots_idx] range has to be set
1370  * appropriately.
1371  */
1372 static void kvm_replace_memslot(struct kvm *kvm,
1373 				struct kvm_memory_slot *old,
1374 				struct kvm_memory_slot *new)
1375 {
1376 	int as_id = kvm_memslots_get_as_id(old, new);
1377 	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1378 	int idx = slots->node_idx;
1379 
1380 	if (old) {
1381 		hash_del(&old->id_node[idx]);
1382 		interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1383 
1384 		if ((long)old == atomic_long_read(&slots->last_used_slot))
1385 			atomic_long_set(&slots->last_used_slot, (long)new);
1386 
1387 		if (!new) {
1388 			kvm_erase_gfn_node(slots, old);
1389 			return;
1390 		}
1391 	}
1392 
1393 	/*
1394 	 * Initialize @new's hva range.  Do this even when replacing an @old
1395 	 * slot, kvm_copy_memslot() deliberately does not touch node data.
1396 	 */
1397 	new->hva_node[idx].start = new->userspace_addr;
1398 	new->hva_node[idx].last = new->userspace_addr +
1399 				  (new->npages << PAGE_SHIFT) - 1;
1400 
1401 	/*
1402 	 * (Re)Add the new memslot.  There is no O(1) interval_tree_replace(),
1403 	 * hva_node needs to be swapped with remove+insert even though hva can't
1404 	 * change when replacing an existing slot.
1405 	 */
1406 	hash_add(slots->id_hash, &new->id_node[idx], new->id);
1407 	interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1408 
1409 	/*
1410 	 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1411 	 * switch the node in the gfn tree instead of removing the old and
1412 	 * inserting the new as two separate operations. Replacement is a
1413 	 * single O(1) operation versus two O(log(n)) operations for
1414 	 * remove+insert.
1415 	 */
1416 	if (old && old->base_gfn == new->base_gfn) {
1417 		kvm_replace_gfn_node(slots, old, new);
1418 	} else {
1419 		if (old)
1420 			kvm_erase_gfn_node(slots, old);
1421 		kvm_insert_gfn_node(slots, new);
1422 	}
1423 }
1424 
1425 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1426 {
1427 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1428 
1429 #ifdef __KVM_HAVE_READONLY_MEM
1430 	valid_flags |= KVM_MEM_READONLY;
1431 #endif
1432 
1433 	if (mem->flags & ~valid_flags)
1434 		return -EINVAL;
1435 
1436 	return 0;
1437 }
1438 
1439 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1440 {
1441 	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1442 
1443 	/* Grab the generation from the activate memslots. */
1444 	u64 gen = __kvm_memslots(kvm, as_id)->generation;
1445 
1446 	WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1447 	slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1448 
1449 	/*
1450 	 * Do not store the new memslots while there are invalidations in
1451 	 * progress, otherwise the locking in invalidate_range_start and
1452 	 * invalidate_range_end will be unbalanced.
1453 	 */
1454 	spin_lock(&kvm->mn_invalidate_lock);
1455 	prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1456 	while (kvm->mn_active_invalidate_count) {
1457 		set_current_state(TASK_UNINTERRUPTIBLE);
1458 		spin_unlock(&kvm->mn_invalidate_lock);
1459 		schedule();
1460 		spin_lock(&kvm->mn_invalidate_lock);
1461 	}
1462 	finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1463 	rcu_assign_pointer(kvm->memslots[as_id], slots);
1464 	spin_unlock(&kvm->mn_invalidate_lock);
1465 
1466 	/*
1467 	 * Acquired in kvm_set_memslot. Must be released before synchronize
1468 	 * SRCU below in order to avoid deadlock with another thread
1469 	 * acquiring the slots_arch_lock in an srcu critical section.
1470 	 */
1471 	mutex_unlock(&kvm->slots_arch_lock);
1472 
1473 	synchronize_srcu_expedited(&kvm->srcu);
1474 
1475 	/*
1476 	 * Increment the new memslot generation a second time, dropping the
1477 	 * update in-progress flag and incrementing the generation based on
1478 	 * the number of address spaces.  This provides a unique and easily
1479 	 * identifiable generation number while the memslots are in flux.
1480 	 */
1481 	gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1482 
1483 	/*
1484 	 * Generations must be unique even across address spaces.  We do not need
1485 	 * a global counter for that, instead the generation space is evenly split
1486 	 * across address spaces.  For example, with two address spaces, address
1487 	 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1488 	 * use generations 1, 3, 5, ...
1489 	 */
1490 	gen += KVM_ADDRESS_SPACE_NUM;
1491 
1492 	kvm_arch_memslots_updated(kvm, gen);
1493 
1494 	slots->generation = gen;
1495 }
1496 
1497 static int kvm_prepare_memory_region(struct kvm *kvm,
1498 				     const struct kvm_memory_slot *old,
1499 				     struct kvm_memory_slot *new,
1500 				     enum kvm_mr_change change)
1501 {
1502 	int r;
1503 
1504 	/*
1505 	 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1506 	 * will be freed on "commit".  If logging is enabled in both old and
1507 	 * new, reuse the existing bitmap.  If logging is enabled only in the
1508 	 * new and KVM isn't using a ring buffer, allocate and initialize a
1509 	 * new bitmap.
1510 	 */
1511 	if (change != KVM_MR_DELETE) {
1512 		if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1513 			new->dirty_bitmap = NULL;
1514 		else if (old && old->dirty_bitmap)
1515 			new->dirty_bitmap = old->dirty_bitmap;
1516 		else if (!kvm->dirty_ring_size) {
1517 			r = kvm_alloc_dirty_bitmap(new);
1518 			if (r)
1519 				return r;
1520 
1521 			if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1522 				bitmap_set(new->dirty_bitmap, 0, new->npages);
1523 		}
1524 	}
1525 
1526 	r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1527 
1528 	/* Free the bitmap on failure if it was allocated above. */
1529 	if (r && new && new->dirty_bitmap && old && !old->dirty_bitmap)
1530 		kvm_destroy_dirty_bitmap(new);
1531 
1532 	return r;
1533 }
1534 
1535 static void kvm_commit_memory_region(struct kvm *kvm,
1536 				     struct kvm_memory_slot *old,
1537 				     const struct kvm_memory_slot *new,
1538 				     enum kvm_mr_change change)
1539 {
1540 	/*
1541 	 * Update the total number of memslot pages before calling the arch
1542 	 * hook so that architectures can consume the result directly.
1543 	 */
1544 	if (change == KVM_MR_DELETE)
1545 		kvm->nr_memslot_pages -= old->npages;
1546 	else if (change == KVM_MR_CREATE)
1547 		kvm->nr_memslot_pages += new->npages;
1548 
1549 	kvm_arch_commit_memory_region(kvm, old, new, change);
1550 
1551 	switch (change) {
1552 	case KVM_MR_CREATE:
1553 		/* Nothing more to do. */
1554 		break;
1555 	case KVM_MR_DELETE:
1556 		/* Free the old memslot and all its metadata. */
1557 		kvm_free_memslot(kvm, old);
1558 		break;
1559 	case KVM_MR_MOVE:
1560 	case KVM_MR_FLAGS_ONLY:
1561 		/*
1562 		 * Free the dirty bitmap as needed; the below check encompasses
1563 		 * both the flags and whether a ring buffer is being used)
1564 		 */
1565 		if (old->dirty_bitmap && !new->dirty_bitmap)
1566 			kvm_destroy_dirty_bitmap(old);
1567 
1568 		/*
1569 		 * The final quirk.  Free the detached, old slot, but only its
1570 		 * memory, not any metadata.  Metadata, including arch specific
1571 		 * data, may be reused by @new.
1572 		 */
1573 		kfree(old);
1574 		break;
1575 	default:
1576 		BUG();
1577 	}
1578 }
1579 
1580 /*
1581  * Activate @new, which must be installed in the inactive slots by the caller,
1582  * by swapping the active slots and then propagating @new to @old once @old is
1583  * unreachable and can be safely modified.
1584  *
1585  * With NULL @old this simply adds @new to @active (while swapping the sets).
1586  * With NULL @new this simply removes @old from @active and frees it
1587  * (while also swapping the sets).
1588  */
1589 static void kvm_activate_memslot(struct kvm *kvm,
1590 				 struct kvm_memory_slot *old,
1591 				 struct kvm_memory_slot *new)
1592 {
1593 	int as_id = kvm_memslots_get_as_id(old, new);
1594 
1595 	kvm_swap_active_memslots(kvm, as_id);
1596 
1597 	/* Propagate the new memslot to the now inactive memslots. */
1598 	kvm_replace_memslot(kvm, old, new);
1599 }
1600 
1601 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1602 			     const struct kvm_memory_slot *src)
1603 {
1604 	dest->base_gfn = src->base_gfn;
1605 	dest->npages = src->npages;
1606 	dest->dirty_bitmap = src->dirty_bitmap;
1607 	dest->arch = src->arch;
1608 	dest->userspace_addr = src->userspace_addr;
1609 	dest->flags = src->flags;
1610 	dest->id = src->id;
1611 	dest->as_id = src->as_id;
1612 }
1613 
1614 static void kvm_invalidate_memslot(struct kvm *kvm,
1615 				   struct kvm_memory_slot *old,
1616 				   struct kvm_memory_slot *invalid_slot)
1617 {
1618 	/*
1619 	 * Mark the current slot INVALID.  As with all memslot modifications,
1620 	 * this must be done on an unreachable slot to avoid modifying the
1621 	 * current slot in the active tree.
1622 	 */
1623 	kvm_copy_memslot(invalid_slot, old);
1624 	invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1625 	kvm_replace_memslot(kvm, old, invalid_slot);
1626 
1627 	/*
1628 	 * Activate the slot that is now marked INVALID, but don't propagate
1629 	 * the slot to the now inactive slots. The slot is either going to be
1630 	 * deleted or recreated as a new slot.
1631 	 */
1632 	kvm_swap_active_memslots(kvm, old->as_id);
1633 
1634 	/*
1635 	 * From this point no new shadow pages pointing to a deleted, or moved,
1636 	 * memslot will be created.  Validation of sp->gfn happens in:
1637 	 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1638 	 *	- kvm_is_visible_gfn (mmu_check_root)
1639 	 */
1640 	kvm_arch_flush_shadow_memslot(kvm, old);
1641 
1642 	/* Was released by kvm_swap_active_memslots, reacquire. */
1643 	mutex_lock(&kvm->slots_arch_lock);
1644 
1645 	/*
1646 	 * Copy the arch-specific field of the newly-installed slot back to the
1647 	 * old slot as the arch data could have changed between releasing
1648 	 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1649 	 * above.  Writers are required to retrieve memslots *after* acquiring
1650 	 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1651 	 */
1652 	old->arch = invalid_slot->arch;
1653 }
1654 
1655 static void kvm_create_memslot(struct kvm *kvm,
1656 			       struct kvm_memory_slot *new)
1657 {
1658 	/* Add the new memslot to the inactive set and activate. */
1659 	kvm_replace_memslot(kvm, NULL, new);
1660 	kvm_activate_memslot(kvm, NULL, new);
1661 }
1662 
1663 static void kvm_delete_memslot(struct kvm *kvm,
1664 			       struct kvm_memory_slot *old,
1665 			       struct kvm_memory_slot *invalid_slot)
1666 {
1667 	/*
1668 	 * Remove the old memslot (in the inactive memslots) by passing NULL as
1669 	 * the "new" slot, and for the invalid version in the active slots.
1670 	 */
1671 	kvm_replace_memslot(kvm, old, NULL);
1672 	kvm_activate_memslot(kvm, invalid_slot, NULL);
1673 }
1674 
1675 static void kvm_move_memslot(struct kvm *kvm,
1676 			     struct kvm_memory_slot *old,
1677 			     struct kvm_memory_slot *new,
1678 			     struct kvm_memory_slot *invalid_slot)
1679 {
1680 	/*
1681 	 * Replace the old memslot in the inactive slots, and then swap slots
1682 	 * and replace the current INVALID with the new as well.
1683 	 */
1684 	kvm_replace_memslot(kvm, old, new);
1685 	kvm_activate_memslot(kvm, invalid_slot, new);
1686 }
1687 
1688 static void kvm_update_flags_memslot(struct kvm *kvm,
1689 				     struct kvm_memory_slot *old,
1690 				     struct kvm_memory_slot *new)
1691 {
1692 	/*
1693 	 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1694 	 * an intermediate step. Instead, the old memslot is simply replaced
1695 	 * with a new, updated copy in both memslot sets.
1696 	 */
1697 	kvm_replace_memslot(kvm, old, new);
1698 	kvm_activate_memslot(kvm, old, new);
1699 }
1700 
1701 static int kvm_set_memslot(struct kvm *kvm,
1702 			   struct kvm_memory_slot *old,
1703 			   struct kvm_memory_slot *new,
1704 			   enum kvm_mr_change change)
1705 {
1706 	struct kvm_memory_slot *invalid_slot;
1707 	int r;
1708 
1709 	/*
1710 	 * Released in kvm_swap_active_memslots.
1711 	 *
1712 	 * Must be held from before the current memslots are copied until
1713 	 * after the new memslots are installed with rcu_assign_pointer,
1714 	 * then released before the synchronize srcu in kvm_swap_active_memslots.
1715 	 *
1716 	 * When modifying memslots outside of the slots_lock, must be held
1717 	 * before reading the pointer to the current memslots until after all
1718 	 * changes to those memslots are complete.
1719 	 *
1720 	 * These rules ensure that installing new memslots does not lose
1721 	 * changes made to the previous memslots.
1722 	 */
1723 	mutex_lock(&kvm->slots_arch_lock);
1724 
1725 	/*
1726 	 * Invalidate the old slot if it's being deleted or moved.  This is
1727 	 * done prior to actually deleting/moving the memslot to allow vCPUs to
1728 	 * continue running by ensuring there are no mappings or shadow pages
1729 	 * for the memslot when it is deleted/moved.  Without pre-invalidation
1730 	 * (and without a lock), a window would exist between effecting the
1731 	 * delete/move and committing the changes in arch code where KVM or a
1732 	 * guest could access a non-existent memslot.
1733 	 *
1734 	 * Modifications are done on a temporary, unreachable slot.  The old
1735 	 * slot needs to be preserved in case a later step fails and the
1736 	 * invalidation needs to be reverted.
1737 	 */
1738 	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1739 		invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1740 		if (!invalid_slot) {
1741 			mutex_unlock(&kvm->slots_arch_lock);
1742 			return -ENOMEM;
1743 		}
1744 		kvm_invalidate_memslot(kvm, old, invalid_slot);
1745 	}
1746 
1747 	r = kvm_prepare_memory_region(kvm, old, new, change);
1748 	if (r) {
1749 		/*
1750 		 * For DELETE/MOVE, revert the above INVALID change.  No
1751 		 * modifications required since the original slot was preserved
1752 		 * in the inactive slots.  Changing the active memslots also
1753 		 * release slots_arch_lock.
1754 		 */
1755 		if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1756 			kvm_activate_memslot(kvm, invalid_slot, old);
1757 			kfree(invalid_slot);
1758 		} else {
1759 			mutex_unlock(&kvm->slots_arch_lock);
1760 		}
1761 		return r;
1762 	}
1763 
1764 	/*
1765 	 * For DELETE and MOVE, the working slot is now active as the INVALID
1766 	 * version of the old slot.  MOVE is particularly special as it reuses
1767 	 * the old slot and returns a copy of the old slot (in working_slot).
1768 	 * For CREATE, there is no old slot.  For DELETE and FLAGS_ONLY, the
1769 	 * old slot is detached but otherwise preserved.
1770 	 */
1771 	if (change == KVM_MR_CREATE)
1772 		kvm_create_memslot(kvm, new);
1773 	else if (change == KVM_MR_DELETE)
1774 		kvm_delete_memslot(kvm, old, invalid_slot);
1775 	else if (change == KVM_MR_MOVE)
1776 		kvm_move_memslot(kvm, old, new, invalid_slot);
1777 	else if (change == KVM_MR_FLAGS_ONLY)
1778 		kvm_update_flags_memslot(kvm, old, new);
1779 	else
1780 		BUG();
1781 
1782 	/* Free the temporary INVALID slot used for DELETE and MOVE. */
1783 	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1784 		kfree(invalid_slot);
1785 
1786 	/*
1787 	 * No need to refresh new->arch, changes after dropping slots_arch_lock
1788 	 * will directly hit the final, active memsot.  Architectures are
1789 	 * responsible for knowing that new->arch may be stale.
1790 	 */
1791 	kvm_commit_memory_region(kvm, old, new, change);
1792 
1793 	return 0;
1794 }
1795 
1796 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1797 				      gfn_t start, gfn_t end)
1798 {
1799 	struct kvm_memslot_iter iter;
1800 
1801 	kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1802 		if (iter.slot->id != id)
1803 			return true;
1804 	}
1805 
1806 	return false;
1807 }
1808 
1809 /*
1810  * Allocate some memory and give it an address in the guest physical address
1811  * space.
1812  *
1813  * Discontiguous memory is allowed, mostly for framebuffers.
1814  *
1815  * Must be called holding kvm->slots_lock for write.
1816  */
1817 int __kvm_set_memory_region(struct kvm *kvm,
1818 			    const struct kvm_userspace_memory_region *mem)
1819 {
1820 	struct kvm_memory_slot *old, *new;
1821 	struct kvm_memslots *slots;
1822 	enum kvm_mr_change change;
1823 	unsigned long npages;
1824 	gfn_t base_gfn;
1825 	int as_id, id;
1826 	int r;
1827 
1828 	r = check_memory_region_flags(mem);
1829 	if (r)
1830 		return r;
1831 
1832 	as_id = mem->slot >> 16;
1833 	id = (u16)mem->slot;
1834 
1835 	/* General sanity checks */
1836 	if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1837 	    (mem->memory_size != (unsigned long)mem->memory_size))
1838 		return -EINVAL;
1839 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1840 		return -EINVAL;
1841 	/* We can read the guest memory with __xxx_user() later on. */
1842 	if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1843 	    (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1844 	     !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1845 			mem->memory_size))
1846 		return -EINVAL;
1847 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1848 		return -EINVAL;
1849 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1850 		return -EINVAL;
1851 	if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1852 		return -EINVAL;
1853 
1854 	slots = __kvm_memslots(kvm, as_id);
1855 
1856 	/*
1857 	 * Note, the old memslot (and the pointer itself!) may be invalidated
1858 	 * and/or destroyed by kvm_set_memslot().
1859 	 */
1860 	old = id_to_memslot(slots, id);
1861 
1862 	if (!mem->memory_size) {
1863 		if (!old || !old->npages)
1864 			return -EINVAL;
1865 
1866 		if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1867 			return -EIO;
1868 
1869 		return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1870 	}
1871 
1872 	base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1873 	npages = (mem->memory_size >> PAGE_SHIFT);
1874 
1875 	if (!old || !old->npages) {
1876 		change = KVM_MR_CREATE;
1877 
1878 		/*
1879 		 * To simplify KVM internals, the total number of pages across
1880 		 * all memslots must fit in an unsigned long.
1881 		 */
1882 		if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1883 			return -EINVAL;
1884 	} else { /* Modify an existing slot. */
1885 		if ((mem->userspace_addr != old->userspace_addr) ||
1886 		    (npages != old->npages) ||
1887 		    ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1888 			return -EINVAL;
1889 
1890 		if (base_gfn != old->base_gfn)
1891 			change = KVM_MR_MOVE;
1892 		else if (mem->flags != old->flags)
1893 			change = KVM_MR_FLAGS_ONLY;
1894 		else /* Nothing to change. */
1895 			return 0;
1896 	}
1897 
1898 	if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1899 	    kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1900 		return -EEXIST;
1901 
1902 	/* Allocate a slot that will persist in the memslot. */
1903 	new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1904 	if (!new)
1905 		return -ENOMEM;
1906 
1907 	new->as_id = as_id;
1908 	new->id = id;
1909 	new->base_gfn = base_gfn;
1910 	new->npages = npages;
1911 	new->flags = mem->flags;
1912 	new->userspace_addr = mem->userspace_addr;
1913 
1914 	r = kvm_set_memslot(kvm, old, new, change);
1915 	if (r)
1916 		kfree(new);
1917 	return r;
1918 }
1919 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1920 
1921 int kvm_set_memory_region(struct kvm *kvm,
1922 			  const struct kvm_userspace_memory_region *mem)
1923 {
1924 	int r;
1925 
1926 	mutex_lock(&kvm->slots_lock);
1927 	r = __kvm_set_memory_region(kvm, mem);
1928 	mutex_unlock(&kvm->slots_lock);
1929 	return r;
1930 }
1931 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1932 
1933 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1934 					  struct kvm_userspace_memory_region *mem)
1935 {
1936 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1937 		return -EINVAL;
1938 
1939 	return kvm_set_memory_region(kvm, mem);
1940 }
1941 
1942 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1943 /**
1944  * kvm_get_dirty_log - get a snapshot of dirty pages
1945  * @kvm:	pointer to kvm instance
1946  * @log:	slot id and address to which we copy the log
1947  * @is_dirty:	set to '1' if any dirty pages were found
1948  * @memslot:	set to the associated memslot, always valid on success
1949  */
1950 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1951 		      int *is_dirty, struct kvm_memory_slot **memslot)
1952 {
1953 	struct kvm_memslots *slots;
1954 	int i, as_id, id;
1955 	unsigned long n;
1956 	unsigned long any = 0;
1957 
1958 	/* Dirty ring tracking is exclusive to dirty log tracking */
1959 	if (kvm->dirty_ring_size)
1960 		return -ENXIO;
1961 
1962 	*memslot = NULL;
1963 	*is_dirty = 0;
1964 
1965 	as_id = log->slot >> 16;
1966 	id = (u16)log->slot;
1967 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1968 		return -EINVAL;
1969 
1970 	slots = __kvm_memslots(kvm, as_id);
1971 	*memslot = id_to_memslot(slots, id);
1972 	if (!(*memslot) || !(*memslot)->dirty_bitmap)
1973 		return -ENOENT;
1974 
1975 	kvm_arch_sync_dirty_log(kvm, *memslot);
1976 
1977 	n = kvm_dirty_bitmap_bytes(*memslot);
1978 
1979 	for (i = 0; !any && i < n/sizeof(long); ++i)
1980 		any = (*memslot)->dirty_bitmap[i];
1981 
1982 	if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
1983 		return -EFAULT;
1984 
1985 	if (any)
1986 		*is_dirty = 1;
1987 	return 0;
1988 }
1989 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1990 
1991 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
1992 /**
1993  * kvm_get_dirty_log_protect - get a snapshot of dirty pages
1994  *	and reenable dirty page tracking for the corresponding pages.
1995  * @kvm:	pointer to kvm instance
1996  * @log:	slot id and address to which we copy the log
1997  *
1998  * We need to keep it in mind that VCPU threads can write to the bitmap
1999  * concurrently. So, to avoid losing track of dirty pages we keep the
2000  * following order:
2001  *
2002  *    1. Take a snapshot of the bit and clear it if needed.
2003  *    2. Write protect the corresponding page.
2004  *    3. Copy the snapshot to the userspace.
2005  *    4. Upon return caller flushes TLB's if needed.
2006  *
2007  * Between 2 and 4, the guest may write to the page using the remaining TLB
2008  * entry.  This is not a problem because the page is reported dirty using
2009  * the snapshot taken before and step 4 ensures that writes done after
2010  * exiting to userspace will be logged for the next call.
2011  *
2012  */
2013 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2014 {
2015 	struct kvm_memslots *slots;
2016 	struct kvm_memory_slot *memslot;
2017 	int i, as_id, id;
2018 	unsigned long n;
2019 	unsigned long *dirty_bitmap;
2020 	unsigned long *dirty_bitmap_buffer;
2021 	bool flush;
2022 
2023 	/* Dirty ring tracking is exclusive to dirty log tracking */
2024 	if (kvm->dirty_ring_size)
2025 		return -ENXIO;
2026 
2027 	as_id = log->slot >> 16;
2028 	id = (u16)log->slot;
2029 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2030 		return -EINVAL;
2031 
2032 	slots = __kvm_memslots(kvm, as_id);
2033 	memslot = id_to_memslot(slots, id);
2034 	if (!memslot || !memslot->dirty_bitmap)
2035 		return -ENOENT;
2036 
2037 	dirty_bitmap = memslot->dirty_bitmap;
2038 
2039 	kvm_arch_sync_dirty_log(kvm, memslot);
2040 
2041 	n = kvm_dirty_bitmap_bytes(memslot);
2042 	flush = false;
2043 	if (kvm->manual_dirty_log_protect) {
2044 		/*
2045 		 * Unlike kvm_get_dirty_log, we always return false in *flush,
2046 		 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
2047 		 * is some code duplication between this function and
2048 		 * kvm_get_dirty_log, but hopefully all architecture
2049 		 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2050 		 * can be eliminated.
2051 		 */
2052 		dirty_bitmap_buffer = dirty_bitmap;
2053 	} else {
2054 		dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2055 		memset(dirty_bitmap_buffer, 0, n);
2056 
2057 		KVM_MMU_LOCK(kvm);
2058 		for (i = 0; i < n / sizeof(long); i++) {
2059 			unsigned long mask;
2060 			gfn_t offset;
2061 
2062 			if (!dirty_bitmap[i])
2063 				continue;
2064 
2065 			flush = true;
2066 			mask = xchg(&dirty_bitmap[i], 0);
2067 			dirty_bitmap_buffer[i] = mask;
2068 
2069 			offset = i * BITS_PER_LONG;
2070 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2071 								offset, mask);
2072 		}
2073 		KVM_MMU_UNLOCK(kvm);
2074 	}
2075 
2076 	if (flush)
2077 		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2078 
2079 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2080 		return -EFAULT;
2081 	return 0;
2082 }
2083 
2084 
2085 /**
2086  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2087  * @kvm: kvm instance
2088  * @log: slot id and address to which we copy the log
2089  *
2090  * Steps 1-4 below provide general overview of dirty page logging. See
2091  * kvm_get_dirty_log_protect() function description for additional details.
2092  *
2093  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2094  * always flush the TLB (step 4) even if previous step failed  and the dirty
2095  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2096  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2097  * writes will be marked dirty for next log read.
2098  *
2099  *   1. Take a snapshot of the bit and clear it if needed.
2100  *   2. Write protect the corresponding page.
2101  *   3. Copy the snapshot to the userspace.
2102  *   4. Flush TLB's if needed.
2103  */
2104 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2105 				      struct kvm_dirty_log *log)
2106 {
2107 	int r;
2108 
2109 	mutex_lock(&kvm->slots_lock);
2110 
2111 	r = kvm_get_dirty_log_protect(kvm, log);
2112 
2113 	mutex_unlock(&kvm->slots_lock);
2114 	return r;
2115 }
2116 
2117 /**
2118  * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2119  *	and reenable dirty page tracking for the corresponding pages.
2120  * @kvm:	pointer to kvm instance
2121  * @log:	slot id and address from which to fetch the bitmap of dirty pages
2122  */
2123 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2124 				       struct kvm_clear_dirty_log *log)
2125 {
2126 	struct kvm_memslots *slots;
2127 	struct kvm_memory_slot *memslot;
2128 	int as_id, id;
2129 	gfn_t offset;
2130 	unsigned long i, n;
2131 	unsigned long *dirty_bitmap;
2132 	unsigned long *dirty_bitmap_buffer;
2133 	bool flush;
2134 
2135 	/* Dirty ring tracking is exclusive to dirty log tracking */
2136 	if (kvm->dirty_ring_size)
2137 		return -ENXIO;
2138 
2139 	as_id = log->slot >> 16;
2140 	id = (u16)log->slot;
2141 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2142 		return -EINVAL;
2143 
2144 	if (log->first_page & 63)
2145 		return -EINVAL;
2146 
2147 	slots = __kvm_memslots(kvm, as_id);
2148 	memslot = id_to_memslot(slots, id);
2149 	if (!memslot || !memslot->dirty_bitmap)
2150 		return -ENOENT;
2151 
2152 	dirty_bitmap = memslot->dirty_bitmap;
2153 
2154 	n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2155 
2156 	if (log->first_page > memslot->npages ||
2157 	    log->num_pages > memslot->npages - log->first_page ||
2158 	    (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2159 	    return -EINVAL;
2160 
2161 	kvm_arch_sync_dirty_log(kvm, memslot);
2162 
2163 	flush = false;
2164 	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2165 	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2166 		return -EFAULT;
2167 
2168 	KVM_MMU_LOCK(kvm);
2169 	for (offset = log->first_page, i = offset / BITS_PER_LONG,
2170 		 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2171 	     i++, offset += BITS_PER_LONG) {
2172 		unsigned long mask = *dirty_bitmap_buffer++;
2173 		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2174 		if (!mask)
2175 			continue;
2176 
2177 		mask &= atomic_long_fetch_andnot(mask, p);
2178 
2179 		/*
2180 		 * mask contains the bits that really have been cleared.  This
2181 		 * never includes any bits beyond the length of the memslot (if
2182 		 * the length is not aligned to 64 pages), therefore it is not
2183 		 * a problem if userspace sets them in log->dirty_bitmap.
2184 		*/
2185 		if (mask) {
2186 			flush = true;
2187 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2188 								offset, mask);
2189 		}
2190 	}
2191 	KVM_MMU_UNLOCK(kvm);
2192 
2193 	if (flush)
2194 		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2195 
2196 	return 0;
2197 }
2198 
2199 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2200 					struct kvm_clear_dirty_log *log)
2201 {
2202 	int r;
2203 
2204 	mutex_lock(&kvm->slots_lock);
2205 
2206 	r = kvm_clear_dirty_log_protect(kvm, log);
2207 
2208 	mutex_unlock(&kvm->slots_lock);
2209 	return r;
2210 }
2211 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2212 
2213 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2214 {
2215 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2216 }
2217 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2218 
2219 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2220 {
2221 	struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2222 	u64 gen = slots->generation;
2223 	struct kvm_memory_slot *slot;
2224 
2225 	/*
2226 	 * This also protects against using a memslot from a different address space,
2227 	 * since different address spaces have different generation numbers.
2228 	 */
2229 	if (unlikely(gen != vcpu->last_used_slot_gen)) {
2230 		vcpu->last_used_slot = NULL;
2231 		vcpu->last_used_slot_gen = gen;
2232 	}
2233 
2234 	slot = try_get_memslot(vcpu->last_used_slot, gfn);
2235 	if (slot)
2236 		return slot;
2237 
2238 	/*
2239 	 * Fall back to searching all memslots. We purposely use
2240 	 * search_memslots() instead of __gfn_to_memslot() to avoid
2241 	 * thrashing the VM-wide last_used_slot in kvm_memslots.
2242 	 */
2243 	slot = search_memslots(slots, gfn, false);
2244 	if (slot) {
2245 		vcpu->last_used_slot = slot;
2246 		return slot;
2247 	}
2248 
2249 	return NULL;
2250 }
2251 
2252 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2253 {
2254 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2255 
2256 	return kvm_is_visible_memslot(memslot);
2257 }
2258 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2259 
2260 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2261 {
2262 	struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2263 
2264 	return kvm_is_visible_memslot(memslot);
2265 }
2266 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2267 
2268 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2269 {
2270 	struct vm_area_struct *vma;
2271 	unsigned long addr, size;
2272 
2273 	size = PAGE_SIZE;
2274 
2275 	addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2276 	if (kvm_is_error_hva(addr))
2277 		return PAGE_SIZE;
2278 
2279 	mmap_read_lock(current->mm);
2280 	vma = find_vma(current->mm, addr);
2281 	if (!vma)
2282 		goto out;
2283 
2284 	size = vma_kernel_pagesize(vma);
2285 
2286 out:
2287 	mmap_read_unlock(current->mm);
2288 
2289 	return size;
2290 }
2291 
2292 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2293 {
2294 	return slot->flags & KVM_MEM_READONLY;
2295 }
2296 
2297 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2298 				       gfn_t *nr_pages, bool write)
2299 {
2300 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2301 		return KVM_HVA_ERR_BAD;
2302 
2303 	if (memslot_is_readonly(slot) && write)
2304 		return KVM_HVA_ERR_RO_BAD;
2305 
2306 	if (nr_pages)
2307 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
2308 
2309 	return __gfn_to_hva_memslot(slot, gfn);
2310 }
2311 
2312 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2313 				     gfn_t *nr_pages)
2314 {
2315 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2316 }
2317 
2318 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2319 					gfn_t gfn)
2320 {
2321 	return gfn_to_hva_many(slot, gfn, NULL);
2322 }
2323 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2324 
2325 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2326 {
2327 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2328 }
2329 EXPORT_SYMBOL_GPL(gfn_to_hva);
2330 
2331 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2332 {
2333 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2334 }
2335 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2336 
2337 /*
2338  * Return the hva of a @gfn and the R/W attribute if possible.
2339  *
2340  * @slot: the kvm_memory_slot which contains @gfn
2341  * @gfn: the gfn to be translated
2342  * @writable: used to return the read/write attribute of the @slot if the hva
2343  * is valid and @writable is not NULL
2344  */
2345 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2346 				      gfn_t gfn, bool *writable)
2347 {
2348 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2349 
2350 	if (!kvm_is_error_hva(hva) && writable)
2351 		*writable = !memslot_is_readonly(slot);
2352 
2353 	return hva;
2354 }
2355 
2356 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2357 {
2358 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2359 
2360 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2361 }
2362 
2363 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2364 {
2365 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2366 
2367 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2368 }
2369 
2370 static inline int check_user_page_hwpoison(unsigned long addr)
2371 {
2372 	int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2373 
2374 	rc = get_user_pages(addr, 1, flags, NULL, NULL);
2375 	return rc == -EHWPOISON;
2376 }
2377 
2378 /*
2379  * The fast path to get the writable pfn which will be stored in @pfn,
2380  * true indicates success, otherwise false is returned.  It's also the
2381  * only part that runs if we can in atomic context.
2382  */
2383 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2384 			    bool *writable, kvm_pfn_t *pfn)
2385 {
2386 	struct page *page[1];
2387 
2388 	/*
2389 	 * Fast pin a writable pfn only if it is a write fault request
2390 	 * or the caller allows to map a writable pfn for a read fault
2391 	 * request.
2392 	 */
2393 	if (!(write_fault || writable))
2394 		return false;
2395 
2396 	if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2397 		*pfn = page_to_pfn(page[0]);
2398 
2399 		if (writable)
2400 			*writable = true;
2401 		return true;
2402 	}
2403 
2404 	return false;
2405 }
2406 
2407 /*
2408  * The slow path to get the pfn of the specified host virtual address,
2409  * 1 indicates success, -errno is returned if error is detected.
2410  */
2411 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2412 			   bool *writable, kvm_pfn_t *pfn)
2413 {
2414 	unsigned int flags = FOLL_HWPOISON;
2415 	struct page *page;
2416 	int npages = 0;
2417 
2418 	might_sleep();
2419 
2420 	if (writable)
2421 		*writable = write_fault;
2422 
2423 	if (write_fault)
2424 		flags |= FOLL_WRITE;
2425 	if (async)
2426 		flags |= FOLL_NOWAIT;
2427 
2428 	npages = get_user_pages_unlocked(addr, 1, &page, flags);
2429 	if (npages != 1)
2430 		return npages;
2431 
2432 	/* map read fault as writable if possible */
2433 	if (unlikely(!write_fault) && writable) {
2434 		struct page *wpage;
2435 
2436 		if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2437 			*writable = true;
2438 			put_page(page);
2439 			page = wpage;
2440 		}
2441 	}
2442 	*pfn = page_to_pfn(page);
2443 	return npages;
2444 }
2445 
2446 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2447 {
2448 	if (unlikely(!(vma->vm_flags & VM_READ)))
2449 		return false;
2450 
2451 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2452 		return false;
2453 
2454 	return true;
2455 }
2456 
2457 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2458 {
2459 	if (kvm_is_reserved_pfn(pfn))
2460 		return 1;
2461 	return get_page_unless_zero(pfn_to_page(pfn));
2462 }
2463 
2464 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2465 			       unsigned long addr, bool write_fault,
2466 			       bool *writable, kvm_pfn_t *p_pfn)
2467 {
2468 	kvm_pfn_t pfn;
2469 	pte_t *ptep;
2470 	spinlock_t *ptl;
2471 	int r;
2472 
2473 	r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2474 	if (r) {
2475 		/*
2476 		 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2477 		 * not call the fault handler, so do it here.
2478 		 */
2479 		bool unlocked = false;
2480 		r = fixup_user_fault(current->mm, addr,
2481 				     (write_fault ? FAULT_FLAG_WRITE : 0),
2482 				     &unlocked);
2483 		if (unlocked)
2484 			return -EAGAIN;
2485 		if (r)
2486 			return r;
2487 
2488 		r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2489 		if (r)
2490 			return r;
2491 	}
2492 
2493 	if (write_fault && !pte_write(*ptep)) {
2494 		pfn = KVM_PFN_ERR_RO_FAULT;
2495 		goto out;
2496 	}
2497 
2498 	if (writable)
2499 		*writable = pte_write(*ptep);
2500 	pfn = pte_pfn(*ptep);
2501 
2502 	/*
2503 	 * Get a reference here because callers of *hva_to_pfn* and
2504 	 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2505 	 * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
2506 	 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2507 	 * simply do nothing for reserved pfns.
2508 	 *
2509 	 * Whoever called remap_pfn_range is also going to call e.g.
2510 	 * unmap_mapping_range before the underlying pages are freed,
2511 	 * causing a call to our MMU notifier.
2512 	 *
2513 	 * Certain IO or PFNMAP mappings can be backed with valid
2514 	 * struct pages, but be allocated without refcounting e.g.,
2515 	 * tail pages of non-compound higher order allocations, which
2516 	 * would then underflow the refcount when the caller does the
2517 	 * required put_page. Don't allow those pages here.
2518 	 */
2519 	if (!kvm_try_get_pfn(pfn))
2520 		r = -EFAULT;
2521 
2522 out:
2523 	pte_unmap_unlock(ptep, ptl);
2524 	*p_pfn = pfn;
2525 
2526 	return r;
2527 }
2528 
2529 /*
2530  * Pin guest page in memory and return its pfn.
2531  * @addr: host virtual address which maps memory to the guest
2532  * @atomic: whether this function can sleep
2533  * @async: whether this function need to wait IO complete if the
2534  *         host page is not in the memory
2535  * @write_fault: whether we should get a writable host page
2536  * @writable: whether it allows to map a writable host page for !@write_fault
2537  *
2538  * The function will map a writable host page for these two cases:
2539  * 1): @write_fault = true
2540  * 2): @write_fault = false && @writable, @writable will tell the caller
2541  *     whether the mapping is writable.
2542  */
2543 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2544 		     bool write_fault, bool *writable)
2545 {
2546 	struct vm_area_struct *vma;
2547 	kvm_pfn_t pfn = 0;
2548 	int npages, r;
2549 
2550 	/* we can do it either atomically or asynchronously, not both */
2551 	BUG_ON(atomic && async);
2552 
2553 	if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2554 		return pfn;
2555 
2556 	if (atomic)
2557 		return KVM_PFN_ERR_FAULT;
2558 
2559 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2560 	if (npages == 1)
2561 		return pfn;
2562 
2563 	mmap_read_lock(current->mm);
2564 	if (npages == -EHWPOISON ||
2565 	      (!async && check_user_page_hwpoison(addr))) {
2566 		pfn = KVM_PFN_ERR_HWPOISON;
2567 		goto exit;
2568 	}
2569 
2570 retry:
2571 	vma = vma_lookup(current->mm, addr);
2572 
2573 	if (vma == NULL)
2574 		pfn = KVM_PFN_ERR_FAULT;
2575 	else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2576 		r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2577 		if (r == -EAGAIN)
2578 			goto retry;
2579 		if (r < 0)
2580 			pfn = KVM_PFN_ERR_FAULT;
2581 	} else {
2582 		if (async && vma_is_valid(vma, write_fault))
2583 			*async = true;
2584 		pfn = KVM_PFN_ERR_FAULT;
2585 	}
2586 exit:
2587 	mmap_read_unlock(current->mm);
2588 	return pfn;
2589 }
2590 
2591 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2592 			       bool atomic, bool *async, bool write_fault,
2593 			       bool *writable, hva_t *hva)
2594 {
2595 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2596 
2597 	if (hva)
2598 		*hva = addr;
2599 
2600 	if (addr == KVM_HVA_ERR_RO_BAD) {
2601 		if (writable)
2602 			*writable = false;
2603 		return KVM_PFN_ERR_RO_FAULT;
2604 	}
2605 
2606 	if (kvm_is_error_hva(addr)) {
2607 		if (writable)
2608 			*writable = false;
2609 		return KVM_PFN_NOSLOT;
2610 	}
2611 
2612 	/* Do not map writable pfn in the readonly memslot. */
2613 	if (writable && memslot_is_readonly(slot)) {
2614 		*writable = false;
2615 		writable = NULL;
2616 	}
2617 
2618 	return hva_to_pfn(addr, atomic, async, write_fault,
2619 			  writable);
2620 }
2621 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2622 
2623 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2624 		      bool *writable)
2625 {
2626 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2627 				    write_fault, writable, NULL);
2628 }
2629 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2630 
2631 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2632 {
2633 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2634 }
2635 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2636 
2637 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2638 {
2639 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2640 }
2641 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2642 
2643 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2644 {
2645 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2646 }
2647 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2648 
2649 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2650 {
2651 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2652 }
2653 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2654 
2655 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2656 {
2657 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2658 }
2659 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2660 
2661 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2662 			    struct page **pages, int nr_pages)
2663 {
2664 	unsigned long addr;
2665 	gfn_t entry = 0;
2666 
2667 	addr = gfn_to_hva_many(slot, gfn, &entry);
2668 	if (kvm_is_error_hva(addr))
2669 		return -1;
2670 
2671 	if (entry < nr_pages)
2672 		return 0;
2673 
2674 	return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2675 }
2676 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2677 
2678 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2679 {
2680 	if (is_error_noslot_pfn(pfn))
2681 		return KVM_ERR_PTR_BAD_PAGE;
2682 
2683 	if (kvm_is_reserved_pfn(pfn)) {
2684 		WARN_ON(1);
2685 		return KVM_ERR_PTR_BAD_PAGE;
2686 	}
2687 
2688 	return pfn_to_page(pfn);
2689 }
2690 
2691 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2692 {
2693 	kvm_pfn_t pfn;
2694 
2695 	pfn = gfn_to_pfn(kvm, gfn);
2696 
2697 	return kvm_pfn_to_page(pfn);
2698 }
2699 EXPORT_SYMBOL_GPL(gfn_to_page);
2700 
2701 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2702 {
2703 	if (pfn == 0)
2704 		return;
2705 
2706 	if (dirty)
2707 		kvm_release_pfn_dirty(pfn);
2708 	else
2709 		kvm_release_pfn_clean(pfn);
2710 }
2711 
2712 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2713 {
2714 	kvm_pfn_t pfn;
2715 	void *hva = NULL;
2716 	struct page *page = KVM_UNMAPPED_PAGE;
2717 
2718 	if (!map)
2719 		return -EINVAL;
2720 
2721 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
2722 	if (is_error_noslot_pfn(pfn))
2723 		return -EINVAL;
2724 
2725 	if (pfn_valid(pfn)) {
2726 		page = pfn_to_page(pfn);
2727 		hva = kmap(page);
2728 #ifdef CONFIG_HAS_IOMEM
2729 	} else {
2730 		hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2731 #endif
2732 	}
2733 
2734 	if (!hva)
2735 		return -EFAULT;
2736 
2737 	map->page = page;
2738 	map->hva = hva;
2739 	map->pfn = pfn;
2740 	map->gfn = gfn;
2741 
2742 	return 0;
2743 }
2744 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2745 
2746 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2747 {
2748 	if (!map)
2749 		return;
2750 
2751 	if (!map->hva)
2752 		return;
2753 
2754 	if (map->page != KVM_UNMAPPED_PAGE)
2755 		kunmap(map->page);
2756 #ifdef CONFIG_HAS_IOMEM
2757 	else
2758 		memunmap(map->hva);
2759 #endif
2760 
2761 	if (dirty)
2762 		kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2763 
2764 	kvm_release_pfn(map->pfn, dirty);
2765 
2766 	map->hva = NULL;
2767 	map->page = NULL;
2768 }
2769 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2770 
2771 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2772 {
2773 	kvm_pfn_t pfn;
2774 
2775 	pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2776 
2777 	return kvm_pfn_to_page(pfn);
2778 }
2779 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2780 
2781 void kvm_release_page_clean(struct page *page)
2782 {
2783 	WARN_ON(is_error_page(page));
2784 
2785 	kvm_release_pfn_clean(page_to_pfn(page));
2786 }
2787 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2788 
2789 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2790 {
2791 	if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2792 		put_page(pfn_to_page(pfn));
2793 }
2794 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2795 
2796 void kvm_release_page_dirty(struct page *page)
2797 {
2798 	WARN_ON(is_error_page(page));
2799 
2800 	kvm_release_pfn_dirty(page_to_pfn(page));
2801 }
2802 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2803 
2804 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2805 {
2806 	kvm_set_pfn_dirty(pfn);
2807 	kvm_release_pfn_clean(pfn);
2808 }
2809 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2810 
2811 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2812 {
2813 	if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2814 		SetPageDirty(pfn_to_page(pfn));
2815 }
2816 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2817 
2818 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2819 {
2820 	if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2821 		mark_page_accessed(pfn_to_page(pfn));
2822 }
2823 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2824 
2825 static int next_segment(unsigned long len, int offset)
2826 {
2827 	if (len > PAGE_SIZE - offset)
2828 		return PAGE_SIZE - offset;
2829 	else
2830 		return len;
2831 }
2832 
2833 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2834 				 void *data, int offset, int len)
2835 {
2836 	int r;
2837 	unsigned long addr;
2838 
2839 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2840 	if (kvm_is_error_hva(addr))
2841 		return -EFAULT;
2842 	r = __copy_from_user(data, (void __user *)addr + offset, len);
2843 	if (r)
2844 		return -EFAULT;
2845 	return 0;
2846 }
2847 
2848 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2849 			int len)
2850 {
2851 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2852 
2853 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
2854 }
2855 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2856 
2857 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2858 			     int offset, int len)
2859 {
2860 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2861 
2862 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
2863 }
2864 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2865 
2866 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2867 {
2868 	gfn_t gfn = gpa >> PAGE_SHIFT;
2869 	int seg;
2870 	int offset = offset_in_page(gpa);
2871 	int ret;
2872 
2873 	while ((seg = next_segment(len, offset)) != 0) {
2874 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2875 		if (ret < 0)
2876 			return ret;
2877 		offset = 0;
2878 		len -= seg;
2879 		data += seg;
2880 		++gfn;
2881 	}
2882 	return 0;
2883 }
2884 EXPORT_SYMBOL_GPL(kvm_read_guest);
2885 
2886 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2887 {
2888 	gfn_t gfn = gpa >> PAGE_SHIFT;
2889 	int seg;
2890 	int offset = offset_in_page(gpa);
2891 	int ret;
2892 
2893 	while ((seg = next_segment(len, offset)) != 0) {
2894 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2895 		if (ret < 0)
2896 			return ret;
2897 		offset = 0;
2898 		len -= seg;
2899 		data += seg;
2900 		++gfn;
2901 	}
2902 	return 0;
2903 }
2904 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2905 
2906 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2907 			           void *data, int offset, unsigned long len)
2908 {
2909 	int r;
2910 	unsigned long addr;
2911 
2912 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2913 	if (kvm_is_error_hva(addr))
2914 		return -EFAULT;
2915 	pagefault_disable();
2916 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2917 	pagefault_enable();
2918 	if (r)
2919 		return -EFAULT;
2920 	return 0;
2921 }
2922 
2923 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2924 			       void *data, unsigned long len)
2925 {
2926 	gfn_t gfn = gpa >> PAGE_SHIFT;
2927 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2928 	int offset = offset_in_page(gpa);
2929 
2930 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2931 }
2932 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2933 
2934 static int __kvm_write_guest_page(struct kvm *kvm,
2935 				  struct kvm_memory_slot *memslot, gfn_t gfn,
2936 			          const void *data, int offset, int len)
2937 {
2938 	int r;
2939 	unsigned long addr;
2940 
2941 	addr = gfn_to_hva_memslot(memslot, gfn);
2942 	if (kvm_is_error_hva(addr))
2943 		return -EFAULT;
2944 	r = __copy_to_user((void __user *)addr + offset, data, len);
2945 	if (r)
2946 		return -EFAULT;
2947 	mark_page_dirty_in_slot(kvm, memslot, gfn);
2948 	return 0;
2949 }
2950 
2951 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2952 			 const void *data, int offset, int len)
2953 {
2954 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2955 
2956 	return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2957 }
2958 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2959 
2960 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2961 			      const void *data, int offset, int len)
2962 {
2963 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2964 
2965 	return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
2966 }
2967 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
2968 
2969 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
2970 		    unsigned long len)
2971 {
2972 	gfn_t gfn = gpa >> PAGE_SHIFT;
2973 	int seg;
2974 	int offset = offset_in_page(gpa);
2975 	int ret;
2976 
2977 	while ((seg = next_segment(len, offset)) != 0) {
2978 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
2979 		if (ret < 0)
2980 			return ret;
2981 		offset = 0;
2982 		len -= seg;
2983 		data += seg;
2984 		++gfn;
2985 	}
2986 	return 0;
2987 }
2988 EXPORT_SYMBOL_GPL(kvm_write_guest);
2989 
2990 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
2991 		         unsigned long len)
2992 {
2993 	gfn_t gfn = gpa >> PAGE_SHIFT;
2994 	int seg;
2995 	int offset = offset_in_page(gpa);
2996 	int ret;
2997 
2998 	while ((seg = next_segment(len, offset)) != 0) {
2999 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3000 		if (ret < 0)
3001 			return ret;
3002 		offset = 0;
3003 		len -= seg;
3004 		data += seg;
3005 		++gfn;
3006 	}
3007 	return 0;
3008 }
3009 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3010 
3011 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3012 				       struct gfn_to_hva_cache *ghc,
3013 				       gpa_t gpa, unsigned long len)
3014 {
3015 	int offset = offset_in_page(gpa);
3016 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
3017 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3018 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3019 	gfn_t nr_pages_avail;
3020 
3021 	/* Update ghc->generation before performing any error checks. */
3022 	ghc->generation = slots->generation;
3023 
3024 	if (start_gfn > end_gfn) {
3025 		ghc->hva = KVM_HVA_ERR_BAD;
3026 		return -EINVAL;
3027 	}
3028 
3029 	/*
3030 	 * If the requested region crosses two memslots, we still
3031 	 * verify that the entire region is valid here.
3032 	 */
3033 	for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3034 		ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3035 		ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3036 					   &nr_pages_avail);
3037 		if (kvm_is_error_hva(ghc->hva))
3038 			return -EFAULT;
3039 	}
3040 
3041 	/* Use the slow path for cross page reads and writes. */
3042 	if (nr_pages_needed == 1)
3043 		ghc->hva += offset;
3044 	else
3045 		ghc->memslot = NULL;
3046 
3047 	ghc->gpa = gpa;
3048 	ghc->len = len;
3049 	return 0;
3050 }
3051 
3052 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3053 			      gpa_t gpa, unsigned long len)
3054 {
3055 	struct kvm_memslots *slots = kvm_memslots(kvm);
3056 	return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3057 }
3058 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3059 
3060 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3061 				  void *data, unsigned int offset,
3062 				  unsigned long len)
3063 {
3064 	struct kvm_memslots *slots = kvm_memslots(kvm);
3065 	int r;
3066 	gpa_t gpa = ghc->gpa + offset;
3067 
3068 	if (WARN_ON_ONCE(len + offset > ghc->len))
3069 		return -EINVAL;
3070 
3071 	if (slots->generation != ghc->generation) {
3072 		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3073 			return -EFAULT;
3074 	}
3075 
3076 	if (kvm_is_error_hva(ghc->hva))
3077 		return -EFAULT;
3078 
3079 	if (unlikely(!ghc->memslot))
3080 		return kvm_write_guest(kvm, gpa, data, len);
3081 
3082 	r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3083 	if (r)
3084 		return -EFAULT;
3085 	mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3086 
3087 	return 0;
3088 }
3089 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3090 
3091 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3092 			   void *data, unsigned long len)
3093 {
3094 	return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3095 }
3096 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3097 
3098 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3099 				 void *data, unsigned int offset,
3100 				 unsigned long len)
3101 {
3102 	struct kvm_memslots *slots = kvm_memslots(kvm);
3103 	int r;
3104 	gpa_t gpa = ghc->gpa + offset;
3105 
3106 	if (WARN_ON_ONCE(len + offset > ghc->len))
3107 		return -EINVAL;
3108 
3109 	if (slots->generation != ghc->generation) {
3110 		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3111 			return -EFAULT;
3112 	}
3113 
3114 	if (kvm_is_error_hva(ghc->hva))
3115 		return -EFAULT;
3116 
3117 	if (unlikely(!ghc->memslot))
3118 		return kvm_read_guest(kvm, gpa, data, len);
3119 
3120 	r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3121 	if (r)
3122 		return -EFAULT;
3123 
3124 	return 0;
3125 }
3126 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3127 
3128 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3129 			  void *data, unsigned long len)
3130 {
3131 	return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3132 }
3133 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3134 
3135 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3136 {
3137 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3138 	gfn_t gfn = gpa >> PAGE_SHIFT;
3139 	int seg;
3140 	int offset = offset_in_page(gpa);
3141 	int ret;
3142 
3143 	while ((seg = next_segment(len, offset)) != 0) {
3144 		ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3145 		if (ret < 0)
3146 			return ret;
3147 		offset = 0;
3148 		len -= seg;
3149 		++gfn;
3150 	}
3151 	return 0;
3152 }
3153 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3154 
3155 void mark_page_dirty_in_slot(struct kvm *kvm,
3156 			     const struct kvm_memory_slot *memslot,
3157 		 	     gfn_t gfn)
3158 {
3159 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3160 
3161 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3162 	if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3163 		return;
3164 #endif
3165 
3166 	if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3167 		unsigned long rel_gfn = gfn - memslot->base_gfn;
3168 		u32 slot = (memslot->as_id << 16) | memslot->id;
3169 
3170 		if (kvm->dirty_ring_size)
3171 			kvm_dirty_ring_push(&vcpu->dirty_ring,
3172 					    slot, rel_gfn);
3173 		else
3174 			set_bit_le(rel_gfn, memslot->dirty_bitmap);
3175 	}
3176 }
3177 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3178 
3179 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3180 {
3181 	struct kvm_memory_slot *memslot;
3182 
3183 	memslot = gfn_to_memslot(kvm, gfn);
3184 	mark_page_dirty_in_slot(kvm, memslot, gfn);
3185 }
3186 EXPORT_SYMBOL_GPL(mark_page_dirty);
3187 
3188 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3189 {
3190 	struct kvm_memory_slot *memslot;
3191 
3192 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3193 	mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3194 }
3195 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3196 
3197 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3198 {
3199 	if (!vcpu->sigset_active)
3200 		return;
3201 
3202 	/*
3203 	 * This does a lockless modification of ->real_blocked, which is fine
3204 	 * because, only current can change ->real_blocked and all readers of
3205 	 * ->real_blocked don't care as long ->real_blocked is always a subset
3206 	 * of ->blocked.
3207 	 */
3208 	sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
3209 }
3210 
3211 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3212 {
3213 	if (!vcpu->sigset_active)
3214 		return;
3215 
3216 	sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
3217 	sigemptyset(&current->real_blocked);
3218 }
3219 
3220 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3221 {
3222 	unsigned int old, val, grow, grow_start;
3223 
3224 	old = val = vcpu->halt_poll_ns;
3225 	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3226 	grow = READ_ONCE(halt_poll_ns_grow);
3227 	if (!grow)
3228 		goto out;
3229 
3230 	val *= grow;
3231 	if (val < grow_start)
3232 		val = grow_start;
3233 
3234 	if (val > vcpu->kvm->max_halt_poll_ns)
3235 		val = vcpu->kvm->max_halt_poll_ns;
3236 
3237 	vcpu->halt_poll_ns = val;
3238 out:
3239 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3240 }
3241 
3242 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3243 {
3244 	unsigned int old, val, shrink, grow_start;
3245 
3246 	old = val = vcpu->halt_poll_ns;
3247 	shrink = READ_ONCE(halt_poll_ns_shrink);
3248 	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3249 	if (shrink == 0)
3250 		val = 0;
3251 	else
3252 		val /= shrink;
3253 
3254 	if (val < grow_start)
3255 		val = 0;
3256 
3257 	vcpu->halt_poll_ns = val;
3258 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3259 }
3260 
3261 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3262 {
3263 	int ret = -EINTR;
3264 	int idx = srcu_read_lock(&vcpu->kvm->srcu);
3265 
3266 	if (kvm_arch_vcpu_runnable(vcpu)) {
3267 		kvm_make_request(KVM_REQ_UNHALT, vcpu);
3268 		goto out;
3269 	}
3270 	if (kvm_cpu_has_pending_timer(vcpu))
3271 		goto out;
3272 	if (signal_pending(current))
3273 		goto out;
3274 	if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3275 		goto out;
3276 
3277 	ret = 0;
3278 out:
3279 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3280 	return ret;
3281 }
3282 
3283 /*
3284  * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3285  * pending.  This is mostly used when halting a vCPU, but may also be used
3286  * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3287  */
3288 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3289 {
3290 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3291 	bool waited = false;
3292 
3293 	vcpu->stat.generic.blocking = 1;
3294 
3295 	kvm_arch_vcpu_blocking(vcpu);
3296 
3297 	prepare_to_rcuwait(wait);
3298 	for (;;) {
3299 		set_current_state(TASK_INTERRUPTIBLE);
3300 
3301 		if (kvm_vcpu_check_block(vcpu) < 0)
3302 			break;
3303 
3304 		waited = true;
3305 		schedule();
3306 	}
3307 	finish_rcuwait(wait);
3308 
3309 	kvm_arch_vcpu_unblocking(vcpu);
3310 
3311 	vcpu->stat.generic.blocking = 0;
3312 
3313 	return waited;
3314 }
3315 
3316 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3317 					  ktime_t end, bool success)
3318 {
3319 	struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3320 	u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3321 
3322 	++vcpu->stat.generic.halt_attempted_poll;
3323 
3324 	if (success) {
3325 		++vcpu->stat.generic.halt_successful_poll;
3326 
3327 		if (!vcpu_valid_wakeup(vcpu))
3328 			++vcpu->stat.generic.halt_poll_invalid;
3329 
3330 		stats->halt_poll_success_ns += poll_ns;
3331 		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3332 	} else {
3333 		stats->halt_poll_fail_ns += poll_ns;
3334 		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3335 	}
3336 }
3337 
3338 /*
3339  * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc...  If halt
3340  * polling is enabled, busy wait for a short time before blocking to avoid the
3341  * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3342  * is halted.
3343  */
3344 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3345 {
3346 	bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3347 	bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3348 	ktime_t start, cur, poll_end;
3349 	bool waited = false;
3350 	u64 halt_ns;
3351 
3352 	start = cur = poll_end = ktime_get();
3353 	if (do_halt_poll) {
3354 		ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3355 
3356 		do {
3357 			/*
3358 			 * This sets KVM_REQ_UNHALT if an interrupt
3359 			 * arrives.
3360 			 */
3361 			if (kvm_vcpu_check_block(vcpu) < 0)
3362 				goto out;
3363 			cpu_relax();
3364 			poll_end = cur = ktime_get();
3365 		} while (kvm_vcpu_can_poll(cur, stop));
3366 	}
3367 
3368 	waited = kvm_vcpu_block(vcpu);
3369 
3370 	cur = ktime_get();
3371 	if (waited) {
3372 		vcpu->stat.generic.halt_wait_ns +=
3373 			ktime_to_ns(cur) - ktime_to_ns(poll_end);
3374 		KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3375 				ktime_to_ns(cur) - ktime_to_ns(poll_end));
3376 	}
3377 out:
3378 	/* The total time the vCPU was "halted", including polling time. */
3379 	halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3380 
3381 	/*
3382 	 * Note, halt-polling is considered successful so long as the vCPU was
3383 	 * never actually scheduled out, i.e. even if the wake event arrived
3384 	 * after of the halt-polling loop itself, but before the full wait.
3385 	 */
3386 	if (do_halt_poll)
3387 		update_halt_poll_stats(vcpu, start, poll_end, !waited);
3388 
3389 	if (halt_poll_allowed) {
3390 		if (!vcpu_valid_wakeup(vcpu)) {
3391 			shrink_halt_poll_ns(vcpu);
3392 		} else if (vcpu->kvm->max_halt_poll_ns) {
3393 			if (halt_ns <= vcpu->halt_poll_ns)
3394 				;
3395 			/* we had a long block, shrink polling */
3396 			else if (vcpu->halt_poll_ns &&
3397 				 halt_ns > vcpu->kvm->max_halt_poll_ns)
3398 				shrink_halt_poll_ns(vcpu);
3399 			/* we had a short halt and our poll time is too small */
3400 			else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3401 				 halt_ns < vcpu->kvm->max_halt_poll_ns)
3402 				grow_halt_poll_ns(vcpu);
3403 		} else {
3404 			vcpu->halt_poll_ns = 0;
3405 		}
3406 	}
3407 
3408 	trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3409 }
3410 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3411 
3412 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3413 {
3414 	if (__kvm_vcpu_wake_up(vcpu)) {
3415 		WRITE_ONCE(vcpu->ready, true);
3416 		++vcpu->stat.generic.halt_wakeup;
3417 		return true;
3418 	}
3419 
3420 	return false;
3421 }
3422 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3423 
3424 #ifndef CONFIG_S390
3425 /*
3426  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3427  */
3428 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3429 {
3430 	int me, cpu;
3431 
3432 	if (kvm_vcpu_wake_up(vcpu))
3433 		return;
3434 
3435 	me = get_cpu();
3436 	/*
3437 	 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3438 	 * to EXITING_GUEST_MODE.  Therefore the moderately expensive "should
3439 	 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3440 	 * within the vCPU thread itself.
3441 	 */
3442 	if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3443 		if (vcpu->mode == IN_GUEST_MODE)
3444 			WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3445 		goto out;
3446 	}
3447 
3448 	/*
3449 	 * Note, the vCPU could get migrated to a different pCPU at any point
3450 	 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3451 	 * IPI to the previous pCPU.  But, that's ok because the purpose of the
3452 	 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3453 	 * vCPU also requires it to leave IN_GUEST_MODE.
3454 	 */
3455 	if (kvm_arch_vcpu_should_kick(vcpu)) {
3456 		cpu = READ_ONCE(vcpu->cpu);
3457 		if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3458 			smp_send_reschedule(cpu);
3459 	}
3460 out:
3461 	put_cpu();
3462 }
3463 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3464 #endif /* !CONFIG_S390 */
3465 
3466 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3467 {
3468 	struct pid *pid;
3469 	struct task_struct *task = NULL;
3470 	int ret = 0;
3471 
3472 	rcu_read_lock();
3473 	pid = rcu_dereference(target->pid);
3474 	if (pid)
3475 		task = get_pid_task(pid, PIDTYPE_PID);
3476 	rcu_read_unlock();
3477 	if (!task)
3478 		return ret;
3479 	ret = yield_to(task, 1);
3480 	put_task_struct(task);
3481 
3482 	return ret;
3483 }
3484 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3485 
3486 /*
3487  * Helper that checks whether a VCPU is eligible for directed yield.
3488  * Most eligible candidate to yield is decided by following heuristics:
3489  *
3490  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3491  *  (preempted lock holder), indicated by @in_spin_loop.
3492  *  Set at the beginning and cleared at the end of interception/PLE handler.
3493  *
3494  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3495  *  chance last time (mostly it has become eligible now since we have probably
3496  *  yielded to lockholder in last iteration. This is done by toggling
3497  *  @dy_eligible each time a VCPU checked for eligibility.)
3498  *
3499  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3500  *  to preempted lock-holder could result in wrong VCPU selection and CPU
3501  *  burning. Giving priority for a potential lock-holder increases lock
3502  *  progress.
3503  *
3504  *  Since algorithm is based on heuristics, accessing another VCPU data without
3505  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
3506  *  and continue with next VCPU and so on.
3507  */
3508 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3509 {
3510 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3511 	bool eligible;
3512 
3513 	eligible = !vcpu->spin_loop.in_spin_loop ||
3514 		    vcpu->spin_loop.dy_eligible;
3515 
3516 	if (vcpu->spin_loop.in_spin_loop)
3517 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3518 
3519 	return eligible;
3520 #else
3521 	return true;
3522 #endif
3523 }
3524 
3525 /*
3526  * Unlike kvm_arch_vcpu_runnable, this function is called outside
3527  * a vcpu_load/vcpu_put pair.  However, for most architectures
3528  * kvm_arch_vcpu_runnable does not require vcpu_load.
3529  */
3530 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3531 {
3532 	return kvm_arch_vcpu_runnable(vcpu);
3533 }
3534 
3535 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3536 {
3537 	if (kvm_arch_dy_runnable(vcpu))
3538 		return true;
3539 
3540 #ifdef CONFIG_KVM_ASYNC_PF
3541 	if (!list_empty_careful(&vcpu->async_pf.done))
3542 		return true;
3543 #endif
3544 
3545 	return false;
3546 }
3547 
3548 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3549 {
3550 	return false;
3551 }
3552 
3553 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3554 {
3555 	struct kvm *kvm = me->kvm;
3556 	struct kvm_vcpu *vcpu;
3557 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3558 	unsigned long i;
3559 	int yielded = 0;
3560 	int try = 3;
3561 	int pass;
3562 
3563 	kvm_vcpu_set_in_spin_loop(me, true);
3564 	/*
3565 	 * We boost the priority of a VCPU that is runnable but not
3566 	 * currently running, because it got preempted by something
3567 	 * else and called schedule in __vcpu_run.  Hopefully that
3568 	 * VCPU is holding the lock that we need and will release it.
3569 	 * We approximate round-robin by starting at the last boosted VCPU.
3570 	 */
3571 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
3572 		kvm_for_each_vcpu(i, vcpu, kvm) {
3573 			if (!pass && i <= last_boosted_vcpu) {
3574 				i = last_boosted_vcpu;
3575 				continue;
3576 			} else if (pass && i > last_boosted_vcpu)
3577 				break;
3578 			if (!READ_ONCE(vcpu->ready))
3579 				continue;
3580 			if (vcpu == me)
3581 				continue;
3582 			if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3583 				continue;
3584 			if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3585 			    !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3586 			    !kvm_arch_vcpu_in_kernel(vcpu))
3587 				continue;
3588 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3589 				continue;
3590 
3591 			yielded = kvm_vcpu_yield_to(vcpu);
3592 			if (yielded > 0) {
3593 				kvm->last_boosted_vcpu = i;
3594 				break;
3595 			} else if (yielded < 0) {
3596 				try--;
3597 				if (!try)
3598 					break;
3599 			}
3600 		}
3601 	}
3602 	kvm_vcpu_set_in_spin_loop(me, false);
3603 
3604 	/* Ensure vcpu is not eligible during next spinloop */
3605 	kvm_vcpu_set_dy_eligible(me, false);
3606 }
3607 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3608 
3609 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3610 {
3611 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3612 	return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3613 	    (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3614 	     kvm->dirty_ring_size / PAGE_SIZE);
3615 #else
3616 	return false;
3617 #endif
3618 }
3619 
3620 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3621 {
3622 	struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3623 	struct page *page;
3624 
3625 	if (vmf->pgoff == 0)
3626 		page = virt_to_page(vcpu->run);
3627 #ifdef CONFIG_X86
3628 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3629 		page = virt_to_page(vcpu->arch.pio_data);
3630 #endif
3631 #ifdef CONFIG_KVM_MMIO
3632 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3633 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3634 #endif
3635 	else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3636 		page = kvm_dirty_ring_get_page(
3637 		    &vcpu->dirty_ring,
3638 		    vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3639 	else
3640 		return kvm_arch_vcpu_fault(vcpu, vmf);
3641 	get_page(page);
3642 	vmf->page = page;
3643 	return 0;
3644 }
3645 
3646 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3647 	.fault = kvm_vcpu_fault,
3648 };
3649 
3650 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3651 {
3652 	struct kvm_vcpu *vcpu = file->private_data;
3653 	unsigned long pages = vma_pages(vma);
3654 
3655 	if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3656 	     kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3657 	    ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3658 		return -EINVAL;
3659 
3660 	vma->vm_ops = &kvm_vcpu_vm_ops;
3661 	return 0;
3662 }
3663 
3664 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3665 {
3666 	struct kvm_vcpu *vcpu = filp->private_data;
3667 
3668 	kvm_put_kvm(vcpu->kvm);
3669 	return 0;
3670 }
3671 
3672 static struct file_operations kvm_vcpu_fops = {
3673 	.release        = kvm_vcpu_release,
3674 	.unlocked_ioctl = kvm_vcpu_ioctl,
3675 	.mmap           = kvm_vcpu_mmap,
3676 	.llseek		= noop_llseek,
3677 	KVM_COMPAT(kvm_vcpu_compat_ioctl),
3678 };
3679 
3680 /*
3681  * Allocates an inode for the vcpu.
3682  */
3683 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3684 {
3685 	char name[8 + 1 + ITOA_MAX_LEN + 1];
3686 
3687 	snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3688 	return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3689 }
3690 
3691 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3692 {
3693 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3694 	struct dentry *debugfs_dentry;
3695 	char dir_name[ITOA_MAX_LEN * 2];
3696 
3697 	if (!debugfs_initialized())
3698 		return;
3699 
3700 	snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3701 	debugfs_dentry = debugfs_create_dir(dir_name,
3702 					    vcpu->kvm->debugfs_dentry);
3703 
3704 	kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3705 #endif
3706 }
3707 
3708 /*
3709  * Creates some virtual cpus.  Good luck creating more than one.
3710  */
3711 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3712 {
3713 	int r;
3714 	struct kvm_vcpu *vcpu;
3715 	struct page *page;
3716 
3717 	if (id >= KVM_MAX_VCPU_IDS)
3718 		return -EINVAL;
3719 
3720 	mutex_lock(&kvm->lock);
3721 	if (kvm->created_vcpus == KVM_MAX_VCPUS) {
3722 		mutex_unlock(&kvm->lock);
3723 		return -EINVAL;
3724 	}
3725 
3726 	kvm->created_vcpus++;
3727 	mutex_unlock(&kvm->lock);
3728 
3729 	r = kvm_arch_vcpu_precreate(kvm, id);
3730 	if (r)
3731 		goto vcpu_decrement;
3732 
3733 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3734 	if (!vcpu) {
3735 		r = -ENOMEM;
3736 		goto vcpu_decrement;
3737 	}
3738 
3739 	BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3740 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3741 	if (!page) {
3742 		r = -ENOMEM;
3743 		goto vcpu_free;
3744 	}
3745 	vcpu->run = page_address(page);
3746 
3747 	kvm_vcpu_init(vcpu, kvm, id);
3748 
3749 	r = kvm_arch_vcpu_create(vcpu);
3750 	if (r)
3751 		goto vcpu_free_run_page;
3752 
3753 	if (kvm->dirty_ring_size) {
3754 		r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3755 					 id, kvm->dirty_ring_size);
3756 		if (r)
3757 			goto arch_vcpu_destroy;
3758 	}
3759 
3760 	mutex_lock(&kvm->lock);
3761 	if (kvm_get_vcpu_by_id(kvm, id)) {
3762 		r = -EEXIST;
3763 		goto unlock_vcpu_destroy;
3764 	}
3765 
3766 	vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3767 	r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3768 	BUG_ON(r == -EBUSY);
3769 	if (r)
3770 		goto unlock_vcpu_destroy;
3771 
3772 	/* Fill the stats id string for the vcpu */
3773 	snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3774 		 task_pid_nr(current), id);
3775 
3776 	/* Now it's all set up, let userspace reach it */
3777 	kvm_get_kvm(kvm);
3778 	r = create_vcpu_fd(vcpu);
3779 	if (r < 0) {
3780 		xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3781 		kvm_put_kvm_no_destroy(kvm);
3782 		goto unlock_vcpu_destroy;
3783 	}
3784 
3785 	/*
3786 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Store the vcpu
3787 	 * pointer before kvm->online_vcpu's incremented value.
3788 	 */
3789 	smp_wmb();
3790 	atomic_inc(&kvm->online_vcpus);
3791 
3792 	mutex_unlock(&kvm->lock);
3793 	kvm_arch_vcpu_postcreate(vcpu);
3794 	kvm_create_vcpu_debugfs(vcpu);
3795 	return r;
3796 
3797 unlock_vcpu_destroy:
3798 	mutex_unlock(&kvm->lock);
3799 	kvm_dirty_ring_free(&vcpu->dirty_ring);
3800 arch_vcpu_destroy:
3801 	kvm_arch_vcpu_destroy(vcpu);
3802 vcpu_free_run_page:
3803 	free_page((unsigned long)vcpu->run);
3804 vcpu_free:
3805 	kmem_cache_free(kvm_vcpu_cache, vcpu);
3806 vcpu_decrement:
3807 	mutex_lock(&kvm->lock);
3808 	kvm->created_vcpus--;
3809 	mutex_unlock(&kvm->lock);
3810 	return r;
3811 }
3812 
3813 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3814 {
3815 	if (sigset) {
3816 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3817 		vcpu->sigset_active = 1;
3818 		vcpu->sigset = *sigset;
3819 	} else
3820 		vcpu->sigset_active = 0;
3821 	return 0;
3822 }
3823 
3824 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3825 			      size_t size, loff_t *offset)
3826 {
3827 	struct kvm_vcpu *vcpu = file->private_data;
3828 
3829 	return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3830 			&kvm_vcpu_stats_desc[0], &vcpu->stat,
3831 			sizeof(vcpu->stat), user_buffer, size, offset);
3832 }
3833 
3834 static const struct file_operations kvm_vcpu_stats_fops = {
3835 	.read = kvm_vcpu_stats_read,
3836 	.llseek = noop_llseek,
3837 };
3838 
3839 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3840 {
3841 	int fd;
3842 	struct file *file;
3843 	char name[15 + ITOA_MAX_LEN + 1];
3844 
3845 	snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3846 
3847 	fd = get_unused_fd_flags(O_CLOEXEC);
3848 	if (fd < 0)
3849 		return fd;
3850 
3851 	file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3852 	if (IS_ERR(file)) {
3853 		put_unused_fd(fd);
3854 		return PTR_ERR(file);
3855 	}
3856 	file->f_mode |= FMODE_PREAD;
3857 	fd_install(fd, file);
3858 
3859 	return fd;
3860 }
3861 
3862 static long kvm_vcpu_ioctl(struct file *filp,
3863 			   unsigned int ioctl, unsigned long arg)
3864 {
3865 	struct kvm_vcpu *vcpu = filp->private_data;
3866 	void __user *argp = (void __user *)arg;
3867 	int r;
3868 	struct kvm_fpu *fpu = NULL;
3869 	struct kvm_sregs *kvm_sregs = NULL;
3870 
3871 	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3872 		return -EIO;
3873 
3874 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3875 		return -EINVAL;
3876 
3877 	/*
3878 	 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3879 	 * execution; mutex_lock() would break them.
3880 	 */
3881 	r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3882 	if (r != -ENOIOCTLCMD)
3883 		return r;
3884 
3885 	if (mutex_lock_killable(&vcpu->mutex))
3886 		return -EINTR;
3887 	switch (ioctl) {
3888 	case KVM_RUN: {
3889 		struct pid *oldpid;
3890 		r = -EINVAL;
3891 		if (arg)
3892 			goto out;
3893 		oldpid = rcu_access_pointer(vcpu->pid);
3894 		if (unlikely(oldpid != task_pid(current))) {
3895 			/* The thread running this VCPU changed. */
3896 			struct pid *newpid;
3897 
3898 			r = kvm_arch_vcpu_run_pid_change(vcpu);
3899 			if (r)
3900 				break;
3901 
3902 			newpid = get_task_pid(current, PIDTYPE_PID);
3903 			rcu_assign_pointer(vcpu->pid, newpid);
3904 			if (oldpid)
3905 				synchronize_rcu();
3906 			put_pid(oldpid);
3907 		}
3908 		r = kvm_arch_vcpu_ioctl_run(vcpu);
3909 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3910 		break;
3911 	}
3912 	case KVM_GET_REGS: {
3913 		struct kvm_regs *kvm_regs;
3914 
3915 		r = -ENOMEM;
3916 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3917 		if (!kvm_regs)
3918 			goto out;
3919 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3920 		if (r)
3921 			goto out_free1;
3922 		r = -EFAULT;
3923 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3924 			goto out_free1;
3925 		r = 0;
3926 out_free1:
3927 		kfree(kvm_regs);
3928 		break;
3929 	}
3930 	case KVM_SET_REGS: {
3931 		struct kvm_regs *kvm_regs;
3932 
3933 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3934 		if (IS_ERR(kvm_regs)) {
3935 			r = PTR_ERR(kvm_regs);
3936 			goto out;
3937 		}
3938 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3939 		kfree(kvm_regs);
3940 		break;
3941 	}
3942 	case KVM_GET_SREGS: {
3943 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3944 				    GFP_KERNEL_ACCOUNT);
3945 		r = -ENOMEM;
3946 		if (!kvm_sregs)
3947 			goto out;
3948 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3949 		if (r)
3950 			goto out;
3951 		r = -EFAULT;
3952 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3953 			goto out;
3954 		r = 0;
3955 		break;
3956 	}
3957 	case KVM_SET_SREGS: {
3958 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3959 		if (IS_ERR(kvm_sregs)) {
3960 			r = PTR_ERR(kvm_sregs);
3961 			kvm_sregs = NULL;
3962 			goto out;
3963 		}
3964 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
3965 		break;
3966 	}
3967 	case KVM_GET_MP_STATE: {
3968 		struct kvm_mp_state mp_state;
3969 
3970 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
3971 		if (r)
3972 			goto out;
3973 		r = -EFAULT;
3974 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
3975 			goto out;
3976 		r = 0;
3977 		break;
3978 	}
3979 	case KVM_SET_MP_STATE: {
3980 		struct kvm_mp_state mp_state;
3981 
3982 		r = -EFAULT;
3983 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
3984 			goto out;
3985 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
3986 		break;
3987 	}
3988 	case KVM_TRANSLATE: {
3989 		struct kvm_translation tr;
3990 
3991 		r = -EFAULT;
3992 		if (copy_from_user(&tr, argp, sizeof(tr)))
3993 			goto out;
3994 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
3995 		if (r)
3996 			goto out;
3997 		r = -EFAULT;
3998 		if (copy_to_user(argp, &tr, sizeof(tr)))
3999 			goto out;
4000 		r = 0;
4001 		break;
4002 	}
4003 	case KVM_SET_GUEST_DEBUG: {
4004 		struct kvm_guest_debug dbg;
4005 
4006 		r = -EFAULT;
4007 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
4008 			goto out;
4009 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4010 		break;
4011 	}
4012 	case KVM_SET_SIGNAL_MASK: {
4013 		struct kvm_signal_mask __user *sigmask_arg = argp;
4014 		struct kvm_signal_mask kvm_sigmask;
4015 		sigset_t sigset, *p;
4016 
4017 		p = NULL;
4018 		if (argp) {
4019 			r = -EFAULT;
4020 			if (copy_from_user(&kvm_sigmask, argp,
4021 					   sizeof(kvm_sigmask)))
4022 				goto out;
4023 			r = -EINVAL;
4024 			if (kvm_sigmask.len != sizeof(sigset))
4025 				goto out;
4026 			r = -EFAULT;
4027 			if (copy_from_user(&sigset, sigmask_arg->sigset,
4028 					   sizeof(sigset)))
4029 				goto out;
4030 			p = &sigset;
4031 		}
4032 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4033 		break;
4034 	}
4035 	case KVM_GET_FPU: {
4036 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4037 		r = -ENOMEM;
4038 		if (!fpu)
4039 			goto out;
4040 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4041 		if (r)
4042 			goto out;
4043 		r = -EFAULT;
4044 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4045 			goto out;
4046 		r = 0;
4047 		break;
4048 	}
4049 	case KVM_SET_FPU: {
4050 		fpu = memdup_user(argp, sizeof(*fpu));
4051 		if (IS_ERR(fpu)) {
4052 			r = PTR_ERR(fpu);
4053 			fpu = NULL;
4054 			goto out;
4055 		}
4056 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4057 		break;
4058 	}
4059 	case KVM_GET_STATS_FD: {
4060 		r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4061 		break;
4062 	}
4063 	default:
4064 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4065 	}
4066 out:
4067 	mutex_unlock(&vcpu->mutex);
4068 	kfree(fpu);
4069 	kfree(kvm_sregs);
4070 	return r;
4071 }
4072 
4073 #ifdef CONFIG_KVM_COMPAT
4074 static long kvm_vcpu_compat_ioctl(struct file *filp,
4075 				  unsigned int ioctl, unsigned long arg)
4076 {
4077 	struct kvm_vcpu *vcpu = filp->private_data;
4078 	void __user *argp = compat_ptr(arg);
4079 	int r;
4080 
4081 	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4082 		return -EIO;
4083 
4084 	switch (ioctl) {
4085 	case KVM_SET_SIGNAL_MASK: {
4086 		struct kvm_signal_mask __user *sigmask_arg = argp;
4087 		struct kvm_signal_mask kvm_sigmask;
4088 		sigset_t sigset;
4089 
4090 		if (argp) {
4091 			r = -EFAULT;
4092 			if (copy_from_user(&kvm_sigmask, argp,
4093 					   sizeof(kvm_sigmask)))
4094 				goto out;
4095 			r = -EINVAL;
4096 			if (kvm_sigmask.len != sizeof(compat_sigset_t))
4097 				goto out;
4098 			r = -EFAULT;
4099 			if (get_compat_sigset(&sigset,
4100 					      (compat_sigset_t __user *)sigmask_arg->sigset))
4101 				goto out;
4102 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4103 		} else
4104 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4105 		break;
4106 	}
4107 	default:
4108 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
4109 	}
4110 
4111 out:
4112 	return r;
4113 }
4114 #endif
4115 
4116 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4117 {
4118 	struct kvm_device *dev = filp->private_data;
4119 
4120 	if (dev->ops->mmap)
4121 		return dev->ops->mmap(dev, vma);
4122 
4123 	return -ENODEV;
4124 }
4125 
4126 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4127 				 int (*accessor)(struct kvm_device *dev,
4128 						 struct kvm_device_attr *attr),
4129 				 unsigned long arg)
4130 {
4131 	struct kvm_device_attr attr;
4132 
4133 	if (!accessor)
4134 		return -EPERM;
4135 
4136 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4137 		return -EFAULT;
4138 
4139 	return accessor(dev, &attr);
4140 }
4141 
4142 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4143 			     unsigned long arg)
4144 {
4145 	struct kvm_device *dev = filp->private_data;
4146 
4147 	if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4148 		return -EIO;
4149 
4150 	switch (ioctl) {
4151 	case KVM_SET_DEVICE_ATTR:
4152 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4153 	case KVM_GET_DEVICE_ATTR:
4154 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4155 	case KVM_HAS_DEVICE_ATTR:
4156 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4157 	default:
4158 		if (dev->ops->ioctl)
4159 			return dev->ops->ioctl(dev, ioctl, arg);
4160 
4161 		return -ENOTTY;
4162 	}
4163 }
4164 
4165 static int kvm_device_release(struct inode *inode, struct file *filp)
4166 {
4167 	struct kvm_device *dev = filp->private_data;
4168 	struct kvm *kvm = dev->kvm;
4169 
4170 	if (dev->ops->release) {
4171 		mutex_lock(&kvm->lock);
4172 		list_del(&dev->vm_node);
4173 		dev->ops->release(dev);
4174 		mutex_unlock(&kvm->lock);
4175 	}
4176 
4177 	kvm_put_kvm(kvm);
4178 	return 0;
4179 }
4180 
4181 static const struct file_operations kvm_device_fops = {
4182 	.unlocked_ioctl = kvm_device_ioctl,
4183 	.release = kvm_device_release,
4184 	KVM_COMPAT(kvm_device_ioctl),
4185 	.mmap = kvm_device_mmap,
4186 };
4187 
4188 struct kvm_device *kvm_device_from_filp(struct file *filp)
4189 {
4190 	if (filp->f_op != &kvm_device_fops)
4191 		return NULL;
4192 
4193 	return filp->private_data;
4194 }
4195 
4196 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4197 #ifdef CONFIG_KVM_MPIC
4198 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
4199 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
4200 #endif
4201 };
4202 
4203 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4204 {
4205 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
4206 		return -ENOSPC;
4207 
4208 	if (kvm_device_ops_table[type] != NULL)
4209 		return -EEXIST;
4210 
4211 	kvm_device_ops_table[type] = ops;
4212 	return 0;
4213 }
4214 
4215 void kvm_unregister_device_ops(u32 type)
4216 {
4217 	if (kvm_device_ops_table[type] != NULL)
4218 		kvm_device_ops_table[type] = NULL;
4219 }
4220 
4221 static int kvm_ioctl_create_device(struct kvm *kvm,
4222 				   struct kvm_create_device *cd)
4223 {
4224 	const struct kvm_device_ops *ops = NULL;
4225 	struct kvm_device *dev;
4226 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4227 	int type;
4228 	int ret;
4229 
4230 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4231 		return -ENODEV;
4232 
4233 	type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4234 	ops = kvm_device_ops_table[type];
4235 	if (ops == NULL)
4236 		return -ENODEV;
4237 
4238 	if (test)
4239 		return 0;
4240 
4241 	dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4242 	if (!dev)
4243 		return -ENOMEM;
4244 
4245 	dev->ops = ops;
4246 	dev->kvm = kvm;
4247 
4248 	mutex_lock(&kvm->lock);
4249 	ret = ops->create(dev, type);
4250 	if (ret < 0) {
4251 		mutex_unlock(&kvm->lock);
4252 		kfree(dev);
4253 		return ret;
4254 	}
4255 	list_add(&dev->vm_node, &kvm->devices);
4256 	mutex_unlock(&kvm->lock);
4257 
4258 	if (ops->init)
4259 		ops->init(dev);
4260 
4261 	kvm_get_kvm(kvm);
4262 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4263 	if (ret < 0) {
4264 		kvm_put_kvm_no_destroy(kvm);
4265 		mutex_lock(&kvm->lock);
4266 		list_del(&dev->vm_node);
4267 		mutex_unlock(&kvm->lock);
4268 		ops->destroy(dev);
4269 		return ret;
4270 	}
4271 
4272 	cd->fd = ret;
4273 	return 0;
4274 }
4275 
4276 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4277 {
4278 	switch (arg) {
4279 	case KVM_CAP_USER_MEMORY:
4280 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4281 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4282 	case KVM_CAP_INTERNAL_ERROR_DATA:
4283 #ifdef CONFIG_HAVE_KVM_MSI
4284 	case KVM_CAP_SIGNAL_MSI:
4285 #endif
4286 #ifdef CONFIG_HAVE_KVM_IRQFD
4287 	case KVM_CAP_IRQFD:
4288 	case KVM_CAP_IRQFD_RESAMPLE:
4289 #endif
4290 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4291 	case KVM_CAP_CHECK_EXTENSION_VM:
4292 	case KVM_CAP_ENABLE_CAP_VM:
4293 	case KVM_CAP_HALT_POLL:
4294 		return 1;
4295 #ifdef CONFIG_KVM_MMIO
4296 	case KVM_CAP_COALESCED_MMIO:
4297 		return KVM_COALESCED_MMIO_PAGE_OFFSET;
4298 	case KVM_CAP_COALESCED_PIO:
4299 		return 1;
4300 #endif
4301 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4302 	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4303 		return KVM_DIRTY_LOG_MANUAL_CAPS;
4304 #endif
4305 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4306 	case KVM_CAP_IRQ_ROUTING:
4307 		return KVM_MAX_IRQ_ROUTES;
4308 #endif
4309 #if KVM_ADDRESS_SPACE_NUM > 1
4310 	case KVM_CAP_MULTI_ADDRESS_SPACE:
4311 		return KVM_ADDRESS_SPACE_NUM;
4312 #endif
4313 	case KVM_CAP_NR_MEMSLOTS:
4314 		return KVM_USER_MEM_SLOTS;
4315 	case KVM_CAP_DIRTY_LOG_RING:
4316 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4317 		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4318 #else
4319 		return 0;
4320 #endif
4321 	case KVM_CAP_BINARY_STATS_FD:
4322 		return 1;
4323 	default:
4324 		break;
4325 	}
4326 	return kvm_vm_ioctl_check_extension(kvm, arg);
4327 }
4328 
4329 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4330 {
4331 	int r;
4332 
4333 	if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4334 		return -EINVAL;
4335 
4336 	/* the size should be power of 2 */
4337 	if (!size || (size & (size - 1)))
4338 		return -EINVAL;
4339 
4340 	/* Should be bigger to keep the reserved entries, or a page */
4341 	if (size < kvm_dirty_ring_get_rsvd_entries() *
4342 	    sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4343 		return -EINVAL;
4344 
4345 	if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4346 	    sizeof(struct kvm_dirty_gfn))
4347 		return -E2BIG;
4348 
4349 	/* We only allow it to set once */
4350 	if (kvm->dirty_ring_size)
4351 		return -EINVAL;
4352 
4353 	mutex_lock(&kvm->lock);
4354 
4355 	if (kvm->created_vcpus) {
4356 		/* We don't allow to change this value after vcpu created */
4357 		r = -EINVAL;
4358 	} else {
4359 		kvm->dirty_ring_size = size;
4360 		r = 0;
4361 	}
4362 
4363 	mutex_unlock(&kvm->lock);
4364 	return r;
4365 }
4366 
4367 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4368 {
4369 	unsigned long i;
4370 	struct kvm_vcpu *vcpu;
4371 	int cleared = 0;
4372 
4373 	if (!kvm->dirty_ring_size)
4374 		return -EINVAL;
4375 
4376 	mutex_lock(&kvm->slots_lock);
4377 
4378 	kvm_for_each_vcpu(i, vcpu, kvm)
4379 		cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4380 
4381 	mutex_unlock(&kvm->slots_lock);
4382 
4383 	if (cleared)
4384 		kvm_flush_remote_tlbs(kvm);
4385 
4386 	return cleared;
4387 }
4388 
4389 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4390 						  struct kvm_enable_cap *cap)
4391 {
4392 	return -EINVAL;
4393 }
4394 
4395 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4396 					   struct kvm_enable_cap *cap)
4397 {
4398 	switch (cap->cap) {
4399 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4400 	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4401 		u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4402 
4403 		if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4404 			allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4405 
4406 		if (cap->flags || (cap->args[0] & ~allowed_options))
4407 			return -EINVAL;
4408 		kvm->manual_dirty_log_protect = cap->args[0];
4409 		return 0;
4410 	}
4411 #endif
4412 	case KVM_CAP_HALT_POLL: {
4413 		if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4414 			return -EINVAL;
4415 
4416 		kvm->max_halt_poll_ns = cap->args[0];
4417 		return 0;
4418 	}
4419 	case KVM_CAP_DIRTY_LOG_RING:
4420 		return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4421 	default:
4422 		return kvm_vm_ioctl_enable_cap(kvm, cap);
4423 	}
4424 }
4425 
4426 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4427 			      size_t size, loff_t *offset)
4428 {
4429 	struct kvm *kvm = file->private_data;
4430 
4431 	return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4432 				&kvm_vm_stats_desc[0], &kvm->stat,
4433 				sizeof(kvm->stat), user_buffer, size, offset);
4434 }
4435 
4436 static const struct file_operations kvm_vm_stats_fops = {
4437 	.read = kvm_vm_stats_read,
4438 	.llseek = noop_llseek,
4439 };
4440 
4441 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4442 {
4443 	int fd;
4444 	struct file *file;
4445 
4446 	fd = get_unused_fd_flags(O_CLOEXEC);
4447 	if (fd < 0)
4448 		return fd;
4449 
4450 	file = anon_inode_getfile("kvm-vm-stats",
4451 			&kvm_vm_stats_fops, kvm, O_RDONLY);
4452 	if (IS_ERR(file)) {
4453 		put_unused_fd(fd);
4454 		return PTR_ERR(file);
4455 	}
4456 	file->f_mode |= FMODE_PREAD;
4457 	fd_install(fd, file);
4458 
4459 	return fd;
4460 }
4461 
4462 static long kvm_vm_ioctl(struct file *filp,
4463 			   unsigned int ioctl, unsigned long arg)
4464 {
4465 	struct kvm *kvm = filp->private_data;
4466 	void __user *argp = (void __user *)arg;
4467 	int r;
4468 
4469 	if (kvm->mm != current->mm || kvm->vm_dead)
4470 		return -EIO;
4471 	switch (ioctl) {
4472 	case KVM_CREATE_VCPU:
4473 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4474 		break;
4475 	case KVM_ENABLE_CAP: {
4476 		struct kvm_enable_cap cap;
4477 
4478 		r = -EFAULT;
4479 		if (copy_from_user(&cap, argp, sizeof(cap)))
4480 			goto out;
4481 		r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4482 		break;
4483 	}
4484 	case KVM_SET_USER_MEMORY_REGION: {
4485 		struct kvm_userspace_memory_region kvm_userspace_mem;
4486 
4487 		r = -EFAULT;
4488 		if (copy_from_user(&kvm_userspace_mem, argp,
4489 						sizeof(kvm_userspace_mem)))
4490 			goto out;
4491 
4492 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4493 		break;
4494 	}
4495 	case KVM_GET_DIRTY_LOG: {
4496 		struct kvm_dirty_log log;
4497 
4498 		r = -EFAULT;
4499 		if (copy_from_user(&log, argp, sizeof(log)))
4500 			goto out;
4501 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4502 		break;
4503 	}
4504 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4505 	case KVM_CLEAR_DIRTY_LOG: {
4506 		struct kvm_clear_dirty_log log;
4507 
4508 		r = -EFAULT;
4509 		if (copy_from_user(&log, argp, sizeof(log)))
4510 			goto out;
4511 		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4512 		break;
4513 	}
4514 #endif
4515 #ifdef CONFIG_KVM_MMIO
4516 	case KVM_REGISTER_COALESCED_MMIO: {
4517 		struct kvm_coalesced_mmio_zone zone;
4518 
4519 		r = -EFAULT;
4520 		if (copy_from_user(&zone, argp, sizeof(zone)))
4521 			goto out;
4522 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4523 		break;
4524 	}
4525 	case KVM_UNREGISTER_COALESCED_MMIO: {
4526 		struct kvm_coalesced_mmio_zone zone;
4527 
4528 		r = -EFAULT;
4529 		if (copy_from_user(&zone, argp, sizeof(zone)))
4530 			goto out;
4531 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4532 		break;
4533 	}
4534 #endif
4535 	case KVM_IRQFD: {
4536 		struct kvm_irqfd data;
4537 
4538 		r = -EFAULT;
4539 		if (copy_from_user(&data, argp, sizeof(data)))
4540 			goto out;
4541 		r = kvm_irqfd(kvm, &data);
4542 		break;
4543 	}
4544 	case KVM_IOEVENTFD: {
4545 		struct kvm_ioeventfd data;
4546 
4547 		r = -EFAULT;
4548 		if (copy_from_user(&data, argp, sizeof(data)))
4549 			goto out;
4550 		r = kvm_ioeventfd(kvm, &data);
4551 		break;
4552 	}
4553 #ifdef CONFIG_HAVE_KVM_MSI
4554 	case KVM_SIGNAL_MSI: {
4555 		struct kvm_msi msi;
4556 
4557 		r = -EFAULT;
4558 		if (copy_from_user(&msi, argp, sizeof(msi)))
4559 			goto out;
4560 		r = kvm_send_userspace_msi(kvm, &msi);
4561 		break;
4562 	}
4563 #endif
4564 #ifdef __KVM_HAVE_IRQ_LINE
4565 	case KVM_IRQ_LINE_STATUS:
4566 	case KVM_IRQ_LINE: {
4567 		struct kvm_irq_level irq_event;
4568 
4569 		r = -EFAULT;
4570 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4571 			goto out;
4572 
4573 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4574 					ioctl == KVM_IRQ_LINE_STATUS);
4575 		if (r)
4576 			goto out;
4577 
4578 		r = -EFAULT;
4579 		if (ioctl == KVM_IRQ_LINE_STATUS) {
4580 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4581 				goto out;
4582 		}
4583 
4584 		r = 0;
4585 		break;
4586 	}
4587 #endif
4588 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4589 	case KVM_SET_GSI_ROUTING: {
4590 		struct kvm_irq_routing routing;
4591 		struct kvm_irq_routing __user *urouting;
4592 		struct kvm_irq_routing_entry *entries = NULL;
4593 
4594 		r = -EFAULT;
4595 		if (copy_from_user(&routing, argp, sizeof(routing)))
4596 			goto out;
4597 		r = -EINVAL;
4598 		if (!kvm_arch_can_set_irq_routing(kvm))
4599 			goto out;
4600 		if (routing.nr > KVM_MAX_IRQ_ROUTES)
4601 			goto out;
4602 		if (routing.flags)
4603 			goto out;
4604 		if (routing.nr) {
4605 			urouting = argp;
4606 			entries = vmemdup_user(urouting->entries,
4607 					       array_size(sizeof(*entries),
4608 							  routing.nr));
4609 			if (IS_ERR(entries)) {
4610 				r = PTR_ERR(entries);
4611 				goto out;
4612 			}
4613 		}
4614 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
4615 					routing.flags);
4616 		kvfree(entries);
4617 		break;
4618 	}
4619 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4620 	case KVM_CREATE_DEVICE: {
4621 		struct kvm_create_device cd;
4622 
4623 		r = -EFAULT;
4624 		if (copy_from_user(&cd, argp, sizeof(cd)))
4625 			goto out;
4626 
4627 		r = kvm_ioctl_create_device(kvm, &cd);
4628 		if (r)
4629 			goto out;
4630 
4631 		r = -EFAULT;
4632 		if (copy_to_user(argp, &cd, sizeof(cd)))
4633 			goto out;
4634 
4635 		r = 0;
4636 		break;
4637 	}
4638 	case KVM_CHECK_EXTENSION:
4639 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4640 		break;
4641 	case KVM_RESET_DIRTY_RINGS:
4642 		r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4643 		break;
4644 	case KVM_GET_STATS_FD:
4645 		r = kvm_vm_ioctl_get_stats_fd(kvm);
4646 		break;
4647 	default:
4648 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4649 	}
4650 out:
4651 	return r;
4652 }
4653 
4654 #ifdef CONFIG_KVM_COMPAT
4655 struct compat_kvm_dirty_log {
4656 	__u32 slot;
4657 	__u32 padding1;
4658 	union {
4659 		compat_uptr_t dirty_bitmap; /* one bit per page */
4660 		__u64 padding2;
4661 	};
4662 };
4663 
4664 struct compat_kvm_clear_dirty_log {
4665 	__u32 slot;
4666 	__u32 num_pages;
4667 	__u64 first_page;
4668 	union {
4669 		compat_uptr_t dirty_bitmap; /* one bit per page */
4670 		__u64 padding2;
4671 	};
4672 };
4673 
4674 static long kvm_vm_compat_ioctl(struct file *filp,
4675 			   unsigned int ioctl, unsigned long arg)
4676 {
4677 	struct kvm *kvm = filp->private_data;
4678 	int r;
4679 
4680 	if (kvm->mm != current->mm || kvm->vm_dead)
4681 		return -EIO;
4682 	switch (ioctl) {
4683 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4684 	case KVM_CLEAR_DIRTY_LOG: {
4685 		struct compat_kvm_clear_dirty_log compat_log;
4686 		struct kvm_clear_dirty_log log;
4687 
4688 		if (copy_from_user(&compat_log, (void __user *)arg,
4689 				   sizeof(compat_log)))
4690 			return -EFAULT;
4691 		log.slot	 = compat_log.slot;
4692 		log.num_pages	 = compat_log.num_pages;
4693 		log.first_page	 = compat_log.first_page;
4694 		log.padding2	 = compat_log.padding2;
4695 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4696 
4697 		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4698 		break;
4699 	}
4700 #endif
4701 	case KVM_GET_DIRTY_LOG: {
4702 		struct compat_kvm_dirty_log compat_log;
4703 		struct kvm_dirty_log log;
4704 
4705 		if (copy_from_user(&compat_log, (void __user *)arg,
4706 				   sizeof(compat_log)))
4707 			return -EFAULT;
4708 		log.slot	 = compat_log.slot;
4709 		log.padding1	 = compat_log.padding1;
4710 		log.padding2	 = compat_log.padding2;
4711 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4712 
4713 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4714 		break;
4715 	}
4716 	default:
4717 		r = kvm_vm_ioctl(filp, ioctl, arg);
4718 	}
4719 	return r;
4720 }
4721 #endif
4722 
4723 static struct file_operations kvm_vm_fops = {
4724 	.release        = kvm_vm_release,
4725 	.unlocked_ioctl = kvm_vm_ioctl,
4726 	.llseek		= noop_llseek,
4727 	KVM_COMPAT(kvm_vm_compat_ioctl),
4728 };
4729 
4730 bool file_is_kvm(struct file *file)
4731 {
4732 	return file && file->f_op == &kvm_vm_fops;
4733 }
4734 EXPORT_SYMBOL_GPL(file_is_kvm);
4735 
4736 static int kvm_dev_ioctl_create_vm(unsigned long type)
4737 {
4738 	int r;
4739 	struct kvm *kvm;
4740 	struct file *file;
4741 
4742 	kvm = kvm_create_vm(type);
4743 	if (IS_ERR(kvm))
4744 		return PTR_ERR(kvm);
4745 #ifdef CONFIG_KVM_MMIO
4746 	r = kvm_coalesced_mmio_init(kvm);
4747 	if (r < 0)
4748 		goto put_kvm;
4749 #endif
4750 	r = get_unused_fd_flags(O_CLOEXEC);
4751 	if (r < 0)
4752 		goto put_kvm;
4753 
4754 	snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4755 			"kvm-%d", task_pid_nr(current));
4756 
4757 	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4758 	if (IS_ERR(file)) {
4759 		put_unused_fd(r);
4760 		r = PTR_ERR(file);
4761 		goto put_kvm;
4762 	}
4763 
4764 	/*
4765 	 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4766 	 * already set, with ->release() being kvm_vm_release().  In error
4767 	 * cases it will be called by the final fput(file) and will take
4768 	 * care of doing kvm_put_kvm(kvm).
4769 	 */
4770 	if (kvm_create_vm_debugfs(kvm, r) < 0) {
4771 		put_unused_fd(r);
4772 		fput(file);
4773 		return -ENOMEM;
4774 	}
4775 	kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4776 
4777 	fd_install(r, file);
4778 	return r;
4779 
4780 put_kvm:
4781 	kvm_put_kvm(kvm);
4782 	return r;
4783 }
4784 
4785 static long kvm_dev_ioctl(struct file *filp,
4786 			  unsigned int ioctl, unsigned long arg)
4787 {
4788 	long r = -EINVAL;
4789 
4790 	switch (ioctl) {
4791 	case KVM_GET_API_VERSION:
4792 		if (arg)
4793 			goto out;
4794 		r = KVM_API_VERSION;
4795 		break;
4796 	case KVM_CREATE_VM:
4797 		r = kvm_dev_ioctl_create_vm(arg);
4798 		break;
4799 	case KVM_CHECK_EXTENSION:
4800 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4801 		break;
4802 	case KVM_GET_VCPU_MMAP_SIZE:
4803 		if (arg)
4804 			goto out;
4805 		r = PAGE_SIZE;     /* struct kvm_run */
4806 #ifdef CONFIG_X86
4807 		r += PAGE_SIZE;    /* pio data page */
4808 #endif
4809 #ifdef CONFIG_KVM_MMIO
4810 		r += PAGE_SIZE;    /* coalesced mmio ring page */
4811 #endif
4812 		break;
4813 	case KVM_TRACE_ENABLE:
4814 	case KVM_TRACE_PAUSE:
4815 	case KVM_TRACE_DISABLE:
4816 		r = -EOPNOTSUPP;
4817 		break;
4818 	default:
4819 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
4820 	}
4821 out:
4822 	return r;
4823 }
4824 
4825 static struct file_operations kvm_chardev_ops = {
4826 	.unlocked_ioctl = kvm_dev_ioctl,
4827 	.llseek		= noop_llseek,
4828 	KVM_COMPAT(kvm_dev_ioctl),
4829 };
4830 
4831 static struct miscdevice kvm_dev = {
4832 	KVM_MINOR,
4833 	"kvm",
4834 	&kvm_chardev_ops,
4835 };
4836 
4837 static void hardware_enable_nolock(void *junk)
4838 {
4839 	int cpu = raw_smp_processor_id();
4840 	int r;
4841 
4842 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4843 		return;
4844 
4845 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
4846 
4847 	r = kvm_arch_hardware_enable();
4848 
4849 	if (r) {
4850 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4851 		atomic_inc(&hardware_enable_failed);
4852 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4853 	}
4854 }
4855 
4856 static int kvm_starting_cpu(unsigned int cpu)
4857 {
4858 	raw_spin_lock(&kvm_count_lock);
4859 	if (kvm_usage_count)
4860 		hardware_enable_nolock(NULL);
4861 	raw_spin_unlock(&kvm_count_lock);
4862 	return 0;
4863 }
4864 
4865 static void hardware_disable_nolock(void *junk)
4866 {
4867 	int cpu = raw_smp_processor_id();
4868 
4869 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4870 		return;
4871 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4872 	kvm_arch_hardware_disable();
4873 }
4874 
4875 static int kvm_dying_cpu(unsigned int cpu)
4876 {
4877 	raw_spin_lock(&kvm_count_lock);
4878 	if (kvm_usage_count)
4879 		hardware_disable_nolock(NULL);
4880 	raw_spin_unlock(&kvm_count_lock);
4881 	return 0;
4882 }
4883 
4884 static void hardware_disable_all_nolock(void)
4885 {
4886 	BUG_ON(!kvm_usage_count);
4887 
4888 	kvm_usage_count--;
4889 	if (!kvm_usage_count)
4890 		on_each_cpu(hardware_disable_nolock, NULL, 1);
4891 }
4892 
4893 static void hardware_disable_all(void)
4894 {
4895 	raw_spin_lock(&kvm_count_lock);
4896 	hardware_disable_all_nolock();
4897 	raw_spin_unlock(&kvm_count_lock);
4898 }
4899 
4900 static int hardware_enable_all(void)
4901 {
4902 	int r = 0;
4903 
4904 	raw_spin_lock(&kvm_count_lock);
4905 
4906 	kvm_usage_count++;
4907 	if (kvm_usage_count == 1) {
4908 		atomic_set(&hardware_enable_failed, 0);
4909 		on_each_cpu(hardware_enable_nolock, NULL, 1);
4910 
4911 		if (atomic_read(&hardware_enable_failed)) {
4912 			hardware_disable_all_nolock();
4913 			r = -EBUSY;
4914 		}
4915 	}
4916 
4917 	raw_spin_unlock(&kvm_count_lock);
4918 
4919 	return r;
4920 }
4921 
4922 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4923 		      void *v)
4924 {
4925 	/*
4926 	 * Some (well, at least mine) BIOSes hang on reboot if
4927 	 * in vmx root mode.
4928 	 *
4929 	 * And Intel TXT required VMX off for all cpu when system shutdown.
4930 	 */
4931 	pr_info("kvm: exiting hardware virtualization\n");
4932 	kvm_rebooting = true;
4933 	on_each_cpu(hardware_disable_nolock, NULL, 1);
4934 	return NOTIFY_OK;
4935 }
4936 
4937 static struct notifier_block kvm_reboot_notifier = {
4938 	.notifier_call = kvm_reboot,
4939 	.priority = 0,
4940 };
4941 
4942 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4943 {
4944 	int i;
4945 
4946 	for (i = 0; i < bus->dev_count; i++) {
4947 		struct kvm_io_device *pos = bus->range[i].dev;
4948 
4949 		kvm_iodevice_destructor(pos);
4950 	}
4951 	kfree(bus);
4952 }
4953 
4954 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4955 				 const struct kvm_io_range *r2)
4956 {
4957 	gpa_t addr1 = r1->addr;
4958 	gpa_t addr2 = r2->addr;
4959 
4960 	if (addr1 < addr2)
4961 		return -1;
4962 
4963 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
4964 	 * accept any overlapping write.  Any order is acceptable for
4965 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
4966 	 * we process all of them.
4967 	 */
4968 	if (r2->len) {
4969 		addr1 += r1->len;
4970 		addr2 += r2->len;
4971 	}
4972 
4973 	if (addr1 > addr2)
4974 		return 1;
4975 
4976 	return 0;
4977 }
4978 
4979 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
4980 {
4981 	return kvm_io_bus_cmp(p1, p2);
4982 }
4983 
4984 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
4985 			     gpa_t addr, int len)
4986 {
4987 	struct kvm_io_range *range, key;
4988 	int off;
4989 
4990 	key = (struct kvm_io_range) {
4991 		.addr = addr,
4992 		.len = len,
4993 	};
4994 
4995 	range = bsearch(&key, bus->range, bus->dev_count,
4996 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
4997 	if (range == NULL)
4998 		return -ENOENT;
4999 
5000 	off = range - bus->range;
5001 
5002 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5003 		off--;
5004 
5005 	return off;
5006 }
5007 
5008 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5009 			      struct kvm_io_range *range, const void *val)
5010 {
5011 	int idx;
5012 
5013 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5014 	if (idx < 0)
5015 		return -EOPNOTSUPP;
5016 
5017 	while (idx < bus->dev_count &&
5018 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5019 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5020 					range->len, val))
5021 			return idx;
5022 		idx++;
5023 	}
5024 
5025 	return -EOPNOTSUPP;
5026 }
5027 
5028 /* kvm_io_bus_write - called under kvm->slots_lock */
5029 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5030 		     int len, const void *val)
5031 {
5032 	struct kvm_io_bus *bus;
5033 	struct kvm_io_range range;
5034 	int r;
5035 
5036 	range = (struct kvm_io_range) {
5037 		.addr = addr,
5038 		.len = len,
5039 	};
5040 
5041 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5042 	if (!bus)
5043 		return -ENOMEM;
5044 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
5045 	return r < 0 ? r : 0;
5046 }
5047 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5048 
5049 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5050 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5051 			    gpa_t addr, int len, const void *val, long cookie)
5052 {
5053 	struct kvm_io_bus *bus;
5054 	struct kvm_io_range range;
5055 
5056 	range = (struct kvm_io_range) {
5057 		.addr = addr,
5058 		.len = len,
5059 	};
5060 
5061 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5062 	if (!bus)
5063 		return -ENOMEM;
5064 
5065 	/* First try the device referenced by cookie. */
5066 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
5067 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5068 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5069 					val))
5070 			return cookie;
5071 
5072 	/*
5073 	 * cookie contained garbage; fall back to search and return the
5074 	 * correct cookie value.
5075 	 */
5076 	return __kvm_io_bus_write(vcpu, bus, &range, val);
5077 }
5078 
5079 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5080 			     struct kvm_io_range *range, void *val)
5081 {
5082 	int idx;
5083 
5084 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5085 	if (idx < 0)
5086 		return -EOPNOTSUPP;
5087 
5088 	while (idx < bus->dev_count &&
5089 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5090 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5091 				       range->len, val))
5092 			return idx;
5093 		idx++;
5094 	}
5095 
5096 	return -EOPNOTSUPP;
5097 }
5098 
5099 /* kvm_io_bus_read - called under kvm->slots_lock */
5100 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5101 		    int len, void *val)
5102 {
5103 	struct kvm_io_bus *bus;
5104 	struct kvm_io_range range;
5105 	int r;
5106 
5107 	range = (struct kvm_io_range) {
5108 		.addr = addr,
5109 		.len = len,
5110 	};
5111 
5112 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5113 	if (!bus)
5114 		return -ENOMEM;
5115 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
5116 	return r < 0 ? r : 0;
5117 }
5118 
5119 /* Caller must hold slots_lock. */
5120 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5121 			    int len, struct kvm_io_device *dev)
5122 {
5123 	int i;
5124 	struct kvm_io_bus *new_bus, *bus;
5125 	struct kvm_io_range range;
5126 
5127 	bus = kvm_get_bus(kvm, bus_idx);
5128 	if (!bus)
5129 		return -ENOMEM;
5130 
5131 	/* exclude ioeventfd which is limited by maximum fd */
5132 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5133 		return -ENOSPC;
5134 
5135 	new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5136 			  GFP_KERNEL_ACCOUNT);
5137 	if (!new_bus)
5138 		return -ENOMEM;
5139 
5140 	range = (struct kvm_io_range) {
5141 		.addr = addr,
5142 		.len = len,
5143 		.dev = dev,
5144 	};
5145 
5146 	for (i = 0; i < bus->dev_count; i++)
5147 		if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5148 			break;
5149 
5150 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5151 	new_bus->dev_count++;
5152 	new_bus->range[i] = range;
5153 	memcpy(new_bus->range + i + 1, bus->range + i,
5154 		(bus->dev_count - i) * sizeof(struct kvm_io_range));
5155 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5156 	synchronize_srcu_expedited(&kvm->srcu);
5157 	kfree(bus);
5158 
5159 	return 0;
5160 }
5161 
5162 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5163 			      struct kvm_io_device *dev)
5164 {
5165 	int i, j;
5166 	struct kvm_io_bus *new_bus, *bus;
5167 
5168 	lockdep_assert_held(&kvm->slots_lock);
5169 
5170 	bus = kvm_get_bus(kvm, bus_idx);
5171 	if (!bus)
5172 		return 0;
5173 
5174 	for (i = 0; i < bus->dev_count; i++) {
5175 		if (bus->range[i].dev == dev) {
5176 			break;
5177 		}
5178 	}
5179 
5180 	if (i == bus->dev_count)
5181 		return 0;
5182 
5183 	new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5184 			  GFP_KERNEL_ACCOUNT);
5185 	if (new_bus) {
5186 		memcpy(new_bus, bus, struct_size(bus, range, i));
5187 		new_bus->dev_count--;
5188 		memcpy(new_bus->range + i, bus->range + i + 1,
5189 				flex_array_size(new_bus, range, new_bus->dev_count - i));
5190 	}
5191 
5192 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5193 	synchronize_srcu_expedited(&kvm->srcu);
5194 
5195 	/* Destroy the old bus _after_ installing the (null) bus. */
5196 	if (!new_bus) {
5197 		pr_err("kvm: failed to shrink bus, removing it completely\n");
5198 		for (j = 0; j < bus->dev_count; j++) {
5199 			if (j == i)
5200 				continue;
5201 			kvm_iodevice_destructor(bus->range[j].dev);
5202 		}
5203 	}
5204 
5205 	kfree(bus);
5206 	return new_bus ? 0 : -ENOMEM;
5207 }
5208 
5209 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5210 					 gpa_t addr)
5211 {
5212 	struct kvm_io_bus *bus;
5213 	int dev_idx, srcu_idx;
5214 	struct kvm_io_device *iodev = NULL;
5215 
5216 	srcu_idx = srcu_read_lock(&kvm->srcu);
5217 
5218 	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5219 	if (!bus)
5220 		goto out_unlock;
5221 
5222 	dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5223 	if (dev_idx < 0)
5224 		goto out_unlock;
5225 
5226 	iodev = bus->range[dev_idx].dev;
5227 
5228 out_unlock:
5229 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5230 
5231 	return iodev;
5232 }
5233 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5234 
5235 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5236 			   int (*get)(void *, u64 *), int (*set)(void *, u64),
5237 			   const char *fmt)
5238 {
5239 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5240 					  inode->i_private;
5241 
5242 	/*
5243 	 * The debugfs files are a reference to the kvm struct which
5244         * is still valid when kvm_destroy_vm is called.  kvm_get_kvm_safe
5245         * avoids the race between open and the removal of the debugfs directory.
5246 	 */
5247 	if (!kvm_get_kvm_safe(stat_data->kvm))
5248 		return -ENOENT;
5249 
5250 	if (simple_attr_open(inode, file, get,
5251 		    kvm_stats_debugfs_mode(stat_data->desc) & 0222
5252 		    ? set : NULL,
5253 		    fmt)) {
5254 		kvm_put_kvm(stat_data->kvm);
5255 		return -ENOMEM;
5256 	}
5257 
5258 	return 0;
5259 }
5260 
5261 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5262 {
5263 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5264 					  inode->i_private;
5265 
5266 	simple_attr_release(inode, file);
5267 	kvm_put_kvm(stat_data->kvm);
5268 
5269 	return 0;
5270 }
5271 
5272 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5273 {
5274 	*val = *(u64 *)((void *)(&kvm->stat) + offset);
5275 
5276 	return 0;
5277 }
5278 
5279 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5280 {
5281 	*(u64 *)((void *)(&kvm->stat) + offset) = 0;
5282 
5283 	return 0;
5284 }
5285 
5286 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5287 {
5288 	unsigned long i;
5289 	struct kvm_vcpu *vcpu;
5290 
5291 	*val = 0;
5292 
5293 	kvm_for_each_vcpu(i, vcpu, kvm)
5294 		*val += *(u64 *)((void *)(&vcpu->stat) + offset);
5295 
5296 	return 0;
5297 }
5298 
5299 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5300 {
5301 	unsigned long i;
5302 	struct kvm_vcpu *vcpu;
5303 
5304 	kvm_for_each_vcpu(i, vcpu, kvm)
5305 		*(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5306 
5307 	return 0;
5308 }
5309 
5310 static int kvm_stat_data_get(void *data, u64 *val)
5311 {
5312 	int r = -EFAULT;
5313 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5314 
5315 	switch (stat_data->kind) {
5316 	case KVM_STAT_VM:
5317 		r = kvm_get_stat_per_vm(stat_data->kvm,
5318 					stat_data->desc->desc.offset, val);
5319 		break;
5320 	case KVM_STAT_VCPU:
5321 		r = kvm_get_stat_per_vcpu(stat_data->kvm,
5322 					  stat_data->desc->desc.offset, val);
5323 		break;
5324 	}
5325 
5326 	return r;
5327 }
5328 
5329 static int kvm_stat_data_clear(void *data, u64 val)
5330 {
5331 	int r = -EFAULT;
5332 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5333 
5334 	if (val)
5335 		return -EINVAL;
5336 
5337 	switch (stat_data->kind) {
5338 	case KVM_STAT_VM:
5339 		r = kvm_clear_stat_per_vm(stat_data->kvm,
5340 					  stat_data->desc->desc.offset);
5341 		break;
5342 	case KVM_STAT_VCPU:
5343 		r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5344 					    stat_data->desc->desc.offset);
5345 		break;
5346 	}
5347 
5348 	return r;
5349 }
5350 
5351 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5352 {
5353 	__simple_attr_check_format("%llu\n", 0ull);
5354 	return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5355 				kvm_stat_data_clear, "%llu\n");
5356 }
5357 
5358 static const struct file_operations stat_fops_per_vm = {
5359 	.owner = THIS_MODULE,
5360 	.open = kvm_stat_data_open,
5361 	.release = kvm_debugfs_release,
5362 	.read = simple_attr_read,
5363 	.write = simple_attr_write,
5364 	.llseek = no_llseek,
5365 };
5366 
5367 static int vm_stat_get(void *_offset, u64 *val)
5368 {
5369 	unsigned offset = (long)_offset;
5370 	struct kvm *kvm;
5371 	u64 tmp_val;
5372 
5373 	*val = 0;
5374 	mutex_lock(&kvm_lock);
5375 	list_for_each_entry(kvm, &vm_list, vm_list) {
5376 		kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5377 		*val += tmp_val;
5378 	}
5379 	mutex_unlock(&kvm_lock);
5380 	return 0;
5381 }
5382 
5383 static int vm_stat_clear(void *_offset, u64 val)
5384 {
5385 	unsigned offset = (long)_offset;
5386 	struct kvm *kvm;
5387 
5388 	if (val)
5389 		return -EINVAL;
5390 
5391 	mutex_lock(&kvm_lock);
5392 	list_for_each_entry(kvm, &vm_list, vm_list) {
5393 		kvm_clear_stat_per_vm(kvm, offset);
5394 	}
5395 	mutex_unlock(&kvm_lock);
5396 
5397 	return 0;
5398 }
5399 
5400 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5401 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5402 
5403 static int vcpu_stat_get(void *_offset, u64 *val)
5404 {
5405 	unsigned offset = (long)_offset;
5406 	struct kvm *kvm;
5407 	u64 tmp_val;
5408 
5409 	*val = 0;
5410 	mutex_lock(&kvm_lock);
5411 	list_for_each_entry(kvm, &vm_list, vm_list) {
5412 		kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5413 		*val += tmp_val;
5414 	}
5415 	mutex_unlock(&kvm_lock);
5416 	return 0;
5417 }
5418 
5419 static int vcpu_stat_clear(void *_offset, u64 val)
5420 {
5421 	unsigned offset = (long)_offset;
5422 	struct kvm *kvm;
5423 
5424 	if (val)
5425 		return -EINVAL;
5426 
5427 	mutex_lock(&kvm_lock);
5428 	list_for_each_entry(kvm, &vm_list, vm_list) {
5429 		kvm_clear_stat_per_vcpu(kvm, offset);
5430 	}
5431 	mutex_unlock(&kvm_lock);
5432 
5433 	return 0;
5434 }
5435 
5436 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5437 			"%llu\n");
5438 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5439 
5440 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5441 {
5442 	struct kobj_uevent_env *env;
5443 	unsigned long long created, active;
5444 
5445 	if (!kvm_dev.this_device || !kvm)
5446 		return;
5447 
5448 	mutex_lock(&kvm_lock);
5449 	if (type == KVM_EVENT_CREATE_VM) {
5450 		kvm_createvm_count++;
5451 		kvm_active_vms++;
5452 	} else if (type == KVM_EVENT_DESTROY_VM) {
5453 		kvm_active_vms--;
5454 	}
5455 	created = kvm_createvm_count;
5456 	active = kvm_active_vms;
5457 	mutex_unlock(&kvm_lock);
5458 
5459 	env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5460 	if (!env)
5461 		return;
5462 
5463 	add_uevent_var(env, "CREATED=%llu", created);
5464 	add_uevent_var(env, "COUNT=%llu", active);
5465 
5466 	if (type == KVM_EVENT_CREATE_VM) {
5467 		add_uevent_var(env, "EVENT=create");
5468 		kvm->userspace_pid = task_pid_nr(current);
5469 	} else if (type == KVM_EVENT_DESTROY_VM) {
5470 		add_uevent_var(env, "EVENT=destroy");
5471 	}
5472 	add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5473 
5474 	if (kvm->debugfs_dentry) {
5475 		char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5476 
5477 		if (p) {
5478 			tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5479 			if (!IS_ERR(tmp))
5480 				add_uevent_var(env, "STATS_PATH=%s", tmp);
5481 			kfree(p);
5482 		}
5483 	}
5484 	/* no need for checks, since we are adding at most only 5 keys */
5485 	env->envp[env->envp_idx++] = NULL;
5486 	kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5487 	kfree(env);
5488 }
5489 
5490 static void kvm_init_debug(void)
5491 {
5492 	const struct file_operations *fops;
5493 	const struct _kvm_stats_desc *pdesc;
5494 	int i;
5495 
5496 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5497 
5498 	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5499 		pdesc = &kvm_vm_stats_desc[i];
5500 		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5501 			fops = &vm_stat_fops;
5502 		else
5503 			fops = &vm_stat_readonly_fops;
5504 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5505 				kvm_debugfs_dir,
5506 				(void *)(long)pdesc->desc.offset, fops);
5507 	}
5508 
5509 	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5510 		pdesc = &kvm_vcpu_stats_desc[i];
5511 		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5512 			fops = &vcpu_stat_fops;
5513 		else
5514 			fops = &vcpu_stat_readonly_fops;
5515 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5516 				kvm_debugfs_dir,
5517 				(void *)(long)pdesc->desc.offset, fops);
5518 	}
5519 }
5520 
5521 static int kvm_suspend(void)
5522 {
5523 	if (kvm_usage_count)
5524 		hardware_disable_nolock(NULL);
5525 	return 0;
5526 }
5527 
5528 static void kvm_resume(void)
5529 {
5530 	if (kvm_usage_count) {
5531 #ifdef CONFIG_LOCKDEP
5532 		WARN_ON(lockdep_is_held(&kvm_count_lock));
5533 #endif
5534 		hardware_enable_nolock(NULL);
5535 	}
5536 }
5537 
5538 static struct syscore_ops kvm_syscore_ops = {
5539 	.suspend = kvm_suspend,
5540 	.resume = kvm_resume,
5541 };
5542 
5543 static inline
5544 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5545 {
5546 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
5547 }
5548 
5549 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5550 {
5551 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5552 
5553 	WRITE_ONCE(vcpu->preempted, false);
5554 	WRITE_ONCE(vcpu->ready, false);
5555 
5556 	__this_cpu_write(kvm_running_vcpu, vcpu);
5557 	kvm_arch_sched_in(vcpu, cpu);
5558 	kvm_arch_vcpu_load(vcpu, cpu);
5559 }
5560 
5561 static void kvm_sched_out(struct preempt_notifier *pn,
5562 			  struct task_struct *next)
5563 {
5564 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5565 
5566 	if (current->on_rq) {
5567 		WRITE_ONCE(vcpu->preempted, true);
5568 		WRITE_ONCE(vcpu->ready, true);
5569 	}
5570 	kvm_arch_vcpu_put(vcpu);
5571 	__this_cpu_write(kvm_running_vcpu, NULL);
5572 }
5573 
5574 /**
5575  * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5576  *
5577  * We can disable preemption locally around accessing the per-CPU variable,
5578  * and use the resolved vcpu pointer after enabling preemption again,
5579  * because even if the current thread is migrated to another CPU, reading
5580  * the per-CPU value later will give us the same value as we update the
5581  * per-CPU variable in the preempt notifier handlers.
5582  */
5583 struct kvm_vcpu *kvm_get_running_vcpu(void)
5584 {
5585 	struct kvm_vcpu *vcpu;
5586 
5587 	preempt_disable();
5588 	vcpu = __this_cpu_read(kvm_running_vcpu);
5589 	preempt_enable();
5590 
5591 	return vcpu;
5592 }
5593 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5594 
5595 /**
5596  * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5597  */
5598 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5599 {
5600         return &kvm_running_vcpu;
5601 }
5602 
5603 #ifdef CONFIG_GUEST_PERF_EVENTS
5604 static unsigned int kvm_guest_state(void)
5605 {
5606 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5607 	unsigned int state;
5608 
5609 	if (!kvm_arch_pmi_in_guest(vcpu))
5610 		return 0;
5611 
5612 	state = PERF_GUEST_ACTIVE;
5613 	if (!kvm_arch_vcpu_in_kernel(vcpu))
5614 		state |= PERF_GUEST_USER;
5615 
5616 	return state;
5617 }
5618 
5619 static unsigned long kvm_guest_get_ip(void)
5620 {
5621 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5622 
5623 	/* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5624 	if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5625 		return 0;
5626 
5627 	return kvm_arch_vcpu_get_ip(vcpu);
5628 }
5629 
5630 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5631 	.state			= kvm_guest_state,
5632 	.get_ip			= kvm_guest_get_ip,
5633 	.handle_intel_pt_intr	= NULL,
5634 };
5635 
5636 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5637 {
5638 	kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5639 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5640 }
5641 void kvm_unregister_perf_callbacks(void)
5642 {
5643 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5644 }
5645 #endif
5646 
5647 struct kvm_cpu_compat_check {
5648 	void *opaque;
5649 	int *ret;
5650 };
5651 
5652 static void check_processor_compat(void *data)
5653 {
5654 	struct kvm_cpu_compat_check *c = data;
5655 
5656 	*c->ret = kvm_arch_check_processor_compat(c->opaque);
5657 }
5658 
5659 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5660 		  struct module *module)
5661 {
5662 	struct kvm_cpu_compat_check c;
5663 	int r;
5664 	int cpu;
5665 
5666 	r = kvm_arch_init(opaque);
5667 	if (r)
5668 		goto out_fail;
5669 
5670 	/*
5671 	 * kvm_arch_init makes sure there's at most one caller
5672 	 * for architectures that support multiple implementations,
5673 	 * like intel and amd on x86.
5674 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5675 	 * conflicts in case kvm is already setup for another implementation.
5676 	 */
5677 	r = kvm_irqfd_init();
5678 	if (r)
5679 		goto out_irqfd;
5680 
5681 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5682 		r = -ENOMEM;
5683 		goto out_free_0;
5684 	}
5685 
5686 	r = kvm_arch_hardware_setup(opaque);
5687 	if (r < 0)
5688 		goto out_free_1;
5689 
5690 	c.ret = &r;
5691 	c.opaque = opaque;
5692 	for_each_online_cpu(cpu) {
5693 		smp_call_function_single(cpu, check_processor_compat, &c, 1);
5694 		if (r < 0)
5695 			goto out_free_2;
5696 	}
5697 
5698 	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5699 				      kvm_starting_cpu, kvm_dying_cpu);
5700 	if (r)
5701 		goto out_free_2;
5702 	register_reboot_notifier(&kvm_reboot_notifier);
5703 
5704 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
5705 	if (!vcpu_align)
5706 		vcpu_align = __alignof__(struct kvm_vcpu);
5707 	kvm_vcpu_cache =
5708 		kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5709 					   SLAB_ACCOUNT,
5710 					   offsetof(struct kvm_vcpu, arch),
5711 					   offsetofend(struct kvm_vcpu, stats_id)
5712 					   - offsetof(struct kvm_vcpu, arch),
5713 					   NULL);
5714 	if (!kvm_vcpu_cache) {
5715 		r = -ENOMEM;
5716 		goto out_free_3;
5717 	}
5718 
5719 	for_each_possible_cpu(cpu) {
5720 		if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5721 					    GFP_KERNEL, cpu_to_node(cpu))) {
5722 			r = -ENOMEM;
5723 			goto out_free_4;
5724 		}
5725 	}
5726 
5727 	r = kvm_async_pf_init();
5728 	if (r)
5729 		goto out_free_5;
5730 
5731 	kvm_chardev_ops.owner = module;
5732 	kvm_vm_fops.owner = module;
5733 	kvm_vcpu_fops.owner = module;
5734 
5735 	r = misc_register(&kvm_dev);
5736 	if (r) {
5737 		pr_err("kvm: misc device register failed\n");
5738 		goto out_unreg;
5739 	}
5740 
5741 	register_syscore_ops(&kvm_syscore_ops);
5742 
5743 	kvm_preempt_ops.sched_in = kvm_sched_in;
5744 	kvm_preempt_ops.sched_out = kvm_sched_out;
5745 
5746 	kvm_init_debug();
5747 
5748 	r = kvm_vfio_ops_init();
5749 	WARN_ON(r);
5750 
5751 	return 0;
5752 
5753 out_unreg:
5754 	kvm_async_pf_deinit();
5755 out_free_5:
5756 	for_each_possible_cpu(cpu)
5757 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5758 out_free_4:
5759 	kmem_cache_destroy(kvm_vcpu_cache);
5760 out_free_3:
5761 	unregister_reboot_notifier(&kvm_reboot_notifier);
5762 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5763 out_free_2:
5764 	kvm_arch_hardware_unsetup();
5765 out_free_1:
5766 	free_cpumask_var(cpus_hardware_enabled);
5767 out_free_0:
5768 	kvm_irqfd_exit();
5769 out_irqfd:
5770 	kvm_arch_exit();
5771 out_fail:
5772 	return r;
5773 }
5774 EXPORT_SYMBOL_GPL(kvm_init);
5775 
5776 void kvm_exit(void)
5777 {
5778 	int cpu;
5779 
5780 	debugfs_remove_recursive(kvm_debugfs_dir);
5781 	misc_deregister(&kvm_dev);
5782 	for_each_possible_cpu(cpu)
5783 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5784 	kmem_cache_destroy(kvm_vcpu_cache);
5785 	kvm_async_pf_deinit();
5786 	unregister_syscore_ops(&kvm_syscore_ops);
5787 	unregister_reboot_notifier(&kvm_reboot_notifier);
5788 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5789 	on_each_cpu(hardware_disable_nolock, NULL, 1);
5790 	kvm_arch_hardware_unsetup();
5791 	kvm_arch_exit();
5792 	kvm_irqfd_exit();
5793 	free_cpumask_var(cpus_hardware_enabled);
5794 	kvm_vfio_ops_exit();
5795 }
5796 EXPORT_SYMBOL_GPL(kvm_exit);
5797 
5798 struct kvm_vm_worker_thread_context {
5799 	struct kvm *kvm;
5800 	struct task_struct *parent;
5801 	struct completion init_done;
5802 	kvm_vm_thread_fn_t thread_fn;
5803 	uintptr_t data;
5804 	int err;
5805 };
5806 
5807 static int kvm_vm_worker_thread(void *context)
5808 {
5809 	/*
5810 	 * The init_context is allocated on the stack of the parent thread, so
5811 	 * we have to locally copy anything that is needed beyond initialization
5812 	 */
5813 	struct kvm_vm_worker_thread_context *init_context = context;
5814 	struct kvm *kvm = init_context->kvm;
5815 	kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5816 	uintptr_t data = init_context->data;
5817 	int err;
5818 
5819 	err = kthread_park(current);
5820 	/* kthread_park(current) is never supposed to return an error */
5821 	WARN_ON(err != 0);
5822 	if (err)
5823 		goto init_complete;
5824 
5825 	err = cgroup_attach_task_all(init_context->parent, current);
5826 	if (err) {
5827 		kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5828 			__func__, err);
5829 		goto init_complete;
5830 	}
5831 
5832 	set_user_nice(current, task_nice(init_context->parent));
5833 
5834 init_complete:
5835 	init_context->err = err;
5836 	complete(&init_context->init_done);
5837 	init_context = NULL;
5838 
5839 	if (err)
5840 		return err;
5841 
5842 	/* Wait to be woken up by the spawner before proceeding. */
5843 	kthread_parkme();
5844 
5845 	if (!kthread_should_stop())
5846 		err = thread_fn(kvm, data);
5847 
5848 	return err;
5849 }
5850 
5851 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5852 				uintptr_t data, const char *name,
5853 				struct task_struct **thread_ptr)
5854 {
5855 	struct kvm_vm_worker_thread_context init_context = {};
5856 	struct task_struct *thread;
5857 
5858 	*thread_ptr = NULL;
5859 	init_context.kvm = kvm;
5860 	init_context.parent = current;
5861 	init_context.thread_fn = thread_fn;
5862 	init_context.data = data;
5863 	init_completion(&init_context.init_done);
5864 
5865 	thread = kthread_run(kvm_vm_worker_thread, &init_context,
5866 			     "%s-%d", name, task_pid_nr(current));
5867 	if (IS_ERR(thread))
5868 		return PTR_ERR(thread);
5869 
5870 	/* kthread_run is never supposed to return NULL */
5871 	WARN_ON(thread == NULL);
5872 
5873 	wait_for_completion(&init_context.init_done);
5874 
5875 	if (!init_context.err)
5876 		*thread_ptr = thread;
5877 
5878 	return init_context.err;
5879 }
5880