xref: /openbmc/linux/virt/kvm/kvm_main.c (revision aac28965)
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 struct file_operations kvm_chardev_ops;
121 
122 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
123 			   unsigned long arg);
124 #ifdef CONFIG_KVM_COMPAT
125 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
126 				  unsigned long arg);
127 #define KVM_COMPAT(c)	.compat_ioctl	= (c)
128 #else
129 /*
130  * For architectures that don't implement a compat infrastructure,
131  * adopt a double line of defense:
132  * - Prevent a compat task from opening /dev/kvm
133  * - If the open has been done by a 64bit task, and the KVM fd
134  *   passed to a compat task, let the ioctls fail.
135  */
136 static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
137 				unsigned long arg) { return -EINVAL; }
138 
139 static int kvm_no_compat_open(struct inode *inode, struct file *file)
140 {
141 	return is_compat_task() ? -ENODEV : 0;
142 }
143 #define KVM_COMPAT(c)	.compat_ioctl	= kvm_no_compat_ioctl,	\
144 			.open		= kvm_no_compat_open
145 #endif
146 static int hardware_enable_all(void);
147 static void hardware_disable_all(void);
148 
149 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
150 
151 __visible bool kvm_rebooting;
152 EXPORT_SYMBOL_GPL(kvm_rebooting);
153 
154 #define KVM_EVENT_CREATE_VM 0
155 #define KVM_EVENT_DESTROY_VM 1
156 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
157 static unsigned long long kvm_createvm_count;
158 static unsigned long long kvm_active_vms;
159 
160 static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
161 
162 __weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
163 						   unsigned long start, unsigned long end)
164 {
165 }
166 
167 __weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
168 {
169 }
170 
171 bool kvm_is_zone_device_pfn(kvm_pfn_t pfn)
172 {
173 	/*
174 	 * The metadata used by is_zone_device_page() to determine whether or
175 	 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
176 	 * the device has been pinned, e.g. by get_user_pages().  WARN if the
177 	 * page_count() is zero to help detect bad usage of this helper.
178 	 */
179 	if (!pfn_valid(pfn) || WARN_ON_ONCE(!page_count(pfn_to_page(pfn))))
180 		return false;
181 
182 	return is_zone_device_page(pfn_to_page(pfn));
183 }
184 
185 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
186 {
187 	/*
188 	 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
189 	 * perspective they are "normal" pages, albeit with slightly different
190 	 * usage rules.
191 	 */
192 	if (pfn_valid(pfn))
193 		return PageReserved(pfn_to_page(pfn)) &&
194 		       !is_zero_pfn(pfn) &&
195 		       !kvm_is_zone_device_pfn(pfn);
196 
197 	return true;
198 }
199 
200 /*
201  * Switches to specified vcpu, until a matching vcpu_put()
202  */
203 void vcpu_load(struct kvm_vcpu *vcpu)
204 {
205 	int cpu = get_cpu();
206 
207 	__this_cpu_write(kvm_running_vcpu, vcpu);
208 	preempt_notifier_register(&vcpu->preempt_notifier);
209 	kvm_arch_vcpu_load(vcpu, cpu);
210 	put_cpu();
211 }
212 EXPORT_SYMBOL_GPL(vcpu_load);
213 
214 void vcpu_put(struct kvm_vcpu *vcpu)
215 {
216 	preempt_disable();
217 	kvm_arch_vcpu_put(vcpu);
218 	preempt_notifier_unregister(&vcpu->preempt_notifier);
219 	__this_cpu_write(kvm_running_vcpu, NULL);
220 	preempt_enable();
221 }
222 EXPORT_SYMBOL_GPL(vcpu_put);
223 
224 /* TODO: merge with kvm_arch_vcpu_should_kick */
225 static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
226 {
227 	int mode = kvm_vcpu_exiting_guest_mode(vcpu);
228 
229 	/*
230 	 * We need to wait for the VCPU to reenable interrupts and get out of
231 	 * READING_SHADOW_PAGE_TABLES mode.
232 	 */
233 	if (req & KVM_REQUEST_WAIT)
234 		return mode != OUTSIDE_GUEST_MODE;
235 
236 	/*
237 	 * Need to kick a running VCPU, but otherwise there is nothing to do.
238 	 */
239 	return mode == IN_GUEST_MODE;
240 }
241 
242 static void ack_flush(void *_completed)
243 {
244 }
245 
246 static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
247 {
248 	if (cpumask_empty(cpus))
249 		return false;
250 
251 	smp_call_function_many(cpus, ack_flush, NULL, wait);
252 	return true;
253 }
254 
255 static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
256 				  struct cpumask *tmp, int current_cpu)
257 {
258 	int cpu;
259 
260 	if (likely(!(req & KVM_REQUEST_NO_ACTION)))
261 		__kvm_make_request(req, vcpu);
262 
263 	if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
264 		return;
265 
266 	/*
267 	 * Note, the vCPU could get migrated to a different pCPU at any point
268 	 * after kvm_request_needs_ipi(), which could result in sending an IPI
269 	 * to the previous pCPU.  But, that's OK because the purpose of the IPI
270 	 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
271 	 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
272 	 * after this point is also OK, as the requirement is only that KVM wait
273 	 * for vCPUs that were reading SPTEs _before_ any changes were
274 	 * finalized. See kvm_vcpu_kick() for more details on handling requests.
275 	 */
276 	if (kvm_request_needs_ipi(vcpu, req)) {
277 		cpu = READ_ONCE(vcpu->cpu);
278 		if (cpu != -1 && cpu != current_cpu)
279 			__cpumask_set_cpu(cpu, tmp);
280 	}
281 }
282 
283 bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
284 				 unsigned long *vcpu_bitmap)
285 {
286 	struct kvm_vcpu *vcpu;
287 	struct cpumask *cpus;
288 	int i, me;
289 	bool called;
290 
291 	me = get_cpu();
292 
293 	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
294 	cpumask_clear(cpus);
295 
296 	for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
297 		vcpu = kvm_get_vcpu(kvm, i);
298 		if (!vcpu)
299 			continue;
300 		kvm_make_vcpu_request(vcpu, req, cpus, me);
301 	}
302 
303 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
304 	put_cpu();
305 
306 	return called;
307 }
308 
309 bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
310 				      struct kvm_vcpu *except)
311 {
312 	struct kvm_vcpu *vcpu;
313 	struct cpumask *cpus;
314 	unsigned long i;
315 	bool called;
316 	int me;
317 
318 	me = get_cpu();
319 
320 	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
321 	cpumask_clear(cpus);
322 
323 	kvm_for_each_vcpu(i, vcpu, kvm) {
324 		if (vcpu == except)
325 			continue;
326 		kvm_make_vcpu_request(vcpu, req, cpus, me);
327 	}
328 
329 	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
330 	put_cpu();
331 
332 	return called;
333 }
334 
335 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
336 {
337 	return kvm_make_all_cpus_request_except(kvm, req, NULL);
338 }
339 EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
340 
341 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
342 void kvm_flush_remote_tlbs(struct kvm *kvm)
343 {
344 	++kvm->stat.generic.remote_tlb_flush_requests;
345 
346 	/*
347 	 * We want to publish modifications to the page tables before reading
348 	 * mode. Pairs with a memory barrier in arch-specific code.
349 	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
350 	 * and smp_mb in walk_shadow_page_lockless_begin/end.
351 	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
352 	 *
353 	 * There is already an smp_mb__after_atomic() before
354 	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
355 	 * barrier here.
356 	 */
357 	if (!kvm_arch_flush_remote_tlb(kvm)
358 	    || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
359 		++kvm->stat.generic.remote_tlb_flush;
360 }
361 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
362 #endif
363 
364 static void kvm_flush_shadow_all(struct kvm *kvm)
365 {
366 	kvm_arch_flush_shadow_all(kvm);
367 	kvm_arch_guest_memory_reclaimed(kvm);
368 }
369 
370 #ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
371 static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
372 					       gfp_t gfp_flags)
373 {
374 	gfp_flags |= mc->gfp_zero;
375 
376 	if (mc->kmem_cache)
377 		return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
378 	else
379 		return (void *)__get_free_page(gfp_flags);
380 }
381 
382 int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
383 {
384 	void *obj;
385 
386 	if (mc->nobjs >= min)
387 		return 0;
388 	while (mc->nobjs < ARRAY_SIZE(mc->objects)) {
389 		obj = mmu_memory_cache_alloc_obj(mc, GFP_KERNEL_ACCOUNT);
390 		if (!obj)
391 			return mc->nobjs >= min ? 0 : -ENOMEM;
392 		mc->objects[mc->nobjs++] = obj;
393 	}
394 	return 0;
395 }
396 
397 int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
398 {
399 	return mc->nobjs;
400 }
401 
402 void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
403 {
404 	while (mc->nobjs) {
405 		if (mc->kmem_cache)
406 			kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
407 		else
408 			free_page((unsigned long)mc->objects[--mc->nobjs]);
409 	}
410 }
411 
412 void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
413 {
414 	void *p;
415 
416 	if (WARN_ON(!mc->nobjs))
417 		p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
418 	else
419 		p = mc->objects[--mc->nobjs];
420 	BUG_ON(!p);
421 	return p;
422 }
423 #endif
424 
425 static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
426 {
427 	mutex_init(&vcpu->mutex);
428 	vcpu->cpu = -1;
429 	vcpu->kvm = kvm;
430 	vcpu->vcpu_id = id;
431 	vcpu->pid = NULL;
432 #ifndef __KVM_HAVE_ARCH_WQP
433 	rcuwait_init(&vcpu->wait);
434 #endif
435 	kvm_async_pf_vcpu_init(vcpu);
436 
437 	kvm_vcpu_set_in_spin_loop(vcpu, false);
438 	kvm_vcpu_set_dy_eligible(vcpu, false);
439 	vcpu->preempted = false;
440 	vcpu->ready = false;
441 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
442 	vcpu->last_used_slot = NULL;
443 }
444 
445 static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
446 {
447 	kvm_arch_vcpu_destroy(vcpu);
448 	kvm_dirty_ring_free(&vcpu->dirty_ring);
449 
450 	/*
451 	 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
452 	 * the vcpu->pid pointer, and at destruction time all file descriptors
453 	 * are already gone.
454 	 */
455 	put_pid(rcu_dereference_protected(vcpu->pid, 1));
456 
457 	free_page((unsigned long)vcpu->run);
458 	kmem_cache_free(kvm_vcpu_cache, vcpu);
459 }
460 
461 void kvm_destroy_vcpus(struct kvm *kvm)
462 {
463 	unsigned long i;
464 	struct kvm_vcpu *vcpu;
465 
466 	kvm_for_each_vcpu(i, vcpu, kvm) {
467 		kvm_vcpu_destroy(vcpu);
468 		xa_erase(&kvm->vcpu_array, i);
469 	}
470 
471 	atomic_set(&kvm->online_vcpus, 0);
472 }
473 EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
474 
475 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
476 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
477 {
478 	return container_of(mn, struct kvm, mmu_notifier);
479 }
480 
481 static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
482 					      struct mm_struct *mm,
483 					      unsigned long start, unsigned long end)
484 {
485 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
486 	int idx;
487 
488 	idx = srcu_read_lock(&kvm->srcu);
489 	kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
490 	srcu_read_unlock(&kvm->srcu, idx);
491 }
492 
493 typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
494 
495 typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
496 			     unsigned long end);
497 
498 typedef void (*on_unlock_fn_t)(struct kvm *kvm);
499 
500 struct kvm_hva_range {
501 	unsigned long start;
502 	unsigned long end;
503 	pte_t pte;
504 	hva_handler_t handler;
505 	on_lock_fn_t on_lock;
506 	on_unlock_fn_t on_unlock;
507 	bool flush_on_ret;
508 	bool may_block;
509 };
510 
511 /*
512  * Use a dedicated stub instead of NULL to indicate that there is no callback
513  * function/handler.  The compiler technically can't guarantee that a real
514  * function will have a non-zero address, and so it will generate code to
515  * check for !NULL, whereas comparing against a stub will be elided at compile
516  * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
517  */
518 static void kvm_null_fn(void)
519 {
520 
521 }
522 #define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
523 
524 /* Iterate over each memslot intersecting [start, last] (inclusive) range */
525 #define kvm_for_each_memslot_in_hva_range(node, slots, start, last)	     \
526 	for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
527 	     node;							     \
528 	     node = interval_tree_iter_next(node, start, last))	     \
529 
530 static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
531 						  const struct kvm_hva_range *range)
532 {
533 	bool ret = false, locked = false;
534 	struct kvm_gfn_range gfn_range;
535 	struct kvm_memory_slot *slot;
536 	struct kvm_memslots *slots;
537 	int i, idx;
538 
539 	if (WARN_ON_ONCE(range->end <= range->start))
540 		return 0;
541 
542 	/* A null handler is allowed if and only if on_lock() is provided. */
543 	if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
544 			 IS_KVM_NULL_FN(range->handler)))
545 		return 0;
546 
547 	idx = srcu_read_lock(&kvm->srcu);
548 
549 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
550 		struct interval_tree_node *node;
551 
552 		slots = __kvm_memslots(kvm, i);
553 		kvm_for_each_memslot_in_hva_range(node, slots,
554 						  range->start, range->end - 1) {
555 			unsigned long hva_start, hva_end;
556 
557 			slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
558 			hva_start = max(range->start, slot->userspace_addr);
559 			hva_end = min(range->end, slot->userspace_addr +
560 						  (slot->npages << PAGE_SHIFT));
561 
562 			/*
563 			 * To optimize for the likely case where the address
564 			 * range is covered by zero or one memslots, don't
565 			 * bother making these conditional (to avoid writes on
566 			 * the second or later invocation of the handler).
567 			 */
568 			gfn_range.pte = range->pte;
569 			gfn_range.may_block = range->may_block;
570 
571 			/*
572 			 * {gfn(page) | page intersects with [hva_start, hva_end)} =
573 			 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
574 			 */
575 			gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
576 			gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
577 			gfn_range.slot = slot;
578 
579 			if (!locked) {
580 				locked = true;
581 				KVM_MMU_LOCK(kvm);
582 				if (!IS_KVM_NULL_FN(range->on_lock))
583 					range->on_lock(kvm, range->start, range->end);
584 				if (IS_KVM_NULL_FN(range->handler))
585 					break;
586 			}
587 			ret |= range->handler(kvm, &gfn_range);
588 		}
589 	}
590 
591 	if (range->flush_on_ret && ret)
592 		kvm_flush_remote_tlbs(kvm);
593 
594 	if (locked) {
595 		KVM_MMU_UNLOCK(kvm);
596 		if (!IS_KVM_NULL_FN(range->on_unlock))
597 			range->on_unlock(kvm);
598 	}
599 
600 	srcu_read_unlock(&kvm->srcu, idx);
601 
602 	/* The notifiers are averse to booleans. :-( */
603 	return (int)ret;
604 }
605 
606 static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
607 						unsigned long start,
608 						unsigned long end,
609 						pte_t pte,
610 						hva_handler_t handler)
611 {
612 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
613 	const struct kvm_hva_range range = {
614 		.start		= start,
615 		.end		= end,
616 		.pte		= pte,
617 		.handler	= handler,
618 		.on_lock	= (void *)kvm_null_fn,
619 		.on_unlock	= (void *)kvm_null_fn,
620 		.flush_on_ret	= true,
621 		.may_block	= false,
622 	};
623 
624 	return __kvm_handle_hva_range(kvm, &range);
625 }
626 
627 static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
628 							 unsigned long start,
629 							 unsigned long end,
630 							 hva_handler_t handler)
631 {
632 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
633 	const struct kvm_hva_range range = {
634 		.start		= start,
635 		.end		= end,
636 		.pte		= __pte(0),
637 		.handler	= handler,
638 		.on_lock	= (void *)kvm_null_fn,
639 		.on_unlock	= (void *)kvm_null_fn,
640 		.flush_on_ret	= false,
641 		.may_block	= false,
642 	};
643 
644 	return __kvm_handle_hva_range(kvm, &range);
645 }
646 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
647 					struct mm_struct *mm,
648 					unsigned long address,
649 					pte_t pte)
650 {
651 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
652 
653 	trace_kvm_set_spte_hva(address);
654 
655 	/*
656 	 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
657 	 * If mmu_notifier_count is zero, then no in-progress invalidations,
658 	 * including this one, found a relevant memslot at start(); rechecking
659 	 * memslots here is unnecessary.  Note, a false positive (count elevated
660 	 * by a different invalidation) is sub-optimal but functionally ok.
661 	 */
662 	WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
663 	if (!READ_ONCE(kvm->mmu_notifier_count))
664 		return;
665 
666 	kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
667 }
668 
669 void kvm_inc_notifier_count(struct kvm *kvm, unsigned long start,
670 				   unsigned long end)
671 {
672 	/*
673 	 * The count increase must become visible at unlock time as no
674 	 * spte can be established without taking the mmu_lock and
675 	 * count is also read inside the mmu_lock critical section.
676 	 */
677 	kvm->mmu_notifier_count++;
678 	if (likely(kvm->mmu_notifier_count == 1)) {
679 		kvm->mmu_notifier_range_start = start;
680 		kvm->mmu_notifier_range_end = end;
681 	} else {
682 		/*
683 		 * Fully tracking multiple concurrent ranges has diminishing
684 		 * returns. Keep things simple and just find the minimal range
685 		 * which includes the current and new ranges. As there won't be
686 		 * enough information to subtract a range after its invalidate
687 		 * completes, any ranges invalidated concurrently will
688 		 * accumulate and persist until all outstanding invalidates
689 		 * complete.
690 		 */
691 		kvm->mmu_notifier_range_start =
692 			min(kvm->mmu_notifier_range_start, start);
693 		kvm->mmu_notifier_range_end =
694 			max(kvm->mmu_notifier_range_end, end);
695 	}
696 }
697 
698 static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
699 					const struct mmu_notifier_range *range)
700 {
701 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
702 	const struct kvm_hva_range hva_range = {
703 		.start		= range->start,
704 		.end		= range->end,
705 		.pte		= __pte(0),
706 		.handler	= kvm_unmap_gfn_range,
707 		.on_lock	= kvm_inc_notifier_count,
708 		.on_unlock	= kvm_arch_guest_memory_reclaimed,
709 		.flush_on_ret	= true,
710 		.may_block	= mmu_notifier_range_blockable(range),
711 	};
712 
713 	trace_kvm_unmap_hva_range(range->start, range->end);
714 
715 	/*
716 	 * Prevent memslot modification between range_start() and range_end()
717 	 * so that conditionally locking provides the same result in both
718 	 * functions.  Without that guarantee, the mmu_notifier_count
719 	 * adjustments will be imbalanced.
720 	 *
721 	 * Pairs with the decrement in range_end().
722 	 */
723 	spin_lock(&kvm->mn_invalidate_lock);
724 	kvm->mn_active_invalidate_count++;
725 	spin_unlock(&kvm->mn_invalidate_lock);
726 
727 	gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
728 					  hva_range.may_block);
729 
730 	__kvm_handle_hva_range(kvm, &hva_range);
731 
732 	return 0;
733 }
734 
735 void kvm_dec_notifier_count(struct kvm *kvm, unsigned long start,
736 				   unsigned long end)
737 {
738 	/*
739 	 * This sequence increase will notify the kvm page fault that
740 	 * the page that is going to be mapped in the spte could have
741 	 * been freed.
742 	 */
743 	kvm->mmu_notifier_seq++;
744 	smp_wmb();
745 	/*
746 	 * The above sequence increase must be visible before the
747 	 * below count decrease, which is ensured by the smp_wmb above
748 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
749 	 */
750 	kvm->mmu_notifier_count--;
751 }
752 
753 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
754 					const struct mmu_notifier_range *range)
755 {
756 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
757 	const struct kvm_hva_range hva_range = {
758 		.start		= range->start,
759 		.end		= range->end,
760 		.pte		= __pte(0),
761 		.handler	= (void *)kvm_null_fn,
762 		.on_lock	= kvm_dec_notifier_count,
763 		.on_unlock	= (void *)kvm_null_fn,
764 		.flush_on_ret	= false,
765 		.may_block	= mmu_notifier_range_blockable(range),
766 	};
767 	bool wake;
768 
769 	__kvm_handle_hva_range(kvm, &hva_range);
770 
771 	/* Pairs with the increment in range_start(). */
772 	spin_lock(&kvm->mn_invalidate_lock);
773 	wake = (--kvm->mn_active_invalidate_count == 0);
774 	spin_unlock(&kvm->mn_invalidate_lock);
775 
776 	/*
777 	 * There can only be one waiter, since the wait happens under
778 	 * slots_lock.
779 	 */
780 	if (wake)
781 		rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
782 
783 	BUG_ON(kvm->mmu_notifier_count < 0);
784 }
785 
786 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
787 					      struct mm_struct *mm,
788 					      unsigned long start,
789 					      unsigned long end)
790 {
791 	trace_kvm_age_hva(start, end);
792 
793 	return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
794 }
795 
796 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
797 					struct mm_struct *mm,
798 					unsigned long start,
799 					unsigned long end)
800 {
801 	trace_kvm_age_hva(start, end);
802 
803 	/*
804 	 * Even though we do not flush TLB, this will still adversely
805 	 * affect performance on pre-Haswell Intel EPT, where there is
806 	 * no EPT Access Bit to clear so that we have to tear down EPT
807 	 * tables instead. If we find this unacceptable, we can always
808 	 * add a parameter to kvm_age_hva so that it effectively doesn't
809 	 * do anything on clear_young.
810 	 *
811 	 * Also note that currently we never issue secondary TLB flushes
812 	 * from clear_young, leaving this job up to the regular system
813 	 * cadence. If we find this inaccurate, we might come up with a
814 	 * more sophisticated heuristic later.
815 	 */
816 	return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
817 }
818 
819 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
820 				       struct mm_struct *mm,
821 				       unsigned long address)
822 {
823 	trace_kvm_test_age_hva(address);
824 
825 	return kvm_handle_hva_range_no_flush(mn, address, address + 1,
826 					     kvm_test_age_gfn);
827 }
828 
829 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
830 				     struct mm_struct *mm)
831 {
832 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
833 	int idx;
834 
835 	idx = srcu_read_lock(&kvm->srcu);
836 	kvm_flush_shadow_all(kvm);
837 	srcu_read_unlock(&kvm->srcu, idx);
838 }
839 
840 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
841 	.invalidate_range	= kvm_mmu_notifier_invalidate_range,
842 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
843 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
844 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
845 	.clear_young		= kvm_mmu_notifier_clear_young,
846 	.test_young		= kvm_mmu_notifier_test_young,
847 	.change_pte		= kvm_mmu_notifier_change_pte,
848 	.release		= kvm_mmu_notifier_release,
849 };
850 
851 static int kvm_init_mmu_notifier(struct kvm *kvm)
852 {
853 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
854 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
855 }
856 
857 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
858 
859 static int kvm_init_mmu_notifier(struct kvm *kvm)
860 {
861 	return 0;
862 }
863 
864 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
865 
866 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
867 static int kvm_pm_notifier_call(struct notifier_block *bl,
868 				unsigned long state,
869 				void *unused)
870 {
871 	struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
872 
873 	return kvm_arch_pm_notifier(kvm, state);
874 }
875 
876 static void kvm_init_pm_notifier(struct kvm *kvm)
877 {
878 	kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
879 	/* Suspend KVM before we suspend ftrace, RCU, etc. */
880 	kvm->pm_notifier.priority = INT_MAX;
881 	register_pm_notifier(&kvm->pm_notifier);
882 }
883 
884 static void kvm_destroy_pm_notifier(struct kvm *kvm)
885 {
886 	unregister_pm_notifier(&kvm->pm_notifier);
887 }
888 #else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
889 static void kvm_init_pm_notifier(struct kvm *kvm)
890 {
891 }
892 
893 static void kvm_destroy_pm_notifier(struct kvm *kvm)
894 {
895 }
896 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
897 
898 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
899 {
900 	if (!memslot->dirty_bitmap)
901 		return;
902 
903 	kvfree(memslot->dirty_bitmap);
904 	memslot->dirty_bitmap = NULL;
905 }
906 
907 /* This does not remove the slot from struct kvm_memslots data structures */
908 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
909 {
910 	kvm_destroy_dirty_bitmap(slot);
911 
912 	kvm_arch_free_memslot(kvm, slot);
913 
914 	kfree(slot);
915 }
916 
917 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
918 {
919 	struct hlist_node *idnode;
920 	struct kvm_memory_slot *memslot;
921 	int bkt;
922 
923 	/*
924 	 * The same memslot objects live in both active and inactive sets,
925 	 * arbitrarily free using index '1' so the second invocation of this
926 	 * function isn't operating over a structure with dangling pointers
927 	 * (even though this function isn't actually touching them).
928 	 */
929 	if (!slots->node_idx)
930 		return;
931 
932 	hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
933 		kvm_free_memslot(kvm, memslot);
934 }
935 
936 static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
937 {
938 	switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
939 	case KVM_STATS_TYPE_INSTANT:
940 		return 0444;
941 	case KVM_STATS_TYPE_CUMULATIVE:
942 	case KVM_STATS_TYPE_PEAK:
943 	default:
944 		return 0644;
945 	}
946 }
947 
948 
949 static void kvm_destroy_vm_debugfs(struct kvm *kvm)
950 {
951 	int i;
952 	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
953 				      kvm_vcpu_stats_header.num_desc;
954 
955 	if (IS_ERR(kvm->debugfs_dentry))
956 		return;
957 
958 	debugfs_remove_recursive(kvm->debugfs_dentry);
959 
960 	if (kvm->debugfs_stat_data) {
961 		for (i = 0; i < kvm_debugfs_num_entries; i++)
962 			kfree(kvm->debugfs_stat_data[i]);
963 		kfree(kvm->debugfs_stat_data);
964 	}
965 }
966 
967 static int kvm_create_vm_debugfs(struct kvm *kvm, int fd)
968 {
969 	static DEFINE_MUTEX(kvm_debugfs_lock);
970 	struct dentry *dent;
971 	char dir_name[ITOA_MAX_LEN * 2];
972 	struct kvm_stat_data *stat_data;
973 	const struct _kvm_stats_desc *pdesc;
974 	int i, ret;
975 	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
976 				      kvm_vcpu_stats_header.num_desc;
977 
978 	if (!debugfs_initialized())
979 		return 0;
980 
981 	snprintf(dir_name, sizeof(dir_name), "%d-%d", task_pid_nr(current), fd);
982 	mutex_lock(&kvm_debugfs_lock);
983 	dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
984 	if (dent) {
985 		pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
986 		dput(dent);
987 		mutex_unlock(&kvm_debugfs_lock);
988 		return 0;
989 	}
990 	dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
991 	mutex_unlock(&kvm_debugfs_lock);
992 	if (IS_ERR(dent))
993 		return 0;
994 
995 	kvm->debugfs_dentry = dent;
996 	kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
997 					 sizeof(*kvm->debugfs_stat_data),
998 					 GFP_KERNEL_ACCOUNT);
999 	if (!kvm->debugfs_stat_data)
1000 		return -ENOMEM;
1001 
1002 	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1003 		pdesc = &kvm_vm_stats_desc[i];
1004 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1005 		if (!stat_data)
1006 			return -ENOMEM;
1007 
1008 		stat_data->kvm = kvm;
1009 		stat_data->desc = pdesc;
1010 		stat_data->kind = KVM_STAT_VM;
1011 		kvm->debugfs_stat_data[i] = stat_data;
1012 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1013 				    kvm->debugfs_dentry, stat_data,
1014 				    &stat_fops_per_vm);
1015 	}
1016 
1017 	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1018 		pdesc = &kvm_vcpu_stats_desc[i];
1019 		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1020 		if (!stat_data)
1021 			return -ENOMEM;
1022 
1023 		stat_data->kvm = kvm;
1024 		stat_data->desc = pdesc;
1025 		stat_data->kind = KVM_STAT_VCPU;
1026 		kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1027 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1028 				    kvm->debugfs_dentry, stat_data,
1029 				    &stat_fops_per_vm);
1030 	}
1031 
1032 	ret = kvm_arch_create_vm_debugfs(kvm);
1033 	if (ret) {
1034 		kvm_destroy_vm_debugfs(kvm);
1035 		return i;
1036 	}
1037 
1038 	return 0;
1039 }
1040 
1041 /*
1042  * Called after the VM is otherwise initialized, but just before adding it to
1043  * the vm_list.
1044  */
1045 int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1046 {
1047 	return 0;
1048 }
1049 
1050 /*
1051  * Called just after removing the VM from the vm_list, but before doing any
1052  * other destruction.
1053  */
1054 void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1055 {
1056 }
1057 
1058 /*
1059  * Called after per-vm debugfs created.  When called kvm->debugfs_dentry should
1060  * be setup already, so we can create arch-specific debugfs entries under it.
1061  * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1062  * a per-arch destroy interface is not needed.
1063  */
1064 int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1065 {
1066 	return 0;
1067 }
1068 
1069 static struct kvm *kvm_create_vm(unsigned long type)
1070 {
1071 	struct kvm *kvm = kvm_arch_alloc_vm();
1072 	struct kvm_memslots *slots;
1073 	int r = -ENOMEM;
1074 	int i, j;
1075 
1076 	if (!kvm)
1077 		return ERR_PTR(-ENOMEM);
1078 
1079 	KVM_MMU_LOCK_INIT(kvm);
1080 	mmgrab(current->mm);
1081 	kvm->mm = current->mm;
1082 	kvm_eventfd_init(kvm);
1083 	mutex_init(&kvm->lock);
1084 	mutex_init(&kvm->irq_lock);
1085 	mutex_init(&kvm->slots_lock);
1086 	mutex_init(&kvm->slots_arch_lock);
1087 	spin_lock_init(&kvm->mn_invalidate_lock);
1088 	rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1089 	xa_init(&kvm->vcpu_array);
1090 
1091 	INIT_LIST_HEAD(&kvm->gpc_list);
1092 	spin_lock_init(&kvm->gpc_lock);
1093 
1094 	INIT_LIST_HEAD(&kvm->devices);
1095 	kvm->max_vcpus = KVM_MAX_VCPUS;
1096 
1097 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1098 
1099 	/*
1100 	 * Force subsequent debugfs file creations to fail if the VM directory
1101 	 * is not created (by kvm_create_vm_debugfs()).
1102 	 */
1103 	kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1104 
1105 	if (init_srcu_struct(&kvm->srcu))
1106 		goto out_err_no_srcu;
1107 	if (init_srcu_struct(&kvm->irq_srcu))
1108 		goto out_err_no_irq_srcu;
1109 
1110 	refcount_set(&kvm->users_count, 1);
1111 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1112 		for (j = 0; j < 2; j++) {
1113 			slots = &kvm->__memslots[i][j];
1114 
1115 			atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1116 			slots->hva_tree = RB_ROOT_CACHED;
1117 			slots->gfn_tree = RB_ROOT;
1118 			hash_init(slots->id_hash);
1119 			slots->node_idx = j;
1120 
1121 			/* Generations must be different for each address space. */
1122 			slots->generation = i;
1123 		}
1124 
1125 		rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1126 	}
1127 
1128 	for (i = 0; i < KVM_NR_BUSES; i++) {
1129 		rcu_assign_pointer(kvm->buses[i],
1130 			kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1131 		if (!kvm->buses[i])
1132 			goto out_err_no_arch_destroy_vm;
1133 	}
1134 
1135 	kvm->max_halt_poll_ns = halt_poll_ns;
1136 
1137 	r = kvm_arch_init_vm(kvm, type);
1138 	if (r)
1139 		goto out_err_no_arch_destroy_vm;
1140 
1141 	r = hardware_enable_all();
1142 	if (r)
1143 		goto out_err_no_disable;
1144 
1145 #ifdef CONFIG_HAVE_KVM_IRQFD
1146 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1147 #endif
1148 
1149 	r = kvm_init_mmu_notifier(kvm);
1150 	if (r)
1151 		goto out_err_no_mmu_notifier;
1152 
1153 	r = kvm_arch_post_init_vm(kvm);
1154 	if (r)
1155 		goto out_err;
1156 
1157 	mutex_lock(&kvm_lock);
1158 	list_add(&kvm->vm_list, &vm_list);
1159 	mutex_unlock(&kvm_lock);
1160 
1161 	preempt_notifier_inc();
1162 	kvm_init_pm_notifier(kvm);
1163 
1164 	/*
1165 	 * When the fd passed to this ioctl() is opened it pins the module,
1166 	 * but try_module_get() also prevents getting a reference if the module
1167 	 * is in MODULE_STATE_GOING (e.g. if someone ran "rmmod --wait").
1168 	 */
1169 	if (!try_module_get(kvm_chardev_ops.owner)) {
1170 		r = -ENODEV;
1171 		goto out_err;
1172 	}
1173 
1174 	return kvm;
1175 
1176 out_err:
1177 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1178 	if (kvm->mmu_notifier.ops)
1179 		mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1180 #endif
1181 out_err_no_mmu_notifier:
1182 	hardware_disable_all();
1183 out_err_no_disable:
1184 	kvm_arch_destroy_vm(kvm);
1185 out_err_no_arch_destroy_vm:
1186 	WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1187 	for (i = 0; i < KVM_NR_BUSES; i++)
1188 		kfree(kvm_get_bus(kvm, i));
1189 	cleanup_srcu_struct(&kvm->irq_srcu);
1190 out_err_no_irq_srcu:
1191 	cleanup_srcu_struct(&kvm->srcu);
1192 out_err_no_srcu:
1193 	kvm_arch_free_vm(kvm);
1194 	mmdrop(current->mm);
1195 	return ERR_PTR(r);
1196 }
1197 
1198 static void kvm_destroy_devices(struct kvm *kvm)
1199 {
1200 	struct kvm_device *dev, *tmp;
1201 
1202 	/*
1203 	 * We do not need to take the kvm->lock here, because nobody else
1204 	 * has a reference to the struct kvm at this point and therefore
1205 	 * cannot access the devices list anyhow.
1206 	 */
1207 	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1208 		list_del(&dev->vm_node);
1209 		dev->ops->destroy(dev);
1210 	}
1211 }
1212 
1213 static void kvm_destroy_vm(struct kvm *kvm)
1214 {
1215 	int i;
1216 	struct mm_struct *mm = kvm->mm;
1217 
1218 	kvm_destroy_pm_notifier(kvm);
1219 	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1220 	kvm_destroy_vm_debugfs(kvm);
1221 	kvm_arch_sync_events(kvm);
1222 	mutex_lock(&kvm_lock);
1223 	list_del(&kvm->vm_list);
1224 	mutex_unlock(&kvm_lock);
1225 	kvm_arch_pre_destroy_vm(kvm);
1226 
1227 	kvm_free_irq_routing(kvm);
1228 	for (i = 0; i < KVM_NR_BUSES; i++) {
1229 		struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1230 
1231 		if (bus)
1232 			kvm_io_bus_destroy(bus);
1233 		kvm->buses[i] = NULL;
1234 	}
1235 	kvm_coalesced_mmio_free(kvm);
1236 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1237 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1238 	/*
1239 	 * At this point, pending calls to invalidate_range_start()
1240 	 * have completed but no more MMU notifiers will run, so
1241 	 * mn_active_invalidate_count may remain unbalanced.
1242 	 * No threads can be waiting in install_new_memslots as the
1243 	 * last reference on KVM has been dropped, but freeing
1244 	 * memslots would deadlock without this manual intervention.
1245 	 */
1246 	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1247 	kvm->mn_active_invalidate_count = 0;
1248 #else
1249 	kvm_flush_shadow_all(kvm);
1250 #endif
1251 	kvm_arch_destroy_vm(kvm);
1252 	kvm_destroy_devices(kvm);
1253 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1254 		kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1255 		kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1256 	}
1257 	cleanup_srcu_struct(&kvm->irq_srcu);
1258 	cleanup_srcu_struct(&kvm->srcu);
1259 	kvm_arch_free_vm(kvm);
1260 	preempt_notifier_dec();
1261 	hardware_disable_all();
1262 	mmdrop(mm);
1263 	module_put(kvm_chardev_ops.owner);
1264 }
1265 
1266 void kvm_get_kvm(struct kvm *kvm)
1267 {
1268 	refcount_inc(&kvm->users_count);
1269 }
1270 EXPORT_SYMBOL_GPL(kvm_get_kvm);
1271 
1272 /*
1273  * Make sure the vm is not during destruction, which is a safe version of
1274  * kvm_get_kvm().  Return true if kvm referenced successfully, false otherwise.
1275  */
1276 bool kvm_get_kvm_safe(struct kvm *kvm)
1277 {
1278 	return refcount_inc_not_zero(&kvm->users_count);
1279 }
1280 EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1281 
1282 void kvm_put_kvm(struct kvm *kvm)
1283 {
1284 	if (refcount_dec_and_test(&kvm->users_count))
1285 		kvm_destroy_vm(kvm);
1286 }
1287 EXPORT_SYMBOL_GPL(kvm_put_kvm);
1288 
1289 /*
1290  * Used to put a reference that was taken on behalf of an object associated
1291  * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1292  * of the new file descriptor fails and the reference cannot be transferred to
1293  * its final owner.  In such cases, the caller is still actively using @kvm and
1294  * will fail miserably if the refcount unexpectedly hits zero.
1295  */
1296 void kvm_put_kvm_no_destroy(struct kvm *kvm)
1297 {
1298 	WARN_ON(refcount_dec_and_test(&kvm->users_count));
1299 }
1300 EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1301 
1302 static int kvm_vm_release(struct inode *inode, struct file *filp)
1303 {
1304 	struct kvm *kvm = filp->private_data;
1305 
1306 	kvm_irqfd_release(kvm);
1307 
1308 	kvm_put_kvm(kvm);
1309 	return 0;
1310 }
1311 
1312 /*
1313  * Allocation size is twice as large as the actual dirty bitmap size.
1314  * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1315  */
1316 static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1317 {
1318 	unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1319 
1320 	memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1321 	if (!memslot->dirty_bitmap)
1322 		return -ENOMEM;
1323 
1324 	return 0;
1325 }
1326 
1327 static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1328 {
1329 	struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1330 	int node_idx_inactive = active->node_idx ^ 1;
1331 
1332 	return &kvm->__memslots[as_id][node_idx_inactive];
1333 }
1334 
1335 /*
1336  * Helper to get the address space ID when one of memslot pointers may be NULL.
1337  * This also serves as a sanity that at least one of the pointers is non-NULL,
1338  * and that their address space IDs don't diverge.
1339  */
1340 static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1341 				  struct kvm_memory_slot *b)
1342 {
1343 	if (WARN_ON_ONCE(!a && !b))
1344 		return 0;
1345 
1346 	if (!a)
1347 		return b->as_id;
1348 	if (!b)
1349 		return a->as_id;
1350 
1351 	WARN_ON_ONCE(a->as_id != b->as_id);
1352 	return a->as_id;
1353 }
1354 
1355 static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1356 				struct kvm_memory_slot *slot)
1357 {
1358 	struct rb_root *gfn_tree = &slots->gfn_tree;
1359 	struct rb_node **node, *parent;
1360 	int idx = slots->node_idx;
1361 
1362 	parent = NULL;
1363 	for (node = &gfn_tree->rb_node; *node; ) {
1364 		struct kvm_memory_slot *tmp;
1365 
1366 		tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1367 		parent = *node;
1368 		if (slot->base_gfn < tmp->base_gfn)
1369 			node = &(*node)->rb_left;
1370 		else if (slot->base_gfn > tmp->base_gfn)
1371 			node = &(*node)->rb_right;
1372 		else
1373 			BUG();
1374 	}
1375 
1376 	rb_link_node(&slot->gfn_node[idx], parent, node);
1377 	rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1378 }
1379 
1380 static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1381 			       struct kvm_memory_slot *slot)
1382 {
1383 	rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1384 }
1385 
1386 static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1387 				 struct kvm_memory_slot *old,
1388 				 struct kvm_memory_slot *new)
1389 {
1390 	int idx = slots->node_idx;
1391 
1392 	WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1393 
1394 	rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1395 			&slots->gfn_tree);
1396 }
1397 
1398 /*
1399  * Replace @old with @new in the inactive memslots.
1400  *
1401  * With NULL @old this simply adds @new.
1402  * With NULL @new this simply removes @old.
1403  *
1404  * If @new is non-NULL its hva_node[slots_idx] range has to be set
1405  * appropriately.
1406  */
1407 static void kvm_replace_memslot(struct kvm *kvm,
1408 				struct kvm_memory_slot *old,
1409 				struct kvm_memory_slot *new)
1410 {
1411 	int as_id = kvm_memslots_get_as_id(old, new);
1412 	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1413 	int idx = slots->node_idx;
1414 
1415 	if (old) {
1416 		hash_del(&old->id_node[idx]);
1417 		interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1418 
1419 		if ((long)old == atomic_long_read(&slots->last_used_slot))
1420 			atomic_long_set(&slots->last_used_slot, (long)new);
1421 
1422 		if (!new) {
1423 			kvm_erase_gfn_node(slots, old);
1424 			return;
1425 		}
1426 	}
1427 
1428 	/*
1429 	 * Initialize @new's hva range.  Do this even when replacing an @old
1430 	 * slot, kvm_copy_memslot() deliberately does not touch node data.
1431 	 */
1432 	new->hva_node[idx].start = new->userspace_addr;
1433 	new->hva_node[idx].last = new->userspace_addr +
1434 				  (new->npages << PAGE_SHIFT) - 1;
1435 
1436 	/*
1437 	 * (Re)Add the new memslot.  There is no O(1) interval_tree_replace(),
1438 	 * hva_node needs to be swapped with remove+insert even though hva can't
1439 	 * change when replacing an existing slot.
1440 	 */
1441 	hash_add(slots->id_hash, &new->id_node[idx], new->id);
1442 	interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1443 
1444 	/*
1445 	 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1446 	 * switch the node in the gfn tree instead of removing the old and
1447 	 * inserting the new as two separate operations. Replacement is a
1448 	 * single O(1) operation versus two O(log(n)) operations for
1449 	 * remove+insert.
1450 	 */
1451 	if (old && old->base_gfn == new->base_gfn) {
1452 		kvm_replace_gfn_node(slots, old, new);
1453 	} else {
1454 		if (old)
1455 			kvm_erase_gfn_node(slots, old);
1456 		kvm_insert_gfn_node(slots, new);
1457 	}
1458 }
1459 
1460 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
1461 {
1462 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1463 
1464 #ifdef __KVM_HAVE_READONLY_MEM
1465 	valid_flags |= KVM_MEM_READONLY;
1466 #endif
1467 
1468 	if (mem->flags & ~valid_flags)
1469 		return -EINVAL;
1470 
1471 	return 0;
1472 }
1473 
1474 static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1475 {
1476 	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1477 
1478 	/* Grab the generation from the activate memslots. */
1479 	u64 gen = __kvm_memslots(kvm, as_id)->generation;
1480 
1481 	WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1482 	slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1483 
1484 	/*
1485 	 * Do not store the new memslots while there are invalidations in
1486 	 * progress, otherwise the locking in invalidate_range_start and
1487 	 * invalidate_range_end will be unbalanced.
1488 	 */
1489 	spin_lock(&kvm->mn_invalidate_lock);
1490 	prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1491 	while (kvm->mn_active_invalidate_count) {
1492 		set_current_state(TASK_UNINTERRUPTIBLE);
1493 		spin_unlock(&kvm->mn_invalidate_lock);
1494 		schedule();
1495 		spin_lock(&kvm->mn_invalidate_lock);
1496 	}
1497 	finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1498 	rcu_assign_pointer(kvm->memslots[as_id], slots);
1499 	spin_unlock(&kvm->mn_invalidate_lock);
1500 
1501 	/*
1502 	 * Acquired in kvm_set_memslot. Must be released before synchronize
1503 	 * SRCU below in order to avoid deadlock with another thread
1504 	 * acquiring the slots_arch_lock in an srcu critical section.
1505 	 */
1506 	mutex_unlock(&kvm->slots_arch_lock);
1507 
1508 	synchronize_srcu_expedited(&kvm->srcu);
1509 
1510 	/*
1511 	 * Increment the new memslot generation a second time, dropping the
1512 	 * update in-progress flag and incrementing the generation based on
1513 	 * the number of address spaces.  This provides a unique and easily
1514 	 * identifiable generation number while the memslots are in flux.
1515 	 */
1516 	gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1517 
1518 	/*
1519 	 * Generations must be unique even across address spaces.  We do not need
1520 	 * a global counter for that, instead the generation space is evenly split
1521 	 * across address spaces.  For example, with two address spaces, address
1522 	 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1523 	 * use generations 1, 3, 5, ...
1524 	 */
1525 	gen += KVM_ADDRESS_SPACE_NUM;
1526 
1527 	kvm_arch_memslots_updated(kvm, gen);
1528 
1529 	slots->generation = gen;
1530 }
1531 
1532 static int kvm_prepare_memory_region(struct kvm *kvm,
1533 				     const struct kvm_memory_slot *old,
1534 				     struct kvm_memory_slot *new,
1535 				     enum kvm_mr_change change)
1536 {
1537 	int r;
1538 
1539 	/*
1540 	 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1541 	 * will be freed on "commit".  If logging is enabled in both old and
1542 	 * new, reuse the existing bitmap.  If logging is enabled only in the
1543 	 * new and KVM isn't using a ring buffer, allocate and initialize a
1544 	 * new bitmap.
1545 	 */
1546 	if (change != KVM_MR_DELETE) {
1547 		if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1548 			new->dirty_bitmap = NULL;
1549 		else if (old && old->dirty_bitmap)
1550 			new->dirty_bitmap = old->dirty_bitmap;
1551 		else if (!kvm->dirty_ring_size) {
1552 			r = kvm_alloc_dirty_bitmap(new);
1553 			if (r)
1554 				return r;
1555 
1556 			if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1557 				bitmap_set(new->dirty_bitmap, 0, new->npages);
1558 		}
1559 	}
1560 
1561 	r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1562 
1563 	/* Free the bitmap on failure if it was allocated above. */
1564 	if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1565 		kvm_destroy_dirty_bitmap(new);
1566 
1567 	return r;
1568 }
1569 
1570 static void kvm_commit_memory_region(struct kvm *kvm,
1571 				     struct kvm_memory_slot *old,
1572 				     const struct kvm_memory_slot *new,
1573 				     enum kvm_mr_change change)
1574 {
1575 	/*
1576 	 * Update the total number of memslot pages before calling the arch
1577 	 * hook so that architectures can consume the result directly.
1578 	 */
1579 	if (change == KVM_MR_DELETE)
1580 		kvm->nr_memslot_pages -= old->npages;
1581 	else if (change == KVM_MR_CREATE)
1582 		kvm->nr_memslot_pages += new->npages;
1583 
1584 	kvm_arch_commit_memory_region(kvm, old, new, change);
1585 
1586 	switch (change) {
1587 	case KVM_MR_CREATE:
1588 		/* Nothing more to do. */
1589 		break;
1590 	case KVM_MR_DELETE:
1591 		/* Free the old memslot and all its metadata. */
1592 		kvm_free_memslot(kvm, old);
1593 		break;
1594 	case KVM_MR_MOVE:
1595 	case KVM_MR_FLAGS_ONLY:
1596 		/*
1597 		 * Free the dirty bitmap as needed; the below check encompasses
1598 		 * both the flags and whether a ring buffer is being used)
1599 		 */
1600 		if (old->dirty_bitmap && !new->dirty_bitmap)
1601 			kvm_destroy_dirty_bitmap(old);
1602 
1603 		/*
1604 		 * The final quirk.  Free the detached, old slot, but only its
1605 		 * memory, not any metadata.  Metadata, including arch specific
1606 		 * data, may be reused by @new.
1607 		 */
1608 		kfree(old);
1609 		break;
1610 	default:
1611 		BUG();
1612 	}
1613 }
1614 
1615 /*
1616  * Activate @new, which must be installed in the inactive slots by the caller,
1617  * by swapping the active slots and then propagating @new to @old once @old is
1618  * unreachable and can be safely modified.
1619  *
1620  * With NULL @old this simply adds @new to @active (while swapping the sets).
1621  * With NULL @new this simply removes @old from @active and frees it
1622  * (while also swapping the sets).
1623  */
1624 static void kvm_activate_memslot(struct kvm *kvm,
1625 				 struct kvm_memory_slot *old,
1626 				 struct kvm_memory_slot *new)
1627 {
1628 	int as_id = kvm_memslots_get_as_id(old, new);
1629 
1630 	kvm_swap_active_memslots(kvm, as_id);
1631 
1632 	/* Propagate the new memslot to the now inactive memslots. */
1633 	kvm_replace_memslot(kvm, old, new);
1634 }
1635 
1636 static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1637 			     const struct kvm_memory_slot *src)
1638 {
1639 	dest->base_gfn = src->base_gfn;
1640 	dest->npages = src->npages;
1641 	dest->dirty_bitmap = src->dirty_bitmap;
1642 	dest->arch = src->arch;
1643 	dest->userspace_addr = src->userspace_addr;
1644 	dest->flags = src->flags;
1645 	dest->id = src->id;
1646 	dest->as_id = src->as_id;
1647 }
1648 
1649 static void kvm_invalidate_memslot(struct kvm *kvm,
1650 				   struct kvm_memory_slot *old,
1651 				   struct kvm_memory_slot *invalid_slot)
1652 {
1653 	/*
1654 	 * Mark the current slot INVALID.  As with all memslot modifications,
1655 	 * this must be done on an unreachable slot to avoid modifying the
1656 	 * current slot in the active tree.
1657 	 */
1658 	kvm_copy_memslot(invalid_slot, old);
1659 	invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1660 	kvm_replace_memslot(kvm, old, invalid_slot);
1661 
1662 	/*
1663 	 * Activate the slot that is now marked INVALID, but don't propagate
1664 	 * the slot to the now inactive slots. The slot is either going to be
1665 	 * deleted or recreated as a new slot.
1666 	 */
1667 	kvm_swap_active_memslots(kvm, old->as_id);
1668 
1669 	/*
1670 	 * From this point no new shadow pages pointing to a deleted, or moved,
1671 	 * memslot will be created.  Validation of sp->gfn happens in:
1672 	 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1673 	 *	- kvm_is_visible_gfn (mmu_check_root)
1674 	 */
1675 	kvm_arch_flush_shadow_memslot(kvm, old);
1676 	kvm_arch_guest_memory_reclaimed(kvm);
1677 
1678 	/* Was released by kvm_swap_active_memslots, reacquire. */
1679 	mutex_lock(&kvm->slots_arch_lock);
1680 
1681 	/*
1682 	 * Copy the arch-specific field of the newly-installed slot back to the
1683 	 * old slot as the arch data could have changed between releasing
1684 	 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1685 	 * above.  Writers are required to retrieve memslots *after* acquiring
1686 	 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1687 	 */
1688 	old->arch = invalid_slot->arch;
1689 }
1690 
1691 static void kvm_create_memslot(struct kvm *kvm,
1692 			       struct kvm_memory_slot *new)
1693 {
1694 	/* Add the new memslot to the inactive set and activate. */
1695 	kvm_replace_memslot(kvm, NULL, new);
1696 	kvm_activate_memslot(kvm, NULL, new);
1697 }
1698 
1699 static void kvm_delete_memslot(struct kvm *kvm,
1700 			       struct kvm_memory_slot *old,
1701 			       struct kvm_memory_slot *invalid_slot)
1702 {
1703 	/*
1704 	 * Remove the old memslot (in the inactive memslots) by passing NULL as
1705 	 * the "new" slot, and for the invalid version in the active slots.
1706 	 */
1707 	kvm_replace_memslot(kvm, old, NULL);
1708 	kvm_activate_memslot(kvm, invalid_slot, NULL);
1709 }
1710 
1711 static void kvm_move_memslot(struct kvm *kvm,
1712 			     struct kvm_memory_slot *old,
1713 			     struct kvm_memory_slot *new,
1714 			     struct kvm_memory_slot *invalid_slot)
1715 {
1716 	/*
1717 	 * Replace the old memslot in the inactive slots, and then swap slots
1718 	 * and replace the current INVALID with the new as well.
1719 	 */
1720 	kvm_replace_memslot(kvm, old, new);
1721 	kvm_activate_memslot(kvm, invalid_slot, new);
1722 }
1723 
1724 static void kvm_update_flags_memslot(struct kvm *kvm,
1725 				     struct kvm_memory_slot *old,
1726 				     struct kvm_memory_slot *new)
1727 {
1728 	/*
1729 	 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1730 	 * an intermediate step. Instead, the old memslot is simply replaced
1731 	 * with a new, updated copy in both memslot sets.
1732 	 */
1733 	kvm_replace_memslot(kvm, old, new);
1734 	kvm_activate_memslot(kvm, old, new);
1735 }
1736 
1737 static int kvm_set_memslot(struct kvm *kvm,
1738 			   struct kvm_memory_slot *old,
1739 			   struct kvm_memory_slot *new,
1740 			   enum kvm_mr_change change)
1741 {
1742 	struct kvm_memory_slot *invalid_slot;
1743 	int r;
1744 
1745 	/*
1746 	 * Released in kvm_swap_active_memslots.
1747 	 *
1748 	 * Must be held from before the current memslots are copied until
1749 	 * after the new memslots are installed with rcu_assign_pointer,
1750 	 * then released before the synchronize srcu in kvm_swap_active_memslots.
1751 	 *
1752 	 * When modifying memslots outside of the slots_lock, must be held
1753 	 * before reading the pointer to the current memslots until after all
1754 	 * changes to those memslots are complete.
1755 	 *
1756 	 * These rules ensure that installing new memslots does not lose
1757 	 * changes made to the previous memslots.
1758 	 */
1759 	mutex_lock(&kvm->slots_arch_lock);
1760 
1761 	/*
1762 	 * Invalidate the old slot if it's being deleted or moved.  This is
1763 	 * done prior to actually deleting/moving the memslot to allow vCPUs to
1764 	 * continue running by ensuring there are no mappings or shadow pages
1765 	 * for the memslot when it is deleted/moved.  Without pre-invalidation
1766 	 * (and without a lock), a window would exist between effecting the
1767 	 * delete/move and committing the changes in arch code where KVM or a
1768 	 * guest could access a non-existent memslot.
1769 	 *
1770 	 * Modifications are done on a temporary, unreachable slot.  The old
1771 	 * slot needs to be preserved in case a later step fails and the
1772 	 * invalidation needs to be reverted.
1773 	 */
1774 	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1775 		invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1776 		if (!invalid_slot) {
1777 			mutex_unlock(&kvm->slots_arch_lock);
1778 			return -ENOMEM;
1779 		}
1780 		kvm_invalidate_memslot(kvm, old, invalid_slot);
1781 	}
1782 
1783 	r = kvm_prepare_memory_region(kvm, old, new, change);
1784 	if (r) {
1785 		/*
1786 		 * For DELETE/MOVE, revert the above INVALID change.  No
1787 		 * modifications required since the original slot was preserved
1788 		 * in the inactive slots.  Changing the active memslots also
1789 		 * release slots_arch_lock.
1790 		 */
1791 		if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1792 			kvm_activate_memslot(kvm, invalid_slot, old);
1793 			kfree(invalid_slot);
1794 		} else {
1795 			mutex_unlock(&kvm->slots_arch_lock);
1796 		}
1797 		return r;
1798 	}
1799 
1800 	/*
1801 	 * For DELETE and MOVE, the working slot is now active as the INVALID
1802 	 * version of the old slot.  MOVE is particularly special as it reuses
1803 	 * the old slot and returns a copy of the old slot (in working_slot).
1804 	 * For CREATE, there is no old slot.  For DELETE and FLAGS_ONLY, the
1805 	 * old slot is detached but otherwise preserved.
1806 	 */
1807 	if (change == KVM_MR_CREATE)
1808 		kvm_create_memslot(kvm, new);
1809 	else if (change == KVM_MR_DELETE)
1810 		kvm_delete_memslot(kvm, old, invalid_slot);
1811 	else if (change == KVM_MR_MOVE)
1812 		kvm_move_memslot(kvm, old, new, invalid_slot);
1813 	else if (change == KVM_MR_FLAGS_ONLY)
1814 		kvm_update_flags_memslot(kvm, old, new);
1815 	else
1816 		BUG();
1817 
1818 	/* Free the temporary INVALID slot used for DELETE and MOVE. */
1819 	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1820 		kfree(invalid_slot);
1821 
1822 	/*
1823 	 * No need to refresh new->arch, changes after dropping slots_arch_lock
1824 	 * will directly hit the final, active memslot.  Architectures are
1825 	 * responsible for knowing that new->arch may be stale.
1826 	 */
1827 	kvm_commit_memory_region(kvm, old, new, change);
1828 
1829 	return 0;
1830 }
1831 
1832 static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1833 				      gfn_t start, gfn_t end)
1834 {
1835 	struct kvm_memslot_iter iter;
1836 
1837 	kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1838 		if (iter.slot->id != id)
1839 			return true;
1840 	}
1841 
1842 	return false;
1843 }
1844 
1845 /*
1846  * Allocate some memory and give it an address in the guest physical address
1847  * space.
1848  *
1849  * Discontiguous memory is allowed, mostly for framebuffers.
1850  *
1851  * Must be called holding kvm->slots_lock for write.
1852  */
1853 int __kvm_set_memory_region(struct kvm *kvm,
1854 			    const struct kvm_userspace_memory_region *mem)
1855 {
1856 	struct kvm_memory_slot *old, *new;
1857 	struct kvm_memslots *slots;
1858 	enum kvm_mr_change change;
1859 	unsigned long npages;
1860 	gfn_t base_gfn;
1861 	int as_id, id;
1862 	int r;
1863 
1864 	r = check_memory_region_flags(mem);
1865 	if (r)
1866 		return r;
1867 
1868 	as_id = mem->slot >> 16;
1869 	id = (u16)mem->slot;
1870 
1871 	/* General sanity checks */
1872 	if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1873 	    (mem->memory_size != (unsigned long)mem->memory_size))
1874 		return -EINVAL;
1875 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1876 		return -EINVAL;
1877 	/* We can read the guest memory with __xxx_user() later on. */
1878 	if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1879 	    (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1880 	     !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1881 			mem->memory_size))
1882 		return -EINVAL;
1883 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
1884 		return -EINVAL;
1885 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1886 		return -EINVAL;
1887 	if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1888 		return -EINVAL;
1889 
1890 	slots = __kvm_memslots(kvm, as_id);
1891 
1892 	/*
1893 	 * Note, the old memslot (and the pointer itself!) may be invalidated
1894 	 * and/or destroyed by kvm_set_memslot().
1895 	 */
1896 	old = id_to_memslot(slots, id);
1897 
1898 	if (!mem->memory_size) {
1899 		if (!old || !old->npages)
1900 			return -EINVAL;
1901 
1902 		if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1903 			return -EIO;
1904 
1905 		return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1906 	}
1907 
1908 	base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1909 	npages = (mem->memory_size >> PAGE_SHIFT);
1910 
1911 	if (!old || !old->npages) {
1912 		change = KVM_MR_CREATE;
1913 
1914 		/*
1915 		 * To simplify KVM internals, the total number of pages across
1916 		 * all memslots must fit in an unsigned long.
1917 		 */
1918 		if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1919 			return -EINVAL;
1920 	} else { /* Modify an existing slot. */
1921 		if ((mem->userspace_addr != old->userspace_addr) ||
1922 		    (npages != old->npages) ||
1923 		    ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1924 			return -EINVAL;
1925 
1926 		if (base_gfn != old->base_gfn)
1927 			change = KVM_MR_MOVE;
1928 		else if (mem->flags != old->flags)
1929 			change = KVM_MR_FLAGS_ONLY;
1930 		else /* Nothing to change. */
1931 			return 0;
1932 	}
1933 
1934 	if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
1935 	    kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
1936 		return -EEXIST;
1937 
1938 	/* Allocate a slot that will persist in the memslot. */
1939 	new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
1940 	if (!new)
1941 		return -ENOMEM;
1942 
1943 	new->as_id = as_id;
1944 	new->id = id;
1945 	new->base_gfn = base_gfn;
1946 	new->npages = npages;
1947 	new->flags = mem->flags;
1948 	new->userspace_addr = mem->userspace_addr;
1949 
1950 	r = kvm_set_memslot(kvm, old, new, change);
1951 	if (r)
1952 		kfree(new);
1953 	return r;
1954 }
1955 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
1956 
1957 int kvm_set_memory_region(struct kvm *kvm,
1958 			  const struct kvm_userspace_memory_region *mem)
1959 {
1960 	int r;
1961 
1962 	mutex_lock(&kvm->slots_lock);
1963 	r = __kvm_set_memory_region(kvm, mem);
1964 	mutex_unlock(&kvm->slots_lock);
1965 	return r;
1966 }
1967 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1968 
1969 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1970 					  struct kvm_userspace_memory_region *mem)
1971 {
1972 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1973 		return -EINVAL;
1974 
1975 	return kvm_set_memory_region(kvm, mem);
1976 }
1977 
1978 #ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1979 /**
1980  * kvm_get_dirty_log - get a snapshot of dirty pages
1981  * @kvm:	pointer to kvm instance
1982  * @log:	slot id and address to which we copy the log
1983  * @is_dirty:	set to '1' if any dirty pages were found
1984  * @memslot:	set to the associated memslot, always valid on success
1985  */
1986 int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
1987 		      int *is_dirty, struct kvm_memory_slot **memslot)
1988 {
1989 	struct kvm_memslots *slots;
1990 	int i, as_id, id;
1991 	unsigned long n;
1992 	unsigned long any = 0;
1993 
1994 	/* Dirty ring tracking is exclusive to dirty log tracking */
1995 	if (kvm->dirty_ring_size)
1996 		return -ENXIO;
1997 
1998 	*memslot = NULL;
1999 	*is_dirty = 0;
2000 
2001 	as_id = log->slot >> 16;
2002 	id = (u16)log->slot;
2003 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2004 		return -EINVAL;
2005 
2006 	slots = __kvm_memslots(kvm, as_id);
2007 	*memslot = id_to_memslot(slots, id);
2008 	if (!(*memslot) || !(*memslot)->dirty_bitmap)
2009 		return -ENOENT;
2010 
2011 	kvm_arch_sync_dirty_log(kvm, *memslot);
2012 
2013 	n = kvm_dirty_bitmap_bytes(*memslot);
2014 
2015 	for (i = 0; !any && i < n/sizeof(long); ++i)
2016 		any = (*memslot)->dirty_bitmap[i];
2017 
2018 	if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2019 		return -EFAULT;
2020 
2021 	if (any)
2022 		*is_dirty = 1;
2023 	return 0;
2024 }
2025 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2026 
2027 #else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2028 /**
2029  * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2030  *	and reenable dirty page tracking for the corresponding pages.
2031  * @kvm:	pointer to kvm instance
2032  * @log:	slot id and address to which we copy the log
2033  *
2034  * We need to keep it in mind that VCPU threads can write to the bitmap
2035  * concurrently. So, to avoid losing track of dirty pages we keep the
2036  * following order:
2037  *
2038  *    1. Take a snapshot of the bit and clear it if needed.
2039  *    2. Write protect the corresponding page.
2040  *    3. Copy the snapshot to the userspace.
2041  *    4. Upon return caller flushes TLB's if needed.
2042  *
2043  * Between 2 and 4, the guest may write to the page using the remaining TLB
2044  * entry.  This is not a problem because the page is reported dirty using
2045  * the snapshot taken before and step 4 ensures that writes done after
2046  * exiting to userspace will be logged for the next call.
2047  *
2048  */
2049 static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2050 {
2051 	struct kvm_memslots *slots;
2052 	struct kvm_memory_slot *memslot;
2053 	int i, as_id, id;
2054 	unsigned long n;
2055 	unsigned long *dirty_bitmap;
2056 	unsigned long *dirty_bitmap_buffer;
2057 	bool flush;
2058 
2059 	/* Dirty ring tracking is exclusive to dirty log tracking */
2060 	if (kvm->dirty_ring_size)
2061 		return -ENXIO;
2062 
2063 	as_id = log->slot >> 16;
2064 	id = (u16)log->slot;
2065 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2066 		return -EINVAL;
2067 
2068 	slots = __kvm_memslots(kvm, as_id);
2069 	memslot = id_to_memslot(slots, id);
2070 	if (!memslot || !memslot->dirty_bitmap)
2071 		return -ENOENT;
2072 
2073 	dirty_bitmap = memslot->dirty_bitmap;
2074 
2075 	kvm_arch_sync_dirty_log(kvm, memslot);
2076 
2077 	n = kvm_dirty_bitmap_bytes(memslot);
2078 	flush = false;
2079 	if (kvm->manual_dirty_log_protect) {
2080 		/*
2081 		 * Unlike kvm_get_dirty_log, we always return false in *flush,
2082 		 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
2083 		 * is some code duplication between this function and
2084 		 * kvm_get_dirty_log, but hopefully all architecture
2085 		 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2086 		 * can be eliminated.
2087 		 */
2088 		dirty_bitmap_buffer = dirty_bitmap;
2089 	} else {
2090 		dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2091 		memset(dirty_bitmap_buffer, 0, n);
2092 
2093 		KVM_MMU_LOCK(kvm);
2094 		for (i = 0; i < n / sizeof(long); i++) {
2095 			unsigned long mask;
2096 			gfn_t offset;
2097 
2098 			if (!dirty_bitmap[i])
2099 				continue;
2100 
2101 			flush = true;
2102 			mask = xchg(&dirty_bitmap[i], 0);
2103 			dirty_bitmap_buffer[i] = mask;
2104 
2105 			offset = i * BITS_PER_LONG;
2106 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2107 								offset, mask);
2108 		}
2109 		KVM_MMU_UNLOCK(kvm);
2110 	}
2111 
2112 	if (flush)
2113 		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2114 
2115 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2116 		return -EFAULT;
2117 	return 0;
2118 }
2119 
2120 
2121 /**
2122  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2123  * @kvm: kvm instance
2124  * @log: slot id and address to which we copy the log
2125  *
2126  * Steps 1-4 below provide general overview of dirty page logging. See
2127  * kvm_get_dirty_log_protect() function description for additional details.
2128  *
2129  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2130  * always flush the TLB (step 4) even if previous step failed  and the dirty
2131  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2132  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2133  * writes will be marked dirty for next log read.
2134  *
2135  *   1. Take a snapshot of the bit and clear it if needed.
2136  *   2. Write protect the corresponding page.
2137  *   3. Copy the snapshot to the userspace.
2138  *   4. Flush TLB's if needed.
2139  */
2140 static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2141 				      struct kvm_dirty_log *log)
2142 {
2143 	int r;
2144 
2145 	mutex_lock(&kvm->slots_lock);
2146 
2147 	r = kvm_get_dirty_log_protect(kvm, log);
2148 
2149 	mutex_unlock(&kvm->slots_lock);
2150 	return r;
2151 }
2152 
2153 /**
2154  * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2155  *	and reenable dirty page tracking for the corresponding pages.
2156  * @kvm:	pointer to kvm instance
2157  * @log:	slot id and address from which to fetch the bitmap of dirty pages
2158  */
2159 static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2160 				       struct kvm_clear_dirty_log *log)
2161 {
2162 	struct kvm_memslots *slots;
2163 	struct kvm_memory_slot *memslot;
2164 	int as_id, id;
2165 	gfn_t offset;
2166 	unsigned long i, n;
2167 	unsigned long *dirty_bitmap;
2168 	unsigned long *dirty_bitmap_buffer;
2169 	bool flush;
2170 
2171 	/* Dirty ring tracking is exclusive to dirty log tracking */
2172 	if (kvm->dirty_ring_size)
2173 		return -ENXIO;
2174 
2175 	as_id = log->slot >> 16;
2176 	id = (u16)log->slot;
2177 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2178 		return -EINVAL;
2179 
2180 	if (log->first_page & 63)
2181 		return -EINVAL;
2182 
2183 	slots = __kvm_memslots(kvm, as_id);
2184 	memslot = id_to_memslot(slots, id);
2185 	if (!memslot || !memslot->dirty_bitmap)
2186 		return -ENOENT;
2187 
2188 	dirty_bitmap = memslot->dirty_bitmap;
2189 
2190 	n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2191 
2192 	if (log->first_page > memslot->npages ||
2193 	    log->num_pages > memslot->npages - log->first_page ||
2194 	    (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2195 	    return -EINVAL;
2196 
2197 	kvm_arch_sync_dirty_log(kvm, memslot);
2198 
2199 	flush = false;
2200 	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2201 	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2202 		return -EFAULT;
2203 
2204 	KVM_MMU_LOCK(kvm);
2205 	for (offset = log->first_page, i = offset / BITS_PER_LONG,
2206 		 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2207 	     i++, offset += BITS_PER_LONG) {
2208 		unsigned long mask = *dirty_bitmap_buffer++;
2209 		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2210 		if (!mask)
2211 			continue;
2212 
2213 		mask &= atomic_long_fetch_andnot(mask, p);
2214 
2215 		/*
2216 		 * mask contains the bits that really have been cleared.  This
2217 		 * never includes any bits beyond the length of the memslot (if
2218 		 * the length is not aligned to 64 pages), therefore it is not
2219 		 * a problem if userspace sets them in log->dirty_bitmap.
2220 		*/
2221 		if (mask) {
2222 			flush = true;
2223 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2224 								offset, mask);
2225 		}
2226 	}
2227 	KVM_MMU_UNLOCK(kvm);
2228 
2229 	if (flush)
2230 		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2231 
2232 	return 0;
2233 }
2234 
2235 static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2236 					struct kvm_clear_dirty_log *log)
2237 {
2238 	int r;
2239 
2240 	mutex_lock(&kvm->slots_lock);
2241 
2242 	r = kvm_clear_dirty_log_protect(kvm, log);
2243 
2244 	mutex_unlock(&kvm->slots_lock);
2245 	return r;
2246 }
2247 #endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2248 
2249 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2250 {
2251 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2252 }
2253 EXPORT_SYMBOL_GPL(gfn_to_memslot);
2254 
2255 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2256 {
2257 	struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2258 	u64 gen = slots->generation;
2259 	struct kvm_memory_slot *slot;
2260 
2261 	/*
2262 	 * This also protects against using a memslot from a different address space,
2263 	 * since different address spaces have different generation numbers.
2264 	 */
2265 	if (unlikely(gen != vcpu->last_used_slot_gen)) {
2266 		vcpu->last_used_slot = NULL;
2267 		vcpu->last_used_slot_gen = gen;
2268 	}
2269 
2270 	slot = try_get_memslot(vcpu->last_used_slot, gfn);
2271 	if (slot)
2272 		return slot;
2273 
2274 	/*
2275 	 * Fall back to searching all memslots. We purposely use
2276 	 * search_memslots() instead of __gfn_to_memslot() to avoid
2277 	 * thrashing the VM-wide last_used_slot in kvm_memslots.
2278 	 */
2279 	slot = search_memslots(slots, gfn, false);
2280 	if (slot) {
2281 		vcpu->last_used_slot = slot;
2282 		return slot;
2283 	}
2284 
2285 	return NULL;
2286 }
2287 
2288 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2289 {
2290 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2291 
2292 	return kvm_is_visible_memslot(memslot);
2293 }
2294 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2295 
2296 bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2297 {
2298 	struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2299 
2300 	return kvm_is_visible_memslot(memslot);
2301 }
2302 EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2303 
2304 unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2305 {
2306 	struct vm_area_struct *vma;
2307 	unsigned long addr, size;
2308 
2309 	size = PAGE_SIZE;
2310 
2311 	addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2312 	if (kvm_is_error_hva(addr))
2313 		return PAGE_SIZE;
2314 
2315 	mmap_read_lock(current->mm);
2316 	vma = find_vma(current->mm, addr);
2317 	if (!vma)
2318 		goto out;
2319 
2320 	size = vma_kernel_pagesize(vma);
2321 
2322 out:
2323 	mmap_read_unlock(current->mm);
2324 
2325 	return size;
2326 }
2327 
2328 static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2329 {
2330 	return slot->flags & KVM_MEM_READONLY;
2331 }
2332 
2333 static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2334 				       gfn_t *nr_pages, bool write)
2335 {
2336 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2337 		return KVM_HVA_ERR_BAD;
2338 
2339 	if (memslot_is_readonly(slot) && write)
2340 		return KVM_HVA_ERR_RO_BAD;
2341 
2342 	if (nr_pages)
2343 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
2344 
2345 	return __gfn_to_hva_memslot(slot, gfn);
2346 }
2347 
2348 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2349 				     gfn_t *nr_pages)
2350 {
2351 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2352 }
2353 
2354 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2355 					gfn_t gfn)
2356 {
2357 	return gfn_to_hva_many(slot, gfn, NULL);
2358 }
2359 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2360 
2361 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2362 {
2363 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2364 }
2365 EXPORT_SYMBOL_GPL(gfn_to_hva);
2366 
2367 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2368 {
2369 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2370 }
2371 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2372 
2373 /*
2374  * Return the hva of a @gfn and the R/W attribute if possible.
2375  *
2376  * @slot: the kvm_memory_slot which contains @gfn
2377  * @gfn: the gfn to be translated
2378  * @writable: used to return the read/write attribute of the @slot if the hva
2379  * is valid and @writable is not NULL
2380  */
2381 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2382 				      gfn_t gfn, bool *writable)
2383 {
2384 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2385 
2386 	if (!kvm_is_error_hva(hva) && writable)
2387 		*writable = !memslot_is_readonly(slot);
2388 
2389 	return hva;
2390 }
2391 
2392 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2393 {
2394 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2395 
2396 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2397 }
2398 
2399 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2400 {
2401 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2402 
2403 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2404 }
2405 
2406 static inline int check_user_page_hwpoison(unsigned long addr)
2407 {
2408 	int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2409 
2410 	rc = get_user_pages(addr, 1, flags, NULL, NULL);
2411 	return rc == -EHWPOISON;
2412 }
2413 
2414 /*
2415  * The fast path to get the writable pfn which will be stored in @pfn,
2416  * true indicates success, otherwise false is returned.  It's also the
2417  * only part that runs if we can in atomic context.
2418  */
2419 static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2420 			    bool *writable, kvm_pfn_t *pfn)
2421 {
2422 	struct page *page[1];
2423 
2424 	/*
2425 	 * Fast pin a writable pfn only if it is a write fault request
2426 	 * or the caller allows to map a writable pfn for a read fault
2427 	 * request.
2428 	 */
2429 	if (!(write_fault || writable))
2430 		return false;
2431 
2432 	if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2433 		*pfn = page_to_pfn(page[0]);
2434 
2435 		if (writable)
2436 			*writable = true;
2437 		return true;
2438 	}
2439 
2440 	return false;
2441 }
2442 
2443 /*
2444  * The slow path to get the pfn of the specified host virtual address,
2445  * 1 indicates success, -errno is returned if error is detected.
2446  */
2447 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2448 			   bool *writable, kvm_pfn_t *pfn)
2449 {
2450 	unsigned int flags = FOLL_HWPOISON;
2451 	struct page *page;
2452 	int npages = 0;
2453 
2454 	might_sleep();
2455 
2456 	if (writable)
2457 		*writable = write_fault;
2458 
2459 	if (write_fault)
2460 		flags |= FOLL_WRITE;
2461 	if (async)
2462 		flags |= FOLL_NOWAIT;
2463 
2464 	npages = get_user_pages_unlocked(addr, 1, &page, flags);
2465 	if (npages != 1)
2466 		return npages;
2467 
2468 	/* map read fault as writable if possible */
2469 	if (unlikely(!write_fault) && writable) {
2470 		struct page *wpage;
2471 
2472 		if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2473 			*writable = true;
2474 			put_page(page);
2475 			page = wpage;
2476 		}
2477 	}
2478 	*pfn = page_to_pfn(page);
2479 	return npages;
2480 }
2481 
2482 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2483 {
2484 	if (unlikely(!(vma->vm_flags & VM_READ)))
2485 		return false;
2486 
2487 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2488 		return false;
2489 
2490 	return true;
2491 }
2492 
2493 static int kvm_try_get_pfn(kvm_pfn_t pfn)
2494 {
2495 	if (kvm_is_reserved_pfn(pfn))
2496 		return 1;
2497 	return get_page_unless_zero(pfn_to_page(pfn));
2498 }
2499 
2500 static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2501 			       unsigned long addr, bool write_fault,
2502 			       bool *writable, kvm_pfn_t *p_pfn)
2503 {
2504 	kvm_pfn_t pfn;
2505 	pte_t *ptep;
2506 	spinlock_t *ptl;
2507 	int r;
2508 
2509 	r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2510 	if (r) {
2511 		/*
2512 		 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2513 		 * not call the fault handler, so do it here.
2514 		 */
2515 		bool unlocked = false;
2516 		r = fixup_user_fault(current->mm, addr,
2517 				     (write_fault ? FAULT_FLAG_WRITE : 0),
2518 				     &unlocked);
2519 		if (unlocked)
2520 			return -EAGAIN;
2521 		if (r)
2522 			return r;
2523 
2524 		r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2525 		if (r)
2526 			return r;
2527 	}
2528 
2529 	if (write_fault && !pte_write(*ptep)) {
2530 		pfn = KVM_PFN_ERR_RO_FAULT;
2531 		goto out;
2532 	}
2533 
2534 	if (writable)
2535 		*writable = pte_write(*ptep);
2536 	pfn = pte_pfn(*ptep);
2537 
2538 	/*
2539 	 * Get a reference here because callers of *hva_to_pfn* and
2540 	 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2541 	 * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
2542 	 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2543 	 * simply do nothing for reserved pfns.
2544 	 *
2545 	 * Whoever called remap_pfn_range is also going to call e.g.
2546 	 * unmap_mapping_range before the underlying pages are freed,
2547 	 * causing a call to our MMU notifier.
2548 	 *
2549 	 * Certain IO or PFNMAP mappings can be backed with valid
2550 	 * struct pages, but be allocated without refcounting e.g.,
2551 	 * tail pages of non-compound higher order allocations, which
2552 	 * would then underflow the refcount when the caller does the
2553 	 * required put_page. Don't allow those pages here.
2554 	 */
2555 	if (!kvm_try_get_pfn(pfn))
2556 		r = -EFAULT;
2557 
2558 out:
2559 	pte_unmap_unlock(ptep, ptl);
2560 	*p_pfn = pfn;
2561 
2562 	return r;
2563 }
2564 
2565 /*
2566  * Pin guest page in memory and return its pfn.
2567  * @addr: host virtual address which maps memory to the guest
2568  * @atomic: whether this function can sleep
2569  * @async: whether this function need to wait IO complete if the
2570  *         host page is not in the memory
2571  * @write_fault: whether we should get a writable host page
2572  * @writable: whether it allows to map a writable host page for !@write_fault
2573  *
2574  * The function will map a writable host page for these two cases:
2575  * 1): @write_fault = true
2576  * 2): @write_fault = false && @writable, @writable will tell the caller
2577  *     whether the mapping is writable.
2578  */
2579 kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
2580 		     bool write_fault, bool *writable)
2581 {
2582 	struct vm_area_struct *vma;
2583 	kvm_pfn_t pfn = 0;
2584 	int npages, r;
2585 
2586 	/* we can do it either atomically or asynchronously, not both */
2587 	BUG_ON(atomic && async);
2588 
2589 	if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2590 		return pfn;
2591 
2592 	if (atomic)
2593 		return KVM_PFN_ERR_FAULT;
2594 
2595 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
2596 	if (npages == 1)
2597 		return pfn;
2598 
2599 	mmap_read_lock(current->mm);
2600 	if (npages == -EHWPOISON ||
2601 	      (!async && check_user_page_hwpoison(addr))) {
2602 		pfn = KVM_PFN_ERR_HWPOISON;
2603 		goto exit;
2604 	}
2605 
2606 retry:
2607 	vma = vma_lookup(current->mm, addr);
2608 
2609 	if (vma == NULL)
2610 		pfn = KVM_PFN_ERR_FAULT;
2611 	else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2612 		r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2613 		if (r == -EAGAIN)
2614 			goto retry;
2615 		if (r < 0)
2616 			pfn = KVM_PFN_ERR_FAULT;
2617 	} else {
2618 		if (async && vma_is_valid(vma, write_fault))
2619 			*async = true;
2620 		pfn = KVM_PFN_ERR_FAULT;
2621 	}
2622 exit:
2623 	mmap_read_unlock(current->mm);
2624 	return pfn;
2625 }
2626 
2627 kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2628 			       bool atomic, bool *async, bool write_fault,
2629 			       bool *writable, hva_t *hva)
2630 {
2631 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2632 
2633 	if (hva)
2634 		*hva = addr;
2635 
2636 	if (addr == KVM_HVA_ERR_RO_BAD) {
2637 		if (writable)
2638 			*writable = false;
2639 		return KVM_PFN_ERR_RO_FAULT;
2640 	}
2641 
2642 	if (kvm_is_error_hva(addr)) {
2643 		if (writable)
2644 			*writable = false;
2645 		return KVM_PFN_NOSLOT;
2646 	}
2647 
2648 	/* Do not map writable pfn in the readonly memslot. */
2649 	if (writable && memslot_is_readonly(slot)) {
2650 		*writable = false;
2651 		writable = NULL;
2652 	}
2653 
2654 	return hva_to_pfn(addr, atomic, async, write_fault,
2655 			  writable);
2656 }
2657 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2658 
2659 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2660 		      bool *writable)
2661 {
2662 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
2663 				    write_fault, writable, NULL);
2664 }
2665 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2666 
2667 kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2668 {
2669 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL, NULL);
2670 }
2671 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2672 
2673 kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2674 {
2675 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL, NULL);
2676 }
2677 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2678 
2679 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2680 {
2681 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2682 }
2683 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2684 
2685 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2686 {
2687 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2688 }
2689 EXPORT_SYMBOL_GPL(gfn_to_pfn);
2690 
2691 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2692 {
2693 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2694 }
2695 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2696 
2697 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2698 			    struct page **pages, int nr_pages)
2699 {
2700 	unsigned long addr;
2701 	gfn_t entry = 0;
2702 
2703 	addr = gfn_to_hva_many(slot, gfn, &entry);
2704 	if (kvm_is_error_hva(addr))
2705 		return -1;
2706 
2707 	if (entry < nr_pages)
2708 		return 0;
2709 
2710 	return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2711 }
2712 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2713 
2714 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
2715 {
2716 	if (is_error_noslot_pfn(pfn))
2717 		return KVM_ERR_PTR_BAD_PAGE;
2718 
2719 	if (kvm_is_reserved_pfn(pfn)) {
2720 		WARN_ON(1);
2721 		return KVM_ERR_PTR_BAD_PAGE;
2722 	}
2723 
2724 	return pfn_to_page(pfn);
2725 }
2726 
2727 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2728 {
2729 	kvm_pfn_t pfn;
2730 
2731 	pfn = gfn_to_pfn(kvm, gfn);
2732 
2733 	return kvm_pfn_to_page(pfn);
2734 }
2735 EXPORT_SYMBOL_GPL(gfn_to_page);
2736 
2737 void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2738 {
2739 	if (pfn == 0)
2740 		return;
2741 
2742 	if (dirty)
2743 		kvm_release_pfn_dirty(pfn);
2744 	else
2745 		kvm_release_pfn_clean(pfn);
2746 }
2747 
2748 int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2749 {
2750 	kvm_pfn_t pfn;
2751 	void *hva = NULL;
2752 	struct page *page = KVM_UNMAPPED_PAGE;
2753 
2754 	if (!map)
2755 		return -EINVAL;
2756 
2757 	pfn = gfn_to_pfn(vcpu->kvm, gfn);
2758 	if (is_error_noslot_pfn(pfn))
2759 		return -EINVAL;
2760 
2761 	if (pfn_valid(pfn)) {
2762 		page = pfn_to_page(pfn);
2763 		hva = kmap(page);
2764 #ifdef CONFIG_HAS_IOMEM
2765 	} else {
2766 		hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2767 #endif
2768 	}
2769 
2770 	if (!hva)
2771 		return -EFAULT;
2772 
2773 	map->page = page;
2774 	map->hva = hva;
2775 	map->pfn = pfn;
2776 	map->gfn = gfn;
2777 
2778 	return 0;
2779 }
2780 EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2781 
2782 void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2783 {
2784 	if (!map)
2785 		return;
2786 
2787 	if (!map->hva)
2788 		return;
2789 
2790 	if (map->page != KVM_UNMAPPED_PAGE)
2791 		kunmap(map->page);
2792 #ifdef CONFIG_HAS_IOMEM
2793 	else
2794 		memunmap(map->hva);
2795 #endif
2796 
2797 	if (dirty)
2798 		kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2799 
2800 	kvm_release_pfn(map->pfn, dirty);
2801 
2802 	map->hva = NULL;
2803 	map->page = NULL;
2804 }
2805 EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2806 
2807 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
2808 {
2809 	kvm_pfn_t pfn;
2810 
2811 	pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
2812 
2813 	return kvm_pfn_to_page(pfn);
2814 }
2815 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
2816 
2817 void kvm_release_page_clean(struct page *page)
2818 {
2819 	WARN_ON(is_error_page(page));
2820 
2821 	kvm_release_pfn_clean(page_to_pfn(page));
2822 }
2823 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2824 
2825 void kvm_release_pfn_clean(kvm_pfn_t pfn)
2826 {
2827 	if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
2828 		put_page(pfn_to_page(pfn));
2829 }
2830 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2831 
2832 void kvm_release_page_dirty(struct page *page)
2833 {
2834 	WARN_ON(is_error_page(page));
2835 
2836 	kvm_release_pfn_dirty(page_to_pfn(page));
2837 }
2838 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2839 
2840 void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2841 {
2842 	kvm_set_pfn_dirty(pfn);
2843 	kvm_release_pfn_clean(pfn);
2844 }
2845 EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2846 
2847 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2848 {
2849 	if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2850 		SetPageDirty(pfn_to_page(pfn));
2851 }
2852 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2853 
2854 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2855 {
2856 	if (!kvm_is_reserved_pfn(pfn) && !kvm_is_zone_device_pfn(pfn))
2857 		mark_page_accessed(pfn_to_page(pfn));
2858 }
2859 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2860 
2861 static int next_segment(unsigned long len, int offset)
2862 {
2863 	if (len > PAGE_SIZE - offset)
2864 		return PAGE_SIZE - offset;
2865 	else
2866 		return len;
2867 }
2868 
2869 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2870 				 void *data, int offset, int len)
2871 {
2872 	int r;
2873 	unsigned long addr;
2874 
2875 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2876 	if (kvm_is_error_hva(addr))
2877 		return -EFAULT;
2878 	r = __copy_from_user(data, (void __user *)addr + offset, len);
2879 	if (r)
2880 		return -EFAULT;
2881 	return 0;
2882 }
2883 
2884 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
2885 			int len)
2886 {
2887 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2888 
2889 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
2890 }
2891 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
2892 
2893 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
2894 			     int offset, int len)
2895 {
2896 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2897 
2898 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
2899 }
2900 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
2901 
2902 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
2903 {
2904 	gfn_t gfn = gpa >> PAGE_SHIFT;
2905 	int seg;
2906 	int offset = offset_in_page(gpa);
2907 	int ret;
2908 
2909 	while ((seg = next_segment(len, offset)) != 0) {
2910 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
2911 		if (ret < 0)
2912 			return ret;
2913 		offset = 0;
2914 		len -= seg;
2915 		data += seg;
2916 		++gfn;
2917 	}
2918 	return 0;
2919 }
2920 EXPORT_SYMBOL_GPL(kvm_read_guest);
2921 
2922 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
2923 {
2924 	gfn_t gfn = gpa >> PAGE_SHIFT;
2925 	int seg;
2926 	int offset = offset_in_page(gpa);
2927 	int ret;
2928 
2929 	while ((seg = next_segment(len, offset)) != 0) {
2930 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
2931 		if (ret < 0)
2932 			return ret;
2933 		offset = 0;
2934 		len -= seg;
2935 		data += seg;
2936 		++gfn;
2937 	}
2938 	return 0;
2939 }
2940 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
2941 
2942 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2943 			           void *data, int offset, unsigned long len)
2944 {
2945 	int r;
2946 	unsigned long addr;
2947 
2948 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
2949 	if (kvm_is_error_hva(addr))
2950 		return -EFAULT;
2951 	pagefault_disable();
2952 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
2953 	pagefault_enable();
2954 	if (r)
2955 		return -EFAULT;
2956 	return 0;
2957 }
2958 
2959 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
2960 			       void *data, unsigned long len)
2961 {
2962 	gfn_t gfn = gpa >> PAGE_SHIFT;
2963 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2964 	int offset = offset_in_page(gpa);
2965 
2966 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
2967 }
2968 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
2969 
2970 static int __kvm_write_guest_page(struct kvm *kvm,
2971 				  struct kvm_memory_slot *memslot, gfn_t gfn,
2972 			          const void *data, int offset, int len)
2973 {
2974 	int r;
2975 	unsigned long addr;
2976 
2977 	addr = gfn_to_hva_memslot(memslot, gfn);
2978 	if (kvm_is_error_hva(addr))
2979 		return -EFAULT;
2980 	r = __copy_to_user((void __user *)addr + offset, data, len);
2981 	if (r)
2982 		return -EFAULT;
2983 	mark_page_dirty_in_slot(kvm, memslot, gfn);
2984 	return 0;
2985 }
2986 
2987 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
2988 			 const void *data, int offset, int len)
2989 {
2990 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2991 
2992 	return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
2993 }
2994 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
2995 
2996 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
2997 			      const void *data, int offset, int len)
2998 {
2999 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3000 
3001 	return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3002 }
3003 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3004 
3005 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3006 		    unsigned long len)
3007 {
3008 	gfn_t gfn = gpa >> PAGE_SHIFT;
3009 	int seg;
3010 	int offset = offset_in_page(gpa);
3011 	int ret;
3012 
3013 	while ((seg = next_segment(len, offset)) != 0) {
3014 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3015 		if (ret < 0)
3016 			return ret;
3017 		offset = 0;
3018 		len -= seg;
3019 		data += seg;
3020 		++gfn;
3021 	}
3022 	return 0;
3023 }
3024 EXPORT_SYMBOL_GPL(kvm_write_guest);
3025 
3026 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3027 		         unsigned long len)
3028 {
3029 	gfn_t gfn = gpa >> PAGE_SHIFT;
3030 	int seg;
3031 	int offset = offset_in_page(gpa);
3032 	int ret;
3033 
3034 	while ((seg = next_segment(len, offset)) != 0) {
3035 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3036 		if (ret < 0)
3037 			return ret;
3038 		offset = 0;
3039 		len -= seg;
3040 		data += seg;
3041 		++gfn;
3042 	}
3043 	return 0;
3044 }
3045 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3046 
3047 static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3048 				       struct gfn_to_hva_cache *ghc,
3049 				       gpa_t gpa, unsigned long len)
3050 {
3051 	int offset = offset_in_page(gpa);
3052 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
3053 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3054 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3055 	gfn_t nr_pages_avail;
3056 
3057 	/* Update ghc->generation before performing any error checks. */
3058 	ghc->generation = slots->generation;
3059 
3060 	if (start_gfn > end_gfn) {
3061 		ghc->hva = KVM_HVA_ERR_BAD;
3062 		return -EINVAL;
3063 	}
3064 
3065 	/*
3066 	 * If the requested region crosses two memslots, we still
3067 	 * verify that the entire region is valid here.
3068 	 */
3069 	for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3070 		ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3071 		ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3072 					   &nr_pages_avail);
3073 		if (kvm_is_error_hva(ghc->hva))
3074 			return -EFAULT;
3075 	}
3076 
3077 	/* Use the slow path for cross page reads and writes. */
3078 	if (nr_pages_needed == 1)
3079 		ghc->hva += offset;
3080 	else
3081 		ghc->memslot = NULL;
3082 
3083 	ghc->gpa = gpa;
3084 	ghc->len = len;
3085 	return 0;
3086 }
3087 
3088 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3089 			      gpa_t gpa, unsigned long len)
3090 {
3091 	struct kvm_memslots *slots = kvm_memslots(kvm);
3092 	return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3093 }
3094 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3095 
3096 int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3097 				  void *data, unsigned int offset,
3098 				  unsigned long len)
3099 {
3100 	struct kvm_memslots *slots = kvm_memslots(kvm);
3101 	int r;
3102 	gpa_t gpa = ghc->gpa + offset;
3103 
3104 	if (WARN_ON_ONCE(len + offset > ghc->len))
3105 		return -EINVAL;
3106 
3107 	if (slots->generation != ghc->generation) {
3108 		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3109 			return -EFAULT;
3110 	}
3111 
3112 	if (kvm_is_error_hva(ghc->hva))
3113 		return -EFAULT;
3114 
3115 	if (unlikely(!ghc->memslot))
3116 		return kvm_write_guest(kvm, gpa, data, len);
3117 
3118 	r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3119 	if (r)
3120 		return -EFAULT;
3121 	mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3122 
3123 	return 0;
3124 }
3125 EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3126 
3127 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3128 			   void *data, unsigned long len)
3129 {
3130 	return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3131 }
3132 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3133 
3134 int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3135 				 void *data, unsigned int offset,
3136 				 unsigned long len)
3137 {
3138 	struct kvm_memslots *slots = kvm_memslots(kvm);
3139 	int r;
3140 	gpa_t gpa = ghc->gpa + offset;
3141 
3142 	if (WARN_ON_ONCE(len + offset > ghc->len))
3143 		return -EINVAL;
3144 
3145 	if (slots->generation != ghc->generation) {
3146 		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3147 			return -EFAULT;
3148 	}
3149 
3150 	if (kvm_is_error_hva(ghc->hva))
3151 		return -EFAULT;
3152 
3153 	if (unlikely(!ghc->memslot))
3154 		return kvm_read_guest(kvm, gpa, data, len);
3155 
3156 	r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3157 	if (r)
3158 		return -EFAULT;
3159 
3160 	return 0;
3161 }
3162 EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3163 
3164 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3165 			  void *data, unsigned long len)
3166 {
3167 	return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3168 }
3169 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3170 
3171 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3172 {
3173 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3174 	gfn_t gfn = gpa >> PAGE_SHIFT;
3175 	int seg;
3176 	int offset = offset_in_page(gpa);
3177 	int ret;
3178 
3179 	while ((seg = next_segment(len, offset)) != 0) {
3180 		ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3181 		if (ret < 0)
3182 			return ret;
3183 		offset = 0;
3184 		len -= seg;
3185 		++gfn;
3186 	}
3187 	return 0;
3188 }
3189 EXPORT_SYMBOL_GPL(kvm_clear_guest);
3190 
3191 void mark_page_dirty_in_slot(struct kvm *kvm,
3192 			     const struct kvm_memory_slot *memslot,
3193 		 	     gfn_t gfn)
3194 {
3195 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3196 
3197 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3198 	if (WARN_ON_ONCE(!vcpu) || WARN_ON_ONCE(vcpu->kvm != kvm))
3199 		return;
3200 #endif
3201 
3202 	if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3203 		unsigned long rel_gfn = gfn - memslot->base_gfn;
3204 		u32 slot = (memslot->as_id << 16) | memslot->id;
3205 
3206 		if (kvm->dirty_ring_size)
3207 			kvm_dirty_ring_push(&vcpu->dirty_ring,
3208 					    slot, rel_gfn);
3209 		else
3210 			set_bit_le(rel_gfn, memslot->dirty_bitmap);
3211 	}
3212 }
3213 EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3214 
3215 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3216 {
3217 	struct kvm_memory_slot *memslot;
3218 
3219 	memslot = gfn_to_memslot(kvm, gfn);
3220 	mark_page_dirty_in_slot(kvm, memslot, gfn);
3221 }
3222 EXPORT_SYMBOL_GPL(mark_page_dirty);
3223 
3224 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3225 {
3226 	struct kvm_memory_slot *memslot;
3227 
3228 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3229 	mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3230 }
3231 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3232 
3233 void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3234 {
3235 	if (!vcpu->sigset_active)
3236 		return;
3237 
3238 	/*
3239 	 * This does a lockless modification of ->real_blocked, which is fine
3240 	 * because, only current can change ->real_blocked and all readers of
3241 	 * ->real_blocked don't care as long ->real_blocked is always a subset
3242 	 * of ->blocked.
3243 	 */
3244 	sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
3245 }
3246 
3247 void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3248 {
3249 	if (!vcpu->sigset_active)
3250 		return;
3251 
3252 	sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
3253 	sigemptyset(&current->real_blocked);
3254 }
3255 
3256 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3257 {
3258 	unsigned int old, val, grow, grow_start;
3259 
3260 	old = val = vcpu->halt_poll_ns;
3261 	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3262 	grow = READ_ONCE(halt_poll_ns_grow);
3263 	if (!grow)
3264 		goto out;
3265 
3266 	val *= grow;
3267 	if (val < grow_start)
3268 		val = grow_start;
3269 
3270 	if (val > vcpu->kvm->max_halt_poll_ns)
3271 		val = vcpu->kvm->max_halt_poll_ns;
3272 
3273 	vcpu->halt_poll_ns = val;
3274 out:
3275 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3276 }
3277 
3278 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3279 {
3280 	unsigned int old, val, shrink, grow_start;
3281 
3282 	old = val = vcpu->halt_poll_ns;
3283 	shrink = READ_ONCE(halt_poll_ns_shrink);
3284 	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3285 	if (shrink == 0)
3286 		val = 0;
3287 	else
3288 		val /= shrink;
3289 
3290 	if (val < grow_start)
3291 		val = 0;
3292 
3293 	vcpu->halt_poll_ns = val;
3294 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3295 }
3296 
3297 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3298 {
3299 	int ret = -EINTR;
3300 	int idx = srcu_read_lock(&vcpu->kvm->srcu);
3301 
3302 	if (kvm_arch_vcpu_runnable(vcpu)) {
3303 		kvm_make_request(KVM_REQ_UNHALT, vcpu);
3304 		goto out;
3305 	}
3306 	if (kvm_cpu_has_pending_timer(vcpu))
3307 		goto out;
3308 	if (signal_pending(current))
3309 		goto out;
3310 	if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3311 		goto out;
3312 
3313 	ret = 0;
3314 out:
3315 	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3316 	return ret;
3317 }
3318 
3319 /*
3320  * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3321  * pending.  This is mostly used when halting a vCPU, but may also be used
3322  * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3323  */
3324 bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3325 {
3326 	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3327 	bool waited = false;
3328 
3329 	vcpu->stat.generic.blocking = 1;
3330 
3331 	preempt_disable();
3332 	kvm_arch_vcpu_blocking(vcpu);
3333 	prepare_to_rcuwait(wait);
3334 	preempt_enable();
3335 
3336 	for (;;) {
3337 		set_current_state(TASK_INTERRUPTIBLE);
3338 
3339 		if (kvm_vcpu_check_block(vcpu) < 0)
3340 			break;
3341 
3342 		waited = true;
3343 		schedule();
3344 	}
3345 
3346 	preempt_disable();
3347 	finish_rcuwait(wait);
3348 	kvm_arch_vcpu_unblocking(vcpu);
3349 	preempt_enable();
3350 
3351 	vcpu->stat.generic.blocking = 0;
3352 
3353 	return waited;
3354 }
3355 
3356 static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3357 					  ktime_t end, bool success)
3358 {
3359 	struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3360 	u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3361 
3362 	++vcpu->stat.generic.halt_attempted_poll;
3363 
3364 	if (success) {
3365 		++vcpu->stat.generic.halt_successful_poll;
3366 
3367 		if (!vcpu_valid_wakeup(vcpu))
3368 			++vcpu->stat.generic.halt_poll_invalid;
3369 
3370 		stats->halt_poll_success_ns += poll_ns;
3371 		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3372 	} else {
3373 		stats->halt_poll_fail_ns += poll_ns;
3374 		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3375 	}
3376 }
3377 
3378 /*
3379  * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc...  If halt
3380  * polling is enabled, busy wait for a short time before blocking to avoid the
3381  * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3382  * is halted.
3383  */
3384 void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3385 {
3386 	bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3387 	bool do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3388 	ktime_t start, cur, poll_end;
3389 	bool waited = false;
3390 	u64 halt_ns;
3391 
3392 	start = cur = poll_end = ktime_get();
3393 	if (do_halt_poll) {
3394 		ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3395 
3396 		do {
3397 			/*
3398 			 * This sets KVM_REQ_UNHALT if an interrupt
3399 			 * arrives.
3400 			 */
3401 			if (kvm_vcpu_check_block(vcpu) < 0)
3402 				goto out;
3403 			cpu_relax();
3404 			poll_end = cur = ktime_get();
3405 		} while (kvm_vcpu_can_poll(cur, stop));
3406 	}
3407 
3408 	waited = kvm_vcpu_block(vcpu);
3409 
3410 	cur = ktime_get();
3411 	if (waited) {
3412 		vcpu->stat.generic.halt_wait_ns +=
3413 			ktime_to_ns(cur) - ktime_to_ns(poll_end);
3414 		KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3415 				ktime_to_ns(cur) - ktime_to_ns(poll_end));
3416 	}
3417 out:
3418 	/* The total time the vCPU was "halted", including polling time. */
3419 	halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3420 
3421 	/*
3422 	 * Note, halt-polling is considered successful so long as the vCPU was
3423 	 * never actually scheduled out, i.e. even if the wake event arrived
3424 	 * after of the halt-polling loop itself, but before the full wait.
3425 	 */
3426 	if (do_halt_poll)
3427 		update_halt_poll_stats(vcpu, start, poll_end, !waited);
3428 
3429 	if (halt_poll_allowed) {
3430 		if (!vcpu_valid_wakeup(vcpu)) {
3431 			shrink_halt_poll_ns(vcpu);
3432 		} else if (vcpu->kvm->max_halt_poll_ns) {
3433 			if (halt_ns <= vcpu->halt_poll_ns)
3434 				;
3435 			/* we had a long block, shrink polling */
3436 			else if (vcpu->halt_poll_ns &&
3437 				 halt_ns > vcpu->kvm->max_halt_poll_ns)
3438 				shrink_halt_poll_ns(vcpu);
3439 			/* we had a short halt and our poll time is too small */
3440 			else if (vcpu->halt_poll_ns < vcpu->kvm->max_halt_poll_ns &&
3441 				 halt_ns < vcpu->kvm->max_halt_poll_ns)
3442 				grow_halt_poll_ns(vcpu);
3443 		} else {
3444 			vcpu->halt_poll_ns = 0;
3445 		}
3446 	}
3447 
3448 	trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3449 }
3450 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3451 
3452 bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3453 {
3454 	if (__kvm_vcpu_wake_up(vcpu)) {
3455 		WRITE_ONCE(vcpu->ready, true);
3456 		++vcpu->stat.generic.halt_wakeup;
3457 		return true;
3458 	}
3459 
3460 	return false;
3461 }
3462 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3463 
3464 #ifndef CONFIG_S390
3465 /*
3466  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3467  */
3468 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3469 {
3470 	int me, cpu;
3471 
3472 	if (kvm_vcpu_wake_up(vcpu))
3473 		return;
3474 
3475 	me = get_cpu();
3476 	/*
3477 	 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3478 	 * to EXITING_GUEST_MODE.  Therefore the moderately expensive "should
3479 	 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3480 	 * within the vCPU thread itself.
3481 	 */
3482 	if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3483 		if (vcpu->mode == IN_GUEST_MODE)
3484 			WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3485 		goto out;
3486 	}
3487 
3488 	/*
3489 	 * Note, the vCPU could get migrated to a different pCPU at any point
3490 	 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3491 	 * IPI to the previous pCPU.  But, that's ok because the purpose of the
3492 	 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3493 	 * vCPU also requires it to leave IN_GUEST_MODE.
3494 	 */
3495 	if (kvm_arch_vcpu_should_kick(vcpu)) {
3496 		cpu = READ_ONCE(vcpu->cpu);
3497 		if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3498 			smp_send_reschedule(cpu);
3499 	}
3500 out:
3501 	put_cpu();
3502 }
3503 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3504 #endif /* !CONFIG_S390 */
3505 
3506 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3507 {
3508 	struct pid *pid;
3509 	struct task_struct *task = NULL;
3510 	int ret = 0;
3511 
3512 	rcu_read_lock();
3513 	pid = rcu_dereference(target->pid);
3514 	if (pid)
3515 		task = get_pid_task(pid, PIDTYPE_PID);
3516 	rcu_read_unlock();
3517 	if (!task)
3518 		return ret;
3519 	ret = yield_to(task, 1);
3520 	put_task_struct(task);
3521 
3522 	return ret;
3523 }
3524 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3525 
3526 /*
3527  * Helper that checks whether a VCPU is eligible for directed yield.
3528  * Most eligible candidate to yield is decided by following heuristics:
3529  *
3530  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3531  *  (preempted lock holder), indicated by @in_spin_loop.
3532  *  Set at the beginning and cleared at the end of interception/PLE handler.
3533  *
3534  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3535  *  chance last time (mostly it has become eligible now since we have probably
3536  *  yielded to lockholder in last iteration. This is done by toggling
3537  *  @dy_eligible each time a VCPU checked for eligibility.)
3538  *
3539  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3540  *  to preempted lock-holder could result in wrong VCPU selection and CPU
3541  *  burning. Giving priority for a potential lock-holder increases lock
3542  *  progress.
3543  *
3544  *  Since algorithm is based on heuristics, accessing another VCPU data without
3545  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
3546  *  and continue with next VCPU and so on.
3547  */
3548 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3549 {
3550 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3551 	bool eligible;
3552 
3553 	eligible = !vcpu->spin_loop.in_spin_loop ||
3554 		    vcpu->spin_loop.dy_eligible;
3555 
3556 	if (vcpu->spin_loop.in_spin_loop)
3557 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3558 
3559 	return eligible;
3560 #else
3561 	return true;
3562 #endif
3563 }
3564 
3565 /*
3566  * Unlike kvm_arch_vcpu_runnable, this function is called outside
3567  * a vcpu_load/vcpu_put pair.  However, for most architectures
3568  * kvm_arch_vcpu_runnable does not require vcpu_load.
3569  */
3570 bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3571 {
3572 	return kvm_arch_vcpu_runnable(vcpu);
3573 }
3574 
3575 static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3576 {
3577 	if (kvm_arch_dy_runnable(vcpu))
3578 		return true;
3579 
3580 #ifdef CONFIG_KVM_ASYNC_PF
3581 	if (!list_empty_careful(&vcpu->async_pf.done))
3582 		return true;
3583 #endif
3584 
3585 	return false;
3586 }
3587 
3588 bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3589 {
3590 	return false;
3591 }
3592 
3593 void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3594 {
3595 	struct kvm *kvm = me->kvm;
3596 	struct kvm_vcpu *vcpu;
3597 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3598 	unsigned long i;
3599 	int yielded = 0;
3600 	int try = 3;
3601 	int pass;
3602 
3603 	kvm_vcpu_set_in_spin_loop(me, true);
3604 	/*
3605 	 * We boost the priority of a VCPU that is runnable but not
3606 	 * currently running, because it got preempted by something
3607 	 * else and called schedule in __vcpu_run.  Hopefully that
3608 	 * VCPU is holding the lock that we need and will release it.
3609 	 * We approximate round-robin by starting at the last boosted VCPU.
3610 	 */
3611 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
3612 		kvm_for_each_vcpu(i, vcpu, kvm) {
3613 			if (!pass && i <= last_boosted_vcpu) {
3614 				i = last_boosted_vcpu;
3615 				continue;
3616 			} else if (pass && i > last_boosted_vcpu)
3617 				break;
3618 			if (!READ_ONCE(vcpu->ready))
3619 				continue;
3620 			if (vcpu == me)
3621 				continue;
3622 			if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3623 				continue;
3624 			if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3625 			    !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3626 			    !kvm_arch_vcpu_in_kernel(vcpu))
3627 				continue;
3628 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3629 				continue;
3630 
3631 			yielded = kvm_vcpu_yield_to(vcpu);
3632 			if (yielded > 0) {
3633 				kvm->last_boosted_vcpu = i;
3634 				break;
3635 			} else if (yielded < 0) {
3636 				try--;
3637 				if (!try)
3638 					break;
3639 			}
3640 		}
3641 	}
3642 	kvm_vcpu_set_in_spin_loop(me, false);
3643 
3644 	/* Ensure vcpu is not eligible during next spinloop */
3645 	kvm_vcpu_set_dy_eligible(me, false);
3646 }
3647 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3648 
3649 static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3650 {
3651 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
3652 	return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3653 	    (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3654 	     kvm->dirty_ring_size / PAGE_SIZE);
3655 #else
3656 	return false;
3657 #endif
3658 }
3659 
3660 static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3661 {
3662 	struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3663 	struct page *page;
3664 
3665 	if (vmf->pgoff == 0)
3666 		page = virt_to_page(vcpu->run);
3667 #ifdef CONFIG_X86
3668 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3669 		page = virt_to_page(vcpu->arch.pio_data);
3670 #endif
3671 #ifdef CONFIG_KVM_MMIO
3672 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3673 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3674 #endif
3675 	else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3676 		page = kvm_dirty_ring_get_page(
3677 		    &vcpu->dirty_ring,
3678 		    vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3679 	else
3680 		return kvm_arch_vcpu_fault(vcpu, vmf);
3681 	get_page(page);
3682 	vmf->page = page;
3683 	return 0;
3684 }
3685 
3686 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3687 	.fault = kvm_vcpu_fault,
3688 };
3689 
3690 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3691 {
3692 	struct kvm_vcpu *vcpu = file->private_data;
3693 	unsigned long pages = vma_pages(vma);
3694 
3695 	if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3696 	     kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3697 	    ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3698 		return -EINVAL;
3699 
3700 	vma->vm_ops = &kvm_vcpu_vm_ops;
3701 	return 0;
3702 }
3703 
3704 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3705 {
3706 	struct kvm_vcpu *vcpu = filp->private_data;
3707 
3708 	kvm_put_kvm(vcpu->kvm);
3709 	return 0;
3710 }
3711 
3712 static const struct file_operations kvm_vcpu_fops = {
3713 	.release        = kvm_vcpu_release,
3714 	.unlocked_ioctl = kvm_vcpu_ioctl,
3715 	.mmap           = kvm_vcpu_mmap,
3716 	.llseek		= noop_llseek,
3717 	KVM_COMPAT(kvm_vcpu_compat_ioctl),
3718 };
3719 
3720 /*
3721  * Allocates an inode for the vcpu.
3722  */
3723 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3724 {
3725 	char name[8 + 1 + ITOA_MAX_LEN + 1];
3726 
3727 	snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3728 	return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3729 }
3730 
3731 static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3732 {
3733 #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3734 	struct dentry *debugfs_dentry;
3735 	char dir_name[ITOA_MAX_LEN * 2];
3736 
3737 	if (!debugfs_initialized())
3738 		return;
3739 
3740 	snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3741 	debugfs_dentry = debugfs_create_dir(dir_name,
3742 					    vcpu->kvm->debugfs_dentry);
3743 
3744 	kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3745 #endif
3746 }
3747 
3748 /*
3749  * Creates some virtual cpus.  Good luck creating more than one.
3750  */
3751 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3752 {
3753 	int r;
3754 	struct kvm_vcpu *vcpu;
3755 	struct page *page;
3756 
3757 	if (id >= KVM_MAX_VCPU_IDS)
3758 		return -EINVAL;
3759 
3760 	mutex_lock(&kvm->lock);
3761 	if (kvm->created_vcpus >= kvm->max_vcpus) {
3762 		mutex_unlock(&kvm->lock);
3763 		return -EINVAL;
3764 	}
3765 
3766 	kvm->created_vcpus++;
3767 	mutex_unlock(&kvm->lock);
3768 
3769 	r = kvm_arch_vcpu_precreate(kvm, id);
3770 	if (r)
3771 		goto vcpu_decrement;
3772 
3773 	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3774 	if (!vcpu) {
3775 		r = -ENOMEM;
3776 		goto vcpu_decrement;
3777 	}
3778 
3779 	BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3780 	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3781 	if (!page) {
3782 		r = -ENOMEM;
3783 		goto vcpu_free;
3784 	}
3785 	vcpu->run = page_address(page);
3786 
3787 	kvm_vcpu_init(vcpu, kvm, id);
3788 
3789 	r = kvm_arch_vcpu_create(vcpu);
3790 	if (r)
3791 		goto vcpu_free_run_page;
3792 
3793 	if (kvm->dirty_ring_size) {
3794 		r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3795 					 id, kvm->dirty_ring_size);
3796 		if (r)
3797 			goto arch_vcpu_destroy;
3798 	}
3799 
3800 	mutex_lock(&kvm->lock);
3801 	if (kvm_get_vcpu_by_id(kvm, id)) {
3802 		r = -EEXIST;
3803 		goto unlock_vcpu_destroy;
3804 	}
3805 
3806 	vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3807 	r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3808 	BUG_ON(r == -EBUSY);
3809 	if (r)
3810 		goto unlock_vcpu_destroy;
3811 
3812 	/* Fill the stats id string for the vcpu */
3813 	snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
3814 		 task_pid_nr(current), id);
3815 
3816 	/* Now it's all set up, let userspace reach it */
3817 	kvm_get_kvm(kvm);
3818 	r = create_vcpu_fd(vcpu);
3819 	if (r < 0) {
3820 		xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3821 		kvm_put_kvm_no_destroy(kvm);
3822 		goto unlock_vcpu_destroy;
3823 	}
3824 
3825 	/*
3826 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Store the vcpu
3827 	 * pointer before kvm->online_vcpu's incremented value.
3828 	 */
3829 	smp_wmb();
3830 	atomic_inc(&kvm->online_vcpus);
3831 
3832 	mutex_unlock(&kvm->lock);
3833 	kvm_arch_vcpu_postcreate(vcpu);
3834 	kvm_create_vcpu_debugfs(vcpu);
3835 	return r;
3836 
3837 unlock_vcpu_destroy:
3838 	mutex_unlock(&kvm->lock);
3839 	kvm_dirty_ring_free(&vcpu->dirty_ring);
3840 arch_vcpu_destroy:
3841 	kvm_arch_vcpu_destroy(vcpu);
3842 vcpu_free_run_page:
3843 	free_page((unsigned long)vcpu->run);
3844 vcpu_free:
3845 	kmem_cache_free(kvm_vcpu_cache, vcpu);
3846 vcpu_decrement:
3847 	mutex_lock(&kvm->lock);
3848 	kvm->created_vcpus--;
3849 	mutex_unlock(&kvm->lock);
3850 	return r;
3851 }
3852 
3853 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
3854 {
3855 	if (sigset) {
3856 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
3857 		vcpu->sigset_active = 1;
3858 		vcpu->sigset = *sigset;
3859 	} else
3860 		vcpu->sigset_active = 0;
3861 	return 0;
3862 }
3863 
3864 static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
3865 			      size_t size, loff_t *offset)
3866 {
3867 	struct kvm_vcpu *vcpu = file->private_data;
3868 
3869 	return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
3870 			&kvm_vcpu_stats_desc[0], &vcpu->stat,
3871 			sizeof(vcpu->stat), user_buffer, size, offset);
3872 }
3873 
3874 static const struct file_operations kvm_vcpu_stats_fops = {
3875 	.read = kvm_vcpu_stats_read,
3876 	.llseek = noop_llseek,
3877 };
3878 
3879 static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
3880 {
3881 	int fd;
3882 	struct file *file;
3883 	char name[15 + ITOA_MAX_LEN + 1];
3884 
3885 	snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
3886 
3887 	fd = get_unused_fd_flags(O_CLOEXEC);
3888 	if (fd < 0)
3889 		return fd;
3890 
3891 	file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
3892 	if (IS_ERR(file)) {
3893 		put_unused_fd(fd);
3894 		return PTR_ERR(file);
3895 	}
3896 	file->f_mode |= FMODE_PREAD;
3897 	fd_install(fd, file);
3898 
3899 	return fd;
3900 }
3901 
3902 static long kvm_vcpu_ioctl(struct file *filp,
3903 			   unsigned int ioctl, unsigned long arg)
3904 {
3905 	struct kvm_vcpu *vcpu = filp->private_data;
3906 	void __user *argp = (void __user *)arg;
3907 	int r;
3908 	struct kvm_fpu *fpu = NULL;
3909 	struct kvm_sregs *kvm_sregs = NULL;
3910 
3911 	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
3912 		return -EIO;
3913 
3914 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
3915 		return -EINVAL;
3916 
3917 	/*
3918 	 * Some architectures have vcpu ioctls that are asynchronous to vcpu
3919 	 * execution; mutex_lock() would break them.
3920 	 */
3921 	r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
3922 	if (r != -ENOIOCTLCMD)
3923 		return r;
3924 
3925 	if (mutex_lock_killable(&vcpu->mutex))
3926 		return -EINTR;
3927 	switch (ioctl) {
3928 	case KVM_RUN: {
3929 		struct pid *oldpid;
3930 		r = -EINVAL;
3931 		if (arg)
3932 			goto out;
3933 		oldpid = rcu_access_pointer(vcpu->pid);
3934 		if (unlikely(oldpid != task_pid(current))) {
3935 			/* The thread running this VCPU changed. */
3936 			struct pid *newpid;
3937 
3938 			r = kvm_arch_vcpu_run_pid_change(vcpu);
3939 			if (r)
3940 				break;
3941 
3942 			newpid = get_task_pid(current, PIDTYPE_PID);
3943 			rcu_assign_pointer(vcpu->pid, newpid);
3944 			if (oldpid)
3945 				synchronize_rcu();
3946 			put_pid(oldpid);
3947 		}
3948 		r = kvm_arch_vcpu_ioctl_run(vcpu);
3949 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
3950 		break;
3951 	}
3952 	case KVM_GET_REGS: {
3953 		struct kvm_regs *kvm_regs;
3954 
3955 		r = -ENOMEM;
3956 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
3957 		if (!kvm_regs)
3958 			goto out;
3959 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
3960 		if (r)
3961 			goto out_free1;
3962 		r = -EFAULT;
3963 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
3964 			goto out_free1;
3965 		r = 0;
3966 out_free1:
3967 		kfree(kvm_regs);
3968 		break;
3969 	}
3970 	case KVM_SET_REGS: {
3971 		struct kvm_regs *kvm_regs;
3972 
3973 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
3974 		if (IS_ERR(kvm_regs)) {
3975 			r = PTR_ERR(kvm_regs);
3976 			goto out;
3977 		}
3978 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
3979 		kfree(kvm_regs);
3980 		break;
3981 	}
3982 	case KVM_GET_SREGS: {
3983 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
3984 				    GFP_KERNEL_ACCOUNT);
3985 		r = -ENOMEM;
3986 		if (!kvm_sregs)
3987 			goto out;
3988 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
3989 		if (r)
3990 			goto out;
3991 		r = -EFAULT;
3992 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
3993 			goto out;
3994 		r = 0;
3995 		break;
3996 	}
3997 	case KVM_SET_SREGS: {
3998 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
3999 		if (IS_ERR(kvm_sregs)) {
4000 			r = PTR_ERR(kvm_sregs);
4001 			kvm_sregs = NULL;
4002 			goto out;
4003 		}
4004 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4005 		break;
4006 	}
4007 	case KVM_GET_MP_STATE: {
4008 		struct kvm_mp_state mp_state;
4009 
4010 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4011 		if (r)
4012 			goto out;
4013 		r = -EFAULT;
4014 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4015 			goto out;
4016 		r = 0;
4017 		break;
4018 	}
4019 	case KVM_SET_MP_STATE: {
4020 		struct kvm_mp_state mp_state;
4021 
4022 		r = -EFAULT;
4023 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4024 			goto out;
4025 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4026 		break;
4027 	}
4028 	case KVM_TRANSLATE: {
4029 		struct kvm_translation tr;
4030 
4031 		r = -EFAULT;
4032 		if (copy_from_user(&tr, argp, sizeof(tr)))
4033 			goto out;
4034 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4035 		if (r)
4036 			goto out;
4037 		r = -EFAULT;
4038 		if (copy_to_user(argp, &tr, sizeof(tr)))
4039 			goto out;
4040 		r = 0;
4041 		break;
4042 	}
4043 	case KVM_SET_GUEST_DEBUG: {
4044 		struct kvm_guest_debug dbg;
4045 
4046 		r = -EFAULT;
4047 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
4048 			goto out;
4049 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4050 		break;
4051 	}
4052 	case KVM_SET_SIGNAL_MASK: {
4053 		struct kvm_signal_mask __user *sigmask_arg = argp;
4054 		struct kvm_signal_mask kvm_sigmask;
4055 		sigset_t sigset, *p;
4056 
4057 		p = NULL;
4058 		if (argp) {
4059 			r = -EFAULT;
4060 			if (copy_from_user(&kvm_sigmask, argp,
4061 					   sizeof(kvm_sigmask)))
4062 				goto out;
4063 			r = -EINVAL;
4064 			if (kvm_sigmask.len != sizeof(sigset))
4065 				goto out;
4066 			r = -EFAULT;
4067 			if (copy_from_user(&sigset, sigmask_arg->sigset,
4068 					   sizeof(sigset)))
4069 				goto out;
4070 			p = &sigset;
4071 		}
4072 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4073 		break;
4074 	}
4075 	case KVM_GET_FPU: {
4076 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4077 		r = -ENOMEM;
4078 		if (!fpu)
4079 			goto out;
4080 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4081 		if (r)
4082 			goto out;
4083 		r = -EFAULT;
4084 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4085 			goto out;
4086 		r = 0;
4087 		break;
4088 	}
4089 	case KVM_SET_FPU: {
4090 		fpu = memdup_user(argp, sizeof(*fpu));
4091 		if (IS_ERR(fpu)) {
4092 			r = PTR_ERR(fpu);
4093 			fpu = NULL;
4094 			goto out;
4095 		}
4096 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4097 		break;
4098 	}
4099 	case KVM_GET_STATS_FD: {
4100 		r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4101 		break;
4102 	}
4103 	default:
4104 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4105 	}
4106 out:
4107 	mutex_unlock(&vcpu->mutex);
4108 	kfree(fpu);
4109 	kfree(kvm_sregs);
4110 	return r;
4111 }
4112 
4113 #ifdef CONFIG_KVM_COMPAT
4114 static long kvm_vcpu_compat_ioctl(struct file *filp,
4115 				  unsigned int ioctl, unsigned long arg)
4116 {
4117 	struct kvm_vcpu *vcpu = filp->private_data;
4118 	void __user *argp = compat_ptr(arg);
4119 	int r;
4120 
4121 	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4122 		return -EIO;
4123 
4124 	switch (ioctl) {
4125 	case KVM_SET_SIGNAL_MASK: {
4126 		struct kvm_signal_mask __user *sigmask_arg = argp;
4127 		struct kvm_signal_mask kvm_sigmask;
4128 		sigset_t sigset;
4129 
4130 		if (argp) {
4131 			r = -EFAULT;
4132 			if (copy_from_user(&kvm_sigmask, argp,
4133 					   sizeof(kvm_sigmask)))
4134 				goto out;
4135 			r = -EINVAL;
4136 			if (kvm_sigmask.len != sizeof(compat_sigset_t))
4137 				goto out;
4138 			r = -EFAULT;
4139 			if (get_compat_sigset(&sigset,
4140 					      (compat_sigset_t __user *)sigmask_arg->sigset))
4141 				goto out;
4142 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4143 		} else
4144 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4145 		break;
4146 	}
4147 	default:
4148 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
4149 	}
4150 
4151 out:
4152 	return r;
4153 }
4154 #endif
4155 
4156 static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4157 {
4158 	struct kvm_device *dev = filp->private_data;
4159 
4160 	if (dev->ops->mmap)
4161 		return dev->ops->mmap(dev, vma);
4162 
4163 	return -ENODEV;
4164 }
4165 
4166 static int kvm_device_ioctl_attr(struct kvm_device *dev,
4167 				 int (*accessor)(struct kvm_device *dev,
4168 						 struct kvm_device_attr *attr),
4169 				 unsigned long arg)
4170 {
4171 	struct kvm_device_attr attr;
4172 
4173 	if (!accessor)
4174 		return -EPERM;
4175 
4176 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4177 		return -EFAULT;
4178 
4179 	return accessor(dev, &attr);
4180 }
4181 
4182 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4183 			     unsigned long arg)
4184 {
4185 	struct kvm_device *dev = filp->private_data;
4186 
4187 	if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4188 		return -EIO;
4189 
4190 	switch (ioctl) {
4191 	case KVM_SET_DEVICE_ATTR:
4192 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4193 	case KVM_GET_DEVICE_ATTR:
4194 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4195 	case KVM_HAS_DEVICE_ATTR:
4196 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4197 	default:
4198 		if (dev->ops->ioctl)
4199 			return dev->ops->ioctl(dev, ioctl, arg);
4200 
4201 		return -ENOTTY;
4202 	}
4203 }
4204 
4205 static int kvm_device_release(struct inode *inode, struct file *filp)
4206 {
4207 	struct kvm_device *dev = filp->private_data;
4208 	struct kvm *kvm = dev->kvm;
4209 
4210 	if (dev->ops->release) {
4211 		mutex_lock(&kvm->lock);
4212 		list_del(&dev->vm_node);
4213 		dev->ops->release(dev);
4214 		mutex_unlock(&kvm->lock);
4215 	}
4216 
4217 	kvm_put_kvm(kvm);
4218 	return 0;
4219 }
4220 
4221 static const struct file_operations kvm_device_fops = {
4222 	.unlocked_ioctl = kvm_device_ioctl,
4223 	.release = kvm_device_release,
4224 	KVM_COMPAT(kvm_device_ioctl),
4225 	.mmap = kvm_device_mmap,
4226 };
4227 
4228 struct kvm_device *kvm_device_from_filp(struct file *filp)
4229 {
4230 	if (filp->f_op != &kvm_device_fops)
4231 		return NULL;
4232 
4233 	return filp->private_data;
4234 }
4235 
4236 static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4237 #ifdef CONFIG_KVM_MPIC
4238 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
4239 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
4240 #endif
4241 };
4242 
4243 int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4244 {
4245 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
4246 		return -ENOSPC;
4247 
4248 	if (kvm_device_ops_table[type] != NULL)
4249 		return -EEXIST;
4250 
4251 	kvm_device_ops_table[type] = ops;
4252 	return 0;
4253 }
4254 
4255 void kvm_unregister_device_ops(u32 type)
4256 {
4257 	if (kvm_device_ops_table[type] != NULL)
4258 		kvm_device_ops_table[type] = NULL;
4259 }
4260 
4261 static int kvm_ioctl_create_device(struct kvm *kvm,
4262 				   struct kvm_create_device *cd)
4263 {
4264 	const struct kvm_device_ops *ops = NULL;
4265 	struct kvm_device *dev;
4266 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4267 	int type;
4268 	int ret;
4269 
4270 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4271 		return -ENODEV;
4272 
4273 	type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4274 	ops = kvm_device_ops_table[type];
4275 	if (ops == NULL)
4276 		return -ENODEV;
4277 
4278 	if (test)
4279 		return 0;
4280 
4281 	dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4282 	if (!dev)
4283 		return -ENOMEM;
4284 
4285 	dev->ops = ops;
4286 	dev->kvm = kvm;
4287 
4288 	mutex_lock(&kvm->lock);
4289 	ret = ops->create(dev, type);
4290 	if (ret < 0) {
4291 		mutex_unlock(&kvm->lock);
4292 		kfree(dev);
4293 		return ret;
4294 	}
4295 	list_add(&dev->vm_node, &kvm->devices);
4296 	mutex_unlock(&kvm->lock);
4297 
4298 	if (ops->init)
4299 		ops->init(dev);
4300 
4301 	kvm_get_kvm(kvm);
4302 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4303 	if (ret < 0) {
4304 		kvm_put_kvm_no_destroy(kvm);
4305 		mutex_lock(&kvm->lock);
4306 		list_del(&dev->vm_node);
4307 		if (ops->release)
4308 			ops->release(dev);
4309 		mutex_unlock(&kvm->lock);
4310 		if (ops->destroy)
4311 			ops->destroy(dev);
4312 		return ret;
4313 	}
4314 
4315 	cd->fd = ret;
4316 	return 0;
4317 }
4318 
4319 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4320 {
4321 	switch (arg) {
4322 	case KVM_CAP_USER_MEMORY:
4323 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4324 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4325 	case KVM_CAP_INTERNAL_ERROR_DATA:
4326 #ifdef CONFIG_HAVE_KVM_MSI
4327 	case KVM_CAP_SIGNAL_MSI:
4328 #endif
4329 #ifdef CONFIG_HAVE_KVM_IRQFD
4330 	case KVM_CAP_IRQFD:
4331 	case KVM_CAP_IRQFD_RESAMPLE:
4332 #endif
4333 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4334 	case KVM_CAP_CHECK_EXTENSION_VM:
4335 	case KVM_CAP_ENABLE_CAP_VM:
4336 	case KVM_CAP_HALT_POLL:
4337 		return 1;
4338 #ifdef CONFIG_KVM_MMIO
4339 	case KVM_CAP_COALESCED_MMIO:
4340 		return KVM_COALESCED_MMIO_PAGE_OFFSET;
4341 	case KVM_CAP_COALESCED_PIO:
4342 		return 1;
4343 #endif
4344 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4345 	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4346 		return KVM_DIRTY_LOG_MANUAL_CAPS;
4347 #endif
4348 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4349 	case KVM_CAP_IRQ_ROUTING:
4350 		return KVM_MAX_IRQ_ROUTES;
4351 #endif
4352 #if KVM_ADDRESS_SPACE_NUM > 1
4353 	case KVM_CAP_MULTI_ADDRESS_SPACE:
4354 		return KVM_ADDRESS_SPACE_NUM;
4355 #endif
4356 	case KVM_CAP_NR_MEMSLOTS:
4357 		return KVM_USER_MEM_SLOTS;
4358 	case KVM_CAP_DIRTY_LOG_RING:
4359 #ifdef CONFIG_HAVE_KVM_DIRTY_RING
4360 		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4361 #else
4362 		return 0;
4363 #endif
4364 	case KVM_CAP_BINARY_STATS_FD:
4365 	case KVM_CAP_SYSTEM_EVENT_DATA:
4366 		return 1;
4367 	default:
4368 		break;
4369 	}
4370 	return kvm_vm_ioctl_check_extension(kvm, arg);
4371 }
4372 
4373 static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4374 {
4375 	int r;
4376 
4377 	if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4378 		return -EINVAL;
4379 
4380 	/* the size should be power of 2 */
4381 	if (!size || (size & (size - 1)))
4382 		return -EINVAL;
4383 
4384 	/* Should be bigger to keep the reserved entries, or a page */
4385 	if (size < kvm_dirty_ring_get_rsvd_entries() *
4386 	    sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4387 		return -EINVAL;
4388 
4389 	if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4390 	    sizeof(struct kvm_dirty_gfn))
4391 		return -E2BIG;
4392 
4393 	/* We only allow it to set once */
4394 	if (kvm->dirty_ring_size)
4395 		return -EINVAL;
4396 
4397 	mutex_lock(&kvm->lock);
4398 
4399 	if (kvm->created_vcpus) {
4400 		/* We don't allow to change this value after vcpu created */
4401 		r = -EINVAL;
4402 	} else {
4403 		kvm->dirty_ring_size = size;
4404 		r = 0;
4405 	}
4406 
4407 	mutex_unlock(&kvm->lock);
4408 	return r;
4409 }
4410 
4411 static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4412 {
4413 	unsigned long i;
4414 	struct kvm_vcpu *vcpu;
4415 	int cleared = 0;
4416 
4417 	if (!kvm->dirty_ring_size)
4418 		return -EINVAL;
4419 
4420 	mutex_lock(&kvm->slots_lock);
4421 
4422 	kvm_for_each_vcpu(i, vcpu, kvm)
4423 		cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4424 
4425 	mutex_unlock(&kvm->slots_lock);
4426 
4427 	if (cleared)
4428 		kvm_flush_remote_tlbs(kvm);
4429 
4430 	return cleared;
4431 }
4432 
4433 int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4434 						  struct kvm_enable_cap *cap)
4435 {
4436 	return -EINVAL;
4437 }
4438 
4439 static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4440 					   struct kvm_enable_cap *cap)
4441 {
4442 	switch (cap->cap) {
4443 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4444 	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4445 		u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4446 
4447 		if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4448 			allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4449 
4450 		if (cap->flags || (cap->args[0] & ~allowed_options))
4451 			return -EINVAL;
4452 		kvm->manual_dirty_log_protect = cap->args[0];
4453 		return 0;
4454 	}
4455 #endif
4456 	case KVM_CAP_HALT_POLL: {
4457 		if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4458 			return -EINVAL;
4459 
4460 		kvm->max_halt_poll_ns = cap->args[0];
4461 		return 0;
4462 	}
4463 	case KVM_CAP_DIRTY_LOG_RING:
4464 		return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4465 	default:
4466 		return kvm_vm_ioctl_enable_cap(kvm, cap);
4467 	}
4468 }
4469 
4470 static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4471 			      size_t size, loff_t *offset)
4472 {
4473 	struct kvm *kvm = file->private_data;
4474 
4475 	return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4476 				&kvm_vm_stats_desc[0], &kvm->stat,
4477 				sizeof(kvm->stat), user_buffer, size, offset);
4478 }
4479 
4480 static const struct file_operations kvm_vm_stats_fops = {
4481 	.read = kvm_vm_stats_read,
4482 	.llseek = noop_llseek,
4483 };
4484 
4485 static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4486 {
4487 	int fd;
4488 	struct file *file;
4489 
4490 	fd = get_unused_fd_flags(O_CLOEXEC);
4491 	if (fd < 0)
4492 		return fd;
4493 
4494 	file = anon_inode_getfile("kvm-vm-stats",
4495 			&kvm_vm_stats_fops, kvm, O_RDONLY);
4496 	if (IS_ERR(file)) {
4497 		put_unused_fd(fd);
4498 		return PTR_ERR(file);
4499 	}
4500 	file->f_mode |= FMODE_PREAD;
4501 	fd_install(fd, file);
4502 
4503 	return fd;
4504 }
4505 
4506 static long kvm_vm_ioctl(struct file *filp,
4507 			   unsigned int ioctl, unsigned long arg)
4508 {
4509 	struct kvm *kvm = filp->private_data;
4510 	void __user *argp = (void __user *)arg;
4511 	int r;
4512 
4513 	if (kvm->mm != current->mm || kvm->vm_dead)
4514 		return -EIO;
4515 	switch (ioctl) {
4516 	case KVM_CREATE_VCPU:
4517 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4518 		break;
4519 	case KVM_ENABLE_CAP: {
4520 		struct kvm_enable_cap cap;
4521 
4522 		r = -EFAULT;
4523 		if (copy_from_user(&cap, argp, sizeof(cap)))
4524 			goto out;
4525 		r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4526 		break;
4527 	}
4528 	case KVM_SET_USER_MEMORY_REGION: {
4529 		struct kvm_userspace_memory_region kvm_userspace_mem;
4530 
4531 		r = -EFAULT;
4532 		if (copy_from_user(&kvm_userspace_mem, argp,
4533 						sizeof(kvm_userspace_mem)))
4534 			goto out;
4535 
4536 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4537 		break;
4538 	}
4539 	case KVM_GET_DIRTY_LOG: {
4540 		struct kvm_dirty_log log;
4541 
4542 		r = -EFAULT;
4543 		if (copy_from_user(&log, argp, sizeof(log)))
4544 			goto out;
4545 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4546 		break;
4547 	}
4548 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4549 	case KVM_CLEAR_DIRTY_LOG: {
4550 		struct kvm_clear_dirty_log log;
4551 
4552 		r = -EFAULT;
4553 		if (copy_from_user(&log, argp, sizeof(log)))
4554 			goto out;
4555 		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4556 		break;
4557 	}
4558 #endif
4559 #ifdef CONFIG_KVM_MMIO
4560 	case KVM_REGISTER_COALESCED_MMIO: {
4561 		struct kvm_coalesced_mmio_zone zone;
4562 
4563 		r = -EFAULT;
4564 		if (copy_from_user(&zone, argp, sizeof(zone)))
4565 			goto out;
4566 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4567 		break;
4568 	}
4569 	case KVM_UNREGISTER_COALESCED_MMIO: {
4570 		struct kvm_coalesced_mmio_zone zone;
4571 
4572 		r = -EFAULT;
4573 		if (copy_from_user(&zone, argp, sizeof(zone)))
4574 			goto out;
4575 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4576 		break;
4577 	}
4578 #endif
4579 	case KVM_IRQFD: {
4580 		struct kvm_irqfd data;
4581 
4582 		r = -EFAULT;
4583 		if (copy_from_user(&data, argp, sizeof(data)))
4584 			goto out;
4585 		r = kvm_irqfd(kvm, &data);
4586 		break;
4587 	}
4588 	case KVM_IOEVENTFD: {
4589 		struct kvm_ioeventfd data;
4590 
4591 		r = -EFAULT;
4592 		if (copy_from_user(&data, argp, sizeof(data)))
4593 			goto out;
4594 		r = kvm_ioeventfd(kvm, &data);
4595 		break;
4596 	}
4597 #ifdef CONFIG_HAVE_KVM_MSI
4598 	case KVM_SIGNAL_MSI: {
4599 		struct kvm_msi msi;
4600 
4601 		r = -EFAULT;
4602 		if (copy_from_user(&msi, argp, sizeof(msi)))
4603 			goto out;
4604 		r = kvm_send_userspace_msi(kvm, &msi);
4605 		break;
4606 	}
4607 #endif
4608 #ifdef __KVM_HAVE_IRQ_LINE
4609 	case KVM_IRQ_LINE_STATUS:
4610 	case KVM_IRQ_LINE: {
4611 		struct kvm_irq_level irq_event;
4612 
4613 		r = -EFAULT;
4614 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4615 			goto out;
4616 
4617 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4618 					ioctl == KVM_IRQ_LINE_STATUS);
4619 		if (r)
4620 			goto out;
4621 
4622 		r = -EFAULT;
4623 		if (ioctl == KVM_IRQ_LINE_STATUS) {
4624 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4625 				goto out;
4626 		}
4627 
4628 		r = 0;
4629 		break;
4630 	}
4631 #endif
4632 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4633 	case KVM_SET_GSI_ROUTING: {
4634 		struct kvm_irq_routing routing;
4635 		struct kvm_irq_routing __user *urouting;
4636 		struct kvm_irq_routing_entry *entries = NULL;
4637 
4638 		r = -EFAULT;
4639 		if (copy_from_user(&routing, argp, sizeof(routing)))
4640 			goto out;
4641 		r = -EINVAL;
4642 		if (!kvm_arch_can_set_irq_routing(kvm))
4643 			goto out;
4644 		if (routing.nr > KVM_MAX_IRQ_ROUTES)
4645 			goto out;
4646 		if (routing.flags)
4647 			goto out;
4648 		if (routing.nr) {
4649 			urouting = argp;
4650 			entries = vmemdup_user(urouting->entries,
4651 					       array_size(sizeof(*entries),
4652 							  routing.nr));
4653 			if (IS_ERR(entries)) {
4654 				r = PTR_ERR(entries);
4655 				goto out;
4656 			}
4657 		}
4658 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
4659 					routing.flags);
4660 		kvfree(entries);
4661 		break;
4662 	}
4663 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
4664 	case KVM_CREATE_DEVICE: {
4665 		struct kvm_create_device cd;
4666 
4667 		r = -EFAULT;
4668 		if (copy_from_user(&cd, argp, sizeof(cd)))
4669 			goto out;
4670 
4671 		r = kvm_ioctl_create_device(kvm, &cd);
4672 		if (r)
4673 			goto out;
4674 
4675 		r = -EFAULT;
4676 		if (copy_to_user(argp, &cd, sizeof(cd)))
4677 			goto out;
4678 
4679 		r = 0;
4680 		break;
4681 	}
4682 	case KVM_CHECK_EXTENSION:
4683 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4684 		break;
4685 	case KVM_RESET_DIRTY_RINGS:
4686 		r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4687 		break;
4688 	case KVM_GET_STATS_FD:
4689 		r = kvm_vm_ioctl_get_stats_fd(kvm);
4690 		break;
4691 	default:
4692 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4693 	}
4694 out:
4695 	return r;
4696 }
4697 
4698 #ifdef CONFIG_KVM_COMPAT
4699 struct compat_kvm_dirty_log {
4700 	__u32 slot;
4701 	__u32 padding1;
4702 	union {
4703 		compat_uptr_t dirty_bitmap; /* one bit per page */
4704 		__u64 padding2;
4705 	};
4706 };
4707 
4708 struct compat_kvm_clear_dirty_log {
4709 	__u32 slot;
4710 	__u32 num_pages;
4711 	__u64 first_page;
4712 	union {
4713 		compat_uptr_t dirty_bitmap; /* one bit per page */
4714 		__u64 padding2;
4715 	};
4716 };
4717 
4718 static long kvm_vm_compat_ioctl(struct file *filp,
4719 			   unsigned int ioctl, unsigned long arg)
4720 {
4721 	struct kvm *kvm = filp->private_data;
4722 	int r;
4723 
4724 	if (kvm->mm != current->mm || kvm->vm_dead)
4725 		return -EIO;
4726 	switch (ioctl) {
4727 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4728 	case KVM_CLEAR_DIRTY_LOG: {
4729 		struct compat_kvm_clear_dirty_log compat_log;
4730 		struct kvm_clear_dirty_log log;
4731 
4732 		if (copy_from_user(&compat_log, (void __user *)arg,
4733 				   sizeof(compat_log)))
4734 			return -EFAULT;
4735 		log.slot	 = compat_log.slot;
4736 		log.num_pages	 = compat_log.num_pages;
4737 		log.first_page	 = compat_log.first_page;
4738 		log.padding2	 = compat_log.padding2;
4739 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4740 
4741 		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4742 		break;
4743 	}
4744 #endif
4745 	case KVM_GET_DIRTY_LOG: {
4746 		struct compat_kvm_dirty_log compat_log;
4747 		struct kvm_dirty_log log;
4748 
4749 		if (copy_from_user(&compat_log, (void __user *)arg,
4750 				   sizeof(compat_log)))
4751 			return -EFAULT;
4752 		log.slot	 = compat_log.slot;
4753 		log.padding1	 = compat_log.padding1;
4754 		log.padding2	 = compat_log.padding2;
4755 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4756 
4757 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4758 		break;
4759 	}
4760 	default:
4761 		r = kvm_vm_ioctl(filp, ioctl, arg);
4762 	}
4763 	return r;
4764 }
4765 #endif
4766 
4767 static const struct file_operations kvm_vm_fops = {
4768 	.release        = kvm_vm_release,
4769 	.unlocked_ioctl = kvm_vm_ioctl,
4770 	.llseek		= noop_llseek,
4771 	KVM_COMPAT(kvm_vm_compat_ioctl),
4772 };
4773 
4774 bool file_is_kvm(struct file *file)
4775 {
4776 	return file && file->f_op == &kvm_vm_fops;
4777 }
4778 EXPORT_SYMBOL_GPL(file_is_kvm);
4779 
4780 static int kvm_dev_ioctl_create_vm(unsigned long type)
4781 {
4782 	int r;
4783 	struct kvm *kvm;
4784 	struct file *file;
4785 
4786 	kvm = kvm_create_vm(type);
4787 	if (IS_ERR(kvm))
4788 		return PTR_ERR(kvm);
4789 #ifdef CONFIG_KVM_MMIO
4790 	r = kvm_coalesced_mmio_init(kvm);
4791 	if (r < 0)
4792 		goto put_kvm;
4793 #endif
4794 	r = get_unused_fd_flags(O_CLOEXEC);
4795 	if (r < 0)
4796 		goto put_kvm;
4797 
4798 	snprintf(kvm->stats_id, sizeof(kvm->stats_id),
4799 			"kvm-%d", task_pid_nr(current));
4800 
4801 	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
4802 	if (IS_ERR(file)) {
4803 		put_unused_fd(r);
4804 		r = PTR_ERR(file);
4805 		goto put_kvm;
4806 	}
4807 
4808 	/*
4809 	 * Don't call kvm_put_kvm anymore at this point; file->f_op is
4810 	 * already set, with ->release() being kvm_vm_release().  In error
4811 	 * cases it will be called by the final fput(file) and will take
4812 	 * care of doing kvm_put_kvm(kvm).
4813 	 */
4814 	if (kvm_create_vm_debugfs(kvm, r) < 0) {
4815 		put_unused_fd(r);
4816 		fput(file);
4817 		return -ENOMEM;
4818 	}
4819 	kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
4820 
4821 	fd_install(r, file);
4822 	return r;
4823 
4824 put_kvm:
4825 	kvm_put_kvm(kvm);
4826 	return r;
4827 }
4828 
4829 static long kvm_dev_ioctl(struct file *filp,
4830 			  unsigned int ioctl, unsigned long arg)
4831 {
4832 	long r = -EINVAL;
4833 
4834 	switch (ioctl) {
4835 	case KVM_GET_API_VERSION:
4836 		if (arg)
4837 			goto out;
4838 		r = KVM_API_VERSION;
4839 		break;
4840 	case KVM_CREATE_VM:
4841 		r = kvm_dev_ioctl_create_vm(arg);
4842 		break;
4843 	case KVM_CHECK_EXTENSION:
4844 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
4845 		break;
4846 	case KVM_GET_VCPU_MMAP_SIZE:
4847 		if (arg)
4848 			goto out;
4849 		r = PAGE_SIZE;     /* struct kvm_run */
4850 #ifdef CONFIG_X86
4851 		r += PAGE_SIZE;    /* pio data page */
4852 #endif
4853 #ifdef CONFIG_KVM_MMIO
4854 		r += PAGE_SIZE;    /* coalesced mmio ring page */
4855 #endif
4856 		break;
4857 	case KVM_TRACE_ENABLE:
4858 	case KVM_TRACE_PAUSE:
4859 	case KVM_TRACE_DISABLE:
4860 		r = -EOPNOTSUPP;
4861 		break;
4862 	default:
4863 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
4864 	}
4865 out:
4866 	return r;
4867 }
4868 
4869 static struct file_operations kvm_chardev_ops = {
4870 	.unlocked_ioctl = kvm_dev_ioctl,
4871 	.llseek		= noop_llseek,
4872 	KVM_COMPAT(kvm_dev_ioctl),
4873 };
4874 
4875 static struct miscdevice kvm_dev = {
4876 	KVM_MINOR,
4877 	"kvm",
4878 	&kvm_chardev_ops,
4879 };
4880 
4881 static void hardware_enable_nolock(void *junk)
4882 {
4883 	int cpu = raw_smp_processor_id();
4884 	int r;
4885 
4886 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
4887 		return;
4888 
4889 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
4890 
4891 	r = kvm_arch_hardware_enable();
4892 
4893 	if (r) {
4894 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4895 		atomic_inc(&hardware_enable_failed);
4896 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
4897 	}
4898 }
4899 
4900 static int kvm_starting_cpu(unsigned int cpu)
4901 {
4902 	raw_spin_lock(&kvm_count_lock);
4903 	if (kvm_usage_count)
4904 		hardware_enable_nolock(NULL);
4905 	raw_spin_unlock(&kvm_count_lock);
4906 	return 0;
4907 }
4908 
4909 static void hardware_disable_nolock(void *junk)
4910 {
4911 	int cpu = raw_smp_processor_id();
4912 
4913 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
4914 		return;
4915 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
4916 	kvm_arch_hardware_disable();
4917 }
4918 
4919 static int kvm_dying_cpu(unsigned int cpu)
4920 {
4921 	raw_spin_lock(&kvm_count_lock);
4922 	if (kvm_usage_count)
4923 		hardware_disable_nolock(NULL);
4924 	raw_spin_unlock(&kvm_count_lock);
4925 	return 0;
4926 }
4927 
4928 static void hardware_disable_all_nolock(void)
4929 {
4930 	BUG_ON(!kvm_usage_count);
4931 
4932 	kvm_usage_count--;
4933 	if (!kvm_usage_count)
4934 		on_each_cpu(hardware_disable_nolock, NULL, 1);
4935 }
4936 
4937 static void hardware_disable_all(void)
4938 {
4939 	raw_spin_lock(&kvm_count_lock);
4940 	hardware_disable_all_nolock();
4941 	raw_spin_unlock(&kvm_count_lock);
4942 }
4943 
4944 static int hardware_enable_all(void)
4945 {
4946 	int r = 0;
4947 
4948 	raw_spin_lock(&kvm_count_lock);
4949 
4950 	kvm_usage_count++;
4951 	if (kvm_usage_count == 1) {
4952 		atomic_set(&hardware_enable_failed, 0);
4953 		on_each_cpu(hardware_enable_nolock, NULL, 1);
4954 
4955 		if (atomic_read(&hardware_enable_failed)) {
4956 			hardware_disable_all_nolock();
4957 			r = -EBUSY;
4958 		}
4959 	}
4960 
4961 	raw_spin_unlock(&kvm_count_lock);
4962 
4963 	return r;
4964 }
4965 
4966 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
4967 		      void *v)
4968 {
4969 	/*
4970 	 * Some (well, at least mine) BIOSes hang on reboot if
4971 	 * in vmx root mode.
4972 	 *
4973 	 * And Intel TXT required VMX off for all cpu when system shutdown.
4974 	 */
4975 	pr_info("kvm: exiting hardware virtualization\n");
4976 	kvm_rebooting = true;
4977 	on_each_cpu(hardware_disable_nolock, NULL, 1);
4978 	return NOTIFY_OK;
4979 }
4980 
4981 static struct notifier_block kvm_reboot_notifier = {
4982 	.notifier_call = kvm_reboot,
4983 	.priority = 0,
4984 };
4985 
4986 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
4987 {
4988 	int i;
4989 
4990 	for (i = 0; i < bus->dev_count; i++) {
4991 		struct kvm_io_device *pos = bus->range[i].dev;
4992 
4993 		kvm_iodevice_destructor(pos);
4994 	}
4995 	kfree(bus);
4996 }
4997 
4998 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
4999 				 const struct kvm_io_range *r2)
5000 {
5001 	gpa_t addr1 = r1->addr;
5002 	gpa_t addr2 = r2->addr;
5003 
5004 	if (addr1 < addr2)
5005 		return -1;
5006 
5007 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
5008 	 * accept any overlapping write.  Any order is acceptable for
5009 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5010 	 * we process all of them.
5011 	 */
5012 	if (r2->len) {
5013 		addr1 += r1->len;
5014 		addr2 += r2->len;
5015 	}
5016 
5017 	if (addr1 > addr2)
5018 		return 1;
5019 
5020 	return 0;
5021 }
5022 
5023 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5024 {
5025 	return kvm_io_bus_cmp(p1, p2);
5026 }
5027 
5028 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5029 			     gpa_t addr, int len)
5030 {
5031 	struct kvm_io_range *range, key;
5032 	int off;
5033 
5034 	key = (struct kvm_io_range) {
5035 		.addr = addr,
5036 		.len = len,
5037 	};
5038 
5039 	range = bsearch(&key, bus->range, bus->dev_count,
5040 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5041 	if (range == NULL)
5042 		return -ENOENT;
5043 
5044 	off = range - bus->range;
5045 
5046 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5047 		off--;
5048 
5049 	return off;
5050 }
5051 
5052 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5053 			      struct kvm_io_range *range, const void *val)
5054 {
5055 	int idx;
5056 
5057 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5058 	if (idx < 0)
5059 		return -EOPNOTSUPP;
5060 
5061 	while (idx < bus->dev_count &&
5062 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5063 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5064 					range->len, val))
5065 			return idx;
5066 		idx++;
5067 	}
5068 
5069 	return -EOPNOTSUPP;
5070 }
5071 
5072 /* kvm_io_bus_write - called under kvm->slots_lock */
5073 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5074 		     int len, const void *val)
5075 {
5076 	struct kvm_io_bus *bus;
5077 	struct kvm_io_range range;
5078 	int r;
5079 
5080 	range = (struct kvm_io_range) {
5081 		.addr = addr,
5082 		.len = len,
5083 	};
5084 
5085 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5086 	if (!bus)
5087 		return -ENOMEM;
5088 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
5089 	return r < 0 ? r : 0;
5090 }
5091 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5092 
5093 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5094 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5095 			    gpa_t addr, int len, const void *val, long cookie)
5096 {
5097 	struct kvm_io_bus *bus;
5098 	struct kvm_io_range range;
5099 
5100 	range = (struct kvm_io_range) {
5101 		.addr = addr,
5102 		.len = len,
5103 	};
5104 
5105 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5106 	if (!bus)
5107 		return -ENOMEM;
5108 
5109 	/* First try the device referenced by cookie. */
5110 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
5111 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5112 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5113 					val))
5114 			return cookie;
5115 
5116 	/*
5117 	 * cookie contained garbage; fall back to search and return the
5118 	 * correct cookie value.
5119 	 */
5120 	return __kvm_io_bus_write(vcpu, bus, &range, val);
5121 }
5122 
5123 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5124 			     struct kvm_io_range *range, void *val)
5125 {
5126 	int idx;
5127 
5128 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5129 	if (idx < 0)
5130 		return -EOPNOTSUPP;
5131 
5132 	while (idx < bus->dev_count &&
5133 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5134 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5135 				       range->len, val))
5136 			return idx;
5137 		idx++;
5138 	}
5139 
5140 	return -EOPNOTSUPP;
5141 }
5142 
5143 /* kvm_io_bus_read - called under kvm->slots_lock */
5144 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5145 		    int len, void *val)
5146 {
5147 	struct kvm_io_bus *bus;
5148 	struct kvm_io_range range;
5149 	int r;
5150 
5151 	range = (struct kvm_io_range) {
5152 		.addr = addr,
5153 		.len = len,
5154 	};
5155 
5156 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5157 	if (!bus)
5158 		return -ENOMEM;
5159 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
5160 	return r < 0 ? r : 0;
5161 }
5162 
5163 /* Caller must hold slots_lock. */
5164 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5165 			    int len, struct kvm_io_device *dev)
5166 {
5167 	int i;
5168 	struct kvm_io_bus *new_bus, *bus;
5169 	struct kvm_io_range range;
5170 
5171 	bus = kvm_get_bus(kvm, bus_idx);
5172 	if (!bus)
5173 		return -ENOMEM;
5174 
5175 	/* exclude ioeventfd which is limited by maximum fd */
5176 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5177 		return -ENOSPC;
5178 
5179 	new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5180 			  GFP_KERNEL_ACCOUNT);
5181 	if (!new_bus)
5182 		return -ENOMEM;
5183 
5184 	range = (struct kvm_io_range) {
5185 		.addr = addr,
5186 		.len = len,
5187 		.dev = dev,
5188 	};
5189 
5190 	for (i = 0; i < bus->dev_count; i++)
5191 		if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5192 			break;
5193 
5194 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5195 	new_bus->dev_count++;
5196 	new_bus->range[i] = range;
5197 	memcpy(new_bus->range + i + 1, bus->range + i,
5198 		(bus->dev_count - i) * sizeof(struct kvm_io_range));
5199 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5200 	synchronize_srcu_expedited(&kvm->srcu);
5201 	kfree(bus);
5202 
5203 	return 0;
5204 }
5205 
5206 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5207 			      struct kvm_io_device *dev)
5208 {
5209 	int i, j;
5210 	struct kvm_io_bus *new_bus, *bus;
5211 
5212 	lockdep_assert_held(&kvm->slots_lock);
5213 
5214 	bus = kvm_get_bus(kvm, bus_idx);
5215 	if (!bus)
5216 		return 0;
5217 
5218 	for (i = 0; i < bus->dev_count; i++) {
5219 		if (bus->range[i].dev == dev) {
5220 			break;
5221 		}
5222 	}
5223 
5224 	if (i == bus->dev_count)
5225 		return 0;
5226 
5227 	new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5228 			  GFP_KERNEL_ACCOUNT);
5229 	if (new_bus) {
5230 		memcpy(new_bus, bus, struct_size(bus, range, i));
5231 		new_bus->dev_count--;
5232 		memcpy(new_bus->range + i, bus->range + i + 1,
5233 				flex_array_size(new_bus, range, new_bus->dev_count - i));
5234 	}
5235 
5236 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5237 	synchronize_srcu_expedited(&kvm->srcu);
5238 
5239 	/* Destroy the old bus _after_ installing the (null) bus. */
5240 	if (!new_bus) {
5241 		pr_err("kvm: failed to shrink bus, removing it completely\n");
5242 		for (j = 0; j < bus->dev_count; j++) {
5243 			if (j == i)
5244 				continue;
5245 			kvm_iodevice_destructor(bus->range[j].dev);
5246 		}
5247 	}
5248 
5249 	kfree(bus);
5250 	return new_bus ? 0 : -ENOMEM;
5251 }
5252 
5253 struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5254 					 gpa_t addr)
5255 {
5256 	struct kvm_io_bus *bus;
5257 	int dev_idx, srcu_idx;
5258 	struct kvm_io_device *iodev = NULL;
5259 
5260 	srcu_idx = srcu_read_lock(&kvm->srcu);
5261 
5262 	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5263 	if (!bus)
5264 		goto out_unlock;
5265 
5266 	dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5267 	if (dev_idx < 0)
5268 		goto out_unlock;
5269 
5270 	iodev = bus->range[dev_idx].dev;
5271 
5272 out_unlock:
5273 	srcu_read_unlock(&kvm->srcu, srcu_idx);
5274 
5275 	return iodev;
5276 }
5277 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5278 
5279 static int kvm_debugfs_open(struct inode *inode, struct file *file,
5280 			   int (*get)(void *, u64 *), int (*set)(void *, u64),
5281 			   const char *fmt)
5282 {
5283 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5284 					  inode->i_private;
5285 
5286 	/*
5287 	 * The debugfs files are a reference to the kvm struct which
5288         * is still valid when kvm_destroy_vm is called.  kvm_get_kvm_safe
5289         * avoids the race between open and the removal of the debugfs directory.
5290 	 */
5291 	if (!kvm_get_kvm_safe(stat_data->kvm))
5292 		return -ENOENT;
5293 
5294 	if (simple_attr_open(inode, file, get,
5295 		    kvm_stats_debugfs_mode(stat_data->desc) & 0222
5296 		    ? set : NULL,
5297 		    fmt)) {
5298 		kvm_put_kvm(stat_data->kvm);
5299 		return -ENOMEM;
5300 	}
5301 
5302 	return 0;
5303 }
5304 
5305 static int kvm_debugfs_release(struct inode *inode, struct file *file)
5306 {
5307 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5308 					  inode->i_private;
5309 
5310 	simple_attr_release(inode, file);
5311 	kvm_put_kvm(stat_data->kvm);
5312 
5313 	return 0;
5314 }
5315 
5316 static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5317 {
5318 	*val = *(u64 *)((void *)(&kvm->stat) + offset);
5319 
5320 	return 0;
5321 }
5322 
5323 static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5324 {
5325 	*(u64 *)((void *)(&kvm->stat) + offset) = 0;
5326 
5327 	return 0;
5328 }
5329 
5330 static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5331 {
5332 	unsigned long i;
5333 	struct kvm_vcpu *vcpu;
5334 
5335 	*val = 0;
5336 
5337 	kvm_for_each_vcpu(i, vcpu, kvm)
5338 		*val += *(u64 *)((void *)(&vcpu->stat) + offset);
5339 
5340 	return 0;
5341 }
5342 
5343 static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5344 {
5345 	unsigned long i;
5346 	struct kvm_vcpu *vcpu;
5347 
5348 	kvm_for_each_vcpu(i, vcpu, kvm)
5349 		*(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5350 
5351 	return 0;
5352 }
5353 
5354 static int kvm_stat_data_get(void *data, u64 *val)
5355 {
5356 	int r = -EFAULT;
5357 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5358 
5359 	switch (stat_data->kind) {
5360 	case KVM_STAT_VM:
5361 		r = kvm_get_stat_per_vm(stat_data->kvm,
5362 					stat_data->desc->desc.offset, val);
5363 		break;
5364 	case KVM_STAT_VCPU:
5365 		r = kvm_get_stat_per_vcpu(stat_data->kvm,
5366 					  stat_data->desc->desc.offset, val);
5367 		break;
5368 	}
5369 
5370 	return r;
5371 }
5372 
5373 static int kvm_stat_data_clear(void *data, u64 val)
5374 {
5375 	int r = -EFAULT;
5376 	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5377 
5378 	if (val)
5379 		return -EINVAL;
5380 
5381 	switch (stat_data->kind) {
5382 	case KVM_STAT_VM:
5383 		r = kvm_clear_stat_per_vm(stat_data->kvm,
5384 					  stat_data->desc->desc.offset);
5385 		break;
5386 	case KVM_STAT_VCPU:
5387 		r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5388 					    stat_data->desc->desc.offset);
5389 		break;
5390 	}
5391 
5392 	return r;
5393 }
5394 
5395 static int kvm_stat_data_open(struct inode *inode, struct file *file)
5396 {
5397 	__simple_attr_check_format("%llu\n", 0ull);
5398 	return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5399 				kvm_stat_data_clear, "%llu\n");
5400 }
5401 
5402 static const struct file_operations stat_fops_per_vm = {
5403 	.owner = THIS_MODULE,
5404 	.open = kvm_stat_data_open,
5405 	.release = kvm_debugfs_release,
5406 	.read = simple_attr_read,
5407 	.write = simple_attr_write,
5408 	.llseek = no_llseek,
5409 };
5410 
5411 static int vm_stat_get(void *_offset, u64 *val)
5412 {
5413 	unsigned offset = (long)_offset;
5414 	struct kvm *kvm;
5415 	u64 tmp_val;
5416 
5417 	*val = 0;
5418 	mutex_lock(&kvm_lock);
5419 	list_for_each_entry(kvm, &vm_list, vm_list) {
5420 		kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5421 		*val += tmp_val;
5422 	}
5423 	mutex_unlock(&kvm_lock);
5424 	return 0;
5425 }
5426 
5427 static int vm_stat_clear(void *_offset, u64 val)
5428 {
5429 	unsigned offset = (long)_offset;
5430 	struct kvm *kvm;
5431 
5432 	if (val)
5433 		return -EINVAL;
5434 
5435 	mutex_lock(&kvm_lock);
5436 	list_for_each_entry(kvm, &vm_list, vm_list) {
5437 		kvm_clear_stat_per_vm(kvm, offset);
5438 	}
5439 	mutex_unlock(&kvm_lock);
5440 
5441 	return 0;
5442 }
5443 
5444 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5445 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5446 
5447 static int vcpu_stat_get(void *_offset, u64 *val)
5448 {
5449 	unsigned offset = (long)_offset;
5450 	struct kvm *kvm;
5451 	u64 tmp_val;
5452 
5453 	*val = 0;
5454 	mutex_lock(&kvm_lock);
5455 	list_for_each_entry(kvm, &vm_list, vm_list) {
5456 		kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5457 		*val += tmp_val;
5458 	}
5459 	mutex_unlock(&kvm_lock);
5460 	return 0;
5461 }
5462 
5463 static int vcpu_stat_clear(void *_offset, u64 val)
5464 {
5465 	unsigned offset = (long)_offset;
5466 	struct kvm *kvm;
5467 
5468 	if (val)
5469 		return -EINVAL;
5470 
5471 	mutex_lock(&kvm_lock);
5472 	list_for_each_entry(kvm, &vm_list, vm_list) {
5473 		kvm_clear_stat_per_vcpu(kvm, offset);
5474 	}
5475 	mutex_unlock(&kvm_lock);
5476 
5477 	return 0;
5478 }
5479 
5480 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5481 			"%llu\n");
5482 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5483 
5484 static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5485 {
5486 	struct kobj_uevent_env *env;
5487 	unsigned long long created, active;
5488 
5489 	if (!kvm_dev.this_device || !kvm)
5490 		return;
5491 
5492 	mutex_lock(&kvm_lock);
5493 	if (type == KVM_EVENT_CREATE_VM) {
5494 		kvm_createvm_count++;
5495 		kvm_active_vms++;
5496 	} else if (type == KVM_EVENT_DESTROY_VM) {
5497 		kvm_active_vms--;
5498 	}
5499 	created = kvm_createvm_count;
5500 	active = kvm_active_vms;
5501 	mutex_unlock(&kvm_lock);
5502 
5503 	env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5504 	if (!env)
5505 		return;
5506 
5507 	add_uevent_var(env, "CREATED=%llu", created);
5508 	add_uevent_var(env, "COUNT=%llu", active);
5509 
5510 	if (type == KVM_EVENT_CREATE_VM) {
5511 		add_uevent_var(env, "EVENT=create");
5512 		kvm->userspace_pid = task_pid_nr(current);
5513 	} else if (type == KVM_EVENT_DESTROY_VM) {
5514 		add_uevent_var(env, "EVENT=destroy");
5515 	}
5516 	add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5517 
5518 	if (!IS_ERR(kvm->debugfs_dentry)) {
5519 		char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5520 
5521 		if (p) {
5522 			tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5523 			if (!IS_ERR(tmp))
5524 				add_uevent_var(env, "STATS_PATH=%s", tmp);
5525 			kfree(p);
5526 		}
5527 	}
5528 	/* no need for checks, since we are adding at most only 5 keys */
5529 	env->envp[env->envp_idx++] = NULL;
5530 	kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5531 	kfree(env);
5532 }
5533 
5534 static void kvm_init_debug(void)
5535 {
5536 	const struct file_operations *fops;
5537 	const struct _kvm_stats_desc *pdesc;
5538 	int i;
5539 
5540 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5541 
5542 	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5543 		pdesc = &kvm_vm_stats_desc[i];
5544 		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5545 			fops = &vm_stat_fops;
5546 		else
5547 			fops = &vm_stat_readonly_fops;
5548 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5549 				kvm_debugfs_dir,
5550 				(void *)(long)pdesc->desc.offset, fops);
5551 	}
5552 
5553 	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5554 		pdesc = &kvm_vcpu_stats_desc[i];
5555 		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5556 			fops = &vcpu_stat_fops;
5557 		else
5558 			fops = &vcpu_stat_readonly_fops;
5559 		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5560 				kvm_debugfs_dir,
5561 				(void *)(long)pdesc->desc.offset, fops);
5562 	}
5563 }
5564 
5565 static int kvm_suspend(void)
5566 {
5567 	if (kvm_usage_count)
5568 		hardware_disable_nolock(NULL);
5569 	return 0;
5570 }
5571 
5572 static void kvm_resume(void)
5573 {
5574 	if (kvm_usage_count) {
5575 		lockdep_assert_not_held(&kvm_count_lock);
5576 		hardware_enable_nolock(NULL);
5577 	}
5578 }
5579 
5580 static struct syscore_ops kvm_syscore_ops = {
5581 	.suspend = kvm_suspend,
5582 	.resume = kvm_resume,
5583 };
5584 
5585 static inline
5586 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5587 {
5588 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
5589 }
5590 
5591 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5592 {
5593 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5594 
5595 	WRITE_ONCE(vcpu->preempted, false);
5596 	WRITE_ONCE(vcpu->ready, false);
5597 
5598 	__this_cpu_write(kvm_running_vcpu, vcpu);
5599 	kvm_arch_sched_in(vcpu, cpu);
5600 	kvm_arch_vcpu_load(vcpu, cpu);
5601 }
5602 
5603 static void kvm_sched_out(struct preempt_notifier *pn,
5604 			  struct task_struct *next)
5605 {
5606 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5607 
5608 	if (current->on_rq) {
5609 		WRITE_ONCE(vcpu->preempted, true);
5610 		WRITE_ONCE(vcpu->ready, true);
5611 	}
5612 	kvm_arch_vcpu_put(vcpu);
5613 	__this_cpu_write(kvm_running_vcpu, NULL);
5614 }
5615 
5616 /**
5617  * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5618  *
5619  * We can disable preemption locally around accessing the per-CPU variable,
5620  * and use the resolved vcpu pointer after enabling preemption again,
5621  * because even if the current thread is migrated to another CPU, reading
5622  * the per-CPU value later will give us the same value as we update the
5623  * per-CPU variable in the preempt notifier handlers.
5624  */
5625 struct kvm_vcpu *kvm_get_running_vcpu(void)
5626 {
5627 	struct kvm_vcpu *vcpu;
5628 
5629 	preempt_disable();
5630 	vcpu = __this_cpu_read(kvm_running_vcpu);
5631 	preempt_enable();
5632 
5633 	return vcpu;
5634 }
5635 EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5636 
5637 /**
5638  * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5639  */
5640 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5641 {
5642         return &kvm_running_vcpu;
5643 }
5644 
5645 #ifdef CONFIG_GUEST_PERF_EVENTS
5646 static unsigned int kvm_guest_state(void)
5647 {
5648 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5649 	unsigned int state;
5650 
5651 	if (!kvm_arch_pmi_in_guest(vcpu))
5652 		return 0;
5653 
5654 	state = PERF_GUEST_ACTIVE;
5655 	if (!kvm_arch_vcpu_in_kernel(vcpu))
5656 		state |= PERF_GUEST_USER;
5657 
5658 	return state;
5659 }
5660 
5661 static unsigned long kvm_guest_get_ip(void)
5662 {
5663 	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5664 
5665 	/* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5666 	if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5667 		return 0;
5668 
5669 	return kvm_arch_vcpu_get_ip(vcpu);
5670 }
5671 
5672 static struct perf_guest_info_callbacks kvm_guest_cbs = {
5673 	.state			= kvm_guest_state,
5674 	.get_ip			= kvm_guest_get_ip,
5675 	.handle_intel_pt_intr	= NULL,
5676 };
5677 
5678 void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5679 {
5680 	kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5681 	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5682 }
5683 void kvm_unregister_perf_callbacks(void)
5684 {
5685 	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5686 }
5687 #endif
5688 
5689 struct kvm_cpu_compat_check {
5690 	void *opaque;
5691 	int *ret;
5692 };
5693 
5694 static void check_processor_compat(void *data)
5695 {
5696 	struct kvm_cpu_compat_check *c = data;
5697 
5698 	*c->ret = kvm_arch_check_processor_compat(c->opaque);
5699 }
5700 
5701 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5702 		  struct module *module)
5703 {
5704 	struct kvm_cpu_compat_check c;
5705 	int r;
5706 	int cpu;
5707 
5708 	r = kvm_arch_init(opaque);
5709 	if (r)
5710 		goto out_fail;
5711 
5712 	/*
5713 	 * kvm_arch_init makes sure there's at most one caller
5714 	 * for architectures that support multiple implementations,
5715 	 * like intel and amd on x86.
5716 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5717 	 * conflicts in case kvm is already setup for another implementation.
5718 	 */
5719 	r = kvm_irqfd_init();
5720 	if (r)
5721 		goto out_irqfd;
5722 
5723 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5724 		r = -ENOMEM;
5725 		goto out_free_0;
5726 	}
5727 
5728 	r = kvm_arch_hardware_setup(opaque);
5729 	if (r < 0)
5730 		goto out_free_1;
5731 
5732 	c.ret = &r;
5733 	c.opaque = opaque;
5734 	for_each_online_cpu(cpu) {
5735 		smp_call_function_single(cpu, check_processor_compat, &c, 1);
5736 		if (r < 0)
5737 			goto out_free_2;
5738 	}
5739 
5740 	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5741 				      kvm_starting_cpu, kvm_dying_cpu);
5742 	if (r)
5743 		goto out_free_2;
5744 	register_reboot_notifier(&kvm_reboot_notifier);
5745 
5746 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
5747 	if (!vcpu_align)
5748 		vcpu_align = __alignof__(struct kvm_vcpu);
5749 	kvm_vcpu_cache =
5750 		kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5751 					   SLAB_ACCOUNT,
5752 					   offsetof(struct kvm_vcpu, arch),
5753 					   offsetofend(struct kvm_vcpu, stats_id)
5754 					   - offsetof(struct kvm_vcpu, arch),
5755 					   NULL);
5756 	if (!kvm_vcpu_cache) {
5757 		r = -ENOMEM;
5758 		goto out_free_3;
5759 	}
5760 
5761 	for_each_possible_cpu(cpu) {
5762 		if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5763 					    GFP_KERNEL, cpu_to_node(cpu))) {
5764 			r = -ENOMEM;
5765 			goto out_free_4;
5766 		}
5767 	}
5768 
5769 	r = kvm_async_pf_init();
5770 	if (r)
5771 		goto out_free_5;
5772 
5773 	kvm_chardev_ops.owner = module;
5774 
5775 	r = misc_register(&kvm_dev);
5776 	if (r) {
5777 		pr_err("kvm: misc device register failed\n");
5778 		goto out_unreg;
5779 	}
5780 
5781 	register_syscore_ops(&kvm_syscore_ops);
5782 
5783 	kvm_preempt_ops.sched_in = kvm_sched_in;
5784 	kvm_preempt_ops.sched_out = kvm_sched_out;
5785 
5786 	kvm_init_debug();
5787 
5788 	r = kvm_vfio_ops_init();
5789 	WARN_ON(r);
5790 
5791 	return 0;
5792 
5793 out_unreg:
5794 	kvm_async_pf_deinit();
5795 out_free_5:
5796 	for_each_possible_cpu(cpu)
5797 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5798 out_free_4:
5799 	kmem_cache_destroy(kvm_vcpu_cache);
5800 out_free_3:
5801 	unregister_reboot_notifier(&kvm_reboot_notifier);
5802 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5803 out_free_2:
5804 	kvm_arch_hardware_unsetup();
5805 out_free_1:
5806 	free_cpumask_var(cpus_hardware_enabled);
5807 out_free_0:
5808 	kvm_irqfd_exit();
5809 out_irqfd:
5810 	kvm_arch_exit();
5811 out_fail:
5812 	return r;
5813 }
5814 EXPORT_SYMBOL_GPL(kvm_init);
5815 
5816 void kvm_exit(void)
5817 {
5818 	int cpu;
5819 
5820 	debugfs_remove_recursive(kvm_debugfs_dir);
5821 	misc_deregister(&kvm_dev);
5822 	for_each_possible_cpu(cpu)
5823 		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
5824 	kmem_cache_destroy(kvm_vcpu_cache);
5825 	kvm_async_pf_deinit();
5826 	unregister_syscore_ops(&kvm_syscore_ops);
5827 	unregister_reboot_notifier(&kvm_reboot_notifier);
5828 	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
5829 	on_each_cpu(hardware_disable_nolock, NULL, 1);
5830 	kvm_arch_hardware_unsetup();
5831 	kvm_arch_exit();
5832 	kvm_irqfd_exit();
5833 	free_cpumask_var(cpus_hardware_enabled);
5834 	kvm_vfio_ops_exit();
5835 }
5836 EXPORT_SYMBOL_GPL(kvm_exit);
5837 
5838 struct kvm_vm_worker_thread_context {
5839 	struct kvm *kvm;
5840 	struct task_struct *parent;
5841 	struct completion init_done;
5842 	kvm_vm_thread_fn_t thread_fn;
5843 	uintptr_t data;
5844 	int err;
5845 };
5846 
5847 static int kvm_vm_worker_thread(void *context)
5848 {
5849 	/*
5850 	 * The init_context is allocated on the stack of the parent thread, so
5851 	 * we have to locally copy anything that is needed beyond initialization
5852 	 */
5853 	struct kvm_vm_worker_thread_context *init_context = context;
5854 	struct task_struct *parent;
5855 	struct kvm *kvm = init_context->kvm;
5856 	kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
5857 	uintptr_t data = init_context->data;
5858 	int err;
5859 
5860 	err = kthread_park(current);
5861 	/* kthread_park(current) is never supposed to return an error */
5862 	WARN_ON(err != 0);
5863 	if (err)
5864 		goto init_complete;
5865 
5866 	err = cgroup_attach_task_all(init_context->parent, current);
5867 	if (err) {
5868 		kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
5869 			__func__, err);
5870 		goto init_complete;
5871 	}
5872 
5873 	set_user_nice(current, task_nice(init_context->parent));
5874 
5875 init_complete:
5876 	init_context->err = err;
5877 	complete(&init_context->init_done);
5878 	init_context = NULL;
5879 
5880 	if (err)
5881 		goto out;
5882 
5883 	/* Wait to be woken up by the spawner before proceeding. */
5884 	kthread_parkme();
5885 
5886 	if (!kthread_should_stop())
5887 		err = thread_fn(kvm, data);
5888 
5889 out:
5890 	/*
5891 	 * Move kthread back to its original cgroup to prevent it lingering in
5892 	 * the cgroup of the VM process, after the latter finishes its
5893 	 * execution.
5894 	 *
5895 	 * kthread_stop() waits on the 'exited' completion condition which is
5896 	 * set in exit_mm(), via mm_release(), in do_exit(). However, the
5897 	 * kthread is removed from the cgroup in the cgroup_exit() which is
5898 	 * called after the exit_mm(). This causes the kthread_stop() to return
5899 	 * before the kthread actually quits the cgroup.
5900 	 */
5901 	rcu_read_lock();
5902 	parent = rcu_dereference(current->real_parent);
5903 	get_task_struct(parent);
5904 	rcu_read_unlock();
5905 	cgroup_attach_task_all(parent, current);
5906 	put_task_struct(parent);
5907 
5908 	return err;
5909 }
5910 
5911 int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
5912 				uintptr_t data, const char *name,
5913 				struct task_struct **thread_ptr)
5914 {
5915 	struct kvm_vm_worker_thread_context init_context = {};
5916 	struct task_struct *thread;
5917 
5918 	*thread_ptr = NULL;
5919 	init_context.kvm = kvm;
5920 	init_context.parent = current;
5921 	init_context.thread_fn = thread_fn;
5922 	init_context.data = data;
5923 	init_completion(&init_context.init_done);
5924 
5925 	thread = kthread_run(kvm_vm_worker_thread, &init_context,
5926 			     "%s-%d", name, task_pid_nr(current));
5927 	if (IS_ERR(thread))
5928 		return PTR_ERR(thread);
5929 
5930 	/* kthread_run is never supposed to return NULL */
5931 	WARN_ON(thread == NULL);
5932 
5933 	wait_for_completion(&init_context.init_done);
5934 
5935 	if (!init_context.err)
5936 		*thread_ptr = thread;
5937 
5938 	return init_context.err;
5939 }
5940