xref: /linux/virt/kvm/kvm_main.c (revision 110e6f26af80dfd90b6e5c645b1aed7228aa580d)
1 /*
2  * Kernel-based Virtual Machine driver for Linux
3  *
4  * This module enables machines with Intel VT-x extensions to run virtual
5  * machines without emulation or binary translation.
6  *
7  * Copyright (C) 2006 Qumranet, Inc.
8  * Copyright 2010 Red Hat, Inc. and/or its affiliates.
9  *
10  * Authors:
11  *   Avi Kivity   <avi@qumranet.com>
12  *   Yaniv Kamay  <yaniv@qumranet.com>
13  *
14  * This work is licensed under the terms of the GNU GPL, version 2.  See
15  * the COPYING file in the top-level directory.
16  *
17  */
18 
19 #include <kvm/iodev.h>
20 
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
26 #include <linux/mm.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
52 
53 #include <asm/processor.h>
54 #include <asm/io.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
58 
59 #include "coalesced_mmio.h"
60 #include "async_pf.h"
61 #include "vfio.h"
62 
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
65 
66 MODULE_AUTHOR("Qumranet");
67 MODULE_LICENSE("GPL");
68 
69 /* Architectures should define their poll value according to the halt latency */
70 static unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
71 module_param(halt_poll_ns, uint, S_IRUGO | S_IWUSR);
72 
73 /* Default doubles per-vcpu halt_poll_ns. */
74 static unsigned int halt_poll_ns_grow = 2;
75 module_param(halt_poll_ns_grow, uint, S_IRUGO | S_IWUSR);
76 
77 /* Default resets per-vcpu halt_poll_ns . */
78 static unsigned int halt_poll_ns_shrink;
79 module_param(halt_poll_ns_shrink, uint, S_IRUGO | S_IWUSR);
80 
81 /*
82  * Ordering of locks:
83  *
84  *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
85  */
86 
87 DEFINE_SPINLOCK(kvm_lock);
88 static DEFINE_RAW_SPINLOCK(kvm_count_lock);
89 LIST_HEAD(vm_list);
90 
91 static cpumask_var_t cpus_hardware_enabled;
92 static int kvm_usage_count;
93 static atomic_t hardware_enable_failed;
94 
95 struct kmem_cache *kvm_vcpu_cache;
96 EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
97 
98 static __read_mostly struct preempt_ops kvm_preempt_ops;
99 
100 struct dentry *kvm_debugfs_dir;
101 EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
102 
103 static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
104 			   unsigned long arg);
105 #ifdef CONFIG_KVM_COMPAT
106 static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
107 				  unsigned long arg);
108 #endif
109 static int hardware_enable_all(void);
110 static void hardware_disable_all(void);
111 
112 static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
113 
114 static void kvm_release_pfn_dirty(kvm_pfn_t pfn);
115 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot, gfn_t gfn);
116 
117 __visible bool kvm_rebooting;
118 EXPORT_SYMBOL_GPL(kvm_rebooting);
119 
120 static bool largepages_enabled = true;
121 
122 bool kvm_is_reserved_pfn(kvm_pfn_t pfn)
123 {
124 	if (pfn_valid(pfn))
125 		return PageReserved(pfn_to_page(pfn));
126 
127 	return true;
128 }
129 
130 /*
131  * Switches to specified vcpu, until a matching vcpu_put()
132  */
133 int vcpu_load(struct kvm_vcpu *vcpu)
134 {
135 	int cpu;
136 
137 	if (mutex_lock_killable(&vcpu->mutex))
138 		return -EINTR;
139 	cpu = get_cpu();
140 	preempt_notifier_register(&vcpu->preempt_notifier);
141 	kvm_arch_vcpu_load(vcpu, cpu);
142 	put_cpu();
143 	return 0;
144 }
145 
146 void vcpu_put(struct kvm_vcpu *vcpu)
147 {
148 	preempt_disable();
149 	kvm_arch_vcpu_put(vcpu);
150 	preempt_notifier_unregister(&vcpu->preempt_notifier);
151 	preempt_enable();
152 	mutex_unlock(&vcpu->mutex);
153 }
154 
155 static void ack_flush(void *_completed)
156 {
157 }
158 
159 bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
160 {
161 	int i, cpu, me;
162 	cpumask_var_t cpus;
163 	bool called = true;
164 	struct kvm_vcpu *vcpu;
165 
166 	zalloc_cpumask_var(&cpus, GFP_ATOMIC);
167 
168 	me = get_cpu();
169 	kvm_for_each_vcpu(i, vcpu, kvm) {
170 		kvm_make_request(req, vcpu);
171 		cpu = vcpu->cpu;
172 
173 		/* Set ->requests bit before we read ->mode. */
174 		smp_mb__after_atomic();
175 
176 		if (cpus != NULL && cpu != -1 && cpu != me &&
177 		      kvm_vcpu_exiting_guest_mode(vcpu) != OUTSIDE_GUEST_MODE)
178 			cpumask_set_cpu(cpu, cpus);
179 	}
180 	if (unlikely(cpus == NULL))
181 		smp_call_function_many(cpu_online_mask, ack_flush, NULL, 1);
182 	else if (!cpumask_empty(cpus))
183 		smp_call_function_many(cpus, ack_flush, NULL, 1);
184 	else
185 		called = false;
186 	put_cpu();
187 	free_cpumask_var(cpus);
188 	return called;
189 }
190 
191 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
192 void kvm_flush_remote_tlbs(struct kvm *kvm)
193 {
194 	/*
195 	 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
196 	 * kvm_make_all_cpus_request.
197 	 */
198 	long dirty_count = smp_load_acquire(&kvm->tlbs_dirty);
199 
200 	/*
201 	 * We want to publish modifications to the page tables before reading
202 	 * mode. Pairs with a memory barrier in arch-specific code.
203 	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
204 	 * and smp_mb in walk_shadow_page_lockless_begin/end.
205 	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
206 	 *
207 	 * There is already an smp_mb__after_atomic() before
208 	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
209 	 * barrier here.
210 	 */
211 	if (kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
212 		++kvm->stat.remote_tlb_flush;
213 	cmpxchg(&kvm->tlbs_dirty, dirty_count, 0);
214 }
215 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
216 #endif
217 
218 void kvm_reload_remote_mmus(struct kvm *kvm)
219 {
220 	kvm_make_all_cpus_request(kvm, KVM_REQ_MMU_RELOAD);
221 }
222 
223 int kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
224 {
225 	struct page *page;
226 	int r;
227 
228 	mutex_init(&vcpu->mutex);
229 	vcpu->cpu = -1;
230 	vcpu->kvm = kvm;
231 	vcpu->vcpu_id = id;
232 	vcpu->pid = NULL;
233 	init_swait_queue_head(&vcpu->wq);
234 	kvm_async_pf_vcpu_init(vcpu);
235 
236 	vcpu->pre_pcpu = -1;
237 	INIT_LIST_HEAD(&vcpu->blocked_vcpu_list);
238 
239 	page = alloc_page(GFP_KERNEL | __GFP_ZERO);
240 	if (!page) {
241 		r = -ENOMEM;
242 		goto fail;
243 	}
244 	vcpu->run = page_address(page);
245 
246 	kvm_vcpu_set_in_spin_loop(vcpu, false);
247 	kvm_vcpu_set_dy_eligible(vcpu, false);
248 	vcpu->preempted = false;
249 
250 	r = kvm_arch_vcpu_init(vcpu);
251 	if (r < 0)
252 		goto fail_free_run;
253 	return 0;
254 
255 fail_free_run:
256 	free_page((unsigned long)vcpu->run);
257 fail:
258 	return r;
259 }
260 EXPORT_SYMBOL_GPL(kvm_vcpu_init);
261 
262 void kvm_vcpu_uninit(struct kvm_vcpu *vcpu)
263 {
264 	put_pid(vcpu->pid);
265 	kvm_arch_vcpu_uninit(vcpu);
266 	free_page((unsigned long)vcpu->run);
267 }
268 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit);
269 
270 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
271 static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
272 {
273 	return container_of(mn, struct kvm, mmu_notifier);
274 }
275 
276 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier *mn,
277 					     struct mm_struct *mm,
278 					     unsigned long address)
279 {
280 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
281 	int need_tlb_flush, idx;
282 
283 	/*
284 	 * When ->invalidate_page runs, the linux pte has been zapped
285 	 * already but the page is still allocated until
286 	 * ->invalidate_page returns. So if we increase the sequence
287 	 * here the kvm page fault will notice if the spte can't be
288 	 * established because the page is going to be freed. If
289 	 * instead the kvm page fault establishes the spte before
290 	 * ->invalidate_page runs, kvm_unmap_hva will release it
291 	 * before returning.
292 	 *
293 	 * The sequence increase only need to be seen at spin_unlock
294 	 * time, and not at spin_lock time.
295 	 *
296 	 * Increasing the sequence after the spin_unlock would be
297 	 * unsafe because the kvm page fault could then establish the
298 	 * pte after kvm_unmap_hva returned, without noticing the page
299 	 * is going to be freed.
300 	 */
301 	idx = srcu_read_lock(&kvm->srcu);
302 	spin_lock(&kvm->mmu_lock);
303 
304 	kvm->mmu_notifier_seq++;
305 	need_tlb_flush = kvm_unmap_hva(kvm, address) | kvm->tlbs_dirty;
306 	/* we've to flush the tlb before the pages can be freed */
307 	if (need_tlb_flush)
308 		kvm_flush_remote_tlbs(kvm);
309 
310 	spin_unlock(&kvm->mmu_lock);
311 
312 	kvm_arch_mmu_notifier_invalidate_page(kvm, address);
313 
314 	srcu_read_unlock(&kvm->srcu, idx);
315 }
316 
317 static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
318 					struct mm_struct *mm,
319 					unsigned long address,
320 					pte_t pte)
321 {
322 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
323 	int idx;
324 
325 	idx = srcu_read_lock(&kvm->srcu);
326 	spin_lock(&kvm->mmu_lock);
327 	kvm->mmu_notifier_seq++;
328 	kvm_set_spte_hva(kvm, address, pte);
329 	spin_unlock(&kvm->mmu_lock);
330 	srcu_read_unlock(&kvm->srcu, idx);
331 }
332 
333 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
334 						    struct mm_struct *mm,
335 						    unsigned long start,
336 						    unsigned long end)
337 {
338 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
339 	int need_tlb_flush = 0, idx;
340 
341 	idx = srcu_read_lock(&kvm->srcu);
342 	spin_lock(&kvm->mmu_lock);
343 	/*
344 	 * The count increase must become visible at unlock time as no
345 	 * spte can be established without taking the mmu_lock and
346 	 * count is also read inside the mmu_lock critical section.
347 	 */
348 	kvm->mmu_notifier_count++;
349 	need_tlb_flush = kvm_unmap_hva_range(kvm, start, end);
350 	need_tlb_flush |= kvm->tlbs_dirty;
351 	/* we've to flush the tlb before the pages can be freed */
352 	if (need_tlb_flush)
353 		kvm_flush_remote_tlbs(kvm);
354 
355 	spin_unlock(&kvm->mmu_lock);
356 	srcu_read_unlock(&kvm->srcu, idx);
357 }
358 
359 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
360 						  struct mm_struct *mm,
361 						  unsigned long start,
362 						  unsigned long end)
363 {
364 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
365 
366 	spin_lock(&kvm->mmu_lock);
367 	/*
368 	 * This sequence increase will notify the kvm page fault that
369 	 * the page that is going to be mapped in the spte could have
370 	 * been freed.
371 	 */
372 	kvm->mmu_notifier_seq++;
373 	smp_wmb();
374 	/*
375 	 * The above sequence increase must be visible before the
376 	 * below count decrease, which is ensured by the smp_wmb above
377 	 * in conjunction with the smp_rmb in mmu_notifier_retry().
378 	 */
379 	kvm->mmu_notifier_count--;
380 	spin_unlock(&kvm->mmu_lock);
381 
382 	BUG_ON(kvm->mmu_notifier_count < 0);
383 }
384 
385 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
386 					      struct mm_struct *mm,
387 					      unsigned long start,
388 					      unsigned long end)
389 {
390 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
391 	int young, idx;
392 
393 	idx = srcu_read_lock(&kvm->srcu);
394 	spin_lock(&kvm->mmu_lock);
395 
396 	young = kvm_age_hva(kvm, start, end);
397 	if (young)
398 		kvm_flush_remote_tlbs(kvm);
399 
400 	spin_unlock(&kvm->mmu_lock);
401 	srcu_read_unlock(&kvm->srcu, idx);
402 
403 	return young;
404 }
405 
406 static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
407 					struct mm_struct *mm,
408 					unsigned long start,
409 					unsigned long end)
410 {
411 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
412 	int young, idx;
413 
414 	idx = srcu_read_lock(&kvm->srcu);
415 	spin_lock(&kvm->mmu_lock);
416 	/*
417 	 * Even though we do not flush TLB, this will still adversely
418 	 * affect performance on pre-Haswell Intel EPT, where there is
419 	 * no EPT Access Bit to clear so that we have to tear down EPT
420 	 * tables instead. If we find this unacceptable, we can always
421 	 * add a parameter to kvm_age_hva so that it effectively doesn't
422 	 * do anything on clear_young.
423 	 *
424 	 * Also note that currently we never issue secondary TLB flushes
425 	 * from clear_young, leaving this job up to the regular system
426 	 * cadence. If we find this inaccurate, we might come up with a
427 	 * more sophisticated heuristic later.
428 	 */
429 	young = kvm_age_hva(kvm, start, end);
430 	spin_unlock(&kvm->mmu_lock);
431 	srcu_read_unlock(&kvm->srcu, idx);
432 
433 	return young;
434 }
435 
436 static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
437 				       struct mm_struct *mm,
438 				       unsigned long address)
439 {
440 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
441 	int young, idx;
442 
443 	idx = srcu_read_lock(&kvm->srcu);
444 	spin_lock(&kvm->mmu_lock);
445 	young = kvm_test_age_hva(kvm, address);
446 	spin_unlock(&kvm->mmu_lock);
447 	srcu_read_unlock(&kvm->srcu, idx);
448 
449 	return young;
450 }
451 
452 static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
453 				     struct mm_struct *mm)
454 {
455 	struct kvm *kvm = mmu_notifier_to_kvm(mn);
456 	int idx;
457 
458 	idx = srcu_read_lock(&kvm->srcu);
459 	kvm_arch_flush_shadow_all(kvm);
460 	srcu_read_unlock(&kvm->srcu, idx);
461 }
462 
463 static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
464 	.invalidate_page	= kvm_mmu_notifier_invalidate_page,
465 	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
466 	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
467 	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
468 	.clear_young		= kvm_mmu_notifier_clear_young,
469 	.test_young		= kvm_mmu_notifier_test_young,
470 	.change_pte		= kvm_mmu_notifier_change_pte,
471 	.release		= kvm_mmu_notifier_release,
472 };
473 
474 static int kvm_init_mmu_notifier(struct kvm *kvm)
475 {
476 	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
477 	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
478 }
479 
480 #else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
481 
482 static int kvm_init_mmu_notifier(struct kvm *kvm)
483 {
484 	return 0;
485 }
486 
487 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
488 
489 static struct kvm_memslots *kvm_alloc_memslots(void)
490 {
491 	int i;
492 	struct kvm_memslots *slots;
493 
494 	slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
495 	if (!slots)
496 		return NULL;
497 
498 	/*
499 	 * Init kvm generation close to the maximum to easily test the
500 	 * code of handling generation number wrap-around.
501 	 */
502 	slots->generation = -150;
503 	for (i = 0; i < KVM_MEM_SLOTS_NUM; i++)
504 		slots->id_to_index[i] = slots->memslots[i].id = i;
505 
506 	return slots;
507 }
508 
509 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
510 {
511 	if (!memslot->dirty_bitmap)
512 		return;
513 
514 	kvfree(memslot->dirty_bitmap);
515 	memslot->dirty_bitmap = NULL;
516 }
517 
518 /*
519  * Free any memory in @free but not in @dont.
520  */
521 static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
522 			      struct kvm_memory_slot *dont)
523 {
524 	if (!dont || free->dirty_bitmap != dont->dirty_bitmap)
525 		kvm_destroy_dirty_bitmap(free);
526 
527 	kvm_arch_free_memslot(kvm, free, dont);
528 
529 	free->npages = 0;
530 }
531 
532 static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
533 {
534 	struct kvm_memory_slot *memslot;
535 
536 	if (!slots)
537 		return;
538 
539 	kvm_for_each_memslot(memslot, slots)
540 		kvm_free_memslot(kvm, memslot, NULL);
541 
542 	kvfree(slots);
543 }
544 
545 static struct kvm *kvm_create_vm(unsigned long type)
546 {
547 	int r, i;
548 	struct kvm *kvm = kvm_arch_alloc_vm();
549 
550 	if (!kvm)
551 		return ERR_PTR(-ENOMEM);
552 
553 	spin_lock_init(&kvm->mmu_lock);
554 	atomic_inc(&current->mm->mm_count);
555 	kvm->mm = current->mm;
556 	kvm_eventfd_init(kvm);
557 	mutex_init(&kvm->lock);
558 	mutex_init(&kvm->irq_lock);
559 	mutex_init(&kvm->slots_lock);
560 	atomic_set(&kvm->users_count, 1);
561 	INIT_LIST_HEAD(&kvm->devices);
562 
563 	r = kvm_arch_init_vm(kvm, type);
564 	if (r)
565 		goto out_err_no_disable;
566 
567 	r = hardware_enable_all();
568 	if (r)
569 		goto out_err_no_disable;
570 
571 #ifdef CONFIG_HAVE_KVM_IRQFD
572 	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
573 #endif
574 
575 	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
576 
577 	r = -ENOMEM;
578 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
579 		kvm->memslots[i] = kvm_alloc_memslots();
580 		if (!kvm->memslots[i])
581 			goto out_err_no_srcu;
582 	}
583 
584 	if (init_srcu_struct(&kvm->srcu))
585 		goto out_err_no_srcu;
586 	if (init_srcu_struct(&kvm->irq_srcu))
587 		goto out_err_no_irq_srcu;
588 	for (i = 0; i < KVM_NR_BUSES; i++) {
589 		kvm->buses[i] = kzalloc(sizeof(struct kvm_io_bus),
590 					GFP_KERNEL);
591 		if (!kvm->buses[i])
592 			goto out_err;
593 	}
594 
595 	r = kvm_init_mmu_notifier(kvm);
596 	if (r)
597 		goto out_err;
598 
599 	spin_lock(&kvm_lock);
600 	list_add(&kvm->vm_list, &vm_list);
601 	spin_unlock(&kvm_lock);
602 
603 	preempt_notifier_inc();
604 
605 	return kvm;
606 
607 out_err:
608 	cleanup_srcu_struct(&kvm->irq_srcu);
609 out_err_no_irq_srcu:
610 	cleanup_srcu_struct(&kvm->srcu);
611 out_err_no_srcu:
612 	hardware_disable_all();
613 out_err_no_disable:
614 	for (i = 0; i < KVM_NR_BUSES; i++)
615 		kfree(kvm->buses[i]);
616 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
617 		kvm_free_memslots(kvm, kvm->memslots[i]);
618 	kvm_arch_free_vm(kvm);
619 	mmdrop(current->mm);
620 	return ERR_PTR(r);
621 }
622 
623 /*
624  * Avoid using vmalloc for a small buffer.
625  * Should not be used when the size is statically known.
626  */
627 void *kvm_kvzalloc(unsigned long size)
628 {
629 	if (size > PAGE_SIZE)
630 		return vzalloc(size);
631 	else
632 		return kzalloc(size, GFP_KERNEL);
633 }
634 
635 static void kvm_destroy_devices(struct kvm *kvm)
636 {
637 	struct kvm_device *dev, *tmp;
638 
639 	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
640 		list_del(&dev->vm_node);
641 		dev->ops->destroy(dev);
642 	}
643 }
644 
645 static void kvm_destroy_vm(struct kvm *kvm)
646 {
647 	int i;
648 	struct mm_struct *mm = kvm->mm;
649 
650 	kvm_arch_sync_events(kvm);
651 	spin_lock(&kvm_lock);
652 	list_del(&kvm->vm_list);
653 	spin_unlock(&kvm_lock);
654 	kvm_free_irq_routing(kvm);
655 	for (i = 0; i < KVM_NR_BUSES; i++)
656 		kvm_io_bus_destroy(kvm->buses[i]);
657 	kvm_coalesced_mmio_free(kvm);
658 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
659 	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
660 #else
661 	kvm_arch_flush_shadow_all(kvm);
662 #endif
663 	kvm_arch_destroy_vm(kvm);
664 	kvm_destroy_devices(kvm);
665 	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++)
666 		kvm_free_memslots(kvm, kvm->memslots[i]);
667 	cleanup_srcu_struct(&kvm->irq_srcu);
668 	cleanup_srcu_struct(&kvm->srcu);
669 	kvm_arch_free_vm(kvm);
670 	preempt_notifier_dec();
671 	hardware_disable_all();
672 	mmdrop(mm);
673 }
674 
675 void kvm_get_kvm(struct kvm *kvm)
676 {
677 	atomic_inc(&kvm->users_count);
678 }
679 EXPORT_SYMBOL_GPL(kvm_get_kvm);
680 
681 void kvm_put_kvm(struct kvm *kvm)
682 {
683 	if (atomic_dec_and_test(&kvm->users_count))
684 		kvm_destroy_vm(kvm);
685 }
686 EXPORT_SYMBOL_GPL(kvm_put_kvm);
687 
688 
689 static int kvm_vm_release(struct inode *inode, struct file *filp)
690 {
691 	struct kvm *kvm = filp->private_data;
692 
693 	kvm_irqfd_release(kvm);
694 
695 	kvm_put_kvm(kvm);
696 	return 0;
697 }
698 
699 /*
700  * Allocation size is twice as large as the actual dirty bitmap size.
701  * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
702  */
703 static int kvm_create_dirty_bitmap(struct kvm_memory_slot *memslot)
704 {
705 	unsigned long dirty_bytes = 2 * kvm_dirty_bitmap_bytes(memslot);
706 
707 	memslot->dirty_bitmap = kvm_kvzalloc(dirty_bytes);
708 	if (!memslot->dirty_bitmap)
709 		return -ENOMEM;
710 
711 	return 0;
712 }
713 
714 /*
715  * Insert memslot and re-sort memslots based on their GFN,
716  * so binary search could be used to lookup GFN.
717  * Sorting algorithm takes advantage of having initially
718  * sorted array and known changed memslot position.
719  */
720 static void update_memslots(struct kvm_memslots *slots,
721 			    struct kvm_memory_slot *new)
722 {
723 	int id = new->id;
724 	int i = slots->id_to_index[id];
725 	struct kvm_memory_slot *mslots = slots->memslots;
726 
727 	WARN_ON(mslots[i].id != id);
728 	if (!new->npages) {
729 		WARN_ON(!mslots[i].npages);
730 		if (mslots[i].npages)
731 			slots->used_slots--;
732 	} else {
733 		if (!mslots[i].npages)
734 			slots->used_slots++;
735 	}
736 
737 	while (i < KVM_MEM_SLOTS_NUM - 1 &&
738 	       new->base_gfn <= mslots[i + 1].base_gfn) {
739 		if (!mslots[i + 1].npages)
740 			break;
741 		mslots[i] = mslots[i + 1];
742 		slots->id_to_index[mslots[i].id] = i;
743 		i++;
744 	}
745 
746 	/*
747 	 * The ">=" is needed when creating a slot with base_gfn == 0,
748 	 * so that it moves before all those with base_gfn == npages == 0.
749 	 *
750 	 * On the other hand, if new->npages is zero, the above loop has
751 	 * already left i pointing to the beginning of the empty part of
752 	 * mslots, and the ">=" would move the hole backwards in this
753 	 * case---which is wrong.  So skip the loop when deleting a slot.
754 	 */
755 	if (new->npages) {
756 		while (i > 0 &&
757 		       new->base_gfn >= mslots[i - 1].base_gfn) {
758 			mslots[i] = mslots[i - 1];
759 			slots->id_to_index[mslots[i].id] = i;
760 			i--;
761 		}
762 	} else
763 		WARN_ON_ONCE(i != slots->used_slots);
764 
765 	mslots[i] = *new;
766 	slots->id_to_index[mslots[i].id] = i;
767 }
768 
769 static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
770 {
771 	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
772 
773 #ifdef __KVM_HAVE_READONLY_MEM
774 	valid_flags |= KVM_MEM_READONLY;
775 #endif
776 
777 	if (mem->flags & ~valid_flags)
778 		return -EINVAL;
779 
780 	return 0;
781 }
782 
783 static struct kvm_memslots *install_new_memslots(struct kvm *kvm,
784 		int as_id, struct kvm_memslots *slots)
785 {
786 	struct kvm_memslots *old_memslots = __kvm_memslots(kvm, as_id);
787 
788 	/*
789 	 * Set the low bit in the generation, which disables SPTE caching
790 	 * until the end of synchronize_srcu_expedited.
791 	 */
792 	WARN_ON(old_memslots->generation & 1);
793 	slots->generation = old_memslots->generation + 1;
794 
795 	rcu_assign_pointer(kvm->memslots[as_id], slots);
796 	synchronize_srcu_expedited(&kvm->srcu);
797 
798 	/*
799 	 * Increment the new memslot generation a second time. This prevents
800 	 * vm exits that race with memslot updates from caching a memslot
801 	 * generation that will (potentially) be valid forever.
802 	 */
803 	slots->generation++;
804 
805 	kvm_arch_memslots_updated(kvm, slots);
806 
807 	return old_memslots;
808 }
809 
810 /*
811  * Allocate some memory and give it an address in the guest physical address
812  * space.
813  *
814  * Discontiguous memory is allowed, mostly for framebuffers.
815  *
816  * Must be called holding kvm->slots_lock for write.
817  */
818 int __kvm_set_memory_region(struct kvm *kvm,
819 			    const struct kvm_userspace_memory_region *mem)
820 {
821 	int r;
822 	gfn_t base_gfn;
823 	unsigned long npages;
824 	struct kvm_memory_slot *slot;
825 	struct kvm_memory_slot old, new;
826 	struct kvm_memslots *slots = NULL, *old_memslots;
827 	int as_id, id;
828 	enum kvm_mr_change change;
829 
830 	r = check_memory_region_flags(mem);
831 	if (r)
832 		goto out;
833 
834 	r = -EINVAL;
835 	as_id = mem->slot >> 16;
836 	id = (u16)mem->slot;
837 
838 	/* General sanity checks */
839 	if (mem->memory_size & (PAGE_SIZE - 1))
840 		goto out;
841 	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
842 		goto out;
843 	/* We can read the guest memory with __xxx_user() later on. */
844 	if ((id < KVM_USER_MEM_SLOTS) &&
845 	    ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
846 	     !access_ok(VERIFY_WRITE,
847 			(void __user *)(unsigned long)mem->userspace_addr,
848 			mem->memory_size)))
849 		goto out;
850 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
851 		goto out;
852 	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
853 		goto out;
854 
855 	slot = id_to_memslot(__kvm_memslots(kvm, as_id), id);
856 	base_gfn = mem->guest_phys_addr >> PAGE_SHIFT;
857 	npages = mem->memory_size >> PAGE_SHIFT;
858 
859 	if (npages > KVM_MEM_MAX_NR_PAGES)
860 		goto out;
861 
862 	new = old = *slot;
863 
864 	new.id = id;
865 	new.base_gfn = base_gfn;
866 	new.npages = npages;
867 	new.flags = mem->flags;
868 
869 	if (npages) {
870 		if (!old.npages)
871 			change = KVM_MR_CREATE;
872 		else { /* Modify an existing slot. */
873 			if ((mem->userspace_addr != old.userspace_addr) ||
874 			    (npages != old.npages) ||
875 			    ((new.flags ^ old.flags) & KVM_MEM_READONLY))
876 				goto out;
877 
878 			if (base_gfn != old.base_gfn)
879 				change = KVM_MR_MOVE;
880 			else if (new.flags != old.flags)
881 				change = KVM_MR_FLAGS_ONLY;
882 			else { /* Nothing to change. */
883 				r = 0;
884 				goto out;
885 			}
886 		}
887 	} else {
888 		if (!old.npages)
889 			goto out;
890 
891 		change = KVM_MR_DELETE;
892 		new.base_gfn = 0;
893 		new.flags = 0;
894 	}
895 
896 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
897 		/* Check for overlaps */
898 		r = -EEXIST;
899 		kvm_for_each_memslot(slot, __kvm_memslots(kvm, as_id)) {
900 			if ((slot->id >= KVM_USER_MEM_SLOTS) ||
901 			    (slot->id == id))
902 				continue;
903 			if (!((base_gfn + npages <= slot->base_gfn) ||
904 			      (base_gfn >= slot->base_gfn + slot->npages)))
905 				goto out;
906 		}
907 	}
908 
909 	/* Free page dirty bitmap if unneeded */
910 	if (!(new.flags & KVM_MEM_LOG_DIRTY_PAGES))
911 		new.dirty_bitmap = NULL;
912 
913 	r = -ENOMEM;
914 	if (change == KVM_MR_CREATE) {
915 		new.userspace_addr = mem->userspace_addr;
916 
917 		if (kvm_arch_create_memslot(kvm, &new, npages))
918 			goto out_free;
919 	}
920 
921 	/* Allocate page dirty bitmap if needed */
922 	if ((new.flags & KVM_MEM_LOG_DIRTY_PAGES) && !new.dirty_bitmap) {
923 		if (kvm_create_dirty_bitmap(&new) < 0)
924 			goto out_free;
925 	}
926 
927 	slots = kvm_kvzalloc(sizeof(struct kvm_memslots));
928 	if (!slots)
929 		goto out_free;
930 	memcpy(slots, __kvm_memslots(kvm, as_id), sizeof(struct kvm_memslots));
931 
932 	if ((change == KVM_MR_DELETE) || (change == KVM_MR_MOVE)) {
933 		slot = id_to_memslot(slots, id);
934 		slot->flags |= KVM_MEMSLOT_INVALID;
935 
936 		old_memslots = install_new_memslots(kvm, as_id, slots);
937 
938 		/* slot was deleted or moved, clear iommu mapping */
939 		kvm_iommu_unmap_pages(kvm, &old);
940 		/* From this point no new shadow pages pointing to a deleted,
941 		 * or moved, memslot will be created.
942 		 *
943 		 * validation of sp->gfn happens in:
944 		 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
945 		 *	- kvm_is_visible_gfn (mmu_check_roots)
946 		 */
947 		kvm_arch_flush_shadow_memslot(kvm, slot);
948 
949 		/*
950 		 * We can re-use the old_memslots from above, the only difference
951 		 * from the currently installed memslots is the invalid flag.  This
952 		 * will get overwritten by update_memslots anyway.
953 		 */
954 		slots = old_memslots;
955 	}
956 
957 	r = kvm_arch_prepare_memory_region(kvm, &new, mem, change);
958 	if (r)
959 		goto out_slots;
960 
961 	/* actual memory is freed via old in kvm_free_memslot below */
962 	if (change == KVM_MR_DELETE) {
963 		new.dirty_bitmap = NULL;
964 		memset(&new.arch, 0, sizeof(new.arch));
965 	}
966 
967 	update_memslots(slots, &new);
968 	old_memslots = install_new_memslots(kvm, as_id, slots);
969 
970 	kvm_arch_commit_memory_region(kvm, mem, &old, &new, change);
971 
972 	kvm_free_memslot(kvm, &old, &new);
973 	kvfree(old_memslots);
974 
975 	/*
976 	 * IOMMU mapping:  New slots need to be mapped.  Old slots need to be
977 	 * un-mapped and re-mapped if their base changes.  Since base change
978 	 * unmapping is handled above with slot deletion, mapping alone is
979 	 * needed here.  Anything else the iommu might care about for existing
980 	 * slots (size changes, userspace addr changes and read-only flag
981 	 * changes) is disallowed above, so any other attribute changes getting
982 	 * here can be skipped.
983 	 */
984 	if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
985 		r = kvm_iommu_map_pages(kvm, &new);
986 		return r;
987 	}
988 
989 	return 0;
990 
991 out_slots:
992 	kvfree(slots);
993 out_free:
994 	kvm_free_memslot(kvm, &new, &old);
995 out:
996 	return r;
997 }
998 EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
999 
1000 int kvm_set_memory_region(struct kvm *kvm,
1001 			  const struct kvm_userspace_memory_region *mem)
1002 {
1003 	int r;
1004 
1005 	mutex_lock(&kvm->slots_lock);
1006 	r = __kvm_set_memory_region(kvm, mem);
1007 	mutex_unlock(&kvm->slots_lock);
1008 	return r;
1009 }
1010 EXPORT_SYMBOL_GPL(kvm_set_memory_region);
1011 
1012 static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
1013 					  struct kvm_userspace_memory_region *mem)
1014 {
1015 	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
1016 		return -EINVAL;
1017 
1018 	return kvm_set_memory_region(kvm, mem);
1019 }
1020 
1021 int kvm_get_dirty_log(struct kvm *kvm,
1022 			struct kvm_dirty_log *log, int *is_dirty)
1023 {
1024 	struct kvm_memslots *slots;
1025 	struct kvm_memory_slot *memslot;
1026 	int r, i, as_id, id;
1027 	unsigned long n;
1028 	unsigned long any = 0;
1029 
1030 	r = -EINVAL;
1031 	as_id = log->slot >> 16;
1032 	id = (u16)log->slot;
1033 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1034 		goto out;
1035 
1036 	slots = __kvm_memslots(kvm, as_id);
1037 	memslot = id_to_memslot(slots, id);
1038 	r = -ENOENT;
1039 	if (!memslot->dirty_bitmap)
1040 		goto out;
1041 
1042 	n = kvm_dirty_bitmap_bytes(memslot);
1043 
1044 	for (i = 0; !any && i < n/sizeof(long); ++i)
1045 		any = memslot->dirty_bitmap[i];
1046 
1047 	r = -EFAULT;
1048 	if (copy_to_user(log->dirty_bitmap, memslot->dirty_bitmap, n))
1049 		goto out;
1050 
1051 	if (any)
1052 		*is_dirty = 1;
1053 
1054 	r = 0;
1055 out:
1056 	return r;
1057 }
1058 EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
1059 
1060 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1061 /**
1062  * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1063  *	are dirty write protect them for next write.
1064  * @kvm:	pointer to kvm instance
1065  * @log:	slot id and address to which we copy the log
1066  * @is_dirty:	flag set if any page is dirty
1067  *
1068  * We need to keep it in mind that VCPU threads can write to the bitmap
1069  * concurrently. So, to avoid losing track of dirty pages we keep the
1070  * following order:
1071  *
1072  *    1. Take a snapshot of the bit and clear it if needed.
1073  *    2. Write protect the corresponding page.
1074  *    3. Copy the snapshot to the userspace.
1075  *    4. Upon return caller flushes TLB's if needed.
1076  *
1077  * Between 2 and 4, the guest may write to the page using the remaining TLB
1078  * entry.  This is not a problem because the page is reported dirty using
1079  * the snapshot taken before and step 4 ensures that writes done after
1080  * exiting to userspace will be logged for the next call.
1081  *
1082  */
1083 int kvm_get_dirty_log_protect(struct kvm *kvm,
1084 			struct kvm_dirty_log *log, bool *is_dirty)
1085 {
1086 	struct kvm_memslots *slots;
1087 	struct kvm_memory_slot *memslot;
1088 	int r, i, as_id, id;
1089 	unsigned long n;
1090 	unsigned long *dirty_bitmap;
1091 	unsigned long *dirty_bitmap_buffer;
1092 
1093 	r = -EINVAL;
1094 	as_id = log->slot >> 16;
1095 	id = (u16)log->slot;
1096 	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
1097 		goto out;
1098 
1099 	slots = __kvm_memslots(kvm, as_id);
1100 	memslot = id_to_memslot(slots, id);
1101 
1102 	dirty_bitmap = memslot->dirty_bitmap;
1103 	r = -ENOENT;
1104 	if (!dirty_bitmap)
1105 		goto out;
1106 
1107 	n = kvm_dirty_bitmap_bytes(memslot);
1108 
1109 	dirty_bitmap_buffer = dirty_bitmap + n / sizeof(long);
1110 	memset(dirty_bitmap_buffer, 0, n);
1111 
1112 	spin_lock(&kvm->mmu_lock);
1113 	*is_dirty = false;
1114 	for (i = 0; i < n / sizeof(long); i++) {
1115 		unsigned long mask;
1116 		gfn_t offset;
1117 
1118 		if (!dirty_bitmap[i])
1119 			continue;
1120 
1121 		*is_dirty = true;
1122 
1123 		mask = xchg(&dirty_bitmap[i], 0);
1124 		dirty_bitmap_buffer[i] = mask;
1125 
1126 		if (mask) {
1127 			offset = i * BITS_PER_LONG;
1128 			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
1129 								offset, mask);
1130 		}
1131 	}
1132 
1133 	spin_unlock(&kvm->mmu_lock);
1134 
1135 	r = -EFAULT;
1136 	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
1137 		goto out;
1138 
1139 	r = 0;
1140 out:
1141 	return r;
1142 }
1143 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect);
1144 #endif
1145 
1146 bool kvm_largepages_enabled(void)
1147 {
1148 	return largepages_enabled;
1149 }
1150 
1151 void kvm_disable_largepages(void)
1152 {
1153 	largepages_enabled = false;
1154 }
1155 EXPORT_SYMBOL_GPL(kvm_disable_largepages);
1156 
1157 struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
1158 {
1159 	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
1160 }
1161 EXPORT_SYMBOL_GPL(gfn_to_memslot);
1162 
1163 struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
1164 {
1165 	return __gfn_to_memslot(kvm_vcpu_memslots(vcpu), gfn);
1166 }
1167 
1168 bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
1169 {
1170 	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
1171 
1172 	if (!memslot || memslot->id >= KVM_USER_MEM_SLOTS ||
1173 	      memslot->flags & KVM_MEMSLOT_INVALID)
1174 		return false;
1175 
1176 	return true;
1177 }
1178 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
1179 
1180 unsigned long kvm_host_page_size(struct kvm *kvm, gfn_t gfn)
1181 {
1182 	struct vm_area_struct *vma;
1183 	unsigned long addr, size;
1184 
1185 	size = PAGE_SIZE;
1186 
1187 	addr = gfn_to_hva(kvm, gfn);
1188 	if (kvm_is_error_hva(addr))
1189 		return PAGE_SIZE;
1190 
1191 	down_read(&current->mm->mmap_sem);
1192 	vma = find_vma(current->mm, addr);
1193 	if (!vma)
1194 		goto out;
1195 
1196 	size = vma_kernel_pagesize(vma);
1197 
1198 out:
1199 	up_read(&current->mm->mmap_sem);
1200 
1201 	return size;
1202 }
1203 
1204 static bool memslot_is_readonly(struct kvm_memory_slot *slot)
1205 {
1206 	return slot->flags & KVM_MEM_READONLY;
1207 }
1208 
1209 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1210 				       gfn_t *nr_pages, bool write)
1211 {
1212 	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
1213 		return KVM_HVA_ERR_BAD;
1214 
1215 	if (memslot_is_readonly(slot) && write)
1216 		return KVM_HVA_ERR_RO_BAD;
1217 
1218 	if (nr_pages)
1219 		*nr_pages = slot->npages - (gfn - slot->base_gfn);
1220 
1221 	return __gfn_to_hva_memslot(slot, gfn);
1222 }
1223 
1224 static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
1225 				     gfn_t *nr_pages)
1226 {
1227 	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
1228 }
1229 
1230 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
1231 					gfn_t gfn)
1232 {
1233 	return gfn_to_hva_many(slot, gfn, NULL);
1234 }
1235 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
1236 
1237 unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
1238 {
1239 	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
1240 }
1241 EXPORT_SYMBOL_GPL(gfn_to_hva);
1242 
1243 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
1244 {
1245 	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
1246 }
1247 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
1248 
1249 /*
1250  * If writable is set to false, the hva returned by this function is only
1251  * allowed to be read.
1252  */
1253 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
1254 				      gfn_t gfn, bool *writable)
1255 {
1256 	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
1257 
1258 	if (!kvm_is_error_hva(hva) && writable)
1259 		*writable = !memslot_is_readonly(slot);
1260 
1261 	return hva;
1262 }
1263 
1264 unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
1265 {
1266 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1267 
1268 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1269 }
1270 
1271 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
1272 {
1273 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1274 
1275 	return gfn_to_hva_memslot_prot(slot, gfn, writable);
1276 }
1277 
1278 static int get_user_page_nowait(unsigned long start, int write,
1279 		struct page **page)
1280 {
1281 	int flags = FOLL_TOUCH | FOLL_NOWAIT | FOLL_HWPOISON | FOLL_GET;
1282 
1283 	if (write)
1284 		flags |= FOLL_WRITE;
1285 
1286 	return __get_user_pages(current, current->mm, start, 1, flags, page,
1287 			NULL, NULL);
1288 }
1289 
1290 static inline int check_user_page_hwpoison(unsigned long addr)
1291 {
1292 	int rc, flags = FOLL_TOUCH | FOLL_HWPOISON | FOLL_WRITE;
1293 
1294 	rc = __get_user_pages(current, current->mm, addr, 1,
1295 			      flags, NULL, NULL, NULL);
1296 	return rc == -EHWPOISON;
1297 }
1298 
1299 /*
1300  * The atomic path to get the writable pfn which will be stored in @pfn,
1301  * true indicates success, otherwise false is returned.
1302  */
1303 static bool hva_to_pfn_fast(unsigned long addr, bool atomic, bool *async,
1304 			    bool write_fault, bool *writable, kvm_pfn_t *pfn)
1305 {
1306 	struct page *page[1];
1307 	int npages;
1308 
1309 	if (!(async || atomic))
1310 		return false;
1311 
1312 	/*
1313 	 * Fast pin a writable pfn only if it is a write fault request
1314 	 * or the caller allows to map a writable pfn for a read fault
1315 	 * request.
1316 	 */
1317 	if (!(write_fault || writable))
1318 		return false;
1319 
1320 	npages = __get_user_pages_fast(addr, 1, 1, page);
1321 	if (npages == 1) {
1322 		*pfn = page_to_pfn(page[0]);
1323 
1324 		if (writable)
1325 			*writable = true;
1326 		return true;
1327 	}
1328 
1329 	return false;
1330 }
1331 
1332 /*
1333  * The slow path to get the pfn of the specified host virtual address,
1334  * 1 indicates success, -errno is returned if error is detected.
1335  */
1336 static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
1337 			   bool *writable, kvm_pfn_t *pfn)
1338 {
1339 	struct page *page[1];
1340 	int npages = 0;
1341 
1342 	might_sleep();
1343 
1344 	if (writable)
1345 		*writable = write_fault;
1346 
1347 	if (async) {
1348 		down_read(&current->mm->mmap_sem);
1349 		npages = get_user_page_nowait(addr, write_fault, page);
1350 		up_read(&current->mm->mmap_sem);
1351 	} else
1352 		npages = __get_user_pages_unlocked(current, current->mm, addr, 1,
1353 						   write_fault, 0, page,
1354 						   FOLL_TOUCH|FOLL_HWPOISON);
1355 	if (npages != 1)
1356 		return npages;
1357 
1358 	/* map read fault as writable if possible */
1359 	if (unlikely(!write_fault) && writable) {
1360 		struct page *wpage[1];
1361 
1362 		npages = __get_user_pages_fast(addr, 1, 1, wpage);
1363 		if (npages == 1) {
1364 			*writable = true;
1365 			put_page(page[0]);
1366 			page[0] = wpage[0];
1367 		}
1368 
1369 		npages = 1;
1370 	}
1371 	*pfn = page_to_pfn(page[0]);
1372 	return npages;
1373 }
1374 
1375 static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
1376 {
1377 	if (unlikely(!(vma->vm_flags & VM_READ)))
1378 		return false;
1379 
1380 	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
1381 		return false;
1382 
1383 	return true;
1384 }
1385 
1386 /*
1387  * Pin guest page in memory and return its pfn.
1388  * @addr: host virtual address which maps memory to the guest
1389  * @atomic: whether this function can sleep
1390  * @async: whether this function need to wait IO complete if the
1391  *         host page is not in the memory
1392  * @write_fault: whether we should get a writable host page
1393  * @writable: whether it allows to map a writable host page for !@write_fault
1394  *
1395  * The function will map a writable host page for these two cases:
1396  * 1): @write_fault = true
1397  * 2): @write_fault = false && @writable, @writable will tell the caller
1398  *     whether the mapping is writable.
1399  */
1400 static kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool *async,
1401 			bool write_fault, bool *writable)
1402 {
1403 	struct vm_area_struct *vma;
1404 	kvm_pfn_t pfn = 0;
1405 	int npages;
1406 
1407 	/* we can do it either atomically or asynchronously, not both */
1408 	BUG_ON(atomic && async);
1409 
1410 	if (hva_to_pfn_fast(addr, atomic, async, write_fault, writable, &pfn))
1411 		return pfn;
1412 
1413 	if (atomic)
1414 		return KVM_PFN_ERR_FAULT;
1415 
1416 	npages = hva_to_pfn_slow(addr, async, write_fault, writable, &pfn);
1417 	if (npages == 1)
1418 		return pfn;
1419 
1420 	down_read(&current->mm->mmap_sem);
1421 	if (npages == -EHWPOISON ||
1422 	      (!async && check_user_page_hwpoison(addr))) {
1423 		pfn = KVM_PFN_ERR_HWPOISON;
1424 		goto exit;
1425 	}
1426 
1427 	vma = find_vma_intersection(current->mm, addr, addr + 1);
1428 
1429 	if (vma == NULL)
1430 		pfn = KVM_PFN_ERR_FAULT;
1431 	else if ((vma->vm_flags & VM_PFNMAP)) {
1432 		pfn = ((addr - vma->vm_start) >> PAGE_SHIFT) +
1433 			vma->vm_pgoff;
1434 		BUG_ON(!kvm_is_reserved_pfn(pfn));
1435 	} else {
1436 		if (async && vma_is_valid(vma, write_fault))
1437 			*async = true;
1438 		pfn = KVM_PFN_ERR_FAULT;
1439 	}
1440 exit:
1441 	up_read(&current->mm->mmap_sem);
1442 	return pfn;
1443 }
1444 
1445 kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
1446 			       bool atomic, bool *async, bool write_fault,
1447 			       bool *writable)
1448 {
1449 	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
1450 
1451 	if (addr == KVM_HVA_ERR_RO_BAD) {
1452 		if (writable)
1453 			*writable = false;
1454 		return KVM_PFN_ERR_RO_FAULT;
1455 	}
1456 
1457 	if (kvm_is_error_hva(addr)) {
1458 		if (writable)
1459 			*writable = false;
1460 		return KVM_PFN_NOSLOT;
1461 	}
1462 
1463 	/* Do not map writable pfn in the readonly memslot. */
1464 	if (writable && memslot_is_readonly(slot)) {
1465 		*writable = false;
1466 		writable = NULL;
1467 	}
1468 
1469 	return hva_to_pfn(addr, atomic, async, write_fault,
1470 			  writable);
1471 }
1472 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
1473 
1474 kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
1475 		      bool *writable)
1476 {
1477 	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, NULL,
1478 				    write_fault, writable);
1479 }
1480 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
1481 
1482 kvm_pfn_t gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn)
1483 {
1484 	return __gfn_to_pfn_memslot(slot, gfn, false, NULL, true, NULL);
1485 }
1486 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
1487 
1488 kvm_pfn_t gfn_to_pfn_memslot_atomic(struct kvm_memory_slot *slot, gfn_t gfn)
1489 {
1490 	return __gfn_to_pfn_memslot(slot, gfn, true, NULL, true, NULL);
1491 }
1492 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
1493 
1494 kvm_pfn_t gfn_to_pfn_atomic(struct kvm *kvm, gfn_t gfn)
1495 {
1496 	return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm, gfn), gfn);
1497 }
1498 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic);
1499 
1500 kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
1501 {
1502 	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1503 }
1504 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
1505 
1506 kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
1507 {
1508 	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
1509 }
1510 EXPORT_SYMBOL_GPL(gfn_to_pfn);
1511 
1512 kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
1513 {
1514 	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
1515 }
1516 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
1517 
1518 int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1519 			    struct page **pages, int nr_pages)
1520 {
1521 	unsigned long addr;
1522 	gfn_t entry;
1523 
1524 	addr = gfn_to_hva_many(slot, gfn, &entry);
1525 	if (kvm_is_error_hva(addr))
1526 		return -1;
1527 
1528 	if (entry < nr_pages)
1529 		return 0;
1530 
1531 	return __get_user_pages_fast(addr, nr_pages, 1, pages);
1532 }
1533 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
1534 
1535 static struct page *kvm_pfn_to_page(kvm_pfn_t pfn)
1536 {
1537 	if (is_error_noslot_pfn(pfn))
1538 		return KVM_ERR_PTR_BAD_PAGE;
1539 
1540 	if (kvm_is_reserved_pfn(pfn)) {
1541 		WARN_ON(1);
1542 		return KVM_ERR_PTR_BAD_PAGE;
1543 	}
1544 
1545 	return pfn_to_page(pfn);
1546 }
1547 
1548 struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
1549 {
1550 	kvm_pfn_t pfn;
1551 
1552 	pfn = gfn_to_pfn(kvm, gfn);
1553 
1554 	return kvm_pfn_to_page(pfn);
1555 }
1556 EXPORT_SYMBOL_GPL(gfn_to_page);
1557 
1558 struct page *kvm_vcpu_gfn_to_page(struct kvm_vcpu *vcpu, gfn_t gfn)
1559 {
1560 	kvm_pfn_t pfn;
1561 
1562 	pfn = kvm_vcpu_gfn_to_pfn(vcpu, gfn);
1563 
1564 	return kvm_pfn_to_page(pfn);
1565 }
1566 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page);
1567 
1568 void kvm_release_page_clean(struct page *page)
1569 {
1570 	WARN_ON(is_error_page(page));
1571 
1572 	kvm_release_pfn_clean(page_to_pfn(page));
1573 }
1574 EXPORT_SYMBOL_GPL(kvm_release_page_clean);
1575 
1576 void kvm_release_pfn_clean(kvm_pfn_t pfn)
1577 {
1578 	if (!is_error_noslot_pfn(pfn) && !kvm_is_reserved_pfn(pfn))
1579 		put_page(pfn_to_page(pfn));
1580 }
1581 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
1582 
1583 void kvm_release_page_dirty(struct page *page)
1584 {
1585 	WARN_ON(is_error_page(page));
1586 
1587 	kvm_release_pfn_dirty(page_to_pfn(page));
1588 }
1589 EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
1590 
1591 static void kvm_release_pfn_dirty(kvm_pfn_t pfn)
1592 {
1593 	kvm_set_pfn_dirty(pfn);
1594 	kvm_release_pfn_clean(pfn);
1595 }
1596 
1597 void kvm_set_pfn_dirty(kvm_pfn_t pfn)
1598 {
1599 	if (!kvm_is_reserved_pfn(pfn)) {
1600 		struct page *page = pfn_to_page(pfn);
1601 
1602 		if (!PageReserved(page))
1603 			SetPageDirty(page);
1604 	}
1605 }
1606 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
1607 
1608 void kvm_set_pfn_accessed(kvm_pfn_t pfn)
1609 {
1610 	if (!kvm_is_reserved_pfn(pfn))
1611 		mark_page_accessed(pfn_to_page(pfn));
1612 }
1613 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
1614 
1615 void kvm_get_pfn(kvm_pfn_t pfn)
1616 {
1617 	if (!kvm_is_reserved_pfn(pfn))
1618 		get_page(pfn_to_page(pfn));
1619 }
1620 EXPORT_SYMBOL_GPL(kvm_get_pfn);
1621 
1622 static int next_segment(unsigned long len, int offset)
1623 {
1624 	if (len > PAGE_SIZE - offset)
1625 		return PAGE_SIZE - offset;
1626 	else
1627 		return len;
1628 }
1629 
1630 static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
1631 				 void *data, int offset, int len)
1632 {
1633 	int r;
1634 	unsigned long addr;
1635 
1636 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1637 	if (kvm_is_error_hva(addr))
1638 		return -EFAULT;
1639 	r = __copy_from_user(data, (void __user *)addr + offset, len);
1640 	if (r)
1641 		return -EFAULT;
1642 	return 0;
1643 }
1644 
1645 int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
1646 			int len)
1647 {
1648 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1649 
1650 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1651 }
1652 EXPORT_SYMBOL_GPL(kvm_read_guest_page);
1653 
1654 int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
1655 			     int offset, int len)
1656 {
1657 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1658 
1659 	return __kvm_read_guest_page(slot, gfn, data, offset, len);
1660 }
1661 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
1662 
1663 int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
1664 {
1665 	gfn_t gfn = gpa >> PAGE_SHIFT;
1666 	int seg;
1667 	int offset = offset_in_page(gpa);
1668 	int ret;
1669 
1670 	while ((seg = next_segment(len, offset)) != 0) {
1671 		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
1672 		if (ret < 0)
1673 			return ret;
1674 		offset = 0;
1675 		len -= seg;
1676 		data += seg;
1677 		++gfn;
1678 	}
1679 	return 0;
1680 }
1681 EXPORT_SYMBOL_GPL(kvm_read_guest);
1682 
1683 int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
1684 {
1685 	gfn_t gfn = gpa >> PAGE_SHIFT;
1686 	int seg;
1687 	int offset = offset_in_page(gpa);
1688 	int ret;
1689 
1690 	while ((seg = next_segment(len, offset)) != 0) {
1691 		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
1692 		if (ret < 0)
1693 			return ret;
1694 		offset = 0;
1695 		len -= seg;
1696 		data += seg;
1697 		++gfn;
1698 	}
1699 	return 0;
1700 }
1701 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
1702 
1703 static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
1704 			           void *data, int offset, unsigned long len)
1705 {
1706 	int r;
1707 	unsigned long addr;
1708 
1709 	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
1710 	if (kvm_is_error_hva(addr))
1711 		return -EFAULT;
1712 	pagefault_disable();
1713 	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
1714 	pagefault_enable();
1715 	if (r)
1716 		return -EFAULT;
1717 	return 0;
1718 }
1719 
1720 int kvm_read_guest_atomic(struct kvm *kvm, gpa_t gpa, void *data,
1721 			  unsigned long len)
1722 {
1723 	gfn_t gfn = gpa >> PAGE_SHIFT;
1724 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1725 	int offset = offset_in_page(gpa);
1726 
1727 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1728 }
1729 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic);
1730 
1731 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
1732 			       void *data, unsigned long len)
1733 {
1734 	gfn_t gfn = gpa >> PAGE_SHIFT;
1735 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1736 	int offset = offset_in_page(gpa);
1737 
1738 	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
1739 }
1740 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
1741 
1742 static int __kvm_write_guest_page(struct kvm_memory_slot *memslot, gfn_t gfn,
1743 			          const void *data, int offset, int len)
1744 {
1745 	int r;
1746 	unsigned long addr;
1747 
1748 	addr = gfn_to_hva_memslot(memslot, gfn);
1749 	if (kvm_is_error_hva(addr))
1750 		return -EFAULT;
1751 	r = __copy_to_user((void __user *)addr + offset, data, len);
1752 	if (r)
1753 		return -EFAULT;
1754 	mark_page_dirty_in_slot(memslot, gfn);
1755 	return 0;
1756 }
1757 
1758 int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
1759 			 const void *data, int offset, int len)
1760 {
1761 	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
1762 
1763 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1764 }
1765 EXPORT_SYMBOL_GPL(kvm_write_guest_page);
1766 
1767 int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
1768 			      const void *data, int offset, int len)
1769 {
1770 	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1771 
1772 	return __kvm_write_guest_page(slot, gfn, data, offset, len);
1773 }
1774 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
1775 
1776 int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
1777 		    unsigned long len)
1778 {
1779 	gfn_t gfn = gpa >> PAGE_SHIFT;
1780 	int seg;
1781 	int offset = offset_in_page(gpa);
1782 	int ret;
1783 
1784 	while ((seg = next_segment(len, offset)) != 0) {
1785 		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
1786 		if (ret < 0)
1787 			return ret;
1788 		offset = 0;
1789 		len -= seg;
1790 		data += seg;
1791 		++gfn;
1792 	}
1793 	return 0;
1794 }
1795 EXPORT_SYMBOL_GPL(kvm_write_guest);
1796 
1797 int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
1798 		         unsigned long len)
1799 {
1800 	gfn_t gfn = gpa >> PAGE_SHIFT;
1801 	int seg;
1802 	int offset = offset_in_page(gpa);
1803 	int ret;
1804 
1805 	while ((seg = next_segment(len, offset)) != 0) {
1806 		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
1807 		if (ret < 0)
1808 			return ret;
1809 		offset = 0;
1810 		len -= seg;
1811 		data += seg;
1812 		++gfn;
1813 	}
1814 	return 0;
1815 }
1816 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
1817 
1818 int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1819 			      gpa_t gpa, unsigned long len)
1820 {
1821 	struct kvm_memslots *slots = kvm_memslots(kvm);
1822 	int offset = offset_in_page(gpa);
1823 	gfn_t start_gfn = gpa >> PAGE_SHIFT;
1824 	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
1825 	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
1826 	gfn_t nr_pages_avail;
1827 
1828 	ghc->gpa = gpa;
1829 	ghc->generation = slots->generation;
1830 	ghc->len = len;
1831 	ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1832 	ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn, NULL);
1833 	if (!kvm_is_error_hva(ghc->hva) && nr_pages_needed <= 1) {
1834 		ghc->hva += offset;
1835 	} else {
1836 		/*
1837 		 * If the requested region crosses two memslots, we still
1838 		 * verify that the entire region is valid here.
1839 		 */
1840 		while (start_gfn <= end_gfn) {
1841 			ghc->memslot = gfn_to_memslot(kvm, start_gfn);
1842 			ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
1843 						   &nr_pages_avail);
1844 			if (kvm_is_error_hva(ghc->hva))
1845 				return -EFAULT;
1846 			start_gfn += nr_pages_avail;
1847 		}
1848 		/* Use the slow path for cross page reads and writes. */
1849 		ghc->memslot = NULL;
1850 	}
1851 	return 0;
1852 }
1853 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
1854 
1855 int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1856 			   void *data, unsigned long len)
1857 {
1858 	struct kvm_memslots *slots = kvm_memslots(kvm);
1859 	int r;
1860 
1861 	BUG_ON(len > ghc->len);
1862 
1863 	if (slots->generation != ghc->generation)
1864 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1865 
1866 	if (unlikely(!ghc->memslot))
1867 		return kvm_write_guest(kvm, ghc->gpa, data, len);
1868 
1869 	if (kvm_is_error_hva(ghc->hva))
1870 		return -EFAULT;
1871 
1872 	r = __copy_to_user((void __user *)ghc->hva, data, len);
1873 	if (r)
1874 		return -EFAULT;
1875 	mark_page_dirty_in_slot(ghc->memslot, ghc->gpa >> PAGE_SHIFT);
1876 
1877 	return 0;
1878 }
1879 EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
1880 
1881 int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
1882 			   void *data, unsigned long len)
1883 {
1884 	struct kvm_memslots *slots = kvm_memslots(kvm);
1885 	int r;
1886 
1887 	BUG_ON(len > ghc->len);
1888 
1889 	if (slots->generation != ghc->generation)
1890 		kvm_gfn_to_hva_cache_init(kvm, ghc, ghc->gpa, ghc->len);
1891 
1892 	if (unlikely(!ghc->memslot))
1893 		return kvm_read_guest(kvm, ghc->gpa, data, len);
1894 
1895 	if (kvm_is_error_hva(ghc->hva))
1896 		return -EFAULT;
1897 
1898 	r = __copy_from_user(data, (void __user *)ghc->hva, len);
1899 	if (r)
1900 		return -EFAULT;
1901 
1902 	return 0;
1903 }
1904 EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
1905 
1906 int kvm_clear_guest_page(struct kvm *kvm, gfn_t gfn, int offset, int len)
1907 {
1908 	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
1909 
1910 	return kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
1911 }
1912 EXPORT_SYMBOL_GPL(kvm_clear_guest_page);
1913 
1914 int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
1915 {
1916 	gfn_t gfn = gpa >> PAGE_SHIFT;
1917 	int seg;
1918 	int offset = offset_in_page(gpa);
1919 	int ret;
1920 
1921 	while ((seg = next_segment(len, offset)) != 0) {
1922 		ret = kvm_clear_guest_page(kvm, gfn, offset, seg);
1923 		if (ret < 0)
1924 			return ret;
1925 		offset = 0;
1926 		len -= seg;
1927 		++gfn;
1928 	}
1929 	return 0;
1930 }
1931 EXPORT_SYMBOL_GPL(kvm_clear_guest);
1932 
1933 static void mark_page_dirty_in_slot(struct kvm_memory_slot *memslot,
1934 				    gfn_t gfn)
1935 {
1936 	if (memslot && memslot->dirty_bitmap) {
1937 		unsigned long rel_gfn = gfn - memslot->base_gfn;
1938 
1939 		set_bit_le(rel_gfn, memslot->dirty_bitmap);
1940 	}
1941 }
1942 
1943 void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
1944 {
1945 	struct kvm_memory_slot *memslot;
1946 
1947 	memslot = gfn_to_memslot(kvm, gfn);
1948 	mark_page_dirty_in_slot(memslot, gfn);
1949 }
1950 EXPORT_SYMBOL_GPL(mark_page_dirty);
1951 
1952 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
1953 {
1954 	struct kvm_memory_slot *memslot;
1955 
1956 	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
1957 	mark_page_dirty_in_slot(memslot, gfn);
1958 }
1959 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
1960 
1961 static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
1962 {
1963 	unsigned int old, val, grow;
1964 
1965 	old = val = vcpu->halt_poll_ns;
1966 	grow = READ_ONCE(halt_poll_ns_grow);
1967 	/* 10us base */
1968 	if (val == 0 && grow)
1969 		val = 10000;
1970 	else
1971 		val *= grow;
1972 
1973 	if (val > halt_poll_ns)
1974 		val = halt_poll_ns;
1975 
1976 	vcpu->halt_poll_ns = val;
1977 	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
1978 }
1979 
1980 static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
1981 {
1982 	unsigned int old, val, shrink;
1983 
1984 	old = val = vcpu->halt_poll_ns;
1985 	shrink = READ_ONCE(halt_poll_ns_shrink);
1986 	if (shrink == 0)
1987 		val = 0;
1988 	else
1989 		val /= shrink;
1990 
1991 	vcpu->halt_poll_ns = val;
1992 	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
1993 }
1994 
1995 static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
1996 {
1997 	if (kvm_arch_vcpu_runnable(vcpu)) {
1998 		kvm_make_request(KVM_REQ_UNHALT, vcpu);
1999 		return -EINTR;
2000 	}
2001 	if (kvm_cpu_has_pending_timer(vcpu))
2002 		return -EINTR;
2003 	if (signal_pending(current))
2004 		return -EINTR;
2005 
2006 	return 0;
2007 }
2008 
2009 /*
2010  * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2011  */
2012 void kvm_vcpu_block(struct kvm_vcpu *vcpu)
2013 {
2014 	ktime_t start, cur;
2015 	DECLARE_SWAITQUEUE(wait);
2016 	bool waited = false;
2017 	u64 block_ns;
2018 
2019 	start = cur = ktime_get();
2020 	if (vcpu->halt_poll_ns) {
2021 		ktime_t stop = ktime_add_ns(ktime_get(), vcpu->halt_poll_ns);
2022 
2023 		++vcpu->stat.halt_attempted_poll;
2024 		do {
2025 			/*
2026 			 * This sets KVM_REQ_UNHALT if an interrupt
2027 			 * arrives.
2028 			 */
2029 			if (kvm_vcpu_check_block(vcpu) < 0) {
2030 				++vcpu->stat.halt_successful_poll;
2031 				goto out;
2032 			}
2033 			cur = ktime_get();
2034 		} while (single_task_running() && ktime_before(cur, stop));
2035 	}
2036 
2037 	kvm_arch_vcpu_blocking(vcpu);
2038 
2039 	for (;;) {
2040 		prepare_to_swait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
2041 
2042 		if (kvm_vcpu_check_block(vcpu) < 0)
2043 			break;
2044 
2045 		waited = true;
2046 		schedule();
2047 	}
2048 
2049 	finish_swait(&vcpu->wq, &wait);
2050 	cur = ktime_get();
2051 
2052 	kvm_arch_vcpu_unblocking(vcpu);
2053 out:
2054 	block_ns = ktime_to_ns(cur) - ktime_to_ns(start);
2055 
2056 	if (halt_poll_ns) {
2057 		if (block_ns <= vcpu->halt_poll_ns)
2058 			;
2059 		/* we had a long block, shrink polling */
2060 		else if (vcpu->halt_poll_ns && block_ns > halt_poll_ns)
2061 			shrink_halt_poll_ns(vcpu);
2062 		/* we had a short halt and our poll time is too small */
2063 		else if (vcpu->halt_poll_ns < halt_poll_ns &&
2064 			block_ns < halt_poll_ns)
2065 			grow_halt_poll_ns(vcpu);
2066 	} else
2067 		vcpu->halt_poll_ns = 0;
2068 
2069 	trace_kvm_vcpu_wakeup(block_ns, waited);
2070 }
2071 EXPORT_SYMBOL_GPL(kvm_vcpu_block);
2072 
2073 #ifndef CONFIG_S390
2074 /*
2075  * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2076  */
2077 void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
2078 {
2079 	int me;
2080 	int cpu = vcpu->cpu;
2081 	struct swait_queue_head *wqp;
2082 
2083 	wqp = kvm_arch_vcpu_wq(vcpu);
2084 	if (swait_active(wqp)) {
2085 		swake_up(wqp);
2086 		++vcpu->stat.halt_wakeup;
2087 	}
2088 
2089 	me = get_cpu();
2090 	if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
2091 		if (kvm_arch_vcpu_should_kick(vcpu))
2092 			smp_send_reschedule(cpu);
2093 	put_cpu();
2094 }
2095 EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
2096 #endif /* !CONFIG_S390 */
2097 
2098 int kvm_vcpu_yield_to(struct kvm_vcpu *target)
2099 {
2100 	struct pid *pid;
2101 	struct task_struct *task = NULL;
2102 	int ret = 0;
2103 
2104 	rcu_read_lock();
2105 	pid = rcu_dereference(target->pid);
2106 	if (pid)
2107 		task = get_pid_task(pid, PIDTYPE_PID);
2108 	rcu_read_unlock();
2109 	if (!task)
2110 		return ret;
2111 	ret = yield_to(task, 1);
2112 	put_task_struct(task);
2113 
2114 	return ret;
2115 }
2116 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
2117 
2118 /*
2119  * Helper that checks whether a VCPU is eligible for directed yield.
2120  * Most eligible candidate to yield is decided by following heuristics:
2121  *
2122  *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2123  *  (preempted lock holder), indicated by @in_spin_loop.
2124  *  Set at the beiginning and cleared at the end of interception/PLE handler.
2125  *
2126  *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2127  *  chance last time (mostly it has become eligible now since we have probably
2128  *  yielded to lockholder in last iteration. This is done by toggling
2129  *  @dy_eligible each time a VCPU checked for eligibility.)
2130  *
2131  *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2132  *  to preempted lock-holder could result in wrong VCPU selection and CPU
2133  *  burning. Giving priority for a potential lock-holder increases lock
2134  *  progress.
2135  *
2136  *  Since algorithm is based on heuristics, accessing another VCPU data without
2137  *  locking does not harm. It may result in trying to yield to  same VCPU, fail
2138  *  and continue with next VCPU and so on.
2139  */
2140 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
2141 {
2142 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2143 	bool eligible;
2144 
2145 	eligible = !vcpu->spin_loop.in_spin_loop ||
2146 		    vcpu->spin_loop.dy_eligible;
2147 
2148 	if (vcpu->spin_loop.in_spin_loop)
2149 		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
2150 
2151 	return eligible;
2152 #else
2153 	return true;
2154 #endif
2155 }
2156 
2157 void kvm_vcpu_on_spin(struct kvm_vcpu *me)
2158 {
2159 	struct kvm *kvm = me->kvm;
2160 	struct kvm_vcpu *vcpu;
2161 	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
2162 	int yielded = 0;
2163 	int try = 3;
2164 	int pass;
2165 	int i;
2166 
2167 	kvm_vcpu_set_in_spin_loop(me, true);
2168 	/*
2169 	 * We boost the priority of a VCPU that is runnable but not
2170 	 * currently running, because it got preempted by something
2171 	 * else and called schedule in __vcpu_run.  Hopefully that
2172 	 * VCPU is holding the lock that we need and will release it.
2173 	 * We approximate round-robin by starting at the last boosted VCPU.
2174 	 */
2175 	for (pass = 0; pass < 2 && !yielded && try; pass++) {
2176 		kvm_for_each_vcpu(i, vcpu, kvm) {
2177 			if (!pass && i <= last_boosted_vcpu) {
2178 				i = last_boosted_vcpu;
2179 				continue;
2180 			} else if (pass && i > last_boosted_vcpu)
2181 				break;
2182 			if (!ACCESS_ONCE(vcpu->preempted))
2183 				continue;
2184 			if (vcpu == me)
2185 				continue;
2186 			if (swait_active(&vcpu->wq) && !kvm_arch_vcpu_runnable(vcpu))
2187 				continue;
2188 			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
2189 				continue;
2190 
2191 			yielded = kvm_vcpu_yield_to(vcpu);
2192 			if (yielded > 0) {
2193 				kvm->last_boosted_vcpu = i;
2194 				break;
2195 			} else if (yielded < 0) {
2196 				try--;
2197 				if (!try)
2198 					break;
2199 			}
2200 		}
2201 	}
2202 	kvm_vcpu_set_in_spin_loop(me, false);
2203 
2204 	/* Ensure vcpu is not eligible during next spinloop */
2205 	kvm_vcpu_set_dy_eligible(me, false);
2206 }
2207 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
2208 
2209 static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2210 {
2211 	struct kvm_vcpu *vcpu = vma->vm_file->private_data;
2212 	struct page *page;
2213 
2214 	if (vmf->pgoff == 0)
2215 		page = virt_to_page(vcpu->run);
2216 #ifdef CONFIG_X86
2217 	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
2218 		page = virt_to_page(vcpu->arch.pio_data);
2219 #endif
2220 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2221 	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
2222 		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
2223 #endif
2224 	else
2225 		return kvm_arch_vcpu_fault(vcpu, vmf);
2226 	get_page(page);
2227 	vmf->page = page;
2228 	return 0;
2229 }
2230 
2231 static const struct vm_operations_struct kvm_vcpu_vm_ops = {
2232 	.fault = kvm_vcpu_fault,
2233 };
2234 
2235 static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
2236 {
2237 	vma->vm_ops = &kvm_vcpu_vm_ops;
2238 	return 0;
2239 }
2240 
2241 static int kvm_vcpu_release(struct inode *inode, struct file *filp)
2242 {
2243 	struct kvm_vcpu *vcpu = filp->private_data;
2244 
2245 	kvm_put_kvm(vcpu->kvm);
2246 	return 0;
2247 }
2248 
2249 static struct file_operations kvm_vcpu_fops = {
2250 	.release        = kvm_vcpu_release,
2251 	.unlocked_ioctl = kvm_vcpu_ioctl,
2252 #ifdef CONFIG_KVM_COMPAT
2253 	.compat_ioctl   = kvm_vcpu_compat_ioctl,
2254 #endif
2255 	.mmap           = kvm_vcpu_mmap,
2256 	.llseek		= noop_llseek,
2257 };
2258 
2259 /*
2260  * Allocates an inode for the vcpu.
2261  */
2262 static int create_vcpu_fd(struct kvm_vcpu *vcpu)
2263 {
2264 	return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
2265 }
2266 
2267 /*
2268  * Creates some virtual cpus.  Good luck creating more than one.
2269  */
2270 static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
2271 {
2272 	int r;
2273 	struct kvm_vcpu *vcpu;
2274 
2275 	if (id >= KVM_MAX_VCPUS)
2276 		return -EINVAL;
2277 
2278 	vcpu = kvm_arch_vcpu_create(kvm, id);
2279 	if (IS_ERR(vcpu))
2280 		return PTR_ERR(vcpu);
2281 
2282 	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
2283 
2284 	r = kvm_arch_vcpu_setup(vcpu);
2285 	if (r)
2286 		goto vcpu_destroy;
2287 
2288 	mutex_lock(&kvm->lock);
2289 	if (!kvm_vcpu_compatible(vcpu)) {
2290 		r = -EINVAL;
2291 		goto unlock_vcpu_destroy;
2292 	}
2293 	if (atomic_read(&kvm->online_vcpus) == KVM_MAX_VCPUS) {
2294 		r = -EINVAL;
2295 		goto unlock_vcpu_destroy;
2296 	}
2297 	if (kvm_get_vcpu_by_id(kvm, id)) {
2298 		r = -EEXIST;
2299 		goto unlock_vcpu_destroy;
2300 	}
2301 
2302 	BUG_ON(kvm->vcpus[atomic_read(&kvm->online_vcpus)]);
2303 
2304 	/* Now it's all set up, let userspace reach it */
2305 	kvm_get_kvm(kvm);
2306 	r = create_vcpu_fd(vcpu);
2307 	if (r < 0) {
2308 		kvm_put_kvm(kvm);
2309 		goto unlock_vcpu_destroy;
2310 	}
2311 
2312 	kvm->vcpus[atomic_read(&kvm->online_vcpus)] = vcpu;
2313 
2314 	/*
2315 	 * Pairs with smp_rmb() in kvm_get_vcpu.  Write kvm->vcpus
2316 	 * before kvm->online_vcpu's incremented value.
2317 	 */
2318 	smp_wmb();
2319 	atomic_inc(&kvm->online_vcpus);
2320 
2321 	mutex_unlock(&kvm->lock);
2322 	kvm_arch_vcpu_postcreate(vcpu);
2323 	return r;
2324 
2325 unlock_vcpu_destroy:
2326 	mutex_unlock(&kvm->lock);
2327 vcpu_destroy:
2328 	kvm_arch_vcpu_destroy(vcpu);
2329 	return r;
2330 }
2331 
2332 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
2333 {
2334 	if (sigset) {
2335 		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
2336 		vcpu->sigset_active = 1;
2337 		vcpu->sigset = *sigset;
2338 	} else
2339 		vcpu->sigset_active = 0;
2340 	return 0;
2341 }
2342 
2343 static long kvm_vcpu_ioctl(struct file *filp,
2344 			   unsigned int ioctl, unsigned long arg)
2345 {
2346 	struct kvm_vcpu *vcpu = filp->private_data;
2347 	void __user *argp = (void __user *)arg;
2348 	int r;
2349 	struct kvm_fpu *fpu = NULL;
2350 	struct kvm_sregs *kvm_sregs = NULL;
2351 
2352 	if (vcpu->kvm->mm != current->mm)
2353 		return -EIO;
2354 
2355 	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
2356 		return -EINVAL;
2357 
2358 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2359 	/*
2360 	 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2361 	 * so vcpu_load() would break it.
2362 	 */
2363 	if (ioctl == KVM_S390_INTERRUPT || ioctl == KVM_S390_IRQ || ioctl == KVM_INTERRUPT)
2364 		return kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2365 #endif
2366 
2367 
2368 	r = vcpu_load(vcpu);
2369 	if (r)
2370 		return r;
2371 	switch (ioctl) {
2372 	case KVM_RUN:
2373 		r = -EINVAL;
2374 		if (arg)
2375 			goto out;
2376 		if (unlikely(vcpu->pid != current->pids[PIDTYPE_PID].pid)) {
2377 			/* The thread running this VCPU changed. */
2378 			struct pid *oldpid = vcpu->pid;
2379 			struct pid *newpid = get_task_pid(current, PIDTYPE_PID);
2380 
2381 			rcu_assign_pointer(vcpu->pid, newpid);
2382 			if (oldpid)
2383 				synchronize_rcu();
2384 			put_pid(oldpid);
2385 		}
2386 		r = kvm_arch_vcpu_ioctl_run(vcpu, vcpu->run);
2387 		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
2388 		break;
2389 	case KVM_GET_REGS: {
2390 		struct kvm_regs *kvm_regs;
2391 
2392 		r = -ENOMEM;
2393 		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL);
2394 		if (!kvm_regs)
2395 			goto out;
2396 		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
2397 		if (r)
2398 			goto out_free1;
2399 		r = -EFAULT;
2400 		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
2401 			goto out_free1;
2402 		r = 0;
2403 out_free1:
2404 		kfree(kvm_regs);
2405 		break;
2406 	}
2407 	case KVM_SET_REGS: {
2408 		struct kvm_regs *kvm_regs;
2409 
2410 		r = -ENOMEM;
2411 		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
2412 		if (IS_ERR(kvm_regs)) {
2413 			r = PTR_ERR(kvm_regs);
2414 			goto out;
2415 		}
2416 		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
2417 		kfree(kvm_regs);
2418 		break;
2419 	}
2420 	case KVM_GET_SREGS: {
2421 		kvm_sregs = kzalloc(sizeof(struct kvm_sregs), GFP_KERNEL);
2422 		r = -ENOMEM;
2423 		if (!kvm_sregs)
2424 			goto out;
2425 		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
2426 		if (r)
2427 			goto out;
2428 		r = -EFAULT;
2429 		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
2430 			goto out;
2431 		r = 0;
2432 		break;
2433 	}
2434 	case KVM_SET_SREGS: {
2435 		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
2436 		if (IS_ERR(kvm_sregs)) {
2437 			r = PTR_ERR(kvm_sregs);
2438 			kvm_sregs = NULL;
2439 			goto out;
2440 		}
2441 		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
2442 		break;
2443 	}
2444 	case KVM_GET_MP_STATE: {
2445 		struct kvm_mp_state mp_state;
2446 
2447 		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
2448 		if (r)
2449 			goto out;
2450 		r = -EFAULT;
2451 		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
2452 			goto out;
2453 		r = 0;
2454 		break;
2455 	}
2456 	case KVM_SET_MP_STATE: {
2457 		struct kvm_mp_state mp_state;
2458 
2459 		r = -EFAULT;
2460 		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
2461 			goto out;
2462 		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
2463 		break;
2464 	}
2465 	case KVM_TRANSLATE: {
2466 		struct kvm_translation tr;
2467 
2468 		r = -EFAULT;
2469 		if (copy_from_user(&tr, argp, sizeof(tr)))
2470 			goto out;
2471 		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
2472 		if (r)
2473 			goto out;
2474 		r = -EFAULT;
2475 		if (copy_to_user(argp, &tr, sizeof(tr)))
2476 			goto out;
2477 		r = 0;
2478 		break;
2479 	}
2480 	case KVM_SET_GUEST_DEBUG: {
2481 		struct kvm_guest_debug dbg;
2482 
2483 		r = -EFAULT;
2484 		if (copy_from_user(&dbg, argp, sizeof(dbg)))
2485 			goto out;
2486 		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
2487 		break;
2488 	}
2489 	case KVM_SET_SIGNAL_MASK: {
2490 		struct kvm_signal_mask __user *sigmask_arg = argp;
2491 		struct kvm_signal_mask kvm_sigmask;
2492 		sigset_t sigset, *p;
2493 
2494 		p = NULL;
2495 		if (argp) {
2496 			r = -EFAULT;
2497 			if (copy_from_user(&kvm_sigmask, argp,
2498 					   sizeof(kvm_sigmask)))
2499 				goto out;
2500 			r = -EINVAL;
2501 			if (kvm_sigmask.len != sizeof(sigset))
2502 				goto out;
2503 			r = -EFAULT;
2504 			if (copy_from_user(&sigset, sigmask_arg->sigset,
2505 					   sizeof(sigset)))
2506 				goto out;
2507 			p = &sigset;
2508 		}
2509 		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
2510 		break;
2511 	}
2512 	case KVM_GET_FPU: {
2513 		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL);
2514 		r = -ENOMEM;
2515 		if (!fpu)
2516 			goto out;
2517 		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
2518 		if (r)
2519 			goto out;
2520 		r = -EFAULT;
2521 		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
2522 			goto out;
2523 		r = 0;
2524 		break;
2525 	}
2526 	case KVM_SET_FPU: {
2527 		fpu = memdup_user(argp, sizeof(*fpu));
2528 		if (IS_ERR(fpu)) {
2529 			r = PTR_ERR(fpu);
2530 			fpu = NULL;
2531 			goto out;
2532 		}
2533 		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
2534 		break;
2535 	}
2536 	default:
2537 		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
2538 	}
2539 out:
2540 	vcpu_put(vcpu);
2541 	kfree(fpu);
2542 	kfree(kvm_sregs);
2543 	return r;
2544 }
2545 
2546 #ifdef CONFIG_KVM_COMPAT
2547 static long kvm_vcpu_compat_ioctl(struct file *filp,
2548 				  unsigned int ioctl, unsigned long arg)
2549 {
2550 	struct kvm_vcpu *vcpu = filp->private_data;
2551 	void __user *argp = compat_ptr(arg);
2552 	int r;
2553 
2554 	if (vcpu->kvm->mm != current->mm)
2555 		return -EIO;
2556 
2557 	switch (ioctl) {
2558 	case KVM_SET_SIGNAL_MASK: {
2559 		struct kvm_signal_mask __user *sigmask_arg = argp;
2560 		struct kvm_signal_mask kvm_sigmask;
2561 		compat_sigset_t csigset;
2562 		sigset_t sigset;
2563 
2564 		if (argp) {
2565 			r = -EFAULT;
2566 			if (copy_from_user(&kvm_sigmask, argp,
2567 					   sizeof(kvm_sigmask)))
2568 				goto out;
2569 			r = -EINVAL;
2570 			if (kvm_sigmask.len != sizeof(csigset))
2571 				goto out;
2572 			r = -EFAULT;
2573 			if (copy_from_user(&csigset, sigmask_arg->sigset,
2574 					   sizeof(csigset)))
2575 				goto out;
2576 			sigset_from_compat(&sigset, &csigset);
2577 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
2578 		} else
2579 			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
2580 		break;
2581 	}
2582 	default:
2583 		r = kvm_vcpu_ioctl(filp, ioctl, arg);
2584 	}
2585 
2586 out:
2587 	return r;
2588 }
2589 #endif
2590 
2591 static int kvm_device_ioctl_attr(struct kvm_device *dev,
2592 				 int (*accessor)(struct kvm_device *dev,
2593 						 struct kvm_device_attr *attr),
2594 				 unsigned long arg)
2595 {
2596 	struct kvm_device_attr attr;
2597 
2598 	if (!accessor)
2599 		return -EPERM;
2600 
2601 	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
2602 		return -EFAULT;
2603 
2604 	return accessor(dev, &attr);
2605 }
2606 
2607 static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
2608 			     unsigned long arg)
2609 {
2610 	struct kvm_device *dev = filp->private_data;
2611 
2612 	switch (ioctl) {
2613 	case KVM_SET_DEVICE_ATTR:
2614 		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
2615 	case KVM_GET_DEVICE_ATTR:
2616 		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
2617 	case KVM_HAS_DEVICE_ATTR:
2618 		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
2619 	default:
2620 		if (dev->ops->ioctl)
2621 			return dev->ops->ioctl(dev, ioctl, arg);
2622 
2623 		return -ENOTTY;
2624 	}
2625 }
2626 
2627 static int kvm_device_release(struct inode *inode, struct file *filp)
2628 {
2629 	struct kvm_device *dev = filp->private_data;
2630 	struct kvm *kvm = dev->kvm;
2631 
2632 	kvm_put_kvm(kvm);
2633 	return 0;
2634 }
2635 
2636 static const struct file_operations kvm_device_fops = {
2637 	.unlocked_ioctl = kvm_device_ioctl,
2638 #ifdef CONFIG_KVM_COMPAT
2639 	.compat_ioctl = kvm_device_ioctl,
2640 #endif
2641 	.release = kvm_device_release,
2642 };
2643 
2644 struct kvm_device *kvm_device_from_filp(struct file *filp)
2645 {
2646 	if (filp->f_op != &kvm_device_fops)
2647 		return NULL;
2648 
2649 	return filp->private_data;
2650 }
2651 
2652 static struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
2653 #ifdef CONFIG_KVM_MPIC
2654 	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
2655 	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
2656 #endif
2657 
2658 #ifdef CONFIG_KVM_XICS
2659 	[KVM_DEV_TYPE_XICS]		= &kvm_xics_ops,
2660 #endif
2661 };
2662 
2663 int kvm_register_device_ops(struct kvm_device_ops *ops, u32 type)
2664 {
2665 	if (type >= ARRAY_SIZE(kvm_device_ops_table))
2666 		return -ENOSPC;
2667 
2668 	if (kvm_device_ops_table[type] != NULL)
2669 		return -EEXIST;
2670 
2671 	kvm_device_ops_table[type] = ops;
2672 	return 0;
2673 }
2674 
2675 void kvm_unregister_device_ops(u32 type)
2676 {
2677 	if (kvm_device_ops_table[type] != NULL)
2678 		kvm_device_ops_table[type] = NULL;
2679 }
2680 
2681 static int kvm_ioctl_create_device(struct kvm *kvm,
2682 				   struct kvm_create_device *cd)
2683 {
2684 	struct kvm_device_ops *ops = NULL;
2685 	struct kvm_device *dev;
2686 	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
2687 	int ret;
2688 
2689 	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
2690 		return -ENODEV;
2691 
2692 	ops = kvm_device_ops_table[cd->type];
2693 	if (ops == NULL)
2694 		return -ENODEV;
2695 
2696 	if (test)
2697 		return 0;
2698 
2699 	dev = kzalloc(sizeof(*dev), GFP_KERNEL);
2700 	if (!dev)
2701 		return -ENOMEM;
2702 
2703 	dev->ops = ops;
2704 	dev->kvm = kvm;
2705 
2706 	ret = ops->create(dev, cd->type);
2707 	if (ret < 0) {
2708 		kfree(dev);
2709 		return ret;
2710 	}
2711 
2712 	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
2713 	if (ret < 0) {
2714 		ops->destroy(dev);
2715 		return ret;
2716 	}
2717 
2718 	list_add(&dev->vm_node, &kvm->devices);
2719 	kvm_get_kvm(kvm);
2720 	cd->fd = ret;
2721 	return 0;
2722 }
2723 
2724 static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
2725 {
2726 	switch (arg) {
2727 	case KVM_CAP_USER_MEMORY:
2728 	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
2729 	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
2730 	case KVM_CAP_INTERNAL_ERROR_DATA:
2731 #ifdef CONFIG_HAVE_KVM_MSI
2732 	case KVM_CAP_SIGNAL_MSI:
2733 #endif
2734 #ifdef CONFIG_HAVE_KVM_IRQFD
2735 	case KVM_CAP_IRQFD:
2736 	case KVM_CAP_IRQFD_RESAMPLE:
2737 #endif
2738 	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
2739 	case KVM_CAP_CHECK_EXTENSION_VM:
2740 		return 1;
2741 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2742 	case KVM_CAP_IRQ_ROUTING:
2743 		return KVM_MAX_IRQ_ROUTES;
2744 #endif
2745 #if KVM_ADDRESS_SPACE_NUM > 1
2746 	case KVM_CAP_MULTI_ADDRESS_SPACE:
2747 		return KVM_ADDRESS_SPACE_NUM;
2748 #endif
2749 	default:
2750 		break;
2751 	}
2752 	return kvm_vm_ioctl_check_extension(kvm, arg);
2753 }
2754 
2755 static long kvm_vm_ioctl(struct file *filp,
2756 			   unsigned int ioctl, unsigned long arg)
2757 {
2758 	struct kvm *kvm = filp->private_data;
2759 	void __user *argp = (void __user *)arg;
2760 	int r;
2761 
2762 	if (kvm->mm != current->mm)
2763 		return -EIO;
2764 	switch (ioctl) {
2765 	case KVM_CREATE_VCPU:
2766 		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
2767 		break;
2768 	case KVM_SET_USER_MEMORY_REGION: {
2769 		struct kvm_userspace_memory_region kvm_userspace_mem;
2770 
2771 		r = -EFAULT;
2772 		if (copy_from_user(&kvm_userspace_mem, argp,
2773 						sizeof(kvm_userspace_mem)))
2774 			goto out;
2775 
2776 		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
2777 		break;
2778 	}
2779 	case KVM_GET_DIRTY_LOG: {
2780 		struct kvm_dirty_log log;
2781 
2782 		r = -EFAULT;
2783 		if (copy_from_user(&log, argp, sizeof(log)))
2784 			goto out;
2785 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2786 		break;
2787 	}
2788 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2789 	case KVM_REGISTER_COALESCED_MMIO: {
2790 		struct kvm_coalesced_mmio_zone zone;
2791 
2792 		r = -EFAULT;
2793 		if (copy_from_user(&zone, argp, sizeof(zone)))
2794 			goto out;
2795 		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
2796 		break;
2797 	}
2798 	case KVM_UNREGISTER_COALESCED_MMIO: {
2799 		struct kvm_coalesced_mmio_zone zone;
2800 
2801 		r = -EFAULT;
2802 		if (copy_from_user(&zone, argp, sizeof(zone)))
2803 			goto out;
2804 		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
2805 		break;
2806 	}
2807 #endif
2808 	case KVM_IRQFD: {
2809 		struct kvm_irqfd data;
2810 
2811 		r = -EFAULT;
2812 		if (copy_from_user(&data, argp, sizeof(data)))
2813 			goto out;
2814 		r = kvm_irqfd(kvm, &data);
2815 		break;
2816 	}
2817 	case KVM_IOEVENTFD: {
2818 		struct kvm_ioeventfd data;
2819 
2820 		r = -EFAULT;
2821 		if (copy_from_user(&data, argp, sizeof(data)))
2822 			goto out;
2823 		r = kvm_ioeventfd(kvm, &data);
2824 		break;
2825 	}
2826 #ifdef CONFIG_HAVE_KVM_MSI
2827 	case KVM_SIGNAL_MSI: {
2828 		struct kvm_msi msi;
2829 
2830 		r = -EFAULT;
2831 		if (copy_from_user(&msi, argp, sizeof(msi)))
2832 			goto out;
2833 		r = kvm_send_userspace_msi(kvm, &msi);
2834 		break;
2835 	}
2836 #endif
2837 #ifdef __KVM_HAVE_IRQ_LINE
2838 	case KVM_IRQ_LINE_STATUS:
2839 	case KVM_IRQ_LINE: {
2840 		struct kvm_irq_level irq_event;
2841 
2842 		r = -EFAULT;
2843 		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
2844 			goto out;
2845 
2846 		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
2847 					ioctl == KVM_IRQ_LINE_STATUS);
2848 		if (r)
2849 			goto out;
2850 
2851 		r = -EFAULT;
2852 		if (ioctl == KVM_IRQ_LINE_STATUS) {
2853 			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
2854 				goto out;
2855 		}
2856 
2857 		r = 0;
2858 		break;
2859 	}
2860 #endif
2861 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2862 	case KVM_SET_GSI_ROUTING: {
2863 		struct kvm_irq_routing routing;
2864 		struct kvm_irq_routing __user *urouting;
2865 		struct kvm_irq_routing_entry *entries;
2866 
2867 		r = -EFAULT;
2868 		if (copy_from_user(&routing, argp, sizeof(routing)))
2869 			goto out;
2870 		r = -EINVAL;
2871 		if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2872 			goto out;
2873 		if (routing.flags)
2874 			goto out;
2875 		r = -ENOMEM;
2876 		entries = vmalloc(routing.nr * sizeof(*entries));
2877 		if (!entries)
2878 			goto out;
2879 		r = -EFAULT;
2880 		urouting = argp;
2881 		if (copy_from_user(entries, urouting->entries,
2882 				   routing.nr * sizeof(*entries)))
2883 			goto out_free_irq_routing;
2884 		r = kvm_set_irq_routing(kvm, entries, routing.nr,
2885 					routing.flags);
2886 out_free_irq_routing:
2887 		vfree(entries);
2888 		break;
2889 	}
2890 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
2891 	case KVM_CREATE_DEVICE: {
2892 		struct kvm_create_device cd;
2893 
2894 		r = -EFAULT;
2895 		if (copy_from_user(&cd, argp, sizeof(cd)))
2896 			goto out;
2897 
2898 		r = kvm_ioctl_create_device(kvm, &cd);
2899 		if (r)
2900 			goto out;
2901 
2902 		r = -EFAULT;
2903 		if (copy_to_user(argp, &cd, sizeof(cd)))
2904 			goto out;
2905 
2906 		r = 0;
2907 		break;
2908 	}
2909 	case KVM_CHECK_EXTENSION:
2910 		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
2911 		break;
2912 	default:
2913 		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
2914 	}
2915 out:
2916 	return r;
2917 }
2918 
2919 #ifdef CONFIG_KVM_COMPAT
2920 struct compat_kvm_dirty_log {
2921 	__u32 slot;
2922 	__u32 padding1;
2923 	union {
2924 		compat_uptr_t dirty_bitmap; /* one bit per page */
2925 		__u64 padding2;
2926 	};
2927 };
2928 
2929 static long kvm_vm_compat_ioctl(struct file *filp,
2930 			   unsigned int ioctl, unsigned long arg)
2931 {
2932 	struct kvm *kvm = filp->private_data;
2933 	int r;
2934 
2935 	if (kvm->mm != current->mm)
2936 		return -EIO;
2937 	switch (ioctl) {
2938 	case KVM_GET_DIRTY_LOG: {
2939 		struct compat_kvm_dirty_log compat_log;
2940 		struct kvm_dirty_log log;
2941 
2942 		r = -EFAULT;
2943 		if (copy_from_user(&compat_log, (void __user *)arg,
2944 				   sizeof(compat_log)))
2945 			goto out;
2946 		log.slot	 = compat_log.slot;
2947 		log.padding1	 = compat_log.padding1;
2948 		log.padding2	 = compat_log.padding2;
2949 		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
2950 
2951 		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
2952 		break;
2953 	}
2954 	default:
2955 		r = kvm_vm_ioctl(filp, ioctl, arg);
2956 	}
2957 
2958 out:
2959 	return r;
2960 }
2961 #endif
2962 
2963 static struct file_operations kvm_vm_fops = {
2964 	.release        = kvm_vm_release,
2965 	.unlocked_ioctl = kvm_vm_ioctl,
2966 #ifdef CONFIG_KVM_COMPAT
2967 	.compat_ioctl   = kvm_vm_compat_ioctl,
2968 #endif
2969 	.llseek		= noop_llseek,
2970 };
2971 
2972 static int kvm_dev_ioctl_create_vm(unsigned long type)
2973 {
2974 	int r;
2975 	struct kvm *kvm;
2976 
2977 	kvm = kvm_create_vm(type);
2978 	if (IS_ERR(kvm))
2979 		return PTR_ERR(kvm);
2980 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2981 	r = kvm_coalesced_mmio_init(kvm);
2982 	if (r < 0) {
2983 		kvm_put_kvm(kvm);
2984 		return r;
2985 	}
2986 #endif
2987 	r = anon_inode_getfd("kvm-vm", &kvm_vm_fops, kvm, O_RDWR | O_CLOEXEC);
2988 	if (r < 0)
2989 		kvm_put_kvm(kvm);
2990 
2991 	return r;
2992 }
2993 
2994 static long kvm_dev_ioctl(struct file *filp,
2995 			  unsigned int ioctl, unsigned long arg)
2996 {
2997 	long r = -EINVAL;
2998 
2999 	switch (ioctl) {
3000 	case KVM_GET_API_VERSION:
3001 		if (arg)
3002 			goto out;
3003 		r = KVM_API_VERSION;
3004 		break;
3005 	case KVM_CREATE_VM:
3006 		r = kvm_dev_ioctl_create_vm(arg);
3007 		break;
3008 	case KVM_CHECK_EXTENSION:
3009 		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
3010 		break;
3011 	case KVM_GET_VCPU_MMAP_SIZE:
3012 		if (arg)
3013 			goto out;
3014 		r = PAGE_SIZE;     /* struct kvm_run */
3015 #ifdef CONFIG_X86
3016 		r += PAGE_SIZE;    /* pio data page */
3017 #endif
3018 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3019 		r += PAGE_SIZE;    /* coalesced mmio ring page */
3020 #endif
3021 		break;
3022 	case KVM_TRACE_ENABLE:
3023 	case KVM_TRACE_PAUSE:
3024 	case KVM_TRACE_DISABLE:
3025 		r = -EOPNOTSUPP;
3026 		break;
3027 	default:
3028 		return kvm_arch_dev_ioctl(filp, ioctl, arg);
3029 	}
3030 out:
3031 	return r;
3032 }
3033 
3034 static struct file_operations kvm_chardev_ops = {
3035 	.unlocked_ioctl = kvm_dev_ioctl,
3036 	.compat_ioctl   = kvm_dev_ioctl,
3037 	.llseek		= noop_llseek,
3038 };
3039 
3040 static struct miscdevice kvm_dev = {
3041 	KVM_MINOR,
3042 	"kvm",
3043 	&kvm_chardev_ops,
3044 };
3045 
3046 static void hardware_enable_nolock(void *junk)
3047 {
3048 	int cpu = raw_smp_processor_id();
3049 	int r;
3050 
3051 	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
3052 		return;
3053 
3054 	cpumask_set_cpu(cpu, cpus_hardware_enabled);
3055 
3056 	r = kvm_arch_hardware_enable();
3057 
3058 	if (r) {
3059 		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3060 		atomic_inc(&hardware_enable_failed);
3061 		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
3062 	}
3063 }
3064 
3065 static void hardware_enable(void)
3066 {
3067 	raw_spin_lock(&kvm_count_lock);
3068 	if (kvm_usage_count)
3069 		hardware_enable_nolock(NULL);
3070 	raw_spin_unlock(&kvm_count_lock);
3071 }
3072 
3073 static void hardware_disable_nolock(void *junk)
3074 {
3075 	int cpu = raw_smp_processor_id();
3076 
3077 	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
3078 		return;
3079 	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
3080 	kvm_arch_hardware_disable();
3081 }
3082 
3083 static void hardware_disable(void)
3084 {
3085 	raw_spin_lock(&kvm_count_lock);
3086 	if (kvm_usage_count)
3087 		hardware_disable_nolock(NULL);
3088 	raw_spin_unlock(&kvm_count_lock);
3089 }
3090 
3091 static void hardware_disable_all_nolock(void)
3092 {
3093 	BUG_ON(!kvm_usage_count);
3094 
3095 	kvm_usage_count--;
3096 	if (!kvm_usage_count)
3097 		on_each_cpu(hardware_disable_nolock, NULL, 1);
3098 }
3099 
3100 static void hardware_disable_all(void)
3101 {
3102 	raw_spin_lock(&kvm_count_lock);
3103 	hardware_disable_all_nolock();
3104 	raw_spin_unlock(&kvm_count_lock);
3105 }
3106 
3107 static int hardware_enable_all(void)
3108 {
3109 	int r = 0;
3110 
3111 	raw_spin_lock(&kvm_count_lock);
3112 
3113 	kvm_usage_count++;
3114 	if (kvm_usage_count == 1) {
3115 		atomic_set(&hardware_enable_failed, 0);
3116 		on_each_cpu(hardware_enable_nolock, NULL, 1);
3117 
3118 		if (atomic_read(&hardware_enable_failed)) {
3119 			hardware_disable_all_nolock();
3120 			r = -EBUSY;
3121 		}
3122 	}
3123 
3124 	raw_spin_unlock(&kvm_count_lock);
3125 
3126 	return r;
3127 }
3128 
3129 static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
3130 			   void *v)
3131 {
3132 	val &= ~CPU_TASKS_FROZEN;
3133 	switch (val) {
3134 	case CPU_DYING:
3135 		hardware_disable();
3136 		break;
3137 	case CPU_STARTING:
3138 		hardware_enable();
3139 		break;
3140 	}
3141 	return NOTIFY_OK;
3142 }
3143 
3144 static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
3145 		      void *v)
3146 {
3147 	/*
3148 	 * Some (well, at least mine) BIOSes hang on reboot if
3149 	 * in vmx root mode.
3150 	 *
3151 	 * And Intel TXT required VMX off for all cpu when system shutdown.
3152 	 */
3153 	pr_info("kvm: exiting hardware virtualization\n");
3154 	kvm_rebooting = true;
3155 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3156 	return NOTIFY_OK;
3157 }
3158 
3159 static struct notifier_block kvm_reboot_notifier = {
3160 	.notifier_call = kvm_reboot,
3161 	.priority = 0,
3162 };
3163 
3164 static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
3165 {
3166 	int i;
3167 
3168 	for (i = 0; i < bus->dev_count; i++) {
3169 		struct kvm_io_device *pos = bus->range[i].dev;
3170 
3171 		kvm_iodevice_destructor(pos);
3172 	}
3173 	kfree(bus);
3174 }
3175 
3176 static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
3177 				 const struct kvm_io_range *r2)
3178 {
3179 	gpa_t addr1 = r1->addr;
3180 	gpa_t addr2 = r2->addr;
3181 
3182 	if (addr1 < addr2)
3183 		return -1;
3184 
3185 	/* If r2->len == 0, match the exact address.  If r2->len != 0,
3186 	 * accept any overlapping write.  Any order is acceptable for
3187 	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3188 	 * we process all of them.
3189 	 */
3190 	if (r2->len) {
3191 		addr1 += r1->len;
3192 		addr2 += r2->len;
3193 	}
3194 
3195 	if (addr1 > addr2)
3196 		return 1;
3197 
3198 	return 0;
3199 }
3200 
3201 static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
3202 {
3203 	return kvm_io_bus_cmp(p1, p2);
3204 }
3205 
3206 static int kvm_io_bus_insert_dev(struct kvm_io_bus *bus, struct kvm_io_device *dev,
3207 			  gpa_t addr, int len)
3208 {
3209 	bus->range[bus->dev_count++] = (struct kvm_io_range) {
3210 		.addr = addr,
3211 		.len = len,
3212 		.dev = dev,
3213 	};
3214 
3215 	sort(bus->range, bus->dev_count, sizeof(struct kvm_io_range),
3216 		kvm_io_bus_sort_cmp, NULL);
3217 
3218 	return 0;
3219 }
3220 
3221 static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
3222 			     gpa_t addr, int len)
3223 {
3224 	struct kvm_io_range *range, key;
3225 	int off;
3226 
3227 	key = (struct kvm_io_range) {
3228 		.addr = addr,
3229 		.len = len,
3230 	};
3231 
3232 	range = bsearch(&key, bus->range, bus->dev_count,
3233 			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
3234 	if (range == NULL)
3235 		return -ENOENT;
3236 
3237 	off = range - bus->range;
3238 
3239 	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
3240 		off--;
3241 
3242 	return off;
3243 }
3244 
3245 static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3246 			      struct kvm_io_range *range, const void *val)
3247 {
3248 	int idx;
3249 
3250 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3251 	if (idx < 0)
3252 		return -EOPNOTSUPP;
3253 
3254 	while (idx < bus->dev_count &&
3255 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3256 		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
3257 					range->len, val))
3258 			return idx;
3259 		idx++;
3260 	}
3261 
3262 	return -EOPNOTSUPP;
3263 }
3264 
3265 /* kvm_io_bus_write - called under kvm->slots_lock */
3266 int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3267 		     int len, const void *val)
3268 {
3269 	struct kvm_io_bus *bus;
3270 	struct kvm_io_range range;
3271 	int r;
3272 
3273 	range = (struct kvm_io_range) {
3274 		.addr = addr,
3275 		.len = len,
3276 	};
3277 
3278 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3279 	r = __kvm_io_bus_write(vcpu, bus, &range, val);
3280 	return r < 0 ? r : 0;
3281 }
3282 
3283 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3284 int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
3285 			    gpa_t addr, int len, const void *val, long cookie)
3286 {
3287 	struct kvm_io_bus *bus;
3288 	struct kvm_io_range range;
3289 
3290 	range = (struct kvm_io_range) {
3291 		.addr = addr,
3292 		.len = len,
3293 	};
3294 
3295 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3296 
3297 	/* First try the device referenced by cookie. */
3298 	if ((cookie >= 0) && (cookie < bus->dev_count) &&
3299 	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
3300 		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
3301 					val))
3302 			return cookie;
3303 
3304 	/*
3305 	 * cookie contained garbage; fall back to search and return the
3306 	 * correct cookie value.
3307 	 */
3308 	return __kvm_io_bus_write(vcpu, bus, &range, val);
3309 }
3310 
3311 static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
3312 			     struct kvm_io_range *range, void *val)
3313 {
3314 	int idx;
3315 
3316 	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
3317 	if (idx < 0)
3318 		return -EOPNOTSUPP;
3319 
3320 	while (idx < bus->dev_count &&
3321 		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
3322 		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
3323 				       range->len, val))
3324 			return idx;
3325 		idx++;
3326 	}
3327 
3328 	return -EOPNOTSUPP;
3329 }
3330 EXPORT_SYMBOL_GPL(kvm_io_bus_write);
3331 
3332 /* kvm_io_bus_read - called under kvm->slots_lock */
3333 int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
3334 		    int len, void *val)
3335 {
3336 	struct kvm_io_bus *bus;
3337 	struct kvm_io_range range;
3338 	int r;
3339 
3340 	range = (struct kvm_io_range) {
3341 		.addr = addr,
3342 		.len = len,
3343 	};
3344 
3345 	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
3346 	r = __kvm_io_bus_read(vcpu, bus, &range, val);
3347 	return r < 0 ? r : 0;
3348 }
3349 
3350 
3351 /* Caller must hold slots_lock. */
3352 int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
3353 			    int len, struct kvm_io_device *dev)
3354 {
3355 	struct kvm_io_bus *new_bus, *bus;
3356 
3357 	bus = kvm->buses[bus_idx];
3358 	/* exclude ioeventfd which is limited by maximum fd */
3359 	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
3360 		return -ENOSPC;
3361 
3362 	new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count + 1) *
3363 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3364 	if (!new_bus)
3365 		return -ENOMEM;
3366 	memcpy(new_bus, bus, sizeof(*bus) + (bus->dev_count *
3367 	       sizeof(struct kvm_io_range)));
3368 	kvm_io_bus_insert_dev(new_bus, dev, addr, len);
3369 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3370 	synchronize_srcu_expedited(&kvm->srcu);
3371 	kfree(bus);
3372 
3373 	return 0;
3374 }
3375 
3376 /* Caller must hold slots_lock. */
3377 int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
3378 			      struct kvm_io_device *dev)
3379 {
3380 	int i, r;
3381 	struct kvm_io_bus *new_bus, *bus;
3382 
3383 	bus = kvm->buses[bus_idx];
3384 	r = -ENOENT;
3385 	for (i = 0; i < bus->dev_count; i++)
3386 		if (bus->range[i].dev == dev) {
3387 			r = 0;
3388 			break;
3389 		}
3390 
3391 	if (r)
3392 		return r;
3393 
3394 	new_bus = kmalloc(sizeof(*bus) + ((bus->dev_count - 1) *
3395 			  sizeof(struct kvm_io_range)), GFP_KERNEL);
3396 	if (!new_bus)
3397 		return -ENOMEM;
3398 
3399 	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
3400 	new_bus->dev_count--;
3401 	memcpy(new_bus->range + i, bus->range + i + 1,
3402 	       (new_bus->dev_count - i) * sizeof(struct kvm_io_range));
3403 
3404 	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
3405 	synchronize_srcu_expedited(&kvm->srcu);
3406 	kfree(bus);
3407 	return r;
3408 }
3409 
3410 static struct notifier_block kvm_cpu_notifier = {
3411 	.notifier_call = kvm_cpu_hotplug,
3412 };
3413 
3414 static int vm_stat_get(void *_offset, u64 *val)
3415 {
3416 	unsigned offset = (long)_offset;
3417 	struct kvm *kvm;
3418 
3419 	*val = 0;
3420 	spin_lock(&kvm_lock);
3421 	list_for_each_entry(kvm, &vm_list, vm_list)
3422 		*val += *(u32 *)((void *)kvm + offset);
3423 	spin_unlock(&kvm_lock);
3424 	return 0;
3425 }
3426 
3427 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, NULL, "%llu\n");
3428 
3429 static int vcpu_stat_get(void *_offset, u64 *val)
3430 {
3431 	unsigned offset = (long)_offset;
3432 	struct kvm *kvm;
3433 	struct kvm_vcpu *vcpu;
3434 	int i;
3435 
3436 	*val = 0;
3437 	spin_lock(&kvm_lock);
3438 	list_for_each_entry(kvm, &vm_list, vm_list)
3439 		kvm_for_each_vcpu(i, vcpu, kvm)
3440 			*val += *(u32 *)((void *)vcpu + offset);
3441 
3442 	spin_unlock(&kvm_lock);
3443 	return 0;
3444 }
3445 
3446 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, NULL, "%llu\n");
3447 
3448 static const struct file_operations *stat_fops[] = {
3449 	[KVM_STAT_VCPU] = &vcpu_stat_fops,
3450 	[KVM_STAT_VM]   = &vm_stat_fops,
3451 };
3452 
3453 static int kvm_init_debug(void)
3454 {
3455 	int r = -EEXIST;
3456 	struct kvm_stats_debugfs_item *p;
3457 
3458 	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
3459 	if (kvm_debugfs_dir == NULL)
3460 		goto out;
3461 
3462 	for (p = debugfs_entries; p->name; ++p) {
3463 		if (!debugfs_create_file(p->name, 0444, kvm_debugfs_dir,
3464 					 (void *)(long)p->offset,
3465 					 stat_fops[p->kind]))
3466 			goto out_dir;
3467 	}
3468 
3469 	return 0;
3470 
3471 out_dir:
3472 	debugfs_remove_recursive(kvm_debugfs_dir);
3473 out:
3474 	return r;
3475 }
3476 
3477 static int kvm_suspend(void)
3478 {
3479 	if (kvm_usage_count)
3480 		hardware_disable_nolock(NULL);
3481 	return 0;
3482 }
3483 
3484 static void kvm_resume(void)
3485 {
3486 	if (kvm_usage_count) {
3487 		WARN_ON(raw_spin_is_locked(&kvm_count_lock));
3488 		hardware_enable_nolock(NULL);
3489 	}
3490 }
3491 
3492 static struct syscore_ops kvm_syscore_ops = {
3493 	.suspend = kvm_suspend,
3494 	.resume = kvm_resume,
3495 };
3496 
3497 static inline
3498 struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
3499 {
3500 	return container_of(pn, struct kvm_vcpu, preempt_notifier);
3501 }
3502 
3503 static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
3504 {
3505 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3506 
3507 	if (vcpu->preempted)
3508 		vcpu->preempted = false;
3509 
3510 	kvm_arch_sched_in(vcpu, cpu);
3511 
3512 	kvm_arch_vcpu_load(vcpu, cpu);
3513 }
3514 
3515 static void kvm_sched_out(struct preempt_notifier *pn,
3516 			  struct task_struct *next)
3517 {
3518 	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
3519 
3520 	if (current->state == TASK_RUNNING)
3521 		vcpu->preempted = true;
3522 	kvm_arch_vcpu_put(vcpu);
3523 }
3524 
3525 int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
3526 		  struct module *module)
3527 {
3528 	int r;
3529 	int cpu;
3530 
3531 	r = kvm_arch_init(opaque);
3532 	if (r)
3533 		goto out_fail;
3534 
3535 	/*
3536 	 * kvm_arch_init makes sure there's at most one caller
3537 	 * for architectures that support multiple implementations,
3538 	 * like intel and amd on x86.
3539 	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3540 	 * conflicts in case kvm is already setup for another implementation.
3541 	 */
3542 	r = kvm_irqfd_init();
3543 	if (r)
3544 		goto out_irqfd;
3545 
3546 	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
3547 		r = -ENOMEM;
3548 		goto out_free_0;
3549 	}
3550 
3551 	r = kvm_arch_hardware_setup();
3552 	if (r < 0)
3553 		goto out_free_0a;
3554 
3555 	for_each_online_cpu(cpu) {
3556 		smp_call_function_single(cpu,
3557 				kvm_arch_check_processor_compat,
3558 				&r, 1);
3559 		if (r < 0)
3560 			goto out_free_1;
3561 	}
3562 
3563 	r = register_cpu_notifier(&kvm_cpu_notifier);
3564 	if (r)
3565 		goto out_free_2;
3566 	register_reboot_notifier(&kvm_reboot_notifier);
3567 
3568 	/* A kmem cache lets us meet the alignment requirements of fx_save. */
3569 	if (!vcpu_align)
3570 		vcpu_align = __alignof__(struct kvm_vcpu);
3571 	kvm_vcpu_cache = kmem_cache_create("kvm_vcpu", vcpu_size, vcpu_align,
3572 					   0, NULL);
3573 	if (!kvm_vcpu_cache) {
3574 		r = -ENOMEM;
3575 		goto out_free_3;
3576 	}
3577 
3578 	r = kvm_async_pf_init();
3579 	if (r)
3580 		goto out_free;
3581 
3582 	kvm_chardev_ops.owner = module;
3583 	kvm_vm_fops.owner = module;
3584 	kvm_vcpu_fops.owner = module;
3585 
3586 	r = misc_register(&kvm_dev);
3587 	if (r) {
3588 		pr_err("kvm: misc device register failed\n");
3589 		goto out_unreg;
3590 	}
3591 
3592 	register_syscore_ops(&kvm_syscore_ops);
3593 
3594 	kvm_preempt_ops.sched_in = kvm_sched_in;
3595 	kvm_preempt_ops.sched_out = kvm_sched_out;
3596 
3597 	r = kvm_init_debug();
3598 	if (r) {
3599 		pr_err("kvm: create debugfs files failed\n");
3600 		goto out_undebugfs;
3601 	}
3602 
3603 	r = kvm_vfio_ops_init();
3604 	WARN_ON(r);
3605 
3606 	return 0;
3607 
3608 out_undebugfs:
3609 	unregister_syscore_ops(&kvm_syscore_ops);
3610 	misc_deregister(&kvm_dev);
3611 out_unreg:
3612 	kvm_async_pf_deinit();
3613 out_free:
3614 	kmem_cache_destroy(kvm_vcpu_cache);
3615 out_free_3:
3616 	unregister_reboot_notifier(&kvm_reboot_notifier);
3617 	unregister_cpu_notifier(&kvm_cpu_notifier);
3618 out_free_2:
3619 out_free_1:
3620 	kvm_arch_hardware_unsetup();
3621 out_free_0a:
3622 	free_cpumask_var(cpus_hardware_enabled);
3623 out_free_0:
3624 	kvm_irqfd_exit();
3625 out_irqfd:
3626 	kvm_arch_exit();
3627 out_fail:
3628 	return r;
3629 }
3630 EXPORT_SYMBOL_GPL(kvm_init);
3631 
3632 void kvm_exit(void)
3633 {
3634 	debugfs_remove_recursive(kvm_debugfs_dir);
3635 	misc_deregister(&kvm_dev);
3636 	kmem_cache_destroy(kvm_vcpu_cache);
3637 	kvm_async_pf_deinit();
3638 	unregister_syscore_ops(&kvm_syscore_ops);
3639 	unregister_reboot_notifier(&kvm_reboot_notifier);
3640 	unregister_cpu_notifier(&kvm_cpu_notifier);
3641 	on_each_cpu(hardware_disable_nolock, NULL, 1);
3642 	kvm_arch_hardware_unsetup();
3643 	kvm_arch_exit();
3644 	kvm_irqfd_exit();
3645 	free_cpumask_var(cpus_hardware_enabled);
3646 	kvm_vfio_ops_exit();
3647 }
3648 EXPORT_SYMBOL_GPL(kvm_exit);
3649