1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Kernel-based Virtual Machine driver for Linux
   4 *
   5 * This module enables machines with Intel VT-x extensions to run virtual
   6 * machines without emulation or binary translation.
   7 *
   8 * Copyright (C) 2006 Qumranet, Inc.
   9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10 *
  11 * Authors:
  12 *   Avi Kivity   <avi@qumranet.com>
  13 *   Yaniv Kamay  <yaniv@qumranet.com>
  14 */
  15
  16#include <kvm/iodev.h>
  17
  18#include <linux/kvm_host.h>
  19#include <linux/kvm.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/percpu.h>
  23#include <linux/mm.h>
  24#include <linux/miscdevice.h>
  25#include <linux/vmalloc.h>
  26#include <linux/reboot.h>
  27#include <linux/debugfs.h>
  28#include <linux/highmem.h>
  29#include <linux/file.h>
  30#include <linux/syscore_ops.h>
  31#include <linux/cpu.h>
  32#include <linux/sched/signal.h>
  33#include <linux/sched/mm.h>
  34#include <linux/sched/stat.h>
  35#include <linux/cpumask.h>
  36#include <linux/smp.h>
  37#include <linux/anon_inodes.h>
  38#include <linux/profile.h>
  39#include <linux/kvm_para.h>
  40#include <linux/pagemap.h>
  41#include <linux/mman.h>
  42#include <linux/swap.h>
  43#include <linux/bitops.h>
  44#include <linux/spinlock.h>
  45#include <linux/compat.h>
  46#include <linux/srcu.h>
  47#include <linux/hugetlb.h>
  48#include <linux/slab.h>
  49#include <linux/sort.h>
  50#include <linux/bsearch.h>
  51#include <linux/io.h>
  52#include <linux/lockdep.h>
  53#include <linux/kthread.h>
  54#include <linux/suspend.h>
  55
  56#include <asm/processor.h>
  57#include <asm/ioctl.h>
  58#include <linux/uaccess.h>
  59
  60#include "coalesced_mmio.h"
  61#include "async_pf.h"
  62#include "kvm_mm.h"
  63#include "vfio.h"
  64
  65#include <trace/events/ipi.h>
  66
  67#define CREATE_TRACE_POINTS
  68#include <trace/events/kvm.h>
  69
  70#include <linux/kvm_dirty_ring.h>
  71
  72
  73/* Worst case buffer size needed for holding an integer. */
  74#define ITOA_MAX_LEN 12
  75
  76MODULE_AUTHOR("Qumranet");
  77MODULE_LICENSE("GPL");
  78
  79/* Architectures should define their poll value according to the halt latency */
  80unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
  81module_param(halt_poll_ns, uint, 0644);
  82EXPORT_SYMBOL_GPL(halt_poll_ns);
  83
  84/* Default doubles per-vcpu halt_poll_ns. */
  85unsigned int halt_poll_ns_grow = 2;
  86module_param(halt_poll_ns_grow, uint, 0644);
  87EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
  88
  89/* The start value to grow halt_poll_ns from */
  90unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
  91module_param(halt_poll_ns_grow_start, uint, 0644);
  92EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
  93
  94/* Default resets per-vcpu halt_poll_ns . */
  95unsigned int halt_poll_ns_shrink;
  96module_param(halt_poll_ns_shrink, uint, 0644);
  97EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
  98
  99/*
 100 * Ordering of locks:
 101 *
 102 *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
 103 */
 104
 105DEFINE_MUTEX(kvm_lock);
 106LIST_HEAD(vm_list);
 107
 108static struct kmem_cache *kvm_vcpu_cache;
 109
 110static __read_mostly struct preempt_ops kvm_preempt_ops;
 111static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
 112
 113struct dentry *kvm_debugfs_dir;
 114EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
 115
 116static const struct file_operations stat_fops_per_vm;
 117
 118static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 119			   unsigned long arg);
 120#ifdef CONFIG_KVM_COMPAT
 121static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
 122				  unsigned long arg);
 123#define KVM_COMPAT(c)	.compat_ioctl	= (c)
 124#else
 125/*
 126 * For architectures that don't implement a compat infrastructure,
 127 * adopt a double line of defense:
 128 * - Prevent a compat task from opening /dev/kvm
 129 * - If the open has been done by a 64bit task, and the KVM fd
 130 *   passed to a compat task, let the ioctls fail.
 131 */
 132static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
 133				unsigned long arg) { return -EINVAL; }
 134
 135static int kvm_no_compat_open(struct inode *inode, struct file *file)
 136{
 137	return is_compat_task() ? -ENODEV : 0;
 138}
 139#define KVM_COMPAT(c)	.compat_ioctl	= kvm_no_compat_ioctl,	\
 140			.open		= kvm_no_compat_open
 141#endif
 142static int hardware_enable_all(void);
 143static void hardware_disable_all(void);
 144
 145static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 146
 147#define KVM_EVENT_CREATE_VM 0
 148#define KVM_EVENT_DESTROY_VM 1
 149static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
 150static unsigned long long kvm_createvm_count;
 151static unsigned long long kvm_active_vms;
 152
 153static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
 154
 155__weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
 156{
 157}
 158
 159bool kvm_is_zone_device_page(struct page *page)
 160{
 161	/*
 162	 * The metadata used by is_zone_device_page() to determine whether or
 163	 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
 164	 * the device has been pinned, e.g. by get_user_pages().  WARN if the
 165	 * page_count() is zero to help detect bad usage of this helper.
 166	 */
 167	if (WARN_ON_ONCE(!page_count(page)))
 168		return false;
 169
 170	return is_zone_device_page(page);
 171}
 172
 173/*
 174 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
 175 * page, NULL otherwise.  Note, the list of refcounted PG_reserved page types
 176 * is likely incomplete, it has been compiled purely through people wanting to
 177 * back guest with a certain type of memory and encountering issues.
 178 */
 179struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
 180{
 181	struct page *page;
 182
 183	if (!pfn_valid(pfn))
 184		return NULL;
 185
 186	page = pfn_to_page(pfn);
 187	if (!PageReserved(page))
 188		return page;
 189
 190	/* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
 191	if (is_zero_pfn(pfn))
 192		return page;
 193
 194	/*
 195	 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
 196	 * perspective they are "normal" pages, albeit with slightly different
 197	 * usage rules.
 198	 */
 199	if (kvm_is_zone_device_page(page))
 200		return page;
 201
 202	return NULL;
 203}
 204
 205/*
 206 * Switches to specified vcpu, until a matching vcpu_put()
 207 */
 208void vcpu_load(struct kvm_vcpu *vcpu)
 209{
 210	int cpu = get_cpu();
 211
 212	__this_cpu_write(kvm_running_vcpu, vcpu);
 213	preempt_notifier_register(&vcpu->preempt_notifier);
 214	kvm_arch_vcpu_load(vcpu, cpu);
 215	put_cpu();
 216}
 217EXPORT_SYMBOL_GPL(vcpu_load);
 218
 219void vcpu_put(struct kvm_vcpu *vcpu)
 220{
 221	preempt_disable();
 222	kvm_arch_vcpu_put(vcpu);
 223	preempt_notifier_unregister(&vcpu->preempt_notifier);
 224	__this_cpu_write(kvm_running_vcpu, NULL);
 225	preempt_enable();
 226}
 227EXPORT_SYMBOL_GPL(vcpu_put);
 228
 229/* TODO: merge with kvm_arch_vcpu_should_kick */
 230static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
 231{
 232	int mode = kvm_vcpu_exiting_guest_mode(vcpu);
 233
 234	/*
 235	 * We need to wait for the VCPU to reenable interrupts and get out of
 236	 * READING_SHADOW_PAGE_TABLES mode.
 237	 */
 238	if (req & KVM_REQUEST_WAIT)
 239		return mode != OUTSIDE_GUEST_MODE;
 240
 241	/*
 242	 * Need to kick a running VCPU, but otherwise there is nothing to do.
 243	 */
 244	return mode == IN_GUEST_MODE;
 245}
 246
 247static void ack_kick(void *_completed)
 248{
 249}
 250
 251static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
 252{
 253	if (cpumask_empty(cpus))
 254		return false;
 255
 256	smp_call_function_many(cpus, ack_kick, NULL, wait);
 257	return true;
 258}
 259
 260static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
 261				  struct cpumask *tmp, int current_cpu)
 262{
 263	int cpu;
 264
 265	if (likely(!(req & KVM_REQUEST_NO_ACTION)))
 266		__kvm_make_request(req, vcpu);
 267
 268	if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
 269		return;
 270
 271	/*
 272	 * Note, the vCPU could get migrated to a different pCPU at any point
 273	 * after kvm_request_needs_ipi(), which could result in sending an IPI
 274	 * to the previous pCPU.  But, that's OK because the purpose of the IPI
 275	 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
 276	 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
 277	 * after this point is also OK, as the requirement is only that KVM wait
 278	 * for vCPUs that were reading SPTEs _before_ any changes were
 279	 * finalized. See kvm_vcpu_kick() for more details on handling requests.
 280	 */
 281	if (kvm_request_needs_ipi(vcpu, req)) {
 282		cpu = READ_ONCE(vcpu->cpu);
 283		if (cpu != -1 && cpu != current_cpu)
 284			__cpumask_set_cpu(cpu, tmp);
 285	}
 286}
 287
 288bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
 289				 unsigned long *vcpu_bitmap)
 290{
 291	struct kvm_vcpu *vcpu;
 292	struct cpumask *cpus;
 293	int i, me;
 294	bool called;
 295
 296	me = get_cpu();
 297
 298	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
 299	cpumask_clear(cpus);
 300
 301	for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
 302		vcpu = kvm_get_vcpu(kvm, i);
 303		if (!vcpu)
 304			continue;
 305		kvm_make_vcpu_request(vcpu, req, cpus, me);
 306	}
 307
 308	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
 309	put_cpu();
 310
 311	return called;
 312}
 313
 314bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
 315				      struct kvm_vcpu *except)
 316{
 317	struct kvm_vcpu *vcpu;
 318	struct cpumask *cpus;
 319	unsigned long i;
 320	bool called;
 321	int me;
 322
 323	me = get_cpu();
 324
 325	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
 326	cpumask_clear(cpus);
 327
 328	kvm_for_each_vcpu(i, vcpu, kvm) {
 329		if (vcpu == except)
 330			continue;
 331		kvm_make_vcpu_request(vcpu, req, cpus, me);
 332	}
 333
 334	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
 335	put_cpu();
 336
 337	return called;
 338}
 339
 340bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 341{
 342	return kvm_make_all_cpus_request_except(kvm, req, NULL);
 343}
 344EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
 345
 346void kvm_flush_remote_tlbs(struct kvm *kvm)
 347{
 348	++kvm->stat.generic.remote_tlb_flush_requests;
 349
 350	/*
 351	 * We want to publish modifications to the page tables before reading
 352	 * mode. Pairs with a memory barrier in arch-specific code.
 353	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
 354	 * and smp_mb in walk_shadow_page_lockless_begin/end.
 355	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
 356	 *
 357	 * There is already an smp_mb__after_atomic() before
 358	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
 359	 * barrier here.
 360	 */
 361	if (!kvm_arch_flush_remote_tlbs(kvm)
 362	    || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 363		++kvm->stat.generic.remote_tlb_flush;
 364}
 365EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 366
 367void kvm_flush_remote_tlbs_range(struct kvm *kvm, gfn_t gfn, u64 nr_pages)
 368{
 369	if (!kvm_arch_flush_remote_tlbs_range(kvm, gfn, nr_pages))
 370		return;
 371
 372	/*
 373	 * Fall back to a flushing entire TLBs if the architecture range-based
 374	 * TLB invalidation is unsupported or can't be performed for whatever
 375	 * reason.
 376	 */
 377	kvm_flush_remote_tlbs(kvm);
 378}
 379
 380void kvm_flush_remote_tlbs_memslot(struct kvm *kvm,
 381				   const struct kvm_memory_slot *memslot)
 382{
 383	/*
 384	 * All current use cases for flushing the TLBs for a specific memslot
 385	 * are related to dirty logging, and many do the TLB flush out of
 386	 * mmu_lock. The interaction between the various operations on memslot
 387	 * must be serialized by slots_locks to ensure the TLB flush from one
 388	 * operation is observed by any other operation on the same memslot.
 389	 */
 390	lockdep_assert_held(&kvm->slots_lock);
 391	kvm_flush_remote_tlbs_range(kvm, memslot->base_gfn, memslot->npages);
 392}
 393
 394static void kvm_flush_shadow_all(struct kvm *kvm)
 395{
 396	kvm_arch_flush_shadow_all(kvm);
 397	kvm_arch_guest_memory_reclaimed(kvm);
 398}
 399
 400#ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
 401static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
 402					       gfp_t gfp_flags)
 403{
 404	gfp_flags |= mc->gfp_zero;
 405
 406	if (mc->kmem_cache)
 407		return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
 408	else
 409		return (void *)__get_free_page(gfp_flags);
 410}
 411
 412int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
 413{
 414	gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
 415	void *obj;
 416
 417	if (mc->nobjs >= min)
 418		return 0;
 419
 420	if (unlikely(!mc->objects)) {
 421		if (WARN_ON_ONCE(!capacity))
 422			return -EIO;
 423
 424		mc->objects = kvmalloc_array(capacity, sizeof(void *), gfp);
 425		if (!mc->objects)
 426			return -ENOMEM;
 427
 428		mc->capacity = capacity;
 429	}
 430
 431	/* It is illegal to request a different capacity across topups. */
 432	if (WARN_ON_ONCE(mc->capacity != capacity))
 433		return -EIO;
 434
 435	while (mc->nobjs < mc->capacity) {
 436		obj = mmu_memory_cache_alloc_obj(mc, gfp);
 437		if (!obj)
 438			return mc->nobjs >= min ? 0 : -ENOMEM;
 439		mc->objects[mc->nobjs++] = obj;
 440	}
 441	return 0;
 442}
 443
 444int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
 445{
 446	return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
 447}
 448
 449int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
 450{
 451	return mc->nobjs;
 452}
 453
 454void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
 455{
 456	while (mc->nobjs) {
 457		if (mc->kmem_cache)
 458			kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
 459		else
 460			free_page((unsigned long)mc->objects[--mc->nobjs]);
 461	}
 462
 463	kvfree(mc->objects);
 464
 465	mc->objects = NULL;
 466	mc->capacity = 0;
 467}
 468
 469void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
 470{
 471	void *p;
 472
 473	if (WARN_ON(!mc->nobjs))
 474		p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
 475	else
 476		p = mc->objects[--mc->nobjs];
 477	BUG_ON(!p);
 478	return p;
 479}
 480#endif
 481
 482static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 483{
 484	mutex_init(&vcpu->mutex);
 485	vcpu->cpu = -1;
 486	vcpu->kvm = kvm;
 487	vcpu->vcpu_id = id;
 488	vcpu->pid = NULL;
 489#ifndef __KVM_HAVE_ARCH_WQP
 490	rcuwait_init(&vcpu->wait);
 491#endif
 492	kvm_async_pf_vcpu_init(vcpu);
 493
 494	kvm_vcpu_set_in_spin_loop(vcpu, false);
 495	kvm_vcpu_set_dy_eligible(vcpu, false);
 496	vcpu->preempted = false;
 497	vcpu->ready = false;
 498	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 499	vcpu->last_used_slot = NULL;
 500
 501	/* Fill the stats id string for the vcpu */
 502	snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
 503		 task_pid_nr(current), id);
 504}
 505
 506static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
 507{
 508	kvm_arch_vcpu_destroy(vcpu);
 509	kvm_dirty_ring_free(&vcpu->dirty_ring);
 510
 511	/*
 512	 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
 513	 * the vcpu->pid pointer, and at destruction time all file descriptors
 514	 * are already gone.
 515	 */
 516	put_pid(rcu_dereference_protected(vcpu->pid, 1));
 517
 518	free_page((unsigned long)vcpu->run);
 519	kmem_cache_free(kvm_vcpu_cache, vcpu);
 520}
 521
 522void kvm_destroy_vcpus(struct kvm *kvm)
 523{
 524	unsigned long i;
 525	struct kvm_vcpu *vcpu;
 526
 527	kvm_for_each_vcpu(i, vcpu, kvm) {
 528		kvm_vcpu_destroy(vcpu);
 529		xa_erase(&kvm->vcpu_array, i);
 530	}
 531
 532	atomic_set(&kvm->online_vcpus, 0);
 533}
 534EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
 535
 536#ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
 537static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 538{
 539	return container_of(mn, struct kvm, mmu_notifier);
 540}
 541
 542typedef bool (*gfn_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
 543
 544typedef void (*on_lock_fn_t)(struct kvm *kvm);
 545
 546struct kvm_mmu_notifier_range {
 547	/*
 548	 * 64-bit addresses, as KVM notifiers can operate on host virtual
 549	 * addresses (unsigned long) and guest physical addresses (64-bit).
 550	 */
 551	u64 start;
 552	u64 end;
 553	union kvm_mmu_notifier_arg arg;
 554	gfn_handler_t handler;
 555	on_lock_fn_t on_lock;
 556	bool flush_on_ret;
 557	bool may_block;
 558};
 559
 560/*
 561 * The inner-most helper returns a tuple containing the return value from the
 562 * arch- and action-specific handler, plus a flag indicating whether or not at
 563 * least one memslot was found, i.e. if the handler found guest memory.
 564 *
 565 * Note, most notifiers are averse to booleans, so even though KVM tracks the
 566 * return from arch code as a bool, outer helpers will cast it to an int. :-(
 567 */
 568typedef struct kvm_mmu_notifier_return {
 569	bool ret;
 570	bool found_memslot;
 571} kvm_mn_ret_t;
 572
 573/*
 574 * Use a dedicated stub instead of NULL to indicate that there is no callback
 575 * function/handler.  The compiler technically can't guarantee that a real
 576 * function will have a non-zero address, and so it will generate code to
 577 * check for !NULL, whereas comparing against a stub will be elided at compile
 578 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
 579 */
 580static void kvm_null_fn(void)
 581{
 582
 583}
 584#define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
 585
 586static const union kvm_mmu_notifier_arg KVM_MMU_NOTIFIER_NO_ARG;
 587
 588/* Iterate over each memslot intersecting [start, last] (inclusive) range */
 589#define kvm_for_each_memslot_in_hva_range(node, slots, start, last)	     \
 590	for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
 591	     node;							     \
 592	     node = interval_tree_iter_next(node, start, last))	     \
 593
 594static __always_inline kvm_mn_ret_t __kvm_handle_hva_range(struct kvm *kvm,
 595							   const struct kvm_mmu_notifier_range *range)
 596{
 597	struct kvm_mmu_notifier_return r = {
 598		.ret = false,
 599		.found_memslot = false,
 600	};
 601	struct kvm_gfn_range gfn_range;
 602	struct kvm_memory_slot *slot;
 603	struct kvm_memslots *slots;
 604	int i, idx;
 605
 606	if (WARN_ON_ONCE(range->end <= range->start))
 607		return r;
 608
 609	/* A null handler is allowed if and only if on_lock() is provided. */
 610	if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
 611			 IS_KVM_NULL_FN(range->handler)))
 612		return r;
 613
 614	idx = srcu_read_lock(&kvm->srcu);
 615
 616	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
 617		struct interval_tree_node *node;
 618
 619		slots = __kvm_memslots(kvm, i);
 620		kvm_for_each_memslot_in_hva_range(node, slots,
 621						  range->start, range->end - 1) {
 622			unsigned long hva_start, hva_end;
 623
 624			slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
 625			hva_start = max_t(unsigned long, range->start, slot->userspace_addr);
 626			hva_end = min_t(unsigned long, range->end,
 627					slot->userspace_addr + (slot->npages << PAGE_SHIFT));
 628
 629			/*
 630			 * To optimize for the likely case where the address
 631			 * range is covered by zero or one memslots, don't
 632			 * bother making these conditional (to avoid writes on
 633			 * the second or later invocation of the handler).
 634			 */
 635			gfn_range.arg = range->arg;
 636			gfn_range.may_block = range->may_block;
 637
 638			/*
 639			 * {gfn(page) | page intersects with [hva_start, hva_end)} =
 640			 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
 641			 */
 642			gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
 643			gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
 644			gfn_range.slot = slot;
 645
 646			if (!r.found_memslot) {
 647				r.found_memslot = true;
 648				KVM_MMU_LOCK(kvm);
 649				if (!IS_KVM_NULL_FN(range->on_lock))
 650					range->on_lock(kvm);
 651
 652				if (IS_KVM_NULL_FN(range->handler))
 653					break;
 654			}
 655			r.ret |= range->handler(kvm, &gfn_range);
 656		}
 657	}
 658
 659	if (range->flush_on_ret && r.ret)
 660		kvm_flush_remote_tlbs(kvm);
 661
 662	if (r.found_memslot)
 663		KVM_MMU_UNLOCK(kvm);
 664
 665	srcu_read_unlock(&kvm->srcu, idx);
 666
 667	return r;
 668}
 669
 670static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
 671						unsigned long start,
 672						unsigned long end,
 673						union kvm_mmu_notifier_arg arg,
 674						gfn_handler_t handler)
 675{
 676	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 677	const struct kvm_mmu_notifier_range range = {
 678		.start		= start,
 679		.end		= end,
 680		.arg		= arg,
 681		.handler	= handler,
 682		.on_lock	= (void *)kvm_null_fn,
 683		.flush_on_ret	= true,
 684		.may_block	= false,
 685	};
 686
 687	return __kvm_handle_hva_range(kvm, &range).ret;
 688}
 689
 690static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
 691							 unsigned long start,
 692							 unsigned long end,
 693							 gfn_handler_t handler)
 694{
 695	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 696	const struct kvm_mmu_notifier_range range = {
 697		.start		= start,
 698		.end		= end,
 699		.handler	= handler,
 700		.on_lock	= (void *)kvm_null_fn,
 701		.flush_on_ret	= false,
 702		.may_block	= false,
 703	};
 704
 705	return __kvm_handle_hva_range(kvm, &range).ret;
 706}
 707
 708static bool kvm_change_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
 709{
 710	/*
 711	 * Skipping invalid memslots is correct if and only change_pte() is
 712	 * surrounded by invalidate_range_{start,end}(), which is currently
 713	 * guaranteed by the primary MMU.  If that ever changes, KVM needs to
 714	 * unmap the memslot instead of skipping the memslot to ensure that KVM
 715	 * doesn't hold references to the old PFN.
 716	 */
 717	WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
 718
 719	if (range->slot->flags & KVM_MEMSLOT_INVALID)
 720		return false;
 721
 722	return kvm_set_spte_gfn(kvm, range);
 723}
 724
 725static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 726					struct mm_struct *mm,
 727					unsigned long address,
 728					pte_t pte)
 729{
 730	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 731	const union kvm_mmu_notifier_arg arg = { .pte = pte };
 732
 733	trace_kvm_set_spte_hva(address);
 734
 735	/*
 736	 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
 737	 * If mmu_invalidate_in_progress is zero, then no in-progress
 738	 * invalidations, including this one, found a relevant memslot at
 739	 * start(); rechecking memslots here is unnecessary.  Note, a false
 740	 * positive (count elevated by a different invalidation) is sub-optimal
 741	 * but functionally ok.
 742	 */
 743	WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
 744	if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
 745		return;
 746
 747	kvm_handle_hva_range(mn, address, address + 1, arg, kvm_change_spte_gfn);
 748}
 749
 750void kvm_mmu_invalidate_begin(struct kvm *kvm)
 751{
 752	lockdep_assert_held_write(&kvm->mmu_lock);
 753	/*
 754	 * The count increase must become visible at unlock time as no
 755	 * spte can be established without taking the mmu_lock and
 756	 * count is also read inside the mmu_lock critical section.
 757	 */
 758	kvm->mmu_invalidate_in_progress++;
 759
 760	if (likely(kvm->mmu_invalidate_in_progress == 1)) {
 761		kvm->mmu_invalidate_range_start = INVALID_GPA;
 762		kvm->mmu_invalidate_range_end = INVALID_GPA;
 763	}
 764}
 765
 766void kvm_mmu_invalidate_range_add(struct kvm *kvm, gfn_t start, gfn_t end)
 767{
 768	lockdep_assert_held_write(&kvm->mmu_lock);
 769
 770	WARN_ON_ONCE(!kvm->mmu_invalidate_in_progress);
 771
 772	if (likely(kvm->mmu_invalidate_range_start == INVALID_GPA)) {
 773		kvm->mmu_invalidate_range_start = start;
 774		kvm->mmu_invalidate_range_end = end;
 775	} else {
 776		/*
 777		 * Fully tracking multiple concurrent ranges has diminishing
 778		 * returns. Keep things simple and just find the minimal range
 779		 * which includes the current and new ranges. As there won't be
 780		 * enough information to subtract a range after its invalidate
 781		 * completes, any ranges invalidated concurrently will
 782		 * accumulate and persist until all outstanding invalidates
 783		 * complete.
 784		 */
 785		kvm->mmu_invalidate_range_start =
 786			min(kvm->mmu_invalidate_range_start, start);
 787		kvm->mmu_invalidate_range_end =
 788			max(kvm->mmu_invalidate_range_end, end);
 789	}
 790}
 791
 792bool kvm_mmu_unmap_gfn_range(struct kvm *kvm, struct kvm_gfn_range *range)
 793{
 794	kvm_mmu_invalidate_range_add(kvm, range->start, range->end);
 795	return kvm_unmap_gfn_range(kvm, range);
 796}
 797
 798static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 799					const struct mmu_notifier_range *range)
 800{
 801	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 802	const struct kvm_mmu_notifier_range hva_range = {
 803		.start		= range->start,
 804		.end		= range->end,
 805		.handler	= kvm_mmu_unmap_gfn_range,
 806		.on_lock	= kvm_mmu_invalidate_begin,
 807		.flush_on_ret	= true,
 808		.may_block	= mmu_notifier_range_blockable(range),
 809	};
 810
 811	trace_kvm_unmap_hva_range(range->start, range->end);
 812
 813	/*
 814	 * Prevent memslot modification between range_start() and range_end()
 815	 * so that conditionally locking provides the same result in both
 816	 * functions.  Without that guarantee, the mmu_invalidate_in_progress
 817	 * adjustments will be imbalanced.
 818	 *
 819	 * Pairs with the decrement in range_end().
 820	 */
 821	spin_lock(&kvm->mn_invalidate_lock);
 822	kvm->mn_active_invalidate_count++;
 823	spin_unlock(&kvm->mn_invalidate_lock);
 824
 825	/*
 826	 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
 827	 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
 828	 * each cache's lock.  There are relatively few caches in existence at
 829	 * any given time, and the caches themselves can check for hva overlap,
 830	 * i.e. don't need to rely on memslot overlap checks for performance.
 831	 * Because this runs without holding mmu_lock, the pfn caches must use
 832	 * mn_active_invalidate_count (see above) instead of
 833	 * mmu_invalidate_in_progress.
 834	 */
 835	gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end);
 
 836
 837	/*
 838	 * If one or more memslots were found and thus zapped, notify arch code
 839	 * that guest memory has been reclaimed.  This needs to be done *after*
 840	 * dropping mmu_lock, as x86's reclaim path is slooooow.
 841	 */
 842	if (__kvm_handle_hva_range(kvm, &hva_range).found_memslot)
 843		kvm_arch_guest_memory_reclaimed(kvm);
 844
 845	return 0;
 846}
 847
 848void kvm_mmu_invalidate_end(struct kvm *kvm)
 849{
 850	lockdep_assert_held_write(&kvm->mmu_lock);
 851
 852	/*
 853	 * This sequence increase will notify the kvm page fault that
 854	 * the page that is going to be mapped in the spte could have
 855	 * been freed.
 856	 */
 857	kvm->mmu_invalidate_seq++;
 858	smp_wmb();
 859	/*
 860	 * The above sequence increase must be visible before the
 861	 * below count decrease, which is ensured by the smp_wmb above
 862	 * in conjunction with the smp_rmb in mmu_invalidate_retry().
 863	 */
 864	kvm->mmu_invalidate_in_progress--;
 865	KVM_BUG_ON(kvm->mmu_invalidate_in_progress < 0, kvm);
 866
 867	/*
 868	 * Assert that at least one range was added between start() and end().
 869	 * Not adding a range isn't fatal, but it is a KVM bug.
 870	 */
 871	WARN_ON_ONCE(kvm->mmu_invalidate_range_start == INVALID_GPA);
 872}
 873
 874static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 875					const struct mmu_notifier_range *range)
 876{
 877	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 878	const struct kvm_mmu_notifier_range hva_range = {
 879		.start		= range->start,
 880		.end		= range->end,
 881		.handler	= (void *)kvm_null_fn,
 882		.on_lock	= kvm_mmu_invalidate_end,
 883		.flush_on_ret	= false,
 884		.may_block	= mmu_notifier_range_blockable(range),
 885	};
 886	bool wake;
 887
 888	__kvm_handle_hva_range(kvm, &hva_range);
 889
 890	/* Pairs with the increment in range_start(). */
 891	spin_lock(&kvm->mn_invalidate_lock);
 892	if (!WARN_ON_ONCE(!kvm->mn_active_invalidate_count))
 893		--kvm->mn_active_invalidate_count;
 894	wake = !kvm->mn_active_invalidate_count;
 895	spin_unlock(&kvm->mn_invalidate_lock);
 896
 897	/*
 898	 * There can only be one waiter, since the wait happens under
 899	 * slots_lock.
 900	 */
 901	if (wake)
 902		rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
 903}
 904
 905static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 906					      struct mm_struct *mm,
 907					      unsigned long start,
 908					      unsigned long end)
 909{
 910	trace_kvm_age_hva(start, end);
 911
 912	return kvm_handle_hva_range(mn, start, end, KVM_MMU_NOTIFIER_NO_ARG,
 913				    kvm_age_gfn);
 914}
 915
 916static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
 917					struct mm_struct *mm,
 918					unsigned long start,
 919					unsigned long end)
 920{
 921	trace_kvm_age_hva(start, end);
 922
 923	/*
 924	 * Even though we do not flush TLB, this will still adversely
 925	 * affect performance on pre-Haswell Intel EPT, where there is
 926	 * no EPT Access Bit to clear so that we have to tear down EPT
 927	 * tables instead. If we find this unacceptable, we can always
 928	 * add a parameter to kvm_age_hva so that it effectively doesn't
 929	 * do anything on clear_young.
 930	 *
 931	 * Also note that currently we never issue secondary TLB flushes
 932	 * from clear_young, leaving this job up to the regular system
 933	 * cadence. If we find this inaccurate, we might come up with a
 934	 * more sophisticated heuristic later.
 935	 */
 936	return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
 937}
 938
 939static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 940				       struct mm_struct *mm,
 941				       unsigned long address)
 942{
 943	trace_kvm_test_age_hva(address);
 944
 945	return kvm_handle_hva_range_no_flush(mn, address, address + 1,
 946					     kvm_test_age_gfn);
 947}
 948
 949static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 950				     struct mm_struct *mm)
 951{
 952	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 953	int idx;
 954
 955	idx = srcu_read_lock(&kvm->srcu);
 956	kvm_flush_shadow_all(kvm);
 957	srcu_read_unlock(&kvm->srcu, idx);
 958}
 959
 960static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 961	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
 962	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
 963	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
 964	.clear_young		= kvm_mmu_notifier_clear_young,
 965	.test_young		= kvm_mmu_notifier_test_young,
 966	.change_pte		= kvm_mmu_notifier_change_pte,
 967	.release		= kvm_mmu_notifier_release,
 968};
 969
 970static int kvm_init_mmu_notifier(struct kvm *kvm)
 971{
 972	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 973	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 974}
 975
 976#else  /* !CONFIG_KVM_GENERIC_MMU_NOTIFIER */
 977
 978static int kvm_init_mmu_notifier(struct kvm *kvm)
 979{
 980	return 0;
 981}
 982
 983#endif /* CONFIG_KVM_GENERIC_MMU_NOTIFIER */
 984
 985#ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
 986static int kvm_pm_notifier_call(struct notifier_block *bl,
 987				unsigned long state,
 988				void *unused)
 989{
 990	struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
 991
 992	return kvm_arch_pm_notifier(kvm, state);
 993}
 994
 995static void kvm_init_pm_notifier(struct kvm *kvm)
 996{
 997	kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
 998	/* Suspend KVM before we suspend ftrace, RCU, etc. */
 999	kvm->pm_notifier.priority = INT_MAX;
1000	register_pm_notifier(&kvm->pm_notifier);
1001}
1002
1003static void kvm_destroy_pm_notifier(struct kvm *kvm)
1004{
1005	unregister_pm_notifier(&kvm->pm_notifier);
1006}
1007#else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
1008static void kvm_init_pm_notifier(struct kvm *kvm)
1009{
1010}
1011
1012static void kvm_destroy_pm_notifier(struct kvm *kvm)
1013{
1014}
1015#endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
1016
1017static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
1018{
1019	if (!memslot->dirty_bitmap)
1020		return;
1021
1022	kvfree(memslot->dirty_bitmap);
1023	memslot->dirty_bitmap = NULL;
1024}
1025
1026/* This does not remove the slot from struct kvm_memslots data structures */
1027static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
1028{
1029	if (slot->flags & KVM_MEM_GUEST_MEMFD)
1030		kvm_gmem_unbind(slot);
1031
1032	kvm_destroy_dirty_bitmap(slot);
1033
1034	kvm_arch_free_memslot(kvm, slot);
1035
1036	kfree(slot);
1037}
1038
1039static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
1040{
1041	struct hlist_node *idnode;
1042	struct kvm_memory_slot *memslot;
1043	int bkt;
1044
1045	/*
1046	 * The same memslot objects live in both active and inactive sets,
1047	 * arbitrarily free using index '1' so the second invocation of this
1048	 * function isn't operating over a structure with dangling pointers
1049	 * (even though this function isn't actually touching them).
1050	 */
1051	if (!slots->node_idx)
1052		return;
1053
1054	hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
1055		kvm_free_memslot(kvm, memslot);
1056}
1057
1058static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
1059{
1060	switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
1061	case KVM_STATS_TYPE_INSTANT:
1062		return 0444;
1063	case KVM_STATS_TYPE_CUMULATIVE:
1064	case KVM_STATS_TYPE_PEAK:
1065	default:
1066		return 0644;
1067	}
1068}
1069
1070
1071static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1072{
1073	int i;
1074	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1075				      kvm_vcpu_stats_header.num_desc;
1076
1077	if (IS_ERR(kvm->debugfs_dentry))
1078		return;
1079
1080	debugfs_remove_recursive(kvm->debugfs_dentry);
1081
1082	if (kvm->debugfs_stat_data) {
1083		for (i = 0; i < kvm_debugfs_num_entries; i++)
1084			kfree(kvm->debugfs_stat_data[i]);
1085		kfree(kvm->debugfs_stat_data);
1086	}
1087}
1088
1089static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1090{
1091	static DEFINE_MUTEX(kvm_debugfs_lock);
1092	struct dentry *dent;
1093	char dir_name[ITOA_MAX_LEN * 2];
1094	struct kvm_stat_data *stat_data;
1095	const struct _kvm_stats_desc *pdesc;
1096	int i, ret = -ENOMEM;
1097	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1098				      kvm_vcpu_stats_header.num_desc;
1099
1100	if (!debugfs_initialized())
1101		return 0;
1102
1103	snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1104	mutex_lock(&kvm_debugfs_lock);
1105	dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1106	if (dent) {
1107		pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1108		dput(dent);
1109		mutex_unlock(&kvm_debugfs_lock);
1110		return 0;
1111	}
1112	dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1113	mutex_unlock(&kvm_debugfs_lock);
1114	if (IS_ERR(dent))
1115		return 0;
1116
1117	kvm->debugfs_dentry = dent;
1118	kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1119					 sizeof(*kvm->debugfs_stat_data),
1120					 GFP_KERNEL_ACCOUNT);
1121	if (!kvm->debugfs_stat_data)
1122		goto out_err;
1123
1124	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1125		pdesc = &kvm_vm_stats_desc[i];
1126		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1127		if (!stat_data)
1128			goto out_err;
1129
1130		stat_data->kvm = kvm;
1131		stat_data->desc = pdesc;
1132		stat_data->kind = KVM_STAT_VM;
1133		kvm->debugfs_stat_data[i] = stat_data;
1134		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1135				    kvm->debugfs_dentry, stat_data,
1136				    &stat_fops_per_vm);
1137	}
1138
1139	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1140		pdesc = &kvm_vcpu_stats_desc[i];
1141		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1142		if (!stat_data)
1143			goto out_err;
1144
1145		stat_data->kvm = kvm;
1146		stat_data->desc = pdesc;
1147		stat_data->kind = KVM_STAT_VCPU;
1148		kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1149		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1150				    kvm->debugfs_dentry, stat_data,
1151				    &stat_fops_per_vm);
1152	}
1153
1154	kvm_arch_create_vm_debugfs(kvm);
 
 
 
1155	return 0;
1156out_err:
1157	kvm_destroy_vm_debugfs(kvm);
1158	return ret;
1159}
1160
1161/*
1162 * Called after the VM is otherwise initialized, but just before adding it to
1163 * the vm_list.
1164 */
1165int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1166{
1167	return 0;
1168}
1169
1170/*
1171 * Called just after removing the VM from the vm_list, but before doing any
1172 * other destruction.
1173 */
1174void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1175{
1176}
1177
1178/*
1179 * Called after per-vm debugfs created.  When called kvm->debugfs_dentry should
1180 * be setup already, so we can create arch-specific debugfs entries under it.
1181 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1182 * a per-arch destroy interface is not needed.
1183 */
1184void __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1185{
 
1186}
1187
1188static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1189{
1190	struct kvm *kvm = kvm_arch_alloc_vm();
1191	struct kvm_memslots *slots;
1192	int r = -ENOMEM;
1193	int i, j;
1194
1195	if (!kvm)
1196		return ERR_PTR(-ENOMEM);
1197
1198	KVM_MMU_LOCK_INIT(kvm);
1199	mmgrab(current->mm);
1200	kvm->mm = current->mm;
1201	kvm_eventfd_init(kvm);
1202	mutex_init(&kvm->lock);
1203	mutex_init(&kvm->irq_lock);
1204	mutex_init(&kvm->slots_lock);
1205	mutex_init(&kvm->slots_arch_lock);
1206	spin_lock_init(&kvm->mn_invalidate_lock);
1207	rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1208	xa_init(&kvm->vcpu_array);
1209#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1210	xa_init(&kvm->mem_attr_array);
1211#endif
1212
1213	INIT_LIST_HEAD(&kvm->gpc_list);
1214	spin_lock_init(&kvm->gpc_lock);
1215
1216	INIT_LIST_HEAD(&kvm->devices);
1217	kvm->max_vcpus = KVM_MAX_VCPUS;
1218
1219	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1220
1221	/*
1222	 * Force subsequent debugfs file creations to fail if the VM directory
1223	 * is not created (by kvm_create_vm_debugfs()).
1224	 */
1225	kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1226
1227	snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1228		 task_pid_nr(current));
1229
1230	if (init_srcu_struct(&kvm->srcu))
1231		goto out_err_no_srcu;
1232	if (init_srcu_struct(&kvm->irq_srcu))
1233		goto out_err_no_irq_srcu;
1234
1235	refcount_set(&kvm->users_count, 1);
1236	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
1237		for (j = 0; j < 2; j++) {
1238			slots = &kvm->__memslots[i][j];
1239
1240			atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1241			slots->hva_tree = RB_ROOT_CACHED;
1242			slots->gfn_tree = RB_ROOT;
1243			hash_init(slots->id_hash);
1244			slots->node_idx = j;
1245
1246			/* Generations must be different for each address space. */
1247			slots->generation = i;
1248		}
1249
1250		rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1251	}
1252
1253	for (i = 0; i < KVM_NR_BUSES; i++) {
1254		rcu_assign_pointer(kvm->buses[i],
1255			kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1256		if (!kvm->buses[i])
1257			goto out_err_no_arch_destroy_vm;
1258	}
1259
1260	r = kvm_arch_init_vm(kvm, type);
1261	if (r)
1262		goto out_err_no_arch_destroy_vm;
1263
1264	r = hardware_enable_all();
1265	if (r)
1266		goto out_err_no_disable;
1267
1268#ifdef CONFIG_HAVE_KVM_IRQCHIP
1269	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1270#endif
1271
1272	r = kvm_init_mmu_notifier(kvm);
1273	if (r)
1274		goto out_err_no_mmu_notifier;
1275
1276	r = kvm_coalesced_mmio_init(kvm);
1277	if (r < 0)
1278		goto out_no_coalesced_mmio;
1279
1280	r = kvm_create_vm_debugfs(kvm, fdname);
1281	if (r)
1282		goto out_err_no_debugfs;
1283
1284	r = kvm_arch_post_init_vm(kvm);
1285	if (r)
1286		goto out_err;
1287
1288	mutex_lock(&kvm_lock);
1289	list_add(&kvm->vm_list, &vm_list);
1290	mutex_unlock(&kvm_lock);
1291
1292	preempt_notifier_inc();
1293	kvm_init_pm_notifier(kvm);
1294
1295	return kvm;
1296
1297out_err:
1298	kvm_destroy_vm_debugfs(kvm);
1299out_err_no_debugfs:
1300	kvm_coalesced_mmio_free(kvm);
1301out_no_coalesced_mmio:
1302#ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1303	if (kvm->mmu_notifier.ops)
1304		mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1305#endif
1306out_err_no_mmu_notifier:
1307	hardware_disable_all();
1308out_err_no_disable:
1309	kvm_arch_destroy_vm(kvm);
1310out_err_no_arch_destroy_vm:
1311	WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1312	for (i = 0; i < KVM_NR_BUSES; i++)
1313		kfree(kvm_get_bus(kvm, i));
1314	cleanup_srcu_struct(&kvm->irq_srcu);
1315out_err_no_irq_srcu:
1316	cleanup_srcu_struct(&kvm->srcu);
1317out_err_no_srcu:
1318	kvm_arch_free_vm(kvm);
1319	mmdrop(current->mm);
1320	return ERR_PTR(r);
1321}
1322
1323static void kvm_destroy_devices(struct kvm *kvm)
1324{
1325	struct kvm_device *dev, *tmp;
1326
1327	/*
1328	 * We do not need to take the kvm->lock here, because nobody else
1329	 * has a reference to the struct kvm at this point and therefore
1330	 * cannot access the devices list anyhow.
1331	 */
1332	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1333		list_del(&dev->vm_node);
1334		dev->ops->destroy(dev);
1335	}
1336}
1337
1338static void kvm_destroy_vm(struct kvm *kvm)
1339{
1340	int i;
1341	struct mm_struct *mm = kvm->mm;
1342
1343	kvm_destroy_pm_notifier(kvm);
1344	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1345	kvm_destroy_vm_debugfs(kvm);
1346	kvm_arch_sync_events(kvm);
1347	mutex_lock(&kvm_lock);
1348	list_del(&kvm->vm_list);
1349	mutex_unlock(&kvm_lock);
1350	kvm_arch_pre_destroy_vm(kvm);
1351
1352	kvm_free_irq_routing(kvm);
1353	for (i = 0; i < KVM_NR_BUSES; i++) {
1354		struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1355
1356		if (bus)
1357			kvm_io_bus_destroy(bus);
1358		kvm->buses[i] = NULL;
1359	}
1360	kvm_coalesced_mmio_free(kvm);
1361#ifdef CONFIG_KVM_GENERIC_MMU_NOTIFIER
1362	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1363	/*
1364	 * At this point, pending calls to invalidate_range_start()
1365	 * have completed but no more MMU notifiers will run, so
1366	 * mn_active_invalidate_count may remain unbalanced.
1367	 * No threads can be waiting in kvm_swap_active_memslots() as the
1368	 * last reference on KVM has been dropped, but freeing
1369	 * memslots would deadlock without this manual intervention.
1370	 *
1371	 * If the count isn't unbalanced, i.e. KVM did NOT unregister its MMU
1372	 * notifier between a start() and end(), then there shouldn't be any
1373	 * in-progress invalidations.
1374	 */
1375	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1376	if (kvm->mn_active_invalidate_count)
1377		kvm->mn_active_invalidate_count = 0;
1378	else
1379		WARN_ON(kvm->mmu_invalidate_in_progress);
1380#else
1381	kvm_flush_shadow_all(kvm);
1382#endif
1383	kvm_arch_destroy_vm(kvm);
1384	kvm_destroy_devices(kvm);
1385	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
1386		kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1387		kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1388	}
1389	cleanup_srcu_struct(&kvm->irq_srcu);
1390	cleanup_srcu_struct(&kvm->srcu);
1391#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
1392	xa_destroy(&kvm->mem_attr_array);
1393#endif
1394	kvm_arch_free_vm(kvm);
1395	preempt_notifier_dec();
1396	hardware_disable_all();
1397	mmdrop(mm);
1398}
1399
1400void kvm_get_kvm(struct kvm *kvm)
1401{
1402	refcount_inc(&kvm->users_count);
1403}
1404EXPORT_SYMBOL_GPL(kvm_get_kvm);
1405
1406/*
1407 * Make sure the vm is not during destruction, which is a safe version of
1408 * kvm_get_kvm().  Return true if kvm referenced successfully, false otherwise.
1409 */
1410bool kvm_get_kvm_safe(struct kvm *kvm)
1411{
1412	return refcount_inc_not_zero(&kvm->users_count);
1413}
1414EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1415
1416void kvm_put_kvm(struct kvm *kvm)
1417{
1418	if (refcount_dec_and_test(&kvm->users_count))
1419		kvm_destroy_vm(kvm);
1420}
1421EXPORT_SYMBOL_GPL(kvm_put_kvm);
1422
1423/*
1424 * Used to put a reference that was taken on behalf of an object associated
1425 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1426 * of the new file descriptor fails and the reference cannot be transferred to
1427 * its final owner.  In such cases, the caller is still actively using @kvm and
1428 * will fail miserably if the refcount unexpectedly hits zero.
1429 */
1430void kvm_put_kvm_no_destroy(struct kvm *kvm)
1431{
1432	WARN_ON(refcount_dec_and_test(&kvm->users_count));
1433}
1434EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1435
1436static int kvm_vm_release(struct inode *inode, struct file *filp)
1437{
1438	struct kvm *kvm = filp->private_data;
1439
1440	kvm_irqfd_release(kvm);
1441
1442	kvm_put_kvm(kvm);
1443	return 0;
1444}
1445
1446/*
1447 * Allocation size is twice as large as the actual dirty bitmap size.
1448 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1449 */
1450static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1451{
1452	unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1453
1454	memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1455	if (!memslot->dirty_bitmap)
1456		return -ENOMEM;
1457
1458	return 0;
1459}
1460
1461static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1462{
1463	struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1464	int node_idx_inactive = active->node_idx ^ 1;
1465
1466	return &kvm->__memslots[as_id][node_idx_inactive];
1467}
1468
1469/*
1470 * Helper to get the address space ID when one of memslot pointers may be NULL.
1471 * This also serves as a sanity that at least one of the pointers is non-NULL,
1472 * and that their address space IDs don't diverge.
1473 */
1474static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1475				  struct kvm_memory_slot *b)
1476{
1477	if (WARN_ON_ONCE(!a && !b))
1478		return 0;
1479
1480	if (!a)
1481		return b->as_id;
1482	if (!b)
1483		return a->as_id;
1484
1485	WARN_ON_ONCE(a->as_id != b->as_id);
1486	return a->as_id;
1487}
1488
1489static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1490				struct kvm_memory_slot *slot)
1491{
1492	struct rb_root *gfn_tree = &slots->gfn_tree;
1493	struct rb_node **node, *parent;
1494	int idx = slots->node_idx;
1495
1496	parent = NULL;
1497	for (node = &gfn_tree->rb_node; *node; ) {
1498		struct kvm_memory_slot *tmp;
1499
1500		tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1501		parent = *node;
1502		if (slot->base_gfn < tmp->base_gfn)
1503			node = &(*node)->rb_left;
1504		else if (slot->base_gfn > tmp->base_gfn)
1505			node = &(*node)->rb_right;
1506		else
1507			BUG();
1508	}
1509
1510	rb_link_node(&slot->gfn_node[idx], parent, node);
1511	rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1512}
1513
1514static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1515			       struct kvm_memory_slot *slot)
1516{
1517	rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1518}
1519
1520static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1521				 struct kvm_memory_slot *old,
1522				 struct kvm_memory_slot *new)
1523{
1524	int idx = slots->node_idx;
1525
1526	WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1527
1528	rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1529			&slots->gfn_tree);
1530}
1531
1532/*
1533 * Replace @old with @new in the inactive memslots.
1534 *
1535 * With NULL @old this simply adds @new.
1536 * With NULL @new this simply removes @old.
1537 *
1538 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1539 * appropriately.
1540 */
1541static void kvm_replace_memslot(struct kvm *kvm,
1542				struct kvm_memory_slot *old,
1543				struct kvm_memory_slot *new)
1544{
1545	int as_id = kvm_memslots_get_as_id(old, new);
1546	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1547	int idx = slots->node_idx;
1548
1549	if (old) {
1550		hash_del(&old->id_node[idx]);
1551		interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1552
1553		if ((long)old == atomic_long_read(&slots->last_used_slot))
1554			atomic_long_set(&slots->last_used_slot, (long)new);
1555
1556		if (!new) {
1557			kvm_erase_gfn_node(slots, old);
1558			return;
1559		}
1560	}
1561
1562	/*
1563	 * Initialize @new's hva range.  Do this even when replacing an @old
1564	 * slot, kvm_copy_memslot() deliberately does not touch node data.
1565	 */
1566	new->hva_node[idx].start = new->userspace_addr;
1567	new->hva_node[idx].last = new->userspace_addr +
1568				  (new->npages << PAGE_SHIFT) - 1;
1569
1570	/*
1571	 * (Re)Add the new memslot.  There is no O(1) interval_tree_replace(),
1572	 * hva_node needs to be swapped with remove+insert even though hva can't
1573	 * change when replacing an existing slot.
1574	 */
1575	hash_add(slots->id_hash, &new->id_node[idx], new->id);
1576	interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1577
1578	/*
1579	 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1580	 * switch the node in the gfn tree instead of removing the old and
1581	 * inserting the new as two separate operations. Replacement is a
1582	 * single O(1) operation versus two O(log(n)) operations for
1583	 * remove+insert.
1584	 */
1585	if (old && old->base_gfn == new->base_gfn) {
1586		kvm_replace_gfn_node(slots, old, new);
1587	} else {
1588		if (old)
1589			kvm_erase_gfn_node(slots, old);
1590		kvm_insert_gfn_node(slots, new);
1591	}
1592}
1593
1594/*
1595 * Flags that do not access any of the extra space of struct
1596 * kvm_userspace_memory_region2.  KVM_SET_USER_MEMORY_REGION_V1_FLAGS
1597 * only allows these.
1598 */
1599#define KVM_SET_USER_MEMORY_REGION_V1_FLAGS \
1600	(KVM_MEM_LOG_DIRTY_PAGES | KVM_MEM_READONLY)
1601
1602static int check_memory_region_flags(struct kvm *kvm,
1603				     const struct kvm_userspace_memory_region2 *mem)
1604{
1605	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1606
1607	if (kvm_arch_has_private_mem(kvm))
1608		valid_flags |= KVM_MEM_GUEST_MEMFD;
1609
1610	/* Dirty logging private memory is not currently supported. */
1611	if (mem->flags & KVM_MEM_GUEST_MEMFD)
1612		valid_flags &= ~KVM_MEM_LOG_DIRTY_PAGES;
1613
1614#ifdef CONFIG_HAVE_KVM_READONLY_MEM
1615	/*
1616	 * GUEST_MEMFD is incompatible with read-only memslots, as writes to
1617	 * read-only memslots have emulated MMIO, not page fault, semantics,
1618	 * and KVM doesn't allow emulated MMIO for private memory.
1619	 */
1620	if (!(mem->flags & KVM_MEM_GUEST_MEMFD))
1621		valid_flags |= KVM_MEM_READONLY;
1622#endif
1623
1624	if (mem->flags & ~valid_flags)
1625		return -EINVAL;
1626
1627	return 0;
1628}
1629
1630static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1631{
1632	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1633
1634	/* Grab the generation from the activate memslots. */
1635	u64 gen = __kvm_memslots(kvm, as_id)->generation;
1636
1637	WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1638	slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1639
1640	/*
1641	 * Do not store the new memslots while there are invalidations in
1642	 * progress, otherwise the locking in invalidate_range_start and
1643	 * invalidate_range_end will be unbalanced.
1644	 */
1645	spin_lock(&kvm->mn_invalidate_lock);
1646	prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1647	while (kvm->mn_active_invalidate_count) {
1648		set_current_state(TASK_UNINTERRUPTIBLE);
1649		spin_unlock(&kvm->mn_invalidate_lock);
1650		schedule();
1651		spin_lock(&kvm->mn_invalidate_lock);
1652	}
1653	finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1654	rcu_assign_pointer(kvm->memslots[as_id], slots);
1655	spin_unlock(&kvm->mn_invalidate_lock);
1656
1657	/*
1658	 * Acquired in kvm_set_memslot. Must be released before synchronize
1659	 * SRCU below in order to avoid deadlock with another thread
1660	 * acquiring the slots_arch_lock in an srcu critical section.
1661	 */
1662	mutex_unlock(&kvm->slots_arch_lock);
1663
1664	synchronize_srcu_expedited(&kvm->srcu);
1665
1666	/*
1667	 * Increment the new memslot generation a second time, dropping the
1668	 * update in-progress flag and incrementing the generation based on
1669	 * the number of address spaces.  This provides a unique and easily
1670	 * identifiable generation number while the memslots are in flux.
1671	 */
1672	gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1673
1674	/*
1675	 * Generations must be unique even across address spaces.  We do not need
1676	 * a global counter for that, instead the generation space is evenly split
1677	 * across address spaces.  For example, with two address spaces, address
1678	 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1679	 * use generations 1, 3, 5, ...
1680	 */
1681	gen += kvm_arch_nr_memslot_as_ids(kvm);
1682
1683	kvm_arch_memslots_updated(kvm, gen);
1684
1685	slots->generation = gen;
1686}
1687
1688static int kvm_prepare_memory_region(struct kvm *kvm,
1689				     const struct kvm_memory_slot *old,
1690				     struct kvm_memory_slot *new,
1691				     enum kvm_mr_change change)
1692{
1693	int r;
1694
1695	/*
1696	 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1697	 * will be freed on "commit".  If logging is enabled in both old and
1698	 * new, reuse the existing bitmap.  If logging is enabled only in the
1699	 * new and KVM isn't using a ring buffer, allocate and initialize a
1700	 * new bitmap.
1701	 */
1702	if (change != KVM_MR_DELETE) {
1703		if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1704			new->dirty_bitmap = NULL;
1705		else if (old && old->dirty_bitmap)
1706			new->dirty_bitmap = old->dirty_bitmap;
1707		else if (kvm_use_dirty_bitmap(kvm)) {
1708			r = kvm_alloc_dirty_bitmap(new);
1709			if (r)
1710				return r;
1711
1712			if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1713				bitmap_set(new->dirty_bitmap, 0, new->npages);
1714		}
1715	}
1716
1717	r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1718
1719	/* Free the bitmap on failure if it was allocated above. */
1720	if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1721		kvm_destroy_dirty_bitmap(new);
1722
1723	return r;
1724}
1725
1726static void kvm_commit_memory_region(struct kvm *kvm,
1727				     struct kvm_memory_slot *old,
1728				     const struct kvm_memory_slot *new,
1729				     enum kvm_mr_change change)
1730{
1731	int old_flags = old ? old->flags : 0;
1732	int new_flags = new ? new->flags : 0;
1733	/*
1734	 * Update the total number of memslot pages before calling the arch
1735	 * hook so that architectures can consume the result directly.
1736	 */
1737	if (change == KVM_MR_DELETE)
1738		kvm->nr_memslot_pages -= old->npages;
1739	else if (change == KVM_MR_CREATE)
1740		kvm->nr_memslot_pages += new->npages;
1741
1742	if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1743		int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1744		atomic_set(&kvm->nr_memslots_dirty_logging,
1745			   atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1746	}
1747
1748	kvm_arch_commit_memory_region(kvm, old, new, change);
1749
1750	switch (change) {
1751	case KVM_MR_CREATE:
1752		/* Nothing more to do. */
1753		break;
1754	case KVM_MR_DELETE:
1755		/* Free the old memslot and all its metadata. */
1756		kvm_free_memslot(kvm, old);
1757		break;
1758	case KVM_MR_MOVE:
1759	case KVM_MR_FLAGS_ONLY:
1760		/*
1761		 * Free the dirty bitmap as needed; the below check encompasses
1762		 * both the flags and whether a ring buffer is being used)
1763		 */
1764		if (old->dirty_bitmap && !new->dirty_bitmap)
1765			kvm_destroy_dirty_bitmap(old);
1766
1767		/*
1768		 * The final quirk.  Free the detached, old slot, but only its
1769		 * memory, not any metadata.  Metadata, including arch specific
1770		 * data, may be reused by @new.
1771		 */
1772		kfree(old);
1773		break;
1774	default:
1775		BUG();
1776	}
1777}
1778
1779/*
1780 * Activate @new, which must be installed in the inactive slots by the caller,
1781 * by swapping the active slots and then propagating @new to @old once @old is
1782 * unreachable and can be safely modified.
1783 *
1784 * With NULL @old this simply adds @new to @active (while swapping the sets).
1785 * With NULL @new this simply removes @old from @active and frees it
1786 * (while also swapping the sets).
1787 */
1788static void kvm_activate_memslot(struct kvm *kvm,
1789				 struct kvm_memory_slot *old,
1790				 struct kvm_memory_slot *new)
1791{
1792	int as_id = kvm_memslots_get_as_id(old, new);
1793
1794	kvm_swap_active_memslots(kvm, as_id);
1795
1796	/* Propagate the new memslot to the now inactive memslots. */
1797	kvm_replace_memslot(kvm, old, new);
1798}
1799
1800static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1801			     const struct kvm_memory_slot *src)
1802{
1803	dest->base_gfn = src->base_gfn;
1804	dest->npages = src->npages;
1805	dest->dirty_bitmap = src->dirty_bitmap;
1806	dest->arch = src->arch;
1807	dest->userspace_addr = src->userspace_addr;
1808	dest->flags = src->flags;
1809	dest->id = src->id;
1810	dest->as_id = src->as_id;
1811}
1812
1813static void kvm_invalidate_memslot(struct kvm *kvm,
1814				   struct kvm_memory_slot *old,
1815				   struct kvm_memory_slot *invalid_slot)
1816{
1817	/*
1818	 * Mark the current slot INVALID.  As with all memslot modifications,
1819	 * this must be done on an unreachable slot to avoid modifying the
1820	 * current slot in the active tree.
1821	 */
1822	kvm_copy_memslot(invalid_slot, old);
1823	invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1824	kvm_replace_memslot(kvm, old, invalid_slot);
1825
1826	/*
1827	 * Activate the slot that is now marked INVALID, but don't propagate
1828	 * the slot to the now inactive slots. The slot is either going to be
1829	 * deleted or recreated as a new slot.
1830	 */
1831	kvm_swap_active_memslots(kvm, old->as_id);
1832
1833	/*
1834	 * From this point no new shadow pages pointing to a deleted, or moved,
1835	 * memslot will be created.  Validation of sp->gfn happens in:
1836	 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1837	 *	- kvm_is_visible_gfn (mmu_check_root)
1838	 */
1839	kvm_arch_flush_shadow_memslot(kvm, old);
1840	kvm_arch_guest_memory_reclaimed(kvm);
1841
1842	/* Was released by kvm_swap_active_memslots(), reacquire. */
1843	mutex_lock(&kvm->slots_arch_lock);
1844
1845	/*
1846	 * Copy the arch-specific field of the newly-installed slot back to the
1847	 * old slot as the arch data could have changed between releasing
1848	 * slots_arch_lock in kvm_swap_active_memslots() and re-acquiring the lock
1849	 * above.  Writers are required to retrieve memslots *after* acquiring
1850	 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1851	 */
1852	old->arch = invalid_slot->arch;
1853}
1854
1855static void kvm_create_memslot(struct kvm *kvm,
1856			       struct kvm_memory_slot *new)
1857{
1858	/* Add the new memslot to the inactive set and activate. */
1859	kvm_replace_memslot(kvm, NULL, new);
1860	kvm_activate_memslot(kvm, NULL, new);
1861}
1862
1863static void kvm_delete_memslot(struct kvm *kvm,
1864			       struct kvm_memory_slot *old,
1865			       struct kvm_memory_slot *invalid_slot)
1866{
1867	/*
1868	 * Remove the old memslot (in the inactive memslots) by passing NULL as
1869	 * the "new" slot, and for the invalid version in the active slots.
1870	 */
1871	kvm_replace_memslot(kvm, old, NULL);
1872	kvm_activate_memslot(kvm, invalid_slot, NULL);
1873}
1874
1875static void kvm_move_memslot(struct kvm *kvm,
1876			     struct kvm_memory_slot *old,
1877			     struct kvm_memory_slot *new,
1878			     struct kvm_memory_slot *invalid_slot)
1879{
1880	/*
1881	 * Replace the old memslot in the inactive slots, and then swap slots
1882	 * and replace the current INVALID with the new as well.
1883	 */
1884	kvm_replace_memslot(kvm, old, new);
1885	kvm_activate_memslot(kvm, invalid_slot, new);
1886}
1887
1888static void kvm_update_flags_memslot(struct kvm *kvm,
1889				     struct kvm_memory_slot *old,
1890				     struct kvm_memory_slot *new)
1891{
1892	/*
1893	 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1894	 * an intermediate step. Instead, the old memslot is simply replaced
1895	 * with a new, updated copy in both memslot sets.
1896	 */
1897	kvm_replace_memslot(kvm, old, new);
1898	kvm_activate_memslot(kvm, old, new);
1899}
1900
1901static int kvm_set_memslot(struct kvm *kvm,
1902			   struct kvm_memory_slot *old,
1903			   struct kvm_memory_slot *new,
1904			   enum kvm_mr_change change)
1905{
1906	struct kvm_memory_slot *invalid_slot;
1907	int r;
1908
1909	/*
1910	 * Released in kvm_swap_active_memslots().
1911	 *
1912	 * Must be held from before the current memslots are copied until after
1913	 * the new memslots are installed with rcu_assign_pointer, then
1914	 * released before the synchronize srcu in kvm_swap_active_memslots().
1915	 *
1916	 * When modifying memslots outside of the slots_lock, must be held
1917	 * before reading the pointer to the current memslots until after all
1918	 * changes to those memslots are complete.
1919	 *
1920	 * These rules ensure that installing new memslots does not lose
1921	 * changes made to the previous memslots.
1922	 */
1923	mutex_lock(&kvm->slots_arch_lock);
1924
1925	/*
1926	 * Invalidate the old slot if it's being deleted or moved.  This is
1927	 * done prior to actually deleting/moving the memslot to allow vCPUs to
1928	 * continue running by ensuring there are no mappings or shadow pages
1929	 * for the memslot when it is deleted/moved.  Without pre-invalidation
1930	 * (and without a lock), a window would exist between effecting the
1931	 * delete/move and committing the changes in arch code where KVM or a
1932	 * guest could access a non-existent memslot.
1933	 *
1934	 * Modifications are done on a temporary, unreachable slot.  The old
1935	 * slot needs to be preserved in case a later step fails and the
1936	 * invalidation needs to be reverted.
1937	 */
1938	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1939		invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1940		if (!invalid_slot) {
1941			mutex_unlock(&kvm->slots_arch_lock);
1942			return -ENOMEM;
1943		}
1944		kvm_invalidate_memslot(kvm, old, invalid_slot);
1945	}
1946
1947	r = kvm_prepare_memory_region(kvm, old, new, change);
1948	if (r) {
1949		/*
1950		 * For DELETE/MOVE, revert the above INVALID change.  No
1951		 * modifications required since the original slot was preserved
1952		 * in the inactive slots.  Changing the active memslots also
1953		 * release slots_arch_lock.
1954		 */
1955		if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1956			kvm_activate_memslot(kvm, invalid_slot, old);
1957			kfree(invalid_slot);
1958		} else {
1959			mutex_unlock(&kvm->slots_arch_lock);
1960		}
1961		return r;
1962	}
1963
1964	/*
1965	 * For DELETE and MOVE, the working slot is now active as the INVALID
1966	 * version of the old slot.  MOVE is particularly special as it reuses
1967	 * the old slot and returns a copy of the old slot (in working_slot).
1968	 * For CREATE, there is no old slot.  For DELETE and FLAGS_ONLY, the
1969	 * old slot is detached but otherwise preserved.
1970	 */
1971	if (change == KVM_MR_CREATE)
1972		kvm_create_memslot(kvm, new);
1973	else if (change == KVM_MR_DELETE)
1974		kvm_delete_memslot(kvm, old, invalid_slot);
1975	else if (change == KVM_MR_MOVE)
1976		kvm_move_memslot(kvm, old, new, invalid_slot);
1977	else if (change == KVM_MR_FLAGS_ONLY)
1978		kvm_update_flags_memslot(kvm, old, new);
1979	else
1980		BUG();
1981
1982	/* Free the temporary INVALID slot used for DELETE and MOVE. */
1983	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1984		kfree(invalid_slot);
1985
1986	/*
1987	 * No need to refresh new->arch, changes after dropping slots_arch_lock
1988	 * will directly hit the final, active memslot.  Architectures are
1989	 * responsible for knowing that new->arch may be stale.
1990	 */
1991	kvm_commit_memory_region(kvm, old, new, change);
1992
1993	return 0;
1994}
1995
1996static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1997				      gfn_t start, gfn_t end)
1998{
1999	struct kvm_memslot_iter iter;
2000
2001	kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
2002		if (iter.slot->id != id)
2003			return true;
2004	}
2005
2006	return false;
2007}
2008
2009/*
2010 * Allocate some memory and give it an address in the guest physical address
2011 * space.
2012 *
2013 * Discontiguous memory is allowed, mostly for framebuffers.
2014 *
2015 * Must be called holding kvm->slots_lock for write.
2016 */
2017int __kvm_set_memory_region(struct kvm *kvm,
2018			    const struct kvm_userspace_memory_region2 *mem)
2019{
2020	struct kvm_memory_slot *old, *new;
2021	struct kvm_memslots *slots;
2022	enum kvm_mr_change change;
2023	unsigned long npages;
2024	gfn_t base_gfn;
2025	int as_id, id;
2026	int r;
2027
2028	r = check_memory_region_flags(kvm, mem);
2029	if (r)
2030		return r;
2031
2032	as_id = mem->slot >> 16;
2033	id = (u16)mem->slot;
2034
2035	/* General sanity checks */
2036	if ((mem->memory_size & (PAGE_SIZE - 1)) ||
2037	    (mem->memory_size != (unsigned long)mem->memory_size))
2038		return -EINVAL;
2039	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
2040		return -EINVAL;
2041	/* We can read the guest memory with __xxx_user() later on. */
2042	if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
2043	    (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
2044	     !access_ok((void __user *)(unsigned long)mem->userspace_addr,
2045			mem->memory_size))
2046		return -EINVAL;
2047	if (mem->flags & KVM_MEM_GUEST_MEMFD &&
2048	    (mem->guest_memfd_offset & (PAGE_SIZE - 1) ||
2049	     mem->guest_memfd_offset + mem->memory_size < mem->guest_memfd_offset))
2050		return -EINVAL;
2051	if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_MEM_SLOTS_NUM)
2052		return -EINVAL;
2053	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
2054		return -EINVAL;
2055	if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
2056		return -EINVAL;
2057
2058	slots = __kvm_memslots(kvm, as_id);
2059
2060	/*
2061	 * Note, the old memslot (and the pointer itself!) may be invalidated
2062	 * and/or destroyed by kvm_set_memslot().
2063	 */
2064	old = id_to_memslot(slots, id);
2065
2066	if (!mem->memory_size) {
2067		if (!old || !old->npages)
2068			return -EINVAL;
2069
2070		if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
2071			return -EIO;
2072
2073		return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
2074	}
2075
2076	base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
2077	npages = (mem->memory_size >> PAGE_SHIFT);
2078
2079	if (!old || !old->npages) {
2080		change = KVM_MR_CREATE;
2081
2082		/*
2083		 * To simplify KVM internals, the total number of pages across
2084		 * all memslots must fit in an unsigned long.
2085		 */
2086		if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
2087			return -EINVAL;
2088	} else { /* Modify an existing slot. */
2089		/* Private memslots are immutable, they can only be deleted. */
2090		if (mem->flags & KVM_MEM_GUEST_MEMFD)
2091			return -EINVAL;
2092		if ((mem->userspace_addr != old->userspace_addr) ||
2093		    (npages != old->npages) ||
2094		    ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
2095			return -EINVAL;
2096
2097		if (base_gfn != old->base_gfn)
2098			change = KVM_MR_MOVE;
2099		else if (mem->flags != old->flags)
2100			change = KVM_MR_FLAGS_ONLY;
2101		else /* Nothing to change. */
2102			return 0;
2103	}
2104
2105	if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2106	    kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2107		return -EEXIST;
2108
2109	/* Allocate a slot that will persist in the memslot. */
2110	new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2111	if (!new)
2112		return -ENOMEM;
2113
2114	new->as_id = as_id;
2115	new->id = id;
2116	new->base_gfn = base_gfn;
2117	new->npages = npages;
2118	new->flags = mem->flags;
2119	new->userspace_addr = mem->userspace_addr;
2120	if (mem->flags & KVM_MEM_GUEST_MEMFD) {
2121		r = kvm_gmem_bind(kvm, new, mem->guest_memfd, mem->guest_memfd_offset);
2122		if (r)
2123			goto out;
2124	}
2125
2126	r = kvm_set_memslot(kvm, old, new, change);
2127	if (r)
2128		goto out_unbind;
2129
2130	return 0;
2131
2132out_unbind:
2133	if (mem->flags & KVM_MEM_GUEST_MEMFD)
2134		kvm_gmem_unbind(new);
2135out:
2136	kfree(new);
2137	return r;
2138}
2139EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2140
2141int kvm_set_memory_region(struct kvm *kvm,
2142			  const struct kvm_userspace_memory_region2 *mem)
2143{
2144	int r;
2145
2146	mutex_lock(&kvm->slots_lock);
2147	r = __kvm_set_memory_region(kvm, mem);
2148	mutex_unlock(&kvm->slots_lock);
2149	return r;
2150}
2151EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2152
2153static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2154					  struct kvm_userspace_memory_region2 *mem)
2155{
2156	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2157		return -EINVAL;
2158
2159	return kvm_set_memory_region(kvm, mem);
2160}
2161
2162#ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2163/**
2164 * kvm_get_dirty_log - get a snapshot of dirty pages
2165 * @kvm:	pointer to kvm instance
2166 * @log:	slot id and address to which we copy the log
2167 * @is_dirty:	set to '1' if any dirty pages were found
2168 * @memslot:	set to the associated memslot, always valid on success
2169 */
2170int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2171		      int *is_dirty, struct kvm_memory_slot **memslot)
2172{
2173	struct kvm_memslots *slots;
2174	int i, as_id, id;
2175	unsigned long n;
2176	unsigned long any = 0;
2177
2178	/* Dirty ring tracking may be exclusive to dirty log tracking */
2179	if (!kvm_use_dirty_bitmap(kvm))
2180		return -ENXIO;
2181
2182	*memslot = NULL;
2183	*is_dirty = 0;
2184
2185	as_id = log->slot >> 16;
2186	id = (u16)log->slot;
2187	if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS)
2188		return -EINVAL;
2189
2190	slots = __kvm_memslots(kvm, as_id);
2191	*memslot = id_to_memslot(slots, id);
2192	if (!(*memslot) || !(*memslot)->dirty_bitmap)
2193		return -ENOENT;
2194
2195	kvm_arch_sync_dirty_log(kvm, *memslot);
2196
2197	n = kvm_dirty_bitmap_bytes(*memslot);
2198
2199	for (i = 0; !any && i < n/sizeof(long); ++i)
2200		any = (*memslot)->dirty_bitmap[i];
2201
2202	if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2203		return -EFAULT;
2204
2205	if (any)
2206		*is_dirty = 1;
2207	return 0;
2208}
2209EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2210
2211#else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2212/**
2213 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2214 *	and reenable dirty page tracking for the corresponding pages.
2215 * @kvm:	pointer to kvm instance
2216 * @log:	slot id and address to which we copy the log
2217 *
2218 * We need to keep it in mind that VCPU threads can write to the bitmap
2219 * concurrently. So, to avoid losing track of dirty pages we keep the
2220 * following order:
2221 *
2222 *    1. Take a snapshot of the bit and clear it if needed.
2223 *    2. Write protect the corresponding page.
2224 *    3. Copy the snapshot to the userspace.
2225 *    4. Upon return caller flushes TLB's if needed.
2226 *
2227 * Between 2 and 4, the guest may write to the page using the remaining TLB
2228 * entry.  This is not a problem because the page is reported dirty using
2229 * the snapshot taken before and step 4 ensures that writes done after
2230 * exiting to userspace will be logged for the next call.
2231 *
2232 */
2233static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2234{
2235	struct kvm_memslots *slots;
2236	struct kvm_memory_slot *memslot;
2237	int i, as_id, id;
2238	unsigned long n;
2239	unsigned long *dirty_bitmap;
2240	unsigned long *dirty_bitmap_buffer;
2241	bool flush;
2242
2243	/* Dirty ring tracking may be exclusive to dirty log tracking */
2244	if (!kvm_use_dirty_bitmap(kvm))
2245		return -ENXIO;
2246
2247	as_id = log->slot >> 16;
2248	id = (u16)log->slot;
2249	if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS)
2250		return -EINVAL;
2251
2252	slots = __kvm_memslots(kvm, as_id);
2253	memslot = id_to_memslot(slots, id);
2254	if (!memslot || !memslot->dirty_bitmap)
2255		return -ENOENT;
2256
2257	dirty_bitmap = memslot->dirty_bitmap;
2258
2259	kvm_arch_sync_dirty_log(kvm, memslot);
2260
2261	n = kvm_dirty_bitmap_bytes(memslot);
2262	flush = false;
2263	if (kvm->manual_dirty_log_protect) {
2264		/*
2265		 * Unlike kvm_get_dirty_log, we always return false in *flush,
2266		 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
2267		 * is some code duplication between this function and
2268		 * kvm_get_dirty_log, but hopefully all architecture
2269		 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2270		 * can be eliminated.
2271		 */
2272		dirty_bitmap_buffer = dirty_bitmap;
2273	} else {
2274		dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2275		memset(dirty_bitmap_buffer, 0, n);
2276
2277		KVM_MMU_LOCK(kvm);
2278		for (i = 0; i < n / sizeof(long); i++) {
2279			unsigned long mask;
2280			gfn_t offset;
2281
2282			if (!dirty_bitmap[i])
2283				continue;
2284
2285			flush = true;
2286			mask = xchg(&dirty_bitmap[i], 0);
2287			dirty_bitmap_buffer[i] = mask;
2288
2289			offset = i * BITS_PER_LONG;
2290			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2291								offset, mask);
2292		}
2293		KVM_MMU_UNLOCK(kvm);
2294	}
2295
2296	if (flush)
2297		kvm_flush_remote_tlbs_memslot(kvm, memslot);
2298
2299	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2300		return -EFAULT;
2301	return 0;
2302}
2303
2304
2305/**
2306 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2307 * @kvm: kvm instance
2308 * @log: slot id and address to which we copy the log
2309 *
2310 * Steps 1-4 below provide general overview of dirty page logging. See
2311 * kvm_get_dirty_log_protect() function description for additional details.
2312 *
2313 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2314 * always flush the TLB (step 4) even if previous step failed  and the dirty
2315 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2316 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2317 * writes will be marked dirty for next log read.
2318 *
2319 *   1. Take a snapshot of the bit and clear it if needed.
2320 *   2. Write protect the corresponding page.
2321 *   3. Copy the snapshot to the userspace.
2322 *   4. Flush TLB's if needed.
2323 */
2324static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2325				      struct kvm_dirty_log *log)
2326{
2327	int r;
2328
2329	mutex_lock(&kvm->slots_lock);
2330
2331	r = kvm_get_dirty_log_protect(kvm, log);
2332
2333	mutex_unlock(&kvm->slots_lock);
2334	return r;
2335}
2336
2337/**
2338 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2339 *	and reenable dirty page tracking for the corresponding pages.
2340 * @kvm:	pointer to kvm instance
2341 * @log:	slot id and address from which to fetch the bitmap of dirty pages
2342 */
2343static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2344				       struct kvm_clear_dirty_log *log)
2345{
2346	struct kvm_memslots *slots;
2347	struct kvm_memory_slot *memslot;
2348	int as_id, id;
2349	gfn_t offset;
2350	unsigned long i, n;
2351	unsigned long *dirty_bitmap;
2352	unsigned long *dirty_bitmap_buffer;
2353	bool flush;
2354
2355	/* Dirty ring tracking may be exclusive to dirty log tracking */
2356	if (!kvm_use_dirty_bitmap(kvm))
2357		return -ENXIO;
2358
2359	as_id = log->slot >> 16;
2360	id = (u16)log->slot;
2361	if (as_id >= kvm_arch_nr_memslot_as_ids(kvm) || id >= KVM_USER_MEM_SLOTS)
2362		return -EINVAL;
2363
2364	if (log->first_page & 63)
2365		return -EINVAL;
2366
2367	slots = __kvm_memslots(kvm, as_id);
2368	memslot = id_to_memslot(slots, id);
2369	if (!memslot || !memslot->dirty_bitmap)
2370		return -ENOENT;
2371
2372	dirty_bitmap = memslot->dirty_bitmap;
2373
2374	n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2375
2376	if (log->first_page > memslot->npages ||
2377	    log->num_pages > memslot->npages - log->first_page ||
2378	    (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2379	    return -EINVAL;
2380
2381	kvm_arch_sync_dirty_log(kvm, memslot);
2382
2383	flush = false;
2384	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2385	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2386		return -EFAULT;
2387
2388	KVM_MMU_LOCK(kvm);
2389	for (offset = log->first_page, i = offset / BITS_PER_LONG,
2390		 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2391	     i++, offset += BITS_PER_LONG) {
2392		unsigned long mask = *dirty_bitmap_buffer++;
2393		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2394		if (!mask)
2395			continue;
2396
2397		mask &= atomic_long_fetch_andnot(mask, p);
2398
2399		/*
2400		 * mask contains the bits that really have been cleared.  This
2401		 * never includes any bits beyond the length of the memslot (if
2402		 * the length is not aligned to 64 pages), therefore it is not
2403		 * a problem if userspace sets them in log->dirty_bitmap.
2404		*/
2405		if (mask) {
2406			flush = true;
2407			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2408								offset, mask);
2409		}
2410	}
2411	KVM_MMU_UNLOCK(kvm);
2412
2413	if (flush)
2414		kvm_flush_remote_tlbs_memslot(kvm, memslot);
2415
2416	return 0;
2417}
2418
2419static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2420					struct kvm_clear_dirty_log *log)
2421{
2422	int r;
2423
2424	mutex_lock(&kvm->slots_lock);
2425
2426	r = kvm_clear_dirty_log_protect(kvm, log);
2427
2428	mutex_unlock(&kvm->slots_lock);
2429	return r;
2430}
2431#endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2432
2433#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
2434/*
2435 * Returns true if _all_ gfns in the range [@start, @end) have attributes
2436 * matching @attrs.
2437 */
2438bool kvm_range_has_memory_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2439				     unsigned long attrs)
2440{
2441	XA_STATE(xas, &kvm->mem_attr_array, start);
2442	unsigned long index;
2443	bool has_attrs;
2444	void *entry;
2445
2446	rcu_read_lock();
2447
2448	if (!attrs) {
2449		has_attrs = !xas_find(&xas, end - 1);
2450		goto out;
2451	}
2452
2453	has_attrs = true;
2454	for (index = start; index < end; index++) {
2455		do {
2456			entry = xas_next(&xas);
2457		} while (xas_retry(&xas, entry));
2458
2459		if (xas.xa_index != index || xa_to_value(entry) != attrs) {
2460			has_attrs = false;
2461			break;
2462		}
2463	}
2464
2465out:
2466	rcu_read_unlock();
2467	return has_attrs;
2468}
2469
2470static u64 kvm_supported_mem_attributes(struct kvm *kvm)
2471{
2472	if (!kvm || kvm_arch_has_private_mem(kvm))
2473		return KVM_MEMORY_ATTRIBUTE_PRIVATE;
2474
2475	return 0;
2476}
2477
2478static __always_inline void kvm_handle_gfn_range(struct kvm *kvm,
2479						 struct kvm_mmu_notifier_range *range)
2480{
2481	struct kvm_gfn_range gfn_range;
2482	struct kvm_memory_slot *slot;
2483	struct kvm_memslots *slots;
2484	struct kvm_memslot_iter iter;
2485	bool found_memslot = false;
2486	bool ret = false;
2487	int i;
2488
2489	gfn_range.arg = range->arg;
2490	gfn_range.may_block = range->may_block;
2491
2492	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
2493		slots = __kvm_memslots(kvm, i);
2494
2495		kvm_for_each_memslot_in_gfn_range(&iter, slots, range->start, range->end) {
2496			slot = iter.slot;
2497			gfn_range.slot = slot;
2498
2499			gfn_range.start = max(range->start, slot->base_gfn);
2500			gfn_range.end = min(range->end, slot->base_gfn + slot->npages);
2501			if (gfn_range.start >= gfn_range.end)
2502				continue;
2503
2504			if (!found_memslot) {
2505				found_memslot = true;
2506				KVM_MMU_LOCK(kvm);
2507				if (!IS_KVM_NULL_FN(range->on_lock))
2508					range->on_lock(kvm);
2509			}
2510
2511			ret |= range->handler(kvm, &gfn_range);
2512		}
2513	}
2514
2515	if (range->flush_on_ret && ret)
2516		kvm_flush_remote_tlbs(kvm);
2517
2518	if (found_memslot)
2519		KVM_MMU_UNLOCK(kvm);
2520}
2521
2522static bool kvm_pre_set_memory_attributes(struct kvm *kvm,
2523					  struct kvm_gfn_range *range)
2524{
2525	/*
2526	 * Unconditionally add the range to the invalidation set, regardless of
2527	 * whether or not the arch callback actually needs to zap SPTEs.  E.g.
2528	 * if KVM supports RWX attributes in the future and the attributes are
2529	 * going from R=>RW, zapping isn't strictly necessary.  Unconditionally
2530	 * adding the range allows KVM to require that MMU invalidations add at
2531	 * least one range between begin() and end(), e.g. allows KVM to detect
2532	 * bugs where the add() is missed.  Relaxing the rule *might* be safe,
2533	 * but it's not obvious that allowing new mappings while the attributes
2534	 * are in flux is desirable or worth the complexity.
2535	 */
2536	kvm_mmu_invalidate_range_add(kvm, range->start, range->end);
2537
2538	return kvm_arch_pre_set_memory_attributes(kvm, range);
2539}
2540
2541/* Set @attributes for the gfn range [@start, @end). */
2542static int kvm_vm_set_mem_attributes(struct kvm *kvm, gfn_t start, gfn_t end,
2543				     unsigned long attributes)
2544{
2545	struct kvm_mmu_notifier_range pre_set_range = {
2546		.start = start,
2547		.end = end,
2548		.handler = kvm_pre_set_memory_attributes,
2549		.on_lock = kvm_mmu_invalidate_begin,
2550		.flush_on_ret = true,
2551		.may_block = true,
2552	};
2553	struct kvm_mmu_notifier_range post_set_range = {
2554		.start = start,
2555		.end = end,
2556		.arg.attributes = attributes,
2557		.handler = kvm_arch_post_set_memory_attributes,
2558		.on_lock = kvm_mmu_invalidate_end,
2559		.may_block = true,
2560	};
2561	unsigned long i;
2562	void *entry;
2563	int r = 0;
2564
2565	entry = attributes ? xa_mk_value(attributes) : NULL;
2566
2567	mutex_lock(&kvm->slots_lock);
2568
2569	/* Nothing to do if the entire range as the desired attributes. */
2570	if (kvm_range_has_memory_attributes(kvm, start, end, attributes))
2571		goto out_unlock;
2572
2573	/*
2574	 * Reserve memory ahead of time to avoid having to deal with failures
2575	 * partway through setting the new attributes.
2576	 */
2577	for (i = start; i < end; i++) {
2578		r = xa_reserve(&kvm->mem_attr_array, i, GFP_KERNEL_ACCOUNT);
2579		if (r)
2580			goto out_unlock;
2581	}
2582
2583	kvm_handle_gfn_range(kvm, &pre_set_range);
2584
2585	for (i = start; i < end; i++) {
2586		r = xa_err(xa_store(&kvm->mem_attr_array, i, entry,
2587				    GFP_KERNEL_ACCOUNT));
2588		KVM_BUG_ON(r, kvm);
2589	}
2590
2591	kvm_handle_gfn_range(kvm, &post_set_range);
2592
2593out_unlock:
2594	mutex_unlock(&kvm->slots_lock);
2595
2596	return r;
2597}
2598static int kvm_vm_ioctl_set_mem_attributes(struct kvm *kvm,
2599					   struct kvm_memory_attributes *attrs)
2600{
2601	gfn_t start, end;
2602
2603	/* flags is currently not used. */
2604	if (attrs->flags)
2605		return -EINVAL;
2606	if (attrs->attributes & ~kvm_supported_mem_attributes(kvm))
2607		return -EINVAL;
2608	if (attrs->size == 0 || attrs->address + attrs->size < attrs->address)
2609		return -EINVAL;
2610	if (!PAGE_ALIGNED(attrs->address) || !PAGE_ALIGNED(attrs->size))
2611		return -EINVAL;
2612
2613	start = attrs->address >> PAGE_SHIFT;
2614	end = (attrs->address + attrs->size) >> PAGE_SHIFT;
2615
2616	/*
2617	 * xarray tracks data using "unsigned long", and as a result so does
2618	 * KVM.  For simplicity, supports generic attributes only on 64-bit
2619	 * architectures.
2620	 */
2621	BUILD_BUG_ON(sizeof(attrs->attributes) != sizeof(unsigned long));
2622
2623	return kvm_vm_set_mem_attributes(kvm, start, end, attrs->attributes);
2624}
2625#endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
2626
2627struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2628{
2629	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2630}
2631EXPORT_SYMBOL_GPL(gfn_to_memslot);
2632
2633struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2634{
2635	struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2636	u64 gen = slots->generation;
2637	struct kvm_memory_slot *slot;
2638
2639	/*
2640	 * This also protects against using a memslot from a different address space,
2641	 * since different address spaces have different generation numbers.
2642	 */
2643	if (unlikely(gen != vcpu->last_used_slot_gen)) {
2644		vcpu->last_used_slot = NULL;
2645		vcpu->last_used_slot_gen = gen;
2646	}
2647
2648	slot = try_get_memslot(vcpu->last_used_slot, gfn);
2649	if (slot)
2650		return slot;
2651
2652	/*
2653	 * Fall back to searching all memslots. We purposely use
2654	 * search_memslots() instead of __gfn_to_memslot() to avoid
2655	 * thrashing the VM-wide last_used_slot in kvm_memslots.
2656	 */
2657	slot = search_memslots(slots, gfn, false);
2658	if (slot) {
2659		vcpu->last_used_slot = slot;
2660		return slot;
2661	}
2662
2663	return NULL;
2664}
2665
2666bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2667{
2668	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2669
2670	return kvm_is_visible_memslot(memslot);
2671}
2672EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2673
2674bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2675{
2676	struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2677
2678	return kvm_is_visible_memslot(memslot);
2679}
2680EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2681
2682unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2683{
2684	struct vm_area_struct *vma;
2685	unsigned long addr, size;
2686
2687	size = PAGE_SIZE;
2688
2689	addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2690	if (kvm_is_error_hva(addr))
2691		return PAGE_SIZE;
2692
2693	mmap_read_lock(current->mm);
2694	vma = find_vma(current->mm, addr);
2695	if (!vma)
2696		goto out;
2697
2698	size = vma_kernel_pagesize(vma);
2699
2700out:
2701	mmap_read_unlock(current->mm);
2702
2703	return size;
2704}
2705
2706static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2707{
2708	return slot->flags & KVM_MEM_READONLY;
2709}
2710
2711static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2712				       gfn_t *nr_pages, bool write)
2713{
2714	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2715		return KVM_HVA_ERR_BAD;
2716
2717	if (memslot_is_readonly(slot) && write)
2718		return KVM_HVA_ERR_RO_BAD;
2719
2720	if (nr_pages)
2721		*nr_pages = slot->npages - (gfn - slot->base_gfn);
2722
2723	return __gfn_to_hva_memslot(slot, gfn);
2724}
2725
2726static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2727				     gfn_t *nr_pages)
2728{
2729	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2730}
2731
2732unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2733					gfn_t gfn)
2734{
2735	return gfn_to_hva_many(slot, gfn, NULL);
2736}
2737EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2738
2739unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2740{
2741	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2742}
2743EXPORT_SYMBOL_GPL(gfn_to_hva);
2744
2745unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2746{
2747	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2748}
2749EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2750
2751/*
2752 * Return the hva of a @gfn and the R/W attribute if possible.
2753 *
2754 * @slot: the kvm_memory_slot which contains @gfn
2755 * @gfn: the gfn to be translated
2756 * @writable: used to return the read/write attribute of the @slot if the hva
2757 * is valid and @writable is not NULL
2758 */
2759unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2760				      gfn_t gfn, bool *writable)
2761{
2762	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2763
2764	if (!kvm_is_error_hva(hva) && writable)
2765		*writable = !memslot_is_readonly(slot);
2766
2767	return hva;
2768}
2769
2770unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2771{
2772	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2773
2774	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2775}
2776
2777unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2778{
2779	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2780
2781	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2782}
2783
2784static inline int check_user_page_hwpoison(unsigned long addr)
2785{
2786	int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2787
2788	rc = get_user_pages(addr, 1, flags, NULL);
2789	return rc == -EHWPOISON;
2790}
2791
2792/*
2793 * The fast path to get the writable pfn which will be stored in @pfn,
2794 * true indicates success, otherwise false is returned.  It's also the
2795 * only part that runs if we can in atomic context.
2796 */
2797static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2798			    bool *writable, kvm_pfn_t *pfn)
2799{
2800	struct page *page[1];
2801
2802	/*
2803	 * Fast pin a writable pfn only if it is a write fault request
2804	 * or the caller allows to map a writable pfn for a read fault
2805	 * request.
2806	 */
2807	if (!(write_fault || writable))
2808		return false;
2809
2810	if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2811		*pfn = page_to_pfn(page[0]);
2812
2813		if (writable)
2814			*writable = true;
2815		return true;
2816	}
2817
2818	return false;
2819}
2820
2821/*
2822 * The slow path to get the pfn of the specified host virtual address,
2823 * 1 indicates success, -errno is returned if error is detected.
2824 */
2825static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2826			   bool interruptible, bool *writable, kvm_pfn_t *pfn)
2827{
2828	/*
2829	 * When a VCPU accesses a page that is not mapped into the secondary
2830	 * MMU, we lookup the page using GUP to map it, so the guest VCPU can
2831	 * make progress. We always want to honor NUMA hinting faults in that
2832	 * case, because GUP usage corresponds to memory accesses from the VCPU.
2833	 * Otherwise, we'd not trigger NUMA hinting faults once a page is
2834	 * mapped into the secondary MMU and gets accessed by a VCPU.
2835	 *
2836	 * Note that get_user_page_fast_only() and FOLL_WRITE for now
2837	 * implicitly honor NUMA hinting faults and don't need this flag.
2838	 */
2839	unsigned int flags = FOLL_HWPOISON | FOLL_HONOR_NUMA_FAULT;
2840	struct page *page;
2841	int npages;
2842
2843	might_sleep();
2844
2845	if (writable)
2846		*writable = write_fault;
2847
2848	if (write_fault)
2849		flags |= FOLL_WRITE;
2850	if (async)
2851		flags |= FOLL_NOWAIT;
2852	if (interruptible)
2853		flags |= FOLL_INTERRUPTIBLE;
2854
2855	npages = get_user_pages_unlocked(addr, 1, &page, flags);
2856	if (npages != 1)
2857		return npages;
2858
2859	/* map read fault as writable if possible */
2860	if (unlikely(!write_fault) && writable) {
2861		struct page *wpage;
2862
2863		if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2864			*writable = true;
2865			put_page(page);
2866			page = wpage;
2867		}
2868	}
2869	*pfn = page_to_pfn(page);
2870	return npages;
2871}
2872
2873static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2874{
2875	if (unlikely(!(vma->vm_flags & VM_READ)))
2876		return false;
2877
2878	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2879		return false;
2880
2881	return true;
2882}
2883
2884static int kvm_try_get_pfn(kvm_pfn_t pfn)
2885{
2886	struct page *page = kvm_pfn_to_refcounted_page(pfn);
2887
2888	if (!page)
2889		return 1;
2890
2891	return get_page_unless_zero(page);
2892}
2893
2894static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2895			       unsigned long addr, bool write_fault,
2896			       bool *writable, kvm_pfn_t *p_pfn)
2897{
2898	kvm_pfn_t pfn;
2899	pte_t *ptep;
2900	pte_t pte;
2901	spinlock_t *ptl;
2902	int r;
2903
2904	r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2905	if (r) {
2906		/*
2907		 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2908		 * not call the fault handler, so do it here.
2909		 */
2910		bool unlocked = false;
2911		r = fixup_user_fault(current->mm, addr,
2912				     (write_fault ? FAULT_FLAG_WRITE : 0),
2913				     &unlocked);
2914		if (unlocked)
2915			return -EAGAIN;
2916		if (r)
2917			return r;
2918
2919		r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2920		if (r)
2921			return r;
2922	}
2923
2924	pte = ptep_get(ptep);
2925
2926	if (write_fault && !pte_write(pte)) {
2927		pfn = KVM_PFN_ERR_RO_FAULT;
2928		goto out;
2929	}
2930
2931	if (writable)
2932		*writable = pte_write(pte);
2933	pfn = pte_pfn(pte);
2934
2935	/*
2936	 * Get a reference here because callers of *hva_to_pfn* and
2937	 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2938	 * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
2939	 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2940	 * simply do nothing for reserved pfns.
2941	 *
2942	 * Whoever called remap_pfn_range is also going to call e.g.
2943	 * unmap_mapping_range before the underlying pages are freed,
2944	 * causing a call to our MMU notifier.
2945	 *
2946	 * Certain IO or PFNMAP mappings can be backed with valid
2947	 * struct pages, but be allocated without refcounting e.g.,
2948	 * tail pages of non-compound higher order allocations, which
2949	 * would then underflow the refcount when the caller does the
2950	 * required put_page. Don't allow those pages here.
2951	 */
2952	if (!kvm_try_get_pfn(pfn))
2953		r = -EFAULT;
2954
2955out:
2956	pte_unmap_unlock(ptep, ptl);
2957	*p_pfn = pfn;
2958
2959	return r;
2960}
2961
2962/*
2963 * Pin guest page in memory and return its pfn.
2964 * @addr: host virtual address which maps memory to the guest
2965 * @atomic: whether this function can sleep
2966 * @interruptible: whether the process can be interrupted by non-fatal signals
2967 * @async: whether this function need to wait IO complete if the
2968 *         host page is not in the memory
2969 * @write_fault: whether we should get a writable host page
2970 * @writable: whether it allows to map a writable host page for !@write_fault
2971 *
2972 * The function will map a writable host page for these two cases:
2973 * 1): @write_fault = true
2974 * 2): @write_fault = false && @writable, @writable will tell the caller
2975 *     whether the mapping is writable.
2976 */
2977kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2978		     bool *async, bool write_fault, bool *writable)
2979{
2980	struct vm_area_struct *vma;
2981	kvm_pfn_t pfn;
2982	int npages, r;
2983
2984	/* we can do it either atomically or asynchronously, not both */
2985	BUG_ON(atomic && async);
2986
2987	if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2988		return pfn;
2989
2990	if (atomic)
2991		return KVM_PFN_ERR_FAULT;
2992
2993	npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
2994				 writable, &pfn);
2995	if (npages == 1)
2996		return pfn;
2997	if (npages == -EINTR)
2998		return KVM_PFN_ERR_SIGPENDING;
2999
3000	mmap_read_lock(current->mm);
3001	if (npages == -EHWPOISON ||
3002	      (!async && check_user_page_hwpoison(addr))) {
3003		pfn = KVM_PFN_ERR_HWPOISON;
3004		goto exit;
3005	}
3006
3007retry:
3008	vma = vma_lookup(current->mm, addr);
3009
3010	if (vma == NULL)
3011		pfn = KVM_PFN_ERR_FAULT;
3012	else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
3013		r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
3014		if (r == -EAGAIN)
3015			goto retry;
3016		if (r < 0)
3017			pfn = KVM_PFN_ERR_FAULT;
3018	} else {
3019		if (async && vma_is_valid(vma, write_fault))
3020			*async = true;
3021		pfn = KVM_PFN_ERR_FAULT;
3022	}
3023exit:
3024	mmap_read_unlock(current->mm);
3025	return pfn;
3026}
3027
3028kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
3029			       bool atomic, bool interruptible, bool *async,
3030			       bool write_fault, bool *writable, hva_t *hva)
3031{
3032	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
3033
3034	if (hva)
3035		*hva = addr;
3036
3037	if (addr == KVM_HVA_ERR_RO_BAD) {
3038		if (writable)
3039			*writable = false;
3040		return KVM_PFN_ERR_RO_FAULT;
3041	}
3042
3043	if (kvm_is_error_hva(addr)) {
3044		if (writable)
3045			*writable = false;
3046		return KVM_PFN_NOSLOT;
3047	}
3048
3049	/* Do not map writable pfn in the readonly memslot. */
3050	if (writable && memslot_is_readonly(slot)) {
3051		*writable = false;
3052		writable = NULL;
3053	}
3054
3055	return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
3056			  writable);
3057}
3058EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
3059
3060kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
3061		      bool *writable)
3062{
3063	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
3064				    NULL, write_fault, writable, NULL);
3065}
3066EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
3067
3068kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
3069{
3070	return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
3071				    NULL, NULL);
3072}
3073EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
3074
3075kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
3076{
3077	return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
3078				    NULL, NULL);
3079}
3080EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
3081
3082kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
3083{
3084	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
3085}
3086EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
3087
3088kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
3089{
3090	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
3091}
3092EXPORT_SYMBOL_GPL(gfn_to_pfn);
3093
3094kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
3095{
3096	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
3097}
3098EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
3099
3100int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3101			    struct page **pages, int nr_pages)
3102{
3103	unsigned long addr;
3104	gfn_t entry = 0;
3105
3106	addr = gfn_to_hva_many(slot, gfn, &entry);
3107	if (kvm_is_error_hva(addr))
3108		return -1;
3109
3110	if (entry < nr_pages)
3111		return 0;
3112
3113	return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
3114}
3115EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
3116
3117/*
3118 * Do not use this helper unless you are absolutely certain the gfn _must_ be
3119 * backed by 'struct page'.  A valid example is if the backing memslot is
3120 * controlled by KVM.  Note, if the returned page is valid, it's refcount has
3121 * been elevated by gfn_to_pfn().
3122 */
3123struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
3124{
3125	struct page *page;
3126	kvm_pfn_t pfn;
3127
3128	pfn = gfn_to_pfn(kvm, gfn);
3129
3130	if (is_error_noslot_pfn(pfn))
3131		return KVM_ERR_PTR_BAD_PAGE;
3132
3133	page = kvm_pfn_to_refcounted_page(pfn);
3134	if (!page)
3135		return KVM_ERR_PTR_BAD_PAGE;
3136
3137	return page;
3138}
3139EXPORT_SYMBOL_GPL(gfn_to_page);
3140
3141void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
3142{
3143	if (dirty)
3144		kvm_release_pfn_dirty(pfn);
3145	else
3146		kvm_release_pfn_clean(pfn);
3147}
3148
3149int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
3150{
3151	kvm_pfn_t pfn;
3152	void *hva = NULL;
3153	struct page *page = KVM_UNMAPPED_PAGE;
3154
3155	if (!map)
3156		return -EINVAL;
3157
3158	pfn = gfn_to_pfn(vcpu->kvm, gfn);
3159	if (is_error_noslot_pfn(pfn))
3160		return -EINVAL;
3161
3162	if (pfn_valid(pfn)) {
3163		page = pfn_to_page(pfn);
3164		hva = kmap(page);
3165#ifdef CONFIG_HAS_IOMEM
3166	} else {
3167		hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
3168#endif
3169	}
3170
3171	if (!hva)
3172		return -EFAULT;
3173
3174	map->page = page;
3175	map->hva = hva;
3176	map->pfn = pfn;
3177	map->gfn = gfn;
3178
3179	return 0;
3180}
3181EXPORT_SYMBOL_GPL(kvm_vcpu_map);
3182
3183void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
3184{
3185	if (!map)
3186		return;
3187
3188	if (!map->hva)
3189		return;
3190
3191	if (map->page != KVM_UNMAPPED_PAGE)
3192		kunmap(map->page);
3193#ifdef CONFIG_HAS_IOMEM
3194	else
3195		memunmap(map->hva);
3196#endif
3197
3198	if (dirty)
3199		kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
3200
3201	kvm_release_pfn(map->pfn, dirty);
3202
3203	map->hva = NULL;
3204	map->page = NULL;
3205}
3206EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
3207
3208static bool kvm_is_ad_tracked_page(struct page *page)
3209{
3210	/*
3211	 * Per page-flags.h, pages tagged PG_reserved "should in general not be
3212	 * touched (e.g. set dirty) except by its owner".
3213	 */
3214	return !PageReserved(page);
3215}
3216
3217static void kvm_set_page_dirty(struct page *page)
3218{
3219	if (kvm_is_ad_tracked_page(page))
3220		SetPageDirty(page);
3221}
3222
3223static void kvm_set_page_accessed(struct page *page)
3224{
3225	if (kvm_is_ad_tracked_page(page))
3226		mark_page_accessed(page);
3227}
3228
3229void kvm_release_page_clean(struct page *page)
3230{
3231	WARN_ON(is_error_page(page));
3232
3233	kvm_set_page_accessed(page);
3234	put_page(page);
3235}
3236EXPORT_SYMBOL_GPL(kvm_release_page_clean);
3237
3238void kvm_release_pfn_clean(kvm_pfn_t pfn)
3239{
3240	struct page *page;
3241
3242	if (is_error_noslot_pfn(pfn))
3243		return;
3244
3245	page = kvm_pfn_to_refcounted_page(pfn);
3246	if (!page)
3247		return;
3248
3249	kvm_release_page_clean(page);
3250}
3251EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
3252
3253void kvm_release_page_dirty(struct page *page)
3254{
3255	WARN_ON(is_error_page(page));
3256
3257	kvm_set_page_dirty(page);
3258	kvm_release_page_clean(page);
3259}
3260EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
3261
3262void kvm_release_pfn_dirty(kvm_pfn_t pfn)
3263{
3264	struct page *page;
3265
3266	if (is_error_noslot_pfn(pfn))
3267		return;
3268
3269	page = kvm_pfn_to_refcounted_page(pfn);
3270	if (!page)
3271		return;
3272
3273	kvm_release_page_dirty(page);
3274}
3275EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
3276
3277/*
3278 * Note, checking for an error/noslot pfn is the caller's responsibility when
3279 * directly marking a page dirty/accessed.  Unlike the "release" helpers, the
3280 * "set" helpers are not to be used when the pfn might point at garbage.
3281 */
3282void kvm_set_pfn_dirty(kvm_pfn_t pfn)
3283{
3284	if (WARN_ON(is_error_noslot_pfn(pfn)))
3285		return;
3286
3287	if (pfn_valid(pfn))
3288		kvm_set_page_dirty(pfn_to_page(pfn));
3289}
3290EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
3291
3292void kvm_set_pfn_accessed(kvm_pfn_t pfn)
3293{
3294	if (WARN_ON(is_error_noslot_pfn(pfn)))
3295		return;
3296
3297	if (pfn_valid(pfn))
3298		kvm_set_page_accessed(pfn_to_page(pfn));
3299}
3300EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
3301
3302static int next_segment(unsigned long len, int offset)
3303{
3304	if (len > PAGE_SIZE - offset)
3305		return PAGE_SIZE - offset;
3306	else
3307		return len;
3308}
3309
3310static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
3311				 void *data, int offset, int len)
3312{
3313	int r;
3314	unsigned long addr;
3315
3316	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3317	if (kvm_is_error_hva(addr))
3318		return -EFAULT;
3319	r = __copy_from_user(data, (void __user *)addr + offset, len);
3320	if (r)
3321		return -EFAULT;
3322	return 0;
3323}
3324
3325int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3326			int len)
3327{
3328	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3329
3330	return __kvm_read_guest_page(slot, gfn, data, offset, len);
3331}
3332EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3333
3334int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3335			     int offset, int len)
3336{
3337	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3338
3339	return __kvm_read_guest_page(slot, gfn, data, offset, len);
3340}
3341EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3342
3343int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3344{
3345	gfn_t gfn = gpa >> PAGE_SHIFT;
3346	int seg;
3347	int offset = offset_in_page(gpa);
3348	int ret;
3349
3350	while ((seg = next_segment(len, offset)) != 0) {
3351		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3352		if (ret < 0)
3353			return ret;
3354		offset = 0;
3355		len -= seg;
3356		data += seg;
3357		++gfn;
3358	}
3359	return 0;
3360}
3361EXPORT_SYMBOL_GPL(kvm_read_guest);
3362
3363int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3364{
3365	gfn_t gfn = gpa >> PAGE_SHIFT;
3366	int seg;
3367	int offset = offset_in_page(gpa);
3368	int ret;
3369
3370	while ((seg = next_segment(len, offset)) != 0) {
3371		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3372		if (ret < 0)
3373			return ret;
3374		offset = 0;
3375		len -= seg;
3376		data += seg;
3377		++gfn;
3378	}
3379	return 0;
3380}
3381EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3382
3383static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3384			           void *data, int offset, unsigned long len)
3385{
3386	int r;
3387	unsigned long addr;
3388
3389	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3390	if (kvm_is_error_hva(addr))
3391		return -EFAULT;
3392	pagefault_disable();
3393	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3394	pagefault_enable();
3395	if (r)
3396		return -EFAULT;
3397	return 0;
3398}
3399
3400int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3401			       void *data, unsigned long len)
3402{
3403	gfn_t gfn = gpa >> PAGE_SHIFT;
3404	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3405	int offset = offset_in_page(gpa);
3406
3407	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3408}
3409EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3410
3411static int __kvm_write_guest_page(struct kvm *kvm,
3412				  struct kvm_memory_slot *memslot, gfn_t gfn,
3413			          const void *data, int offset, int len)
3414{
3415	int r;
3416	unsigned long addr;
3417
3418	addr = gfn_to_hva_memslot(memslot, gfn);
3419	if (kvm_is_error_hva(addr))
3420		return -EFAULT;
3421	r = __copy_to_user((void __user *)addr + offset, data, len);
3422	if (r)
3423		return -EFAULT;
3424	mark_page_dirty_in_slot(kvm, memslot, gfn);
3425	return 0;
3426}
3427
3428int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3429			 const void *data, int offset, int len)
3430{
3431	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3432
3433	return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3434}
3435EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3436
3437int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3438			      const void *data, int offset, int len)
3439{
3440	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3441
3442	return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3443}
3444EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3445
3446int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3447		    unsigned long len)
3448{
3449	gfn_t gfn = gpa >> PAGE_SHIFT;
3450	int seg;
3451	int offset = offset_in_page(gpa);
3452	int ret;
3453
3454	while ((seg = next_segment(len, offset)) != 0) {
3455		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3456		if (ret < 0)
3457			return ret;
3458		offset = 0;
3459		len -= seg;
3460		data += seg;
3461		++gfn;
3462	}
3463	return 0;
3464}
3465EXPORT_SYMBOL_GPL(kvm_write_guest);
3466
3467int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3468		         unsigned long len)
3469{
3470	gfn_t gfn = gpa >> PAGE_SHIFT;
3471	int seg;
3472	int offset = offset_in_page(gpa);
3473	int ret;
3474
3475	while ((seg = next_segment(len, offset)) != 0) {
3476		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3477		if (ret < 0)
3478			return ret;
3479		offset = 0;
3480		len -= seg;
3481		data += seg;
3482		++gfn;
3483	}
3484	return 0;
3485}
3486EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3487
3488static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3489				       struct gfn_to_hva_cache *ghc,
3490				       gpa_t gpa, unsigned long len)
3491{
3492	int offset = offset_in_page(gpa);
3493	gfn_t start_gfn = gpa >> PAGE_SHIFT;
3494	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3495	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3496	gfn_t nr_pages_avail;
3497
3498	/* Update ghc->generation before performing any error checks. */
3499	ghc->generation = slots->generation;
3500
3501	if (start_gfn > end_gfn) {
3502		ghc->hva = KVM_HVA_ERR_BAD;
3503		return -EINVAL;
3504	}
3505
3506	/*
3507	 * If the requested region crosses two memslots, we still
3508	 * verify that the entire region is valid here.
3509	 */
3510	for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3511		ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3512		ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3513					   &nr_pages_avail);
3514		if (kvm_is_error_hva(ghc->hva))
3515			return -EFAULT;
3516	}
3517
3518	/* Use the slow path for cross page reads and writes. */
3519	if (nr_pages_needed == 1)
3520		ghc->hva += offset;
3521	else
3522		ghc->memslot = NULL;
3523
3524	ghc->gpa = gpa;
3525	ghc->len = len;
3526	return 0;
3527}
3528
3529int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3530			      gpa_t gpa, unsigned long len)
3531{
3532	struct kvm_memslots *slots = kvm_memslots(kvm);
3533	return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3534}
3535EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3536
3537int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3538				  void *data, unsigned int offset,
3539				  unsigned long len)
3540{
3541	struct kvm_memslots *slots = kvm_memslots(kvm);
3542	int r;
3543	gpa_t gpa = ghc->gpa + offset;
3544
3545	if (WARN_ON_ONCE(len + offset > ghc->len))
3546		return -EINVAL;
3547
3548	if (slots->generation != ghc->generation) {
3549		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3550			return -EFAULT;
3551	}
3552
3553	if (kvm_is_error_hva(ghc->hva))
3554		return -EFAULT;
3555
3556	if (unlikely(!ghc->memslot))
3557		return kvm_write_guest(kvm, gpa, data, len);
3558
3559	r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3560	if (r)
3561		return -EFAULT;
3562	mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3563
3564	return 0;
3565}
3566EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3567
3568int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3569			   void *data, unsigned long len)
3570{
3571	return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3572}
3573EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3574
3575int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3576				 void *data, unsigned int offset,
3577				 unsigned long len)
3578{
3579	struct kvm_memslots *slots = kvm_memslots(kvm);
3580	int r;
3581	gpa_t gpa = ghc->gpa + offset;
3582
3583	if (WARN_ON_ONCE(len + offset > ghc->len))
3584		return -EINVAL;
3585
3586	if (slots->generation != ghc->generation) {
3587		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3588			return -EFAULT;
3589	}
3590
3591	if (kvm_is_error_hva(ghc->hva))
3592		return -EFAULT;
3593
3594	if (unlikely(!ghc->memslot))
3595		return kvm_read_guest(kvm, gpa, data, len);
3596
3597	r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3598	if (r)
3599		return -EFAULT;
3600
3601	return 0;
3602}
3603EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3604
3605int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3606			  void *data, unsigned long len)
3607{
3608	return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3609}
3610EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3611
3612int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3613{
3614	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3615	gfn_t gfn = gpa >> PAGE_SHIFT;
3616	int seg;
3617	int offset = offset_in_page(gpa);
3618	int ret;
3619
3620	while ((seg = next_segment(len, offset)) != 0) {
3621		ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3622		if (ret < 0)
3623			return ret;
3624		offset = 0;
3625		len -= seg;
3626		++gfn;
3627	}
3628	return 0;
3629}
3630EXPORT_SYMBOL_GPL(kvm_clear_guest);
3631
3632void mark_page_dirty_in_slot(struct kvm *kvm,
3633			     const struct kvm_memory_slot *memslot,
3634		 	     gfn_t gfn)
3635{
3636	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3637
3638#ifdef CONFIG_HAVE_KVM_DIRTY_RING
3639	if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3640		return;
3641
3642	WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3643#endif
3644
3645	if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3646		unsigned long rel_gfn = gfn - memslot->base_gfn;
3647		u32 slot = (memslot->as_id << 16) | memslot->id;
3648
3649		if (kvm->dirty_ring_size && vcpu)
3650			kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3651		else if (memslot->dirty_bitmap)
3652			set_bit_le(rel_gfn, memslot->dirty_bitmap);
3653	}
3654}
3655EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3656
3657void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3658{
3659	struct kvm_memory_slot *memslot;
3660
3661	memslot = gfn_to_memslot(kvm, gfn);
3662	mark_page_dirty_in_slot(kvm, memslot, gfn);
3663}
3664EXPORT_SYMBOL_GPL(mark_page_dirty);
3665
3666void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3667{
3668	struct kvm_memory_slot *memslot;
3669
3670	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3671	mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3672}
3673EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3674
3675void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3676{
3677	if (!vcpu->sigset_active)
3678		return;
3679
3680	/*
3681	 * This does a lockless modification of ->real_blocked, which is fine
3682	 * because, only current can change ->real_blocked and all readers of
3683	 * ->real_blocked don't care as long ->real_blocked is always a subset
3684	 * of ->blocked.
3685	 */
3686	sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
3687}
3688
3689void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3690{
3691	if (!vcpu->sigset_active)
3692		return;
3693
3694	sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
3695	sigemptyset(&current->real_blocked);
3696}
3697
3698static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3699{
3700	unsigned int old, val, grow, grow_start;
3701
3702	old = val = vcpu->halt_poll_ns;
3703	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3704	grow = READ_ONCE(halt_poll_ns_grow);
3705	if (!grow)
3706		goto out;
3707
3708	val *= grow;
3709	if (val < grow_start)
3710		val = grow_start;
3711
3712	vcpu->halt_poll_ns = val;
3713out:
3714	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3715}
3716
3717static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3718{
3719	unsigned int old, val, shrink, grow_start;
3720
3721	old = val = vcpu->halt_poll_ns;
3722	shrink = READ_ONCE(halt_poll_ns_shrink);
3723	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3724	if (shrink == 0)
3725		val = 0;
3726	else
3727		val /= shrink;
3728
3729	if (val < grow_start)
3730		val = 0;
3731
3732	vcpu->halt_poll_ns = val;
3733	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3734}
3735
3736static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3737{
3738	int ret = -EINTR;
3739	int idx = srcu_read_lock(&vcpu->kvm->srcu);
3740
3741	if (kvm_arch_vcpu_runnable(vcpu))
3742		goto out;
3743	if (kvm_cpu_has_pending_timer(vcpu))
3744		goto out;
3745	if (signal_pending(current))
3746		goto out;
3747	if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3748		goto out;
3749
3750	ret = 0;
3751out:
3752	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3753	return ret;
3754}
3755
3756/*
3757 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3758 * pending.  This is mostly used when halting a vCPU, but may also be used
3759 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3760 */
3761bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3762{
3763	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3764	bool waited = false;
3765
3766	vcpu->stat.generic.blocking = 1;
3767
3768	preempt_disable();
3769	kvm_arch_vcpu_blocking(vcpu);
3770	prepare_to_rcuwait(wait);
3771	preempt_enable();
3772
3773	for (;;) {
3774		set_current_state(TASK_INTERRUPTIBLE);
3775
3776		if (kvm_vcpu_check_block(vcpu) < 0)
3777			break;
3778
3779		waited = true;
3780		schedule();
3781	}
3782
3783	preempt_disable();
3784	finish_rcuwait(wait);
3785	kvm_arch_vcpu_unblocking(vcpu);
3786	preempt_enable();
3787
3788	vcpu->stat.generic.blocking = 0;
3789
3790	return waited;
3791}
3792
3793static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3794					  ktime_t end, bool success)
3795{
3796	struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3797	u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3798
3799	++vcpu->stat.generic.halt_attempted_poll;
3800
3801	if (success) {
3802		++vcpu->stat.generic.halt_successful_poll;
3803
3804		if (!vcpu_valid_wakeup(vcpu))
3805			++vcpu->stat.generic.halt_poll_invalid;
3806
3807		stats->halt_poll_success_ns += poll_ns;
3808		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3809	} else {
3810		stats->halt_poll_fail_ns += poll_ns;
3811		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3812	}
3813}
3814
3815static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3816{
3817	struct kvm *kvm = vcpu->kvm;
3818
3819	if (kvm->override_halt_poll_ns) {
3820		/*
3821		 * Ensure kvm->max_halt_poll_ns is not read before
3822		 * kvm->override_halt_poll_ns.
3823		 *
3824		 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3825		 */
3826		smp_rmb();
3827		return READ_ONCE(kvm->max_halt_poll_ns);
3828	}
3829
3830	return READ_ONCE(halt_poll_ns);
3831}
3832
3833/*
3834 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc...  If halt
3835 * polling is enabled, busy wait for a short time before blocking to avoid the
3836 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3837 * is halted.
3838 */
3839void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3840{
3841	unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3842	bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3843	ktime_t start, cur, poll_end;
3844	bool waited = false;
3845	bool do_halt_poll;
3846	u64 halt_ns;
3847
3848	if (vcpu->halt_poll_ns > max_halt_poll_ns)
3849		vcpu->halt_poll_ns = max_halt_poll_ns;
3850
3851	do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3852
3853	start = cur = poll_end = ktime_get();
3854	if (do_halt_poll) {
3855		ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3856
3857		do {
3858			if (kvm_vcpu_check_block(vcpu) < 0)
3859				goto out;
3860			cpu_relax();
3861			poll_end = cur = ktime_get();
3862		} while (kvm_vcpu_can_poll(cur, stop));
3863	}
3864
3865	waited = kvm_vcpu_block(vcpu);
3866
3867	cur = ktime_get();
3868	if (waited) {
3869		vcpu->stat.generic.halt_wait_ns +=
3870			ktime_to_ns(cur) - ktime_to_ns(poll_end);
3871		KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3872				ktime_to_ns(cur) - ktime_to_ns(poll_end));
3873	}
3874out:
3875	/* The total time the vCPU was "halted", including polling time. */
3876	halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3877
3878	/*
3879	 * Note, halt-polling is considered successful so long as the vCPU was
3880	 * never actually scheduled out, i.e. even if the wake event arrived
3881	 * after of the halt-polling loop itself, but before the full wait.
3882	 */
3883	if (do_halt_poll)
3884		update_halt_poll_stats(vcpu, start, poll_end, !waited);
3885
3886	if (halt_poll_allowed) {
3887		/* Recompute the max halt poll time in case it changed. */
3888		max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3889
3890		if (!vcpu_valid_wakeup(vcpu)) {
3891			shrink_halt_poll_ns(vcpu);
3892		} else if (max_halt_poll_ns) {
3893			if (halt_ns <= vcpu->halt_poll_ns)
3894				;
3895			/* we had a long block, shrink polling */
3896			else if (vcpu->halt_poll_ns &&
3897				 halt_ns > max_halt_poll_ns)
3898				shrink_halt_poll_ns(vcpu);
3899			/* we had a short halt and our poll time is too small */
3900			else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3901				 halt_ns < max_halt_poll_ns)
3902				grow_halt_poll_ns(vcpu);
3903		} else {
3904			vcpu->halt_poll_ns = 0;
3905		}
3906	}
3907
3908	trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3909}
3910EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3911
3912bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3913{
3914	if (__kvm_vcpu_wake_up(vcpu)) {
3915		WRITE_ONCE(vcpu->ready, true);
3916		++vcpu->stat.generic.halt_wakeup;
3917		return true;
3918	}
3919
3920	return false;
3921}
3922EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3923
3924#ifndef CONFIG_S390
3925/*
3926 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3927 */
3928void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3929{
3930	int me, cpu;
3931
3932	if (kvm_vcpu_wake_up(vcpu))
3933		return;
3934
3935	me = get_cpu();
3936	/*
3937	 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3938	 * to EXITING_GUEST_MODE.  Therefore the moderately expensive "should
3939	 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3940	 * within the vCPU thread itself.
3941	 */
3942	if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3943		if (vcpu->mode == IN_GUEST_MODE)
3944			WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3945		goto out;
3946	}
3947
3948	/*
3949	 * Note, the vCPU could get migrated to a different pCPU at any point
3950	 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3951	 * IPI to the previous pCPU.  But, that's ok because the purpose of the
3952	 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3953	 * vCPU also requires it to leave IN_GUEST_MODE.
3954	 */
3955	if (kvm_arch_vcpu_should_kick(vcpu)) {
3956		cpu = READ_ONCE(vcpu->cpu);
3957		if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3958			smp_send_reschedule(cpu);
3959	}
3960out:
3961	put_cpu();
3962}
3963EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3964#endif /* !CONFIG_S390 */
3965
3966int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3967{
3968	struct pid *pid;
3969	struct task_struct *task = NULL;
3970	int ret = 0;
3971
3972	rcu_read_lock();
3973	pid = rcu_dereference(target->pid);
3974	if (pid)
3975		task = get_pid_task(pid, PIDTYPE_PID);
3976	rcu_read_unlock();
3977	if (!task)
3978		return ret;
3979	ret = yield_to(task, 1);
3980	put_task_struct(task);
3981
3982	return ret;
3983}
3984EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3985
3986/*
3987 * Helper that checks whether a VCPU is eligible for directed yield.
3988 * Most eligible candidate to yield is decided by following heuristics:
3989 *
3990 *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3991 *  (preempted lock holder), indicated by @in_spin_loop.
3992 *  Set at the beginning and cleared at the end of interception/PLE handler.
3993 *
3994 *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3995 *  chance last time (mostly it has become eligible now since we have probably
3996 *  yielded to lockholder in last iteration. This is done by toggling
3997 *  @dy_eligible each time a VCPU checked for eligibility.)
3998 *
3999 *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
4000 *  to preempted lock-holder could result in wrong VCPU selection and CPU
4001 *  burning. Giving priority for a potential lock-holder increases lock
4002 *  progress.
4003 *
4004 *  Since algorithm is based on heuristics, accessing another VCPU data without
4005 *  locking does not harm. It may result in trying to yield to  same VCPU, fail
4006 *  and continue with next VCPU and so on.
4007 */
4008static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
4009{
4010#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
4011	bool eligible;
4012
4013	eligible = !vcpu->spin_loop.in_spin_loop ||
4014		    vcpu->spin_loop.dy_eligible;
4015
4016	if (vcpu->spin_loop.in_spin_loop)
4017		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
4018
4019	return eligible;
4020#else
4021	return true;
4022#endif
4023}
4024
4025/*
4026 * Unlike kvm_arch_vcpu_runnable, this function is called outside
4027 * a vcpu_load/vcpu_put pair.  However, for most architectures
4028 * kvm_arch_vcpu_runnable does not require vcpu_load.
4029 */
4030bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
4031{
4032	return kvm_arch_vcpu_runnable(vcpu);
4033}
4034
4035static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
4036{
4037	if (kvm_arch_dy_runnable(vcpu))
4038		return true;
4039
4040#ifdef CONFIG_KVM_ASYNC_PF
4041	if (!list_empty_careful(&vcpu->async_pf.done))
4042		return true;
4043#endif
4044
4045	return false;
4046}
4047
4048/*
4049 * By default, simply query the target vCPU's current mode when checking if a
4050 * vCPU was preempted in kernel mode.  All architectures except x86 (or more
4051 * specifical, except VMX) allow querying whether or not a vCPU is in kernel
4052 * mode even if the vCPU is NOT loaded, i.e. using kvm_arch_vcpu_in_kernel()
4053 * directly for cross-vCPU checks is functionally correct and accurate.
4054 */
4055bool __weak kvm_arch_vcpu_preempted_in_kernel(struct kvm_vcpu *vcpu)
4056{
4057	return kvm_arch_vcpu_in_kernel(vcpu);
4058}
4059
4060bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
4061{
4062	return false;
4063}
4064
4065void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
4066{
4067	struct kvm *kvm = me->kvm;
4068	struct kvm_vcpu *vcpu;
4069	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
4070	unsigned long i;
4071	int yielded = 0;
4072	int try = 3;
4073	int pass;
4074
4075	kvm_vcpu_set_in_spin_loop(me, true);
4076	/*
4077	 * We boost the priority of a VCPU that is runnable but not
4078	 * currently running, because it got preempted by something
4079	 * else and called schedule in __vcpu_run.  Hopefully that
4080	 * VCPU is holding the lock that we need and will release it.
4081	 * We approximate round-robin by starting at the last boosted VCPU.
4082	 */
4083	for (pass = 0; pass < 2 && !yielded && try; pass++) {
4084		kvm_for_each_vcpu(i, vcpu, kvm) {
4085			if (!pass && i <= last_boosted_vcpu) {
4086				i = last_boosted_vcpu;
4087				continue;
4088			} else if (pass && i > last_boosted_vcpu)
4089				break;
4090			if (!READ_ONCE(vcpu->ready))
4091				continue;
4092			if (vcpu == me)
4093				continue;
4094			if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
4095				continue;
4096
4097			/*
4098			 * Treat the target vCPU as being in-kernel if it has a
4099			 * pending interrupt, as the vCPU trying to yield may
4100			 * be spinning waiting on IPI delivery, i.e. the target
4101			 * vCPU is in-kernel for the purposes of directed yield.
4102			 */
4103			if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
4104			    !kvm_arch_dy_has_pending_interrupt(vcpu) &&
4105			    !kvm_arch_vcpu_preempted_in_kernel(vcpu))
4106				continue;
4107			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
4108				continue;
4109
4110			yielded = kvm_vcpu_yield_to(vcpu);
4111			if (yielded > 0) {
4112				kvm->last_boosted_vcpu = i;
4113				break;
4114			} else if (yielded < 0) {
4115				try--;
4116				if (!try)
4117					break;
4118			}
4119		}
4120	}
4121	kvm_vcpu_set_in_spin_loop(me, false);
4122
4123	/* Ensure vcpu is not eligible during next spinloop */
4124	kvm_vcpu_set_dy_eligible(me, false);
4125}
4126EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
4127
4128static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
4129{
4130#ifdef CONFIG_HAVE_KVM_DIRTY_RING
4131	return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
4132	    (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
4133	     kvm->dirty_ring_size / PAGE_SIZE);
4134#else
4135	return false;
4136#endif
4137}
4138
4139static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
4140{
4141	struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
4142	struct page *page;
4143
4144	if (vmf->pgoff == 0)
4145		page = virt_to_page(vcpu->run);
4146#ifdef CONFIG_X86
4147	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
4148		page = virt_to_page(vcpu->arch.pio_data);
4149#endif
4150#ifdef CONFIG_KVM_MMIO
4151	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
4152		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
4153#endif
4154	else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
4155		page = kvm_dirty_ring_get_page(
4156		    &vcpu->dirty_ring,
4157		    vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
4158	else
4159		return kvm_arch_vcpu_fault(vcpu, vmf);
4160	get_page(page);
4161	vmf->page = page;
4162	return 0;
4163}
4164
4165static const struct vm_operations_struct kvm_vcpu_vm_ops = {
4166	.fault = kvm_vcpu_fault,
4167};
4168
4169static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
4170{
4171	struct kvm_vcpu *vcpu = file->private_data;
4172	unsigned long pages = vma_pages(vma);
4173
4174	if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
4175	     kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
4176	    ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
4177		return -EINVAL;
4178
4179	vma->vm_ops = &kvm_vcpu_vm_ops;
4180	return 0;
4181}
4182
4183static int kvm_vcpu_release(struct inode *inode, struct file *filp)
4184{
4185	struct kvm_vcpu *vcpu = filp->private_data;
4186
4187	kvm_put_kvm(vcpu->kvm);
4188	return 0;
4189}
4190
4191static struct file_operations kvm_vcpu_fops = {
4192	.release        = kvm_vcpu_release,
4193	.unlocked_ioctl = kvm_vcpu_ioctl,
4194	.mmap           = kvm_vcpu_mmap,
4195	.llseek		= noop_llseek,
4196	KVM_COMPAT(kvm_vcpu_compat_ioctl),
4197};
4198
4199/*
4200 * Allocates an inode for the vcpu.
4201 */
4202static int create_vcpu_fd(struct kvm_vcpu *vcpu)
4203{
4204	char name[8 + 1 + ITOA_MAX_LEN + 1];
4205
4206	snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
4207	return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
4208}
4209
4210#ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
4211static int vcpu_get_pid(void *data, u64 *val)
4212{
4213	struct kvm_vcpu *vcpu = data;
4214
4215	rcu_read_lock();
4216	*val = pid_nr(rcu_dereference(vcpu->pid));
4217	rcu_read_unlock();
4218	return 0;
4219}
4220
4221DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
4222
4223static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
4224{
4225	struct dentry *debugfs_dentry;
4226	char dir_name[ITOA_MAX_LEN * 2];
4227
4228	if (!debugfs_initialized())
4229		return;
4230
4231	snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
4232	debugfs_dentry = debugfs_create_dir(dir_name,
4233					    vcpu->kvm->debugfs_dentry);
4234	debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
4235			    &vcpu_get_pid_fops);
4236
4237	kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
4238}
4239#endif
4240
4241/*
4242 * Creates some virtual cpus.  Good luck creating more than one.
4243 */
4244static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
4245{
4246	int r;
4247	struct kvm_vcpu *vcpu;
4248	struct page *page;
4249
4250	if (id >= KVM_MAX_VCPU_IDS)
4251		return -EINVAL;
4252
4253	mutex_lock(&kvm->lock);
4254	if (kvm->created_vcpus >= kvm->max_vcpus) {
4255		mutex_unlock(&kvm->lock);
4256		return -EINVAL;
4257	}
4258
4259	r = kvm_arch_vcpu_precreate(kvm, id);
4260	if (r) {
4261		mutex_unlock(&kvm->lock);
4262		return r;
4263	}
4264
4265	kvm->created_vcpus++;
4266	mutex_unlock(&kvm->lock);
4267
4268	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
4269	if (!vcpu) {
4270		r = -ENOMEM;
4271		goto vcpu_decrement;
4272	}
4273
4274	BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
4275	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
4276	if (!page) {
4277		r = -ENOMEM;
4278		goto vcpu_free;
4279	}
4280	vcpu->run = page_address(page);
4281
4282	kvm_vcpu_init(vcpu, kvm, id);
4283
4284	r = kvm_arch_vcpu_create(vcpu);
4285	if (r)
4286		goto vcpu_free_run_page;
4287
4288	if (kvm->dirty_ring_size) {
4289		r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
4290					 id, kvm->dirty_ring_size);
4291		if (r)
4292			goto arch_vcpu_destroy;
4293	}
4294
4295	mutex_lock(&kvm->lock);
4296
4297#ifdef CONFIG_LOCKDEP
4298	/* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
4299	mutex_lock(&vcpu->mutex);
4300	mutex_unlock(&vcpu->mutex);
4301#endif
4302
4303	if (kvm_get_vcpu_by_id(kvm, id)) {
4304		r = -EEXIST;
4305		goto unlock_vcpu_destroy;
4306	}
4307
4308	vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
4309	r = xa_reserve(&kvm->vcpu_array, vcpu->vcpu_idx, GFP_KERNEL_ACCOUNT);
4310	if (r)
4311		goto unlock_vcpu_destroy;
4312
4313	/* Now it's all set up, let userspace reach it */
4314	kvm_get_kvm(kvm);
4315	r = create_vcpu_fd(vcpu);
4316	if (r < 0)
4317		goto kvm_put_xa_release;
4318
4319	if (KVM_BUG_ON(xa_store(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, 0), kvm)) {
4320		r = -EINVAL;
4321		goto kvm_put_xa_release;
4322	}
4323
4324	/*
4325	 * Pairs with smp_rmb() in kvm_get_vcpu.  Store the vcpu
4326	 * pointer before kvm->online_vcpu's incremented value.
4327	 */
4328	smp_wmb();
4329	atomic_inc(&kvm->online_vcpus);
4330
4331	mutex_unlock(&kvm->lock);
4332	kvm_arch_vcpu_postcreate(vcpu);
4333	kvm_create_vcpu_debugfs(vcpu);
4334	return r;
4335
4336kvm_put_xa_release:
4337	kvm_put_kvm_no_destroy(kvm);
4338	xa_release(&kvm->vcpu_array, vcpu->vcpu_idx);
4339unlock_vcpu_destroy:
4340	mutex_unlock(&kvm->lock);
4341	kvm_dirty_ring_free(&vcpu->dirty_ring);
4342arch_vcpu_destroy:
4343	kvm_arch_vcpu_destroy(vcpu);
4344vcpu_free_run_page:
4345	free_page((unsigned long)vcpu->run);
4346vcpu_free:
4347	kmem_cache_free(kvm_vcpu_cache, vcpu);
4348vcpu_decrement:
4349	mutex_lock(&kvm->lock);
4350	kvm->created_vcpus--;
4351	mutex_unlock(&kvm->lock);
4352	return r;
4353}
4354
4355static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4356{
4357	if (sigset) {
4358		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4359		vcpu->sigset_active = 1;
4360		vcpu->sigset = *sigset;
4361	} else
4362		vcpu->sigset_active = 0;
4363	return 0;
4364}
4365
4366static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4367			      size_t size, loff_t *offset)
4368{
4369	struct kvm_vcpu *vcpu = file->private_data;
4370
4371	return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4372			&kvm_vcpu_stats_desc[0], &vcpu->stat,
4373			sizeof(vcpu->stat), user_buffer, size, offset);
4374}
4375
4376static int kvm_vcpu_stats_release(struct inode *inode, struct file *file)
4377{
4378	struct kvm_vcpu *vcpu = file->private_data;
4379
4380	kvm_put_kvm(vcpu->kvm);
4381	return 0;
4382}
4383
4384static const struct file_operations kvm_vcpu_stats_fops = {
4385	.owner = THIS_MODULE,
4386	.read = kvm_vcpu_stats_read,
4387	.release = kvm_vcpu_stats_release,
4388	.llseek = noop_llseek,
4389};
4390
4391static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4392{
4393	int fd;
4394	struct file *file;
4395	char name[15 + ITOA_MAX_LEN + 1];
4396
4397	snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4398
4399	fd = get_unused_fd_flags(O_CLOEXEC);
4400	if (fd < 0)
4401		return fd;
4402
4403	file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4404	if (IS_ERR(file)) {
4405		put_unused_fd(fd);
4406		return PTR_ERR(file);
4407	}
4408
4409	kvm_get_kvm(vcpu->kvm);
4410
4411	file->f_mode |= FMODE_PREAD;
4412	fd_install(fd, file);
4413
4414	return fd;
4415}
4416
4417static long kvm_vcpu_ioctl(struct file *filp,
4418			   unsigned int ioctl, unsigned long arg)
4419{
4420	struct kvm_vcpu *vcpu = filp->private_data;
4421	void __user *argp = (void __user *)arg;
4422	int r;
4423	struct kvm_fpu *fpu = NULL;
4424	struct kvm_sregs *kvm_sregs = NULL;
4425
4426	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4427		return -EIO;
4428
4429	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4430		return -EINVAL;
4431
4432	/*
4433	 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4434	 * execution; mutex_lock() would break them.
4435	 */
4436	r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4437	if (r != -ENOIOCTLCMD)
4438		return r;
4439
4440	if (mutex_lock_killable(&vcpu->mutex))
4441		return -EINTR;
4442	switch (ioctl) {
4443	case KVM_RUN: {
4444		struct pid *oldpid;
4445		r = -EINVAL;
4446		if (arg)
4447			goto out;
4448		oldpid = rcu_access_pointer(vcpu->pid);
4449		if (unlikely(oldpid != task_pid(current))) {
4450			/* The thread running this VCPU changed. */
4451			struct pid *newpid;
4452
4453			r = kvm_arch_vcpu_run_pid_change(vcpu);
4454			if (r)
4455				break;
4456
4457			newpid = get_task_pid(current, PIDTYPE_PID);
4458			rcu_assign_pointer(vcpu->pid, newpid);
4459			if (oldpid)
4460				synchronize_rcu();
4461			put_pid(oldpid);
4462		}
4463		r = kvm_arch_vcpu_ioctl_run(vcpu);
4464		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4465		break;
4466	}
4467	case KVM_GET_REGS: {
4468		struct kvm_regs *kvm_regs;
4469
4470		r = -ENOMEM;
4471		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4472		if (!kvm_regs)
4473			goto out;
4474		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4475		if (r)
4476			goto out_free1;
4477		r = -EFAULT;
4478		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4479			goto out_free1;
4480		r = 0;
4481out_free1:
4482		kfree(kvm_regs);
4483		break;
4484	}
4485	case KVM_SET_REGS: {
4486		struct kvm_regs *kvm_regs;
4487
4488		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4489		if (IS_ERR(kvm_regs)) {
4490			r = PTR_ERR(kvm_regs);
4491			goto out;
4492		}
4493		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4494		kfree(kvm_regs);
4495		break;
4496	}
4497	case KVM_GET_SREGS: {
4498		kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4499				    GFP_KERNEL_ACCOUNT);
4500		r = -ENOMEM;
4501		if (!kvm_sregs)
4502			goto out;
4503		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4504		if (r)
4505			goto out;
4506		r = -EFAULT;
4507		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4508			goto out;
4509		r = 0;
4510		break;
4511	}
4512	case KVM_SET_SREGS: {
4513		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4514		if (IS_ERR(kvm_sregs)) {
4515			r = PTR_ERR(kvm_sregs);
4516			kvm_sregs = NULL;
4517			goto out;
4518		}
4519		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4520		break;
4521	}
4522	case KVM_GET_MP_STATE: {
4523		struct kvm_mp_state mp_state;
4524
4525		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4526		if (r)
4527			goto out;
4528		r = -EFAULT;
4529		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4530			goto out;
4531		r = 0;
4532		break;
4533	}
4534	case KVM_SET_MP_STATE: {
4535		struct kvm_mp_state mp_state;
4536
4537		r = -EFAULT;
4538		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4539			goto out;
4540		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4541		break;
4542	}
4543	case KVM_TRANSLATE: {
4544		struct kvm_translation tr;
4545
4546		r = -EFAULT;
4547		if (copy_from_user(&tr, argp, sizeof(tr)))
4548			goto out;
4549		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4550		if (r)
4551			goto out;
4552		r = -EFAULT;
4553		if (copy_to_user(argp, &tr, sizeof(tr)))
4554			goto out;
4555		r = 0;
4556		break;
4557	}
4558	case KVM_SET_GUEST_DEBUG: {
4559		struct kvm_guest_debug dbg;
4560
4561		r = -EFAULT;
4562		if (copy_from_user(&dbg, argp, sizeof(dbg)))
4563			goto out;
4564		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4565		break;
4566	}
4567	case KVM_SET_SIGNAL_MASK: {
4568		struct kvm_signal_mask __user *sigmask_arg = argp;
4569		struct kvm_signal_mask kvm_sigmask;
4570		sigset_t sigset, *p;
4571
4572		p = NULL;
4573		if (argp) {
4574			r = -EFAULT;
4575			if (copy_from_user(&kvm_sigmask, argp,
4576					   sizeof(kvm_sigmask)))
4577				goto out;
4578			r = -EINVAL;
4579			if (kvm_sigmask.len != sizeof(sigset))
4580				goto out;
4581			r = -EFAULT;
4582			if (copy_from_user(&sigset, sigmask_arg->sigset,
4583					   sizeof(sigset)))
4584				goto out;
4585			p = &sigset;
4586		}
4587		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4588		break;
4589	}
4590	case KVM_GET_FPU: {
4591		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4592		r = -ENOMEM;
4593		if (!fpu)
4594			goto out;
4595		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4596		if (r)
4597			goto out;
4598		r = -EFAULT;
4599		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4600			goto out;
4601		r = 0;
4602		break;
4603	}
4604	case KVM_SET_FPU: {
4605		fpu = memdup_user(argp, sizeof(*fpu));
4606		if (IS_ERR(fpu)) {
4607			r = PTR_ERR(fpu);
4608			fpu = NULL;
4609			goto out;
4610		}
4611		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4612		break;
4613	}
4614	case KVM_GET_STATS_FD: {
4615		r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4616		break;
4617	}
4618	default:
4619		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4620	}
4621out:
4622	mutex_unlock(&vcpu->mutex);
4623	kfree(fpu);
4624	kfree(kvm_sregs);
4625	return r;
4626}
4627
4628#ifdef CONFIG_KVM_COMPAT
4629static long kvm_vcpu_compat_ioctl(struct file *filp,
4630				  unsigned int ioctl, unsigned long arg)
4631{
4632	struct kvm_vcpu *vcpu = filp->private_data;
4633	void __user *argp = compat_ptr(arg);
4634	int r;
4635
4636	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4637		return -EIO;
4638
4639	switch (ioctl) {
4640	case KVM_SET_SIGNAL_MASK: {
4641		struct kvm_signal_mask __user *sigmask_arg = argp;
4642		struct kvm_signal_mask kvm_sigmask;
4643		sigset_t sigset;
4644
4645		if (argp) {
4646			r = -EFAULT;
4647			if (copy_from_user(&kvm_sigmask, argp,
4648					   sizeof(kvm_sigmask)))
4649				goto out;
4650			r = -EINVAL;
4651			if (kvm_sigmask.len != sizeof(compat_sigset_t))
4652				goto out;
4653			r = -EFAULT;
4654			if (get_compat_sigset(&sigset,
4655					      (compat_sigset_t __user *)sigmask_arg->sigset))
4656				goto out;
4657			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4658		} else
4659			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4660		break;
4661	}
4662	default:
4663		r = kvm_vcpu_ioctl(filp, ioctl, arg);
4664	}
4665
4666out:
4667	return r;
4668}
4669#endif
4670
4671static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4672{
4673	struct kvm_device *dev = filp->private_data;
4674
4675	if (dev->ops->mmap)
4676		return dev->ops->mmap(dev, vma);
4677
4678	return -ENODEV;
4679}
4680
4681static int kvm_device_ioctl_attr(struct kvm_device *dev,
4682				 int (*accessor)(struct kvm_device *dev,
4683						 struct kvm_device_attr *attr),
4684				 unsigned long arg)
4685{
4686	struct kvm_device_attr attr;
4687
4688	if (!accessor)
4689		return -EPERM;
4690
4691	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4692		return -EFAULT;
4693
4694	return accessor(dev, &attr);
4695}
4696
4697static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4698			     unsigned long arg)
4699{
4700	struct kvm_device *dev = filp->private_data;
4701
4702	if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4703		return -EIO;
4704
4705	switch (ioctl) {
4706	case KVM_SET_DEVICE_ATTR:
4707		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4708	case KVM_GET_DEVICE_ATTR:
4709		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4710	case KVM_HAS_DEVICE_ATTR:
4711		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4712	default:
4713		if (dev->ops->ioctl)
4714			return dev->ops->ioctl(dev, ioctl, arg);
4715
4716		return -ENOTTY;
4717	}
4718}
4719
4720static int kvm_device_release(struct inode *inode, struct file *filp)
4721{
4722	struct kvm_device *dev = filp->private_data;
4723	struct kvm *kvm = dev->kvm;
4724
4725	if (dev->ops->release) {
4726		mutex_lock(&kvm->lock);
4727		list_del(&dev->vm_node);
4728		dev->ops->release(dev);
4729		mutex_unlock(&kvm->lock);
4730	}
4731
4732	kvm_put_kvm(kvm);
4733	return 0;
4734}
4735
4736static struct file_operations kvm_device_fops = {
4737	.unlocked_ioctl = kvm_device_ioctl,
4738	.release = kvm_device_release,
4739	KVM_COMPAT(kvm_device_ioctl),
4740	.mmap = kvm_device_mmap,
4741};
4742
4743struct kvm_device *kvm_device_from_filp(struct file *filp)
4744{
4745	if (filp->f_op != &kvm_device_fops)
4746		return NULL;
4747
4748	return filp->private_data;
4749}
4750
4751static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4752#ifdef CONFIG_KVM_MPIC
4753	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
4754	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
4755#endif
4756};
4757
4758int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4759{
4760	if (type >= ARRAY_SIZE(kvm_device_ops_table))
4761		return -ENOSPC;
4762
4763	if (kvm_device_ops_table[type] != NULL)
4764		return -EEXIST;
4765
4766	kvm_device_ops_table[type] = ops;
4767	return 0;
4768}
4769
4770void kvm_unregister_device_ops(u32 type)
4771{
4772	if (kvm_device_ops_table[type] != NULL)
4773		kvm_device_ops_table[type] = NULL;
4774}
4775
4776static int kvm_ioctl_create_device(struct kvm *kvm,
4777				   struct kvm_create_device *cd)
4778{
4779	const struct kvm_device_ops *ops;
4780	struct kvm_device *dev;
4781	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4782	int type;
4783	int ret;
4784
4785	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4786		return -ENODEV;
4787
4788	type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4789	ops = kvm_device_ops_table[type];
4790	if (ops == NULL)
4791		return -ENODEV;
4792
4793	if (test)
4794		return 0;
4795
4796	dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4797	if (!dev)
4798		return -ENOMEM;
4799
4800	dev->ops = ops;
4801	dev->kvm = kvm;
4802
4803	mutex_lock(&kvm->lock);
4804	ret = ops->create(dev, type);
4805	if (ret < 0) {
4806		mutex_unlock(&kvm->lock);
4807		kfree(dev);
4808		return ret;
4809	}
4810	list_add(&dev->vm_node, &kvm->devices);
4811	mutex_unlock(&kvm->lock);
4812
4813	if (ops->init)
4814		ops->init(dev);
4815
4816	kvm_get_kvm(kvm);
4817	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4818	if (ret < 0) {
4819		kvm_put_kvm_no_destroy(kvm);
4820		mutex_lock(&kvm->lock);
4821		list_del(&dev->vm_node);
4822		if (ops->release)
4823			ops->release(dev);
4824		mutex_unlock(&kvm->lock);
4825		if (ops->destroy)
4826			ops->destroy(dev);
4827		return ret;
4828	}
4829
4830	cd->fd = ret;
4831	return 0;
4832}
4833
4834static int kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4835{
4836	switch (arg) {
4837	case KVM_CAP_USER_MEMORY:
4838	case KVM_CAP_USER_MEMORY2:
4839	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4840	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4841	case KVM_CAP_INTERNAL_ERROR_DATA:
4842#ifdef CONFIG_HAVE_KVM_MSI
4843	case KVM_CAP_SIGNAL_MSI:
4844#endif
4845#ifdef CONFIG_HAVE_KVM_IRQCHIP
4846	case KVM_CAP_IRQFD:
4847#endif
4848	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4849	case KVM_CAP_CHECK_EXTENSION_VM:
4850	case KVM_CAP_ENABLE_CAP_VM:
4851	case KVM_CAP_HALT_POLL:
4852		return 1;
4853#ifdef CONFIG_KVM_MMIO
4854	case KVM_CAP_COALESCED_MMIO:
4855		return KVM_COALESCED_MMIO_PAGE_OFFSET;
4856	case KVM_CAP_COALESCED_PIO:
4857		return 1;
4858#endif
4859#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4860	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4861		return KVM_DIRTY_LOG_MANUAL_CAPS;
4862#endif
4863#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4864	case KVM_CAP_IRQ_ROUTING:
4865		return KVM_MAX_IRQ_ROUTES;
4866#endif
4867#if KVM_MAX_NR_ADDRESS_SPACES > 1
4868	case KVM_CAP_MULTI_ADDRESS_SPACE:
4869		if (kvm)
4870			return kvm_arch_nr_memslot_as_ids(kvm);
4871		return KVM_MAX_NR_ADDRESS_SPACES;
4872#endif
4873	case KVM_CAP_NR_MEMSLOTS:
4874		return KVM_USER_MEM_SLOTS;
4875	case KVM_CAP_DIRTY_LOG_RING:
4876#ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4877		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4878#else
4879		return 0;
4880#endif
4881	case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4882#ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4883		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4884#else
4885		return 0;
4886#endif
4887#ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4888	case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4889#endif
4890	case KVM_CAP_BINARY_STATS_FD:
4891	case KVM_CAP_SYSTEM_EVENT_DATA:
4892	case KVM_CAP_DEVICE_CTRL:
4893		return 1;
4894#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
4895	case KVM_CAP_MEMORY_ATTRIBUTES:
4896		return kvm_supported_mem_attributes(kvm);
4897#endif
4898#ifdef CONFIG_KVM_PRIVATE_MEM
4899	case KVM_CAP_GUEST_MEMFD:
4900		return !kvm || kvm_arch_has_private_mem(kvm);
4901#endif
4902	default:
4903		break;
4904	}
4905	return kvm_vm_ioctl_check_extension(kvm, arg);
4906}
4907
4908static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4909{
4910	int r;
4911
4912	if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4913		return -EINVAL;
4914
4915	/* the size should be power of 2 */
4916	if (!size || (size & (size - 1)))
4917		return -EINVAL;
4918
4919	/* Should be bigger to keep the reserved entries, or a page */
4920	if (size < kvm_dirty_ring_get_rsvd_entries() *
4921	    sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4922		return -EINVAL;
4923
4924	if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4925	    sizeof(struct kvm_dirty_gfn))
4926		return -E2BIG;
4927
4928	/* We only allow it to set once */
4929	if (kvm->dirty_ring_size)
4930		return -EINVAL;
4931
4932	mutex_lock(&kvm->lock);
4933
4934	if (kvm->created_vcpus) {
4935		/* We don't allow to change this value after vcpu created */
4936		r = -EINVAL;
4937	} else {
4938		kvm->dirty_ring_size = size;
4939		r = 0;
4940	}
4941
4942	mutex_unlock(&kvm->lock);
4943	return r;
4944}
4945
4946static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4947{
4948	unsigned long i;
4949	struct kvm_vcpu *vcpu;
4950	int cleared = 0;
4951
4952	if (!kvm->dirty_ring_size)
4953		return -EINVAL;
4954
4955	mutex_lock(&kvm->slots_lock);
4956
4957	kvm_for_each_vcpu(i, vcpu, kvm)
4958		cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4959
4960	mutex_unlock(&kvm->slots_lock);
4961
4962	if (cleared)
4963		kvm_flush_remote_tlbs(kvm);
4964
4965	return cleared;
4966}
4967
4968int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4969						  struct kvm_enable_cap *cap)
4970{
4971	return -EINVAL;
4972}
4973
4974bool kvm_are_all_memslots_empty(struct kvm *kvm)
4975{
4976	int i;
4977
4978	lockdep_assert_held(&kvm->slots_lock);
4979
4980	for (i = 0; i < kvm_arch_nr_memslot_as_ids(kvm); i++) {
4981		if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4982			return false;
4983	}
4984
4985	return true;
4986}
4987EXPORT_SYMBOL_GPL(kvm_are_all_memslots_empty);
4988
4989static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4990					   struct kvm_enable_cap *cap)
4991{
4992	switch (cap->cap) {
4993#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4994	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4995		u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4996
4997		if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4998			allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4999
5000		if (cap->flags || (cap->args[0] & ~allowed_options))
5001			return -EINVAL;
5002		kvm->manual_dirty_log_protect = cap->args[0];
5003		return 0;
5004	}
5005#endif
5006	case KVM_CAP_HALT_POLL: {
5007		if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
5008			return -EINVAL;
5009
5010		kvm->max_halt_poll_ns = cap->args[0];
5011
5012		/*
5013		 * Ensure kvm->override_halt_poll_ns does not become visible
5014		 * before kvm->max_halt_poll_ns.
5015		 *
5016		 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
5017		 */
5018		smp_wmb();
5019		kvm->override_halt_poll_ns = true;
5020
5021		return 0;
5022	}
5023	case KVM_CAP_DIRTY_LOG_RING:
5024	case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
5025		if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
5026			return -EINVAL;
5027
5028		return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
5029	case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
5030		int r = -EINVAL;
5031
5032		if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
5033		    !kvm->dirty_ring_size || cap->flags)
5034			return r;
5035
5036		mutex_lock(&kvm->slots_lock);
5037
5038		/*
5039		 * For simplicity, allow enabling ring+bitmap if and only if
5040		 * there are no memslots, e.g. to ensure all memslots allocate
5041		 * a bitmap after the capability is enabled.
5042		 */
5043		if (kvm_are_all_memslots_empty(kvm)) {
5044			kvm->dirty_ring_with_bitmap = true;
5045			r = 0;
5046		}
5047
5048		mutex_unlock(&kvm->slots_lock);
5049
5050		return r;
5051	}
5052	default:
5053		return kvm_vm_ioctl_enable_cap(kvm, cap);
5054	}
5055}
5056
5057static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
5058			      size_t size, loff_t *offset)
5059{
5060	struct kvm *kvm = file->private_data;
5061
5062	return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
5063				&kvm_vm_stats_desc[0], &kvm->stat,
5064				sizeof(kvm->stat), user_buffer, size, offset);
5065}
5066
5067static int kvm_vm_stats_release(struct inode *inode, struct file *file)
5068{
5069	struct kvm *kvm = file->private_data;
5070
5071	kvm_put_kvm(kvm);
5072	return 0;
5073}
5074
5075static const struct file_operations kvm_vm_stats_fops = {
5076	.owner = THIS_MODULE,
5077	.read = kvm_vm_stats_read,
5078	.release = kvm_vm_stats_release,
5079	.llseek = noop_llseek,
5080};
5081
5082static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
5083{
5084	int fd;
5085	struct file *file;
5086
5087	fd = get_unused_fd_flags(O_CLOEXEC);
5088	if (fd < 0)
5089		return fd;
5090
5091	file = anon_inode_getfile("kvm-vm-stats",
5092			&kvm_vm_stats_fops, kvm, O_RDONLY);
5093	if (IS_ERR(file)) {
5094		put_unused_fd(fd);
5095		return PTR_ERR(file);
5096	}
5097
5098	kvm_get_kvm(kvm);
5099
5100	file->f_mode |= FMODE_PREAD;
5101	fd_install(fd, file);
5102
5103	return fd;
5104}
5105
5106#define SANITY_CHECK_MEM_REGION_FIELD(field)					\
5107do {										\
5108	BUILD_BUG_ON(offsetof(struct kvm_userspace_memory_region, field) !=		\
5109		     offsetof(struct kvm_userspace_memory_region2, field));	\
5110	BUILD_BUG_ON(sizeof_field(struct kvm_userspace_memory_region, field) !=		\
5111		     sizeof_field(struct kvm_userspace_memory_region2, field));	\
5112} while (0)
5113
5114static long kvm_vm_ioctl(struct file *filp,
5115			   unsigned int ioctl, unsigned long arg)
5116{
5117	struct kvm *kvm = filp->private_data;
5118	void __user *argp = (void __user *)arg;
5119	int r;
5120
5121	if (kvm->mm != current->mm || kvm->vm_dead)
5122		return -EIO;
5123	switch (ioctl) {
5124	case KVM_CREATE_VCPU:
5125		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
5126		break;
5127	case KVM_ENABLE_CAP: {
5128		struct kvm_enable_cap cap;
5129
5130		r = -EFAULT;
5131		if (copy_from_user(&cap, argp, sizeof(cap)))
5132			goto out;
5133		r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
5134		break;
5135	}
5136	case KVM_SET_USER_MEMORY_REGION2:
5137	case KVM_SET_USER_MEMORY_REGION: {
5138		struct kvm_userspace_memory_region2 mem;
5139		unsigned long size;
5140
5141		if (ioctl == KVM_SET_USER_MEMORY_REGION) {
5142			/*
5143			 * Fields beyond struct kvm_userspace_memory_region shouldn't be
5144			 * accessed, but avoid leaking kernel memory in case of a bug.
5145			 */
5146			memset(&mem, 0, sizeof(mem));
5147			size = sizeof(struct kvm_userspace_memory_region);
5148		} else {
5149			size = sizeof(struct kvm_userspace_memory_region2);
5150		}
5151
5152		/* Ensure the common parts of the two structs are identical. */
5153		SANITY_CHECK_MEM_REGION_FIELD(slot);
5154		SANITY_CHECK_MEM_REGION_FIELD(flags);
5155		SANITY_CHECK_MEM_REGION_FIELD(guest_phys_addr);
5156		SANITY_CHECK_MEM_REGION_FIELD(memory_size);
5157		SANITY_CHECK_MEM_REGION_FIELD(userspace_addr);
5158
5159		r = -EFAULT;
5160		if (copy_from_user(&mem, argp, size))
5161			goto out;
5162
5163		r = -EINVAL;
5164		if (ioctl == KVM_SET_USER_MEMORY_REGION &&
5165		    (mem.flags & ~KVM_SET_USER_MEMORY_REGION_V1_FLAGS))
5166			goto out;
5167
5168		r = kvm_vm_ioctl_set_memory_region(kvm, &mem);
5169		break;
5170	}
5171	case KVM_GET_DIRTY_LOG: {
5172		struct kvm_dirty_log log;
5173
5174		r = -EFAULT;
5175		if (copy_from_user(&log, argp, sizeof(log)))
5176			goto out;
5177		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5178		break;
5179	}
5180#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5181	case KVM_CLEAR_DIRTY_LOG: {
5182		struct kvm_clear_dirty_log log;
5183
5184		r = -EFAULT;
5185		if (copy_from_user(&log, argp, sizeof(log)))
5186			goto out;
5187		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
5188		break;
5189	}
5190#endif
5191#ifdef CONFIG_KVM_MMIO
5192	case KVM_REGISTER_COALESCED_MMIO: {
5193		struct kvm_coalesced_mmio_zone zone;
5194
5195		r = -EFAULT;
5196		if (copy_from_user(&zone, argp, sizeof(zone)))
5197			goto out;
5198		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
5199		break;
5200	}
5201	case KVM_UNREGISTER_COALESCED_MMIO: {
5202		struct kvm_coalesced_mmio_zone zone;
5203
5204		r = -EFAULT;
5205		if (copy_from_user(&zone, argp, sizeof(zone)))
5206			goto out;
5207		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
5208		break;
5209	}
5210#endif
5211	case KVM_IRQFD: {
5212		struct kvm_irqfd data;
5213
5214		r = -EFAULT;
5215		if (copy_from_user(&data, argp, sizeof(data)))
5216			goto out;
5217		r = kvm_irqfd(kvm, &data);
5218		break;
5219	}
5220	case KVM_IOEVENTFD: {
5221		struct kvm_ioeventfd data;
5222
5223		r = -EFAULT;
5224		if (copy_from_user(&data, argp, sizeof(data)))
5225			goto out;
5226		r = kvm_ioeventfd(kvm, &data);
5227		break;
5228	}
5229#ifdef CONFIG_HAVE_KVM_MSI
5230	case KVM_SIGNAL_MSI: {
5231		struct kvm_msi msi;
5232
5233		r = -EFAULT;
5234		if (copy_from_user(&msi, argp, sizeof(msi)))
5235			goto out;
5236		r = kvm_send_userspace_msi(kvm, &msi);
5237		break;
5238	}
5239#endif
5240#ifdef __KVM_HAVE_IRQ_LINE
5241	case KVM_IRQ_LINE_STATUS:
5242	case KVM_IRQ_LINE: {
5243		struct kvm_irq_level irq_event;
5244
5245		r = -EFAULT;
5246		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
5247			goto out;
5248
5249		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
5250					ioctl == KVM_IRQ_LINE_STATUS);
5251		if (r)
5252			goto out;
5253
5254		r = -EFAULT;
5255		if (ioctl == KVM_IRQ_LINE_STATUS) {
5256			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
5257				goto out;
5258		}
5259
5260		r = 0;
5261		break;
5262	}
5263#endif
5264#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
5265	case KVM_SET_GSI_ROUTING: {
5266		struct kvm_irq_routing routing;
5267		struct kvm_irq_routing __user *urouting;
5268		struct kvm_irq_routing_entry *entries = NULL;
5269
5270		r = -EFAULT;
5271		if (copy_from_user(&routing, argp, sizeof(routing)))
5272			goto out;
5273		r = -EINVAL;
5274		if (!kvm_arch_can_set_irq_routing(kvm))
5275			goto out;
5276		if (routing.nr > KVM_MAX_IRQ_ROUTES)
5277			goto out;
5278		if (routing.flags)
5279			goto out;
5280		if (routing.nr) {
5281			urouting = argp;
5282			entries = vmemdup_array_user(urouting->entries,
5283						     routing.nr, sizeof(*entries));
5284			if (IS_ERR(entries)) {
5285				r = PTR_ERR(entries);
5286				goto out;
5287			}
5288		}
5289		r = kvm_set_irq_routing(kvm, entries, routing.nr,
5290					routing.flags);
5291		kvfree(entries);
5292		break;
5293	}
5294#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
5295#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
5296	case KVM_SET_MEMORY_ATTRIBUTES: {
5297		struct kvm_memory_attributes attrs;
5298
5299		r = -EFAULT;
5300		if (copy_from_user(&attrs, argp, sizeof(attrs)))
5301			goto out;
5302
5303		r = kvm_vm_ioctl_set_mem_attributes(kvm, &attrs);
5304		break;
5305	}
5306#endif /* CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES */
5307	case KVM_CREATE_DEVICE: {
5308		struct kvm_create_device cd;
5309
5310		r = -EFAULT;
5311		if (copy_from_user(&cd, argp, sizeof(cd)))
5312			goto out;
5313
5314		r = kvm_ioctl_create_device(kvm, &cd);
5315		if (r)
5316			goto out;
5317
5318		r = -EFAULT;
5319		if (copy_to_user(argp, &cd, sizeof(cd)))
5320			goto out;
5321
5322		r = 0;
5323		break;
5324	}
5325	case KVM_CHECK_EXTENSION:
5326		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
5327		break;
5328	case KVM_RESET_DIRTY_RINGS:
5329		r = kvm_vm_ioctl_reset_dirty_pages(kvm);
5330		break;
5331	case KVM_GET_STATS_FD:
5332		r = kvm_vm_ioctl_get_stats_fd(kvm);
5333		break;
5334#ifdef CONFIG_KVM_PRIVATE_MEM
5335	case KVM_CREATE_GUEST_MEMFD: {
5336		struct kvm_create_guest_memfd guest_memfd;
5337
5338		r = -EFAULT;
5339		if (copy_from_user(&guest_memfd, argp, sizeof(guest_memfd)))
5340			goto out;
5341
5342		r = kvm_gmem_create(kvm, &guest_memfd);
5343		break;
5344	}
5345#endif
5346	default:
5347		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
5348	}
5349out:
5350	return r;
5351}
5352
5353#ifdef CONFIG_KVM_COMPAT
5354struct compat_kvm_dirty_log {
5355	__u32 slot;
5356	__u32 padding1;
5357	union {
5358		compat_uptr_t dirty_bitmap; /* one bit per page */
5359		__u64 padding2;
5360	};
5361};
5362
5363struct compat_kvm_clear_dirty_log {
5364	__u32 slot;
5365	__u32 num_pages;
5366	__u64 first_page;
5367	union {
5368		compat_uptr_t dirty_bitmap; /* one bit per page */
5369		__u64 padding2;
5370	};
5371};
5372
5373long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
5374				     unsigned long arg)
5375{
5376	return -ENOTTY;
5377}
5378
5379static long kvm_vm_compat_ioctl(struct file *filp,
5380			   unsigned int ioctl, unsigned long arg)
5381{
5382	struct kvm *kvm = filp->private_data;
5383	int r;
5384
5385	if (kvm->mm != current->mm || kvm->vm_dead)
5386		return -EIO;
5387
5388	r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
5389	if (r != -ENOTTY)
5390		return r;
5391
5392	switch (ioctl) {
5393#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
5394	case KVM_CLEAR_DIRTY_LOG: {
5395		struct compat_kvm_clear_dirty_log compat_log;
5396		struct kvm_clear_dirty_log log;
5397
5398		if (copy_from_user(&compat_log, (void __user *)arg,
5399				   sizeof(compat_log)))
5400			return -EFAULT;
5401		log.slot	 = compat_log.slot;
5402		log.num_pages	 = compat_log.num_pages;
5403		log.first_page	 = compat_log.first_page;
5404		log.padding2	 = compat_log.padding2;
5405		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5406
5407		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
5408		break;
5409	}
5410#endif
5411	case KVM_GET_DIRTY_LOG: {
5412		struct compat_kvm_dirty_log compat_log;
5413		struct kvm_dirty_log log;
5414
5415		if (copy_from_user(&compat_log, (void __user *)arg,
5416				   sizeof(compat_log)))
5417			return -EFAULT;
5418		log.slot	 = compat_log.slot;
5419		log.padding1	 = compat_log.padding1;
5420		log.padding2	 = compat_log.padding2;
5421		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
5422
5423		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
5424		break;
5425	}
5426	default:
5427		r = kvm_vm_ioctl(filp, ioctl, arg);
5428	}
5429	return r;
5430}
5431#endif
5432
5433static struct file_operations kvm_vm_fops = {
5434	.release        = kvm_vm_release,
5435	.unlocked_ioctl = kvm_vm_ioctl,
5436	.llseek		= noop_llseek,
5437	KVM_COMPAT(kvm_vm_compat_ioctl),
5438};
5439
5440bool file_is_kvm(struct file *file)
5441{
5442	return file && file->f_op == &kvm_vm_fops;
5443}
5444EXPORT_SYMBOL_GPL(file_is_kvm);
5445
5446static int kvm_dev_ioctl_create_vm(unsigned long type)
5447{
5448	char fdname[ITOA_MAX_LEN + 1];
5449	int r, fd;
5450	struct kvm *kvm;
5451	struct file *file;
5452
5453	fd = get_unused_fd_flags(O_CLOEXEC);
5454	if (fd < 0)
5455		return fd;
5456
5457	snprintf(fdname, sizeof(fdname), "%d", fd);
5458
5459	kvm = kvm_create_vm(type, fdname);
5460	if (IS_ERR(kvm)) {
5461		r = PTR_ERR(kvm);
5462		goto put_fd;
5463	}
5464
5465	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5466	if (IS_ERR(file)) {
5467		r = PTR_ERR(file);
5468		goto put_kvm;
5469	}
5470
5471	/*
5472	 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5473	 * already set, with ->release() being kvm_vm_release().  In error
5474	 * cases it will be called by the final fput(file) and will take
5475	 * care of doing kvm_put_kvm(kvm).
5476	 */
5477	kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5478
5479	fd_install(fd, file);
5480	return fd;
5481
5482put_kvm:
5483	kvm_put_kvm(kvm);
5484put_fd:
5485	put_unused_fd(fd);
5486	return r;
5487}
5488
5489static long kvm_dev_ioctl(struct file *filp,
5490			  unsigned int ioctl, unsigned long arg)
5491{
5492	int r = -EINVAL;
5493
5494	switch (ioctl) {
5495	case KVM_GET_API_VERSION:
5496		if (arg)
5497			goto out;
5498		r = KVM_API_VERSION;
5499		break;
5500	case KVM_CREATE_VM:
5501		r = kvm_dev_ioctl_create_vm(arg);
5502		break;
5503	case KVM_CHECK_EXTENSION:
5504		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5505		break;
5506	case KVM_GET_VCPU_MMAP_SIZE:
5507		if (arg)
5508			goto out;
5509		r = PAGE_SIZE;     /* struct kvm_run */
5510#ifdef CONFIG_X86
5511		r += PAGE_SIZE;    /* pio data page */
5512#endif
5513#ifdef CONFIG_KVM_MMIO
5514		r += PAGE_SIZE;    /* coalesced mmio ring page */
5515#endif
5516		break;
5517	default:
5518		return kvm_arch_dev_ioctl(filp, ioctl, arg);
5519	}
5520out:
5521	return r;
5522}
5523
5524static struct file_operations kvm_chardev_ops = {
5525	.unlocked_ioctl = kvm_dev_ioctl,
5526	.llseek		= noop_llseek,
5527	KVM_COMPAT(kvm_dev_ioctl),
5528};
5529
5530static struct miscdevice kvm_dev = {
5531	KVM_MINOR,
5532	"kvm",
5533	&kvm_chardev_ops,
5534};
5535
5536#ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
5537__visible bool kvm_rebooting;
5538EXPORT_SYMBOL_GPL(kvm_rebooting);
5539
5540static DEFINE_PER_CPU(bool, hardware_enabled);
5541static int kvm_usage_count;
5542
5543static int __hardware_enable_nolock(void)
5544{
5545	if (__this_cpu_read(hardware_enabled))
5546		return 0;
5547
5548	if (kvm_arch_hardware_enable()) {
5549		pr_info("kvm: enabling virtualization on CPU%d failed\n",
5550			raw_smp_processor_id());
5551		return -EIO;
5552	}
5553
5554	__this_cpu_write(hardware_enabled, true);
5555	return 0;
5556}
5557
5558static void hardware_enable_nolock(void *failed)
5559{
5560	if (__hardware_enable_nolock())
5561		atomic_inc(failed);
5562}
5563
5564static int kvm_online_cpu(unsigned int cpu)
5565{
5566	int ret = 0;
5567
5568	/*
5569	 * Abort the CPU online process if hardware virtualization cannot
5570	 * be enabled. Otherwise running VMs would encounter unrecoverable
5571	 * errors when scheduled to this CPU.
5572	 */
5573	mutex_lock(&kvm_lock);
5574	if (kvm_usage_count)
5575		ret = __hardware_enable_nolock();
5576	mutex_unlock(&kvm_lock);
5577	return ret;
5578}
5579
5580static void hardware_disable_nolock(void *junk)
5581{
5582	/*
5583	 * Note, hardware_disable_all_nolock() tells all online CPUs to disable
5584	 * hardware, not just CPUs that successfully enabled hardware!
5585	 */
5586	if (!__this_cpu_read(hardware_enabled))
5587		return;
5588
5589	kvm_arch_hardware_disable();
5590
5591	__this_cpu_write(hardware_enabled, false);
5592}
5593
5594static int kvm_offline_cpu(unsigned int cpu)
5595{
5596	mutex_lock(&kvm_lock);
5597	if (kvm_usage_count)
5598		hardware_disable_nolock(NULL);
5599	mutex_unlock(&kvm_lock);
5600	return 0;
5601}
5602
5603static void hardware_disable_all_nolock(void)
5604{
5605	BUG_ON(!kvm_usage_count);
5606
5607	kvm_usage_count--;
5608	if (!kvm_usage_count)
5609		on_each_cpu(hardware_disable_nolock, NULL, 1);
5610}
5611
5612static void hardware_disable_all(void)
5613{
5614	cpus_read_lock();
5615	mutex_lock(&kvm_lock);
5616	hardware_disable_all_nolock();
5617	mutex_unlock(&kvm_lock);
5618	cpus_read_unlock();
5619}
5620
5621static int hardware_enable_all(void)
5622{
5623	atomic_t failed = ATOMIC_INIT(0);
5624	int r;
5625
5626	/*
5627	 * Do not enable hardware virtualization if the system is going down.
5628	 * If userspace initiated a forced reboot, e.g. reboot -f, then it's
5629	 * possible for an in-flight KVM_CREATE_VM to trigger hardware enabling
5630	 * after kvm_reboot() is called.  Note, this relies on system_state
5631	 * being set _before_ kvm_reboot(), which is why KVM uses a syscore ops
5632	 * hook instead of registering a dedicated reboot notifier (the latter
5633	 * runs before system_state is updated).
5634	 */
5635	if (system_state == SYSTEM_HALT || system_state == SYSTEM_POWER_OFF ||
5636	    system_state == SYSTEM_RESTART)
5637		return -EBUSY;
5638
5639	/*
5640	 * When onlining a CPU, cpu_online_mask is set before kvm_online_cpu()
5641	 * is called, and so on_each_cpu() between them includes the CPU that
5642	 * is being onlined.  As a result, hardware_enable_nolock() may get
5643	 * invoked before kvm_online_cpu(), which also enables hardware if the
5644	 * usage count is non-zero.  Disable CPU hotplug to avoid attempting to
5645	 * enable hardware multiple times.
5646	 */
5647	cpus_read_lock();
5648	mutex_lock(&kvm_lock);
5649
5650	r = 0;
5651
5652	kvm_usage_count++;
5653	if (kvm_usage_count == 1) {
5654		on_each_cpu(hardware_enable_nolock, &failed, 1);
5655
5656		if (atomic_read(&failed)) {
5657			hardware_disable_all_nolock();
5658			r = -EBUSY;
5659		}
5660	}
5661
5662	mutex_unlock(&kvm_lock);
5663	cpus_read_unlock();
5664
5665	return r;
5666}
5667
5668static void kvm_shutdown(void)
5669{
5670	/*
5671	 * Disable hardware virtualization and set kvm_rebooting to indicate
5672	 * that KVM has asynchronously disabled hardware virtualization, i.e.
5673	 * that relevant errors and exceptions aren't entirely unexpected.
5674	 * Some flavors of hardware virtualization need to be disabled before
5675	 * transferring control to firmware (to perform shutdown/reboot), e.g.
5676	 * on x86, virtualization can block INIT interrupts, which are used by
5677	 * firmware to pull APs back under firmware control.  Note, this path
5678	 * is used for both shutdown and reboot scenarios, i.e. neither name is
5679	 * 100% comprehensive.
5680	 */
5681	pr_info("kvm: exiting hardware virtualization\n");
5682	kvm_rebooting = true;
5683	on_each_cpu(hardware_disable_nolock, NULL, 1);
5684}
5685
5686static int kvm_suspend(void)
5687{
5688	/*
5689	 * Secondary CPUs and CPU hotplug are disabled across the suspend/resume
5690	 * callbacks, i.e. no need to acquire kvm_lock to ensure the usage count
5691	 * is stable.  Assert that kvm_lock is not held to ensure the system
5692	 * isn't suspended while KVM is enabling hardware.  Hardware enabling
5693	 * can be preempted, but the task cannot be frozen until it has dropped
5694	 * all locks (userspace tasks are frozen via a fake signal).
5695	 */
5696	lockdep_assert_not_held(&kvm_lock);
5697	lockdep_assert_irqs_disabled();
5698
5699	if (kvm_usage_count)
5700		hardware_disable_nolock(NULL);
5701	return 0;
5702}
5703
5704static void kvm_resume(void)
5705{
5706	lockdep_assert_not_held(&kvm_lock);
5707	lockdep_assert_irqs_disabled();
5708
5709	if (kvm_usage_count)
5710		WARN_ON_ONCE(__hardware_enable_nolock());
5711}
5712
5713static struct syscore_ops kvm_syscore_ops = {
5714	.suspend = kvm_suspend,
5715	.resume = kvm_resume,
5716	.shutdown = kvm_shutdown,
5717};
5718#else /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5719static int hardware_enable_all(void)
5720{
5721	return 0;
5722}
5723
5724static void hardware_disable_all(void)
5725{
5726
5727}
5728#endif /* CONFIG_KVM_GENERIC_HARDWARE_ENABLING */
5729
5730static void kvm_iodevice_destructor(struct kvm_io_device *dev)
5731{
5732	if (dev->ops->destructor)
5733		dev->ops->destructor(dev);
5734}
5735
5736static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5737{
5738	int i;
5739
5740	for (i = 0; i < bus->dev_count; i++) {
5741		struct kvm_io_device *pos = bus->range[i].dev;
5742
5743		kvm_iodevice_destructor(pos);
5744	}
5745	kfree(bus);
5746}
5747
5748static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5749				 const struct kvm_io_range *r2)
5750{
5751	gpa_t addr1 = r1->addr;
5752	gpa_t addr2 = r2->addr;
5753
5754	if (addr1 < addr2)
5755		return -1;
5756
5757	/* If r2->len == 0, match the exact address.  If r2->len != 0,
5758	 * accept any overlapping write.  Any order is acceptable for
5759	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5760	 * we process all of them.
5761	 */
5762	if (r2->len) {
5763		addr1 += r1->len;
5764		addr2 += r2->len;
5765	}
5766
5767	if (addr1 > addr2)
5768		return 1;
5769
5770	return 0;
5771}
5772
5773static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5774{
5775	return kvm_io_bus_cmp(p1, p2);
5776}
5777
5778static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5779			     gpa_t addr, int len)
5780{
5781	struct kvm_io_range *range, key;
5782	int off;
5783
5784	key = (struct kvm_io_range) {
5785		.addr = addr,
5786		.len = len,
5787	};
5788
5789	range = bsearch(&key, bus->range, bus->dev_count,
5790			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5791	if (range == NULL)
5792		return -ENOENT;
5793
5794	off = range - bus->range;
5795
5796	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5797		off--;
5798
5799	return off;
5800}
5801
5802static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5803			      struct kvm_io_range *range, const void *val)
5804{
5805	int idx;
5806
5807	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5808	if (idx < 0)
5809		return -EOPNOTSUPP;
5810
5811	while (idx < bus->dev_count &&
5812		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5813		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5814					range->len, val))
5815			return idx;
5816		idx++;
5817	}
5818
5819	return -EOPNOTSUPP;
5820}
5821
5822/* kvm_io_bus_write - called under kvm->slots_lock */
5823int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5824		     int len, const void *val)
5825{
5826	struct kvm_io_bus *bus;
5827	struct kvm_io_range range;
5828	int r;
5829
5830	range = (struct kvm_io_range) {
5831		.addr = addr,
5832		.len = len,
5833	};
5834
5835	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5836	if (!bus)
5837		return -ENOMEM;
5838	r = __kvm_io_bus_write(vcpu, bus, &range, val);
5839	return r < 0 ? r : 0;
5840}
5841EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5842
5843/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5844int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5845			    gpa_t addr, int len, const void *val, long cookie)
5846{
5847	struct kvm_io_bus *bus;
5848	struct kvm_io_range range;
5849
5850	range = (struct kvm_io_range) {
5851		.addr = addr,
5852		.len = len,
5853	};
5854
5855	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5856	if (!bus)
5857		return -ENOMEM;
5858
5859	/* First try the device referenced by cookie. */
5860	if ((cookie >= 0) && (cookie < bus->dev_count) &&
5861	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5862		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5863					val))
5864			return cookie;
5865
5866	/*
5867	 * cookie contained garbage; fall back to search and return the
5868	 * correct cookie value.
5869	 */
5870	return __kvm_io_bus_write(vcpu, bus, &range, val);
5871}
5872
5873static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5874			     struct kvm_io_range *range, void *val)
5875{
5876	int idx;
5877
5878	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5879	if (idx < 0)
5880		return -EOPNOTSUPP;
5881
5882	while (idx < bus->dev_count &&
5883		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5884		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5885				       range->len, val))
5886			return idx;
5887		idx++;
5888	}
5889
5890	return -EOPNOTSUPP;
5891}
5892
5893/* kvm_io_bus_read - called under kvm->slots_lock */
5894int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5895		    int len, void *val)
5896{
5897	struct kvm_io_bus *bus;
5898	struct kvm_io_range range;
5899	int r;
5900
5901	range = (struct kvm_io_range) {
5902		.addr = addr,
5903		.len = len,
5904	};
5905
5906	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5907	if (!bus)
5908		return -ENOMEM;
5909	r = __kvm_io_bus_read(vcpu, bus, &range, val);
5910	return r < 0 ? r : 0;
5911}
5912
5913int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5914			    int len, struct kvm_io_device *dev)
5915{
5916	int i;
5917	struct kvm_io_bus *new_bus, *bus;
5918	struct kvm_io_range range;
5919
5920	lockdep_assert_held(&kvm->slots_lock);
5921
5922	bus = kvm_get_bus(kvm, bus_idx);
5923	if (!bus)
5924		return -ENOMEM;
5925
5926	/* exclude ioeventfd which is limited by maximum fd */
5927	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5928		return -ENOSPC;
5929
5930	new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5931			  GFP_KERNEL_ACCOUNT);
5932	if (!new_bus)
5933		return -ENOMEM;
5934
5935	range = (struct kvm_io_range) {
5936		.addr = addr,
5937		.len = len,
5938		.dev = dev,
5939	};
5940
5941	for (i = 0; i < bus->dev_count; i++)
5942		if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5943			break;
5944
5945	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5946	new_bus->dev_count++;
5947	new_bus->range[i] = range;
5948	memcpy(new_bus->range + i + 1, bus->range + i,
5949		(bus->dev_count - i) * sizeof(struct kvm_io_range));
5950	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5951	synchronize_srcu_expedited(&kvm->srcu);
5952	kfree(bus);
5953
5954	return 0;
5955}
5956
5957int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5958			      struct kvm_io_device *dev)
5959{
5960	int i;
5961	struct kvm_io_bus *new_bus, *bus;
5962
5963	lockdep_assert_held(&kvm->slots_lock);
5964
5965	bus = kvm_get_bus(kvm, bus_idx);
5966	if (!bus)
5967		return 0;
5968
5969	for (i = 0; i < bus->dev_count; i++) {
5970		if (bus->range[i].dev == dev) {
5971			break;
5972		}
5973	}
5974
5975	if (i == bus->dev_count)
5976		return 0;
5977
5978	new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5979			  GFP_KERNEL_ACCOUNT);
5980	if (new_bus) {
5981		memcpy(new_bus, bus, struct_size(bus, range, i));
5982		new_bus->dev_count--;
5983		memcpy(new_bus->range + i, bus->range + i + 1,
5984				flex_array_size(new_bus, range, new_bus->dev_count - i));
5985	}
5986
5987	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5988	synchronize_srcu_expedited(&kvm->srcu);
5989
5990	/*
5991	 * If NULL bus is installed, destroy the old bus, including all the
5992	 * attached devices. Otherwise, destroy the caller's device only.
5993	 */
5994	if (!new_bus) {
5995		pr_err("kvm: failed to shrink bus, removing it completely\n");
5996		kvm_io_bus_destroy(bus);
5997		return -ENOMEM;
5998	}
5999
6000	kvm_iodevice_destructor(dev);
6001	kfree(bus);
6002	return 0;
6003}
6004
6005struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
6006					 gpa_t addr)
6007{
6008	struct kvm_io_bus *bus;
6009	int dev_idx, srcu_idx;
6010	struct kvm_io_device *iodev = NULL;
6011
6012	srcu_idx = srcu_read_lock(&kvm->srcu);
6013
6014	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
6015	if (!bus)
6016		goto out_unlock;
6017
6018	dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
6019	if (dev_idx < 0)
6020		goto out_unlock;
6021
6022	iodev = bus->range[dev_idx].dev;
6023
6024out_unlock:
6025	srcu_read_unlock(&kvm->srcu, srcu_idx);
6026
6027	return iodev;
6028}
6029EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
6030
6031static int kvm_debugfs_open(struct inode *inode, struct file *file,
6032			   int (*get)(void *, u64 *), int (*set)(void *, u64),
6033			   const char *fmt)
6034{
6035	int ret;
6036	struct kvm_stat_data *stat_data = inode->i_private;
6037
6038	/*
6039	 * The debugfs files are a reference to the kvm struct which
6040        * is still valid when kvm_destroy_vm is called.  kvm_get_kvm_safe
6041        * avoids the race between open and the removal of the debugfs directory.
6042	 */
6043	if (!kvm_get_kvm_safe(stat_data->kvm))
6044		return -ENOENT;
6045
6046	ret = simple_attr_open(inode, file, get,
6047			       kvm_stats_debugfs_mode(stat_data->desc) & 0222
6048			       ? set : NULL, fmt);
6049	if (ret)
6050		kvm_put_kvm(stat_data->kvm);
6051
6052	return ret;
6053}
6054
6055static int kvm_debugfs_release(struct inode *inode, struct file *file)
6056{
6057	struct kvm_stat_data *stat_data = inode->i_private;
6058
6059	simple_attr_release(inode, file);
6060	kvm_put_kvm(stat_data->kvm);
6061
6062	return 0;
6063}
6064
6065static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
6066{
6067	*val = *(u64 *)((void *)(&kvm->stat) + offset);
6068
6069	return 0;
6070}
6071
6072static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
6073{
6074	*(u64 *)((void *)(&kvm->stat) + offset) = 0;
6075
6076	return 0;
6077}
6078
6079static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
6080{
6081	unsigned long i;
6082	struct kvm_vcpu *vcpu;
6083
6084	*val = 0;
6085
6086	kvm_for_each_vcpu(i, vcpu, kvm)
6087		*val += *(u64 *)((void *)(&vcpu->stat) + offset);
6088
6089	return 0;
6090}
6091
6092static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
6093{
6094	unsigned long i;
6095	struct kvm_vcpu *vcpu;
6096
6097	kvm_for_each_vcpu(i, vcpu, kvm)
6098		*(u64 *)((void *)(&vcpu->stat) + offset) = 0;
6099
6100	return 0;
6101}
6102
6103static int kvm_stat_data_get(void *data, u64 *val)
6104{
6105	int r = -EFAULT;
6106	struct kvm_stat_data *stat_data = data;
6107
6108	switch (stat_data->kind) {
6109	case KVM_STAT_VM:
6110		r = kvm_get_stat_per_vm(stat_data->kvm,
6111					stat_data->desc->desc.offset, val);
6112		break;
6113	case KVM_STAT_VCPU:
6114		r = kvm_get_stat_per_vcpu(stat_data->kvm,
6115					  stat_data->desc->desc.offset, val);
6116		break;
6117	}
6118
6119	return r;
6120}
6121
6122static int kvm_stat_data_clear(void *data, u64 val)
6123{
6124	int r = -EFAULT;
6125	struct kvm_stat_data *stat_data = data;
6126
6127	if (val)
6128		return -EINVAL;
6129
6130	switch (stat_data->kind) {
6131	case KVM_STAT_VM:
6132		r = kvm_clear_stat_per_vm(stat_data->kvm,
6133					  stat_data->desc->desc.offset);
6134		break;
6135	case KVM_STAT_VCPU:
6136		r = kvm_clear_stat_per_vcpu(stat_data->kvm,
6137					    stat_data->desc->desc.offset);
6138		break;
6139	}
6140
6141	return r;
6142}
6143
6144static int kvm_stat_data_open(struct inode *inode, struct file *file)
6145{
6146	__simple_attr_check_format("%llu\n", 0ull);
6147	return kvm_debugfs_open(inode, file, kvm_stat_data_get,
6148				kvm_stat_data_clear, "%llu\n");
6149}
6150
6151static const struct file_operations stat_fops_per_vm = {
6152	.owner = THIS_MODULE,
6153	.open = kvm_stat_data_open,
6154	.release = kvm_debugfs_release,
6155	.read = simple_attr_read,
6156	.write = simple_attr_write,
6157	.llseek = no_llseek,
6158};
6159
6160static int vm_stat_get(void *_offset, u64 *val)
6161{
6162	unsigned offset = (long)_offset;
6163	struct kvm *kvm;
6164	u64 tmp_val;
6165
6166	*val = 0;
6167	mutex_lock(&kvm_lock);
6168	list_for_each_entry(kvm, &vm_list, vm_list) {
6169		kvm_get_stat_per_vm(kvm, offset, &tmp_val);
6170		*val += tmp_val;
6171	}
6172	mutex_unlock(&kvm_lock);
6173	return 0;
6174}
6175
6176static int vm_stat_clear(void *_offset, u64 val)
6177{
6178	unsigned offset = (long)_offset;
6179	struct kvm *kvm;
6180
6181	if (val)
6182		return -EINVAL;
6183
6184	mutex_lock(&kvm_lock);
6185	list_for_each_entry(kvm, &vm_list, vm_list) {
6186		kvm_clear_stat_per_vm(kvm, offset);
6187	}
6188	mutex_unlock(&kvm_lock);
6189
6190	return 0;
6191}
6192
6193DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
6194DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
6195
6196static int vcpu_stat_get(void *_offset, u64 *val)
6197{
6198	unsigned offset = (long)_offset;
6199	struct kvm *kvm;
6200	u64 tmp_val;
6201
6202	*val = 0;
6203	mutex_lock(&kvm_lock);
6204	list_for_each_entry(kvm, &vm_list, vm_list) {
6205		kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
6206		*val += tmp_val;
6207	}
6208	mutex_unlock(&kvm_lock);
6209	return 0;
6210}
6211
6212static int vcpu_stat_clear(void *_offset, u64 val)
6213{
6214	unsigned offset = (long)_offset;
6215	struct kvm *kvm;
6216
6217	if (val)
6218		return -EINVAL;
6219
6220	mutex_lock(&kvm_lock);
6221	list_for_each_entry(kvm, &vm_list, vm_list) {
6222		kvm_clear_stat_per_vcpu(kvm, offset);
6223	}
6224	mutex_unlock(&kvm_lock);
6225
6226	return 0;
6227}
6228
6229DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
6230			"%llu\n");
6231DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
6232
6233static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
6234{
6235	struct kobj_uevent_env *env;
6236	unsigned long long created, active;
6237
6238	if (!kvm_dev.this_device || !kvm)
6239		return;
6240
6241	mutex_lock(&kvm_lock);
6242	if (type == KVM_EVENT_CREATE_VM) {
6243		kvm_createvm_count++;
6244		kvm_active_vms++;
6245	} else if (type == KVM_EVENT_DESTROY_VM) {
6246		kvm_active_vms--;
6247	}
6248	created = kvm_createvm_count;
6249	active = kvm_active_vms;
6250	mutex_unlock(&kvm_lock);
6251
6252	env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
6253	if (!env)
6254		return;
6255
6256	add_uevent_var(env, "CREATED=%llu", created);
6257	add_uevent_var(env, "COUNT=%llu", active);
6258
6259	if (type == KVM_EVENT_CREATE_VM) {
6260		add_uevent_var(env, "EVENT=create");
6261		kvm->userspace_pid = task_pid_nr(current);
6262	} else if (type == KVM_EVENT_DESTROY_VM) {
6263		add_uevent_var(env, "EVENT=destroy");
6264	}
6265	add_uevent_var(env, "PID=%d", kvm->userspace_pid);
6266
6267	if (!IS_ERR(kvm->debugfs_dentry)) {
6268		char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
6269
6270		if (p) {
6271			tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
6272			if (!IS_ERR(tmp))
6273				add_uevent_var(env, "STATS_PATH=%s", tmp);
6274			kfree(p);
6275		}
6276	}
6277	/* no need for checks, since we are adding at most only 5 keys */
6278	env->envp[env->envp_idx++] = NULL;
6279	kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
6280	kfree(env);
6281}
6282
6283static void kvm_init_debug(void)
6284{
6285	const struct file_operations *fops;
6286	const struct _kvm_stats_desc *pdesc;
6287	int i;
6288
6289	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
6290
6291	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
6292		pdesc = &kvm_vm_stats_desc[i];
6293		if (kvm_stats_debugfs_mode(pdesc) & 0222)
6294			fops = &vm_stat_fops;
6295		else
6296			fops = &vm_stat_readonly_fops;
6297		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
6298				kvm_debugfs_dir,
6299				(void *)(long)pdesc->desc.offset, fops);
6300	}
6301
6302	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
6303		pdesc = &kvm_vcpu_stats_desc[i];
6304		if (kvm_stats_debugfs_mode(pdesc) & 0222)
6305			fops = &vcpu_stat_fops;
6306		else
6307			fops = &vcpu_stat_readonly_fops;
6308		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
6309				kvm_debugfs_dir,
6310				(void *)(long)pdesc->desc.offset, fops);
6311	}
6312}
6313
6314static inline
6315struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
6316{
6317	return container_of(pn, struct kvm_vcpu, preempt_notifier);
6318}
6319
6320static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
6321{
6322	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
6323
6324	WRITE_ONCE(vcpu->preempted, false);
6325	WRITE_ONCE(vcpu->ready, false);
6326
6327	__this_cpu_write(kvm_running_vcpu, vcpu);
6328	kvm_arch_sched_in(vcpu, cpu);
6329	kvm_arch_vcpu_load(vcpu, cpu);
6330}
6331
6332static void kvm_sched_out(struct preempt_notifier *pn,
6333			  struct task_struct *next)
6334{
6335	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
6336
6337	if (current->on_rq) {
6338		WRITE_ONCE(vcpu->preempted, true);
6339		WRITE_ONCE(vcpu->ready, true);
6340	}
6341	kvm_arch_vcpu_put(vcpu);
6342	__this_cpu_write(kvm_running_vcpu, NULL);
6343}
6344
6345/**
6346 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
6347 *
6348 * We can disable preemption locally around accessing the per-CPU variable,
6349 * and use the resolved vcpu pointer after enabling preemption again,
6350 * because even if the current thread is migrated to another CPU, reading
6351 * the per-CPU value later will give us the same value as we update the
6352 * per-CPU variable in the preempt notifier handlers.
6353 */
6354struct kvm_vcpu *kvm_get_running_vcpu(void)
6355{
6356	struct kvm_vcpu *vcpu;
6357
6358	preempt_disable();
6359	vcpu = __this_cpu_read(kvm_running_vcpu);
6360	preempt_enable();
6361
6362	return vcpu;
6363}
6364EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
6365
6366/**
6367 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
6368 */
6369struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
6370{
6371        return &kvm_running_vcpu;
6372}
6373
6374#ifdef CONFIG_GUEST_PERF_EVENTS
6375static unsigned int kvm_guest_state(void)
6376{
6377	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
6378	unsigned int state;
6379
6380	if (!kvm_arch_pmi_in_guest(vcpu))
6381		return 0;
6382
6383	state = PERF_GUEST_ACTIVE;
6384	if (!kvm_arch_vcpu_in_kernel(vcpu))
6385		state |= PERF_GUEST_USER;
6386
6387	return state;
6388}
6389
6390static unsigned long kvm_guest_get_ip(void)
6391{
6392	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
6393
6394	/* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
6395	if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
6396		return 0;
6397
6398	return kvm_arch_vcpu_get_ip(vcpu);
6399}
6400
6401static struct perf_guest_info_callbacks kvm_guest_cbs = {
6402	.state			= kvm_guest_state,
6403	.get_ip			= kvm_guest_get_ip,
6404	.handle_intel_pt_intr	= NULL,
6405};
6406
6407void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
6408{
6409	kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
6410	perf_register_guest_info_callbacks(&kvm_guest_cbs);
6411}
6412void kvm_unregister_perf_callbacks(void)
6413{
6414	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6415}
6416#endif
6417
6418int kvm_init(unsigned vcpu_size, unsigned vcpu_align, struct module *module)
6419{
6420	int r;
6421	int cpu;
6422
6423#ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6424	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_ONLINE, "kvm/cpu:online",
6425				      kvm_online_cpu, kvm_offline_cpu);
6426	if (r)
6427		return r;
6428
6429	register_syscore_ops(&kvm_syscore_ops);
6430#endif
6431
6432	/* A kmem cache lets us meet the alignment requirements of fx_save. */
6433	if (!vcpu_align)
6434		vcpu_align = __alignof__(struct kvm_vcpu);
6435	kvm_vcpu_cache =
6436		kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
6437					   SLAB_ACCOUNT,
6438					   offsetof(struct kvm_vcpu, arch),
6439					   offsetofend(struct kvm_vcpu, stats_id)
6440					   - offsetof(struct kvm_vcpu, arch),
6441					   NULL);
6442	if (!kvm_vcpu_cache) {
6443		r = -ENOMEM;
6444		goto err_vcpu_cache;
6445	}
6446
6447	for_each_possible_cpu(cpu) {
6448		if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
6449					    GFP_KERNEL, cpu_to_node(cpu))) {
6450			r = -ENOMEM;
6451			goto err_cpu_kick_mask;
6452		}
6453	}
6454
6455	r = kvm_irqfd_init();
6456	if (r)
6457		goto err_irqfd;
6458
6459	r = kvm_async_pf_init();
6460	if (r)
6461		goto err_async_pf;
6462
6463	kvm_chardev_ops.owner = module;
6464	kvm_vm_fops.owner = module;
6465	kvm_vcpu_fops.owner = module;
6466	kvm_device_fops.owner = module;
6467
6468	kvm_preempt_ops.sched_in = kvm_sched_in;
6469	kvm_preempt_ops.sched_out = kvm_sched_out;
6470
6471	kvm_init_debug();
6472
6473	r = kvm_vfio_ops_init();
6474	if (WARN_ON_ONCE(r))
6475		goto err_vfio;
6476
6477	kvm_gmem_init(module);
6478
6479	/*
6480	 * Registration _must_ be the very last thing done, as this exposes
6481	 * /dev/kvm to userspace, i.e. all infrastructure must be setup!
6482	 */
6483	r = misc_register(&kvm_dev);
6484	if (r) {
6485		pr_err("kvm: misc device register failed\n");
6486		goto err_register;
6487	}
6488
6489	return 0;
6490
6491err_register:
6492	kvm_vfio_ops_exit();
6493err_vfio:
6494	kvm_async_pf_deinit();
6495err_async_pf:
6496	kvm_irqfd_exit();
6497err_irqfd:
6498err_cpu_kick_mask:
6499	for_each_possible_cpu(cpu)
6500		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6501	kmem_cache_destroy(kvm_vcpu_cache);
6502err_vcpu_cache:
6503#ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6504	unregister_syscore_ops(&kvm_syscore_ops);
6505	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6506#endif
6507	return r;
6508}
6509EXPORT_SYMBOL_GPL(kvm_init);
6510
6511void kvm_exit(void)
6512{
6513	int cpu;
6514
6515	/*
6516	 * Note, unregistering /dev/kvm doesn't strictly need to come first,
6517	 * fops_get(), a.k.a. try_module_get(), prevents acquiring references
6518	 * to KVM while the module is being stopped.
6519	 */
6520	misc_deregister(&kvm_dev);
6521
6522	debugfs_remove_recursive(kvm_debugfs_dir);
6523	for_each_possible_cpu(cpu)
6524		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6525	kmem_cache_destroy(kvm_vcpu_cache);
6526	kvm_vfio_ops_exit();
6527	kvm_async_pf_deinit();
6528#ifdef CONFIG_KVM_GENERIC_HARDWARE_ENABLING
6529	unregister_syscore_ops(&kvm_syscore_ops);
6530	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_ONLINE);
6531#endif
6532	kvm_irqfd_exit();
6533}
6534EXPORT_SYMBOL_GPL(kvm_exit);
6535
6536struct kvm_vm_worker_thread_context {
6537	struct kvm *kvm;
6538	struct task_struct *parent;
6539	struct completion init_done;
6540	kvm_vm_thread_fn_t thread_fn;
6541	uintptr_t data;
6542	int err;
6543};
6544
6545static int kvm_vm_worker_thread(void *context)
6546{
6547	/*
6548	 * The init_context is allocated on the stack of the parent thread, so
6549	 * we have to locally copy anything that is needed beyond initialization
6550	 */
6551	struct kvm_vm_worker_thread_context *init_context = context;
6552	struct task_struct *parent;
6553	struct kvm *kvm = init_context->kvm;
6554	kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6555	uintptr_t data = init_context->data;
6556	int err;
6557
6558	err = kthread_park(current);
6559	/* kthread_park(current) is never supposed to return an error */
6560	WARN_ON(err != 0);
6561	if (err)
6562		goto init_complete;
6563
6564	err = cgroup_attach_task_all(init_context->parent, current);
6565	if (err) {
6566		kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6567			__func__, err);
6568		goto init_complete;
6569	}
6570
6571	set_user_nice(current, task_nice(init_context->parent));
6572
6573init_complete:
6574	init_context->err = err;
6575	complete(&init_context->init_done);
6576	init_context = NULL;
6577
6578	if (err)
6579		goto out;
6580
6581	/* Wait to be woken up by the spawner before proceeding. */
6582	kthread_parkme();
6583
6584	if (!kthread_should_stop())
6585		err = thread_fn(kvm, data);
6586
6587out:
6588	/*
6589	 * Move kthread back to its original cgroup to prevent it lingering in
6590	 * the cgroup of the VM process, after the latter finishes its
6591	 * execution.
6592	 *
6593	 * kthread_stop() waits on the 'exited' completion condition which is
6594	 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6595	 * kthread is removed from the cgroup in the cgroup_exit() which is
6596	 * called after the exit_mm(). This causes the kthread_stop() to return
6597	 * before the kthread actually quits the cgroup.
6598	 */
6599	rcu_read_lock();
6600	parent = rcu_dereference(current->real_parent);
6601	get_task_struct(parent);
6602	rcu_read_unlock();
6603	cgroup_attach_task_all(parent, current);
6604	put_task_struct(parent);
6605
6606	return err;
6607}
6608
6609int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6610				uintptr_t data, const char *name,
6611				struct task_struct **thread_ptr)
6612{
6613	struct kvm_vm_worker_thread_context init_context = {};
6614	struct task_struct *thread;
6615
6616	*thread_ptr = NULL;
6617	init_context.kvm = kvm;
6618	init_context.parent = current;
6619	init_context.thread_fn = thread_fn;
6620	init_context.data = data;
6621	init_completion(&init_context.init_done);
6622
6623	thread = kthread_run(kvm_vm_worker_thread, &init_context,
6624			     "%s-%d", name, task_pid_nr(current));
6625	if (IS_ERR(thread))
6626		return PTR_ERR(thread);
6627
6628	/* kthread_run is never supposed to return NULL */
6629	WARN_ON(thread == NULL);
6630
6631	wait_for_completion(&init_context.init_done);
6632
6633	if (!init_context.err)
6634		*thread_ptr = thread;
6635
6636	return init_context.err;
6637}