1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Kernel-based Virtual Machine driver for Linux
   4 *
   5 * This module enables machines with Intel VT-x extensions to run virtual
   6 * machines without emulation or binary translation.
   7 *
   8 * Copyright (C) 2006 Qumranet, Inc.
   9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  10 *
  11 * Authors:
  12 *   Avi Kivity   <avi@qumranet.com>
  13 *   Yaniv Kamay  <yaniv@qumranet.com>
  14 */
  15
  16#include <kvm/iodev.h>
  17
  18#include <linux/kvm_host.h>
  19#include <linux/kvm.h>
  20#include <linux/module.h>
  21#include <linux/errno.h>
  22#include <linux/percpu.h>
  23#include <linux/mm.h>
  24#include <linux/miscdevice.h>
  25#include <linux/vmalloc.h>
  26#include <linux/reboot.h>
  27#include <linux/debugfs.h>
  28#include <linux/highmem.h>
  29#include <linux/file.h>
  30#include <linux/syscore_ops.h>
  31#include <linux/cpu.h>
  32#include <linux/sched/signal.h>
  33#include <linux/sched/mm.h>
  34#include <linux/sched/stat.h>
  35#include <linux/cpumask.h>
  36#include <linux/smp.h>
  37#include <linux/anon_inodes.h>
  38#include <linux/profile.h>
  39#include <linux/kvm_para.h>
  40#include <linux/pagemap.h>
  41#include <linux/mman.h>
  42#include <linux/swap.h>
  43#include <linux/bitops.h>
  44#include <linux/spinlock.h>
  45#include <linux/compat.h>
  46#include <linux/srcu.h>
  47#include <linux/hugetlb.h>
  48#include <linux/slab.h>
  49#include <linux/sort.h>
  50#include <linux/bsearch.h>
  51#include <linux/io.h>
  52#include <linux/lockdep.h>
  53#include <linux/kthread.h>
  54#include <linux/suspend.h>
  55
  56#include <asm/processor.h>
  57#include <asm/ioctl.h>
  58#include <linux/uaccess.h>
  59
  60#include "coalesced_mmio.h"
  61#include "async_pf.h"
  62#include "kvm_mm.h"
  63#include "vfio.h"
  64
 
 
  65#define CREATE_TRACE_POINTS
  66#include <trace/events/kvm.h>
  67
  68#include <linux/kvm_dirty_ring.h>
  69
 
  70/* Worst case buffer size needed for holding an integer. */
  71#define ITOA_MAX_LEN 12
  72
  73MODULE_AUTHOR("Qumranet");
  74MODULE_LICENSE("GPL");
  75
  76/* Architectures should define their poll value according to the halt latency */
  77unsigned int halt_poll_ns = KVM_HALT_POLL_NS_DEFAULT;
  78module_param(halt_poll_ns, uint, 0644);
  79EXPORT_SYMBOL_GPL(halt_poll_ns);
  80
  81/* Default doubles per-vcpu halt_poll_ns. */
  82unsigned int halt_poll_ns_grow = 2;
  83module_param(halt_poll_ns_grow, uint, 0644);
  84EXPORT_SYMBOL_GPL(halt_poll_ns_grow);
  85
  86/* The start value to grow halt_poll_ns from */
  87unsigned int halt_poll_ns_grow_start = 10000; /* 10us */
  88module_param(halt_poll_ns_grow_start, uint, 0644);
  89EXPORT_SYMBOL_GPL(halt_poll_ns_grow_start);
  90
  91/* Default resets per-vcpu halt_poll_ns . */
  92unsigned int halt_poll_ns_shrink;
  93module_param(halt_poll_ns_shrink, uint, 0644);
  94EXPORT_SYMBOL_GPL(halt_poll_ns_shrink);
  95
  96/*
  97 * Ordering of locks:
  98 *
  99 *	kvm->lock --> kvm->slots_lock --> kvm->irq_lock
 100 */
 101
 102DEFINE_MUTEX(kvm_lock);
 103static DEFINE_RAW_SPINLOCK(kvm_count_lock);
 104LIST_HEAD(vm_list);
 105
 106static cpumask_var_t cpus_hardware_enabled;
 107static int kvm_usage_count;
 108static atomic_t hardware_enable_failed;
 109
 110static struct kmem_cache *kvm_vcpu_cache;
 111
 112static __read_mostly struct preempt_ops kvm_preempt_ops;
 113static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_running_vcpu);
 114
 115struct dentry *kvm_debugfs_dir;
 116EXPORT_SYMBOL_GPL(kvm_debugfs_dir);
 117
 118static const struct file_operations stat_fops_per_vm;
 119
 120static struct file_operations kvm_chardev_ops;
 121
 122static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
 123			   unsigned long arg);
 124#ifdef CONFIG_KVM_COMPAT
 125static long kvm_vcpu_compat_ioctl(struct file *file, unsigned int ioctl,
 126				  unsigned long arg);
 127#define KVM_COMPAT(c)	.compat_ioctl	= (c)
 128#else
 129/*
 130 * For architectures that don't implement a compat infrastructure,
 131 * adopt a double line of defense:
 132 * - Prevent a compat task from opening /dev/kvm
 133 * - If the open has been done by a 64bit task, and the KVM fd
 134 *   passed to a compat task, let the ioctls fail.
 135 */
 136static long kvm_no_compat_ioctl(struct file *file, unsigned int ioctl,
 137				unsigned long arg) { return -EINVAL; }
 138
 139static int kvm_no_compat_open(struct inode *inode, struct file *file)
 140{
 141	return is_compat_task() ? -ENODEV : 0;
 142}
 143#define KVM_COMPAT(c)	.compat_ioctl	= kvm_no_compat_ioctl,	\
 144			.open		= kvm_no_compat_open
 145#endif
 146static int hardware_enable_all(void);
 147static void hardware_disable_all(void);
 148
 149static void kvm_io_bus_destroy(struct kvm_io_bus *bus);
 150
 151__visible bool kvm_rebooting;
 152EXPORT_SYMBOL_GPL(kvm_rebooting);
 153
 154#define KVM_EVENT_CREATE_VM 0
 155#define KVM_EVENT_DESTROY_VM 1
 156static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm);
 157static unsigned long long kvm_createvm_count;
 158static unsigned long long kvm_active_vms;
 159
 160static DEFINE_PER_CPU(cpumask_var_t, cpu_kick_mask);
 161
 162__weak void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
 163						   unsigned long start, unsigned long end)
 164{
 165}
 166
 167__weak void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
 168{
 169}
 170
 171bool kvm_is_zone_device_page(struct page *page)
 172{
 173	/*
 174	 * The metadata used by is_zone_device_page() to determine whether or
 175	 * not a page is ZONE_DEVICE is guaranteed to be valid if and only if
 176	 * the device has been pinned, e.g. by get_user_pages().  WARN if the
 177	 * page_count() is zero to help detect bad usage of this helper.
 178	 */
 179	if (WARN_ON_ONCE(!page_count(page)))
 180		return false;
 181
 182	return is_zone_device_page(page);
 183}
 184
 185/*
 186 * Returns a 'struct page' if the pfn is "valid" and backed by a refcounted
 187 * page, NULL otherwise.  Note, the list of refcounted PG_reserved page types
 188 * is likely incomplete, it has been compiled purely through people wanting to
 189 * back guest with a certain type of memory and encountering issues.
 190 */
 191struct page *kvm_pfn_to_refcounted_page(kvm_pfn_t pfn)
 192{
 193	struct page *page;
 194
 195	if (!pfn_valid(pfn))
 196		return NULL;
 197
 198	page = pfn_to_page(pfn);
 199	if (!PageReserved(page))
 200		return page;
 201
 202	/* The ZERO_PAGE(s) is marked PG_reserved, but is refcounted. */
 203	if (is_zero_pfn(pfn))
 204		return page;
 205
 206	/*
 207	 * ZONE_DEVICE pages currently set PG_reserved, but from a refcounting
 208	 * perspective they are "normal" pages, albeit with slightly different
 209	 * usage rules.
 210	 */
 211	if (kvm_is_zone_device_page(page))
 212		return page;
 213
 214	return NULL;
 215}
 216
 217/*
 218 * Switches to specified vcpu, until a matching vcpu_put()
 219 */
 220void vcpu_load(struct kvm_vcpu *vcpu)
 221{
 222	int cpu = get_cpu();
 223
 224	__this_cpu_write(kvm_running_vcpu, vcpu);
 225	preempt_notifier_register(&vcpu->preempt_notifier);
 226	kvm_arch_vcpu_load(vcpu, cpu);
 227	put_cpu();
 228}
 229EXPORT_SYMBOL_GPL(vcpu_load);
 230
 231void vcpu_put(struct kvm_vcpu *vcpu)
 232{
 233	preempt_disable();
 234	kvm_arch_vcpu_put(vcpu);
 235	preempt_notifier_unregister(&vcpu->preempt_notifier);
 236	__this_cpu_write(kvm_running_vcpu, NULL);
 237	preempt_enable();
 238}
 239EXPORT_SYMBOL_GPL(vcpu_put);
 240
 241/* TODO: merge with kvm_arch_vcpu_should_kick */
 242static bool kvm_request_needs_ipi(struct kvm_vcpu *vcpu, unsigned req)
 243{
 244	int mode = kvm_vcpu_exiting_guest_mode(vcpu);
 245
 246	/*
 247	 * We need to wait for the VCPU to reenable interrupts and get out of
 248	 * READING_SHADOW_PAGE_TABLES mode.
 249	 */
 250	if (req & KVM_REQUEST_WAIT)
 251		return mode != OUTSIDE_GUEST_MODE;
 252
 253	/*
 254	 * Need to kick a running VCPU, but otherwise there is nothing to do.
 255	 */
 256	return mode == IN_GUEST_MODE;
 257}
 258
 259static void ack_kick(void *_completed)
 260{
 261}
 262
 263static inline bool kvm_kick_many_cpus(struct cpumask *cpus, bool wait)
 264{
 265	if (cpumask_empty(cpus))
 266		return false;
 267
 268	smp_call_function_many(cpus, ack_kick, NULL, wait);
 269	return true;
 270}
 271
 272static void kvm_make_vcpu_request(struct kvm_vcpu *vcpu, unsigned int req,
 273				  struct cpumask *tmp, int current_cpu)
 274{
 275	int cpu;
 276
 277	if (likely(!(req & KVM_REQUEST_NO_ACTION)))
 278		__kvm_make_request(req, vcpu);
 279
 280	if (!(req & KVM_REQUEST_NO_WAKEUP) && kvm_vcpu_wake_up(vcpu))
 281		return;
 282
 283	/*
 284	 * Note, the vCPU could get migrated to a different pCPU at any point
 285	 * after kvm_request_needs_ipi(), which could result in sending an IPI
 286	 * to the previous pCPU.  But, that's OK because the purpose of the IPI
 287	 * is to ensure the vCPU returns to OUTSIDE_GUEST_MODE, which is
 288	 * satisfied if the vCPU migrates. Entering READING_SHADOW_PAGE_TABLES
 289	 * after this point is also OK, as the requirement is only that KVM wait
 290	 * for vCPUs that were reading SPTEs _before_ any changes were
 291	 * finalized. See kvm_vcpu_kick() for more details on handling requests.
 292	 */
 293	if (kvm_request_needs_ipi(vcpu, req)) {
 294		cpu = READ_ONCE(vcpu->cpu);
 295		if (cpu != -1 && cpu != current_cpu)
 296			__cpumask_set_cpu(cpu, tmp);
 297	}
 298}
 299
 300bool kvm_make_vcpus_request_mask(struct kvm *kvm, unsigned int req,
 301				 unsigned long *vcpu_bitmap)
 302{
 303	struct kvm_vcpu *vcpu;
 304	struct cpumask *cpus;
 305	int i, me;
 306	bool called;
 307
 308	me = get_cpu();
 309
 310	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
 311	cpumask_clear(cpus);
 312
 313	for_each_set_bit(i, vcpu_bitmap, KVM_MAX_VCPUS) {
 314		vcpu = kvm_get_vcpu(kvm, i);
 315		if (!vcpu)
 316			continue;
 317		kvm_make_vcpu_request(vcpu, req, cpus, me);
 318	}
 319
 320	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
 321	put_cpu();
 322
 323	return called;
 324}
 325
 326bool kvm_make_all_cpus_request_except(struct kvm *kvm, unsigned int req,
 327				      struct kvm_vcpu *except)
 328{
 329	struct kvm_vcpu *vcpu;
 330	struct cpumask *cpus;
 331	unsigned long i;
 332	bool called;
 333	int me;
 334
 335	me = get_cpu();
 336
 337	cpus = this_cpu_cpumask_var_ptr(cpu_kick_mask);
 338	cpumask_clear(cpus);
 339
 340	kvm_for_each_vcpu(i, vcpu, kvm) {
 341		if (vcpu == except)
 342			continue;
 343		kvm_make_vcpu_request(vcpu, req, cpus, me);
 344	}
 345
 346	called = kvm_kick_many_cpus(cpus, !!(req & KVM_REQUEST_WAIT));
 347	put_cpu();
 348
 349	return called;
 350}
 351
 352bool kvm_make_all_cpus_request(struct kvm *kvm, unsigned int req)
 353{
 354	return kvm_make_all_cpus_request_except(kvm, req, NULL);
 355}
 356EXPORT_SYMBOL_GPL(kvm_make_all_cpus_request);
 357
 358#ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
 359void kvm_flush_remote_tlbs(struct kvm *kvm)
 360{
 361	++kvm->stat.generic.remote_tlb_flush_requests;
 362
 363	/*
 364	 * We want to publish modifications to the page tables before reading
 365	 * mode. Pairs with a memory barrier in arch-specific code.
 366	 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
 367	 * and smp_mb in walk_shadow_page_lockless_begin/end.
 368	 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
 369	 *
 370	 * There is already an smp_mb__after_atomic() before
 371	 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
 372	 * barrier here.
 373	 */
 374	if (!kvm_arch_flush_remote_tlb(kvm)
 375	    || kvm_make_all_cpus_request(kvm, KVM_REQ_TLB_FLUSH))
 376		++kvm->stat.generic.remote_tlb_flush;
 377}
 378EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs);
 379#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 380
 381static void kvm_flush_shadow_all(struct kvm *kvm)
 382{
 383	kvm_arch_flush_shadow_all(kvm);
 384	kvm_arch_guest_memory_reclaimed(kvm);
 385}
 386
 387#ifdef KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE
 388static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
 389					       gfp_t gfp_flags)
 390{
 391	gfp_flags |= mc->gfp_zero;
 392
 393	if (mc->kmem_cache)
 394		return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
 395	else
 396		return (void *)__get_free_page(gfp_flags);
 397}
 398
 399int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
 400{
 401	gfp_t gfp = mc->gfp_custom ? mc->gfp_custom : GFP_KERNEL_ACCOUNT;
 402	void *obj;
 403
 404	if (mc->nobjs >= min)
 405		return 0;
 406
 407	if (unlikely(!mc->objects)) {
 408		if (WARN_ON_ONCE(!capacity))
 409			return -EIO;
 410
 411		mc->objects = kvmalloc_array(sizeof(void *), capacity, gfp);
 412		if (!mc->objects)
 413			return -ENOMEM;
 414
 415		mc->capacity = capacity;
 416	}
 417
 418	/* It is illegal to request a different capacity across topups. */
 419	if (WARN_ON_ONCE(mc->capacity != capacity))
 420		return -EIO;
 421
 422	while (mc->nobjs < mc->capacity) {
 423		obj = mmu_memory_cache_alloc_obj(mc, gfp);
 424		if (!obj)
 425			return mc->nobjs >= min ? 0 : -ENOMEM;
 426		mc->objects[mc->nobjs++] = obj;
 427	}
 428	return 0;
 429}
 430
 431int kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int min)
 432{
 433	return __kvm_mmu_topup_memory_cache(mc, KVM_ARCH_NR_OBJS_PER_MEMORY_CACHE, min);
 434}
 435
 436int kvm_mmu_memory_cache_nr_free_objects(struct kvm_mmu_memory_cache *mc)
 437{
 438	return mc->nobjs;
 439}
 440
 441void kvm_mmu_free_memory_cache(struct kvm_mmu_memory_cache *mc)
 442{
 443	while (mc->nobjs) {
 444		if (mc->kmem_cache)
 445			kmem_cache_free(mc->kmem_cache, mc->objects[--mc->nobjs]);
 446		else
 447			free_page((unsigned long)mc->objects[--mc->nobjs]);
 448	}
 449
 450	kvfree(mc->objects);
 451
 452	mc->objects = NULL;
 453	mc->capacity = 0;
 454}
 455
 456void *kvm_mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
 457{
 458	void *p;
 459
 460	if (WARN_ON(!mc->nobjs))
 461		p = mmu_memory_cache_alloc_obj(mc, GFP_ATOMIC | __GFP_ACCOUNT);
 462	else
 463		p = mc->objects[--mc->nobjs];
 464	BUG_ON(!p);
 465	return p;
 466}
 467#endif
 468
 469static void kvm_vcpu_init(struct kvm_vcpu *vcpu, struct kvm *kvm, unsigned id)
 470{
 471	mutex_init(&vcpu->mutex);
 472	vcpu->cpu = -1;
 473	vcpu->kvm = kvm;
 474	vcpu->vcpu_id = id;
 475	vcpu->pid = NULL;
 476#ifndef __KVM_HAVE_ARCH_WQP
 477	rcuwait_init(&vcpu->wait);
 478#endif
 479	kvm_async_pf_vcpu_init(vcpu);
 480
 481	kvm_vcpu_set_in_spin_loop(vcpu, false);
 482	kvm_vcpu_set_dy_eligible(vcpu, false);
 483	vcpu->preempted = false;
 484	vcpu->ready = false;
 485	preempt_notifier_init(&vcpu->preempt_notifier, &kvm_preempt_ops);
 486	vcpu->last_used_slot = NULL;
 487
 488	/* Fill the stats id string for the vcpu */
 489	snprintf(vcpu->stats_id, sizeof(vcpu->stats_id), "kvm-%d/vcpu-%d",
 490		 task_pid_nr(current), id);
 491}
 492
 493static void kvm_vcpu_destroy(struct kvm_vcpu *vcpu)
 494{
 495	kvm_arch_vcpu_destroy(vcpu);
 496	kvm_dirty_ring_free(&vcpu->dirty_ring);
 497
 498	/*
 499	 * No need for rcu_read_lock as VCPU_RUN is the only place that changes
 500	 * the vcpu->pid pointer, and at destruction time all file descriptors
 501	 * are already gone.
 502	 */
 503	put_pid(rcu_dereference_protected(vcpu->pid, 1));
 504
 505	free_page((unsigned long)vcpu->run);
 506	kmem_cache_free(kvm_vcpu_cache, vcpu);
 507}
 508
 509void kvm_destroy_vcpus(struct kvm *kvm)
 510{
 511	unsigned long i;
 512	struct kvm_vcpu *vcpu;
 513
 514	kvm_for_each_vcpu(i, vcpu, kvm) {
 515		kvm_vcpu_destroy(vcpu);
 516		xa_erase(&kvm->vcpu_array, i);
 517	}
 518
 519	atomic_set(&kvm->online_vcpus, 0);
 520}
 521EXPORT_SYMBOL_GPL(kvm_destroy_vcpus);
 522
 523#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
 524static inline struct kvm *mmu_notifier_to_kvm(struct mmu_notifier *mn)
 525{
 526	return container_of(mn, struct kvm, mmu_notifier);
 527}
 528
 529static void kvm_mmu_notifier_invalidate_range(struct mmu_notifier *mn,
 530					      struct mm_struct *mm,
 531					      unsigned long start, unsigned long end)
 532{
 533	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 534	int idx;
 535
 536	idx = srcu_read_lock(&kvm->srcu);
 537	kvm_arch_mmu_notifier_invalidate_range(kvm, start, end);
 538	srcu_read_unlock(&kvm->srcu, idx);
 539}
 540
 541typedef bool (*hva_handler_t)(struct kvm *kvm, struct kvm_gfn_range *range);
 542
 543typedef void (*on_lock_fn_t)(struct kvm *kvm, unsigned long start,
 544			     unsigned long end);
 545
 546typedef void (*on_unlock_fn_t)(struct kvm *kvm);
 547
 548struct kvm_hva_range {
 549	unsigned long start;
 550	unsigned long end;
 551	pte_t pte;
 552	hva_handler_t handler;
 
 
 553	on_lock_fn_t on_lock;
 554	on_unlock_fn_t on_unlock;
 555	bool flush_on_ret;
 556	bool may_block;
 557};
 558
 559/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 560 * Use a dedicated stub instead of NULL to indicate that there is no callback
 561 * function/handler.  The compiler technically can't guarantee that a real
 562 * function will have a non-zero address, and so it will generate code to
 563 * check for !NULL, whereas comparing against a stub will be elided at compile
 564 * time (unless the compiler is getting long in the tooth, e.g. gcc 4.9).
 565 */
 566static void kvm_null_fn(void)
 567{
 568
 569}
 570#define IS_KVM_NULL_FN(fn) ((fn) == (void *)kvm_null_fn)
 571
 
 
 572/* Iterate over each memslot intersecting [start, last] (inclusive) range */
 573#define kvm_for_each_memslot_in_hva_range(node, slots, start, last)	     \
 574	for (node = interval_tree_iter_first(&slots->hva_tree, start, last); \
 575	     node;							     \
 576	     node = interval_tree_iter_next(node, start, last))	     \
 577
 578static __always_inline int __kvm_handle_hva_range(struct kvm *kvm,
 579						  const struct kvm_hva_range *range)
 580{
 581	bool ret = false, locked = false;
 
 
 
 582	struct kvm_gfn_range gfn_range;
 583	struct kvm_memory_slot *slot;
 584	struct kvm_memslots *slots;
 585	int i, idx;
 586
 587	if (WARN_ON_ONCE(range->end <= range->start))
 588		return 0;
 589
 590	/* A null handler is allowed if and only if on_lock() is provided. */
 591	if (WARN_ON_ONCE(IS_KVM_NULL_FN(range->on_lock) &&
 592			 IS_KVM_NULL_FN(range->handler)))
 593		return 0;
 594
 595	idx = srcu_read_lock(&kvm->srcu);
 596
 597	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
 598		struct interval_tree_node *node;
 599
 600		slots = __kvm_memslots(kvm, i);
 601		kvm_for_each_memslot_in_hva_range(node, slots,
 602						  range->start, range->end - 1) {
 603			unsigned long hva_start, hva_end;
 604
 605			slot = container_of(node, struct kvm_memory_slot, hva_node[slots->node_idx]);
 606			hva_start = max(range->start, slot->userspace_addr);
 607			hva_end = min(range->end, slot->userspace_addr +
 608						  (slot->npages << PAGE_SHIFT));
 609
 610			/*
 611			 * To optimize for the likely case where the address
 612			 * range is covered by zero or one memslots, don't
 613			 * bother making these conditional (to avoid writes on
 614			 * the second or later invocation of the handler).
 615			 */
 616			gfn_range.pte = range->pte;
 617			gfn_range.may_block = range->may_block;
 618
 619			/*
 620			 * {gfn(page) | page intersects with [hva_start, hva_end)} =
 621			 * {gfn_start, gfn_start+1, ..., gfn_end-1}.
 622			 */
 623			gfn_range.start = hva_to_gfn_memslot(hva_start, slot);
 624			gfn_range.end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, slot);
 625			gfn_range.slot = slot;
 626
 627			if (!locked) {
 628				locked = true;
 629				KVM_MMU_LOCK(kvm);
 630				if (!IS_KVM_NULL_FN(range->on_lock))
 631					range->on_lock(kvm, range->start, range->end);
 
 632				if (IS_KVM_NULL_FN(range->handler))
 633					break;
 634			}
 635			ret |= range->handler(kvm, &gfn_range);
 636		}
 637	}
 638
 639	if (range->flush_on_ret && ret)
 640		kvm_flush_remote_tlbs(kvm);
 641
 642	if (locked) {
 643		KVM_MMU_UNLOCK(kvm);
 644		if (!IS_KVM_NULL_FN(range->on_unlock))
 645			range->on_unlock(kvm);
 646	}
 647
 648	srcu_read_unlock(&kvm->srcu, idx);
 649
 650	/* The notifiers are averse to booleans. :-( */
 651	return (int)ret;
 652}
 653
 654static __always_inline int kvm_handle_hva_range(struct mmu_notifier *mn,
 655						unsigned long start,
 656						unsigned long end,
 657						pte_t pte,
 658						hva_handler_t handler)
 659{
 660	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 661	const struct kvm_hva_range range = {
 662		.start		= start,
 663		.end		= end,
 664		.pte		= pte,
 665		.handler	= handler,
 666		.on_lock	= (void *)kvm_null_fn,
 667		.on_unlock	= (void *)kvm_null_fn,
 668		.flush_on_ret	= true,
 669		.may_block	= false,
 670	};
 671
 672	return __kvm_handle_hva_range(kvm, &range);
 673}
 674
 675static __always_inline int kvm_handle_hva_range_no_flush(struct mmu_notifier *mn,
 676							 unsigned long start,
 677							 unsigned long end,
 678							 hva_handler_t handler)
 679{
 680	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 681	const struct kvm_hva_range range = {
 682		.start		= start,
 683		.end		= end,
 684		.pte		= __pte(0),
 685		.handler	= handler,
 686		.on_lock	= (void *)kvm_null_fn,
 687		.on_unlock	= (void *)kvm_null_fn,
 688		.flush_on_ret	= false,
 689		.may_block	= false,
 690	};
 691
 692	return __kvm_handle_hva_range(kvm, &range);
 693}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 694static void kvm_mmu_notifier_change_pte(struct mmu_notifier *mn,
 695					struct mm_struct *mm,
 696					unsigned long address,
 697					pte_t pte)
 698{
 699	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 
 700
 701	trace_kvm_set_spte_hva(address);
 702
 703	/*
 704	 * .change_pte() must be surrounded by .invalidate_range_{start,end}().
 705	 * If mmu_invalidate_in_progress is zero, then no in-progress
 706	 * invalidations, including this one, found a relevant memslot at
 707	 * start(); rechecking memslots here is unnecessary.  Note, a false
 708	 * positive (count elevated by a different invalidation) is sub-optimal
 709	 * but functionally ok.
 710	 */
 711	WARN_ON_ONCE(!READ_ONCE(kvm->mn_active_invalidate_count));
 712	if (!READ_ONCE(kvm->mmu_invalidate_in_progress))
 713		return;
 714
 715	kvm_handle_hva_range(mn, address, address + 1, pte, kvm_set_spte_gfn);
 716}
 717
 718void kvm_mmu_invalidate_begin(struct kvm *kvm, unsigned long start,
 719			      unsigned long end)
 720{
 
 721	/*
 722	 * The count increase must become visible at unlock time as no
 723	 * spte can be established without taking the mmu_lock and
 724	 * count is also read inside the mmu_lock critical section.
 725	 */
 726	kvm->mmu_invalidate_in_progress++;
 
 727	if (likely(kvm->mmu_invalidate_in_progress == 1)) {
 
 
 
 
 
 
 
 
 
 
 
 
 728		kvm->mmu_invalidate_range_start = start;
 729		kvm->mmu_invalidate_range_end = end;
 730	} else {
 731		/*
 732		 * Fully tracking multiple concurrent ranges has diminishing
 733		 * returns. Keep things simple and just find the minimal range
 734		 * which includes the current and new ranges. As there won't be
 735		 * enough information to subtract a range after its invalidate
 736		 * completes, any ranges invalidated concurrently will
 737		 * accumulate and persist until all outstanding invalidates
 738		 * complete.
 739		 */
 740		kvm->mmu_invalidate_range_start =
 741			min(kvm->mmu_invalidate_range_start, start);
 742		kvm->mmu_invalidate_range_end =
 743			max(kvm->mmu_invalidate_range_end, end);
 744	}
 745}
 746
 
 
 
 
 
 
 747static int kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier *mn,
 748					const struct mmu_notifier_range *range)
 749{
 750	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 751	const struct kvm_hva_range hva_range = {
 752		.start		= range->start,
 753		.end		= range->end,
 754		.pte		= __pte(0),
 755		.handler	= kvm_unmap_gfn_range,
 756		.on_lock	= kvm_mmu_invalidate_begin,
 757		.on_unlock	= kvm_arch_guest_memory_reclaimed,
 758		.flush_on_ret	= true,
 759		.may_block	= mmu_notifier_range_blockable(range),
 760	};
 761
 762	trace_kvm_unmap_hva_range(range->start, range->end);
 763
 764	/*
 765	 * Prevent memslot modification between range_start() and range_end()
 766	 * so that conditionally locking provides the same result in both
 767	 * functions.  Without that guarantee, the mmu_invalidate_in_progress
 768	 * adjustments will be imbalanced.
 769	 *
 770	 * Pairs with the decrement in range_end().
 771	 */
 772	spin_lock(&kvm->mn_invalidate_lock);
 773	kvm->mn_active_invalidate_count++;
 774	spin_unlock(&kvm->mn_invalidate_lock);
 775
 776	/*
 777	 * Invalidate pfn caches _before_ invalidating the secondary MMUs, i.e.
 778	 * before acquiring mmu_lock, to avoid holding mmu_lock while acquiring
 779	 * each cache's lock.  There are relatively few caches in existence at
 780	 * any given time, and the caches themselves can check for hva overlap,
 781	 * i.e. don't need to rely on memslot overlap checks for performance.
 782	 * Because this runs without holding mmu_lock, the pfn caches must use
 783	 * mn_active_invalidate_count (see above) instead of
 784	 * mmu_invalidate_in_progress.
 785	 */
 786	gfn_to_pfn_cache_invalidate_start(kvm, range->start, range->end,
 787					  hva_range.may_block);
 788
 789	__kvm_handle_hva_range(kvm, &hva_range);
 
 
 
 
 
 
 790
 791	return 0;
 792}
 793
 794void kvm_mmu_invalidate_end(struct kvm *kvm, unsigned long start,
 795			    unsigned long end)
 796{
 
 
 797	/*
 798	 * This sequence increase will notify the kvm page fault that
 799	 * the page that is going to be mapped in the spte could have
 800	 * been freed.
 801	 */
 802	kvm->mmu_invalidate_seq++;
 803	smp_wmb();
 804	/*
 805	 * The above sequence increase must be visible before the
 806	 * below count decrease, which is ensured by the smp_wmb above
 807	 * in conjunction with the smp_rmb in mmu_invalidate_retry().
 808	 */
 809	kvm->mmu_invalidate_in_progress--;
 
 
 
 
 
 
 
 810}
 811
 812static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier *mn,
 813					const struct mmu_notifier_range *range)
 814{
 815	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 816	const struct kvm_hva_range hva_range = {
 817		.start		= range->start,
 818		.end		= range->end,
 819		.pte		= __pte(0),
 820		.handler	= (void *)kvm_null_fn,
 821		.on_lock	= kvm_mmu_invalidate_end,
 822		.on_unlock	= (void *)kvm_null_fn,
 823		.flush_on_ret	= false,
 824		.may_block	= mmu_notifier_range_blockable(range),
 825	};
 826	bool wake;
 827
 828	__kvm_handle_hva_range(kvm, &hva_range);
 829
 830	/* Pairs with the increment in range_start(). */
 831	spin_lock(&kvm->mn_invalidate_lock);
 832	wake = (--kvm->mn_active_invalidate_count == 0);
 
 
 833	spin_unlock(&kvm->mn_invalidate_lock);
 834
 835	/*
 836	 * There can only be one waiter, since the wait happens under
 837	 * slots_lock.
 838	 */
 839	if (wake)
 840		rcuwait_wake_up(&kvm->mn_memslots_update_rcuwait);
 841
 842	BUG_ON(kvm->mmu_invalidate_in_progress < 0);
 843}
 844
 845static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier *mn,
 846					      struct mm_struct *mm,
 847					      unsigned long start,
 848					      unsigned long end)
 849{
 850	trace_kvm_age_hva(start, end);
 851
 852	return kvm_handle_hva_range(mn, start, end, __pte(0), kvm_age_gfn);
 
 853}
 854
 855static int kvm_mmu_notifier_clear_young(struct mmu_notifier *mn,
 856					struct mm_struct *mm,
 857					unsigned long start,
 858					unsigned long end)
 859{
 860	trace_kvm_age_hva(start, end);
 861
 862	/*
 863	 * Even though we do not flush TLB, this will still adversely
 864	 * affect performance on pre-Haswell Intel EPT, where there is
 865	 * no EPT Access Bit to clear so that we have to tear down EPT
 866	 * tables instead. If we find this unacceptable, we can always
 867	 * add a parameter to kvm_age_hva so that it effectively doesn't
 868	 * do anything on clear_young.
 869	 *
 870	 * Also note that currently we never issue secondary TLB flushes
 871	 * from clear_young, leaving this job up to the regular system
 872	 * cadence. If we find this inaccurate, we might come up with a
 873	 * more sophisticated heuristic later.
 874	 */
 875	return kvm_handle_hva_range_no_flush(mn, start, end, kvm_age_gfn);
 876}
 877
 878static int kvm_mmu_notifier_test_young(struct mmu_notifier *mn,
 879				       struct mm_struct *mm,
 880				       unsigned long address)
 881{
 882	trace_kvm_test_age_hva(address);
 883
 884	return kvm_handle_hva_range_no_flush(mn, address, address + 1,
 885					     kvm_test_age_gfn);
 886}
 887
 888static void kvm_mmu_notifier_release(struct mmu_notifier *mn,
 889				     struct mm_struct *mm)
 890{
 891	struct kvm *kvm = mmu_notifier_to_kvm(mn);
 892	int idx;
 893
 894	idx = srcu_read_lock(&kvm->srcu);
 895	kvm_flush_shadow_all(kvm);
 896	srcu_read_unlock(&kvm->srcu, idx);
 897}
 898
 899static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
 900	.invalidate_range	= kvm_mmu_notifier_invalidate_range,
 901	.invalidate_range_start	= kvm_mmu_notifier_invalidate_range_start,
 902	.invalidate_range_end	= kvm_mmu_notifier_invalidate_range_end,
 903	.clear_flush_young	= kvm_mmu_notifier_clear_flush_young,
 904	.clear_young		= kvm_mmu_notifier_clear_young,
 905	.test_young		= kvm_mmu_notifier_test_young,
 906	.change_pte		= kvm_mmu_notifier_change_pte,
 907	.release		= kvm_mmu_notifier_release,
 908};
 909
 910static int kvm_init_mmu_notifier(struct kvm *kvm)
 911{
 912	kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
 913	return mmu_notifier_register(&kvm->mmu_notifier, current->mm);
 914}
 915
 916#else  /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
 917
 918static int kvm_init_mmu_notifier(struct kvm *kvm)
 919{
 920	return 0;
 921}
 922
 923#endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
 924
 925#ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
 926static int kvm_pm_notifier_call(struct notifier_block *bl,
 927				unsigned long state,
 928				void *unused)
 929{
 930	struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
 931
 932	return kvm_arch_pm_notifier(kvm, state);
 933}
 934
 935static void kvm_init_pm_notifier(struct kvm *kvm)
 936{
 937	kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
 938	/* Suspend KVM before we suspend ftrace, RCU, etc. */
 939	kvm->pm_notifier.priority = INT_MAX;
 940	register_pm_notifier(&kvm->pm_notifier);
 941}
 942
 943static void kvm_destroy_pm_notifier(struct kvm *kvm)
 944{
 945	unregister_pm_notifier(&kvm->pm_notifier);
 946}
 947#else /* !CONFIG_HAVE_KVM_PM_NOTIFIER */
 948static void kvm_init_pm_notifier(struct kvm *kvm)
 949{
 950}
 951
 952static void kvm_destroy_pm_notifier(struct kvm *kvm)
 953{
 954}
 955#endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
 956
 957static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot *memslot)
 958{
 959	if (!memslot->dirty_bitmap)
 960		return;
 961
 962	kvfree(memslot->dirty_bitmap);
 963	memslot->dirty_bitmap = NULL;
 964}
 965
 966/* This does not remove the slot from struct kvm_memslots data structures */
 967static void kvm_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
 968{
 
 
 
 969	kvm_destroy_dirty_bitmap(slot);
 970
 971	kvm_arch_free_memslot(kvm, slot);
 972
 973	kfree(slot);
 974}
 975
 976static void kvm_free_memslots(struct kvm *kvm, struct kvm_memslots *slots)
 977{
 978	struct hlist_node *idnode;
 979	struct kvm_memory_slot *memslot;
 980	int bkt;
 981
 982	/*
 983	 * The same memslot objects live in both active and inactive sets,
 984	 * arbitrarily free using index '1' so the second invocation of this
 985	 * function isn't operating over a structure with dangling pointers
 986	 * (even though this function isn't actually touching them).
 987	 */
 988	if (!slots->node_idx)
 989		return;
 990
 991	hash_for_each_safe(slots->id_hash, bkt, idnode, memslot, id_node[1])
 992		kvm_free_memslot(kvm, memslot);
 993}
 994
 995static umode_t kvm_stats_debugfs_mode(const struct _kvm_stats_desc *pdesc)
 996{
 997	switch (pdesc->desc.flags & KVM_STATS_TYPE_MASK) {
 998	case KVM_STATS_TYPE_INSTANT:
 999		return 0444;
1000	case KVM_STATS_TYPE_CUMULATIVE:
1001	case KVM_STATS_TYPE_PEAK:
1002	default:
1003		return 0644;
1004	}
1005}
1006
1007
1008static void kvm_destroy_vm_debugfs(struct kvm *kvm)
1009{
1010	int i;
1011	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1012				      kvm_vcpu_stats_header.num_desc;
1013
1014	if (IS_ERR(kvm->debugfs_dentry))
1015		return;
1016
1017	debugfs_remove_recursive(kvm->debugfs_dentry);
1018
1019	if (kvm->debugfs_stat_data) {
1020		for (i = 0; i < kvm_debugfs_num_entries; i++)
1021			kfree(kvm->debugfs_stat_data[i]);
1022		kfree(kvm->debugfs_stat_data);
1023	}
1024}
1025
1026static int kvm_create_vm_debugfs(struct kvm *kvm, const char *fdname)
1027{
1028	static DEFINE_MUTEX(kvm_debugfs_lock);
1029	struct dentry *dent;
1030	char dir_name[ITOA_MAX_LEN * 2];
1031	struct kvm_stat_data *stat_data;
1032	const struct _kvm_stats_desc *pdesc;
1033	int i, ret = -ENOMEM;
1034	int kvm_debugfs_num_entries = kvm_vm_stats_header.num_desc +
1035				      kvm_vcpu_stats_header.num_desc;
1036
1037	if (!debugfs_initialized())
1038		return 0;
1039
1040	snprintf(dir_name, sizeof(dir_name), "%d-%s", task_pid_nr(current), fdname);
1041	mutex_lock(&kvm_debugfs_lock);
1042	dent = debugfs_lookup(dir_name, kvm_debugfs_dir);
1043	if (dent) {
1044		pr_warn_ratelimited("KVM: debugfs: duplicate directory %s\n", dir_name);
1045		dput(dent);
1046		mutex_unlock(&kvm_debugfs_lock);
1047		return 0;
1048	}
1049	dent = debugfs_create_dir(dir_name, kvm_debugfs_dir);
1050	mutex_unlock(&kvm_debugfs_lock);
1051	if (IS_ERR(dent))
1052		return 0;
1053
1054	kvm->debugfs_dentry = dent;
1055	kvm->debugfs_stat_data = kcalloc(kvm_debugfs_num_entries,
1056					 sizeof(*kvm->debugfs_stat_data),
1057					 GFP_KERNEL_ACCOUNT);
1058	if (!kvm->debugfs_stat_data)
1059		goto out_err;
1060
1061	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
1062		pdesc = &kvm_vm_stats_desc[i];
1063		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1064		if (!stat_data)
1065			goto out_err;
1066
1067		stat_data->kvm = kvm;
1068		stat_data->desc = pdesc;
1069		stat_data->kind = KVM_STAT_VM;
1070		kvm->debugfs_stat_data[i] = stat_data;
1071		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1072				    kvm->debugfs_dentry, stat_data,
1073				    &stat_fops_per_vm);
1074	}
1075
1076	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
1077		pdesc = &kvm_vcpu_stats_desc[i];
1078		stat_data = kzalloc(sizeof(*stat_data), GFP_KERNEL_ACCOUNT);
1079		if (!stat_data)
1080			goto out_err;
1081
1082		stat_data->kvm = kvm;
1083		stat_data->desc = pdesc;
1084		stat_data->kind = KVM_STAT_VCPU;
1085		kvm->debugfs_stat_data[i + kvm_vm_stats_header.num_desc] = stat_data;
1086		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
1087				    kvm->debugfs_dentry, stat_data,
1088				    &stat_fops_per_vm);
1089	}
1090
1091	ret = kvm_arch_create_vm_debugfs(kvm);
1092	if (ret)
1093		goto out_err;
1094
1095	return 0;
1096out_err:
1097	kvm_destroy_vm_debugfs(kvm);
1098	return ret;
1099}
1100
1101/*
1102 * Called after the VM is otherwise initialized, but just before adding it to
1103 * the vm_list.
1104 */
1105int __weak kvm_arch_post_init_vm(struct kvm *kvm)
1106{
1107	return 0;
1108}
1109
1110/*
1111 * Called just after removing the VM from the vm_list, but before doing any
1112 * other destruction.
1113 */
1114void __weak kvm_arch_pre_destroy_vm(struct kvm *kvm)
1115{
1116}
1117
1118/*
1119 * Called after per-vm debugfs created.  When called kvm->debugfs_dentry should
1120 * be setup already, so we can create arch-specific debugfs entries under it.
1121 * Cleanup should be automatic done in kvm_destroy_vm_debugfs() recursively, so
1122 * a per-arch destroy interface is not needed.
1123 */
1124int __weak kvm_arch_create_vm_debugfs(struct kvm *kvm)
1125{
1126	return 0;
1127}
1128
1129static struct kvm *kvm_create_vm(unsigned long type, const char *fdname)
1130{
1131	struct kvm *kvm = kvm_arch_alloc_vm();
1132	struct kvm_memslots *slots;
1133	int r = -ENOMEM;
1134	int i, j;
1135
1136	if (!kvm)
1137		return ERR_PTR(-ENOMEM);
1138
1139	/* KVM is pinned via open("/dev/kvm"), the fd passed to this ioctl(). */
1140	__module_get(kvm_chardev_ops.owner);
1141
1142	KVM_MMU_LOCK_INIT(kvm);
1143	mmgrab(current->mm);
1144	kvm->mm = current->mm;
1145	kvm_eventfd_init(kvm);
1146	mutex_init(&kvm->lock);
1147	mutex_init(&kvm->irq_lock);
1148	mutex_init(&kvm->slots_lock);
1149	mutex_init(&kvm->slots_arch_lock);
1150	spin_lock_init(&kvm->mn_invalidate_lock);
1151	rcuwait_init(&kvm->mn_memslots_update_rcuwait);
1152	xa_init(&kvm->vcpu_array);
 
 
 
1153
1154	INIT_LIST_HEAD(&kvm->gpc_list);
1155	spin_lock_init(&kvm->gpc_lock);
1156
1157	INIT_LIST_HEAD(&kvm->devices);
1158	kvm->max_vcpus = KVM_MAX_VCPUS;
1159
1160	BUILD_BUG_ON(KVM_MEM_SLOTS_NUM > SHRT_MAX);
1161
1162	/*
1163	 * Force subsequent debugfs file creations to fail if the VM directory
1164	 * is not created (by kvm_create_vm_debugfs()).
1165	 */
1166	kvm->debugfs_dentry = ERR_PTR(-ENOENT);
1167
1168	snprintf(kvm->stats_id, sizeof(kvm->stats_id), "kvm-%d",
1169		 task_pid_nr(current));
1170
1171	if (init_srcu_struct(&kvm->srcu))
1172		goto out_err_no_srcu;
1173	if (init_srcu_struct(&kvm->irq_srcu))
1174		goto out_err_no_irq_srcu;
1175
1176	refcount_set(&kvm->users_count, 1);
1177	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1178		for (j = 0; j < 2; j++) {
1179			slots = &kvm->__memslots[i][j];
1180
1181			atomic_long_set(&slots->last_used_slot, (unsigned long)NULL);
1182			slots->hva_tree = RB_ROOT_CACHED;
1183			slots->gfn_tree = RB_ROOT;
1184			hash_init(slots->id_hash);
1185			slots->node_idx = j;
1186
1187			/* Generations must be different for each address space. */
1188			slots->generation = i;
1189		}
1190
1191		rcu_assign_pointer(kvm->memslots[i], &kvm->__memslots[i][0]);
1192	}
1193
1194	for (i = 0; i < KVM_NR_BUSES; i++) {
1195		rcu_assign_pointer(kvm->buses[i],
1196			kzalloc(sizeof(struct kvm_io_bus), GFP_KERNEL_ACCOUNT));
1197		if (!kvm->buses[i])
1198			goto out_err_no_arch_destroy_vm;
1199	}
1200
1201	r = kvm_arch_init_vm(kvm, type);
1202	if (r)
1203		goto out_err_no_arch_destroy_vm;
1204
1205	r = hardware_enable_all();
1206	if (r)
1207		goto out_err_no_disable;
1208
1209#ifdef CONFIG_HAVE_KVM_IRQFD
1210	INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
1211#endif
1212
1213	r = kvm_init_mmu_notifier(kvm);
1214	if (r)
1215		goto out_err_no_mmu_notifier;
1216
1217	r = kvm_coalesced_mmio_init(kvm);
1218	if (r < 0)
1219		goto out_no_coalesced_mmio;
1220
1221	r = kvm_create_vm_debugfs(kvm, fdname);
1222	if (r)
1223		goto out_err_no_debugfs;
1224
1225	r = kvm_arch_post_init_vm(kvm);
1226	if (r)
1227		goto out_err;
1228
1229	mutex_lock(&kvm_lock);
1230	list_add(&kvm->vm_list, &vm_list);
1231	mutex_unlock(&kvm_lock);
1232
1233	preempt_notifier_inc();
1234	kvm_init_pm_notifier(kvm);
1235
1236	return kvm;
1237
1238out_err:
1239	kvm_destroy_vm_debugfs(kvm);
1240out_err_no_debugfs:
1241	kvm_coalesced_mmio_free(kvm);
1242out_no_coalesced_mmio:
1243#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1244	if (kvm->mmu_notifier.ops)
1245		mmu_notifier_unregister(&kvm->mmu_notifier, current->mm);
1246#endif
1247out_err_no_mmu_notifier:
1248	hardware_disable_all();
1249out_err_no_disable:
1250	kvm_arch_destroy_vm(kvm);
1251out_err_no_arch_destroy_vm:
1252	WARN_ON_ONCE(!refcount_dec_and_test(&kvm->users_count));
1253	for (i = 0; i < KVM_NR_BUSES; i++)
1254		kfree(kvm_get_bus(kvm, i));
1255	cleanup_srcu_struct(&kvm->irq_srcu);
1256out_err_no_irq_srcu:
1257	cleanup_srcu_struct(&kvm->srcu);
1258out_err_no_srcu:
1259	kvm_arch_free_vm(kvm);
1260	mmdrop(current->mm);
1261	module_put(kvm_chardev_ops.owner);
1262	return ERR_PTR(r);
1263}
1264
1265static void kvm_destroy_devices(struct kvm *kvm)
1266{
1267	struct kvm_device *dev, *tmp;
1268
1269	/*
1270	 * We do not need to take the kvm->lock here, because nobody else
1271	 * has a reference to the struct kvm at this point and therefore
1272	 * cannot access the devices list anyhow.
1273	 */
1274	list_for_each_entry_safe(dev, tmp, &kvm->devices, vm_node) {
1275		list_del(&dev->vm_node);
1276		dev->ops->destroy(dev);
1277	}
1278}
1279
1280static void kvm_destroy_vm(struct kvm *kvm)
1281{
1282	int i;
1283	struct mm_struct *mm = kvm->mm;
1284
1285	kvm_destroy_pm_notifier(kvm);
1286	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
1287	kvm_destroy_vm_debugfs(kvm);
1288	kvm_arch_sync_events(kvm);
1289	mutex_lock(&kvm_lock);
1290	list_del(&kvm->vm_list);
1291	mutex_unlock(&kvm_lock);
1292	kvm_arch_pre_destroy_vm(kvm);
1293
1294	kvm_free_irq_routing(kvm);
1295	for (i = 0; i < KVM_NR_BUSES; i++) {
1296		struct kvm_io_bus *bus = kvm_get_bus(kvm, i);
1297
1298		if (bus)
1299			kvm_io_bus_destroy(bus);
1300		kvm->buses[i] = NULL;
1301	}
1302	kvm_coalesced_mmio_free(kvm);
1303#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
1304	mmu_notifier_unregister(&kvm->mmu_notifier, kvm->mm);
1305	/*
1306	 * At this point, pending calls to invalidate_range_start()
1307	 * have completed but no more MMU notifiers will run, so
1308	 * mn_active_invalidate_count may remain unbalanced.
1309	 * No threads can be waiting in install_new_memslots as the
1310	 * last reference on KVM has been dropped, but freeing
1311	 * memslots would deadlock without this manual intervention.
 
 
 
 
1312	 */
1313	WARN_ON(rcuwait_active(&kvm->mn_memslots_update_rcuwait));
1314	kvm->mn_active_invalidate_count = 0;
 
 
 
1315#else
1316	kvm_flush_shadow_all(kvm);
1317#endif
1318	kvm_arch_destroy_vm(kvm);
1319	kvm_destroy_devices(kvm);
1320	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
1321		kvm_free_memslots(kvm, &kvm->__memslots[i][0]);
1322		kvm_free_memslots(kvm, &kvm->__memslots[i][1]);
1323	}
1324	cleanup_srcu_struct(&kvm->irq_srcu);
1325	cleanup_srcu_struct(&kvm->srcu);
 
 
 
1326	kvm_arch_free_vm(kvm);
1327	preempt_notifier_dec();
1328	hardware_disable_all();
1329	mmdrop(mm);
1330	module_put(kvm_chardev_ops.owner);
1331}
1332
1333void kvm_get_kvm(struct kvm *kvm)
1334{
1335	refcount_inc(&kvm->users_count);
1336}
1337EXPORT_SYMBOL_GPL(kvm_get_kvm);
1338
1339/*
1340 * Make sure the vm is not during destruction, which is a safe version of
1341 * kvm_get_kvm().  Return true if kvm referenced successfully, false otherwise.
1342 */
1343bool kvm_get_kvm_safe(struct kvm *kvm)
1344{
1345	return refcount_inc_not_zero(&kvm->users_count);
1346}
1347EXPORT_SYMBOL_GPL(kvm_get_kvm_safe);
1348
1349void kvm_put_kvm(struct kvm *kvm)
1350{
1351	if (refcount_dec_and_test(&kvm->users_count))
1352		kvm_destroy_vm(kvm);
1353}
1354EXPORT_SYMBOL_GPL(kvm_put_kvm);
1355
1356/*
1357 * Used to put a reference that was taken on behalf of an object associated
1358 * with a user-visible file descriptor, e.g. a vcpu or device, if installation
1359 * of the new file descriptor fails and the reference cannot be transferred to
1360 * its final owner.  In such cases, the caller is still actively using @kvm and
1361 * will fail miserably if the refcount unexpectedly hits zero.
1362 */
1363void kvm_put_kvm_no_destroy(struct kvm *kvm)
1364{
1365	WARN_ON(refcount_dec_and_test(&kvm->users_count));
1366}
1367EXPORT_SYMBOL_GPL(kvm_put_kvm_no_destroy);
1368
1369static int kvm_vm_release(struct inode *inode, struct file *filp)
1370{
1371	struct kvm *kvm = filp->private_data;
1372
1373	kvm_irqfd_release(kvm);
1374
1375	kvm_put_kvm(kvm);
1376	return 0;
1377}
1378
1379/*
1380 * Allocation size is twice as large as the actual dirty bitmap size.
1381 * See kvm_vm_ioctl_get_dirty_log() why this is needed.
1382 */
1383static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
1384{
1385	unsigned long dirty_bytes = kvm_dirty_bitmap_bytes(memslot);
1386
1387	memslot->dirty_bitmap = __vcalloc(2, dirty_bytes, GFP_KERNEL_ACCOUNT);
1388	if (!memslot->dirty_bitmap)
1389		return -ENOMEM;
1390
1391	return 0;
1392}
1393
1394static struct kvm_memslots *kvm_get_inactive_memslots(struct kvm *kvm, int as_id)
1395{
1396	struct kvm_memslots *active = __kvm_memslots(kvm, as_id);
1397	int node_idx_inactive = active->node_idx ^ 1;
1398
1399	return &kvm->__memslots[as_id][node_idx_inactive];
1400}
1401
1402/*
1403 * Helper to get the address space ID when one of memslot pointers may be NULL.
1404 * This also serves as a sanity that at least one of the pointers is non-NULL,
1405 * and that their address space IDs don't diverge.
1406 */
1407static int kvm_memslots_get_as_id(struct kvm_memory_slot *a,
1408				  struct kvm_memory_slot *b)
1409{
1410	if (WARN_ON_ONCE(!a && !b))
1411		return 0;
1412
1413	if (!a)
1414		return b->as_id;
1415	if (!b)
1416		return a->as_id;
1417
1418	WARN_ON_ONCE(a->as_id != b->as_id);
1419	return a->as_id;
1420}
1421
1422static void kvm_insert_gfn_node(struct kvm_memslots *slots,
1423				struct kvm_memory_slot *slot)
1424{
1425	struct rb_root *gfn_tree = &slots->gfn_tree;
1426	struct rb_node **node, *parent;
1427	int idx = slots->node_idx;
1428
1429	parent = NULL;
1430	for (node = &gfn_tree->rb_node; *node; ) {
1431		struct kvm_memory_slot *tmp;
1432
1433		tmp = container_of(*node, struct kvm_memory_slot, gfn_node[idx]);
1434		parent = *node;
1435		if (slot->base_gfn < tmp->base_gfn)
1436			node = &(*node)->rb_left;
1437		else if (slot->base_gfn > tmp->base_gfn)
1438			node = &(*node)->rb_right;
1439		else
1440			BUG();
1441	}
1442
1443	rb_link_node(&slot->gfn_node[idx], parent, node);
1444	rb_insert_color(&slot->gfn_node[idx], gfn_tree);
1445}
1446
1447static void kvm_erase_gfn_node(struct kvm_memslots *slots,
1448			       struct kvm_memory_slot *slot)
1449{
1450	rb_erase(&slot->gfn_node[slots->node_idx], &slots->gfn_tree);
1451}
1452
1453static void kvm_replace_gfn_node(struct kvm_memslots *slots,
1454				 struct kvm_memory_slot *old,
1455				 struct kvm_memory_slot *new)
1456{
1457	int idx = slots->node_idx;
1458
1459	WARN_ON_ONCE(old->base_gfn != new->base_gfn);
1460
1461	rb_replace_node(&old->gfn_node[idx], &new->gfn_node[idx],
1462			&slots->gfn_tree);
1463}
1464
1465/*
1466 * Replace @old with @new in the inactive memslots.
1467 *
1468 * With NULL @old this simply adds @new.
1469 * With NULL @new this simply removes @old.
1470 *
1471 * If @new is non-NULL its hva_node[slots_idx] range has to be set
1472 * appropriately.
1473 */
1474static void kvm_replace_memslot(struct kvm *kvm,
1475				struct kvm_memory_slot *old,
1476				struct kvm_memory_slot *new)
1477{
1478	int as_id = kvm_memslots_get_as_id(old, new);
1479	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1480	int idx = slots->node_idx;
1481
1482	if (old) {
1483		hash_del(&old->id_node[idx]);
1484		interval_tree_remove(&old->hva_node[idx], &slots->hva_tree);
1485
1486		if ((long)old == atomic_long_read(&slots->last_used_slot))
1487			atomic_long_set(&slots->last_used_slot, (long)new);
1488
1489		if (!new) {
1490			kvm_erase_gfn_node(slots, old);
1491			return;
1492		}
1493	}
1494
1495	/*
1496	 * Initialize @new's hva range.  Do this even when replacing an @old
1497	 * slot, kvm_copy_memslot() deliberately does not touch node data.
1498	 */
1499	new->hva_node[idx].start = new->userspace_addr;
1500	new->hva_node[idx].last = new->userspace_addr +
1501				  (new->npages << PAGE_SHIFT) - 1;
1502
1503	/*
1504	 * (Re)Add the new memslot.  There is no O(1) interval_tree_replace(),
1505	 * hva_node needs to be swapped with remove+insert even though hva can't
1506	 * change when replacing an existing slot.
1507	 */
1508	hash_add(slots->id_hash, &new->id_node[idx], new->id);
1509	interval_tree_insert(&new->hva_node[idx], &slots->hva_tree);
1510
1511	/*
1512	 * If the memslot gfn is unchanged, rb_replace_node() can be used to
1513	 * switch the node in the gfn tree instead of removing the old and
1514	 * inserting the new as two separate operations. Replacement is a
1515	 * single O(1) operation versus two O(log(n)) operations for
1516	 * remove+insert.
1517	 */
1518	if (old && old->base_gfn == new->base_gfn) {
1519		kvm_replace_gfn_node(slots, old, new);
1520	} else {
1521		if (old)
1522			kvm_erase_gfn_node(slots, old);
1523		kvm_insert_gfn_node(slots, new);
1524	}
1525}
1526
1527static int check_memory_region_flags(const struct kvm_userspace_memory_region *mem)
 
 
 
 
 
 
 
 
 
1528{
1529	u32 valid_flags = KVM_MEM_LOG_DIRTY_PAGES;
1530
1531#ifdef __KVM_HAVE_READONLY_MEM
1532	valid_flags |= KVM_MEM_READONLY;
 
 
 
 
 
 
 
 
 
 
 
 
 
1533#endif
1534
1535	if (mem->flags & ~valid_flags)
1536		return -EINVAL;
1537
1538	return 0;
1539}
1540
1541static void kvm_swap_active_memslots(struct kvm *kvm, int as_id)
1542{
1543	struct kvm_memslots *slots = kvm_get_inactive_memslots(kvm, as_id);
1544
1545	/* Grab the generation from the activate memslots. */
1546	u64 gen = __kvm_memslots(kvm, as_id)->generation;
1547
1548	WARN_ON(gen & KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS);
1549	slots->generation = gen | KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1550
1551	/*
1552	 * Do not store the new memslots while there are invalidations in
1553	 * progress, otherwise the locking in invalidate_range_start and
1554	 * invalidate_range_end will be unbalanced.
1555	 */
1556	spin_lock(&kvm->mn_invalidate_lock);
1557	prepare_to_rcuwait(&kvm->mn_memslots_update_rcuwait);
1558	while (kvm->mn_active_invalidate_count) {
1559		set_current_state(TASK_UNINTERRUPTIBLE);
1560		spin_unlock(&kvm->mn_invalidate_lock);
1561		schedule();
1562		spin_lock(&kvm->mn_invalidate_lock);
1563	}
1564	finish_rcuwait(&kvm->mn_memslots_update_rcuwait);
1565	rcu_assign_pointer(kvm->memslots[as_id], slots);
1566	spin_unlock(&kvm->mn_invalidate_lock);
1567
1568	/*
1569	 * Acquired in kvm_set_memslot. Must be released before synchronize
1570	 * SRCU below in order to avoid deadlock with another thread
1571	 * acquiring the slots_arch_lock in an srcu critical section.
1572	 */
1573	mutex_unlock(&kvm->slots_arch_lock);
1574
1575	synchronize_srcu_expedited(&kvm->srcu);
1576
1577	/*
1578	 * Increment the new memslot generation a second time, dropping the
1579	 * update in-progress flag and incrementing the generation based on
1580	 * the number of address spaces.  This provides a unique and easily
1581	 * identifiable generation number while the memslots are in flux.
1582	 */
1583	gen = slots->generation & ~KVM_MEMSLOT_GEN_UPDATE_IN_PROGRESS;
1584
1585	/*
1586	 * Generations must be unique even across address spaces.  We do not need
1587	 * a global counter for that, instead the generation space is evenly split
1588	 * across address spaces.  For example, with two address spaces, address
1589	 * space 0 will use generations 0, 2, 4, ... while address space 1 will
1590	 * use generations 1, 3, 5, ...
1591	 */
1592	gen += KVM_ADDRESS_SPACE_NUM;
1593
1594	kvm_arch_memslots_updated(kvm, gen);
1595
1596	slots->generation = gen;
1597}
1598
1599static int kvm_prepare_memory_region(struct kvm *kvm,
1600				     const struct kvm_memory_slot *old,
1601				     struct kvm_memory_slot *new,
1602				     enum kvm_mr_change change)
1603{
1604	int r;
1605
1606	/*
1607	 * If dirty logging is disabled, nullify the bitmap; the old bitmap
1608	 * will be freed on "commit".  If logging is enabled in both old and
1609	 * new, reuse the existing bitmap.  If logging is enabled only in the
1610	 * new and KVM isn't using a ring buffer, allocate and initialize a
1611	 * new bitmap.
1612	 */
1613	if (change != KVM_MR_DELETE) {
1614		if (!(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
1615			new->dirty_bitmap = NULL;
1616		else if (old && old->dirty_bitmap)
1617			new->dirty_bitmap = old->dirty_bitmap;
1618		else if (kvm_use_dirty_bitmap(kvm)) {
1619			r = kvm_alloc_dirty_bitmap(new);
1620			if (r)
1621				return r;
1622
1623			if (kvm_dirty_log_manual_protect_and_init_set(kvm))
1624				bitmap_set(new->dirty_bitmap, 0, new->npages);
1625		}
1626	}
1627
1628	r = kvm_arch_prepare_memory_region(kvm, old, new, change);
1629
1630	/* Free the bitmap on failure if it was allocated above. */
1631	if (r && new && new->dirty_bitmap && (!old || !old->dirty_bitmap))
1632		kvm_destroy_dirty_bitmap(new);
1633
1634	return r;
1635}
1636
1637static void kvm_commit_memory_region(struct kvm *kvm,
1638				     struct kvm_memory_slot *old,
1639				     const struct kvm_memory_slot *new,
1640				     enum kvm_mr_change change)
1641{
1642	int old_flags = old ? old->flags : 0;
1643	int new_flags = new ? new->flags : 0;
1644	/*
1645	 * Update the total number of memslot pages before calling the arch
1646	 * hook so that architectures can consume the result directly.
1647	 */
1648	if (change == KVM_MR_DELETE)
1649		kvm->nr_memslot_pages -= old->npages;
1650	else if (change == KVM_MR_CREATE)
1651		kvm->nr_memslot_pages += new->npages;
1652
1653	if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES) {
1654		int change = (new_flags & KVM_MEM_LOG_DIRTY_PAGES) ? 1 : -1;
1655		atomic_set(&kvm->nr_memslots_dirty_logging,
1656			   atomic_read(&kvm->nr_memslots_dirty_logging) + change);
1657	}
1658
1659	kvm_arch_commit_memory_region(kvm, old, new, change);
1660
1661	switch (change) {
1662	case KVM_MR_CREATE:
1663		/* Nothing more to do. */
1664		break;
1665	case KVM_MR_DELETE:
1666		/* Free the old memslot and all its metadata. */
1667		kvm_free_memslot(kvm, old);
1668		break;
1669	case KVM_MR_MOVE:
1670	case KVM_MR_FLAGS_ONLY:
1671		/*
1672		 * Free the dirty bitmap as needed; the below check encompasses
1673		 * both the flags and whether a ring buffer is being used)
1674		 */
1675		if (old->dirty_bitmap && !new->dirty_bitmap)
1676			kvm_destroy_dirty_bitmap(old);
1677
1678		/*
1679		 * The final quirk.  Free the detached, old slot, but only its
1680		 * memory, not any metadata.  Metadata, including arch specific
1681		 * data, may be reused by @new.
1682		 */
1683		kfree(old);
1684		break;
1685	default:
1686		BUG();
1687	}
1688}
1689
1690/*
1691 * Activate @new, which must be installed in the inactive slots by the caller,
1692 * by swapping the active slots and then propagating @new to @old once @old is
1693 * unreachable and can be safely modified.
1694 *
1695 * With NULL @old this simply adds @new to @active (while swapping the sets).
1696 * With NULL @new this simply removes @old from @active and frees it
1697 * (while also swapping the sets).
1698 */
1699static void kvm_activate_memslot(struct kvm *kvm,
1700				 struct kvm_memory_slot *old,
1701				 struct kvm_memory_slot *new)
1702{
1703	int as_id = kvm_memslots_get_as_id(old, new);
1704
1705	kvm_swap_active_memslots(kvm, as_id);
1706
1707	/* Propagate the new memslot to the now inactive memslots. */
1708	kvm_replace_memslot(kvm, old, new);
1709}
1710
1711static void kvm_copy_memslot(struct kvm_memory_slot *dest,
1712			     const struct kvm_memory_slot *src)
1713{
1714	dest->base_gfn = src->base_gfn;
1715	dest->npages = src->npages;
1716	dest->dirty_bitmap = src->dirty_bitmap;
1717	dest->arch = src->arch;
1718	dest->userspace_addr = src->userspace_addr;
1719	dest->flags = src->flags;
1720	dest->id = src->id;
1721	dest->as_id = src->as_id;
1722}
1723
1724static void kvm_invalidate_memslot(struct kvm *kvm,
1725				   struct kvm_memory_slot *old,
1726				   struct kvm_memory_slot *invalid_slot)
1727{
1728	/*
1729	 * Mark the current slot INVALID.  As with all memslot modifications,
1730	 * this must be done on an unreachable slot to avoid modifying the
1731	 * current slot in the active tree.
1732	 */
1733	kvm_copy_memslot(invalid_slot, old);
1734	invalid_slot->flags |= KVM_MEMSLOT_INVALID;
1735	kvm_replace_memslot(kvm, old, invalid_slot);
1736
1737	/*
1738	 * Activate the slot that is now marked INVALID, but don't propagate
1739	 * the slot to the now inactive slots. The slot is either going to be
1740	 * deleted or recreated as a new slot.
1741	 */
1742	kvm_swap_active_memslots(kvm, old->as_id);
1743
1744	/*
1745	 * From this point no new shadow pages pointing to a deleted, or moved,
1746	 * memslot will be created.  Validation of sp->gfn happens in:
1747	 *	- gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1748	 *	- kvm_is_visible_gfn (mmu_check_root)
1749	 */
1750	kvm_arch_flush_shadow_memslot(kvm, old);
1751	kvm_arch_guest_memory_reclaimed(kvm);
1752
1753	/* Was released by kvm_swap_active_memslots, reacquire. */
1754	mutex_lock(&kvm->slots_arch_lock);
1755
1756	/*
1757	 * Copy the arch-specific field of the newly-installed slot back to the
1758	 * old slot as the arch data could have changed between releasing
1759	 * slots_arch_lock in install_new_memslots() and re-acquiring the lock
1760	 * above.  Writers are required to retrieve memslots *after* acquiring
1761	 * slots_arch_lock, thus the active slot's data is guaranteed to be fresh.
1762	 */
1763	old->arch = invalid_slot->arch;
1764}
1765
1766static void kvm_create_memslot(struct kvm *kvm,
1767			       struct kvm_memory_slot *new)
1768{
1769	/* Add the new memslot to the inactive set and activate. */
1770	kvm_replace_memslot(kvm, NULL, new);
1771	kvm_activate_memslot(kvm, NULL, new);
1772}
1773
1774static void kvm_delete_memslot(struct kvm *kvm,
1775			       struct kvm_memory_slot *old,
1776			       struct kvm_memory_slot *invalid_slot)
1777{
1778	/*
1779	 * Remove the old memslot (in the inactive memslots) by passing NULL as
1780	 * the "new" slot, and for the invalid version in the active slots.
1781	 */
1782	kvm_replace_memslot(kvm, old, NULL);
1783	kvm_activate_memslot(kvm, invalid_slot, NULL);
1784}
1785
1786static void kvm_move_memslot(struct kvm *kvm,
1787			     struct kvm_memory_slot *old,
1788			     struct kvm_memory_slot *new,
1789			     struct kvm_memory_slot *invalid_slot)
1790{
1791	/*
1792	 * Replace the old memslot in the inactive slots, and then swap slots
1793	 * and replace the current INVALID with the new as well.
1794	 */
1795	kvm_replace_memslot(kvm, old, new);
1796	kvm_activate_memslot(kvm, invalid_slot, new);
1797}
1798
1799static void kvm_update_flags_memslot(struct kvm *kvm,
1800				     struct kvm_memory_slot *old,
1801				     struct kvm_memory_slot *new)
1802{
1803	/*
1804	 * Similar to the MOVE case, but the slot doesn't need to be zapped as
1805	 * an intermediate step. Instead, the old memslot is simply replaced
1806	 * with a new, updated copy in both memslot sets.
1807	 */
1808	kvm_replace_memslot(kvm, old, new);
1809	kvm_activate_memslot(kvm, old, new);
1810}
1811
1812static int kvm_set_memslot(struct kvm *kvm,
1813			   struct kvm_memory_slot *old,
1814			   struct kvm_memory_slot *new,
1815			   enum kvm_mr_change change)
1816{
1817	struct kvm_memory_slot *invalid_slot;
1818	int r;
1819
1820	/*
1821	 * Released in kvm_swap_active_memslots.
1822	 *
1823	 * Must be held from before the current memslots are copied until
1824	 * after the new memslots are installed with rcu_assign_pointer,
1825	 * then released before the synchronize srcu in kvm_swap_active_memslots.
1826	 *
1827	 * When modifying memslots outside of the slots_lock, must be held
1828	 * before reading the pointer to the current memslots until after all
1829	 * changes to those memslots are complete.
1830	 *
1831	 * These rules ensure that installing new memslots does not lose
1832	 * changes made to the previous memslots.
1833	 */
1834	mutex_lock(&kvm->slots_arch_lock);
1835
1836	/*
1837	 * Invalidate the old slot if it's being deleted or moved.  This is
1838	 * done prior to actually deleting/moving the memslot to allow vCPUs to
1839	 * continue running by ensuring there are no mappings or shadow pages
1840	 * for the memslot when it is deleted/moved.  Without pre-invalidation
1841	 * (and without a lock), a window would exist between effecting the
1842	 * delete/move and committing the changes in arch code where KVM or a
1843	 * guest could access a non-existent memslot.
1844	 *
1845	 * Modifications are done on a temporary, unreachable slot.  The old
1846	 * slot needs to be preserved in case a later step fails and the
1847	 * invalidation needs to be reverted.
1848	 */
1849	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1850		invalid_slot = kzalloc(sizeof(*invalid_slot), GFP_KERNEL_ACCOUNT);
1851		if (!invalid_slot) {
1852			mutex_unlock(&kvm->slots_arch_lock);
1853			return -ENOMEM;
1854		}
1855		kvm_invalidate_memslot(kvm, old, invalid_slot);
1856	}
1857
1858	r = kvm_prepare_memory_region(kvm, old, new, change);
1859	if (r) {
1860		/*
1861		 * For DELETE/MOVE, revert the above INVALID change.  No
1862		 * modifications required since the original slot was preserved
1863		 * in the inactive slots.  Changing the active memslots also
1864		 * release slots_arch_lock.
1865		 */
1866		if (change == KVM_MR_DELETE || change == KVM_MR_MOVE) {
1867			kvm_activate_memslot(kvm, invalid_slot, old);
1868			kfree(invalid_slot);
1869		} else {
1870			mutex_unlock(&kvm->slots_arch_lock);
1871		}
1872		return r;
1873	}
1874
1875	/*
1876	 * For DELETE and MOVE, the working slot is now active as the INVALID
1877	 * version of the old slot.  MOVE is particularly special as it reuses
1878	 * the old slot and returns a copy of the old slot (in working_slot).
1879	 * For CREATE, there is no old slot.  For DELETE and FLAGS_ONLY, the
1880	 * old slot is detached but otherwise preserved.
1881	 */
1882	if (change == KVM_MR_CREATE)
1883		kvm_create_memslot(kvm, new);
1884	else if (change == KVM_MR_DELETE)
1885		kvm_delete_memslot(kvm, old, invalid_slot);
1886	else if (change == KVM_MR_MOVE)
1887		kvm_move_memslot(kvm, old, new, invalid_slot);
1888	else if (change == KVM_MR_FLAGS_ONLY)
1889		kvm_update_flags_memslot(kvm, old, new);
1890	else
1891		BUG();
1892
1893	/* Free the temporary INVALID slot used for DELETE and MOVE. */
1894	if (change == KVM_MR_DELETE || change == KVM_MR_MOVE)
1895		kfree(invalid_slot);
1896
1897	/*
1898	 * No need to refresh new->arch, changes after dropping slots_arch_lock
1899	 * will directly hit the final, active memslot.  Architectures are
1900	 * responsible for knowing that new->arch may be stale.
1901	 */
1902	kvm_commit_memory_region(kvm, old, new, change);
1903
1904	return 0;
1905}
1906
1907static bool kvm_check_memslot_overlap(struct kvm_memslots *slots, int id,
1908				      gfn_t start, gfn_t end)
1909{
1910	struct kvm_memslot_iter iter;
1911
1912	kvm_for_each_memslot_in_gfn_range(&iter, slots, start, end) {
1913		if (iter.slot->id != id)
1914			return true;
1915	}
1916
1917	return false;
1918}
1919
1920/*
1921 * Allocate some memory and give it an address in the guest physical address
1922 * space.
1923 *
1924 * Discontiguous memory is allowed, mostly for framebuffers.
1925 *
1926 * Must be called holding kvm->slots_lock for write.
1927 */
1928int __kvm_set_memory_region(struct kvm *kvm,
1929			    const struct kvm_userspace_memory_region *mem)
1930{
1931	struct kvm_memory_slot *old, *new;
1932	struct kvm_memslots *slots;
1933	enum kvm_mr_change change;
1934	unsigned long npages;
1935	gfn_t base_gfn;
1936	int as_id, id;
1937	int r;
1938
1939	r = check_memory_region_flags(mem);
1940	if (r)
1941		return r;
1942
1943	as_id = mem->slot >> 16;
1944	id = (u16)mem->slot;
1945
1946	/* General sanity checks */
1947	if ((mem->memory_size & (PAGE_SIZE - 1)) ||
1948	    (mem->memory_size != (unsigned long)mem->memory_size))
1949		return -EINVAL;
1950	if (mem->guest_phys_addr & (PAGE_SIZE - 1))
1951		return -EINVAL;
1952	/* We can read the guest memory with __xxx_user() later on. */
1953	if ((mem->userspace_addr & (PAGE_SIZE - 1)) ||
1954	    (mem->userspace_addr != untagged_addr(mem->userspace_addr)) ||
1955	     !access_ok((void __user *)(unsigned long)mem->userspace_addr,
1956			mem->memory_size))
1957		return -EINVAL;
1958	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_MEM_SLOTS_NUM)
 
 
 
 
1959		return -EINVAL;
1960	if (mem->guest_phys_addr + mem->memory_size < mem->guest_phys_addr)
1961		return -EINVAL;
1962	if ((mem->memory_size >> PAGE_SHIFT) > KVM_MEM_MAX_NR_PAGES)
1963		return -EINVAL;
1964
1965	slots = __kvm_memslots(kvm, as_id);
1966
1967	/*
1968	 * Note, the old memslot (and the pointer itself!) may be invalidated
1969	 * and/or destroyed by kvm_set_memslot().
1970	 */
1971	old = id_to_memslot(slots, id);
1972
1973	if (!mem->memory_size) {
1974		if (!old || !old->npages)
1975			return -EINVAL;
1976
1977		if (WARN_ON_ONCE(kvm->nr_memslot_pages < old->npages))
1978			return -EIO;
1979
1980		return kvm_set_memslot(kvm, old, NULL, KVM_MR_DELETE);
1981	}
1982
1983	base_gfn = (mem->guest_phys_addr >> PAGE_SHIFT);
1984	npages = (mem->memory_size >> PAGE_SHIFT);
1985
1986	if (!old || !old->npages) {
1987		change = KVM_MR_CREATE;
1988
1989		/*
1990		 * To simplify KVM internals, the total number of pages across
1991		 * all memslots must fit in an unsigned long.
1992		 */
1993		if ((kvm->nr_memslot_pages + npages) < kvm->nr_memslot_pages)
1994			return -EINVAL;
1995	} else { /* Modify an existing slot. */
 
 
 
1996		if ((mem->userspace_addr != old->userspace_addr) ||
1997		    (npages != old->npages) ||
1998		    ((mem->flags ^ old->flags) & KVM_MEM_READONLY))
1999			return -EINVAL;
2000
2001		if (base_gfn != old->base_gfn)
2002			change = KVM_MR_MOVE;
2003		else if (mem->flags != old->flags)
2004			change = KVM_MR_FLAGS_ONLY;
2005		else /* Nothing to change. */
2006			return 0;
2007	}
2008
2009	if ((change == KVM_MR_CREATE || change == KVM_MR_MOVE) &&
2010	    kvm_check_memslot_overlap(slots, id, base_gfn, base_gfn + npages))
2011		return -EEXIST;
2012
2013	/* Allocate a slot that will persist in the memslot. */
2014	new = kzalloc(sizeof(*new), GFP_KERNEL_ACCOUNT);
2015	if (!new)
2016		return -ENOMEM;
2017
2018	new->as_id = as_id;
2019	new->id = id;
2020	new->base_gfn = base_gfn;
2021	new->npages = npages;
2022	new->flags = mem->flags;
2023	new->userspace_addr = mem->userspace_addr;
 
 
 
 
 
2024
2025	r = kvm_set_memslot(kvm, old, new, change);
2026	if (r)
2027		kfree(new);
 
 
 
 
 
 
 
 
2028	return r;
2029}
2030EXPORT_SYMBOL_GPL(__kvm_set_memory_region);
2031
2032int kvm_set_memory_region(struct kvm *kvm,
2033			  const struct kvm_userspace_memory_region *mem)
2034{
2035	int r;
2036
2037	mutex_lock(&kvm->slots_lock);
2038	r = __kvm_set_memory_region(kvm, mem);
2039	mutex_unlock(&kvm->slots_lock);
2040	return r;
2041}
2042EXPORT_SYMBOL_GPL(kvm_set_memory_region);
2043
2044static int kvm_vm_ioctl_set_memory_region(struct kvm *kvm,
2045					  struct kvm_userspace_memory_region *mem)
2046{
2047	if ((u16)mem->slot >= KVM_USER_MEM_SLOTS)
2048		return -EINVAL;
2049
2050	return kvm_set_memory_region(kvm, mem);
2051}
2052
2053#ifndef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
2054/**
2055 * kvm_get_dirty_log - get a snapshot of dirty pages
2056 * @kvm:	pointer to kvm instance
2057 * @log:	slot id and address to which we copy the log
2058 * @is_dirty:	set to '1' if any dirty pages were found
2059 * @memslot:	set to the associated memslot, always valid on success
2060 */
2061int kvm_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log,
2062		      int *is_dirty, struct kvm_memory_slot **memslot)
2063{
2064	struct kvm_memslots *slots;
2065	int i, as_id, id;
2066	unsigned long n;
2067	unsigned long any = 0;
2068
2069	/* Dirty ring tracking may be exclusive to dirty log tracking */
2070	if (!kvm_use_dirty_bitmap(kvm))
2071		return -ENXIO;
2072
2073	*memslot = NULL;
2074	*is_dirty = 0;
2075
2076	as_id = log->slot >> 16;
2077	id = (u16)log->slot;
2078	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2079		return -EINVAL;
2080
2081	slots = __kvm_memslots(kvm, as_id);
2082	*memslot = id_to_memslot(slots, id);
2083	if (!(*memslot) || !(*memslot)->dirty_bitmap)
2084		return -ENOENT;
2085
2086	kvm_arch_sync_dirty_log(kvm, *memslot);
2087
2088	n = kvm_dirty_bitmap_bytes(*memslot);
2089
2090	for (i = 0; !any && i < n/sizeof(long); ++i)
2091		any = (*memslot)->dirty_bitmap[i];
2092
2093	if (copy_to_user(log->dirty_bitmap, (*memslot)->dirty_bitmap, n))
2094		return -EFAULT;
2095
2096	if (any)
2097		*is_dirty = 1;
2098	return 0;
2099}
2100EXPORT_SYMBOL_GPL(kvm_get_dirty_log);
2101
2102#else /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2103/**
2104 * kvm_get_dirty_log_protect - get a snapshot of dirty pages
2105 *	and reenable dirty page tracking for the corresponding pages.
2106 * @kvm:	pointer to kvm instance
2107 * @log:	slot id and address to which we copy the log
2108 *
2109 * We need to keep it in mind that VCPU threads can write to the bitmap
2110 * concurrently. So, to avoid losing track of dirty pages we keep the
2111 * following order:
2112 *
2113 *    1. Take a snapshot of the bit and clear it if needed.
2114 *    2. Write protect the corresponding page.
2115 *    3. Copy the snapshot to the userspace.
2116 *    4. Upon return caller flushes TLB's if needed.
2117 *
2118 * Between 2 and 4, the guest may write to the page using the remaining TLB
2119 * entry.  This is not a problem because the page is reported dirty using
2120 * the snapshot taken before and step 4 ensures that writes done after
2121 * exiting to userspace will be logged for the next call.
2122 *
2123 */
2124static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
2125{
2126	struct kvm_memslots *slots;
2127	struct kvm_memory_slot *memslot;
2128	int i, as_id, id;
2129	unsigned long n;
2130	unsigned long *dirty_bitmap;
2131	unsigned long *dirty_bitmap_buffer;
2132	bool flush;
2133
2134	/* Dirty ring tracking may be exclusive to dirty log tracking */
2135	if (!kvm_use_dirty_bitmap(kvm))
2136		return -ENXIO;
2137
2138	as_id = log->slot >> 16;
2139	id = (u16)log->slot;
2140	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2141		return -EINVAL;
2142
2143	slots = __kvm_memslots(kvm, as_id);
2144	memslot = id_to_memslot(slots, id);
2145	if (!memslot || !memslot->dirty_bitmap)
2146		return -ENOENT;
2147
2148	dirty_bitmap = memslot->dirty_bitmap;
2149
2150	kvm_arch_sync_dirty_log(kvm, memslot);
2151
2152	n = kvm_dirty_bitmap_bytes(memslot);
2153	flush = false;
2154	if (kvm->manual_dirty_log_protect) {
2155		/*
2156		 * Unlike kvm_get_dirty_log, we always return false in *flush,
2157		 * because no flush is needed until KVM_CLEAR_DIRTY_LOG.  There
2158		 * is some code duplication between this function and
2159		 * kvm_get_dirty_log, but hopefully all architecture
2160		 * transition to kvm_get_dirty_log_protect and kvm_get_dirty_log
2161		 * can be eliminated.
2162		 */
2163		dirty_bitmap_buffer = dirty_bitmap;
2164	} else {
2165		dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2166		memset(dirty_bitmap_buffer, 0, n);
2167
2168		KVM_MMU_LOCK(kvm);
2169		for (i = 0; i < n / sizeof(long); i++) {
2170			unsigned long mask;
2171			gfn_t offset;
2172
2173			if (!dirty_bitmap[i])
2174				continue;
2175
2176			flush = true;
2177			mask = xchg(&dirty_bitmap[i], 0);
2178			dirty_bitmap_buffer[i] = mask;
2179
2180			offset = i * BITS_PER_LONG;
2181			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2182								offset, mask);
2183		}
2184		KVM_MMU_UNLOCK(kvm);
2185	}
2186
2187	if (flush)
2188		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2189
2190	if (copy_to_user(log->dirty_bitmap, dirty_bitmap_buffer, n))
2191		return -EFAULT;
2192	return 0;
2193}
2194
2195
2196/**
2197 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
2198 * @kvm: kvm instance
2199 * @log: slot id and address to which we copy the log
2200 *
2201 * Steps 1-4 below provide general overview of dirty page logging. See
2202 * kvm_get_dirty_log_protect() function description for additional details.
2203 *
2204 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
2205 * always flush the TLB (step 4) even if previous step failed  and the dirty
2206 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
2207 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
2208 * writes will be marked dirty for next log read.
2209 *
2210 *   1. Take a snapshot of the bit and clear it if needed.
2211 *   2. Write protect the corresponding page.
2212 *   3. Copy the snapshot to the userspace.
2213 *   4. Flush TLB's if needed.
2214 */
2215static int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm,
2216				      struct kvm_dirty_log *log)
2217{
2218	int r;
2219
2220	mutex_lock(&kvm->slots_lock);
2221
2222	r = kvm_get_dirty_log_protect(kvm, log);
2223
2224	mutex_unlock(&kvm->slots_lock);
2225	return r;
2226}
2227
2228/**
2229 * kvm_clear_dirty_log_protect - clear dirty bits in the bitmap
2230 *	and reenable dirty page tracking for the corresponding pages.
2231 * @kvm:	pointer to kvm instance
2232 * @log:	slot id and address from which to fetch the bitmap of dirty pages
2233 */
2234static int kvm_clear_dirty_log_protect(struct kvm *kvm,
2235				       struct kvm_clear_dirty_log *log)
2236{
2237	struct kvm_memslots *slots;
2238	struct kvm_memory_slot *memslot;
2239	int as_id, id;
2240	gfn_t offset;
2241	unsigned long i, n;
2242	unsigned long *dirty_bitmap;
2243	unsigned long *dirty_bitmap_buffer;
2244	bool flush;
2245
2246	/* Dirty ring tracking may be exclusive to dirty log tracking */
2247	if (!kvm_use_dirty_bitmap(kvm))
2248		return -ENXIO;
2249
2250	as_id = log->slot >> 16;
2251	id = (u16)log->slot;
2252	if (as_id >= KVM_ADDRESS_SPACE_NUM || id >= KVM_USER_MEM_SLOTS)
2253		return -EINVAL;
2254
2255	if (log->first_page & 63)
2256		return -EINVAL;
2257
2258	slots = __kvm_memslots(kvm, as_id);
2259	memslot = id_to_memslot(slots, id);
2260	if (!memslot || !memslot->dirty_bitmap)
2261		return -ENOENT;
2262
2263	dirty_bitmap = memslot->dirty_bitmap;
2264
2265	n = ALIGN(log->num_pages, BITS_PER_LONG) / 8;
2266
2267	if (log->first_page > memslot->npages ||
2268	    log->num_pages > memslot->npages - log->first_page ||
2269	    (log->num_pages < memslot->npages - log->first_page && (log->num_pages & 63)))
2270	    return -EINVAL;
2271
2272	kvm_arch_sync_dirty_log(kvm, memslot);
2273
2274	flush = false;
2275	dirty_bitmap_buffer = kvm_second_dirty_bitmap(memslot);
2276	if (copy_from_user(dirty_bitmap_buffer, log->dirty_bitmap, n))
2277		return -EFAULT;
2278
2279	KVM_MMU_LOCK(kvm);
2280	for (offset = log->first_page, i = offset / BITS_PER_LONG,
2281		 n = DIV_ROUND_UP(log->num_pages, BITS_PER_LONG); n--;
2282	     i++, offset += BITS_PER_LONG) {
2283		unsigned long mask = *dirty_bitmap_buffer++;
2284		atomic_long_t *p = (atomic_long_t *) &dirty_bitmap[i];
2285		if (!mask)
2286			continue;
2287
2288		mask &= atomic_long_fetch_andnot(mask, p);
2289
2290		/*
2291		 * mask contains the bits that really have been cleared.  This
2292		 * never includes any bits beyond the length of the memslot (if
2293		 * the length is not aligned to 64 pages), therefore it is not
2294		 * a problem if userspace sets them in log->dirty_bitmap.
2295		*/
2296		if (mask) {
2297			flush = true;
2298			kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
2299								offset, mask);
2300		}
2301	}
2302	KVM_MMU_UNLOCK(kvm);
2303
2304	if (flush)
2305		kvm_arch_flush_remote_tlbs_memslot(kvm, memslot);
2306
2307	return 0;
2308}
2309
2310static int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm,
2311					struct kvm_clear_dirty_log *log)
2312{
2313	int r;
2314
2315	mutex_lock(&kvm->slots_lock);
2316
2317	r = kvm_clear_dirty_log_protect(kvm, log);
2318
2319	mutex_unlock(&kvm->slots_lock);
2320	return r;
2321}
2322#endif /* CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT */
2323
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2324struct kvm_memory_slot *gfn_to_memslot(struct kvm *kvm, gfn_t gfn)
2325{
2326	return __gfn_to_memslot(kvm_memslots(kvm), gfn);
2327}
2328EXPORT_SYMBOL_GPL(gfn_to_memslot);
2329
2330struct kvm_memory_slot *kvm_vcpu_gfn_to_memslot(struct kvm_vcpu *vcpu, gfn_t gfn)
2331{
2332	struct kvm_memslots *slots = kvm_vcpu_memslots(vcpu);
2333	u64 gen = slots->generation;
2334	struct kvm_memory_slot *slot;
2335
2336	/*
2337	 * This also protects against using a memslot from a different address space,
2338	 * since different address spaces have different generation numbers.
2339	 */
2340	if (unlikely(gen != vcpu->last_used_slot_gen)) {
2341		vcpu->last_used_slot = NULL;
2342		vcpu->last_used_slot_gen = gen;
2343	}
2344
2345	slot = try_get_memslot(vcpu->last_used_slot, gfn);
2346	if (slot)
2347		return slot;
2348
2349	/*
2350	 * Fall back to searching all memslots. We purposely use
2351	 * search_memslots() instead of __gfn_to_memslot() to avoid
2352	 * thrashing the VM-wide last_used_slot in kvm_memslots.
2353	 */
2354	slot = search_memslots(slots, gfn, false);
2355	if (slot) {
2356		vcpu->last_used_slot = slot;
2357		return slot;
2358	}
2359
2360	return NULL;
2361}
2362
2363bool kvm_is_visible_gfn(struct kvm *kvm, gfn_t gfn)
2364{
2365	struct kvm_memory_slot *memslot = gfn_to_memslot(kvm, gfn);
2366
2367	return kvm_is_visible_memslot(memslot);
2368}
2369EXPORT_SYMBOL_GPL(kvm_is_visible_gfn);
2370
2371bool kvm_vcpu_is_visible_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2372{
2373	struct kvm_memory_slot *memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2374
2375	return kvm_is_visible_memslot(memslot);
2376}
2377EXPORT_SYMBOL_GPL(kvm_vcpu_is_visible_gfn);
2378
2379unsigned long kvm_host_page_size(struct kvm_vcpu *vcpu, gfn_t gfn)
2380{
2381	struct vm_area_struct *vma;
2382	unsigned long addr, size;
2383
2384	size = PAGE_SIZE;
2385
2386	addr = kvm_vcpu_gfn_to_hva_prot(vcpu, gfn, NULL);
2387	if (kvm_is_error_hva(addr))
2388		return PAGE_SIZE;
2389
2390	mmap_read_lock(current->mm);
2391	vma = find_vma(current->mm, addr);
2392	if (!vma)
2393		goto out;
2394
2395	size = vma_kernel_pagesize(vma);
2396
2397out:
2398	mmap_read_unlock(current->mm);
2399
2400	return size;
2401}
2402
2403static bool memslot_is_readonly(const struct kvm_memory_slot *slot)
2404{
2405	return slot->flags & KVM_MEM_READONLY;
2406}
2407
2408static unsigned long __gfn_to_hva_many(const struct kvm_memory_slot *slot, gfn_t gfn,
2409				       gfn_t *nr_pages, bool write)
2410{
2411	if (!slot || slot->flags & KVM_MEMSLOT_INVALID)
2412		return KVM_HVA_ERR_BAD;
2413
2414	if (memslot_is_readonly(slot) && write)
2415		return KVM_HVA_ERR_RO_BAD;
2416
2417	if (nr_pages)
2418		*nr_pages = slot->npages - (gfn - slot->base_gfn);
2419
2420	return __gfn_to_hva_memslot(slot, gfn);
2421}
2422
2423static unsigned long gfn_to_hva_many(struct kvm_memory_slot *slot, gfn_t gfn,
2424				     gfn_t *nr_pages)
2425{
2426	return __gfn_to_hva_many(slot, gfn, nr_pages, true);
2427}
2428
2429unsigned long gfn_to_hva_memslot(struct kvm_memory_slot *slot,
2430					gfn_t gfn)
2431{
2432	return gfn_to_hva_many(slot, gfn, NULL);
2433}
2434EXPORT_SYMBOL_GPL(gfn_to_hva_memslot);
2435
2436unsigned long gfn_to_hva(struct kvm *kvm, gfn_t gfn)
2437{
2438	return gfn_to_hva_many(gfn_to_memslot(kvm, gfn), gfn, NULL);
2439}
2440EXPORT_SYMBOL_GPL(gfn_to_hva);
2441
2442unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu *vcpu, gfn_t gfn)
2443{
2444	return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn, NULL);
2445}
2446EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva);
2447
2448/*
2449 * Return the hva of a @gfn and the R/W attribute if possible.
2450 *
2451 * @slot: the kvm_memory_slot which contains @gfn
2452 * @gfn: the gfn to be translated
2453 * @writable: used to return the read/write attribute of the @slot if the hva
2454 * is valid and @writable is not NULL
2455 */
2456unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot *slot,
2457				      gfn_t gfn, bool *writable)
2458{
2459	unsigned long hva = __gfn_to_hva_many(slot, gfn, NULL, false);
2460
2461	if (!kvm_is_error_hva(hva) && writable)
2462		*writable = !memslot_is_readonly(slot);
2463
2464	return hva;
2465}
2466
2467unsigned long gfn_to_hva_prot(struct kvm *kvm, gfn_t gfn, bool *writable)
2468{
2469	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
2470
2471	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2472}
2473
2474unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu *vcpu, gfn_t gfn, bool *writable)
2475{
2476	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
2477
2478	return gfn_to_hva_memslot_prot(slot, gfn, writable);
2479}
2480
2481static inline int check_user_page_hwpoison(unsigned long addr)
2482{
2483	int rc, flags = FOLL_HWPOISON | FOLL_WRITE;
2484
2485	rc = get_user_pages(addr, 1, flags, NULL, NULL);
2486	return rc == -EHWPOISON;
2487}
2488
2489/*
2490 * The fast path to get the writable pfn which will be stored in @pfn,
2491 * true indicates success, otherwise false is returned.  It's also the
2492 * only part that runs if we can in atomic context.
2493 */
2494static bool hva_to_pfn_fast(unsigned long addr, bool write_fault,
2495			    bool *writable, kvm_pfn_t *pfn)
2496{
2497	struct page *page[1];
2498
2499	/*
2500	 * Fast pin a writable pfn only if it is a write fault request
2501	 * or the caller allows to map a writable pfn for a read fault
2502	 * request.
2503	 */
2504	if (!(write_fault || writable))
2505		return false;
2506
2507	if (get_user_page_fast_only(addr, FOLL_WRITE, page)) {
2508		*pfn = page_to_pfn(page[0]);
2509
2510		if (writable)
2511			*writable = true;
2512		return true;
2513	}
2514
2515	return false;
2516}
2517
2518/*
2519 * The slow path to get the pfn of the specified host virtual address,
2520 * 1 indicates success, -errno is returned if error is detected.
2521 */
2522static int hva_to_pfn_slow(unsigned long addr, bool *async, bool write_fault,
2523			   bool interruptible, bool *writable, kvm_pfn_t *pfn)
2524{
2525	unsigned int flags = FOLL_HWPOISON;
 
 
 
 
 
 
 
 
 
 
 
2526	struct page *page;
2527	int npages;
2528
2529	might_sleep();
2530
2531	if (writable)
2532		*writable = write_fault;
2533
2534	if (write_fault)
2535		flags |= FOLL_WRITE;
2536	if (async)
2537		flags |= FOLL_NOWAIT;
2538	if (interruptible)
2539		flags |= FOLL_INTERRUPTIBLE;
2540
2541	npages = get_user_pages_unlocked(addr, 1, &page, flags);
2542	if (npages != 1)
2543		return npages;
2544
2545	/* map read fault as writable if possible */
2546	if (unlikely(!write_fault) && writable) {
2547		struct page *wpage;
2548
2549		if (get_user_page_fast_only(addr, FOLL_WRITE, &wpage)) {
2550			*writable = true;
2551			put_page(page);
2552			page = wpage;
2553		}
2554	}
2555	*pfn = page_to_pfn(page);
2556	return npages;
2557}
2558
2559static bool vma_is_valid(struct vm_area_struct *vma, bool write_fault)
2560{
2561	if (unlikely(!(vma->vm_flags & VM_READ)))
2562		return false;
2563
2564	if (write_fault && (unlikely(!(vma->vm_flags & VM_WRITE))))
2565		return false;
2566
2567	return true;
2568}
2569
2570static int kvm_try_get_pfn(kvm_pfn_t pfn)
2571{
2572	struct page *page = kvm_pfn_to_refcounted_page(pfn);
2573
2574	if (!page)
2575		return 1;
2576
2577	return get_page_unless_zero(page);
2578}
2579
2580static int hva_to_pfn_remapped(struct vm_area_struct *vma,
2581			       unsigned long addr, bool write_fault,
2582			       bool *writable, kvm_pfn_t *p_pfn)
2583{
2584	kvm_pfn_t pfn;
2585	pte_t *ptep;
 
2586	spinlock_t *ptl;
2587	int r;
2588
2589	r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2590	if (r) {
2591		/*
2592		 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
2593		 * not call the fault handler, so do it here.
2594		 */
2595		bool unlocked = false;
2596		r = fixup_user_fault(current->mm, addr,
2597				     (write_fault ? FAULT_FLAG_WRITE : 0),
2598				     &unlocked);
2599		if (unlocked)
2600			return -EAGAIN;
2601		if (r)
2602			return r;
2603
2604		r = follow_pte(vma->vm_mm, addr, &ptep, &ptl);
2605		if (r)
2606			return r;
2607	}
2608
2609	if (write_fault && !pte_write(*ptep)) {
 
 
2610		pfn = KVM_PFN_ERR_RO_FAULT;
2611		goto out;
2612	}
2613
2614	if (writable)
2615		*writable = pte_write(*ptep);
2616	pfn = pte_pfn(*ptep);
2617
2618	/*
2619	 * Get a reference here because callers of *hva_to_pfn* and
2620	 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
2621	 * returned pfn.  This is only needed if the VMA has VM_MIXEDMAP
2622	 * set, but the kvm_try_get_pfn/kvm_release_pfn_clean pair will
2623	 * simply do nothing for reserved pfns.
2624	 *
2625	 * Whoever called remap_pfn_range is also going to call e.g.
2626	 * unmap_mapping_range before the underlying pages are freed,
2627	 * causing a call to our MMU notifier.
2628	 *
2629	 * Certain IO or PFNMAP mappings can be backed with valid
2630	 * struct pages, but be allocated without refcounting e.g.,
2631	 * tail pages of non-compound higher order allocations, which
2632	 * would then underflow the refcount when the caller does the
2633	 * required put_page. Don't allow those pages here.
2634	 */ 
2635	if (!kvm_try_get_pfn(pfn))
2636		r = -EFAULT;
2637
2638out:
2639	pte_unmap_unlock(ptep, ptl);
2640	*p_pfn = pfn;
2641
2642	return r;
2643}
2644
2645/*
2646 * Pin guest page in memory and return its pfn.
2647 * @addr: host virtual address which maps memory to the guest
2648 * @atomic: whether this function can sleep
2649 * @interruptible: whether the process can be interrupted by non-fatal signals
2650 * @async: whether this function need to wait IO complete if the
2651 *         host page is not in the memory
2652 * @write_fault: whether we should get a writable host page
2653 * @writable: whether it allows to map a writable host page for !@write_fault
2654 *
2655 * The function will map a writable host page for these two cases:
2656 * 1): @write_fault = true
2657 * 2): @write_fault = false && @writable, @writable will tell the caller
2658 *     whether the mapping is writable.
2659 */
2660kvm_pfn_t hva_to_pfn(unsigned long addr, bool atomic, bool interruptible,
2661		     bool *async, bool write_fault, bool *writable)
2662{
2663	struct vm_area_struct *vma;
2664	kvm_pfn_t pfn;
2665	int npages, r;
2666
2667	/* we can do it either atomically or asynchronously, not both */
2668	BUG_ON(atomic && async);
2669
2670	if (hva_to_pfn_fast(addr, write_fault, writable, &pfn))
2671		return pfn;
2672
2673	if (atomic)
2674		return KVM_PFN_ERR_FAULT;
2675
2676	npages = hva_to_pfn_slow(addr, async, write_fault, interruptible,
2677				 writable, &pfn);
2678	if (npages == 1)
2679		return pfn;
2680	if (npages == -EINTR)
2681		return KVM_PFN_ERR_SIGPENDING;
2682
2683	mmap_read_lock(current->mm);
2684	if (npages == -EHWPOISON ||
2685	      (!async && check_user_page_hwpoison(addr))) {
2686		pfn = KVM_PFN_ERR_HWPOISON;
2687		goto exit;
2688	}
2689
2690retry:
2691	vma = vma_lookup(current->mm, addr);
2692
2693	if (vma == NULL)
2694		pfn = KVM_PFN_ERR_FAULT;
2695	else if (vma->vm_flags & (VM_IO | VM_PFNMAP)) {
2696		r = hva_to_pfn_remapped(vma, addr, write_fault, writable, &pfn);
2697		if (r == -EAGAIN)
2698			goto retry;
2699		if (r < 0)
2700			pfn = KVM_PFN_ERR_FAULT;
2701	} else {
2702		if (async && vma_is_valid(vma, write_fault))
2703			*async = true;
2704		pfn = KVM_PFN_ERR_FAULT;
2705	}
2706exit:
2707	mmap_read_unlock(current->mm);
2708	return pfn;
2709}
2710
2711kvm_pfn_t __gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn,
2712			       bool atomic, bool interruptible, bool *async,
2713			       bool write_fault, bool *writable, hva_t *hva)
2714{
2715	unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
2716
2717	if (hva)
2718		*hva = addr;
2719
2720	if (addr == KVM_HVA_ERR_RO_BAD) {
2721		if (writable)
2722			*writable = false;
2723		return KVM_PFN_ERR_RO_FAULT;
2724	}
2725
2726	if (kvm_is_error_hva(addr)) {
2727		if (writable)
2728			*writable = false;
2729		return KVM_PFN_NOSLOT;
2730	}
2731
2732	/* Do not map writable pfn in the readonly memslot. */
2733	if (writable && memslot_is_readonly(slot)) {
2734		*writable = false;
2735		writable = NULL;
2736	}
2737
2738	return hva_to_pfn(addr, atomic, interruptible, async, write_fault,
2739			  writable);
2740}
2741EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot);
2742
2743kvm_pfn_t gfn_to_pfn_prot(struct kvm *kvm, gfn_t gfn, bool write_fault,
2744		      bool *writable)
2745{
2746	return __gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn, false, false,
2747				    NULL, write_fault, writable, NULL);
2748}
2749EXPORT_SYMBOL_GPL(gfn_to_pfn_prot);
2750
2751kvm_pfn_t gfn_to_pfn_memslot(const struct kvm_memory_slot *slot, gfn_t gfn)
2752{
2753	return __gfn_to_pfn_memslot(slot, gfn, false, false, NULL, true,
2754				    NULL, NULL);
2755}
2756EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot);
2757
2758kvm_pfn_t gfn_to_pfn_memslot_atomic(const struct kvm_memory_slot *slot, gfn_t gfn)
2759{
2760	return __gfn_to_pfn_memslot(slot, gfn, true, false, NULL, true,
2761				    NULL, NULL);
2762}
2763EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic);
2764
2765kvm_pfn_t kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu *vcpu, gfn_t gfn)
2766{
2767	return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2768}
2769EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic);
2770
2771kvm_pfn_t gfn_to_pfn(struct kvm *kvm, gfn_t gfn)
2772{
2773	return gfn_to_pfn_memslot(gfn_to_memslot(kvm, gfn), gfn);
2774}
2775EXPORT_SYMBOL_GPL(gfn_to_pfn);
2776
2777kvm_pfn_t kvm_vcpu_gfn_to_pfn(struct kvm_vcpu *vcpu, gfn_t gfn)
2778{
2779	return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu, gfn), gfn);
2780}
2781EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn);
2782
2783int gfn_to_page_many_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
2784			    struct page **pages, int nr_pages)
2785{
2786	unsigned long addr;
2787	gfn_t entry = 0;
2788
2789	addr = gfn_to_hva_many(slot, gfn, &entry);
2790	if (kvm_is_error_hva(addr))
2791		return -1;
2792
2793	if (entry < nr_pages)
2794		return 0;
2795
2796	return get_user_pages_fast_only(addr, nr_pages, FOLL_WRITE, pages);
2797}
2798EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic);
2799
2800/*
2801 * Do not use this helper unless you are absolutely certain the gfn _must_ be
2802 * backed by 'struct page'.  A valid example is if the backing memslot is
2803 * controlled by KVM.  Note, if the returned page is valid, it's refcount has
2804 * been elevated by gfn_to_pfn().
2805 */
2806struct page *gfn_to_page(struct kvm *kvm, gfn_t gfn)
2807{
2808	struct page *page;
2809	kvm_pfn_t pfn;
2810
2811	pfn = gfn_to_pfn(kvm, gfn);
2812
2813	if (is_error_noslot_pfn(pfn))
2814		return KVM_ERR_PTR_BAD_PAGE;
2815
2816	page = kvm_pfn_to_refcounted_page(pfn);
2817	if (!page)
2818		return KVM_ERR_PTR_BAD_PAGE;
2819
2820	return page;
2821}
2822EXPORT_SYMBOL_GPL(gfn_to_page);
2823
2824void kvm_release_pfn(kvm_pfn_t pfn, bool dirty)
2825{
2826	if (dirty)
2827		kvm_release_pfn_dirty(pfn);
2828	else
2829		kvm_release_pfn_clean(pfn);
2830}
2831
2832int kvm_vcpu_map(struct kvm_vcpu *vcpu, gfn_t gfn, struct kvm_host_map *map)
2833{
2834	kvm_pfn_t pfn;
2835	void *hva = NULL;
2836	struct page *page = KVM_UNMAPPED_PAGE;
2837
2838	if (!map)
2839		return -EINVAL;
2840
2841	pfn = gfn_to_pfn(vcpu->kvm, gfn);
2842	if (is_error_noslot_pfn(pfn))
2843		return -EINVAL;
2844
2845	if (pfn_valid(pfn)) {
2846		page = pfn_to_page(pfn);
2847		hva = kmap(page);
2848#ifdef CONFIG_HAS_IOMEM
2849	} else {
2850		hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE, MEMREMAP_WB);
2851#endif
2852	}
2853
2854	if (!hva)
2855		return -EFAULT;
2856
2857	map->page = page;
2858	map->hva = hva;
2859	map->pfn = pfn;
2860	map->gfn = gfn;
2861
2862	return 0;
2863}
2864EXPORT_SYMBOL_GPL(kvm_vcpu_map);
2865
2866void kvm_vcpu_unmap(struct kvm_vcpu *vcpu, struct kvm_host_map *map, bool dirty)
2867{
2868	if (!map)
2869		return;
2870
2871	if (!map->hva)
2872		return;
2873
2874	if (map->page != KVM_UNMAPPED_PAGE)
2875		kunmap(map->page);
2876#ifdef CONFIG_HAS_IOMEM
2877	else
2878		memunmap(map->hva);
2879#endif
2880
2881	if (dirty)
2882		kvm_vcpu_mark_page_dirty(vcpu, map->gfn);
2883
2884	kvm_release_pfn(map->pfn, dirty);
2885
2886	map->hva = NULL;
2887	map->page = NULL;
2888}
2889EXPORT_SYMBOL_GPL(kvm_vcpu_unmap);
2890
2891static bool kvm_is_ad_tracked_page(struct page *page)
2892{
2893	/*
2894	 * Per page-flags.h, pages tagged PG_reserved "should in general not be
2895	 * touched (e.g. set dirty) except by its owner".
2896	 */
2897	return !PageReserved(page);
2898}
2899
2900static void kvm_set_page_dirty(struct page *page)
2901{
2902	if (kvm_is_ad_tracked_page(page))
2903		SetPageDirty(page);
2904}
2905
2906static void kvm_set_page_accessed(struct page *page)
2907{
2908	if (kvm_is_ad_tracked_page(page))
2909		mark_page_accessed(page);
2910}
2911
2912void kvm_release_page_clean(struct page *page)
2913{
2914	WARN_ON(is_error_page(page));
2915
2916	kvm_set_page_accessed(page);
2917	put_page(page);
2918}
2919EXPORT_SYMBOL_GPL(kvm_release_page_clean);
2920
2921void kvm_release_pfn_clean(kvm_pfn_t pfn)
2922{
2923	struct page *page;
2924
2925	if (is_error_noslot_pfn(pfn))
2926		return;
2927
2928	page = kvm_pfn_to_refcounted_page(pfn);
2929	if (!page)
2930		return;
2931
2932	kvm_release_page_clean(page);
2933}
2934EXPORT_SYMBOL_GPL(kvm_release_pfn_clean);
2935
2936void kvm_release_page_dirty(struct page *page)
2937{
2938	WARN_ON(is_error_page(page));
2939
2940	kvm_set_page_dirty(page);
2941	kvm_release_page_clean(page);
2942}
2943EXPORT_SYMBOL_GPL(kvm_release_page_dirty);
2944
2945void kvm_release_pfn_dirty(kvm_pfn_t pfn)
2946{
2947	struct page *page;
2948
2949	if (is_error_noslot_pfn(pfn))
2950		return;
2951
2952	page = kvm_pfn_to_refcounted_page(pfn);
2953	if (!page)
2954		return;
2955
2956	kvm_release_page_dirty(page);
2957}
2958EXPORT_SYMBOL_GPL(kvm_release_pfn_dirty);
2959
2960/*
2961 * Note, checking for an error/noslot pfn is the caller's responsibility when
2962 * directly marking a page dirty/accessed.  Unlike the "release" helpers, the
2963 * "set" helpers are not to be used when the pfn might point at garbage.
2964 */
2965void kvm_set_pfn_dirty(kvm_pfn_t pfn)
2966{
2967	if (WARN_ON(is_error_noslot_pfn(pfn)))
2968		return;
2969
2970	if (pfn_valid(pfn))
2971		kvm_set_page_dirty(pfn_to_page(pfn));
2972}
2973EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty);
2974
2975void kvm_set_pfn_accessed(kvm_pfn_t pfn)
2976{
2977	if (WARN_ON(is_error_noslot_pfn(pfn)))
2978		return;
2979
2980	if (pfn_valid(pfn))
2981		kvm_set_page_accessed(pfn_to_page(pfn));
2982}
2983EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed);
2984
2985static int next_segment(unsigned long len, int offset)
2986{
2987	if (len > PAGE_SIZE - offset)
2988		return PAGE_SIZE - offset;
2989	else
2990		return len;
2991}
2992
2993static int __kvm_read_guest_page(struct kvm_memory_slot *slot, gfn_t gfn,
2994				 void *data, int offset, int len)
2995{
2996	int r;
2997	unsigned long addr;
2998
2999	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3000	if (kvm_is_error_hva(addr))
3001		return -EFAULT;
3002	r = __copy_from_user(data, (void __user *)addr + offset, len);
3003	if (r)
3004		return -EFAULT;
3005	return 0;
3006}
3007
3008int kvm_read_guest_page(struct kvm *kvm, gfn_t gfn, void *data, int offset,
3009			int len)
3010{
3011	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3012
3013	return __kvm_read_guest_page(slot, gfn, data, offset, len);
3014}
3015EXPORT_SYMBOL_GPL(kvm_read_guest_page);
3016
3017int kvm_vcpu_read_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn, void *data,
3018			     int offset, int len)
3019{
3020	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3021
3022	return __kvm_read_guest_page(slot, gfn, data, offset, len);
3023}
3024EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page);
3025
3026int kvm_read_guest(struct kvm *kvm, gpa_t gpa, void *data, unsigned long len)
3027{
3028	gfn_t gfn = gpa >> PAGE_SHIFT;
3029	int seg;
3030	int offset = offset_in_page(gpa);
3031	int ret;
3032
3033	while ((seg = next_segment(len, offset)) != 0) {
3034		ret = kvm_read_guest_page(kvm, gfn, data, offset, seg);
3035		if (ret < 0)
3036			return ret;
3037		offset = 0;
3038		len -= seg;
3039		data += seg;
3040		++gfn;
3041	}
3042	return 0;
3043}
3044EXPORT_SYMBOL_GPL(kvm_read_guest);
3045
3046int kvm_vcpu_read_guest(struct kvm_vcpu *vcpu, gpa_t gpa, void *data, unsigned long len)
3047{
3048	gfn_t gfn = gpa >> PAGE_SHIFT;
3049	int seg;
3050	int offset = offset_in_page(gpa);
3051	int ret;
3052
3053	while ((seg = next_segment(len, offset)) != 0) {
3054		ret = kvm_vcpu_read_guest_page(vcpu, gfn, data, offset, seg);
3055		if (ret < 0)
3056			return ret;
3057		offset = 0;
3058		len -= seg;
3059		data += seg;
3060		++gfn;
3061	}
3062	return 0;
3063}
3064EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest);
3065
3066static int __kvm_read_guest_atomic(struct kvm_memory_slot *slot, gfn_t gfn,
3067			           void *data, int offset, unsigned long len)
3068{
3069	int r;
3070	unsigned long addr;
3071
3072	addr = gfn_to_hva_memslot_prot(slot, gfn, NULL);
3073	if (kvm_is_error_hva(addr))
3074		return -EFAULT;
3075	pagefault_disable();
3076	r = __copy_from_user_inatomic(data, (void __user *)addr + offset, len);
3077	pagefault_enable();
3078	if (r)
3079		return -EFAULT;
3080	return 0;
3081}
3082
3083int kvm_vcpu_read_guest_atomic(struct kvm_vcpu *vcpu, gpa_t gpa,
3084			       void *data, unsigned long len)
3085{
3086	gfn_t gfn = gpa >> PAGE_SHIFT;
3087	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3088	int offset = offset_in_page(gpa);
3089
3090	return __kvm_read_guest_atomic(slot, gfn, data, offset, len);
3091}
3092EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic);
3093
3094static int __kvm_write_guest_page(struct kvm *kvm,
3095				  struct kvm_memory_slot *memslot, gfn_t gfn,
3096			          const void *data, int offset, int len)
3097{
3098	int r;
3099	unsigned long addr;
3100
3101	addr = gfn_to_hva_memslot(memslot, gfn);
3102	if (kvm_is_error_hva(addr))
3103		return -EFAULT;
3104	r = __copy_to_user((void __user *)addr + offset, data, len);
3105	if (r)
3106		return -EFAULT;
3107	mark_page_dirty_in_slot(kvm, memslot, gfn);
3108	return 0;
3109}
3110
3111int kvm_write_guest_page(struct kvm *kvm, gfn_t gfn,
3112			 const void *data, int offset, int len)
3113{
3114	struct kvm_memory_slot *slot = gfn_to_memslot(kvm, gfn);
3115
3116	return __kvm_write_guest_page(kvm, slot, gfn, data, offset, len);
3117}
3118EXPORT_SYMBOL_GPL(kvm_write_guest_page);
3119
3120int kvm_vcpu_write_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
3121			      const void *data, int offset, int len)
3122{
3123	struct kvm_memory_slot *slot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3124
3125	return __kvm_write_guest_page(vcpu->kvm, slot, gfn, data, offset, len);
3126}
3127EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page);
3128
3129int kvm_write_guest(struct kvm *kvm, gpa_t gpa, const void *data,
3130		    unsigned long len)
3131{
3132	gfn_t gfn = gpa >> PAGE_SHIFT;
3133	int seg;
3134	int offset = offset_in_page(gpa);
3135	int ret;
3136
3137	while ((seg = next_segment(len, offset)) != 0) {
3138		ret = kvm_write_guest_page(kvm, gfn, data, offset, seg);
3139		if (ret < 0)
3140			return ret;
3141		offset = 0;
3142		len -= seg;
3143		data += seg;
3144		++gfn;
3145	}
3146	return 0;
3147}
3148EXPORT_SYMBOL_GPL(kvm_write_guest);
3149
3150int kvm_vcpu_write_guest(struct kvm_vcpu *vcpu, gpa_t gpa, const void *data,
3151		         unsigned long len)
3152{
3153	gfn_t gfn = gpa >> PAGE_SHIFT;
3154	int seg;
3155	int offset = offset_in_page(gpa);
3156	int ret;
3157
3158	while ((seg = next_segment(len, offset)) != 0) {
3159		ret = kvm_vcpu_write_guest_page(vcpu, gfn, data, offset, seg);
3160		if (ret < 0)
3161			return ret;
3162		offset = 0;
3163		len -= seg;
3164		data += seg;
3165		++gfn;
3166	}
3167	return 0;
3168}
3169EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest);
3170
3171static int __kvm_gfn_to_hva_cache_init(struct kvm_memslots *slots,
3172				       struct gfn_to_hva_cache *ghc,
3173				       gpa_t gpa, unsigned long len)
3174{
3175	int offset = offset_in_page(gpa);
3176	gfn_t start_gfn = gpa >> PAGE_SHIFT;
3177	gfn_t end_gfn = (gpa + len - 1) >> PAGE_SHIFT;
3178	gfn_t nr_pages_needed = end_gfn - start_gfn + 1;
3179	gfn_t nr_pages_avail;
3180
3181	/* Update ghc->generation before performing any error checks. */
3182	ghc->generation = slots->generation;
3183
3184	if (start_gfn > end_gfn) {
3185		ghc->hva = KVM_HVA_ERR_BAD;
3186		return -EINVAL;
3187	}
3188
3189	/*
3190	 * If the requested region crosses two memslots, we still
3191	 * verify that the entire region is valid here.
3192	 */
3193	for ( ; start_gfn <= end_gfn; start_gfn += nr_pages_avail) {
3194		ghc->memslot = __gfn_to_memslot(slots, start_gfn);
3195		ghc->hva = gfn_to_hva_many(ghc->memslot, start_gfn,
3196					   &nr_pages_avail);
3197		if (kvm_is_error_hva(ghc->hva))
3198			return -EFAULT;
3199	}
3200
3201	/* Use the slow path for cross page reads and writes. */
3202	if (nr_pages_needed == 1)
3203		ghc->hva += offset;
3204	else
3205		ghc->memslot = NULL;
3206
3207	ghc->gpa = gpa;
3208	ghc->len = len;
3209	return 0;
3210}
3211
3212int kvm_gfn_to_hva_cache_init(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3213			      gpa_t gpa, unsigned long len)
3214{
3215	struct kvm_memslots *slots = kvm_memslots(kvm);
3216	return __kvm_gfn_to_hva_cache_init(slots, ghc, gpa, len);
3217}
3218EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init);
3219
3220int kvm_write_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3221				  void *data, unsigned int offset,
3222				  unsigned long len)
3223{
3224	struct kvm_memslots *slots = kvm_memslots(kvm);
3225	int r;
3226	gpa_t gpa = ghc->gpa + offset;
3227
3228	if (WARN_ON_ONCE(len + offset > ghc->len))
3229		return -EINVAL;
3230
3231	if (slots->generation != ghc->generation) {
3232		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3233			return -EFAULT;
3234	}
3235
3236	if (kvm_is_error_hva(ghc->hva))
3237		return -EFAULT;
3238
3239	if (unlikely(!ghc->memslot))
3240		return kvm_write_guest(kvm, gpa, data, len);
3241
3242	r = __copy_to_user((void __user *)ghc->hva + offset, data, len);
3243	if (r)
3244		return -EFAULT;
3245	mark_page_dirty_in_slot(kvm, ghc->memslot, gpa >> PAGE_SHIFT);
3246
3247	return 0;
3248}
3249EXPORT_SYMBOL_GPL(kvm_write_guest_offset_cached);
3250
3251int kvm_write_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3252			   void *data, unsigned long len)
3253{
3254	return kvm_write_guest_offset_cached(kvm, ghc, data, 0, len);
3255}
3256EXPORT_SYMBOL_GPL(kvm_write_guest_cached);
3257
3258int kvm_read_guest_offset_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3259				 void *data, unsigned int offset,
3260				 unsigned long len)
3261{
3262	struct kvm_memslots *slots = kvm_memslots(kvm);
3263	int r;
3264	gpa_t gpa = ghc->gpa + offset;
3265
3266	if (WARN_ON_ONCE(len + offset > ghc->len))
3267		return -EINVAL;
3268
3269	if (slots->generation != ghc->generation) {
3270		if (__kvm_gfn_to_hva_cache_init(slots, ghc, ghc->gpa, ghc->len))
3271			return -EFAULT;
3272	}
3273
3274	if (kvm_is_error_hva(ghc->hva))
3275		return -EFAULT;
3276
3277	if (unlikely(!ghc->memslot))
3278		return kvm_read_guest(kvm, gpa, data, len);
3279
3280	r = __copy_from_user(data, (void __user *)ghc->hva + offset, len);
3281	if (r)
3282		return -EFAULT;
3283
3284	return 0;
3285}
3286EXPORT_SYMBOL_GPL(kvm_read_guest_offset_cached);
3287
3288int kvm_read_guest_cached(struct kvm *kvm, struct gfn_to_hva_cache *ghc,
3289			  void *data, unsigned long len)
3290{
3291	return kvm_read_guest_offset_cached(kvm, ghc, data, 0, len);
3292}
3293EXPORT_SYMBOL_GPL(kvm_read_guest_cached);
3294
3295int kvm_clear_guest(struct kvm *kvm, gpa_t gpa, unsigned long len)
3296{
3297	const void *zero_page = (const void *) __va(page_to_phys(ZERO_PAGE(0)));
3298	gfn_t gfn = gpa >> PAGE_SHIFT;
3299	int seg;
3300	int offset = offset_in_page(gpa);
3301	int ret;
3302
3303	while ((seg = next_segment(len, offset)) != 0) {
3304		ret = kvm_write_guest_page(kvm, gfn, zero_page, offset, len);
3305		if (ret < 0)
3306			return ret;
3307		offset = 0;
3308		len -= seg;
3309		++gfn;
3310	}
3311	return 0;
3312}
3313EXPORT_SYMBOL_GPL(kvm_clear_guest);
3314
3315void mark_page_dirty_in_slot(struct kvm *kvm,
3316			     const struct kvm_memory_slot *memslot,
3317		 	     gfn_t gfn)
3318{
3319	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
3320
3321#ifdef CONFIG_HAVE_KVM_DIRTY_RING
3322	if (WARN_ON_ONCE(vcpu && vcpu->kvm != kvm))
3323		return;
3324
3325	WARN_ON_ONCE(!vcpu && !kvm_arch_allow_write_without_running_vcpu(kvm));
3326#endif
3327
3328	if (memslot && kvm_slot_dirty_track_enabled(memslot)) {
3329		unsigned long rel_gfn = gfn - memslot->base_gfn;
3330		u32 slot = (memslot->as_id << 16) | memslot->id;
3331
3332		if (kvm->dirty_ring_size && vcpu)
3333			kvm_dirty_ring_push(vcpu, slot, rel_gfn);
3334		else if (memslot->dirty_bitmap)
3335			set_bit_le(rel_gfn, memslot->dirty_bitmap);
3336	}
3337}
3338EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
3339
3340void mark_page_dirty(struct kvm *kvm, gfn_t gfn)
3341{
3342	struct kvm_memory_slot *memslot;
3343
3344	memslot = gfn_to_memslot(kvm, gfn);
3345	mark_page_dirty_in_slot(kvm, memslot, gfn);
3346}
3347EXPORT_SYMBOL_GPL(mark_page_dirty);
3348
3349void kvm_vcpu_mark_page_dirty(struct kvm_vcpu *vcpu, gfn_t gfn)
3350{
3351	struct kvm_memory_slot *memslot;
3352
3353	memslot = kvm_vcpu_gfn_to_memslot(vcpu, gfn);
3354	mark_page_dirty_in_slot(vcpu->kvm, memslot, gfn);
3355}
3356EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty);
3357
3358void kvm_sigset_activate(struct kvm_vcpu *vcpu)
3359{
3360	if (!vcpu->sigset_active)
3361		return;
3362
3363	/*
3364	 * This does a lockless modification of ->real_blocked, which is fine
3365	 * because, only current can change ->real_blocked and all readers of
3366	 * ->real_blocked don't care as long ->real_blocked is always a subset
3367	 * of ->blocked.
3368	 */
3369	sigprocmask(SIG_SETMASK, &vcpu->sigset, &current->real_blocked);
3370}
3371
3372void kvm_sigset_deactivate(struct kvm_vcpu *vcpu)
3373{
3374	if (!vcpu->sigset_active)
3375		return;
3376
3377	sigprocmask(SIG_SETMASK, &current->real_blocked, NULL);
3378	sigemptyset(&current->real_blocked);
3379}
3380
3381static void grow_halt_poll_ns(struct kvm_vcpu *vcpu)
3382{
3383	unsigned int old, val, grow, grow_start;
3384
3385	old = val = vcpu->halt_poll_ns;
3386	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3387	grow = READ_ONCE(halt_poll_ns_grow);
3388	if (!grow)
3389		goto out;
3390
3391	val *= grow;
3392	if (val < grow_start)
3393		val = grow_start;
3394
3395	vcpu->halt_poll_ns = val;
3396out:
3397	trace_kvm_halt_poll_ns_grow(vcpu->vcpu_id, val, old);
3398}
3399
3400static void shrink_halt_poll_ns(struct kvm_vcpu *vcpu)
3401{
3402	unsigned int old, val, shrink, grow_start;
3403
3404	old = val = vcpu->halt_poll_ns;
3405	shrink = READ_ONCE(halt_poll_ns_shrink);
3406	grow_start = READ_ONCE(halt_poll_ns_grow_start);
3407	if (shrink == 0)
3408		val = 0;
3409	else
3410		val /= shrink;
3411
3412	if (val < grow_start)
3413		val = 0;
3414
3415	vcpu->halt_poll_ns = val;
3416	trace_kvm_halt_poll_ns_shrink(vcpu->vcpu_id, val, old);
3417}
3418
3419static int kvm_vcpu_check_block(struct kvm_vcpu *vcpu)
3420{
3421	int ret = -EINTR;
3422	int idx = srcu_read_lock(&vcpu->kvm->srcu);
3423
3424	if (kvm_arch_vcpu_runnable(vcpu))
3425		goto out;
3426	if (kvm_cpu_has_pending_timer(vcpu))
3427		goto out;
3428	if (signal_pending(current))
3429		goto out;
3430	if (kvm_check_request(KVM_REQ_UNBLOCK, vcpu))
3431		goto out;
3432
3433	ret = 0;
3434out:
3435	srcu_read_unlock(&vcpu->kvm->srcu, idx);
3436	return ret;
3437}
3438
3439/*
3440 * Block the vCPU until the vCPU is runnable, an event arrives, or a signal is
3441 * pending.  This is mostly used when halting a vCPU, but may also be used
3442 * directly for other vCPU non-runnable states, e.g. x86's Wait-For-SIPI.
3443 */
3444bool kvm_vcpu_block(struct kvm_vcpu *vcpu)
3445{
3446	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
3447	bool waited = false;
3448
3449	vcpu->stat.generic.blocking = 1;
3450
3451	preempt_disable();
3452	kvm_arch_vcpu_blocking(vcpu);
3453	prepare_to_rcuwait(wait);
3454	preempt_enable();
3455
3456	for (;;) {
3457		set_current_state(TASK_INTERRUPTIBLE);
3458
3459		if (kvm_vcpu_check_block(vcpu) < 0)
3460			break;
3461
3462		waited = true;
3463		schedule();
3464	}
3465
3466	preempt_disable();
3467	finish_rcuwait(wait);
3468	kvm_arch_vcpu_unblocking(vcpu);
3469	preempt_enable();
3470
3471	vcpu->stat.generic.blocking = 0;
3472
3473	return waited;
3474}
3475
3476static inline void update_halt_poll_stats(struct kvm_vcpu *vcpu, ktime_t start,
3477					  ktime_t end, bool success)
3478{
3479	struct kvm_vcpu_stat_generic *stats = &vcpu->stat.generic;
3480	u64 poll_ns = ktime_to_ns(ktime_sub(end, start));
3481
3482	++vcpu->stat.generic.halt_attempted_poll;
3483
3484	if (success) {
3485		++vcpu->stat.generic.halt_successful_poll;
3486
3487		if (!vcpu_valid_wakeup(vcpu))
3488			++vcpu->stat.generic.halt_poll_invalid;
3489
3490		stats->halt_poll_success_ns += poll_ns;
3491		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_success_hist, poll_ns);
3492	} else {
3493		stats->halt_poll_fail_ns += poll_ns;
3494		KVM_STATS_LOG_HIST_UPDATE(stats->halt_poll_fail_hist, poll_ns);
3495	}
3496}
3497
3498static unsigned int kvm_vcpu_max_halt_poll_ns(struct kvm_vcpu *vcpu)
3499{
3500	struct kvm *kvm = vcpu->kvm;
3501
3502	if (kvm->override_halt_poll_ns) {
3503		/*
3504		 * Ensure kvm->max_halt_poll_ns is not read before
3505		 * kvm->override_halt_poll_ns.
3506		 *
3507		 * Pairs with the smp_wmb() when enabling KVM_CAP_HALT_POLL.
3508		 */
3509		smp_rmb();
3510		return READ_ONCE(kvm->max_halt_poll_ns);
3511	}
3512
3513	return READ_ONCE(halt_poll_ns);
3514}
3515
3516/*
3517 * Emulate a vCPU halt condition, e.g. HLT on x86, WFI on arm, etc...  If halt
3518 * polling is enabled, busy wait for a short time before blocking to avoid the
3519 * expensive block+unblock sequence if a wake event arrives soon after the vCPU
3520 * is halted.
3521 */
3522void kvm_vcpu_halt(struct kvm_vcpu *vcpu)
3523{
3524	unsigned int max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3525	bool halt_poll_allowed = !kvm_arch_no_poll(vcpu);
3526	ktime_t start, cur, poll_end;
3527	bool waited = false;
3528	bool do_halt_poll;
3529	u64 halt_ns;
3530
3531	if (vcpu->halt_poll_ns > max_halt_poll_ns)
3532		vcpu->halt_poll_ns = max_halt_poll_ns;
3533
3534	do_halt_poll = halt_poll_allowed && vcpu->halt_poll_ns;
3535
3536	start = cur = poll_end = ktime_get();
3537	if (do_halt_poll) {
3538		ktime_t stop = ktime_add_ns(start, vcpu->halt_poll_ns);
3539
3540		do {
3541			if (kvm_vcpu_check_block(vcpu) < 0)
3542				goto out;
3543			cpu_relax();
3544			poll_end = cur = ktime_get();
3545		} while (kvm_vcpu_can_poll(cur, stop));
3546	}
3547
3548	waited = kvm_vcpu_block(vcpu);
3549
3550	cur = ktime_get();
3551	if (waited) {
3552		vcpu->stat.generic.halt_wait_ns +=
3553			ktime_to_ns(cur) - ktime_to_ns(poll_end);
3554		KVM_STATS_LOG_HIST_UPDATE(vcpu->stat.generic.halt_wait_hist,
3555				ktime_to_ns(cur) - ktime_to_ns(poll_end));
3556	}
3557out:
3558	/* The total time the vCPU was "halted", including polling time. */
3559	halt_ns = ktime_to_ns(cur) - ktime_to_ns(start);
3560
3561	/*
3562	 * Note, halt-polling is considered successful so long as the vCPU was
3563	 * never actually scheduled out, i.e. even if the wake event arrived
3564	 * after of the halt-polling loop itself, but before the full wait.
3565	 */
3566	if (do_halt_poll)
3567		update_halt_poll_stats(vcpu, start, poll_end, !waited);
3568
3569	if (halt_poll_allowed) {
3570		/* Recompute the max halt poll time in case it changed. */
3571		max_halt_poll_ns = kvm_vcpu_max_halt_poll_ns(vcpu);
3572
3573		if (!vcpu_valid_wakeup(vcpu)) {
3574			shrink_halt_poll_ns(vcpu);
3575		} else if (max_halt_poll_ns) {
3576			if (halt_ns <= vcpu->halt_poll_ns)
3577				;
3578			/* we had a long block, shrink polling */
3579			else if (vcpu->halt_poll_ns &&
3580				 halt_ns > max_halt_poll_ns)
3581				shrink_halt_poll_ns(vcpu);
3582			/* we had a short halt and our poll time is too small */
3583			else if (vcpu->halt_poll_ns < max_halt_poll_ns &&
3584				 halt_ns < max_halt_poll_ns)
3585				grow_halt_poll_ns(vcpu);
3586		} else {
3587			vcpu->halt_poll_ns = 0;
3588		}
3589	}
3590
3591	trace_kvm_vcpu_wakeup(halt_ns, waited, vcpu_valid_wakeup(vcpu));
3592}
3593EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
3594
3595bool kvm_vcpu_wake_up(struct kvm_vcpu *vcpu)
3596{
3597	if (__kvm_vcpu_wake_up(vcpu)) {
3598		WRITE_ONCE(vcpu->ready, true);
3599		++vcpu->stat.generic.halt_wakeup;
3600		return true;
3601	}
3602
3603	return false;
3604}
3605EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up);
3606
3607#ifndef CONFIG_S390
3608/*
3609 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
3610 */
3611void kvm_vcpu_kick(struct kvm_vcpu *vcpu)
3612{
3613	int me, cpu;
3614
3615	if (kvm_vcpu_wake_up(vcpu))
3616		return;
3617
3618	me = get_cpu();
3619	/*
3620	 * The only state change done outside the vcpu mutex is IN_GUEST_MODE
3621	 * to EXITING_GUEST_MODE.  Therefore the moderately expensive "should
3622	 * kick" check does not need atomic operations if kvm_vcpu_kick is used
3623	 * within the vCPU thread itself.
3624	 */
3625	if (vcpu == __this_cpu_read(kvm_running_vcpu)) {
3626		if (vcpu->mode == IN_GUEST_MODE)
3627			WRITE_ONCE(vcpu->mode, EXITING_GUEST_MODE);
3628		goto out;
3629	}
3630
3631	/*
3632	 * Note, the vCPU could get migrated to a different pCPU at any point
3633	 * after kvm_arch_vcpu_should_kick(), which could result in sending an
3634	 * IPI to the previous pCPU.  But, that's ok because the purpose of the
3635	 * IPI is to force the vCPU to leave IN_GUEST_MODE, and migrating the
3636	 * vCPU also requires it to leave IN_GUEST_MODE.
3637	 */
3638	if (kvm_arch_vcpu_should_kick(vcpu)) {
3639		cpu = READ_ONCE(vcpu->cpu);
3640		if (cpu != me && (unsigned)cpu < nr_cpu_ids && cpu_online(cpu))
3641			smp_send_reschedule(cpu);
3642	}
3643out:
3644	put_cpu();
3645}
3646EXPORT_SYMBOL_GPL(kvm_vcpu_kick);
3647#endif /* !CONFIG_S390 */
3648
3649int kvm_vcpu_yield_to(struct kvm_vcpu *target)
3650{
3651	struct pid *pid;
3652	struct task_struct *task = NULL;
3653	int ret = 0;
3654
3655	rcu_read_lock();
3656	pid = rcu_dereference(target->pid);
3657	if (pid)
3658		task = get_pid_task(pid, PIDTYPE_PID);
3659	rcu_read_unlock();
3660	if (!task)
3661		return ret;
3662	ret = yield_to(task, 1);
3663	put_task_struct(task);
3664
3665	return ret;
3666}
3667EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to);
3668
3669/*
3670 * Helper that checks whether a VCPU is eligible for directed yield.
3671 * Most eligible candidate to yield is decided by following heuristics:
3672 *
3673 *  (a) VCPU which has not done pl-exit or cpu relax intercepted recently
3674 *  (preempted lock holder), indicated by @in_spin_loop.
3675 *  Set at the beginning and cleared at the end of interception/PLE handler.
3676 *
3677 *  (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
3678 *  chance last time (mostly it has become eligible now since we have probably
3679 *  yielded to lockholder in last iteration. This is done by toggling
3680 *  @dy_eligible each time a VCPU checked for eligibility.)
3681 *
3682 *  Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
3683 *  to preempted lock-holder could result in wrong VCPU selection and CPU
3684 *  burning. Giving priority for a potential lock-holder increases lock
3685 *  progress.
3686 *
3687 *  Since algorithm is based on heuristics, accessing another VCPU data without
3688 *  locking does not harm. It may result in trying to yield to  same VCPU, fail
3689 *  and continue with next VCPU and so on.
3690 */
3691static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu *vcpu)
3692{
3693#ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
3694	bool eligible;
3695
3696	eligible = !vcpu->spin_loop.in_spin_loop ||
3697		    vcpu->spin_loop.dy_eligible;
3698
3699	if (vcpu->spin_loop.in_spin_loop)
3700		kvm_vcpu_set_dy_eligible(vcpu, !vcpu->spin_loop.dy_eligible);
3701
3702	return eligible;
3703#else
3704	return true;
3705#endif
3706}
3707
3708/*
3709 * Unlike kvm_arch_vcpu_runnable, this function is called outside
3710 * a vcpu_load/vcpu_put pair.  However, for most architectures
3711 * kvm_arch_vcpu_runnable does not require vcpu_load.
3712 */
3713bool __weak kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
3714{
3715	return kvm_arch_vcpu_runnable(vcpu);
3716}
3717
3718static bool vcpu_dy_runnable(struct kvm_vcpu *vcpu)
3719{
3720	if (kvm_arch_dy_runnable(vcpu))
3721		return true;
3722
3723#ifdef CONFIG_KVM_ASYNC_PF
3724	if (!list_empty_careful(&vcpu->async_pf.done))
3725		return true;
3726#endif
3727
3728	return false;
3729}
3730
 
 
 
 
 
 
 
 
 
 
 
 
3731bool __weak kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
3732{
3733	return false;
3734}
3735
3736void kvm_vcpu_on_spin(struct kvm_vcpu *me, bool yield_to_kernel_mode)
3737{
3738	struct kvm *kvm = me->kvm;
3739	struct kvm_vcpu *vcpu;
3740	int last_boosted_vcpu = me->kvm->last_boosted_vcpu;
3741	unsigned long i;
3742	int yielded = 0;
3743	int try = 3;
3744	int pass;
3745
3746	kvm_vcpu_set_in_spin_loop(me, true);
3747	/*
3748	 * We boost the priority of a VCPU that is runnable but not
3749	 * currently running, because it got preempted by something
3750	 * else and called schedule in __vcpu_run.  Hopefully that
3751	 * VCPU is holding the lock that we need and will release it.
3752	 * We approximate round-robin by starting at the last boosted VCPU.
3753	 */
3754	for (pass = 0; pass < 2 && !yielded && try; pass++) {
3755		kvm_for_each_vcpu(i, vcpu, kvm) {
3756			if (!pass && i <= last_boosted_vcpu) {
3757				i = last_boosted_vcpu;
3758				continue;
3759			} else if (pass && i > last_boosted_vcpu)
3760				break;
3761			if (!READ_ONCE(vcpu->ready))
3762				continue;
3763			if (vcpu == me)
3764				continue;
3765			if (kvm_vcpu_is_blocking(vcpu) && !vcpu_dy_runnable(vcpu))
3766				continue;
 
 
 
 
 
 
 
3767			if (READ_ONCE(vcpu->preempted) && yield_to_kernel_mode &&
3768			    !kvm_arch_dy_has_pending_interrupt(vcpu) &&
3769			    !kvm_arch_vcpu_in_kernel(vcpu))
3770				continue;
3771			if (!kvm_vcpu_eligible_for_directed_yield(vcpu))
3772				continue;
3773
3774			yielded = kvm_vcpu_yield_to(vcpu);
3775			if (yielded > 0) {
3776				kvm->last_boosted_vcpu = i;
3777				break;
3778			} else if (yielded < 0) {
3779				try--;
3780				if (!try)
3781					break;
3782			}
3783		}
3784	}
3785	kvm_vcpu_set_in_spin_loop(me, false);
3786
3787	/* Ensure vcpu is not eligible during next spinloop */
3788	kvm_vcpu_set_dy_eligible(me, false);
3789}
3790EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
3791
3792static bool kvm_page_in_dirty_ring(struct kvm *kvm, unsigned long pgoff)
3793{
3794#ifdef CONFIG_HAVE_KVM_DIRTY_RING
3795	return (pgoff >= KVM_DIRTY_LOG_PAGE_OFFSET) &&
3796	    (pgoff < KVM_DIRTY_LOG_PAGE_OFFSET +
3797	     kvm->dirty_ring_size / PAGE_SIZE);
3798#else
3799	return false;
3800#endif
3801}
3802
3803static vm_fault_t kvm_vcpu_fault(struct vm_fault *vmf)
3804{
3805	struct kvm_vcpu *vcpu = vmf->vma->vm_file->private_data;
3806	struct page *page;
3807
3808	if (vmf->pgoff == 0)
3809		page = virt_to_page(vcpu->run);
3810#ifdef CONFIG_X86
3811	else if (vmf->pgoff == KVM_PIO_PAGE_OFFSET)
3812		page = virt_to_page(vcpu->arch.pio_data);
3813#endif
3814#ifdef CONFIG_KVM_MMIO
3815	else if (vmf->pgoff == KVM_COALESCED_MMIO_PAGE_OFFSET)
3816		page = virt_to_page(vcpu->kvm->coalesced_mmio_ring);
3817#endif
3818	else if (kvm_page_in_dirty_ring(vcpu->kvm, vmf->pgoff))
3819		page = kvm_dirty_ring_get_page(
3820		    &vcpu->dirty_ring,
3821		    vmf->pgoff - KVM_DIRTY_LOG_PAGE_OFFSET);
3822	else
3823		return kvm_arch_vcpu_fault(vcpu, vmf);
3824	get_page(page);
3825	vmf->page = page;
3826	return 0;
3827}
3828
3829static const struct vm_operations_struct kvm_vcpu_vm_ops = {
3830	.fault = kvm_vcpu_fault,
3831};
3832
3833static int kvm_vcpu_mmap(struct file *file, struct vm_area_struct *vma)
3834{
3835	struct kvm_vcpu *vcpu = file->private_data;
3836	unsigned long pages = vma_pages(vma);
3837
3838	if ((kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff) ||
3839	     kvm_page_in_dirty_ring(vcpu->kvm, vma->vm_pgoff + pages - 1)) &&
3840	    ((vma->vm_flags & VM_EXEC) || !(vma->vm_flags & VM_SHARED)))
3841		return -EINVAL;
3842
3843	vma->vm_ops = &kvm_vcpu_vm_ops;
3844	return 0;
3845}
3846
3847static int kvm_vcpu_release(struct inode *inode, struct file *filp)
3848{
3849	struct kvm_vcpu *vcpu = filp->private_data;
3850
3851	kvm_put_kvm(vcpu->kvm);
3852	return 0;
3853}
3854
3855static const struct file_operations kvm_vcpu_fops = {
3856	.release        = kvm_vcpu_release,
3857	.unlocked_ioctl = kvm_vcpu_ioctl,
3858	.mmap           = kvm_vcpu_mmap,
3859	.llseek		= noop_llseek,
3860	KVM_COMPAT(kvm_vcpu_compat_ioctl),
3861};
3862
3863/*
3864 * Allocates an inode for the vcpu.
3865 */
3866static int create_vcpu_fd(struct kvm_vcpu *vcpu)
3867{
3868	char name[8 + 1 + ITOA_MAX_LEN + 1];
3869
3870	snprintf(name, sizeof(name), "kvm-vcpu:%d", vcpu->vcpu_id);
3871	return anon_inode_getfd(name, &kvm_vcpu_fops, vcpu, O_RDWR | O_CLOEXEC);
3872}
3873
3874#ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
3875static int vcpu_get_pid(void *data, u64 *val)
3876{
3877	struct kvm_vcpu *vcpu = (struct kvm_vcpu *) data;
3878	*val = pid_nr(rcu_access_pointer(vcpu->pid));
 
 
 
3879	return 0;
3880}
3881
3882DEFINE_SIMPLE_ATTRIBUTE(vcpu_get_pid_fops, vcpu_get_pid, NULL, "%llu\n");
3883
3884static void kvm_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
3885{
3886	struct dentry *debugfs_dentry;
3887	char dir_name[ITOA_MAX_LEN * 2];
3888
3889	if (!debugfs_initialized())
3890		return;
3891
3892	snprintf(dir_name, sizeof(dir_name), "vcpu%d", vcpu->vcpu_id);
3893	debugfs_dentry = debugfs_create_dir(dir_name,
3894					    vcpu->kvm->debugfs_dentry);
3895	debugfs_create_file("pid", 0444, debugfs_dentry, vcpu,
3896			    &vcpu_get_pid_fops);
3897
3898	kvm_arch_create_vcpu_debugfs(vcpu, debugfs_dentry);
3899}
3900#endif
3901
3902/*
3903 * Creates some virtual cpus.  Good luck creating more than one.
3904 */
3905static int kvm_vm_ioctl_create_vcpu(struct kvm *kvm, u32 id)
3906{
3907	int r;
3908	struct kvm_vcpu *vcpu;
3909	struct page *page;
3910
3911	if (id >= KVM_MAX_VCPU_IDS)
3912		return -EINVAL;
3913
3914	mutex_lock(&kvm->lock);
3915	if (kvm->created_vcpus >= kvm->max_vcpus) {
3916		mutex_unlock(&kvm->lock);
3917		return -EINVAL;
3918	}
3919
3920	r = kvm_arch_vcpu_precreate(kvm, id);
3921	if (r) {
3922		mutex_unlock(&kvm->lock);
3923		return r;
3924	}
3925
3926	kvm->created_vcpus++;
3927	mutex_unlock(&kvm->lock);
3928
3929	vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL_ACCOUNT);
3930	if (!vcpu) {
3931		r = -ENOMEM;
3932		goto vcpu_decrement;
3933	}
3934
3935	BUILD_BUG_ON(sizeof(struct kvm_run) > PAGE_SIZE);
3936	page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
3937	if (!page) {
3938		r = -ENOMEM;
3939		goto vcpu_free;
3940	}
3941	vcpu->run = page_address(page);
3942
3943	kvm_vcpu_init(vcpu, kvm, id);
3944
3945	r = kvm_arch_vcpu_create(vcpu);
3946	if (r)
3947		goto vcpu_free_run_page;
3948
3949	if (kvm->dirty_ring_size) {
3950		r = kvm_dirty_ring_alloc(&vcpu->dirty_ring,
3951					 id, kvm->dirty_ring_size);
3952		if (r)
3953			goto arch_vcpu_destroy;
3954	}
3955
3956	mutex_lock(&kvm->lock);
3957
3958#ifdef CONFIG_LOCKDEP
3959	/* Ensure that lockdep knows vcpu->mutex is taken *inside* kvm->lock */
3960	mutex_lock(&vcpu->mutex);
3961	mutex_unlock(&vcpu->mutex);
3962#endif
3963
3964	if (kvm_get_vcpu_by_id(kvm, id)) {
3965		r = -EEXIST;
3966		goto unlock_vcpu_destroy;
3967	}
3968
3969	vcpu->vcpu_idx = atomic_read(&kvm->online_vcpus);
3970	r = xa_insert(&kvm->vcpu_array, vcpu->vcpu_idx, vcpu, GFP_KERNEL_ACCOUNT);
3971	BUG_ON(r == -EBUSY);
3972	if (r)
3973		goto unlock_vcpu_destroy;
3974
3975	/* Now it's all set up, let userspace reach it */
3976	kvm_get_kvm(kvm);
3977	r = create_vcpu_fd(vcpu);
3978	if (r < 0) {
3979		xa_erase(&kvm->vcpu_array, vcpu->vcpu_idx);
3980		kvm_put_kvm_no_destroy(kvm);
3981		goto unlock_vcpu_destroy;
 
 
3982	}
3983
3984	/*
3985	 * Pairs with smp_rmb() in kvm_get_vcpu.  Store the vcpu
3986	 * pointer before kvm->online_vcpu's incremented value.
3987	 */
3988	smp_wmb();
3989	atomic_inc(&kvm->online_vcpus);
3990
3991	mutex_unlock(&kvm->lock);
3992	kvm_arch_vcpu_postcreate(vcpu);
3993	kvm_create_vcpu_debugfs(vcpu);
3994	return r;
3995
 
 
 
3996unlock_vcpu_destroy:
3997	mutex_unlock(&kvm->lock);
3998	kvm_dirty_ring_free(&vcpu->dirty_ring);
3999arch_vcpu_destroy:
4000	kvm_arch_vcpu_destroy(vcpu);
4001vcpu_free_run_page:
4002	free_page((unsigned long)vcpu->run);
4003vcpu_free:
4004	kmem_cache_free(kvm_vcpu_cache, vcpu);
4005vcpu_decrement:
4006	mutex_lock(&kvm->lock);
4007	kvm->created_vcpus--;
4008	mutex_unlock(&kvm->lock);
4009	return r;
4010}
4011
4012static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
4013{
4014	if (sigset) {
4015		sigdelsetmask(sigset, sigmask(SIGKILL)|sigmask(SIGSTOP));
4016		vcpu->sigset_active = 1;
4017		vcpu->sigset = *sigset;
4018	} else
4019		vcpu->sigset_active = 0;
4020	return 0;
4021}
4022
4023static ssize_t kvm_vcpu_stats_read(struct file *file, char __user *user_buffer,
4024			      size_t size, loff_t *offset)
4025{
4026	struct kvm_vcpu *vcpu = file->private_data;
4027
4028	return kvm_stats_read(vcpu->stats_id, &kvm_vcpu_stats_header,
4029			&kvm_vcpu_stats_desc[0], &vcpu->stat,
4030			sizeof(vcpu->stat), user_buffer, size, offset);
4031}
4032
 
 
 
 
 
 
 
 
4033static const struct file_operations kvm_vcpu_stats_fops = {
 
4034	.read = kvm_vcpu_stats_read,
 
4035	.llseek = noop_llseek,
4036};
4037
4038static int kvm_vcpu_ioctl_get_stats_fd(struct kvm_vcpu *vcpu)
4039{
4040	int fd;
4041	struct file *file;
4042	char name[15 + ITOA_MAX_LEN + 1];
4043
4044	snprintf(name, sizeof(name), "kvm-vcpu-stats:%d", vcpu->vcpu_id);
4045
4046	fd = get_unused_fd_flags(O_CLOEXEC);
4047	if (fd < 0)
4048		return fd;
4049
4050	file = anon_inode_getfile(name, &kvm_vcpu_stats_fops, vcpu, O_RDONLY);
4051	if (IS_ERR(file)) {
4052		put_unused_fd(fd);
4053		return PTR_ERR(file);
4054	}
 
 
 
4055	file->f_mode |= FMODE_PREAD;
4056	fd_install(fd, file);
4057
4058	return fd;
4059}
4060
4061static long kvm_vcpu_ioctl(struct file *filp,
4062			   unsigned int ioctl, unsigned long arg)
4063{
4064	struct kvm_vcpu *vcpu = filp->private_data;
4065	void __user *argp = (void __user *)arg;
4066	int r;
4067	struct kvm_fpu *fpu = NULL;
4068	struct kvm_sregs *kvm_sregs = NULL;
4069
4070	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4071		return -EIO;
4072
4073	if (unlikely(_IOC_TYPE(ioctl) != KVMIO))
4074		return -EINVAL;
4075
4076	/*
4077	 * Some architectures have vcpu ioctls that are asynchronous to vcpu
4078	 * execution; mutex_lock() would break them.
4079	 */
4080	r = kvm_arch_vcpu_async_ioctl(filp, ioctl, arg);
4081	if (r != -ENOIOCTLCMD)
4082		return r;
4083
4084	if (mutex_lock_killable(&vcpu->mutex))
4085		return -EINTR;
4086	switch (ioctl) {
4087	case KVM_RUN: {
4088		struct pid *oldpid;
4089		r = -EINVAL;
4090		if (arg)
4091			goto out;
4092		oldpid = rcu_access_pointer(vcpu->pid);
4093		if (unlikely(oldpid != task_pid(current))) {
4094			/* The thread running this VCPU changed. */
4095			struct pid *newpid;
4096
4097			r = kvm_arch_vcpu_run_pid_change(vcpu);
4098			if (r)
4099				break;
4100
4101			newpid = get_task_pid(current, PIDTYPE_PID);
4102			rcu_assign_pointer(vcpu->pid, newpid);
4103			if (oldpid)
4104				synchronize_rcu();
4105			put_pid(oldpid);
4106		}
4107		r = kvm_arch_vcpu_ioctl_run(vcpu);
4108		trace_kvm_userspace_exit(vcpu->run->exit_reason, r);
4109		break;
4110	}
4111	case KVM_GET_REGS: {
4112		struct kvm_regs *kvm_regs;
4113
4114		r = -ENOMEM;
4115		kvm_regs = kzalloc(sizeof(struct kvm_regs), GFP_KERNEL_ACCOUNT);
4116		if (!kvm_regs)
4117			goto out;
4118		r = kvm_arch_vcpu_ioctl_get_regs(vcpu, kvm_regs);
4119		if (r)
4120			goto out_free1;
4121		r = -EFAULT;
4122		if (copy_to_user(argp, kvm_regs, sizeof(struct kvm_regs)))
4123			goto out_free1;
4124		r = 0;
4125out_free1:
4126		kfree(kvm_regs);
4127		break;
4128	}
4129	case KVM_SET_REGS: {
4130		struct kvm_regs *kvm_regs;
4131
4132		kvm_regs = memdup_user(argp, sizeof(*kvm_regs));
4133		if (IS_ERR(kvm_regs)) {
4134			r = PTR_ERR(kvm_regs);
4135			goto out;
4136		}
4137		r = kvm_arch_vcpu_ioctl_set_regs(vcpu, kvm_regs);
4138		kfree(kvm_regs);
4139		break;
4140	}
4141	case KVM_GET_SREGS: {
4142		kvm_sregs = kzalloc(sizeof(struct kvm_sregs),
4143				    GFP_KERNEL_ACCOUNT);
4144		r = -ENOMEM;
4145		if (!kvm_sregs)
4146			goto out;
4147		r = kvm_arch_vcpu_ioctl_get_sregs(vcpu, kvm_sregs);
4148		if (r)
4149			goto out;
4150		r = -EFAULT;
4151		if (copy_to_user(argp, kvm_sregs, sizeof(struct kvm_sregs)))
4152			goto out;
4153		r = 0;
4154		break;
4155	}
4156	case KVM_SET_SREGS: {
4157		kvm_sregs = memdup_user(argp, sizeof(*kvm_sregs));
4158		if (IS_ERR(kvm_sregs)) {
4159			r = PTR_ERR(kvm_sregs);
4160			kvm_sregs = NULL;
4161			goto out;
4162		}
4163		r = kvm_arch_vcpu_ioctl_set_sregs(vcpu, kvm_sregs);
4164		break;
4165	}
4166	case KVM_GET_MP_STATE: {
4167		struct kvm_mp_state mp_state;
4168
4169		r = kvm_arch_vcpu_ioctl_get_mpstate(vcpu, &mp_state);
4170		if (r)
4171			goto out;
4172		r = -EFAULT;
4173		if (copy_to_user(argp, &mp_state, sizeof(mp_state)))
4174			goto out;
4175		r = 0;
4176		break;
4177	}
4178	case KVM_SET_MP_STATE: {
4179		struct kvm_mp_state mp_state;
4180
4181		r = -EFAULT;
4182		if (copy_from_user(&mp_state, argp, sizeof(mp_state)))
4183			goto out;
4184		r = kvm_arch_vcpu_ioctl_set_mpstate(vcpu, &mp_state);
4185		break;
4186	}
4187	case KVM_TRANSLATE: {
4188		struct kvm_translation tr;
4189
4190		r = -EFAULT;
4191		if (copy_from_user(&tr, argp, sizeof(tr)))
4192			goto out;
4193		r = kvm_arch_vcpu_ioctl_translate(vcpu, &tr);
4194		if (r)
4195			goto out;
4196		r = -EFAULT;
4197		if (copy_to_user(argp, &tr, sizeof(tr)))
4198			goto out;
4199		r = 0;
4200		break;
4201	}
4202	case KVM_SET_GUEST_DEBUG: {
4203		struct kvm_guest_debug dbg;
4204
4205		r = -EFAULT;
4206		if (copy_from_user(&dbg, argp, sizeof(dbg)))
4207			goto out;
4208		r = kvm_arch_vcpu_ioctl_set_guest_debug(vcpu, &dbg);
4209		break;
4210	}
4211	case KVM_SET_SIGNAL_MASK: {
4212		struct kvm_signal_mask __user *sigmask_arg = argp;
4213		struct kvm_signal_mask kvm_sigmask;
4214		sigset_t sigset, *p;
4215
4216		p = NULL;
4217		if (argp) {
4218			r = -EFAULT;
4219			if (copy_from_user(&kvm_sigmask, argp,
4220					   sizeof(kvm_sigmask)))
4221				goto out;
4222			r = -EINVAL;
4223			if (kvm_sigmask.len != sizeof(sigset))
4224				goto out;
4225			r = -EFAULT;
4226			if (copy_from_user(&sigset, sigmask_arg->sigset,
4227					   sizeof(sigset)))
4228				goto out;
4229			p = &sigset;
4230		}
4231		r = kvm_vcpu_ioctl_set_sigmask(vcpu, p);
4232		break;
4233	}
4234	case KVM_GET_FPU: {
4235		fpu = kzalloc(sizeof(struct kvm_fpu), GFP_KERNEL_ACCOUNT);
4236		r = -ENOMEM;
4237		if (!fpu)
4238			goto out;
4239		r = kvm_arch_vcpu_ioctl_get_fpu(vcpu, fpu);
4240		if (r)
4241			goto out;
4242		r = -EFAULT;
4243		if (copy_to_user(argp, fpu, sizeof(struct kvm_fpu)))
4244			goto out;
4245		r = 0;
4246		break;
4247	}
4248	case KVM_SET_FPU: {
4249		fpu = memdup_user(argp, sizeof(*fpu));
4250		if (IS_ERR(fpu)) {
4251			r = PTR_ERR(fpu);
4252			fpu = NULL;
4253			goto out;
4254		}
4255		r = kvm_arch_vcpu_ioctl_set_fpu(vcpu, fpu);
4256		break;
4257	}
4258	case KVM_GET_STATS_FD: {
4259		r = kvm_vcpu_ioctl_get_stats_fd(vcpu);
4260		break;
4261	}
4262	default:
4263		r = kvm_arch_vcpu_ioctl(filp, ioctl, arg);
4264	}
4265out:
4266	mutex_unlock(&vcpu->mutex);
4267	kfree(fpu);
4268	kfree(kvm_sregs);
4269	return r;
4270}
4271
4272#ifdef CONFIG_KVM_COMPAT
4273static long kvm_vcpu_compat_ioctl(struct file *filp,
4274				  unsigned int ioctl, unsigned long arg)
4275{
4276	struct kvm_vcpu *vcpu = filp->private_data;
4277	void __user *argp = compat_ptr(arg);
4278	int r;
4279
4280	if (vcpu->kvm->mm != current->mm || vcpu->kvm->vm_dead)
4281		return -EIO;
4282
4283	switch (ioctl) {
4284	case KVM_SET_SIGNAL_MASK: {
4285		struct kvm_signal_mask __user *sigmask_arg = argp;
4286		struct kvm_signal_mask kvm_sigmask;
4287		sigset_t sigset;
4288
4289		if (argp) {
4290			r = -EFAULT;
4291			if (copy_from_user(&kvm_sigmask, argp,
4292					   sizeof(kvm_sigmask)))
4293				goto out;
4294			r = -EINVAL;
4295			if (kvm_sigmask.len != sizeof(compat_sigset_t))
4296				goto out;
4297			r = -EFAULT;
4298			if (get_compat_sigset(&sigset,
4299					      (compat_sigset_t __user *)sigmask_arg->sigset))
4300				goto out;
4301			r = kvm_vcpu_ioctl_set_sigmask(vcpu, &sigset);
4302		} else
4303			r = kvm_vcpu_ioctl_set_sigmask(vcpu, NULL);
4304		break;
4305	}
4306	default:
4307		r = kvm_vcpu_ioctl(filp, ioctl, arg);
4308	}
4309
4310out:
4311	return r;
4312}
4313#endif
4314
4315static int kvm_device_mmap(struct file *filp, struct vm_area_struct *vma)
4316{
4317	struct kvm_device *dev = filp->private_data;
4318
4319	if (dev->ops->mmap)
4320		return dev->ops->mmap(dev, vma);
4321
4322	return -ENODEV;
4323}
4324
4325static int kvm_device_ioctl_attr(struct kvm_device *dev,
4326				 int (*accessor)(struct kvm_device *dev,
4327						 struct kvm_device_attr *attr),
4328				 unsigned long arg)
4329{
4330	struct kvm_device_attr attr;
4331
4332	if (!accessor)
4333		return -EPERM;
4334
4335	if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4336		return -EFAULT;
4337
4338	return accessor(dev, &attr);
4339}
4340
4341static long kvm_device_ioctl(struct file *filp, unsigned int ioctl,
4342			     unsigned long arg)
4343{
4344	struct kvm_device *dev = filp->private_data;
4345
4346	if (dev->kvm->mm != current->mm || dev->kvm->vm_dead)
4347		return -EIO;
4348
4349	switch (ioctl) {
4350	case KVM_SET_DEVICE_ATTR:
4351		return kvm_device_ioctl_attr(dev, dev->ops->set_attr, arg);
4352	case KVM_GET_DEVICE_ATTR:
4353		return kvm_device_ioctl_attr(dev, dev->ops->get_attr, arg);
4354	case KVM_HAS_DEVICE_ATTR:
4355		return kvm_device_ioctl_attr(dev, dev->ops->has_attr, arg);
4356	default:
4357		if (dev->ops->ioctl)
4358			return dev->ops->ioctl(dev, ioctl, arg);
4359
4360		return -ENOTTY;
4361	}
4362}
4363
4364static int kvm_device_release(struct inode *inode, struct file *filp)
4365{
4366	struct kvm_device *dev = filp->private_data;
4367	struct kvm *kvm = dev->kvm;
4368
4369	if (dev->ops->release) {
4370		mutex_lock(&kvm->lock);
4371		list_del(&dev->vm_node);
4372		dev->ops->release(dev);
4373		mutex_unlock(&kvm->lock);
4374	}
4375
4376	kvm_put_kvm(kvm);
4377	return 0;
4378}
4379
4380static const struct file_operations kvm_device_fops = {
4381	.unlocked_ioctl = kvm_device_ioctl,
4382	.release = kvm_device_release,
4383	KVM_COMPAT(kvm_device_ioctl),
4384	.mmap = kvm_device_mmap,
4385};
4386
4387struct kvm_device *kvm_device_from_filp(struct file *filp)
4388{
4389	if (filp->f_op != &kvm_device_fops)
4390		return NULL;
4391
4392	return filp->private_data;
4393}
4394
4395static const struct kvm_device_ops *kvm_device_ops_table[KVM_DEV_TYPE_MAX] = {
4396#ifdef CONFIG_KVM_MPIC
4397	[KVM_DEV_TYPE_FSL_MPIC_20]	= &kvm_mpic_ops,
4398	[KVM_DEV_TYPE_FSL_MPIC_42]	= &kvm_mpic_ops,
4399#endif
4400};
4401
4402int kvm_register_device_ops(const struct kvm_device_ops *ops, u32 type)
4403{
4404	if (type >= ARRAY_SIZE(kvm_device_ops_table))
4405		return -ENOSPC;
4406
4407	if (kvm_device_ops_table[type] != NULL)
4408		return -EEXIST;
4409
4410	kvm_device_ops_table[type] = ops;
4411	return 0;
4412}
4413
4414void kvm_unregister_device_ops(u32 type)
4415{
4416	if (kvm_device_ops_table[type] != NULL)
4417		kvm_device_ops_table[type] = NULL;
4418}
4419
4420static int kvm_ioctl_create_device(struct kvm *kvm,
4421				   struct kvm_create_device *cd)
4422{
4423	const struct kvm_device_ops *ops;
4424	struct kvm_device *dev;
4425	bool test = cd->flags & KVM_CREATE_DEVICE_TEST;
4426	int type;
4427	int ret;
4428
4429	if (cd->type >= ARRAY_SIZE(kvm_device_ops_table))
4430		return -ENODEV;
4431
4432	type = array_index_nospec(cd->type, ARRAY_SIZE(kvm_device_ops_table));
4433	ops = kvm_device_ops_table[type];
4434	if (ops == NULL)
4435		return -ENODEV;
4436
4437	if (test)
4438		return 0;
4439
4440	dev = kzalloc(sizeof(*dev), GFP_KERNEL_ACCOUNT);
4441	if (!dev)
4442		return -ENOMEM;
4443
4444	dev->ops = ops;
4445	dev->kvm = kvm;
4446
4447	mutex_lock(&kvm->lock);
4448	ret = ops->create(dev, type);
4449	if (ret < 0) {
4450		mutex_unlock(&kvm->lock);
4451		kfree(dev);
4452		return ret;
4453	}
4454	list_add(&dev->vm_node, &kvm->devices);
4455	mutex_unlock(&kvm->lock);
4456
4457	if (ops->init)
4458		ops->init(dev);
4459
4460	kvm_get_kvm(kvm);
4461	ret = anon_inode_getfd(ops->name, &kvm_device_fops, dev, O_RDWR | O_CLOEXEC);
4462	if (ret < 0) {
4463		kvm_put_kvm_no_destroy(kvm);
4464		mutex_lock(&kvm->lock);
4465		list_del(&dev->vm_node);
4466		if (ops->release)
4467			ops->release(dev);
4468		mutex_unlock(&kvm->lock);
4469		if (ops->destroy)
4470			ops->destroy(dev);
4471		return ret;
4472	}
4473
4474	cd->fd = ret;
4475	return 0;
4476}
4477
4478static long kvm_vm_ioctl_check_extension_generic(struct kvm *kvm, long arg)
4479{
4480	switch (arg) {
4481	case KVM_CAP_USER_MEMORY:
 
4482	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
4483	case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS:
4484	case KVM_CAP_INTERNAL_ERROR_DATA:
4485#ifdef CONFIG_HAVE_KVM_MSI
4486	case KVM_CAP_SIGNAL_MSI:
4487#endif
4488#ifdef CONFIG_HAVE_KVM_IRQFD
4489	case KVM_CAP_IRQFD:
4490	case KVM_CAP_IRQFD_RESAMPLE:
4491#endif
4492	case KVM_CAP_IOEVENTFD_ANY_LENGTH:
4493	case KVM_CAP_CHECK_EXTENSION_VM:
4494	case KVM_CAP_ENABLE_CAP_VM:
4495	case KVM_CAP_HALT_POLL:
4496		return 1;
4497#ifdef CONFIG_KVM_MMIO
4498	case KVM_CAP_COALESCED_MMIO:
4499		return KVM_COALESCED_MMIO_PAGE_OFFSET;
4500	case KVM_CAP_COALESCED_PIO:
4501		return 1;
4502#endif
4503#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4504	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2:
4505		return KVM_DIRTY_LOG_MANUAL_CAPS;
4506#endif
4507#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4508	case KVM_CAP_IRQ_ROUTING:
4509		return KVM_MAX_IRQ_ROUTES;
4510#endif
4511#if KVM_ADDRESS_SPACE_NUM > 1
4512	case KVM_CAP_MULTI_ADDRESS_SPACE:
4513		return KVM_ADDRESS_SPACE_NUM;
 
 
4514#endif
4515	case KVM_CAP_NR_MEMSLOTS:
4516		return KVM_USER_MEM_SLOTS;
4517	case KVM_CAP_DIRTY_LOG_RING:
4518#ifdef CONFIG_HAVE_KVM_DIRTY_RING_TSO
4519		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4520#else
4521		return 0;
4522#endif
4523	case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4524#ifdef CONFIG_HAVE_KVM_DIRTY_RING_ACQ_REL
4525		return KVM_DIRTY_RING_MAX_ENTRIES * sizeof(struct kvm_dirty_gfn);
4526#else
4527		return 0;
4528#endif
4529#ifdef CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP
4530	case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP:
4531#endif
4532	case KVM_CAP_BINARY_STATS_FD:
4533	case KVM_CAP_SYSTEM_EVENT_DATA:
 
4534		return 1;
 
 
 
 
 
 
 
 
4535	default:
4536		break;
4537	}
4538	return kvm_vm_ioctl_check_extension(kvm, arg);
4539}
4540
4541static int kvm_vm_ioctl_enable_dirty_log_ring(struct kvm *kvm, u32 size)
4542{
4543	int r;
4544
4545	if (!KVM_DIRTY_LOG_PAGE_OFFSET)
4546		return -EINVAL;
4547
4548	/* the size should be power of 2 */
4549	if (!size || (size & (size - 1)))
4550		return -EINVAL;
4551
4552	/* Should be bigger to keep the reserved entries, or a page */
4553	if (size < kvm_dirty_ring_get_rsvd_entries() *
4554	    sizeof(struct kvm_dirty_gfn) || size < PAGE_SIZE)
4555		return -EINVAL;
4556
4557	if (size > KVM_DIRTY_RING_MAX_ENTRIES *
4558	    sizeof(struct kvm_dirty_gfn))
4559		return -E2BIG;
4560
4561	/* We only allow it to set once */
4562	if (kvm->dirty_ring_size)
4563		return -EINVAL;
4564
4565	mutex_lock(&kvm->lock);
4566
4567	if (kvm->created_vcpus) {
4568		/* We don't allow to change this value after vcpu created */
4569		r = -EINVAL;
4570	} else {
4571		kvm->dirty_ring_size = size;
4572		r = 0;
4573	}
4574
4575	mutex_unlock(&kvm->lock);
4576	return r;
4577}
4578
4579static int kvm_vm_ioctl_reset_dirty_pages(struct kvm *kvm)
4580{
4581	unsigned long i;
4582	struct kvm_vcpu *vcpu;
4583	int cleared = 0;
4584
4585	if (!kvm->dirty_ring_size)
4586		return -EINVAL;
4587
4588	mutex_lock(&kvm->slots_lock);
4589
4590	kvm_for_each_vcpu(i, vcpu, kvm)
4591		cleared += kvm_dirty_ring_reset(vcpu->kvm, &vcpu->dirty_ring);
4592
4593	mutex_unlock(&kvm->slots_lock);
4594
4595	if (cleared)
4596		kvm_flush_remote_tlbs(kvm);
4597
4598	return cleared;
4599}
4600
4601int __attribute__((weak)) kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4602						  struct kvm_enable_cap *cap)
4603{
4604	return -EINVAL;
4605}
4606
4607static bool kvm_are_all_memslots_empty(struct kvm *kvm)
4608{
4609	int i;
4610
4611	lockdep_assert_held(&kvm->slots_lock);
4612
4613	for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
4614		if (!kvm_memslots_empty(__kvm_memslots(kvm, i)))
4615			return false;
4616	}
4617
4618	return true;
4619}
 
4620
4621static int kvm_vm_ioctl_enable_cap_generic(struct kvm *kvm,
4622					   struct kvm_enable_cap *cap)
4623{
4624	switch (cap->cap) {
4625#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4626	case KVM_CAP_MANUAL_DIRTY_LOG_PROTECT2: {
4627		u64 allowed_options = KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE;
4628
4629		if (cap->args[0] & KVM_DIRTY_LOG_MANUAL_PROTECT_ENABLE)
4630			allowed_options = KVM_DIRTY_LOG_MANUAL_CAPS;
4631
4632		if (cap->flags || (cap->args[0] & ~allowed_options))
4633			return -EINVAL;
4634		kvm->manual_dirty_log_protect = cap->args[0];
4635		return 0;
4636	}
4637#endif
4638	case KVM_CAP_HALT_POLL: {
4639		if (cap->flags || cap->args[0] != (unsigned int)cap->args[0])
4640			return -EINVAL;
4641
4642		kvm->max_halt_poll_ns = cap->args[0];
4643
4644		/*
4645		 * Ensure kvm->override_halt_poll_ns does not become visible
4646		 * before kvm->max_halt_poll_ns.
4647		 *
4648		 * Pairs with the smp_rmb() in kvm_vcpu_max_halt_poll_ns().
4649		 */
4650		smp_wmb();
4651		kvm->override_halt_poll_ns = true;
4652
4653		return 0;
4654	}
4655	case KVM_CAP_DIRTY_LOG_RING:
4656	case KVM_CAP_DIRTY_LOG_RING_ACQ_REL:
4657		if (!kvm_vm_ioctl_check_extension_generic(kvm, cap->cap))
4658			return -EINVAL;
4659
4660		return kvm_vm_ioctl_enable_dirty_log_ring(kvm, cap->args[0]);
4661	case KVM_CAP_DIRTY_LOG_RING_WITH_BITMAP: {
4662		int r = -EINVAL;
4663
4664		if (!IS_ENABLED(CONFIG_NEED_KVM_DIRTY_RING_WITH_BITMAP) ||
4665		    !kvm->dirty_ring_size || cap->flags)
4666			return r;
4667
4668		mutex_lock(&kvm->slots_lock);
4669
4670		/*
4671		 * For simplicity, allow enabling ring+bitmap if and only if
4672		 * there are no memslots, e.g. to ensure all memslots allocate
4673		 * a bitmap after the capability is enabled.
4674		 */
4675		if (kvm_are_all_memslots_empty(kvm)) {
4676			kvm->dirty_ring_with_bitmap = true;
4677			r = 0;
4678		}
4679
4680		mutex_unlock(&kvm->slots_lock);
4681
4682		return r;
4683	}
4684	default:
4685		return kvm_vm_ioctl_enable_cap(kvm, cap);
4686	}
4687}
4688
4689static ssize_t kvm_vm_stats_read(struct file *file, char __user *user_buffer,
4690			      size_t size, loff_t *offset)
4691{
4692	struct kvm *kvm = file->private_data;
4693
4694	return kvm_stats_read(kvm->stats_id, &kvm_vm_stats_header,
4695				&kvm_vm_stats_desc[0], &kvm->stat,
4696				sizeof(kvm->stat), user_buffer, size, offset);
4697}
4698
 
 
 
 
 
 
 
 
4699static const struct file_operations kvm_vm_stats_fops = {
 
4700	.read = kvm_vm_stats_read,
 
4701	.llseek = noop_llseek,
4702};
4703
4704static int kvm_vm_ioctl_get_stats_fd(struct kvm *kvm)
4705{
4706	int fd;
4707	struct file *file;
4708
4709	fd = get_unused_fd_flags(O_CLOEXEC);
4710	if (fd < 0)
4711		return fd;
4712
4713	file = anon_inode_getfile("kvm-vm-stats",
4714			&kvm_vm_stats_fops, kvm, O_RDONLY);
4715	if (IS_ERR(file)) {
4716		put_unused_fd(fd);
4717		return PTR_ERR(file);
4718	}
 
 
 
4719	file->f_mode |= FMODE_PREAD;
4720	fd_install(fd, file);
4721
4722	return fd;
4723}
4724
 
 
 
 
 
 
 
 
4725static long kvm_vm_ioctl(struct file *filp,
4726			   unsigned int ioctl, unsigned long arg)
4727{
4728	struct kvm *kvm = filp->private_data;
4729	void __user *argp = (void __user *)arg;
4730	int r;
4731
4732	if (kvm->mm != current->mm || kvm->vm_dead)
4733		return -EIO;
4734	switch (ioctl) {
4735	case KVM_CREATE_VCPU:
4736		r = kvm_vm_ioctl_create_vcpu(kvm, arg);
4737		break;
4738	case KVM_ENABLE_CAP: {
4739		struct kvm_enable_cap cap;
4740
4741		r = -EFAULT;
4742		if (copy_from_user(&cap, argp, sizeof(cap)))
4743			goto out;
4744		r = kvm_vm_ioctl_enable_cap_generic(kvm, &cap);
4745		break;
4746	}
 
4747	case KVM_SET_USER_MEMORY_REGION: {
4748		struct kvm_userspace_memory_region kvm_userspace_mem;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4749
4750		r = -EFAULT;
4751		if (copy_from_user(&kvm_userspace_mem, argp,
4752						sizeof(kvm_userspace_mem)))
 
 
 
 
4753			goto out;
4754
4755		r = kvm_vm_ioctl_set_memory_region(kvm, &kvm_userspace_mem);
4756		break;
4757	}
4758	case KVM_GET_DIRTY_LOG: {
4759		struct kvm_dirty_log log;
4760
4761		r = -EFAULT;
4762		if (copy_from_user(&log, argp, sizeof(log)))
4763			goto out;
4764		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4765		break;
4766	}
4767#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4768	case KVM_CLEAR_DIRTY_LOG: {
4769		struct kvm_clear_dirty_log log;
4770
4771		r = -EFAULT;
4772		if (copy_from_user(&log, argp, sizeof(log)))
4773			goto out;
4774		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4775		break;
4776	}
4777#endif
4778#ifdef CONFIG_KVM_MMIO
4779	case KVM_REGISTER_COALESCED_MMIO: {
4780		struct kvm_coalesced_mmio_zone zone;
4781
4782		r = -EFAULT;
4783		if (copy_from_user(&zone, argp, sizeof(zone)))
4784			goto out;
4785		r = kvm_vm_ioctl_register_coalesced_mmio(kvm, &zone);
4786		break;
4787	}
4788	case KVM_UNREGISTER_COALESCED_MMIO: {
4789		struct kvm_coalesced_mmio_zone zone;
4790
4791		r = -EFAULT;
4792		if (copy_from_user(&zone, argp, sizeof(zone)))
4793			goto out;
4794		r = kvm_vm_ioctl_unregister_coalesced_mmio(kvm, &zone);
4795		break;
4796	}
4797#endif
4798	case KVM_IRQFD: {
4799		struct kvm_irqfd data;
4800
4801		r = -EFAULT;
4802		if (copy_from_user(&data, argp, sizeof(data)))
4803			goto out;
4804		r = kvm_irqfd(kvm, &data);
4805		break;
4806	}
4807	case KVM_IOEVENTFD: {
4808		struct kvm_ioeventfd data;
4809
4810		r = -EFAULT;
4811		if (copy_from_user(&data, argp, sizeof(data)))
4812			goto out;
4813		r = kvm_ioeventfd(kvm, &data);
4814		break;
4815	}
4816#ifdef CONFIG_HAVE_KVM_MSI
4817	case KVM_SIGNAL_MSI: {
4818		struct kvm_msi msi;
4819
4820		r = -EFAULT;
4821		if (copy_from_user(&msi, argp, sizeof(msi)))
4822			goto out;
4823		r = kvm_send_userspace_msi(kvm, &msi);
4824		break;
4825	}
4826#endif
4827#ifdef __KVM_HAVE_IRQ_LINE
4828	case KVM_IRQ_LINE_STATUS:
4829	case KVM_IRQ_LINE: {
4830		struct kvm_irq_level irq_event;
4831
4832		r = -EFAULT;
4833		if (copy_from_user(&irq_event, argp, sizeof(irq_event)))
4834			goto out;
4835
4836		r = kvm_vm_ioctl_irq_line(kvm, &irq_event,
4837					ioctl == KVM_IRQ_LINE_STATUS);
4838		if (r)
4839			goto out;
4840
4841		r = -EFAULT;
4842		if (ioctl == KVM_IRQ_LINE_STATUS) {
4843			if (copy_to_user(argp, &irq_event, sizeof(irq_event)))
4844				goto out;
4845		}
4846
4847		r = 0;
4848		break;
4849	}
4850#endif
4851#ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
4852	case KVM_SET_GSI_ROUTING: {
4853		struct kvm_irq_routing routing;
4854		struct kvm_irq_routing __user *urouting;
4855		struct kvm_irq_routing_entry *entries = NULL;
4856
4857		r = -EFAULT;
4858		if (copy_from_user(&routing, argp, sizeof(routing)))
4859			goto out;
4860		r = -EINVAL;
4861		if (!kvm_arch_can_set_irq_routing(kvm))
4862			goto out;
4863		if (routing.nr > KVM_MAX_IRQ_ROUTES)
4864			goto out;
4865		if (routing.flags)
4866			goto out;
4867		if (routing.nr) {
4868			urouting = argp;
4869			entries = vmemdup_user(urouting->entries,
4870					       array_size(sizeof(*entries),
4871							  routing.nr));
4872			if (IS_ERR(entries)) {
4873				r = PTR_ERR(entries);
4874				goto out;
4875			}
4876		}
4877		r = kvm_set_irq_routing(kvm, entries, routing.nr,
4878					routing.flags);
4879		kvfree(entries);
4880		break;
4881	}
4882#endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
 
 
 
 
 
 
 
 
 
 
 
 
4883	case KVM_CREATE_DEVICE: {
4884		struct kvm_create_device cd;
4885
4886		r = -EFAULT;
4887		if (copy_from_user(&cd, argp, sizeof(cd)))
4888			goto out;
4889
4890		r = kvm_ioctl_create_device(kvm, &cd);
4891		if (r)
4892			goto out;
4893
4894		r = -EFAULT;
4895		if (copy_to_user(argp, &cd, sizeof(cd)))
4896			goto out;
4897
4898		r = 0;
4899		break;
4900	}
4901	case KVM_CHECK_EXTENSION:
4902		r = kvm_vm_ioctl_check_extension_generic(kvm, arg);
4903		break;
4904	case KVM_RESET_DIRTY_RINGS:
4905		r = kvm_vm_ioctl_reset_dirty_pages(kvm);
4906		break;
4907	case KVM_GET_STATS_FD:
4908		r = kvm_vm_ioctl_get_stats_fd(kvm);
4909		break;
 
 
 
 
 
 
 
 
 
 
 
 
4910	default:
4911		r = kvm_arch_vm_ioctl(filp, ioctl, arg);
4912	}
4913out:
4914	return r;
4915}
4916
4917#ifdef CONFIG_KVM_COMPAT
4918struct compat_kvm_dirty_log {
4919	__u32 slot;
4920	__u32 padding1;
4921	union {
4922		compat_uptr_t dirty_bitmap; /* one bit per page */
4923		__u64 padding2;
4924	};
4925};
4926
4927struct compat_kvm_clear_dirty_log {
4928	__u32 slot;
4929	__u32 num_pages;
4930	__u64 first_page;
4931	union {
4932		compat_uptr_t dirty_bitmap; /* one bit per page */
4933		__u64 padding2;
4934	};
4935};
4936
4937long __weak kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
4938				     unsigned long arg)
4939{
4940	return -ENOTTY;
4941}
4942
4943static long kvm_vm_compat_ioctl(struct file *filp,
4944			   unsigned int ioctl, unsigned long arg)
4945{
4946	struct kvm *kvm = filp->private_data;
4947	int r;
4948
4949	if (kvm->mm != current->mm || kvm->vm_dead)
4950		return -EIO;
4951
4952	r = kvm_arch_vm_compat_ioctl(filp, ioctl, arg);
4953	if (r != -ENOTTY)
4954		return r;
4955
4956	switch (ioctl) {
4957#ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
4958	case KVM_CLEAR_DIRTY_LOG: {
4959		struct compat_kvm_clear_dirty_log compat_log;
4960		struct kvm_clear_dirty_log log;
4961
4962		if (copy_from_user(&compat_log, (void __user *)arg,
4963				   sizeof(compat_log)))
4964			return -EFAULT;
4965		log.slot	 = compat_log.slot;
4966		log.num_pages	 = compat_log.num_pages;
4967		log.first_page	 = compat_log.first_page;
4968		log.padding2	 = compat_log.padding2;
4969		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4970
4971		r = kvm_vm_ioctl_clear_dirty_log(kvm, &log);
4972		break;
4973	}
4974#endif
4975	case KVM_GET_DIRTY_LOG: {
4976		struct compat_kvm_dirty_log compat_log;
4977		struct kvm_dirty_log log;
4978
4979		if (copy_from_user(&compat_log, (void __user *)arg,
4980				   sizeof(compat_log)))
4981			return -EFAULT;
4982		log.slot	 = compat_log.slot;
4983		log.padding1	 = compat_log.padding1;
4984		log.padding2	 = compat_log.padding2;
4985		log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
4986
4987		r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
4988		break;
4989	}
4990	default:
4991		r = kvm_vm_ioctl(filp, ioctl, arg);
4992	}
4993	return r;
4994}
4995#endif
4996
4997static const struct file_operations kvm_vm_fops = {
4998	.release        = kvm_vm_release,
4999	.unlocked_ioctl = kvm_vm_ioctl,
5000	.llseek		= noop_llseek,
5001	KVM_COMPAT(kvm_vm_compat_ioctl),
5002};
5003
5004bool file_is_kvm(struct file *file)
5005{
5006	return file && file->f_op == &kvm_vm_fops;
5007}
5008EXPORT_SYMBOL_GPL(file_is_kvm);
5009
5010static int kvm_dev_ioctl_create_vm(unsigned long type)
5011{
5012	char fdname[ITOA_MAX_LEN + 1];
5013	int r, fd;
5014	struct kvm *kvm;
5015	struct file *file;
5016
5017	fd = get_unused_fd_flags(O_CLOEXEC);
5018	if (fd < 0)
5019		return fd;
5020
5021	snprintf(fdname, sizeof(fdname), "%d", fd);
5022
5023	kvm = kvm_create_vm(type, fdname);
5024	if (IS_ERR(kvm)) {
5025		r = PTR_ERR(kvm);
5026		goto put_fd;
5027	}
5028
5029	file = anon_inode_getfile("kvm-vm", &kvm_vm_fops, kvm, O_RDWR);
5030	if (IS_ERR(file)) {
5031		r = PTR_ERR(file);
5032		goto put_kvm;
5033	}
5034
5035	/*
5036	 * Don't call kvm_put_kvm anymore at this point; file->f_op is
5037	 * already set, with ->release() being kvm_vm_release().  In error
5038	 * cases it will be called by the final fput(file) and will take
5039	 * care of doing kvm_put_kvm(kvm).
5040	 */
5041	kvm_uevent_notify_change(KVM_EVENT_CREATE_VM, kvm);
5042
5043	fd_install(fd, file);
5044	return fd;
5045
5046put_kvm:
5047	kvm_put_kvm(kvm);
5048put_fd:
5049	put_unused_fd(fd);
5050	return r;
5051}
5052
5053static long kvm_dev_ioctl(struct file *filp,
5054			  unsigned int ioctl, unsigned long arg)
5055{
5056	long r = -EINVAL;
5057
5058	switch (ioctl) {
5059	case KVM_GET_API_VERSION:
5060		if (arg)
5061			goto out;
5062		r = KVM_API_VERSION;
5063		break;
5064	case KVM_CREATE_VM:
5065		r = kvm_dev_ioctl_create_vm(arg);
5066		break;
5067	case KVM_CHECK_EXTENSION:
5068		r = kvm_vm_ioctl_check_extension_generic(NULL, arg);
5069		break;
5070	case KVM_GET_VCPU_MMAP_SIZE:
5071		if (arg)
5072			goto out;
5073		r = PAGE_SIZE;     /* struct kvm_run */
5074#ifdef CONFIG_X86
5075		r += PAGE_SIZE;    /* pio data page */
5076#endif
5077#ifdef CONFIG_KVM_MMIO
5078		r += PAGE_SIZE;    /* coalesced mmio ring page */
5079#endif
5080		break;
5081	case KVM_TRACE_ENABLE:
5082	case KVM_TRACE_PAUSE:
5083	case KVM_TRACE_DISABLE:
5084		r = -EOPNOTSUPP;
5085		break;
5086	default:
5087		return kvm_arch_dev_ioctl(filp, ioctl, arg);
5088	}
5089out:
5090	return r;
5091}
5092
5093static struct file_operations kvm_chardev_ops = {
5094	.unlocked_ioctl = kvm_dev_ioctl,
5095	.llseek		= noop_llseek,
5096	KVM_COMPAT(kvm_dev_ioctl),
5097};
5098
5099static struct miscdevice kvm_dev = {
5100	KVM_MINOR,
5101	"kvm",
5102	&kvm_chardev_ops,
5103};
5104
5105static void hardware_enable_nolock(void *junk)
5106{
5107	int cpu = raw_smp_processor_id();
5108	int r;
5109
5110	if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
5111		return;
5112
5113	cpumask_set_cpu(cpu, cpus_hardware_enabled);
 
5114
5115	r = kvm_arch_hardware_enable();
 
 
 
5116
5117	if (r) {
5118		cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5119		atomic_inc(&hardware_enable_failed);
5120		pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu);
5121	}
 
 
 
5122}
5123
5124static int kvm_starting_cpu(unsigned int cpu)
5125{
5126	raw_spin_lock(&kvm_count_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
5127	if (kvm_usage_count)
5128		hardware_enable_nolock(NULL);
5129	raw_spin_unlock(&kvm_count_lock);
5130	return 0;
5131}
5132
5133static void hardware_disable_nolock(void *junk)
5134{
5135	int cpu = raw_smp_processor_id();
5136
5137	if (!cpumask_test_cpu(cpu, cpus_hardware_enabled))
 
 
5138		return;
5139	cpumask_clear_cpu(cpu, cpus_hardware_enabled);
5140	kvm_arch_hardware_disable();
 
 
5141}
5142
5143static int kvm_dying_cpu(unsigned int cpu)
5144{
5145	raw_spin_lock(&kvm_count_lock);
5146	if (kvm_usage_count)
5147		hardware_disable_nolock(NULL);
5148	raw_spin_unlock(&kvm_count_lock);
5149	return 0;
5150}
5151
5152static void hardware_disable_all_nolock(void)
5153{
5154	BUG_ON(!kvm_usage_count);
5155
5156	kvm_usage_count--;
5157	if (!kvm_usage_count)
5158		on_each_cpu(hardware_disable_nolock, NULL, 1);
5159}
5160
5161static void hardware_disable_all(void)
5162{
5163	raw_spin_lock(&kvm_count_lock);
 
5164	hardware_disable_all_nolock();
5165	raw_spin_unlock(&kvm_count_lock);
 
5166}
5167
5168static int hardware_enable_all(void)
5169{
5170	int r = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5171
5172	raw_spin_lock(&kvm_count_lock);
5173
5174	kvm_usage_count++;
5175	if (kvm_usage_count == 1) {
5176		atomic_set(&hardware_enable_failed, 0);
5177		on_each_cpu(hardware_enable_nolock, NULL, 1);
5178
5179		if (atomic_read(&hardware_enable_failed)) {
5180			hardware_disable_all_nolock();
5181			r = -EBUSY;
5182		}
5183	}
5184
5185	raw_spin_unlock(&kvm_count_lock);
 
5186
5187	return r;
5188}
5189
5190static int kvm_reboot(struct notifier_block *notifier, unsigned long val,
5191		      void *v)
5192{
5193	/*
5194	 * Some (well, at least mine) BIOSes hang on reboot if
5195	 * in vmx root mode.
5196	 *
5197	 * And Intel TXT required VMX off for all cpu when system shutdown.
 
 
 
 
 
5198	 */
5199	pr_info("kvm: exiting hardware virtualization\n");
5200	kvm_rebooting = true;
5201	on_each_cpu(hardware_disable_nolock, NULL, 1);
5202	return NOTIFY_OK;
5203}
5204
5205static struct notifier_block kvm_reboot_notifier = {
5206	.notifier_call = kvm_reboot,
5207	.priority = 0,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5208};
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5209
5210static void kvm_io_bus_destroy(struct kvm_io_bus *bus)
5211{
5212	int i;
5213
5214	for (i = 0; i < bus->dev_count; i++) {
5215		struct kvm_io_device *pos = bus->range[i].dev;
5216
5217		kvm_iodevice_destructor(pos);
5218	}
5219	kfree(bus);
5220}
5221
5222static inline int kvm_io_bus_cmp(const struct kvm_io_range *r1,
5223				 const struct kvm_io_range *r2)
5224{
5225	gpa_t addr1 = r1->addr;
5226	gpa_t addr2 = r2->addr;
5227
5228	if (addr1 < addr2)
5229		return -1;
5230
5231	/* If r2->len == 0, match the exact address.  If r2->len != 0,
5232	 * accept any overlapping write.  Any order is acceptable for
5233	 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
5234	 * we process all of them.
5235	 */
5236	if (r2->len) {
5237		addr1 += r1->len;
5238		addr2 += r2->len;
5239	}
5240
5241	if (addr1 > addr2)
5242		return 1;
5243
5244	return 0;
5245}
5246
5247static int kvm_io_bus_sort_cmp(const void *p1, const void *p2)
5248{
5249	return kvm_io_bus_cmp(p1, p2);
5250}
5251
5252static int kvm_io_bus_get_first_dev(struct kvm_io_bus *bus,
5253			     gpa_t addr, int len)
5254{
5255	struct kvm_io_range *range, key;
5256	int off;
5257
5258	key = (struct kvm_io_range) {
5259		.addr = addr,
5260		.len = len,
5261	};
5262
5263	range = bsearch(&key, bus->range, bus->dev_count,
5264			sizeof(struct kvm_io_range), kvm_io_bus_sort_cmp);
5265	if (range == NULL)
5266		return -ENOENT;
5267
5268	off = range - bus->range;
5269
5270	while (off > 0 && kvm_io_bus_cmp(&key, &bus->range[off-1]) == 0)
5271		off--;
5272
5273	return off;
5274}
5275
5276static int __kvm_io_bus_write(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5277			      struct kvm_io_range *range, const void *val)
5278{
5279	int idx;
5280
5281	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5282	if (idx < 0)
5283		return -EOPNOTSUPP;
5284
5285	while (idx < bus->dev_count &&
5286		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5287		if (!kvm_iodevice_write(vcpu, bus->range[idx].dev, range->addr,
5288					range->len, val))
5289			return idx;
5290		idx++;
5291	}
5292
5293	return -EOPNOTSUPP;
5294}
5295
5296/* kvm_io_bus_write - called under kvm->slots_lock */
5297int kvm_io_bus_write(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5298		     int len, const void *val)
5299{
5300	struct kvm_io_bus *bus;
5301	struct kvm_io_range range;
5302	int r;
5303
5304	range = (struct kvm_io_range) {
5305		.addr = addr,
5306		.len = len,
5307	};
5308
5309	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5310	if (!bus)
5311		return -ENOMEM;
5312	r = __kvm_io_bus_write(vcpu, bus, &range, val);
5313	return r < 0 ? r : 0;
5314}
5315EXPORT_SYMBOL_GPL(kvm_io_bus_write);
5316
5317/* kvm_io_bus_write_cookie - called under kvm->slots_lock */
5318int kvm_io_bus_write_cookie(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx,
5319			    gpa_t addr, int len, const void *val, long cookie)
5320{
5321	struct kvm_io_bus *bus;
5322	struct kvm_io_range range;
5323
5324	range = (struct kvm_io_range) {
5325		.addr = addr,
5326		.len = len,
5327	};
5328
5329	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5330	if (!bus)
5331		return -ENOMEM;
5332
5333	/* First try the device referenced by cookie. */
5334	if ((cookie >= 0) && (cookie < bus->dev_count) &&
5335	    (kvm_io_bus_cmp(&range, &bus->range[cookie]) == 0))
5336		if (!kvm_iodevice_write(vcpu, bus->range[cookie].dev, addr, len,
5337					val))
5338			return cookie;
5339
5340	/*
5341	 * cookie contained garbage; fall back to search and return the
5342	 * correct cookie value.
5343	 */
5344	return __kvm_io_bus_write(vcpu, bus, &range, val);
5345}
5346
5347static int __kvm_io_bus_read(struct kvm_vcpu *vcpu, struct kvm_io_bus *bus,
5348			     struct kvm_io_range *range, void *val)
5349{
5350	int idx;
5351
5352	idx = kvm_io_bus_get_first_dev(bus, range->addr, range->len);
5353	if (idx < 0)
5354		return -EOPNOTSUPP;
5355
5356	while (idx < bus->dev_count &&
5357		kvm_io_bus_cmp(range, &bus->range[idx]) == 0) {
5358		if (!kvm_iodevice_read(vcpu, bus->range[idx].dev, range->addr,
5359				       range->len, val))
5360			return idx;
5361		idx++;
5362	}
5363
5364	return -EOPNOTSUPP;
5365}
5366
5367/* kvm_io_bus_read - called under kvm->slots_lock */
5368int kvm_io_bus_read(struct kvm_vcpu *vcpu, enum kvm_bus bus_idx, gpa_t addr,
5369		    int len, void *val)
5370{
5371	struct kvm_io_bus *bus;
5372	struct kvm_io_range range;
5373	int r;
5374
5375	range = (struct kvm_io_range) {
5376		.addr = addr,
5377		.len = len,
5378	};
5379
5380	bus = srcu_dereference(vcpu->kvm->buses[bus_idx], &vcpu->kvm->srcu);
5381	if (!bus)
5382		return -ENOMEM;
5383	r = __kvm_io_bus_read(vcpu, bus, &range, val);
5384	return r < 0 ? r : 0;
5385}
5386
5387/* Caller must hold slots_lock. */
5388int kvm_io_bus_register_dev(struct kvm *kvm, enum kvm_bus bus_idx, gpa_t addr,
5389			    int len, struct kvm_io_device *dev)
5390{
5391	int i;
5392	struct kvm_io_bus *new_bus, *bus;
5393	struct kvm_io_range range;
5394
 
 
5395	bus = kvm_get_bus(kvm, bus_idx);
5396	if (!bus)
5397		return -ENOMEM;
5398
5399	/* exclude ioeventfd which is limited by maximum fd */
5400	if (bus->dev_count - bus->ioeventfd_count > NR_IOBUS_DEVS - 1)
5401		return -ENOSPC;
5402
5403	new_bus = kmalloc(struct_size(bus, range, bus->dev_count + 1),
5404			  GFP_KERNEL_ACCOUNT);
5405	if (!new_bus)
5406		return -ENOMEM;
5407
5408	range = (struct kvm_io_range) {
5409		.addr = addr,
5410		.len = len,
5411		.dev = dev,
5412	};
5413
5414	for (i = 0; i < bus->dev_count; i++)
5415		if (kvm_io_bus_cmp(&bus->range[i], &range) > 0)
5416			break;
5417
5418	memcpy(new_bus, bus, sizeof(*bus) + i * sizeof(struct kvm_io_range));
5419	new_bus->dev_count++;
5420	new_bus->range[i] = range;
5421	memcpy(new_bus->range + i + 1, bus->range + i,
5422		(bus->dev_count - i) * sizeof(struct kvm_io_range));
5423	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5424	synchronize_srcu_expedited(&kvm->srcu);
5425	kfree(bus);
5426
5427	return 0;
5428}
5429
5430int kvm_io_bus_unregister_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5431			      struct kvm_io_device *dev)
5432{
5433	int i, j;
5434	struct kvm_io_bus *new_bus, *bus;
5435
5436	lockdep_assert_held(&kvm->slots_lock);
5437
5438	bus = kvm_get_bus(kvm, bus_idx);
5439	if (!bus)
5440		return 0;
5441
5442	for (i = 0; i < bus->dev_count; i++) {
5443		if (bus->range[i].dev == dev) {
5444			break;
5445		}
5446	}
5447
5448	if (i == bus->dev_count)
5449		return 0;
5450
5451	new_bus = kmalloc(struct_size(bus, range, bus->dev_count - 1),
5452			  GFP_KERNEL_ACCOUNT);
5453	if (new_bus) {
5454		memcpy(new_bus, bus, struct_size(bus, range, i));
5455		new_bus->dev_count--;
5456		memcpy(new_bus->range + i, bus->range + i + 1,
5457				flex_array_size(new_bus, range, new_bus->dev_count - i));
5458	}
5459
5460	rcu_assign_pointer(kvm->buses[bus_idx], new_bus);
5461	synchronize_srcu_expedited(&kvm->srcu);
5462
5463	/* Destroy the old bus _after_ installing the (null) bus. */
 
 
 
5464	if (!new_bus) {
5465		pr_err("kvm: failed to shrink bus, removing it completely\n");
5466		for (j = 0; j < bus->dev_count; j++) {
5467			if (j == i)
5468				continue;
5469			kvm_iodevice_destructor(bus->range[j].dev);
5470		}
5471	}
5472
 
5473	kfree(bus);
5474	return new_bus ? 0 : -ENOMEM;
5475}
5476
5477struct kvm_io_device *kvm_io_bus_get_dev(struct kvm *kvm, enum kvm_bus bus_idx,
5478					 gpa_t addr)
5479{
5480	struct kvm_io_bus *bus;
5481	int dev_idx, srcu_idx;
5482	struct kvm_io_device *iodev = NULL;
5483
5484	srcu_idx = srcu_read_lock(&kvm->srcu);
5485
5486	bus = srcu_dereference(kvm->buses[bus_idx], &kvm->srcu);
5487	if (!bus)
5488		goto out_unlock;
5489
5490	dev_idx = kvm_io_bus_get_first_dev(bus, addr, 1);
5491	if (dev_idx < 0)
5492		goto out_unlock;
5493
5494	iodev = bus->range[dev_idx].dev;
5495
5496out_unlock:
5497	srcu_read_unlock(&kvm->srcu, srcu_idx);
5498
5499	return iodev;
5500}
5501EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev);
5502
5503static int kvm_debugfs_open(struct inode *inode, struct file *file,
5504			   int (*get)(void *, u64 *), int (*set)(void *, u64),
5505			   const char *fmt)
5506{
5507	int ret;
5508	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5509					  inode->i_private;
5510
5511	/*
5512	 * The debugfs files are a reference to the kvm struct which
5513        * is still valid when kvm_destroy_vm is called.  kvm_get_kvm_safe
5514        * avoids the race between open and the removal of the debugfs directory.
5515	 */
5516	if (!kvm_get_kvm_safe(stat_data->kvm))
5517		return -ENOENT;
5518
5519	ret = simple_attr_open(inode, file, get,
5520			       kvm_stats_debugfs_mode(stat_data->desc) & 0222
5521			       ? set : NULL, fmt);
5522	if (ret)
5523		kvm_put_kvm(stat_data->kvm);
5524
5525	return ret;
5526}
5527
5528static int kvm_debugfs_release(struct inode *inode, struct file *file)
5529{
5530	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)
5531					  inode->i_private;
5532
5533	simple_attr_release(inode, file);
5534	kvm_put_kvm(stat_data->kvm);
5535
5536	return 0;
5537}
5538
5539static int kvm_get_stat_per_vm(struct kvm *kvm, size_t offset, u64 *val)
5540{
5541	*val = *(u64 *)((void *)(&kvm->stat) + offset);
5542
5543	return 0;
5544}
5545
5546static int kvm_clear_stat_per_vm(struct kvm *kvm, size_t offset)
5547{
5548	*(u64 *)((void *)(&kvm->stat) + offset) = 0;
5549
5550	return 0;
5551}
5552
5553static int kvm_get_stat_per_vcpu(struct kvm *kvm, size_t offset, u64 *val)
5554{
5555	unsigned long i;
5556	struct kvm_vcpu *vcpu;
5557
5558	*val = 0;
5559
5560	kvm_for_each_vcpu(i, vcpu, kvm)
5561		*val += *(u64 *)((void *)(&vcpu->stat) + offset);
5562
5563	return 0;
5564}
5565
5566static int kvm_clear_stat_per_vcpu(struct kvm *kvm, size_t offset)
5567{
5568	unsigned long i;
5569	struct kvm_vcpu *vcpu;
5570
5571	kvm_for_each_vcpu(i, vcpu, kvm)
5572		*(u64 *)((void *)(&vcpu->stat) + offset) = 0;
5573
5574	return 0;
5575}
5576
5577static int kvm_stat_data_get(void *data, u64 *val)
5578{
5579	int r = -EFAULT;
5580	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5581
5582	switch (stat_data->kind) {
5583	case KVM_STAT_VM:
5584		r = kvm_get_stat_per_vm(stat_data->kvm,
5585					stat_data->desc->desc.offset, val);
5586		break;
5587	case KVM_STAT_VCPU:
5588		r = kvm_get_stat_per_vcpu(stat_data->kvm,
5589					  stat_data->desc->desc.offset, val);
5590		break;
5591	}
5592
5593	return r;
5594}
5595
5596static int kvm_stat_data_clear(void *data, u64 val)
5597{
5598	int r = -EFAULT;
5599	struct kvm_stat_data *stat_data = (struct kvm_stat_data *)data;
5600
5601	if (val)
5602		return -EINVAL;
5603
5604	switch (stat_data->kind) {
5605	case KVM_STAT_VM:
5606		r = kvm_clear_stat_per_vm(stat_data->kvm,
5607					  stat_data->desc->desc.offset);
5608		break;
5609	case KVM_STAT_VCPU:
5610		r = kvm_clear_stat_per_vcpu(stat_data->kvm,
5611					    stat_data->desc->desc.offset);
5612		break;
5613	}
5614
5615	return r;
5616}
5617
5618static int kvm_stat_data_open(struct inode *inode, struct file *file)
5619{
5620	__simple_attr_check_format("%llu\n", 0ull);
5621	return kvm_debugfs_open(inode, file, kvm_stat_data_get,
5622				kvm_stat_data_clear, "%llu\n");
5623}
5624
5625static const struct file_operations stat_fops_per_vm = {
5626	.owner = THIS_MODULE,
5627	.open = kvm_stat_data_open,
5628	.release = kvm_debugfs_release,
5629	.read = simple_attr_read,
5630	.write = simple_attr_write,
5631	.llseek = no_llseek,
5632};
5633
5634static int vm_stat_get(void *_offset, u64 *val)
5635{
5636	unsigned offset = (long)_offset;
5637	struct kvm *kvm;
5638	u64 tmp_val;
5639
5640	*val = 0;
5641	mutex_lock(&kvm_lock);
5642	list_for_each_entry(kvm, &vm_list, vm_list) {
5643		kvm_get_stat_per_vm(kvm, offset, &tmp_val);
5644		*val += tmp_val;
5645	}
5646	mutex_unlock(&kvm_lock);
5647	return 0;
5648}
5649
5650static int vm_stat_clear(void *_offset, u64 val)
5651{
5652	unsigned offset = (long)_offset;
5653	struct kvm *kvm;
5654
5655	if (val)
5656		return -EINVAL;
5657
5658	mutex_lock(&kvm_lock);
5659	list_for_each_entry(kvm, &vm_list, vm_list) {
5660		kvm_clear_stat_per_vm(kvm, offset);
5661	}
5662	mutex_unlock(&kvm_lock);
5663
5664	return 0;
5665}
5666
5667DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops, vm_stat_get, vm_stat_clear, "%llu\n");
5668DEFINE_SIMPLE_ATTRIBUTE(vm_stat_readonly_fops, vm_stat_get, NULL, "%llu\n");
5669
5670static int vcpu_stat_get(void *_offset, u64 *val)
5671{
5672	unsigned offset = (long)_offset;
5673	struct kvm *kvm;
5674	u64 tmp_val;
5675
5676	*val = 0;
5677	mutex_lock(&kvm_lock);
5678	list_for_each_entry(kvm, &vm_list, vm_list) {
5679		kvm_get_stat_per_vcpu(kvm, offset, &tmp_val);
5680		*val += tmp_val;
5681	}
5682	mutex_unlock(&kvm_lock);
5683	return 0;
5684}
5685
5686static int vcpu_stat_clear(void *_offset, u64 val)
5687{
5688	unsigned offset = (long)_offset;
5689	struct kvm *kvm;
5690
5691	if (val)
5692		return -EINVAL;
5693
5694	mutex_lock(&kvm_lock);
5695	list_for_each_entry(kvm, &vm_list, vm_list) {
5696		kvm_clear_stat_per_vcpu(kvm, offset);
5697	}
5698	mutex_unlock(&kvm_lock);
5699
5700	return 0;
5701}
5702
5703DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops, vcpu_stat_get, vcpu_stat_clear,
5704			"%llu\n");
5705DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_readonly_fops, vcpu_stat_get, NULL, "%llu\n");
5706
5707static void kvm_uevent_notify_change(unsigned int type, struct kvm *kvm)
5708{
5709	struct kobj_uevent_env *env;
5710	unsigned long long created, active;
5711
5712	if (!kvm_dev.this_device || !kvm)
5713		return;
5714
5715	mutex_lock(&kvm_lock);
5716	if (type == KVM_EVENT_CREATE_VM) {
5717		kvm_createvm_count++;
5718		kvm_active_vms++;
5719	} else if (type == KVM_EVENT_DESTROY_VM) {
5720		kvm_active_vms--;
5721	}
5722	created = kvm_createvm_count;
5723	active = kvm_active_vms;
5724	mutex_unlock(&kvm_lock);
5725
5726	env = kzalloc(sizeof(*env), GFP_KERNEL_ACCOUNT);
5727	if (!env)
5728		return;
5729
5730	add_uevent_var(env, "CREATED=%llu", created);
5731	add_uevent_var(env, "COUNT=%llu", active);
5732
5733	if (type == KVM_EVENT_CREATE_VM) {
5734		add_uevent_var(env, "EVENT=create");
5735		kvm->userspace_pid = task_pid_nr(current);
5736	} else if (type == KVM_EVENT_DESTROY_VM) {
5737		add_uevent_var(env, "EVENT=destroy");
5738	}
5739	add_uevent_var(env, "PID=%d", kvm->userspace_pid);
5740
5741	if (!IS_ERR(kvm->debugfs_dentry)) {
5742		char *tmp, *p = kmalloc(PATH_MAX, GFP_KERNEL_ACCOUNT);
5743
5744		if (p) {
5745			tmp = dentry_path_raw(kvm->debugfs_dentry, p, PATH_MAX);
5746			if (!IS_ERR(tmp))
5747				add_uevent_var(env, "STATS_PATH=%s", tmp);
5748			kfree(p);
5749		}
5750	}
5751	/* no need for checks, since we are adding at most only 5 keys */
5752	env->envp[env->envp_idx++] = NULL;
5753	kobject_uevent_env(&kvm_dev.this_device->kobj, KOBJ_CHANGE, env->envp);
5754	kfree(env);
5755}
5756
5757static void kvm_init_debug(void)
5758{
5759	const struct file_operations *fops;
5760	const struct _kvm_stats_desc *pdesc;
5761	int i;
5762
5763	kvm_debugfs_dir = debugfs_create_dir("kvm", NULL);
5764
5765	for (i = 0; i < kvm_vm_stats_header.num_desc; ++i) {
5766		pdesc = &kvm_vm_stats_desc[i];
5767		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5768			fops = &vm_stat_fops;
5769		else
5770			fops = &vm_stat_readonly_fops;
5771		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5772				kvm_debugfs_dir,
5773				(void *)(long)pdesc->desc.offset, fops);
5774	}
5775
5776	for (i = 0; i < kvm_vcpu_stats_header.num_desc; ++i) {
5777		pdesc = &kvm_vcpu_stats_desc[i];
5778		if (kvm_stats_debugfs_mode(pdesc) & 0222)
5779			fops = &vcpu_stat_fops;
5780		else
5781			fops = &vcpu_stat_readonly_fops;
5782		debugfs_create_file(pdesc->name, kvm_stats_debugfs_mode(pdesc),
5783				kvm_debugfs_dir,
5784				(void *)(long)pdesc->desc.offset, fops);
5785	}
5786}
5787
5788static int kvm_suspend(void)
5789{
5790	if (kvm_usage_count)
5791		hardware_disable_nolock(NULL);
5792	return 0;
5793}
5794
5795static void kvm_resume(void)
5796{
5797	if (kvm_usage_count) {
5798		lockdep_assert_not_held(&kvm_count_lock);
5799		hardware_enable_nolock(NULL);
5800	}
5801}
5802
5803static struct syscore_ops kvm_syscore_ops = {
5804	.suspend = kvm_suspend,
5805	.resume = kvm_resume,
5806};
5807
5808static inline
5809struct kvm_vcpu *preempt_notifier_to_vcpu(struct preempt_notifier *pn)
5810{
5811	return container_of(pn, struct kvm_vcpu, preempt_notifier);
5812}
5813
5814static void kvm_sched_in(struct preempt_notifier *pn, int cpu)
5815{
5816	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5817
5818	WRITE_ONCE(vcpu->preempted, false);
5819	WRITE_ONCE(vcpu->ready, false);
5820
5821	__this_cpu_write(kvm_running_vcpu, vcpu);
5822	kvm_arch_sched_in(vcpu, cpu);
5823	kvm_arch_vcpu_load(vcpu, cpu);
5824}
5825
5826static void kvm_sched_out(struct preempt_notifier *pn,
5827			  struct task_struct *next)
5828{
5829	struct kvm_vcpu *vcpu = preempt_notifier_to_vcpu(pn);
5830
5831	if (current->on_rq) {
5832		WRITE_ONCE(vcpu->preempted, true);
5833		WRITE_ONCE(vcpu->ready, true);
5834	}
5835	kvm_arch_vcpu_put(vcpu);
5836	__this_cpu_write(kvm_running_vcpu, NULL);
5837}
5838
5839/**
5840 * kvm_get_running_vcpu - get the vcpu running on the current CPU.
5841 *
5842 * We can disable preemption locally around accessing the per-CPU variable,
5843 * and use the resolved vcpu pointer after enabling preemption again,
5844 * because even if the current thread is migrated to another CPU, reading
5845 * the per-CPU value later will give us the same value as we update the
5846 * per-CPU variable in the preempt notifier handlers.
5847 */
5848struct kvm_vcpu *kvm_get_running_vcpu(void)
5849{
5850	struct kvm_vcpu *vcpu;
5851
5852	preempt_disable();
5853	vcpu = __this_cpu_read(kvm_running_vcpu);
5854	preempt_enable();
5855
5856	return vcpu;
5857}
5858EXPORT_SYMBOL_GPL(kvm_get_running_vcpu);
5859
5860/**
5861 * kvm_get_running_vcpus - get the per-CPU array of currently running vcpus.
5862 */
5863struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
5864{
5865        return &kvm_running_vcpu;
5866}
5867
5868#ifdef CONFIG_GUEST_PERF_EVENTS
5869static unsigned int kvm_guest_state(void)
5870{
5871	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5872	unsigned int state;
5873
5874	if (!kvm_arch_pmi_in_guest(vcpu))
5875		return 0;
5876
5877	state = PERF_GUEST_ACTIVE;
5878	if (!kvm_arch_vcpu_in_kernel(vcpu))
5879		state |= PERF_GUEST_USER;
5880
5881	return state;
5882}
5883
5884static unsigned long kvm_guest_get_ip(void)
5885{
5886	struct kvm_vcpu *vcpu = kvm_get_running_vcpu();
5887
5888	/* Retrieving the IP must be guarded by a call to kvm_guest_state(). */
5889	if (WARN_ON_ONCE(!kvm_arch_pmi_in_guest(vcpu)))
5890		return 0;
5891
5892	return kvm_arch_vcpu_get_ip(vcpu);
5893}
5894
5895static struct perf_guest_info_callbacks kvm_guest_cbs = {
5896	.state			= kvm_guest_state,
5897	.get_ip			= kvm_guest_get_ip,
5898	.handle_intel_pt_intr	= NULL,
5899};
5900
5901void kvm_register_perf_callbacks(unsigned int (*pt_intr_handler)(void))
5902{
5903	kvm_guest_cbs.handle_intel_pt_intr = pt_intr_handler;
5904	perf_register_guest_info_callbacks(&kvm_guest_cbs);
5905}
5906void kvm_unregister_perf_callbacks(void)
5907{
5908	perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
5909}
5910#endif
5911
5912struct kvm_cpu_compat_check {
5913	void *opaque;
5914	int *ret;
5915};
5916
5917static void check_processor_compat(void *data)
5918{
5919	struct kvm_cpu_compat_check *c = data;
5920
5921	*c->ret = kvm_arch_check_processor_compat(c->opaque);
5922}
5923
5924int kvm_init(void *opaque, unsigned vcpu_size, unsigned vcpu_align,
5925		  struct module *module)
5926{
5927	struct kvm_cpu_compat_check c;
5928	int r;
5929	int cpu;
5930
5931	r = kvm_arch_init(opaque);
 
 
5932	if (r)
5933		goto out_fail;
5934
5935	/*
5936	 * kvm_arch_init makes sure there's at most one caller
5937	 * for architectures that support multiple implementations,
5938	 * like intel and amd on x86.
5939	 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
5940	 * conflicts in case kvm is already setup for another implementation.
5941	 */
5942	r = kvm_irqfd_init();
5943	if (r)
5944		goto out_irqfd;
5945
5946	if (!zalloc_cpumask_var(&cpus_hardware_enabled, GFP_KERNEL)) {
5947		r = -ENOMEM;
5948		goto out_free_0;
5949	}
5950
5951	r = kvm_arch_hardware_setup(opaque);
5952	if (r < 0)
5953		goto out_free_1;
5954
5955	c.ret = &r;
5956	c.opaque = opaque;
5957	for_each_online_cpu(cpu) {
5958		smp_call_function_single(cpu, check_processor_compat, &c, 1);
5959		if (r < 0)
5960			goto out_free_2;
5961	}
5962
5963	r = cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING, "kvm/cpu:starting",
5964				      kvm_starting_cpu, kvm_dying_cpu);
5965	if (r)
5966		goto out_free_2;
5967	register_reboot_notifier(&kvm_reboot_notifier);
5968
5969	/* A kmem cache lets us meet the alignment requirements of fx_save. */
5970	if (!vcpu_align)
5971		vcpu_align = __alignof__(struct kvm_vcpu);
5972	kvm_vcpu_cache =
5973		kmem_cache_create_usercopy("kvm_vcpu", vcpu_size, vcpu_align,
5974					   SLAB_ACCOUNT,
5975					   offsetof(struct kvm_vcpu, arch),
5976					   offsetofend(struct kvm_vcpu, stats_id)
5977					   - offsetof(struct kvm_vcpu, arch),
5978					   NULL);
5979	if (!kvm_vcpu_cache) {
5980		r = -ENOMEM;
5981		goto out_free_3;
5982	}
5983
5984	for_each_possible_cpu(cpu) {
5985		if (!alloc_cpumask_var_node(&per_cpu(cpu_kick_mask, cpu),
5986					    GFP_KERNEL, cpu_to_node(cpu))) {
5987			r = -ENOMEM;
5988			goto out_free_4;
5989		}
5990	}
5991
 
 
 
 
5992	r = kvm_async_pf_init();
5993	if (r)
5994		goto out_free_4;
5995
5996	kvm_chardev_ops.owner = module;
5997
5998	r = misc_register(&kvm_dev);
5999	if (r) {
6000		pr_err("kvm: misc device register failed\n");
6001		goto out_unreg;
6002	}
6003
6004	register_syscore_ops(&kvm_syscore_ops);
6005
6006	kvm_preempt_ops.sched_in = kvm_sched_in;
6007	kvm_preempt_ops.sched_out = kvm_sched_out;
6008
6009	kvm_init_debug();
6010
6011	r = kvm_vfio_ops_init();
6012	WARN_ON(r);
 
 
 
 
 
 
 
 
 
 
 
 
 
6013
6014	return 0;
6015
6016out_unreg:
 
 
6017	kvm_async_pf_deinit();
6018out_free_4:
 
 
 
6019	for_each_possible_cpu(cpu)
6020		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6021	kmem_cache_destroy(kvm_vcpu_cache);
6022out_free_3:
6023	unregister_reboot_notifier(&kvm_reboot_notifier);
6024	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
6025out_free_2:
6026	kvm_arch_hardware_unsetup();
6027out_free_1:
6028	free_cpumask_var(cpus_hardware_enabled);
6029out_free_0:
6030	kvm_irqfd_exit();
6031out_irqfd:
6032	kvm_arch_exit();
6033out_fail:
6034	return r;
6035}
6036EXPORT_SYMBOL_GPL(kvm_init);
6037
6038void kvm_exit(void)
6039{
6040	int cpu;
6041
6042	debugfs_remove_recursive(kvm_debugfs_dir);
 
 
 
 
6043	misc_deregister(&kvm_dev);
 
 
6044	for_each_possible_cpu(cpu)
6045		free_cpumask_var(per_cpu(cpu_kick_mask, cpu));
6046	kmem_cache_destroy(kvm_vcpu_cache);
 
6047	kvm_async_pf_deinit();
 
6048	unregister_syscore_ops(&kvm_syscore_ops);
6049	unregister_reboot_notifier(&kvm_reboot_notifier);
6050	cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING);
6051	on_each_cpu(hardware_disable_nolock, NULL, 1);
6052	kvm_arch_hardware_unsetup();
6053	kvm_arch_exit();
6054	kvm_irqfd_exit();
6055	free_cpumask_var(cpus_hardware_enabled);
6056	kvm_vfio_ops_exit();
6057}
6058EXPORT_SYMBOL_GPL(kvm_exit);
6059
6060struct kvm_vm_worker_thread_context {
6061	struct kvm *kvm;
6062	struct task_struct *parent;
6063	struct completion init_done;
6064	kvm_vm_thread_fn_t thread_fn;
6065	uintptr_t data;
6066	int err;
6067};
6068
6069static int kvm_vm_worker_thread(void *context)
6070{
6071	/*
6072	 * The init_context is allocated on the stack of the parent thread, so
6073	 * we have to locally copy anything that is needed beyond initialization
6074	 */
6075	struct kvm_vm_worker_thread_context *init_context = context;
6076	struct task_struct *parent;
6077	struct kvm *kvm = init_context->kvm;
6078	kvm_vm_thread_fn_t thread_fn = init_context->thread_fn;
6079	uintptr_t data = init_context->data;
6080	int err;
6081
6082	err = kthread_park(current);
6083	/* kthread_park(current) is never supposed to return an error */
6084	WARN_ON(err != 0);
6085	if (err)
6086		goto init_complete;
6087
6088	err = cgroup_attach_task_all(init_context->parent, current);
6089	if (err) {
6090		kvm_err("%s: cgroup_attach_task_all failed with err %d\n",
6091			__func__, err);
6092		goto init_complete;
6093	}
6094
6095	set_user_nice(current, task_nice(init_context->parent));
6096
6097init_complete:
6098	init_context->err = err;
6099	complete(&init_context->init_done);
6100	init_context = NULL;
6101
6102	if (err)
6103		goto out;
6104
6105	/* Wait to be woken up by the spawner before proceeding. */
6106	kthread_parkme();
6107
6108	if (!kthread_should_stop())
6109		err = thread_fn(kvm, data);
6110
6111out:
6112	/*
6113	 * Move kthread back to its original cgroup to prevent it lingering in
6114	 * the cgroup of the VM process, after the latter finishes its
6115	 * execution.
6116	 *
6117	 * kthread_stop() waits on the 'exited' completion condition which is
6118	 * set in exit_mm(), via mm_release(), in do_exit(). However, the
6119	 * kthread is removed from the cgroup in the cgroup_exit() which is
6120	 * called after the exit_mm(). This causes the kthread_stop() to return
6121	 * before the kthread actually quits the cgroup.
6122	 */
6123	rcu_read_lock();
6124	parent = rcu_dereference(current->real_parent);
6125	get_task_struct(parent);
6126	rcu_read_unlock();
6127	cgroup_attach_task_all(parent, current);
6128	put_task_struct(parent);
6129
6130	return err;
6131}
6132
6133int kvm_vm_create_worker_thread(struct kvm *kvm, kvm_vm_thread_fn_t thread_fn,
6134				uintptr_t data, const char *name,
6135				struct task_struct **thread_ptr)
6136{
6137	struct kvm_vm_worker_thread_context init_context = {};
6138	struct task_struct *thread;
6139
6140	*thread_ptr = NULL;
6141	init_context.kvm = kvm;
6142	init_context.parent = current;
6143	init_context.thread_fn = thread_fn;
6144	init_context.data = data;
6145	init_completion(&init_context.init_done);
6146
6147	thread = kthread_run(kvm_vm_worker_thread, &init_context,
6148			     "%s-%d", name, task_pid_nr(current));
6149	if (IS_ERR(thread))
6150		return PTR_ERR(thread);
6151
6152	/* kthread_run is never supposed to return NULL */
6153	WARN_ON(thread == NULL);
6154
6155	wait_for_completion(&init_context.init_done);
6156
6157	if (!init_context.err)
6158		*thread_ptr = thread;
6159
6160	return init_context.err;
6161}