Linux Audio

Check our new training course

Loading...
Note: File does not exist in v4.17.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   4 * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   5 */
   6
   7#include <linux/bug.h>
   8#include <linux/cpu_pm.h>
   9#include <linux/entry-kvm.h>
  10#include <linux/errno.h>
  11#include <linux/err.h>
  12#include <linux/kvm_host.h>
  13#include <linux/list.h>
  14#include <linux/module.h>
  15#include <linux/vmalloc.h>
  16#include <linux/fs.h>
  17#include <linux/mman.h>
  18#include <linux/sched.h>
  19#include <linux/kvm.h>
  20#include <linux/kvm_irqfd.h>
  21#include <linux/irqbypass.h>
  22#include <linux/sched/stat.h>
  23#include <linux/psci.h>
  24#include <trace/events/kvm.h>
  25
  26#define CREATE_TRACE_POINTS
  27#include "trace_arm.h"
  28
  29#include <linux/uaccess.h>
  30#include <asm/ptrace.h>
  31#include <asm/mman.h>
  32#include <asm/tlbflush.h>
  33#include <asm/cacheflush.h>
  34#include <asm/cpufeature.h>
  35#include <asm/virt.h>
  36#include <asm/kvm_arm.h>
  37#include <asm/kvm_asm.h>
  38#include <asm/kvm_emulate.h>
  39#include <asm/kvm_mmu.h>
  40#include <asm/kvm_nested.h>
  41#include <asm/kvm_pkvm.h>
  42#include <asm/kvm_ptrauth.h>
  43#include <asm/sections.h>
  44
  45#include <kvm/arm_hypercalls.h>
  46#include <kvm/arm_pmu.h>
  47#include <kvm/arm_psci.h>
  48
  49#include "sys_regs.h"
  50
  51static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT;
  52
  53enum kvm_wfx_trap_policy {
  54	KVM_WFX_NOTRAP_SINGLE_TASK, /* Default option */
  55	KVM_WFX_NOTRAP,
  56	KVM_WFX_TRAP,
  57};
  58
  59static enum kvm_wfx_trap_policy kvm_wfi_trap_policy __read_mostly = KVM_WFX_NOTRAP_SINGLE_TASK;
  60static enum kvm_wfx_trap_policy kvm_wfe_trap_policy __read_mostly = KVM_WFX_NOTRAP_SINGLE_TASK;
  61
  62DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector);
  63
  64DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  65DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params);
  66
  67DECLARE_KVM_NVHE_PER_CPU(struct kvm_cpu_context, kvm_hyp_ctxt);
  68
  69static bool vgic_present, kvm_arm_initialised;
  70
  71static DEFINE_PER_CPU(unsigned char, kvm_hyp_initialized);
  72
  73bool is_kvm_arm_initialised(void)
  74{
  75	return kvm_arm_initialised;
  76}
  77
  78int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  79{
  80	return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  81}
  82
  83/*
  84 * This functions as an allow-list of protected VM capabilities.
  85 * Features not explicitly allowed by this function are denied.
  86 */
  87static bool pkvm_ext_allowed(struct kvm *kvm, long ext)
  88{
  89	switch (ext) {
  90	case KVM_CAP_IRQCHIP:
  91	case KVM_CAP_ARM_PSCI:
  92	case KVM_CAP_ARM_PSCI_0_2:
  93	case KVM_CAP_NR_VCPUS:
  94	case KVM_CAP_MAX_VCPUS:
  95	case KVM_CAP_MAX_VCPU_ID:
  96	case KVM_CAP_MSI_DEVID:
  97	case KVM_CAP_ARM_VM_IPA_SIZE:
  98	case KVM_CAP_ARM_PMU_V3:
  99	case KVM_CAP_ARM_SVE:
 100	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
 101	case KVM_CAP_ARM_PTRAUTH_GENERIC:
 102		return true;
 103	default:
 104		return false;
 105	}
 106}
 107
 108int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 109			    struct kvm_enable_cap *cap)
 110{
 111	int r = -EINVAL;
 112
 113	if (cap->flags)
 114		return -EINVAL;
 115
 116	if (kvm_vm_is_protected(kvm) && !pkvm_ext_allowed(kvm, cap->cap))
 117		return -EINVAL;
 118
 119	switch (cap->cap) {
 120	case KVM_CAP_ARM_NISV_TO_USER:
 121		r = 0;
 122		set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER,
 123			&kvm->arch.flags);
 124		break;
 125	case KVM_CAP_ARM_MTE:
 126		mutex_lock(&kvm->lock);
 127		if (system_supports_mte() && !kvm->created_vcpus) {
 128			r = 0;
 129			set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags);
 130		}
 131		mutex_unlock(&kvm->lock);
 132		break;
 133	case KVM_CAP_ARM_SYSTEM_SUSPEND:
 134		r = 0;
 135		set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags);
 136		break;
 137	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
 138		mutex_lock(&kvm->slots_lock);
 139		/*
 140		 * To keep things simple, allow changing the chunk
 141		 * size only when no memory slots have been created.
 142		 */
 143		if (kvm_are_all_memslots_empty(kvm)) {
 144			u64 new_cap = cap->args[0];
 145
 146			if (!new_cap || kvm_is_block_size_supported(new_cap)) {
 147				r = 0;
 148				kvm->arch.mmu.split_page_chunk_size = new_cap;
 149			}
 150		}
 151		mutex_unlock(&kvm->slots_lock);
 152		break;
 153	default:
 154		break;
 155	}
 156
 157	return r;
 158}
 159
 160static int kvm_arm_default_max_vcpus(void)
 161{
 162	return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 163}
 164
 165/**
 166 * kvm_arch_init_vm - initializes a VM data structure
 167 * @kvm:	pointer to the KVM struct
 168 * @type:	kvm device type
 169 */
 170int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 171{
 172	int ret;
 173
 174	mutex_init(&kvm->arch.config_lock);
 175
 176#ifdef CONFIG_LOCKDEP
 177	/* Clue in lockdep that the config_lock must be taken inside kvm->lock */
 178	mutex_lock(&kvm->lock);
 179	mutex_lock(&kvm->arch.config_lock);
 180	mutex_unlock(&kvm->arch.config_lock);
 181	mutex_unlock(&kvm->lock);
 182#endif
 183
 184	kvm_init_nested(kvm);
 185
 186	ret = kvm_share_hyp(kvm, kvm + 1);
 187	if (ret)
 188		return ret;
 189
 190	ret = pkvm_init_host_vm(kvm);
 191	if (ret)
 192		goto err_unshare_kvm;
 193
 194	if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL_ACCOUNT)) {
 195		ret = -ENOMEM;
 196		goto err_unshare_kvm;
 197	}
 198	cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask);
 199
 200	ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type);
 201	if (ret)
 202		goto err_free_cpumask;
 203
 204	kvm_vgic_early_init(kvm);
 205
 206	kvm_timer_init_vm(kvm);
 207
 208	/* The maximum number of VCPUs is limited by the host's GIC model */
 209	kvm->max_vcpus = kvm_arm_default_max_vcpus();
 210
 211	kvm_arm_init_hypercalls(kvm);
 212
 213	bitmap_zero(kvm->arch.vcpu_features, KVM_VCPU_MAX_FEATURES);
 214
 215	return 0;
 216
 217err_free_cpumask:
 218	free_cpumask_var(kvm->arch.supported_cpus);
 219err_unshare_kvm:
 220	kvm_unshare_hyp(kvm, kvm + 1);
 221	return ret;
 222}
 223
 224vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 225{
 226	return VM_FAULT_SIGBUS;
 227}
 228
 229void kvm_arch_create_vm_debugfs(struct kvm *kvm)
 230{
 231	kvm_sys_regs_create_debugfs(kvm);
 232	kvm_s2_ptdump_create_debugfs(kvm);
 233}
 234
 235static void kvm_destroy_mpidr_data(struct kvm *kvm)
 236{
 237	struct kvm_mpidr_data *data;
 238
 239	mutex_lock(&kvm->arch.config_lock);
 240
 241	data = rcu_dereference_protected(kvm->arch.mpidr_data,
 242					 lockdep_is_held(&kvm->arch.config_lock));
 243	if (data) {
 244		rcu_assign_pointer(kvm->arch.mpidr_data, NULL);
 245		synchronize_rcu();
 246		kfree(data);
 247	}
 248
 249	mutex_unlock(&kvm->arch.config_lock);
 250}
 251
 252/**
 253 * kvm_arch_destroy_vm - destroy the VM data structure
 254 * @kvm:	pointer to the KVM struct
 255 */
 256void kvm_arch_destroy_vm(struct kvm *kvm)
 257{
 258	bitmap_free(kvm->arch.pmu_filter);
 259	free_cpumask_var(kvm->arch.supported_cpus);
 260
 261	kvm_vgic_destroy(kvm);
 262
 263	if (is_protected_kvm_enabled())
 264		pkvm_destroy_hyp_vm(kvm);
 265
 266	kvm_destroy_mpidr_data(kvm);
 267
 268	kfree(kvm->arch.sysreg_masks);
 269	kvm_destroy_vcpus(kvm);
 270
 271	kvm_unshare_hyp(kvm, kvm + 1);
 272
 273	kvm_arm_teardown_hypercalls(kvm);
 274}
 275
 276static bool kvm_has_full_ptr_auth(void)
 277{
 278	bool apa, gpa, api, gpi, apa3, gpa3;
 279	u64 isar1, isar2, val;
 280
 281	/*
 282	 * Check that:
 283	 *
 284	 * - both Address and Generic auth are implemented for a given
 285         *   algorithm (Q5, IMPDEF or Q3)
 286	 * - only a single algorithm is implemented.
 287	 */
 288	if (!system_has_full_ptr_auth())
 289		return false;
 290
 291	isar1 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
 292	isar2 = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
 293
 294	apa = !!FIELD_GET(ID_AA64ISAR1_EL1_APA_MASK, isar1);
 295	val = FIELD_GET(ID_AA64ISAR1_EL1_GPA_MASK, isar1);
 296	gpa = (val == ID_AA64ISAR1_EL1_GPA_IMP);
 297
 298	api = !!FIELD_GET(ID_AA64ISAR1_EL1_API_MASK, isar1);
 299	val = FIELD_GET(ID_AA64ISAR1_EL1_GPI_MASK, isar1);
 300	gpi = (val == ID_AA64ISAR1_EL1_GPI_IMP);
 301
 302	apa3 = !!FIELD_GET(ID_AA64ISAR2_EL1_APA3_MASK, isar2);
 303	val  = FIELD_GET(ID_AA64ISAR2_EL1_GPA3_MASK, isar2);
 304	gpa3 = (val == ID_AA64ISAR2_EL1_GPA3_IMP);
 305
 306	return (apa == gpa && api == gpi && apa3 == gpa3 &&
 307		(apa + api + apa3) == 1);
 308}
 309
 310int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 311{
 312	int r;
 313
 314	if (kvm && kvm_vm_is_protected(kvm) && !pkvm_ext_allowed(kvm, ext))
 315		return 0;
 316
 317	switch (ext) {
 318	case KVM_CAP_IRQCHIP:
 319		r = vgic_present;
 320		break;
 321	case KVM_CAP_IOEVENTFD:
 322	case KVM_CAP_USER_MEMORY:
 323	case KVM_CAP_SYNC_MMU:
 324	case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 325	case KVM_CAP_ONE_REG:
 326	case KVM_CAP_ARM_PSCI:
 327	case KVM_CAP_ARM_PSCI_0_2:
 328	case KVM_CAP_READONLY_MEM:
 329	case KVM_CAP_MP_STATE:
 330	case KVM_CAP_IMMEDIATE_EXIT:
 331	case KVM_CAP_VCPU_EVENTS:
 332	case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
 333	case KVM_CAP_ARM_NISV_TO_USER:
 334	case KVM_CAP_ARM_INJECT_EXT_DABT:
 335	case KVM_CAP_SET_GUEST_DEBUG:
 336	case KVM_CAP_VCPU_ATTRIBUTES:
 337	case KVM_CAP_PTP_KVM:
 338	case KVM_CAP_ARM_SYSTEM_SUSPEND:
 339	case KVM_CAP_IRQFD_RESAMPLE:
 340	case KVM_CAP_COUNTER_OFFSET:
 341		r = 1;
 342		break;
 343	case KVM_CAP_SET_GUEST_DEBUG2:
 344		return KVM_GUESTDBG_VALID_MASK;
 345	case KVM_CAP_ARM_SET_DEVICE_ADDR:
 346		r = 1;
 347		break;
 348	case KVM_CAP_NR_VCPUS:
 349		/*
 350		 * ARM64 treats KVM_CAP_NR_CPUS differently from all other
 351		 * architectures, as it does not always bound it to
 352		 * KVM_CAP_MAX_VCPUS. It should not matter much because
 353		 * this is just an advisory value.
 354		 */
 355		r = min_t(unsigned int, num_online_cpus(),
 356			  kvm_arm_default_max_vcpus());
 357		break;
 358	case KVM_CAP_MAX_VCPUS:
 359	case KVM_CAP_MAX_VCPU_ID:
 360		if (kvm)
 361			r = kvm->max_vcpus;
 362		else
 363			r = kvm_arm_default_max_vcpus();
 364		break;
 365	case KVM_CAP_MSI_DEVID:
 366		if (!kvm)
 367			r = -EINVAL;
 368		else
 369			r = kvm->arch.vgic.msis_require_devid;
 370		break;
 371	case KVM_CAP_ARM_USER_IRQ:
 372		/*
 373		 * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 374		 * (bump this number if adding more devices)
 375		 */
 376		r = 1;
 377		break;
 378	case KVM_CAP_ARM_MTE:
 379		r = system_supports_mte();
 380		break;
 381	case KVM_CAP_STEAL_TIME:
 382		r = kvm_arm_pvtime_supported();
 383		break;
 384	case KVM_CAP_ARM_EL1_32BIT:
 385		r = cpus_have_final_cap(ARM64_HAS_32BIT_EL1);
 386		break;
 387	case KVM_CAP_GUEST_DEBUG_HW_BPS:
 388		r = get_num_brps();
 389		break;
 390	case KVM_CAP_GUEST_DEBUG_HW_WPS:
 391		r = get_num_wrps();
 392		break;
 393	case KVM_CAP_ARM_PMU_V3:
 394		r = kvm_arm_support_pmu_v3();
 395		break;
 396	case KVM_CAP_ARM_INJECT_SERROR_ESR:
 397		r = cpus_have_final_cap(ARM64_HAS_RAS_EXTN);
 398		break;
 399	case KVM_CAP_ARM_VM_IPA_SIZE:
 400		r = get_kvm_ipa_limit();
 401		break;
 402	case KVM_CAP_ARM_SVE:
 403		r = system_supports_sve();
 404		break;
 405	case KVM_CAP_ARM_PTRAUTH_ADDRESS:
 406	case KVM_CAP_ARM_PTRAUTH_GENERIC:
 407		r = kvm_has_full_ptr_auth();
 408		break;
 409	case KVM_CAP_ARM_EAGER_SPLIT_CHUNK_SIZE:
 410		if (kvm)
 411			r = kvm->arch.mmu.split_page_chunk_size;
 412		else
 413			r = KVM_ARM_EAGER_SPLIT_CHUNK_SIZE_DEFAULT;
 414		break;
 415	case KVM_CAP_ARM_SUPPORTED_BLOCK_SIZES:
 416		r = kvm_supported_block_sizes();
 417		break;
 418	case KVM_CAP_ARM_SUPPORTED_REG_MASK_RANGES:
 419		r = BIT(0);
 420		break;
 421	default:
 422		r = 0;
 423	}
 424
 425	return r;
 426}
 427
 428long kvm_arch_dev_ioctl(struct file *filp,
 429			unsigned int ioctl, unsigned long arg)
 430{
 431	return -EINVAL;
 432}
 433
 434struct kvm *kvm_arch_alloc_vm(void)
 435{
 436	size_t sz = sizeof(struct kvm);
 437
 438	if (!has_vhe())
 439		return kzalloc(sz, GFP_KERNEL_ACCOUNT);
 440
 441	return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO);
 442}
 443
 444int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
 445{
 446	if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
 447		return -EBUSY;
 448
 449	if (id >= kvm->max_vcpus)
 450		return -EINVAL;
 451
 452	return 0;
 453}
 454
 455int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
 456{
 457	int err;
 458
 459	spin_lock_init(&vcpu->arch.mp_state_lock);
 460
 461#ifdef CONFIG_LOCKDEP
 462	/* Inform lockdep that the config_lock is acquired after vcpu->mutex */
 463	mutex_lock(&vcpu->mutex);
 464	mutex_lock(&vcpu->kvm->arch.config_lock);
 465	mutex_unlock(&vcpu->kvm->arch.config_lock);
 466	mutex_unlock(&vcpu->mutex);
 467#endif
 468
 469	/* Force users to call KVM_ARM_VCPU_INIT */
 470	vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
 471
 472	vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO;
 473
 474	/* Set up the timer */
 475	kvm_timer_vcpu_init(vcpu);
 476
 477	kvm_pmu_vcpu_init(vcpu);
 478
 479	kvm_arm_reset_debug_ptr(vcpu);
 480
 481	kvm_arm_pvtime_vcpu_init(&vcpu->arch);
 482
 483	vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu;
 484
 485	/*
 486	 * This vCPU may have been created after mpidr_data was initialized.
 487	 * Throw out the pre-computed mappings if that is the case which forces
 488	 * KVM to fall back to iteratively searching the vCPUs.
 489	 */
 490	kvm_destroy_mpidr_data(vcpu->kvm);
 491
 492	err = kvm_vgic_vcpu_init(vcpu);
 493	if (err)
 494		return err;
 495
 496	return kvm_share_hyp(vcpu, vcpu + 1);
 497}
 498
 499void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 500{
 501}
 502
 503void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 504{
 505	kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache);
 506	kvm_timer_vcpu_terminate(vcpu);
 507	kvm_pmu_vcpu_destroy(vcpu);
 508	kvm_vgic_vcpu_destroy(vcpu);
 509	kvm_arm_vcpu_destroy(vcpu);
 510}
 511
 512void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 513{
 514
 515}
 516
 517void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 518{
 519
 520}
 521
 522static void vcpu_set_pauth_traps(struct kvm_vcpu *vcpu)
 523{
 524	if (vcpu_has_ptrauth(vcpu) && !is_protected_kvm_enabled()) {
 525		/*
 526		 * Either we're running an L2 guest, and the API/APK bits come
 527		 * from L1's HCR_EL2, or API/APK are both set.
 528		 */
 529		if (unlikely(vcpu_has_nv(vcpu) && !is_hyp_ctxt(vcpu))) {
 530			u64 val;
 531
 532			val = __vcpu_sys_reg(vcpu, HCR_EL2);
 533			val &= (HCR_API | HCR_APK);
 534			vcpu->arch.hcr_el2 &= ~(HCR_API | HCR_APK);
 535			vcpu->arch.hcr_el2 |= val;
 536		} else {
 537			vcpu->arch.hcr_el2 |= (HCR_API | HCR_APK);
 538		}
 539
 540		/*
 541		 * Save the host keys if there is any chance for the guest
 542		 * to use pauth, as the entry code will reload the guest
 543		 * keys in that case.
 544		 */
 545		if (vcpu->arch.hcr_el2 & (HCR_API | HCR_APK)) {
 546			struct kvm_cpu_context *ctxt;
 547
 548			ctxt = this_cpu_ptr_hyp_sym(kvm_hyp_ctxt);
 549			ptrauth_save_keys(ctxt);
 550		}
 551	}
 552}
 553
 554static bool kvm_vcpu_should_clear_twi(struct kvm_vcpu *vcpu)
 555{
 556	if (unlikely(kvm_wfi_trap_policy != KVM_WFX_NOTRAP_SINGLE_TASK))
 557		return kvm_wfi_trap_policy == KVM_WFX_NOTRAP;
 558
 559	return single_task_running() &&
 560	       (atomic_read(&vcpu->arch.vgic_cpu.vgic_v3.its_vpe.vlpi_count) ||
 561		vcpu->kvm->arch.vgic.nassgireq);
 562}
 563
 564static bool kvm_vcpu_should_clear_twe(struct kvm_vcpu *vcpu)
 565{
 566	if (unlikely(kvm_wfe_trap_policy != KVM_WFX_NOTRAP_SINGLE_TASK))
 567		return kvm_wfe_trap_policy == KVM_WFX_NOTRAP;
 568
 569	return single_task_running();
 570}
 571
 572void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 573{
 574	struct kvm_s2_mmu *mmu;
 575	int *last_ran;
 576
 577	if (vcpu_has_nv(vcpu))
 578		kvm_vcpu_load_hw_mmu(vcpu);
 579
 580	mmu = vcpu->arch.hw_mmu;
 581	last_ran = this_cpu_ptr(mmu->last_vcpu_ran);
 582
 583	/*
 584	 * Ensure a VMID is allocated for the MMU before programming VTTBR_EL2,
 585	 * which happens eagerly in VHE.
 586	 *
 587	 * Also, the VMID allocator only preserves VMIDs that are active at the
 588	 * time of rollover, so KVM might need to grab a new VMID for the MMU if
 589	 * this is called from kvm_sched_in().
 590	 */
 591	kvm_arm_vmid_update(&mmu->vmid);
 592
 593	/*
 594	 * We guarantee that both TLBs and I-cache are private to each
 595	 * vcpu. If detecting that a vcpu from the same VM has
 596	 * previously run on the same physical CPU, call into the
 597	 * hypervisor code to nuke the relevant contexts.
 598	 *
 599	 * We might get preempted before the vCPU actually runs, but
 600	 * over-invalidation doesn't affect correctness.
 601	 */
 602	if (*last_ran != vcpu->vcpu_idx) {
 603		kvm_call_hyp(__kvm_flush_cpu_context, mmu);
 604		*last_ran = vcpu->vcpu_idx;
 605	}
 606
 607	vcpu->cpu = cpu;
 608
 609	kvm_vgic_load(vcpu);
 610	kvm_timer_vcpu_load(vcpu);
 611	if (has_vhe())
 612		kvm_vcpu_load_vhe(vcpu);
 613	kvm_arch_vcpu_load_fp(vcpu);
 614	kvm_vcpu_pmu_restore_guest(vcpu);
 615	if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
 616		kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
 617
 618	if (kvm_vcpu_should_clear_twe(vcpu))
 619		vcpu->arch.hcr_el2 &= ~HCR_TWE;
 620	else
 621		vcpu->arch.hcr_el2 |= HCR_TWE;
 622
 623	if (kvm_vcpu_should_clear_twi(vcpu))
 624		vcpu->arch.hcr_el2 &= ~HCR_TWI;
 625	else
 626		vcpu->arch.hcr_el2 |= HCR_TWI;
 627
 628	vcpu_set_pauth_traps(vcpu);
 629
 630	kvm_arch_vcpu_load_debug_state_flags(vcpu);
 631
 632	if (!cpumask_test_cpu(cpu, vcpu->kvm->arch.supported_cpus))
 633		vcpu_set_on_unsupported_cpu(vcpu);
 634}
 635
 636void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 637{
 638	kvm_arch_vcpu_put_debug_state_flags(vcpu);
 639	kvm_arch_vcpu_put_fp(vcpu);
 640	if (has_vhe())
 641		kvm_vcpu_put_vhe(vcpu);
 642	kvm_timer_vcpu_put(vcpu);
 643	kvm_vgic_put(vcpu);
 644	kvm_vcpu_pmu_restore_host(vcpu);
 645	if (vcpu_has_nv(vcpu))
 646		kvm_vcpu_put_hw_mmu(vcpu);
 647	kvm_arm_vmid_clear_active();
 648
 649	vcpu_clear_on_unsupported_cpu(vcpu);
 650	vcpu->cpu = -1;
 651}
 652
 653static void __kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
 654{
 655	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_STOPPED);
 656	kvm_make_request(KVM_REQ_SLEEP, vcpu);
 657	kvm_vcpu_kick(vcpu);
 658}
 659
 660void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu)
 661{
 662	spin_lock(&vcpu->arch.mp_state_lock);
 663	__kvm_arm_vcpu_power_off(vcpu);
 664	spin_unlock(&vcpu->arch.mp_state_lock);
 665}
 666
 667bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu)
 668{
 669	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_STOPPED;
 670}
 671
 672static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu)
 673{
 674	WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_SUSPENDED);
 675	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
 676	kvm_vcpu_kick(vcpu);
 677}
 678
 679static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu)
 680{
 681	return READ_ONCE(vcpu->arch.mp_state.mp_state) == KVM_MP_STATE_SUSPENDED;
 682}
 683
 684int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 685				    struct kvm_mp_state *mp_state)
 686{
 687	*mp_state = READ_ONCE(vcpu->arch.mp_state);
 688
 689	return 0;
 690}
 691
 692int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 693				    struct kvm_mp_state *mp_state)
 694{
 695	int ret = 0;
 696
 697	spin_lock(&vcpu->arch.mp_state_lock);
 698
 699	switch (mp_state->mp_state) {
 700	case KVM_MP_STATE_RUNNABLE:
 701		WRITE_ONCE(vcpu->arch.mp_state, *mp_state);
 702		break;
 703	case KVM_MP_STATE_STOPPED:
 704		__kvm_arm_vcpu_power_off(vcpu);
 705		break;
 706	case KVM_MP_STATE_SUSPENDED:
 707		kvm_arm_vcpu_suspend(vcpu);
 708		break;
 709	default:
 710		ret = -EINVAL;
 711	}
 712
 713	spin_unlock(&vcpu->arch.mp_state_lock);
 714
 715	return ret;
 716}
 717
 718/**
 719 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 720 * @v:		The VCPU pointer
 721 *
 722 * If the guest CPU is not waiting for interrupts or an interrupt line is
 723 * asserted, the CPU is by definition runnable.
 724 */
 725int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 726{
 727	bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 728	return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 729		&& !kvm_arm_vcpu_stopped(v) && !v->arch.pause);
 730}
 731
 732bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 733{
 734	return vcpu_mode_priv(vcpu);
 735}
 736
 737#ifdef CONFIG_GUEST_PERF_EVENTS
 738unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
 739{
 740	return *vcpu_pc(vcpu);
 741}
 742#endif
 743
 744static void kvm_init_mpidr_data(struct kvm *kvm)
 745{
 746	struct kvm_mpidr_data *data = NULL;
 747	unsigned long c, mask, nr_entries;
 748	u64 aff_set = 0, aff_clr = ~0UL;
 749	struct kvm_vcpu *vcpu;
 750
 751	mutex_lock(&kvm->arch.config_lock);
 752
 753	if (rcu_access_pointer(kvm->arch.mpidr_data) ||
 754	    atomic_read(&kvm->online_vcpus) == 1)
 755		goto out;
 756
 757	kvm_for_each_vcpu(c, vcpu, kvm) {
 758		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
 759		aff_set |= aff;
 760		aff_clr &= aff;
 761	}
 762
 763	/*
 764	 * A significant bit can be either 0 or 1, and will only appear in
 765	 * aff_set. Use aff_clr to weed out the useless stuff.
 766	 */
 767	mask = aff_set ^ aff_clr;
 768	nr_entries = BIT_ULL(hweight_long(mask));
 769
 770	/*
 771	 * Don't let userspace fool us. If we need more than a single page
 772	 * to describe the compressed MPIDR array, just fall back to the
 773	 * iterative method. Single vcpu VMs do not need this either.
 774	 */
 775	if (struct_size(data, cmpidr_to_idx, nr_entries) <= PAGE_SIZE)
 776		data = kzalloc(struct_size(data, cmpidr_to_idx, nr_entries),
 777			       GFP_KERNEL_ACCOUNT);
 778
 779	if (!data)
 780		goto out;
 781
 782	data->mpidr_mask = mask;
 783
 784	kvm_for_each_vcpu(c, vcpu, kvm) {
 785		u64 aff = kvm_vcpu_get_mpidr_aff(vcpu);
 786		u16 index = kvm_mpidr_index(data, aff);
 787
 788		data->cmpidr_to_idx[index] = c;
 789	}
 790
 791	rcu_assign_pointer(kvm->arch.mpidr_data, data);
 792out:
 793	mutex_unlock(&kvm->arch.config_lock);
 794}
 795
 796/*
 797 * Handle both the initialisation that is being done when the vcpu is
 798 * run for the first time, as well as the updates that must be
 799 * performed each time we get a new thread dealing with this vcpu.
 800 */
 801int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
 802{
 803	struct kvm *kvm = vcpu->kvm;
 804	int ret;
 805
 806	if (!kvm_vcpu_initialized(vcpu))
 807		return -ENOEXEC;
 808
 809	if (!kvm_arm_vcpu_is_finalized(vcpu))
 810		return -EPERM;
 811
 812	ret = kvm_arch_vcpu_run_map_fp(vcpu);
 813	if (ret)
 814		return ret;
 815
 816	if (likely(vcpu_has_run_once(vcpu)))
 817		return 0;
 818
 819	kvm_init_mpidr_data(kvm);
 820
 821	kvm_arm_vcpu_init_debug(vcpu);
 822
 823	if (likely(irqchip_in_kernel(kvm))) {
 824		/*
 825		 * Map the VGIC hardware resources before running a vcpu the
 826		 * first time on this VM.
 827		 */
 828		ret = kvm_vgic_map_resources(kvm);
 829		if (ret)
 830			return ret;
 831	}
 832
 833	ret = kvm_finalize_sys_regs(vcpu);
 834	if (ret)
 835		return ret;
 836
 837	/*
 838	 * This needs to happen after any restriction has been applied
 839	 * to the feature set.
 840	 */
 841	kvm_calculate_traps(vcpu);
 842
 843	ret = kvm_timer_enable(vcpu);
 844	if (ret)
 845		return ret;
 846
 847	ret = kvm_arm_pmu_v3_enable(vcpu);
 848	if (ret)
 849		return ret;
 850
 851	if (is_protected_kvm_enabled()) {
 852		ret = pkvm_create_hyp_vm(kvm);
 853		if (ret)
 854			return ret;
 855	}
 856
 857	mutex_lock(&kvm->arch.config_lock);
 858	set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags);
 859	mutex_unlock(&kvm->arch.config_lock);
 860
 861	return ret;
 862}
 863
 864bool kvm_arch_intc_initialized(struct kvm *kvm)
 865{
 866	return vgic_initialized(kvm);
 867}
 868
 869void kvm_arm_halt_guest(struct kvm *kvm)
 870{
 871	unsigned long i;
 872	struct kvm_vcpu *vcpu;
 873
 874	kvm_for_each_vcpu(i, vcpu, kvm)
 875		vcpu->arch.pause = true;
 876	kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 877}
 878
 879void kvm_arm_resume_guest(struct kvm *kvm)
 880{
 881	unsigned long i;
 882	struct kvm_vcpu *vcpu;
 883
 884	kvm_for_each_vcpu(i, vcpu, kvm) {
 885		vcpu->arch.pause = false;
 886		__kvm_vcpu_wake_up(vcpu);
 887	}
 888}
 889
 890static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu)
 891{
 892	struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu);
 893
 894	rcuwait_wait_event(wait,
 895			   (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause),
 896			   TASK_INTERRUPTIBLE);
 897
 898	if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) {
 899		/* Awaken to handle a signal, request we sleep again later. */
 900		kvm_make_request(KVM_REQ_SLEEP, vcpu);
 901	}
 902
 903	/*
 904	 * Make sure we will observe a potential reset request if we've
 905	 * observed a change to the power state. Pairs with the smp_wmb() in
 906	 * kvm_psci_vcpu_on().
 907	 */
 908	smp_rmb();
 909}
 910
 911/**
 912 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior
 913 * @vcpu:	The VCPU pointer
 914 *
 915 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until
 916 * the vCPU is runnable.  The vCPU may or may not be scheduled out, depending
 917 * on when a wake event arrives, e.g. there may already be a pending wake event.
 918 */
 919void kvm_vcpu_wfi(struct kvm_vcpu *vcpu)
 920{
 921	/*
 922	 * Sync back the state of the GIC CPU interface so that we have
 923	 * the latest PMR and group enables. This ensures that
 924	 * kvm_arch_vcpu_runnable has up-to-date data to decide whether
 925	 * we have pending interrupts, e.g. when determining if the
 926	 * vCPU should block.
 927	 *
 928	 * For the same reason, we want to tell GICv4 that we need
 929	 * doorbells to be signalled, should an interrupt become pending.
 930	 */
 931	preempt_disable();
 932	vcpu_set_flag(vcpu, IN_WFI);
 933	kvm_vgic_put(vcpu);
 934	preempt_enable();
 935
 936	kvm_vcpu_halt(vcpu);
 937	vcpu_clear_flag(vcpu, IN_WFIT);
 938
 939	preempt_disable();
 940	vcpu_clear_flag(vcpu, IN_WFI);
 941	kvm_vgic_load(vcpu);
 942	preempt_enable();
 943}
 944
 945static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu)
 946{
 947	if (!kvm_arm_vcpu_suspended(vcpu))
 948		return 1;
 949
 950	kvm_vcpu_wfi(vcpu);
 951
 952	/*
 953	 * The suspend state is sticky; we do not leave it until userspace
 954	 * explicitly marks the vCPU as runnable. Request that we suspend again
 955	 * later.
 956	 */
 957	kvm_make_request(KVM_REQ_SUSPEND, vcpu);
 958
 959	/*
 960	 * Check to make sure the vCPU is actually runnable. If so, exit to
 961	 * userspace informing it of the wakeup condition.
 962	 */
 963	if (kvm_arch_vcpu_runnable(vcpu)) {
 964		memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event));
 965		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP;
 966		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
 967		return 0;
 968	}
 969
 970	/*
 971	 * Otherwise, we were unblocked to process a different event, such as a
 972	 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to
 973	 * process the event.
 974	 */
 975	return 1;
 976}
 977
 978/**
 979 * check_vcpu_requests - check and handle pending vCPU requests
 980 * @vcpu:	the VCPU pointer
 981 *
 982 * Return: 1 if we should enter the guest
 983 *	   0 if we should exit to userspace
 984 *	   < 0 if we should exit to userspace, where the return value indicates
 985 *	   an error
 986 */
 987static int check_vcpu_requests(struct kvm_vcpu *vcpu)
 988{
 989	if (kvm_request_pending(vcpu)) {
 990		if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu))
 991			return -EIO;
 992
 993		if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 994			kvm_vcpu_sleep(vcpu);
 995
 996		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 997			kvm_reset_vcpu(vcpu);
 998
 999		/*
1000		 * Clear IRQ_PENDING requests that were made to guarantee
1001		 * that a VCPU sees new virtual interrupts.
1002		 */
1003		kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
1004
1005		if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
1006			kvm_update_stolen_time(vcpu);
1007
1008		if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
1009			/* The distributor enable bits were changed */
1010			preempt_disable();
1011			vgic_v4_put(vcpu);
1012			vgic_v4_load(vcpu);
1013			preempt_enable();
1014		}
1015
1016		if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu))
1017			kvm_vcpu_reload_pmu(vcpu);
1018
1019		if (kvm_check_request(KVM_REQ_RESYNC_PMU_EL0, vcpu))
1020			kvm_vcpu_pmu_restore_guest(vcpu);
1021
1022		if (kvm_check_request(KVM_REQ_SUSPEND, vcpu))
1023			return kvm_vcpu_suspend(vcpu);
1024
1025		if (kvm_dirty_ring_check_request(vcpu))
1026			return 0;
1027
1028		check_nested_vcpu_requests(vcpu);
1029	}
1030
1031	return 1;
1032}
1033
1034static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu)
1035{
1036	if (likely(!vcpu_mode_is_32bit(vcpu)))
1037		return false;
1038
1039	if (vcpu_has_nv(vcpu))
1040		return true;
1041
1042	return !kvm_supports_32bit_el0();
1043}
1044
1045/**
1046 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest
1047 * @vcpu:	The VCPU pointer
1048 * @ret:	Pointer to write optional return code
1049 *
1050 * Returns: true if the VCPU needs to return to a preemptible + interruptible
1051 *	    and skip guest entry.
1052 *
1053 * This function disambiguates between two different types of exits: exits to a
1054 * preemptible + interruptible kernel context and exits to userspace. For an
1055 * exit to userspace, this function will write the return code to ret and return
1056 * true. For an exit to preemptible + interruptible kernel context (i.e. check
1057 * for pending work and re-enter), return true without writing to ret.
1058 */
1059static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret)
1060{
1061	struct kvm_run *run = vcpu->run;
1062
1063	/*
1064	 * If we're using a userspace irqchip, then check if we need
1065	 * to tell a userspace irqchip about timer or PMU level
1066	 * changes and if so, exit to userspace (the actual level
1067	 * state gets updated in kvm_timer_update_run and
1068	 * kvm_pmu_update_run below).
1069	 */
1070	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1071		if (kvm_timer_should_notify_user(vcpu) ||
1072		    kvm_pmu_should_notify_user(vcpu)) {
1073			*ret = -EINTR;
1074			run->exit_reason = KVM_EXIT_INTR;
1075			return true;
1076		}
1077	}
1078
1079	if (unlikely(vcpu_on_unsupported_cpu(vcpu))) {
1080		run->exit_reason = KVM_EXIT_FAIL_ENTRY;
1081		run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED;
1082		run->fail_entry.cpu = smp_processor_id();
1083		*ret = 0;
1084		return true;
1085	}
1086
1087	return kvm_request_pending(vcpu) ||
1088			xfer_to_guest_mode_work_pending();
1089}
1090
1091/*
1092 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while
1093 * the vCPU is running.
1094 *
1095 * This must be noinstr as instrumentation may make use of RCU, and this is not
1096 * safe during the EQS.
1097 */
1098static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu)
1099{
1100	int ret;
1101
1102	guest_state_enter_irqoff();
1103	ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu);
1104	guest_state_exit_irqoff();
1105
1106	return ret;
1107}
1108
1109/**
1110 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
1111 * @vcpu:	The VCPU pointer
1112 *
1113 * This function is called through the VCPU_RUN ioctl called from user space. It
1114 * will execute VM code in a loop until the time slice for the process is used
1115 * or some emulation is needed from user space in which case the function will
1116 * return with return value 0 and with the kvm_run structure filled in with the
1117 * required data for the requested emulation.
1118 */
1119int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
1120{
1121	struct kvm_run *run = vcpu->run;
1122	int ret;
1123
1124	if (run->exit_reason == KVM_EXIT_MMIO) {
1125		ret = kvm_handle_mmio_return(vcpu);
1126		if (ret <= 0)
1127			return ret;
1128	}
1129
1130	vcpu_load(vcpu);
1131
1132	if (!vcpu->wants_to_run) {
1133		ret = -EINTR;
1134		goto out;
1135	}
1136
1137	kvm_sigset_activate(vcpu);
1138
1139	ret = 1;
1140	run->exit_reason = KVM_EXIT_UNKNOWN;
1141	run->flags = 0;
1142	while (ret > 0) {
1143		/*
1144		 * Check conditions before entering the guest
1145		 */
1146		ret = xfer_to_guest_mode_handle_work(vcpu);
1147		if (!ret)
1148			ret = 1;
1149
1150		if (ret > 0)
1151			ret = check_vcpu_requests(vcpu);
1152
1153		/*
1154		 * Preparing the interrupts to be injected also
1155		 * involves poking the GIC, which must be done in a
1156		 * non-preemptible context.
1157		 */
1158		preempt_disable();
1159
1160		kvm_pmu_flush_hwstate(vcpu);
1161
1162		local_irq_disable();
1163
1164		kvm_vgic_flush_hwstate(vcpu);
1165
1166		kvm_pmu_update_vcpu_events(vcpu);
1167
1168		/*
1169		 * Ensure we set mode to IN_GUEST_MODE after we disable
1170		 * interrupts and before the final VCPU requests check.
1171		 * See the comment in kvm_vcpu_exiting_guest_mode() and
1172		 * Documentation/virt/kvm/vcpu-requests.rst
1173		 */
1174		smp_store_mb(vcpu->mode, IN_GUEST_MODE);
1175
1176		if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) {
1177			vcpu->mode = OUTSIDE_GUEST_MODE;
1178			isb(); /* Ensure work in x_flush_hwstate is committed */
1179			kvm_pmu_sync_hwstate(vcpu);
1180			if (unlikely(!irqchip_in_kernel(vcpu->kvm)))
1181				kvm_timer_sync_user(vcpu);
1182			kvm_vgic_sync_hwstate(vcpu);
1183			local_irq_enable();
1184			preempt_enable();
1185			continue;
1186		}
1187
1188		kvm_arm_setup_debug(vcpu);
1189		kvm_arch_vcpu_ctxflush_fp(vcpu);
1190
1191		/**************************************************************
1192		 * Enter the guest
1193		 */
1194		trace_kvm_entry(*vcpu_pc(vcpu));
1195		guest_timing_enter_irqoff();
1196
1197		ret = kvm_arm_vcpu_enter_exit(vcpu);
1198
1199		vcpu->mode = OUTSIDE_GUEST_MODE;
1200		vcpu->stat.exits++;
1201		/*
1202		 * Back from guest
1203		 *************************************************************/
1204
1205		kvm_arm_clear_debug(vcpu);
1206
1207		/*
1208		 * We must sync the PMU state before the vgic state so
1209		 * that the vgic can properly sample the updated state of the
1210		 * interrupt line.
1211		 */
1212		kvm_pmu_sync_hwstate(vcpu);
1213
1214		/*
1215		 * Sync the vgic state before syncing the timer state because
1216		 * the timer code needs to know if the virtual timer
1217		 * interrupts are active.
1218		 */
1219		kvm_vgic_sync_hwstate(vcpu);
1220
1221		/*
1222		 * Sync the timer hardware state before enabling interrupts as
1223		 * we don't want vtimer interrupts to race with syncing the
1224		 * timer virtual interrupt state.
1225		 */
1226		if (unlikely(!irqchip_in_kernel(vcpu->kvm)))
1227			kvm_timer_sync_user(vcpu);
1228
1229		kvm_arch_vcpu_ctxsync_fp(vcpu);
1230
1231		/*
1232		 * We must ensure that any pending interrupts are taken before
1233		 * we exit guest timing so that timer ticks are accounted as
1234		 * guest time. Transiently unmask interrupts so that any
1235		 * pending interrupts are taken.
1236		 *
1237		 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other
1238		 * context synchronization event) is necessary to ensure that
1239		 * pending interrupts are taken.
1240		 */
1241		if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) {
1242			local_irq_enable();
1243			isb();
1244			local_irq_disable();
1245		}
1246
1247		guest_timing_exit_irqoff();
1248
1249		local_irq_enable();
1250
1251		trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
1252
1253		/* Exit types that need handling before we can be preempted */
1254		handle_exit_early(vcpu, ret);
1255
1256		preempt_enable();
1257
1258		/*
1259		 * The ARMv8 architecture doesn't give the hypervisor
1260		 * a mechanism to prevent a guest from dropping to AArch32 EL0
1261		 * if implemented by the CPU. If we spot the guest in such
1262		 * state and that we decided it wasn't supposed to do so (like
1263		 * with the asymmetric AArch32 case), return to userspace with
1264		 * a fatal error.
1265		 */
1266		if (vcpu_mode_is_bad_32bit(vcpu)) {
1267			/*
1268			 * As we have caught the guest red-handed, decide that
1269			 * it isn't fit for purpose anymore by making the vcpu
1270			 * invalid. The VMM can try and fix it by issuing  a
1271			 * KVM_ARM_VCPU_INIT if it really wants to.
1272			 */
1273			vcpu_clear_flag(vcpu, VCPU_INITIALIZED);
1274			ret = ARM_EXCEPTION_IL;
1275		}
1276
1277		ret = handle_exit(vcpu, ret);
1278	}
1279
1280	/* Tell userspace about in-kernel device output levels */
1281	if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
1282		kvm_timer_update_run(vcpu);
1283		kvm_pmu_update_run(vcpu);
1284	}
1285
1286	kvm_sigset_deactivate(vcpu);
1287
1288out:
1289	/*
1290	 * In the unlikely event that we are returning to userspace
1291	 * with pending exceptions or PC adjustment, commit these
1292	 * adjustments in order to give userspace a consistent view of
1293	 * the vcpu state. Note that this relies on __kvm_adjust_pc()
1294	 * being preempt-safe on VHE.
1295	 */
1296	if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) ||
1297		     vcpu_get_flag(vcpu, INCREMENT_PC)))
1298		kvm_call_hyp(__kvm_adjust_pc, vcpu);
1299
1300	vcpu_put(vcpu);
1301	return ret;
1302}
1303
1304static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
1305{
1306	int bit_index;
1307	bool set;
1308	unsigned long *hcr;
1309
1310	if (number == KVM_ARM_IRQ_CPU_IRQ)
1311		bit_index = __ffs(HCR_VI);
1312	else /* KVM_ARM_IRQ_CPU_FIQ */
1313		bit_index = __ffs(HCR_VF);
1314
1315	hcr = vcpu_hcr(vcpu);
1316	if (level)
1317		set = test_and_set_bit(bit_index, hcr);
1318	else
1319		set = test_and_clear_bit(bit_index, hcr);
1320
1321	/*
1322	 * If we didn't change anything, no need to wake up or kick other CPUs
1323	 */
1324	if (set == level)
1325		return 0;
1326
1327	/*
1328	 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
1329	 * trigger a world-switch round on the running physical CPU to set the
1330	 * virtual IRQ/FIQ fields in the HCR appropriately.
1331	 */
1332	kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
1333	kvm_vcpu_kick(vcpu);
1334
1335	return 0;
1336}
1337
1338int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
1339			  bool line_status)
1340{
1341	u32 irq = irq_level->irq;
1342	unsigned int irq_type, vcpu_id, irq_num;
1343	struct kvm_vcpu *vcpu = NULL;
1344	bool level = irq_level->level;
1345
1346	irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
1347	vcpu_id = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
1348	vcpu_id += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
1349	irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
1350
1351	trace_kvm_irq_line(irq_type, vcpu_id, irq_num, irq_level->level);
1352
1353	switch (irq_type) {
1354	case KVM_ARM_IRQ_TYPE_CPU:
1355		if (irqchip_in_kernel(kvm))
1356			return -ENXIO;
1357
1358		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1359		if (!vcpu)
1360			return -EINVAL;
1361
1362		if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
1363			return -EINVAL;
1364
1365		return vcpu_interrupt_line(vcpu, irq_num, level);
1366	case KVM_ARM_IRQ_TYPE_PPI:
1367		if (!irqchip_in_kernel(kvm))
1368			return -ENXIO;
1369
1370		vcpu = kvm_get_vcpu_by_id(kvm, vcpu_id);
1371		if (!vcpu)
1372			return -EINVAL;
1373
1374		if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
1375			return -EINVAL;
1376
1377		return kvm_vgic_inject_irq(kvm, vcpu, irq_num, level, NULL);
1378	case KVM_ARM_IRQ_TYPE_SPI:
1379		if (!irqchip_in_kernel(kvm))
1380			return -ENXIO;
1381
1382		if (irq_num < VGIC_NR_PRIVATE_IRQS)
1383			return -EINVAL;
1384
1385		return kvm_vgic_inject_irq(kvm, NULL, irq_num, level, NULL);
1386	}
1387
1388	return -EINVAL;
1389}
1390
1391static unsigned long system_supported_vcpu_features(void)
1392{
1393	unsigned long features = KVM_VCPU_VALID_FEATURES;
1394
1395	if (!cpus_have_final_cap(ARM64_HAS_32BIT_EL1))
1396		clear_bit(KVM_ARM_VCPU_EL1_32BIT, &features);
1397
1398	if (!kvm_arm_support_pmu_v3())
1399		clear_bit(KVM_ARM_VCPU_PMU_V3, &features);
1400
1401	if (!system_supports_sve())
1402		clear_bit(KVM_ARM_VCPU_SVE, &features);
1403
1404	if (!kvm_has_full_ptr_auth()) {
1405		clear_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features);
1406		clear_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features);
1407	}
1408
1409	if (!cpus_have_final_cap(ARM64_HAS_NESTED_VIRT))
1410		clear_bit(KVM_ARM_VCPU_HAS_EL2, &features);
1411
1412	return features;
1413}
1414
1415static int kvm_vcpu_init_check_features(struct kvm_vcpu *vcpu,
1416					const struct kvm_vcpu_init *init)
1417{
1418	unsigned long features = init->features[0];
1419	int i;
1420
1421	if (features & ~KVM_VCPU_VALID_FEATURES)
1422		return -ENOENT;
1423
1424	for (i = 1; i < ARRAY_SIZE(init->features); i++) {
1425		if (init->features[i])
1426			return -ENOENT;
1427	}
1428
1429	if (features & ~system_supported_vcpu_features())
1430		return -EINVAL;
1431
1432	/*
1433	 * For now make sure that both address/generic pointer authentication
1434	 * features are requested by the userspace together.
1435	 */
1436	if (test_bit(KVM_ARM_VCPU_PTRAUTH_ADDRESS, &features) !=
1437	    test_bit(KVM_ARM_VCPU_PTRAUTH_GENERIC, &features))
1438		return -EINVAL;
1439
1440	if (!test_bit(KVM_ARM_VCPU_EL1_32BIT, &features))
1441		return 0;
1442
1443	/* MTE is incompatible with AArch32 */
1444	if (kvm_has_mte(vcpu->kvm))
1445		return -EINVAL;
1446
1447	/* NV is incompatible with AArch32 */
1448	if (test_bit(KVM_ARM_VCPU_HAS_EL2, &features))
1449		return -EINVAL;
1450
1451	return 0;
1452}
1453
1454static bool kvm_vcpu_init_changed(struct kvm_vcpu *vcpu,
1455				  const struct kvm_vcpu_init *init)
1456{
1457	unsigned long features = init->features[0];
1458
1459	return !bitmap_equal(vcpu->kvm->arch.vcpu_features, &features,
1460			     KVM_VCPU_MAX_FEATURES);
1461}
1462
1463static int kvm_setup_vcpu(struct kvm_vcpu *vcpu)
1464{
1465	struct kvm *kvm = vcpu->kvm;
1466	int ret = 0;
1467
1468	/*
1469	 * When the vCPU has a PMU, but no PMU is set for the guest
1470	 * yet, set the default one.
1471	 */
1472	if (kvm_vcpu_has_pmu(vcpu) && !kvm->arch.arm_pmu)
1473		ret = kvm_arm_set_default_pmu(kvm);
1474
1475	/* Prepare for nested if required */
1476	if (!ret && vcpu_has_nv(vcpu))
1477		ret = kvm_vcpu_init_nested(vcpu);
1478
1479	return ret;
1480}
1481
1482static int __kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1483				 const struct kvm_vcpu_init *init)
1484{
1485	unsigned long features = init->features[0];
1486	struct kvm *kvm = vcpu->kvm;
1487	int ret = -EINVAL;
1488
1489	mutex_lock(&kvm->arch.config_lock);
1490
1491	if (test_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags) &&
1492	    kvm_vcpu_init_changed(vcpu, init))
1493		goto out_unlock;
1494
1495	bitmap_copy(kvm->arch.vcpu_features, &features, KVM_VCPU_MAX_FEATURES);
1496
1497	ret = kvm_setup_vcpu(vcpu);
1498	if (ret)
1499		goto out_unlock;
1500
1501	/* Now we know what it is, we can reset it. */
1502	kvm_reset_vcpu(vcpu);
1503
1504	set_bit(KVM_ARCH_FLAG_VCPU_FEATURES_CONFIGURED, &kvm->arch.flags);
1505	vcpu_set_flag(vcpu, VCPU_INITIALIZED);
1506	ret = 0;
1507out_unlock:
1508	mutex_unlock(&kvm->arch.config_lock);
1509	return ret;
1510}
1511
1512static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
1513			       const struct kvm_vcpu_init *init)
1514{
1515	int ret;
1516
1517	if (init->target != KVM_ARM_TARGET_GENERIC_V8 &&
1518	    init->target != kvm_target_cpu())
1519		return -EINVAL;
1520
1521	ret = kvm_vcpu_init_check_features(vcpu, init);
1522	if (ret)
1523		return ret;
1524
1525	if (!kvm_vcpu_initialized(vcpu))
1526		return __kvm_vcpu_set_target(vcpu, init);
1527
1528	if (kvm_vcpu_init_changed(vcpu, init))
1529		return -EINVAL;
1530
1531	kvm_reset_vcpu(vcpu);
1532	return 0;
1533}
1534
1535static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1536					 struct kvm_vcpu_init *init)
1537{
1538	bool power_off = false;
1539	int ret;
1540
1541	/*
1542	 * Treat the power-off vCPU feature as ephemeral. Clear the bit to avoid
1543	 * reflecting it in the finalized feature set, thus limiting its scope
1544	 * to a single KVM_ARM_VCPU_INIT call.
1545	 */
1546	if (init->features[0] & BIT(KVM_ARM_VCPU_POWER_OFF)) {
1547		init->features[0] &= ~BIT(KVM_ARM_VCPU_POWER_OFF);
1548		power_off = true;
1549	}
1550
1551	ret = kvm_vcpu_set_target(vcpu, init);
1552	if (ret)
1553		return ret;
1554
1555	/*
1556	 * Ensure a rebooted VM will fault in RAM pages and detect if the
1557	 * guest MMU is turned off and flush the caches as needed.
1558	 *
1559	 * S2FWB enforces all memory accesses to RAM being cacheable,
1560	 * ensuring that the data side is always coherent. We still
1561	 * need to invalidate the I-cache though, as FWB does *not*
1562	 * imply CTR_EL0.DIC.
1563	 */
1564	if (vcpu_has_run_once(vcpu)) {
1565		if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB))
1566			stage2_unmap_vm(vcpu->kvm);
1567		else
1568			icache_inval_all_pou();
1569	}
1570
1571	vcpu_reset_hcr(vcpu);
1572	vcpu->arch.cptr_el2 = kvm_get_reset_cptr_el2(vcpu);
1573
1574	/*
1575	 * Handle the "start in power-off" case.
1576	 */
1577	spin_lock(&vcpu->arch.mp_state_lock);
1578
1579	if (power_off)
1580		__kvm_arm_vcpu_power_off(vcpu);
1581	else
1582		WRITE_ONCE(vcpu->arch.mp_state.mp_state, KVM_MP_STATE_RUNNABLE);
1583
1584	spin_unlock(&vcpu->arch.mp_state_lock);
1585
1586	return 0;
1587}
1588
1589static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1590				 struct kvm_device_attr *attr)
1591{
1592	int ret = -ENXIO;
1593
1594	switch (attr->group) {
1595	default:
1596		ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1597		break;
1598	}
1599
1600	return ret;
1601}
1602
1603static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1604				 struct kvm_device_attr *attr)
1605{
1606	int ret = -ENXIO;
1607
1608	switch (attr->group) {
1609	default:
1610		ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1611		break;
1612	}
1613
1614	return ret;
1615}
1616
1617static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1618				 struct kvm_device_attr *attr)
1619{
1620	int ret = -ENXIO;
1621
1622	switch (attr->group) {
1623	default:
1624		ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1625		break;
1626	}
1627
1628	return ret;
1629}
1630
1631static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1632				   struct kvm_vcpu_events *events)
1633{
1634	memset(events, 0, sizeof(*events));
1635
1636	return __kvm_arm_vcpu_get_events(vcpu, events);
1637}
1638
1639static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1640				   struct kvm_vcpu_events *events)
1641{
1642	int i;
1643
1644	/* check whether the reserved field is zero */
1645	for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1646		if (events->reserved[i])
1647			return -EINVAL;
1648
1649	/* check whether the pad field is zero */
1650	for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1651		if (events->exception.pad[i])
1652			return -EINVAL;
1653
1654	return __kvm_arm_vcpu_set_events(vcpu, events);
1655}
1656
1657long kvm_arch_vcpu_ioctl(struct file *filp,
1658			 unsigned int ioctl, unsigned long arg)
1659{
1660	struct kvm_vcpu *vcpu = filp->private_data;
1661	void __user *argp = (void __user *)arg;
1662	struct kvm_device_attr attr;
1663	long r;
1664
1665	switch (ioctl) {
1666	case KVM_ARM_VCPU_INIT: {
1667		struct kvm_vcpu_init init;
1668
1669		r = -EFAULT;
1670		if (copy_from_user(&init, argp, sizeof(init)))
1671			break;
1672
1673		r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1674		break;
1675	}
1676	case KVM_SET_ONE_REG:
1677	case KVM_GET_ONE_REG: {
1678		struct kvm_one_reg reg;
1679
1680		r = -ENOEXEC;
1681		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1682			break;
1683
1684		r = -EFAULT;
1685		if (copy_from_user(&reg, argp, sizeof(reg)))
1686			break;
1687
1688		/*
1689		 * We could owe a reset due to PSCI. Handle the pending reset
1690		 * here to ensure userspace register accesses are ordered after
1691		 * the reset.
1692		 */
1693		if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
1694			kvm_reset_vcpu(vcpu);
1695
1696		if (ioctl == KVM_SET_ONE_REG)
1697			r = kvm_arm_set_reg(vcpu, &reg);
1698		else
1699			r = kvm_arm_get_reg(vcpu, &reg);
1700		break;
1701	}
1702	case KVM_GET_REG_LIST: {
1703		struct kvm_reg_list __user *user_list = argp;
1704		struct kvm_reg_list reg_list;
1705		unsigned n;
1706
1707		r = -ENOEXEC;
1708		if (unlikely(!kvm_vcpu_initialized(vcpu)))
1709			break;
1710
1711		r = -EPERM;
1712		if (!kvm_arm_vcpu_is_finalized(vcpu))
1713			break;
1714
1715		r = -EFAULT;
1716		if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1717			break;
1718		n = reg_list.n;
1719		reg_list.n = kvm_arm_num_regs(vcpu);
1720		if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1721			break;
1722		r = -E2BIG;
1723		if (n < reg_list.n)
1724			break;
1725		r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1726		break;
1727	}
1728	case KVM_SET_DEVICE_ATTR: {
1729		r = -EFAULT;
1730		if (copy_from_user(&attr, argp, sizeof(attr)))
1731			break;
1732		r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1733		break;
1734	}
1735	case KVM_GET_DEVICE_ATTR: {
1736		r = -EFAULT;
1737		if (copy_from_user(&attr, argp, sizeof(attr)))
1738			break;
1739		r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1740		break;
1741	}
1742	case KVM_HAS_DEVICE_ATTR: {
1743		r = -EFAULT;
1744		if (copy_from_user(&attr, argp, sizeof(attr)))
1745			break;
1746		r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1747		break;
1748	}
1749	case KVM_GET_VCPU_EVENTS: {
1750		struct kvm_vcpu_events events;
1751
1752		if (kvm_arm_vcpu_get_events(vcpu, &events))
1753			return -EINVAL;
1754
1755		if (copy_to_user(argp, &events, sizeof(events)))
1756			return -EFAULT;
1757
1758		return 0;
1759	}
1760	case KVM_SET_VCPU_EVENTS: {
1761		struct kvm_vcpu_events events;
1762
1763		if (copy_from_user(&events, argp, sizeof(events)))
1764			return -EFAULT;
1765
1766		return kvm_arm_vcpu_set_events(vcpu, &events);
1767	}
1768	case KVM_ARM_VCPU_FINALIZE: {
1769		int what;
1770
1771		if (!kvm_vcpu_initialized(vcpu))
1772			return -ENOEXEC;
1773
1774		if (get_user(what, (const int __user *)argp))
1775			return -EFAULT;
1776
1777		return kvm_arm_vcpu_finalize(vcpu, what);
1778	}
1779	default:
1780		r = -EINVAL;
1781	}
1782
1783	return r;
1784}
1785
1786void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1787{
1788
1789}
1790
1791static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1792					struct kvm_arm_device_addr *dev_addr)
1793{
1794	switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) {
1795	case KVM_ARM_DEVICE_VGIC_V2:
1796		if (!vgic_present)
1797			return -ENXIO;
1798		return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr);
1799	default:
1800		return -ENODEV;
1801	}
1802}
1803
1804static int kvm_vm_has_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1805{
1806	switch (attr->group) {
1807	case KVM_ARM_VM_SMCCC_CTRL:
1808		return kvm_vm_smccc_has_attr(kvm, attr);
1809	default:
1810		return -ENXIO;
1811	}
1812}
1813
1814static int kvm_vm_set_attr(struct kvm *kvm, struct kvm_device_attr *attr)
1815{
1816	switch (attr->group) {
1817	case KVM_ARM_VM_SMCCC_CTRL:
1818		return kvm_vm_smccc_set_attr(kvm, attr);
1819	default:
1820		return -ENXIO;
1821	}
1822}
1823
1824int kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
1825{
1826	struct kvm *kvm = filp->private_data;
1827	void __user *argp = (void __user *)arg;
1828	struct kvm_device_attr attr;
1829
1830	switch (ioctl) {
1831	case KVM_CREATE_IRQCHIP: {
1832		int ret;
1833		if (!vgic_present)
1834			return -ENXIO;
1835		mutex_lock(&kvm->lock);
1836		ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1837		mutex_unlock(&kvm->lock);
1838		return ret;
1839	}
1840	case KVM_ARM_SET_DEVICE_ADDR: {
1841		struct kvm_arm_device_addr dev_addr;
1842
1843		if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1844			return -EFAULT;
1845		return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1846	}
1847	case KVM_ARM_PREFERRED_TARGET: {
1848		struct kvm_vcpu_init init = {
1849			.target = KVM_ARM_TARGET_GENERIC_V8,
1850		};
1851
1852		if (copy_to_user(argp, &init, sizeof(init)))
1853			return -EFAULT;
1854
1855		return 0;
1856	}
1857	case KVM_ARM_MTE_COPY_TAGS: {
1858		struct kvm_arm_copy_mte_tags copy_tags;
1859
1860		if (copy_from_user(&copy_tags, argp, sizeof(copy_tags)))
1861			return -EFAULT;
1862		return kvm_vm_ioctl_mte_copy_tags(kvm, &copy_tags);
1863	}
1864	case KVM_ARM_SET_COUNTER_OFFSET: {
1865		struct kvm_arm_counter_offset offset;
1866
1867		if (copy_from_user(&offset, argp, sizeof(offset)))
1868			return -EFAULT;
1869		return kvm_vm_ioctl_set_counter_offset(kvm, &offset);
1870	}
1871	case KVM_HAS_DEVICE_ATTR: {
1872		if (copy_from_user(&attr, argp, sizeof(attr)))
1873			return -EFAULT;
1874
1875		return kvm_vm_has_attr(kvm, &attr);
1876	}
1877	case KVM_SET_DEVICE_ATTR: {
1878		if (copy_from_user(&attr, argp, sizeof(attr)))
1879			return -EFAULT;
1880
1881		return kvm_vm_set_attr(kvm, &attr);
1882	}
1883	case KVM_ARM_GET_REG_WRITABLE_MASKS: {
1884		struct reg_mask_range range;
1885
1886		if (copy_from_user(&range, argp, sizeof(range)))
1887			return -EFAULT;
1888		return kvm_vm_ioctl_get_reg_writable_masks(kvm, &range);
1889	}
1890	default:
1891		return -EINVAL;
1892	}
1893}
1894
1895/* unlocks vcpus from @vcpu_lock_idx and smaller */
1896static void unlock_vcpus(struct kvm *kvm, int vcpu_lock_idx)
1897{
1898	struct kvm_vcpu *tmp_vcpu;
1899
1900	for (; vcpu_lock_idx >= 0; vcpu_lock_idx--) {
1901		tmp_vcpu = kvm_get_vcpu(kvm, vcpu_lock_idx);
1902		mutex_unlock(&tmp_vcpu->mutex);
1903	}
1904}
1905
1906void unlock_all_vcpus(struct kvm *kvm)
1907{
1908	lockdep_assert_held(&kvm->lock);
1909
1910	unlock_vcpus(kvm, atomic_read(&kvm->online_vcpus) - 1);
1911}
1912
1913/* Returns true if all vcpus were locked, false otherwise */
1914bool lock_all_vcpus(struct kvm *kvm)
1915{
1916	struct kvm_vcpu *tmp_vcpu;
1917	unsigned long c;
1918
1919	lockdep_assert_held(&kvm->lock);
1920
1921	/*
1922	 * Any time a vcpu is in an ioctl (including running), the
1923	 * core KVM code tries to grab the vcpu->mutex.
1924	 *
1925	 * By grabbing the vcpu->mutex of all VCPUs we ensure that no
1926	 * other VCPUs can fiddle with the state while we access it.
1927	 */
1928	kvm_for_each_vcpu(c, tmp_vcpu, kvm) {
1929		if (!mutex_trylock(&tmp_vcpu->mutex)) {
1930			unlock_vcpus(kvm, c - 1);
1931			return false;
1932		}
1933	}
1934
1935	return true;
1936}
1937
1938static unsigned long nvhe_percpu_size(void)
1939{
1940	return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) -
1941		(unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start);
1942}
1943
1944static unsigned long nvhe_percpu_order(void)
1945{
1946	unsigned long size = nvhe_percpu_size();
1947
1948	return size ? get_order(size) : 0;
1949}
1950
1951static size_t pkvm_host_sve_state_order(void)
1952{
1953	return get_order(pkvm_host_sve_state_size());
1954}
1955
1956/* A lookup table holding the hypervisor VA for each vector slot */
1957static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS];
1958
1959static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot)
1960{
1961	hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot);
1962}
1963
1964static int kvm_init_vector_slots(void)
1965{
1966	int err;
1967	void *base;
1968
1969	base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector));
1970	kvm_init_vector_slot(base, HYP_VECTOR_DIRECT);
1971
1972	base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs));
1973	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT);
1974
1975	if (kvm_system_needs_idmapped_vectors() &&
1976	    !is_protected_kvm_enabled()) {
1977		err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs),
1978					       __BP_HARDEN_HYP_VECS_SZ, &base);
1979		if (err)
1980			return err;
1981	}
1982
1983	kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT);
1984	kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT);
1985	return 0;
1986}
1987
1988static void __init cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits)
1989{
1990	struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
1991	unsigned long tcr, ips;
1992
1993	/*
1994	 * Calculate the raw per-cpu offset without a translation from the
1995	 * kernel's mapping to the linear mapping, and store it in tpidr_el2
1996	 * so that we can use adr_l to access per-cpu variables in EL2.
1997	 * Also drop the KASAN tag which gets in the way...
1998	 */
1999	params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) -
2000			    (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start));
2001
2002	params->mair_el2 = read_sysreg(mair_el1);
2003
2004	tcr = read_sysreg(tcr_el1);
2005	ips = FIELD_GET(TCR_IPS_MASK, tcr);
2006	if (cpus_have_final_cap(ARM64_KVM_HVHE)) {
2007		tcr |= TCR_EPD1_MASK;
2008	} else {
2009		tcr &= TCR_EL2_MASK;
2010		tcr |= TCR_EL2_RES1;
2011	}
2012	tcr &= ~TCR_T0SZ_MASK;
2013	tcr |= TCR_T0SZ(hyp_va_bits);
2014	tcr &= ~TCR_EL2_PS_MASK;
2015	tcr |= FIELD_PREP(TCR_EL2_PS_MASK, ips);
2016	if (lpa2_is_enabled())
2017		tcr |= TCR_EL2_DS;
2018	params->tcr_el2 = tcr;
2019
2020	params->pgd_pa = kvm_mmu_get_httbr();
2021	if (is_protected_kvm_enabled())
2022		params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS;
2023	else
2024		params->hcr_el2 = HCR_HOST_NVHE_FLAGS;
2025	if (cpus_have_final_cap(ARM64_KVM_HVHE))
2026		params->hcr_el2 |= HCR_E2H;
2027	params->vttbr = params->vtcr = 0;
2028
2029	/*
2030	 * Flush the init params from the data cache because the struct will
2031	 * be read while the MMU is off.
2032	 */
2033	kvm_flush_dcache_to_poc(params, sizeof(*params));
2034}
2035
2036static void hyp_install_host_vector(void)
2037{
2038	struct kvm_nvhe_init_params *params;
2039	struct arm_smccc_res res;
2040
2041	/* Switch from the HYP stub to our own HYP init vector */
2042	__hyp_set_vectors(kvm_get_idmap_vector());
2043
2044	/*
2045	 * Call initialization code, and switch to the full blown HYP code.
2046	 * If the cpucaps haven't been finalized yet, something has gone very
2047	 * wrong, and hyp will crash and burn when it uses any
2048	 * cpus_have_*_cap() wrapper.
2049	 */
2050	BUG_ON(!system_capabilities_finalized());
2051	params = this_cpu_ptr_nvhe_sym(kvm_init_params);
2052	arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res);
2053	WARN_ON(res.a0 != SMCCC_RET_SUCCESS);
2054}
2055
2056static void cpu_init_hyp_mode(void)
2057{
2058	hyp_install_host_vector();
2059
2060	/*
2061	 * Disabling SSBD on a non-VHE system requires us to enable SSBS
2062	 * at EL2.
2063	 */
2064	if (this_cpu_has_cap(ARM64_SSBS) &&
2065	    arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) {
2066		kvm_call_hyp_nvhe(__kvm_enable_ssbs);
2067	}
2068}
2069
2070static void cpu_hyp_reset(void)
2071{
2072	if (!is_kernel_in_hyp_mode())
2073		__hyp_reset_vectors();
2074}
2075
2076/*
2077 * EL2 vectors can be mapped and rerouted in a number of ways,
2078 * depending on the kernel configuration and CPU present:
2079 *
2080 * - If the CPU is affected by Spectre-v2, the hardening sequence is
2081 *   placed in one of the vector slots, which is executed before jumping
2082 *   to the real vectors.
2083 *
2084 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot
2085 *   containing the hardening sequence is mapped next to the idmap page,
2086 *   and executed before jumping to the real vectors.
2087 *
2088 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an
2089 *   empty slot is selected, mapped next to the idmap page, and
2090 *   executed before jumping to the real vectors.
2091 *
2092 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with
2093 * VHE, as we don't have hypervisor-specific mappings. If the system
2094 * is VHE and yet selects this capability, it will be ignored.
2095 */
2096static void cpu_set_hyp_vector(void)
2097{
2098	struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data);
2099	void *vector = hyp_spectre_vector_selector[data->slot];
2100
2101	if (!is_protected_kvm_enabled())
2102		*this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector;
2103	else
2104		kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot);
2105}
2106
2107static void cpu_hyp_init_context(void)
2108{
2109	kvm_init_host_cpu_context(host_data_ptr(host_ctxt));
2110
2111	if (!is_kernel_in_hyp_mode())
2112		cpu_init_hyp_mode();
2113}
2114
2115static void cpu_hyp_init_features(void)
2116{
2117	cpu_set_hyp_vector();
2118	kvm_arm_init_debug();
2119
2120	if (is_kernel_in_hyp_mode())
2121		kvm_timer_init_vhe();
2122
2123	if (vgic_present)
2124		kvm_vgic_init_cpu_hardware();
2125}
2126
2127static void cpu_hyp_reinit(void)
2128{
2129	cpu_hyp_reset();
2130	cpu_hyp_init_context();
2131	cpu_hyp_init_features();
2132}
2133
2134static void cpu_hyp_init(void *discard)
2135{
2136	if (!__this_cpu_read(kvm_hyp_initialized)) {
2137		cpu_hyp_reinit();
2138		__this_cpu_write(kvm_hyp_initialized, 1);
2139	}
2140}
2141
2142static void cpu_hyp_uninit(void *discard)
2143{
2144	if (__this_cpu_read(kvm_hyp_initialized)) {
2145		cpu_hyp_reset();
2146		__this_cpu_write(kvm_hyp_initialized, 0);
2147	}
2148}
2149
2150int kvm_arch_enable_virtualization_cpu(void)
2151{
2152	/*
2153	 * Most calls to this function are made with migration
2154	 * disabled, but not with preemption disabled. The former is
2155	 * enough to ensure correctness, but most of the helpers
2156	 * expect the later and will throw a tantrum otherwise.
2157	 */
2158	preempt_disable();
2159
2160	cpu_hyp_init(NULL);
2161
2162	kvm_vgic_cpu_up();
2163	kvm_timer_cpu_up();
2164
2165	preempt_enable();
2166
2167	return 0;
2168}
2169
2170void kvm_arch_disable_virtualization_cpu(void)
2171{
2172	kvm_timer_cpu_down();
2173	kvm_vgic_cpu_down();
2174
2175	if (!is_protected_kvm_enabled())
2176		cpu_hyp_uninit(NULL);
2177}
2178
2179#ifdef CONFIG_CPU_PM
2180static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
2181				    unsigned long cmd,
2182				    void *v)
2183{
2184	/*
2185	 * kvm_hyp_initialized is left with its old value over
2186	 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
2187	 * re-enable hyp.
2188	 */
2189	switch (cmd) {
2190	case CPU_PM_ENTER:
2191		if (__this_cpu_read(kvm_hyp_initialized))
2192			/*
2193			 * don't update kvm_hyp_initialized here
2194			 * so that the hyp will be re-enabled
2195			 * when we resume. See below.
2196			 */
2197			cpu_hyp_reset();
2198
2199		return NOTIFY_OK;
2200	case CPU_PM_ENTER_FAILED:
2201	case CPU_PM_EXIT:
2202		if (__this_cpu_read(kvm_hyp_initialized))
2203			/* The hyp was enabled before suspend. */
2204			cpu_hyp_reinit();
2205
2206		return NOTIFY_OK;
2207
2208	default:
2209		return NOTIFY_DONE;
2210	}
2211}
2212
2213static struct notifier_block hyp_init_cpu_pm_nb = {
2214	.notifier_call = hyp_init_cpu_pm_notifier,
2215};
2216
2217static void __init hyp_cpu_pm_init(void)
2218{
2219	if (!is_protected_kvm_enabled())
2220		cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
2221}
2222static void __init hyp_cpu_pm_exit(void)
2223{
2224	if (!is_protected_kvm_enabled())
2225		cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
2226}
2227#else
2228static inline void __init hyp_cpu_pm_init(void)
2229{
2230}
2231static inline void __init hyp_cpu_pm_exit(void)
2232{
2233}
2234#endif
2235
2236static void __init init_cpu_logical_map(void)
2237{
2238	unsigned int cpu;
2239
2240	/*
2241	 * Copy the MPIDR <-> logical CPU ID mapping to hyp.
2242	 * Only copy the set of online CPUs whose features have been checked
2243	 * against the finalized system capabilities. The hypervisor will not
2244	 * allow any other CPUs from the `possible` set to boot.
2245	 */
2246	for_each_online_cpu(cpu)
2247		hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu);
2248}
2249
2250#define init_psci_0_1_impl_state(config, what)	\
2251	config.psci_0_1_ ## what ## _implemented = psci_ops.what
2252
2253static bool __init init_psci_relay(void)
2254{
2255	/*
2256	 * If PSCI has not been initialized, protected KVM cannot install
2257	 * itself on newly booted CPUs.
2258	 */
2259	if (!psci_ops.get_version) {
2260		kvm_err("Cannot initialize protected mode without PSCI\n");
2261		return false;
2262	}
2263
2264	kvm_host_psci_config.version = psci_ops.get_version();
2265	kvm_host_psci_config.smccc_version = arm_smccc_get_version();
2266
2267	if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) {
2268		kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids();
2269		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend);
2270		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on);
2271		init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off);
2272		init_psci_0_1_impl_state(kvm_host_psci_config, migrate);
2273	}
2274	return true;
2275}
2276
2277static int __init init_subsystems(void)
2278{
2279	int err = 0;
2280
2281	/*
2282	 * Enable hardware so that subsystem initialisation can access EL2.
2283	 */
2284	on_each_cpu(cpu_hyp_init, NULL, 1);
2285
2286	/*
2287	 * Register CPU lower-power notifier
2288	 */
2289	hyp_cpu_pm_init();
2290
2291	/*
2292	 * Init HYP view of VGIC
2293	 */
2294	err = kvm_vgic_hyp_init();
2295	switch (err) {
2296	case 0:
2297		vgic_present = true;
2298		break;
2299	case -ENODEV:
2300	case -ENXIO:
2301		vgic_present = false;
2302		err = 0;
2303		break;
2304	default:
2305		goto out;
2306	}
2307
2308	/*
2309	 * Init HYP architected timer support
2310	 */
2311	err = kvm_timer_hyp_init(vgic_present);
2312	if (err)
2313		goto out;
2314
2315	kvm_register_perf_callbacks(NULL);
2316
2317out:
2318	if (err)
2319		hyp_cpu_pm_exit();
2320
2321	if (err || !is_protected_kvm_enabled())
2322		on_each_cpu(cpu_hyp_uninit, NULL, 1);
2323
2324	return err;
2325}
2326
2327static void __init teardown_subsystems(void)
2328{
2329	kvm_unregister_perf_callbacks();
2330	hyp_cpu_pm_exit();
2331}
2332
2333static void __init teardown_hyp_mode(void)
2334{
2335	bool free_sve = system_supports_sve() && is_protected_kvm_enabled();
2336	int cpu;
2337
2338	free_hyp_pgds();
2339	for_each_possible_cpu(cpu) {
2340		free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
2341		free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order());
2342
2343		if (free_sve) {
2344			struct cpu_sve_state *sve_state;
2345
2346			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2347			free_pages((unsigned long) sve_state, pkvm_host_sve_state_order());
2348		}
2349	}
2350}
2351
2352static int __init do_pkvm_init(u32 hyp_va_bits)
2353{
2354	void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base));
2355	int ret;
2356
2357	preempt_disable();
2358	cpu_hyp_init_context();
2359	ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size,
2360				num_possible_cpus(), kern_hyp_va(per_cpu_base),
2361				hyp_va_bits);
2362	cpu_hyp_init_features();
2363
2364	/*
2365	 * The stub hypercalls are now disabled, so set our local flag to
2366	 * prevent a later re-init attempt in kvm_arch_enable_virtualization_cpu().
2367	 */
2368	__this_cpu_write(kvm_hyp_initialized, 1);
2369	preempt_enable();
2370
2371	return ret;
2372}
2373
2374static u64 get_hyp_id_aa64pfr0_el1(void)
2375{
2376	/*
2377	 * Track whether the system isn't affected by spectre/meltdown in the
2378	 * hypervisor's view of id_aa64pfr0_el1, used for protected VMs.
2379	 * Although this is per-CPU, we make it global for simplicity, e.g., not
2380	 * to have to worry about vcpu migration.
2381	 *
2382	 * Unlike for non-protected VMs, userspace cannot override this for
2383	 * protected VMs.
2384	 */
2385	u64 val = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1);
2386
2387	val &= ~(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2) |
2388		 ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3));
2389
2390	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV2),
2391			  arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED);
2392	val |= FIELD_PREP(ARM64_FEATURE_MASK(ID_AA64PFR0_EL1_CSV3),
2393			  arm64_get_meltdown_state() == SPECTRE_UNAFFECTED);
2394
2395	return val;
2396}
2397
2398static void kvm_hyp_init_symbols(void)
2399{
2400	kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = get_hyp_id_aa64pfr0_el1();
2401	kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1);
2402	kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1);
2403	kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1);
2404	kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1);
2405	kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1);
2406	kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1);
2407	kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1);
2408	kvm_nvhe_sym(id_aa64smfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64SMFR0_EL1);
2409	kvm_nvhe_sym(__icache_flags) = __icache_flags;
2410	kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits;
2411}
2412
2413static int __init kvm_hyp_init_protection(u32 hyp_va_bits)
2414{
2415	void *addr = phys_to_virt(hyp_mem_base);
2416	int ret;
2417
2418	ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP);
2419	if (ret)
2420		return ret;
2421
2422	ret = do_pkvm_init(hyp_va_bits);
2423	if (ret)
2424		return ret;
2425
2426	free_hyp_pgds();
2427
2428	return 0;
2429}
2430
2431static int init_pkvm_host_sve_state(void)
2432{
2433	int cpu;
2434
2435	if (!system_supports_sve())
2436		return 0;
2437
2438	/* Allocate pages for host sve state in protected mode. */
2439	for_each_possible_cpu(cpu) {
2440		struct page *page = alloc_pages(GFP_KERNEL, pkvm_host_sve_state_order());
2441
2442		if (!page)
2443			return -ENOMEM;
2444
2445		per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state = page_address(page);
2446	}
2447
2448	/*
2449	 * Don't map the pages in hyp since these are only used in protected
2450	 * mode, which will (re)create its own mapping when initialized.
2451	 */
2452
2453	return 0;
2454}
2455
2456/*
2457 * Finalizes the initialization of hyp mode, once everything else is initialized
2458 * and the initialziation process cannot fail.
2459 */
2460static void finalize_init_hyp_mode(void)
2461{
2462	int cpu;
2463
2464	if (system_supports_sve() && is_protected_kvm_enabled()) {
2465		for_each_possible_cpu(cpu) {
2466			struct cpu_sve_state *sve_state;
2467
2468			sve_state = per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state;
2469			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->sve_state =
2470				kern_hyp_va(sve_state);
2471		}
2472	} else {
2473		for_each_possible_cpu(cpu) {
2474			struct user_fpsimd_state *fpsimd_state;
2475
2476			fpsimd_state = &per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->host_ctxt.fp_regs;
2477			per_cpu_ptr_nvhe_sym(kvm_host_data, cpu)->fpsimd_state =
2478				kern_hyp_va(fpsimd_state);
2479		}
2480	}
2481}
2482
2483static void pkvm_hyp_init_ptrauth(void)
2484{
2485	struct kvm_cpu_context *hyp_ctxt;
2486	int cpu;
2487
2488	for_each_possible_cpu(cpu) {
2489		hyp_ctxt = per_cpu_ptr_nvhe_sym(kvm_hyp_ctxt, cpu);
2490		hyp_ctxt->sys_regs[APIAKEYLO_EL1] = get_random_long();
2491		hyp_ctxt->sys_regs[APIAKEYHI_EL1] = get_random_long();
2492		hyp_ctxt->sys_regs[APIBKEYLO_EL1] = get_random_long();
2493		hyp_ctxt->sys_regs[APIBKEYHI_EL1] = get_random_long();
2494		hyp_ctxt->sys_regs[APDAKEYLO_EL1] = get_random_long();
2495		hyp_ctxt->sys_regs[APDAKEYHI_EL1] = get_random_long();
2496		hyp_ctxt->sys_regs[APDBKEYLO_EL1] = get_random_long();
2497		hyp_ctxt->sys_regs[APDBKEYHI_EL1] = get_random_long();
2498		hyp_ctxt->sys_regs[APGAKEYLO_EL1] = get_random_long();
2499		hyp_ctxt->sys_regs[APGAKEYHI_EL1] = get_random_long();
2500	}
2501}
2502
2503/* Inits Hyp-mode on all online CPUs */
2504static int __init init_hyp_mode(void)
2505{
2506	u32 hyp_va_bits;
2507	int cpu;
2508	int err = -ENOMEM;
2509
2510	/*
2511	 * The protected Hyp-mode cannot be initialized if the memory pool
2512	 * allocation has failed.
2513	 */
2514	if (is_protected_kvm_enabled() && !hyp_mem_base)
2515		goto out_err;
2516
2517	/*
2518	 * Allocate Hyp PGD and setup Hyp identity mapping
2519	 */
2520	err = kvm_mmu_init(&hyp_va_bits);
2521	if (err)
2522		goto out_err;
2523
2524	/*
2525	 * Allocate stack pages for Hypervisor-mode
2526	 */
2527	for_each_possible_cpu(cpu) {
2528		unsigned long stack_page;
2529
2530		stack_page = __get_free_page(GFP_KERNEL);
2531		if (!stack_page) {
2532			err = -ENOMEM;
2533			goto out_err;
2534		}
2535
2536		per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
2537	}
2538
2539	/*
2540	 * Allocate and initialize pages for Hypervisor-mode percpu regions.
2541	 */
2542	for_each_possible_cpu(cpu) {
2543		struct page *page;
2544		void *page_addr;
2545
2546		page = alloc_pages(GFP_KERNEL, nvhe_percpu_order());
2547		if (!page) {
2548			err = -ENOMEM;
2549			goto out_err;
2550		}
2551
2552		page_addr = page_address(page);
2553		memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size());
2554		kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr;
2555	}
2556
2557	/*
2558	 * Map the Hyp-code called directly from the host
2559	 */
2560	err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
2561				  kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
2562	if (err) {
2563		kvm_err("Cannot map world-switch code\n");
2564		goto out_err;
2565	}
2566
2567	err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start),
2568				  kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO);
2569	if (err) {
2570		kvm_err("Cannot map .hyp.rodata section\n");
2571		goto out_err;
2572	}
2573
2574	err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
2575				  kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
2576	if (err) {
2577		kvm_err("Cannot map rodata section\n");
2578		goto out_err;
2579	}
2580
2581	/*
2582	 * .hyp.bss is guaranteed to be placed at the beginning of the .bss
2583	 * section thanks to an assertion in the linker script. Map it RW and
2584	 * the rest of .bss RO.
2585	 */
2586	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start),
2587				  kvm_ksym_ref(__hyp_bss_end), PAGE_HYP);
2588	if (err) {
2589		kvm_err("Cannot map hyp bss section: %d\n", err);
2590		goto out_err;
2591	}
2592
2593	err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end),
2594				  kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
2595	if (err) {
2596		kvm_err("Cannot map bss section\n");
2597		goto out_err;
2598	}
2599
2600	/*
2601	 * Map the Hyp stack pages
2602	 */
2603	for_each_possible_cpu(cpu) {
2604		struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu);
2605		char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
2606
2607		err = create_hyp_stack(__pa(stack_page), &params->stack_hyp_va);
2608		if (err) {
2609			kvm_err("Cannot map hyp stack\n");
2610			goto out_err;
2611		}
2612
2613		/*
2614		 * Save the stack PA in nvhe_init_params. This will be needed
2615		 * to recreate the stack mapping in protected nVHE mode.
2616		 * __hyp_pa() won't do the right thing there, since the stack
2617		 * has been mapped in the flexible private VA space.
2618		 */
2619		params->stack_pa = __pa(stack_page);
2620	}
2621
2622	for_each_possible_cpu(cpu) {
2623		char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu];
2624		char *percpu_end = percpu_begin + nvhe_percpu_size();
2625
2626		/* Map Hyp percpu pages */
2627		err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP);
2628		if (err) {
2629			kvm_err("Cannot map hyp percpu region\n");
2630			goto out_err;
2631		}
2632
2633		/* Prepare the CPU initialization parameters */
2634		cpu_prepare_hyp_mode(cpu, hyp_va_bits);
2635	}
2636
2637	kvm_hyp_init_symbols();
2638
2639	if (is_protected_kvm_enabled()) {
2640		if (IS_ENABLED(CONFIG_ARM64_PTR_AUTH_KERNEL) &&
2641		    cpus_have_final_cap(ARM64_HAS_ADDRESS_AUTH))
2642			pkvm_hyp_init_ptrauth();
2643
2644		init_cpu_logical_map();
2645
2646		if (!init_psci_relay()) {
2647			err = -ENODEV;
2648			goto out_err;
2649		}
2650
2651		err = init_pkvm_host_sve_state();
2652		if (err)
2653			goto out_err;
2654
2655		err = kvm_hyp_init_protection(hyp_va_bits);
2656		if (err) {
2657			kvm_err("Failed to init hyp memory protection\n");
2658			goto out_err;
2659		}
2660	}
2661
2662	return 0;
2663
2664out_err:
2665	teardown_hyp_mode();
2666	kvm_err("error initializing Hyp mode: %d\n", err);
2667	return err;
2668}
2669
2670struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
2671{
2672	struct kvm_vcpu *vcpu = NULL;
2673	struct kvm_mpidr_data *data;
2674	unsigned long i;
2675
2676	mpidr &= MPIDR_HWID_BITMASK;
2677
2678	rcu_read_lock();
2679	data = rcu_dereference(kvm->arch.mpidr_data);
2680
2681	if (data) {
2682		u16 idx = kvm_mpidr_index(data, mpidr);
2683
2684		vcpu = kvm_get_vcpu(kvm, data->cmpidr_to_idx[idx]);
2685		if (mpidr != kvm_vcpu_get_mpidr_aff(vcpu))
2686			vcpu = NULL;
2687	}
2688
2689	rcu_read_unlock();
2690
2691	if (vcpu)
2692		return vcpu;
2693
2694	kvm_for_each_vcpu(i, vcpu, kvm) {
2695		if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
2696			return vcpu;
2697	}
2698	return NULL;
2699}
2700
2701bool kvm_arch_irqchip_in_kernel(struct kvm *kvm)
2702{
2703	return irqchip_in_kernel(kvm);
2704}
2705
2706bool kvm_arch_has_irq_bypass(void)
2707{
2708	return true;
2709}
2710
2711int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
2712				      struct irq_bypass_producer *prod)
2713{
2714	struct kvm_kernel_irqfd *irqfd =
2715		container_of(cons, struct kvm_kernel_irqfd, consumer);
2716
2717	return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
2718					  &irqfd->irq_entry);
2719}
2720void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
2721				      struct irq_bypass_producer *prod)
2722{
2723	struct kvm_kernel_irqfd *irqfd =
2724		container_of(cons, struct kvm_kernel_irqfd, consumer);
2725
2726	kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
2727				     &irqfd->irq_entry);
2728}
2729
2730void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
2731{
2732	struct kvm_kernel_irqfd *irqfd =
2733		container_of(cons, struct kvm_kernel_irqfd, consumer);
2734
2735	kvm_arm_halt_guest(irqfd->kvm);
2736}
2737
2738void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
2739{
2740	struct kvm_kernel_irqfd *irqfd =
2741		container_of(cons, struct kvm_kernel_irqfd, consumer);
2742
2743	kvm_arm_resume_guest(irqfd->kvm);
2744}
2745
2746/* Initialize Hyp-mode and memory mappings on all CPUs */
2747static __init int kvm_arm_init(void)
2748{
2749	int err;
2750	bool in_hyp_mode;
2751
2752	if (!is_hyp_mode_available()) {
2753		kvm_info("HYP mode not available\n");
2754		return -ENODEV;
2755	}
2756
2757	if (kvm_get_mode() == KVM_MODE_NONE) {
2758		kvm_info("KVM disabled from command line\n");
2759		return -ENODEV;
2760	}
2761
2762	err = kvm_sys_reg_table_init();
2763	if (err) {
2764		kvm_info("Error initializing system register tables");
2765		return err;
2766	}
2767
2768	in_hyp_mode = is_kernel_in_hyp_mode();
2769
2770	if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) ||
2771	    cpus_have_final_cap(ARM64_WORKAROUND_1508412))
2772		kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \
2773			 "Only trusted guests should be used on this system.\n");
2774
2775	err = kvm_set_ipa_limit();
2776	if (err)
2777		return err;
2778
2779	err = kvm_arm_init_sve();
2780	if (err)
2781		return err;
2782
2783	err = kvm_arm_vmid_alloc_init();
2784	if (err) {
2785		kvm_err("Failed to initialize VMID allocator.\n");
2786		return err;
2787	}
2788
2789	if (!in_hyp_mode) {
2790		err = init_hyp_mode();
2791		if (err)
2792			goto out_err;
2793	}
2794
2795	err = kvm_init_vector_slots();
2796	if (err) {
2797		kvm_err("Cannot initialise vector slots\n");
2798		goto out_hyp;
2799	}
2800
2801	err = init_subsystems();
2802	if (err)
2803		goto out_hyp;
2804
2805	kvm_info("%s%sVHE mode initialized successfully\n",
2806		 in_hyp_mode ? "" : (is_protected_kvm_enabled() ?
2807				     "Protected " : "Hyp "),
2808		 in_hyp_mode ? "" : (cpus_have_final_cap(ARM64_KVM_HVHE) ?
2809				     "h" : "n"));
2810
2811	/*
2812	 * FIXME: Do something reasonable if kvm_init() fails after pKVM
2813	 * hypervisor protection is finalized.
2814	 */
2815	err = kvm_init(sizeof(struct kvm_vcpu), 0, THIS_MODULE);
2816	if (err)
2817		goto out_subs;
2818
2819	/*
2820	 * This should be called after initialization is done and failure isn't
2821	 * possible anymore.
2822	 */
2823	if (!in_hyp_mode)
2824		finalize_init_hyp_mode();
2825
2826	kvm_arm_initialised = true;
2827
2828	return 0;
2829
2830out_subs:
2831	teardown_subsystems();
2832out_hyp:
2833	if (!in_hyp_mode)
2834		teardown_hyp_mode();
2835out_err:
2836	kvm_arm_vmid_alloc_free();
2837	return err;
2838}
2839
2840static int __init early_kvm_mode_cfg(char *arg)
2841{
2842	if (!arg)
2843		return -EINVAL;
2844
2845	if (strcmp(arg, "none") == 0) {
2846		kvm_mode = KVM_MODE_NONE;
2847		return 0;
2848	}
2849
2850	if (!is_hyp_mode_available()) {
2851		pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n");
2852		return 0;
2853	}
2854
2855	if (strcmp(arg, "protected") == 0) {
2856		if (!is_kernel_in_hyp_mode())
2857			kvm_mode = KVM_MODE_PROTECTED;
2858		else
2859			pr_warn_once("Protected KVM not available with VHE\n");
2860
2861		return 0;
2862	}
2863
2864	if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) {
2865		kvm_mode = KVM_MODE_DEFAULT;
2866		return 0;
2867	}
2868
2869	if (strcmp(arg, "nested") == 0 && !WARN_ON(!is_kernel_in_hyp_mode())) {
2870		kvm_mode = KVM_MODE_NV;
2871		return 0;
2872	}
2873
2874	return -EINVAL;
2875}
2876early_param("kvm-arm.mode", early_kvm_mode_cfg);
2877
2878static int __init early_kvm_wfx_trap_policy_cfg(char *arg, enum kvm_wfx_trap_policy *p)
2879{
2880	if (!arg)
2881		return -EINVAL;
2882
2883	if (strcmp(arg, "trap") == 0) {
2884		*p = KVM_WFX_TRAP;
2885		return 0;
2886	}
2887
2888	if (strcmp(arg, "notrap") == 0) {
2889		*p = KVM_WFX_NOTRAP;
2890		return 0;
2891	}
2892
2893	return -EINVAL;
2894}
2895
2896static int __init early_kvm_wfi_trap_policy_cfg(char *arg)
2897{
2898	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfi_trap_policy);
2899}
2900early_param("kvm-arm.wfi_trap_policy", early_kvm_wfi_trap_policy_cfg);
2901
2902static int __init early_kvm_wfe_trap_policy_cfg(char *arg)
2903{
2904	return early_kvm_wfx_trap_policy_cfg(arg, &kvm_wfe_trap_policy);
2905}
2906early_param("kvm-arm.wfe_trap_policy", early_kvm_wfe_trap_policy_cfg);
2907
2908enum kvm_mode kvm_get_mode(void)
2909{
2910	return kvm_mode;
2911}
2912
2913module_init(kvm_arm_init);