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/kmemleak.h> 20 #include <linux/kvm.h> 21 #include <linux/kvm_irqfd.h> 22 #include <linux/irqbypass.h> 23 #include <linux/sched/stat.h> 24 #include <linux/psci.h> 25 #include <trace/events/kvm.h> 26 27 #define CREATE_TRACE_POINTS 28 #include "trace_arm.h" 29 30 #include <linux/uaccess.h> 31 #include <asm/ptrace.h> 32 #include <asm/mman.h> 33 #include <asm/tlbflush.h> 34 #include <asm/cacheflush.h> 35 #include <asm/cpufeature.h> 36 #include <asm/virt.h> 37 #include <asm/kvm_arm.h> 38 #include <asm/kvm_asm.h> 39 #include <asm/kvm_mmu.h> 40 #include <asm/kvm_pkvm.h> 41 #include <asm/kvm_emulate.h> 42 #include <asm/sections.h> 43 44 #include <kvm/arm_hypercalls.h> 45 #include <kvm/arm_pmu.h> 46 #include <kvm/arm_psci.h> 47 48 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT; 49 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized); 50 51 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector); 52 53 DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); 54 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params); 55 56 static bool vgic_present; 57 58 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); 59 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); 60 61 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 62 { 63 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; 64 } 65 66 int kvm_arch_check_processor_compat(void *opaque) 67 { 68 return 0; 69 } 70 71 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 72 struct kvm_enable_cap *cap) 73 { 74 int r; 75 76 if (cap->flags) 77 return -EINVAL; 78 79 switch (cap->cap) { 80 case KVM_CAP_ARM_NISV_TO_USER: 81 r = 0; 82 set_bit(KVM_ARCH_FLAG_RETURN_NISV_IO_ABORT_TO_USER, 83 &kvm->arch.flags); 84 break; 85 case KVM_CAP_ARM_MTE: 86 mutex_lock(&kvm->lock); 87 if (!system_supports_mte() || kvm->created_vcpus) { 88 r = -EINVAL; 89 } else { 90 r = 0; 91 set_bit(KVM_ARCH_FLAG_MTE_ENABLED, &kvm->arch.flags); 92 } 93 mutex_unlock(&kvm->lock); 94 break; 95 case KVM_CAP_ARM_SYSTEM_SUSPEND: 96 r = 0; 97 set_bit(KVM_ARCH_FLAG_SYSTEM_SUSPEND_ENABLED, &kvm->arch.flags); 98 break; 99 default: 100 r = -EINVAL; 101 break; 102 } 103 104 return r; 105 } 106 107 static int kvm_arm_default_max_vcpus(void) 108 { 109 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; 110 } 111 112 static void set_default_spectre(struct kvm *kvm) 113 { 114 /* 115 * The default is to expose CSV2 == 1 if the HW isn't affected. 116 * Although this is a per-CPU feature, we make it global because 117 * asymmetric systems are just a nuisance. 118 * 119 * Userspace can override this as long as it doesn't promise 120 * the impossible. 121 */ 122 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) 123 kvm->arch.pfr0_csv2 = 1; 124 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) 125 kvm->arch.pfr0_csv3 = 1; 126 } 127 128 /** 129 * kvm_arch_init_vm - initializes a VM data structure 130 * @kvm: pointer to the KVM struct 131 */ 132 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 133 { 134 int ret; 135 136 ret = kvm_share_hyp(kvm, kvm + 1); 137 if (ret) 138 return ret; 139 140 ret = pkvm_init_host_vm(kvm); 141 if (ret) 142 goto err_unshare_kvm; 143 144 if (!zalloc_cpumask_var(&kvm->arch.supported_cpus, GFP_KERNEL)) { 145 ret = -ENOMEM; 146 goto err_unshare_kvm; 147 } 148 cpumask_copy(kvm->arch.supported_cpus, cpu_possible_mask); 149 150 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu, type); 151 if (ret) 152 goto err_free_cpumask; 153 154 kvm_vgic_early_init(kvm); 155 156 /* The maximum number of VCPUs is limited by the host's GIC model */ 157 kvm->max_vcpus = kvm_arm_default_max_vcpus(); 158 159 set_default_spectre(kvm); 160 kvm_arm_init_hypercalls(kvm); 161 162 /* 163 * Initialise the default PMUver before there is a chance to 164 * create an actual PMU. 165 */ 166 kvm->arch.dfr0_pmuver.imp = kvm_arm_pmu_get_pmuver_limit(); 167 168 return 0; 169 170 err_free_cpumask: 171 free_cpumask_var(kvm->arch.supported_cpus); 172 err_unshare_kvm: 173 kvm_unshare_hyp(kvm, kvm + 1); 174 return ret; 175 } 176 177 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 178 { 179 return VM_FAULT_SIGBUS; 180 } 181 182 183 /** 184 * kvm_arch_destroy_vm - destroy the VM data structure 185 * @kvm: pointer to the KVM struct 186 */ 187 void kvm_arch_destroy_vm(struct kvm *kvm) 188 { 189 bitmap_free(kvm->arch.pmu_filter); 190 free_cpumask_var(kvm->arch.supported_cpus); 191 192 kvm_vgic_destroy(kvm); 193 194 if (is_protected_kvm_enabled()) 195 pkvm_destroy_hyp_vm(kvm); 196 197 kvm_destroy_vcpus(kvm); 198 199 kvm_unshare_hyp(kvm, kvm + 1); 200 } 201 202 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 203 { 204 int r; 205 switch (ext) { 206 case KVM_CAP_IRQCHIP: 207 r = vgic_present; 208 break; 209 case KVM_CAP_IOEVENTFD: 210 case KVM_CAP_DEVICE_CTRL: 211 case KVM_CAP_USER_MEMORY: 212 case KVM_CAP_SYNC_MMU: 213 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: 214 case KVM_CAP_ONE_REG: 215 case KVM_CAP_ARM_PSCI: 216 case KVM_CAP_ARM_PSCI_0_2: 217 case KVM_CAP_READONLY_MEM: 218 case KVM_CAP_MP_STATE: 219 case KVM_CAP_IMMEDIATE_EXIT: 220 case KVM_CAP_VCPU_EVENTS: 221 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: 222 case KVM_CAP_ARM_NISV_TO_USER: 223 case KVM_CAP_ARM_INJECT_EXT_DABT: 224 case KVM_CAP_SET_GUEST_DEBUG: 225 case KVM_CAP_VCPU_ATTRIBUTES: 226 case KVM_CAP_PTP_KVM: 227 case KVM_CAP_ARM_SYSTEM_SUSPEND: 228 r = 1; 229 break; 230 case KVM_CAP_SET_GUEST_DEBUG2: 231 return KVM_GUESTDBG_VALID_MASK; 232 case KVM_CAP_ARM_SET_DEVICE_ADDR: 233 r = 1; 234 break; 235 case KVM_CAP_NR_VCPUS: 236 /* 237 * ARM64 treats KVM_CAP_NR_CPUS differently from all other 238 * architectures, as it does not always bound it to 239 * KVM_CAP_MAX_VCPUS. It should not matter much because 240 * this is just an advisory value. 241 */ 242 r = min_t(unsigned int, num_online_cpus(), 243 kvm_arm_default_max_vcpus()); 244 break; 245 case KVM_CAP_MAX_VCPUS: 246 case KVM_CAP_MAX_VCPU_ID: 247 if (kvm) 248 r = kvm->max_vcpus; 249 else 250 r = kvm_arm_default_max_vcpus(); 251 break; 252 case KVM_CAP_MSI_DEVID: 253 if (!kvm) 254 r = -EINVAL; 255 else 256 r = kvm->arch.vgic.msis_require_devid; 257 break; 258 case KVM_CAP_ARM_USER_IRQ: 259 /* 260 * 1: EL1_VTIMER, EL1_PTIMER, and PMU. 261 * (bump this number if adding more devices) 262 */ 263 r = 1; 264 break; 265 case KVM_CAP_ARM_MTE: 266 r = system_supports_mte(); 267 break; 268 case KVM_CAP_STEAL_TIME: 269 r = kvm_arm_pvtime_supported(); 270 break; 271 case KVM_CAP_ARM_EL1_32BIT: 272 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1); 273 break; 274 case KVM_CAP_GUEST_DEBUG_HW_BPS: 275 r = get_num_brps(); 276 break; 277 case KVM_CAP_GUEST_DEBUG_HW_WPS: 278 r = get_num_wrps(); 279 break; 280 case KVM_CAP_ARM_PMU_V3: 281 r = kvm_arm_support_pmu_v3(); 282 break; 283 case KVM_CAP_ARM_INJECT_SERROR_ESR: 284 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN); 285 break; 286 case KVM_CAP_ARM_VM_IPA_SIZE: 287 r = get_kvm_ipa_limit(); 288 break; 289 case KVM_CAP_ARM_SVE: 290 r = system_supports_sve(); 291 break; 292 case KVM_CAP_ARM_PTRAUTH_ADDRESS: 293 case KVM_CAP_ARM_PTRAUTH_GENERIC: 294 r = system_has_full_ptr_auth(); 295 break; 296 default: 297 r = 0; 298 } 299 300 return r; 301 } 302 303 long kvm_arch_dev_ioctl(struct file *filp, 304 unsigned int ioctl, unsigned long arg) 305 { 306 return -EINVAL; 307 } 308 309 struct kvm *kvm_arch_alloc_vm(void) 310 { 311 size_t sz = sizeof(struct kvm); 312 313 if (!has_vhe()) 314 return kzalloc(sz, GFP_KERNEL_ACCOUNT); 315 316 return __vmalloc(sz, GFP_KERNEL_ACCOUNT | __GFP_HIGHMEM | __GFP_ZERO); 317 } 318 319 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 320 { 321 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) 322 return -EBUSY; 323 324 if (id >= kvm->max_vcpus) 325 return -EINVAL; 326 327 return 0; 328 } 329 330 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 331 { 332 int err; 333 334 /* Force users to call KVM_ARM_VCPU_INIT */ 335 vcpu->arch.target = -1; 336 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 337 338 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO; 339 340 /* 341 * Default value for the FP state, will be overloaded at load 342 * time if we support FP (pretty likely) 343 */ 344 vcpu->arch.fp_state = FP_STATE_FREE; 345 346 /* Set up the timer */ 347 kvm_timer_vcpu_init(vcpu); 348 349 kvm_pmu_vcpu_init(vcpu); 350 351 kvm_arm_reset_debug_ptr(vcpu); 352 353 kvm_arm_pvtime_vcpu_init(&vcpu->arch); 354 355 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu; 356 357 err = kvm_vgic_vcpu_init(vcpu); 358 if (err) 359 return err; 360 361 return kvm_share_hyp(vcpu, vcpu + 1); 362 } 363 364 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 365 { 366 } 367 368 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 369 { 370 if (vcpu_has_run_once(vcpu) && unlikely(!irqchip_in_kernel(vcpu->kvm))) 371 static_branch_dec(&userspace_irqchip_in_use); 372 373 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); 374 kvm_timer_vcpu_terminate(vcpu); 375 kvm_pmu_vcpu_destroy(vcpu); 376 377 kvm_arm_vcpu_destroy(vcpu); 378 } 379 380 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) 381 { 382 383 } 384 385 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) 386 { 387 388 } 389 390 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 391 { 392 struct kvm_s2_mmu *mmu; 393 int *last_ran; 394 395 mmu = vcpu->arch.hw_mmu; 396 last_ran = this_cpu_ptr(mmu->last_vcpu_ran); 397 398 /* 399 * We guarantee that both TLBs and I-cache are private to each 400 * vcpu. If detecting that a vcpu from the same VM has 401 * previously run on the same physical CPU, call into the 402 * hypervisor code to nuke the relevant contexts. 403 * 404 * We might get preempted before the vCPU actually runs, but 405 * over-invalidation doesn't affect correctness. 406 */ 407 if (*last_ran != vcpu->vcpu_id) { 408 kvm_call_hyp(__kvm_flush_cpu_context, mmu); 409 *last_ran = vcpu->vcpu_id; 410 } 411 412 vcpu->cpu = cpu; 413 414 kvm_vgic_load(vcpu); 415 kvm_timer_vcpu_load(vcpu); 416 if (has_vhe()) 417 kvm_vcpu_load_sysregs_vhe(vcpu); 418 kvm_arch_vcpu_load_fp(vcpu); 419 kvm_vcpu_pmu_restore_guest(vcpu); 420 if (kvm_arm_is_pvtime_enabled(&vcpu->arch)) 421 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu); 422 423 if (single_task_running()) 424 vcpu_clear_wfx_traps(vcpu); 425 else 426 vcpu_set_wfx_traps(vcpu); 427 428 if (vcpu_has_ptrauth(vcpu)) 429 vcpu_ptrauth_disable(vcpu); 430 kvm_arch_vcpu_load_debug_state_flags(vcpu); 431 432 if (!cpumask_test_cpu(smp_processor_id(), vcpu->kvm->arch.supported_cpus)) 433 vcpu_set_on_unsupported_cpu(vcpu); 434 } 435 436 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 437 { 438 kvm_arch_vcpu_put_debug_state_flags(vcpu); 439 kvm_arch_vcpu_put_fp(vcpu); 440 if (has_vhe()) 441 kvm_vcpu_put_sysregs_vhe(vcpu); 442 kvm_timer_vcpu_put(vcpu); 443 kvm_vgic_put(vcpu); 444 kvm_vcpu_pmu_restore_host(vcpu); 445 kvm_arm_vmid_clear_active(); 446 447 vcpu_clear_on_unsupported_cpu(vcpu); 448 vcpu->cpu = -1; 449 } 450 451 void kvm_arm_vcpu_power_off(struct kvm_vcpu *vcpu) 452 { 453 vcpu->arch.mp_state.mp_state = KVM_MP_STATE_STOPPED; 454 kvm_make_request(KVM_REQ_SLEEP, vcpu); 455 kvm_vcpu_kick(vcpu); 456 } 457 458 bool kvm_arm_vcpu_stopped(struct kvm_vcpu *vcpu) 459 { 460 return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_STOPPED; 461 } 462 463 static void kvm_arm_vcpu_suspend(struct kvm_vcpu *vcpu) 464 { 465 vcpu->arch.mp_state.mp_state = KVM_MP_STATE_SUSPENDED; 466 kvm_make_request(KVM_REQ_SUSPEND, vcpu); 467 kvm_vcpu_kick(vcpu); 468 } 469 470 static bool kvm_arm_vcpu_suspended(struct kvm_vcpu *vcpu) 471 { 472 return vcpu->arch.mp_state.mp_state == KVM_MP_STATE_SUSPENDED; 473 } 474 475 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 476 struct kvm_mp_state *mp_state) 477 { 478 *mp_state = vcpu->arch.mp_state; 479 480 return 0; 481 } 482 483 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 484 struct kvm_mp_state *mp_state) 485 { 486 int ret = 0; 487 488 switch (mp_state->mp_state) { 489 case KVM_MP_STATE_RUNNABLE: 490 vcpu->arch.mp_state = *mp_state; 491 break; 492 case KVM_MP_STATE_STOPPED: 493 kvm_arm_vcpu_power_off(vcpu); 494 break; 495 case KVM_MP_STATE_SUSPENDED: 496 kvm_arm_vcpu_suspend(vcpu); 497 break; 498 default: 499 ret = -EINVAL; 500 } 501 502 return ret; 503 } 504 505 /** 506 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled 507 * @v: The VCPU pointer 508 * 509 * If the guest CPU is not waiting for interrupts or an interrupt line is 510 * asserted, the CPU is by definition runnable. 511 */ 512 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 513 { 514 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); 515 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) 516 && !kvm_arm_vcpu_stopped(v) && !v->arch.pause); 517 } 518 519 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 520 { 521 return vcpu_mode_priv(vcpu); 522 } 523 524 #ifdef CONFIG_GUEST_PERF_EVENTS 525 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu) 526 { 527 return *vcpu_pc(vcpu); 528 } 529 #endif 530 531 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) 532 { 533 return vcpu->arch.target >= 0; 534 } 535 536 /* 537 * Handle both the initialisation that is being done when the vcpu is 538 * run for the first time, as well as the updates that must be 539 * performed each time we get a new thread dealing with this vcpu. 540 */ 541 int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu) 542 { 543 struct kvm *kvm = vcpu->kvm; 544 int ret; 545 546 if (!kvm_vcpu_initialized(vcpu)) 547 return -ENOEXEC; 548 549 if (!kvm_arm_vcpu_is_finalized(vcpu)) 550 return -EPERM; 551 552 ret = kvm_arch_vcpu_run_map_fp(vcpu); 553 if (ret) 554 return ret; 555 556 if (likely(vcpu_has_run_once(vcpu))) 557 return 0; 558 559 kvm_arm_vcpu_init_debug(vcpu); 560 561 if (likely(irqchip_in_kernel(kvm))) { 562 /* 563 * Map the VGIC hardware resources before running a vcpu the 564 * first time on this VM. 565 */ 566 ret = kvm_vgic_map_resources(kvm); 567 if (ret) 568 return ret; 569 } 570 571 ret = kvm_timer_enable(vcpu); 572 if (ret) 573 return ret; 574 575 ret = kvm_arm_pmu_v3_enable(vcpu); 576 if (ret) 577 return ret; 578 579 if (is_protected_kvm_enabled()) { 580 ret = pkvm_create_hyp_vm(kvm); 581 if (ret) 582 return ret; 583 } 584 585 if (!irqchip_in_kernel(kvm)) { 586 /* 587 * Tell the rest of the code that there are userspace irqchip 588 * VMs in the wild. 589 */ 590 static_branch_inc(&userspace_irqchip_in_use); 591 } 592 593 /* 594 * Initialize traps for protected VMs. 595 * NOTE: Move to run in EL2 directly, rather than via a hypercall, once 596 * the code is in place for first run initialization at EL2. 597 */ 598 if (kvm_vm_is_protected(kvm)) 599 kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu); 600 601 mutex_lock(&kvm->lock); 602 set_bit(KVM_ARCH_FLAG_HAS_RAN_ONCE, &kvm->arch.flags); 603 mutex_unlock(&kvm->lock); 604 605 return ret; 606 } 607 608 bool kvm_arch_intc_initialized(struct kvm *kvm) 609 { 610 return vgic_initialized(kvm); 611 } 612 613 void kvm_arm_halt_guest(struct kvm *kvm) 614 { 615 unsigned long i; 616 struct kvm_vcpu *vcpu; 617 618 kvm_for_each_vcpu(i, vcpu, kvm) 619 vcpu->arch.pause = true; 620 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); 621 } 622 623 void kvm_arm_resume_guest(struct kvm *kvm) 624 { 625 unsigned long i; 626 struct kvm_vcpu *vcpu; 627 628 kvm_for_each_vcpu(i, vcpu, kvm) { 629 vcpu->arch.pause = false; 630 __kvm_vcpu_wake_up(vcpu); 631 } 632 } 633 634 static void kvm_vcpu_sleep(struct kvm_vcpu *vcpu) 635 { 636 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 637 638 rcuwait_wait_event(wait, 639 (!kvm_arm_vcpu_stopped(vcpu)) && (!vcpu->arch.pause), 640 TASK_INTERRUPTIBLE); 641 642 if (kvm_arm_vcpu_stopped(vcpu) || vcpu->arch.pause) { 643 /* Awaken to handle a signal, request we sleep again later. */ 644 kvm_make_request(KVM_REQ_SLEEP, vcpu); 645 } 646 647 /* 648 * Make sure we will observe a potential reset request if we've 649 * observed a change to the power state. Pairs with the smp_wmb() in 650 * kvm_psci_vcpu_on(). 651 */ 652 smp_rmb(); 653 } 654 655 /** 656 * kvm_vcpu_wfi - emulate Wait-For-Interrupt behavior 657 * @vcpu: The VCPU pointer 658 * 659 * Suspend execution of a vCPU until a valid wake event is detected, i.e. until 660 * the vCPU is runnable. The vCPU may or may not be scheduled out, depending 661 * on when a wake event arrives, e.g. there may already be a pending wake event. 662 */ 663 void kvm_vcpu_wfi(struct kvm_vcpu *vcpu) 664 { 665 /* 666 * Sync back the state of the GIC CPU interface so that we have 667 * the latest PMR and group enables. This ensures that 668 * kvm_arch_vcpu_runnable has up-to-date data to decide whether 669 * we have pending interrupts, e.g. when determining if the 670 * vCPU should block. 671 * 672 * For the same reason, we want to tell GICv4 that we need 673 * doorbells to be signalled, should an interrupt become pending. 674 */ 675 preempt_disable(); 676 kvm_vgic_vmcr_sync(vcpu); 677 vgic_v4_put(vcpu, true); 678 preempt_enable(); 679 680 kvm_vcpu_halt(vcpu); 681 vcpu_clear_flag(vcpu, IN_WFIT); 682 683 preempt_disable(); 684 vgic_v4_load(vcpu); 685 preempt_enable(); 686 } 687 688 static int kvm_vcpu_suspend(struct kvm_vcpu *vcpu) 689 { 690 if (!kvm_arm_vcpu_suspended(vcpu)) 691 return 1; 692 693 kvm_vcpu_wfi(vcpu); 694 695 /* 696 * The suspend state is sticky; we do not leave it until userspace 697 * explicitly marks the vCPU as runnable. Request that we suspend again 698 * later. 699 */ 700 kvm_make_request(KVM_REQ_SUSPEND, vcpu); 701 702 /* 703 * Check to make sure the vCPU is actually runnable. If so, exit to 704 * userspace informing it of the wakeup condition. 705 */ 706 if (kvm_arch_vcpu_runnable(vcpu)) { 707 memset(&vcpu->run->system_event, 0, sizeof(vcpu->run->system_event)); 708 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_WAKEUP; 709 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT; 710 return 0; 711 } 712 713 /* 714 * Otherwise, we were unblocked to process a different event, such as a 715 * pending signal. Return 1 and allow kvm_arch_vcpu_ioctl_run() to 716 * process the event. 717 */ 718 return 1; 719 } 720 721 /** 722 * check_vcpu_requests - check and handle pending vCPU requests 723 * @vcpu: the VCPU pointer 724 * 725 * Return: 1 if we should enter the guest 726 * 0 if we should exit to userspace 727 * < 0 if we should exit to userspace, where the return value indicates 728 * an error 729 */ 730 static int check_vcpu_requests(struct kvm_vcpu *vcpu) 731 { 732 if (kvm_request_pending(vcpu)) { 733 if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) 734 kvm_vcpu_sleep(vcpu); 735 736 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 737 kvm_reset_vcpu(vcpu); 738 739 /* 740 * Clear IRQ_PENDING requests that were made to guarantee 741 * that a VCPU sees new virtual interrupts. 742 */ 743 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); 744 745 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu)) 746 kvm_update_stolen_time(vcpu); 747 748 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) { 749 /* The distributor enable bits were changed */ 750 preempt_disable(); 751 vgic_v4_put(vcpu, false); 752 vgic_v4_load(vcpu); 753 preempt_enable(); 754 } 755 756 if (kvm_check_request(KVM_REQ_RELOAD_PMU, vcpu)) 757 kvm_pmu_handle_pmcr(vcpu, 758 __vcpu_sys_reg(vcpu, PMCR_EL0)); 759 760 if (kvm_check_request(KVM_REQ_SUSPEND, vcpu)) 761 return kvm_vcpu_suspend(vcpu); 762 763 if (kvm_dirty_ring_check_request(vcpu)) 764 return 0; 765 } 766 767 return 1; 768 } 769 770 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu) 771 { 772 if (likely(!vcpu_mode_is_32bit(vcpu))) 773 return false; 774 775 return !kvm_supports_32bit_el0(); 776 } 777 778 /** 779 * kvm_vcpu_exit_request - returns true if the VCPU should *not* enter the guest 780 * @vcpu: The VCPU pointer 781 * @ret: Pointer to write optional return code 782 * 783 * Returns: true if the VCPU needs to return to a preemptible + interruptible 784 * and skip guest entry. 785 * 786 * This function disambiguates between two different types of exits: exits to a 787 * preemptible + interruptible kernel context and exits to userspace. For an 788 * exit to userspace, this function will write the return code to ret and return 789 * true. For an exit to preemptible + interruptible kernel context (i.e. check 790 * for pending work and re-enter), return true without writing to ret. 791 */ 792 static bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu, int *ret) 793 { 794 struct kvm_run *run = vcpu->run; 795 796 /* 797 * If we're using a userspace irqchip, then check if we need 798 * to tell a userspace irqchip about timer or PMU level 799 * changes and if so, exit to userspace (the actual level 800 * state gets updated in kvm_timer_update_run and 801 * kvm_pmu_update_run below). 802 */ 803 if (static_branch_unlikely(&userspace_irqchip_in_use)) { 804 if (kvm_timer_should_notify_user(vcpu) || 805 kvm_pmu_should_notify_user(vcpu)) { 806 *ret = -EINTR; 807 run->exit_reason = KVM_EXIT_INTR; 808 return true; 809 } 810 } 811 812 if (unlikely(vcpu_on_unsupported_cpu(vcpu))) { 813 run->exit_reason = KVM_EXIT_FAIL_ENTRY; 814 run->fail_entry.hardware_entry_failure_reason = KVM_EXIT_FAIL_ENTRY_CPU_UNSUPPORTED; 815 run->fail_entry.cpu = smp_processor_id(); 816 *ret = 0; 817 return true; 818 } 819 820 return kvm_request_pending(vcpu) || 821 xfer_to_guest_mode_work_pending(); 822 } 823 824 /* 825 * Actually run the vCPU, entering an RCU extended quiescent state (EQS) while 826 * the vCPU is running. 827 * 828 * This must be noinstr as instrumentation may make use of RCU, and this is not 829 * safe during the EQS. 830 */ 831 static int noinstr kvm_arm_vcpu_enter_exit(struct kvm_vcpu *vcpu) 832 { 833 int ret; 834 835 guest_state_enter_irqoff(); 836 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu); 837 guest_state_exit_irqoff(); 838 839 return ret; 840 } 841 842 /** 843 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code 844 * @vcpu: The VCPU pointer 845 * 846 * This function is called through the VCPU_RUN ioctl called from user space. It 847 * will execute VM code in a loop until the time slice for the process is used 848 * or some emulation is needed from user space in which case the function will 849 * return with return value 0 and with the kvm_run structure filled in with the 850 * required data for the requested emulation. 851 */ 852 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 853 { 854 struct kvm_run *run = vcpu->run; 855 int ret; 856 857 if (run->exit_reason == KVM_EXIT_MMIO) { 858 ret = kvm_handle_mmio_return(vcpu); 859 if (ret) 860 return ret; 861 } 862 863 vcpu_load(vcpu); 864 865 if (run->immediate_exit) { 866 ret = -EINTR; 867 goto out; 868 } 869 870 kvm_sigset_activate(vcpu); 871 872 ret = 1; 873 run->exit_reason = KVM_EXIT_UNKNOWN; 874 run->flags = 0; 875 while (ret > 0) { 876 /* 877 * Check conditions before entering the guest 878 */ 879 ret = xfer_to_guest_mode_handle_work(vcpu); 880 if (!ret) 881 ret = 1; 882 883 if (ret > 0) 884 ret = check_vcpu_requests(vcpu); 885 886 /* 887 * Preparing the interrupts to be injected also 888 * involves poking the GIC, which must be done in a 889 * non-preemptible context. 890 */ 891 preempt_disable(); 892 893 /* 894 * The VMID allocator only tracks active VMIDs per 895 * physical CPU, and therefore the VMID allocated may not be 896 * preserved on VMID roll-over if the task was preempted, 897 * making a thread's VMID inactive. So we need to call 898 * kvm_arm_vmid_update() in non-premptible context. 899 */ 900 kvm_arm_vmid_update(&vcpu->arch.hw_mmu->vmid); 901 902 kvm_pmu_flush_hwstate(vcpu); 903 904 local_irq_disable(); 905 906 kvm_vgic_flush_hwstate(vcpu); 907 908 kvm_pmu_update_vcpu_events(vcpu); 909 910 /* 911 * Ensure we set mode to IN_GUEST_MODE after we disable 912 * interrupts and before the final VCPU requests check. 913 * See the comment in kvm_vcpu_exiting_guest_mode() and 914 * Documentation/virt/kvm/vcpu-requests.rst 915 */ 916 smp_store_mb(vcpu->mode, IN_GUEST_MODE); 917 918 if (ret <= 0 || kvm_vcpu_exit_request(vcpu, &ret)) { 919 vcpu->mode = OUTSIDE_GUEST_MODE; 920 isb(); /* Ensure work in x_flush_hwstate is committed */ 921 kvm_pmu_sync_hwstate(vcpu); 922 if (static_branch_unlikely(&userspace_irqchip_in_use)) 923 kvm_timer_sync_user(vcpu); 924 kvm_vgic_sync_hwstate(vcpu); 925 local_irq_enable(); 926 preempt_enable(); 927 continue; 928 } 929 930 kvm_arm_setup_debug(vcpu); 931 kvm_arch_vcpu_ctxflush_fp(vcpu); 932 933 /************************************************************** 934 * Enter the guest 935 */ 936 trace_kvm_entry(*vcpu_pc(vcpu)); 937 guest_timing_enter_irqoff(); 938 939 ret = kvm_arm_vcpu_enter_exit(vcpu); 940 941 vcpu->mode = OUTSIDE_GUEST_MODE; 942 vcpu->stat.exits++; 943 /* 944 * Back from guest 945 *************************************************************/ 946 947 kvm_arm_clear_debug(vcpu); 948 949 /* 950 * We must sync the PMU state before the vgic state so 951 * that the vgic can properly sample the updated state of the 952 * interrupt line. 953 */ 954 kvm_pmu_sync_hwstate(vcpu); 955 956 /* 957 * Sync the vgic state before syncing the timer state because 958 * the timer code needs to know if the virtual timer 959 * interrupts are active. 960 */ 961 kvm_vgic_sync_hwstate(vcpu); 962 963 /* 964 * Sync the timer hardware state before enabling interrupts as 965 * we don't want vtimer interrupts to race with syncing the 966 * timer virtual interrupt state. 967 */ 968 if (static_branch_unlikely(&userspace_irqchip_in_use)) 969 kvm_timer_sync_user(vcpu); 970 971 kvm_arch_vcpu_ctxsync_fp(vcpu); 972 973 /* 974 * We must ensure that any pending interrupts are taken before 975 * we exit guest timing so that timer ticks are accounted as 976 * guest time. Transiently unmask interrupts so that any 977 * pending interrupts are taken. 978 * 979 * Per ARM DDI 0487G.b section D1.13.4, an ISB (or other 980 * context synchronization event) is necessary to ensure that 981 * pending interrupts are taken. 982 */ 983 if (ARM_EXCEPTION_CODE(ret) == ARM_EXCEPTION_IRQ) { 984 local_irq_enable(); 985 isb(); 986 local_irq_disable(); 987 } 988 989 guest_timing_exit_irqoff(); 990 991 local_irq_enable(); 992 993 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); 994 995 /* Exit types that need handling before we can be preempted */ 996 handle_exit_early(vcpu, ret); 997 998 preempt_enable(); 999 1000 /* 1001 * The ARMv8 architecture doesn't give the hypervisor 1002 * a mechanism to prevent a guest from dropping to AArch32 EL0 1003 * if implemented by the CPU. If we spot the guest in such 1004 * state and that we decided it wasn't supposed to do so (like 1005 * with the asymmetric AArch32 case), return to userspace with 1006 * a fatal error. 1007 */ 1008 if (vcpu_mode_is_bad_32bit(vcpu)) { 1009 /* 1010 * As we have caught the guest red-handed, decide that 1011 * it isn't fit for purpose anymore by making the vcpu 1012 * invalid. The VMM can try and fix it by issuing a 1013 * KVM_ARM_VCPU_INIT if it really wants to. 1014 */ 1015 vcpu->arch.target = -1; 1016 ret = ARM_EXCEPTION_IL; 1017 } 1018 1019 ret = handle_exit(vcpu, ret); 1020 } 1021 1022 /* Tell userspace about in-kernel device output levels */ 1023 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { 1024 kvm_timer_update_run(vcpu); 1025 kvm_pmu_update_run(vcpu); 1026 } 1027 1028 kvm_sigset_deactivate(vcpu); 1029 1030 out: 1031 /* 1032 * In the unlikely event that we are returning to userspace 1033 * with pending exceptions or PC adjustment, commit these 1034 * adjustments in order to give userspace a consistent view of 1035 * the vcpu state. Note that this relies on __kvm_adjust_pc() 1036 * being preempt-safe on VHE. 1037 */ 1038 if (unlikely(vcpu_get_flag(vcpu, PENDING_EXCEPTION) || 1039 vcpu_get_flag(vcpu, INCREMENT_PC))) 1040 kvm_call_hyp(__kvm_adjust_pc, vcpu); 1041 1042 vcpu_put(vcpu); 1043 return ret; 1044 } 1045 1046 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) 1047 { 1048 int bit_index; 1049 bool set; 1050 unsigned long *hcr; 1051 1052 if (number == KVM_ARM_IRQ_CPU_IRQ) 1053 bit_index = __ffs(HCR_VI); 1054 else /* KVM_ARM_IRQ_CPU_FIQ */ 1055 bit_index = __ffs(HCR_VF); 1056 1057 hcr = vcpu_hcr(vcpu); 1058 if (level) 1059 set = test_and_set_bit(bit_index, hcr); 1060 else 1061 set = test_and_clear_bit(bit_index, hcr); 1062 1063 /* 1064 * If we didn't change anything, no need to wake up or kick other CPUs 1065 */ 1066 if (set == level) 1067 return 0; 1068 1069 /* 1070 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and 1071 * trigger a world-switch round on the running physical CPU to set the 1072 * virtual IRQ/FIQ fields in the HCR appropriately. 1073 */ 1074 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); 1075 kvm_vcpu_kick(vcpu); 1076 1077 return 0; 1078 } 1079 1080 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 1081 bool line_status) 1082 { 1083 u32 irq = irq_level->irq; 1084 unsigned int irq_type, vcpu_idx, irq_num; 1085 int nrcpus = atomic_read(&kvm->online_vcpus); 1086 struct kvm_vcpu *vcpu = NULL; 1087 bool level = irq_level->level; 1088 1089 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; 1090 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; 1091 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); 1092 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; 1093 1094 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); 1095 1096 switch (irq_type) { 1097 case KVM_ARM_IRQ_TYPE_CPU: 1098 if (irqchip_in_kernel(kvm)) 1099 return -ENXIO; 1100 1101 if (vcpu_idx >= nrcpus) 1102 return -EINVAL; 1103 1104 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 1105 if (!vcpu) 1106 return -EINVAL; 1107 1108 if (irq_num > KVM_ARM_IRQ_CPU_FIQ) 1109 return -EINVAL; 1110 1111 return vcpu_interrupt_line(vcpu, irq_num, level); 1112 case KVM_ARM_IRQ_TYPE_PPI: 1113 if (!irqchip_in_kernel(kvm)) 1114 return -ENXIO; 1115 1116 if (vcpu_idx >= nrcpus) 1117 return -EINVAL; 1118 1119 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 1120 if (!vcpu) 1121 return -EINVAL; 1122 1123 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) 1124 return -EINVAL; 1125 1126 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); 1127 case KVM_ARM_IRQ_TYPE_SPI: 1128 if (!irqchip_in_kernel(kvm)) 1129 return -ENXIO; 1130 1131 if (irq_num < VGIC_NR_PRIVATE_IRQS) 1132 return -EINVAL; 1133 1134 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); 1135 } 1136 1137 return -EINVAL; 1138 } 1139 1140 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, 1141 const struct kvm_vcpu_init *init) 1142 { 1143 unsigned int i, ret; 1144 u32 phys_target = kvm_target_cpu(); 1145 1146 if (init->target != phys_target) 1147 return -EINVAL; 1148 1149 /* 1150 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 1151 * use the same target. 1152 */ 1153 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) 1154 return -EINVAL; 1155 1156 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ 1157 for (i = 0; i < sizeof(init->features) * 8; i++) { 1158 bool set = (init->features[i / 32] & (1 << (i % 32))); 1159 1160 if (set && i >= KVM_VCPU_MAX_FEATURES) 1161 return -ENOENT; 1162 1163 /* 1164 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 1165 * use the same feature set. 1166 */ 1167 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && 1168 test_bit(i, vcpu->arch.features) != set) 1169 return -EINVAL; 1170 1171 if (set) 1172 set_bit(i, vcpu->arch.features); 1173 } 1174 1175 vcpu->arch.target = phys_target; 1176 1177 /* Now we know what it is, we can reset it. */ 1178 ret = kvm_reset_vcpu(vcpu); 1179 if (ret) { 1180 vcpu->arch.target = -1; 1181 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 1182 } 1183 1184 return ret; 1185 } 1186 1187 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, 1188 struct kvm_vcpu_init *init) 1189 { 1190 int ret; 1191 1192 ret = kvm_vcpu_set_target(vcpu, init); 1193 if (ret) 1194 return ret; 1195 1196 /* 1197 * Ensure a rebooted VM will fault in RAM pages and detect if the 1198 * guest MMU is turned off and flush the caches as needed. 1199 * 1200 * S2FWB enforces all memory accesses to RAM being cacheable, 1201 * ensuring that the data side is always coherent. We still 1202 * need to invalidate the I-cache though, as FWB does *not* 1203 * imply CTR_EL0.DIC. 1204 */ 1205 if (vcpu_has_run_once(vcpu)) { 1206 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB)) 1207 stage2_unmap_vm(vcpu->kvm); 1208 else 1209 icache_inval_all_pou(); 1210 } 1211 1212 vcpu_reset_hcr(vcpu); 1213 vcpu->arch.cptr_el2 = CPTR_EL2_DEFAULT; 1214 1215 /* 1216 * Handle the "start in power-off" case. 1217 */ 1218 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) 1219 kvm_arm_vcpu_power_off(vcpu); 1220 else 1221 vcpu->arch.mp_state.mp_state = KVM_MP_STATE_RUNNABLE; 1222 1223 return 0; 1224 } 1225 1226 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, 1227 struct kvm_device_attr *attr) 1228 { 1229 int ret = -ENXIO; 1230 1231 switch (attr->group) { 1232 default: 1233 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); 1234 break; 1235 } 1236 1237 return ret; 1238 } 1239 1240 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, 1241 struct kvm_device_attr *attr) 1242 { 1243 int ret = -ENXIO; 1244 1245 switch (attr->group) { 1246 default: 1247 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); 1248 break; 1249 } 1250 1251 return ret; 1252 } 1253 1254 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, 1255 struct kvm_device_attr *attr) 1256 { 1257 int ret = -ENXIO; 1258 1259 switch (attr->group) { 1260 default: 1261 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); 1262 break; 1263 } 1264 1265 return ret; 1266 } 1267 1268 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1269 struct kvm_vcpu_events *events) 1270 { 1271 memset(events, 0, sizeof(*events)); 1272 1273 return __kvm_arm_vcpu_get_events(vcpu, events); 1274 } 1275 1276 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1277 struct kvm_vcpu_events *events) 1278 { 1279 int i; 1280 1281 /* check whether the reserved field is zero */ 1282 for (i = 0; i < ARRAY_SIZE(events->reserved); i++) 1283 if (events->reserved[i]) 1284 return -EINVAL; 1285 1286 /* check whether the pad field is zero */ 1287 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) 1288 if (events->exception.pad[i]) 1289 return -EINVAL; 1290 1291 return __kvm_arm_vcpu_set_events(vcpu, events); 1292 } 1293 1294 long kvm_arch_vcpu_ioctl(struct file *filp, 1295 unsigned int ioctl, unsigned long arg) 1296 { 1297 struct kvm_vcpu *vcpu = filp->private_data; 1298 void __user *argp = (void __user *)arg; 1299 struct kvm_device_attr attr; 1300 long r; 1301 1302 switch (ioctl) { 1303 case KVM_ARM_VCPU_INIT: { 1304 struct kvm_vcpu_init init; 1305 1306 r = -EFAULT; 1307 if (copy_from_user(&init, argp, sizeof(init))) 1308 break; 1309 1310 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); 1311 break; 1312 } 1313 case KVM_SET_ONE_REG: 1314 case KVM_GET_ONE_REG: { 1315 struct kvm_one_reg reg; 1316 1317 r = -ENOEXEC; 1318 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1319 break; 1320 1321 r = -EFAULT; 1322 if (copy_from_user(®, argp, sizeof(reg))) 1323 break; 1324 1325 /* 1326 * We could owe a reset due to PSCI. Handle the pending reset 1327 * here to ensure userspace register accesses are ordered after 1328 * the reset. 1329 */ 1330 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 1331 kvm_reset_vcpu(vcpu); 1332 1333 if (ioctl == KVM_SET_ONE_REG) 1334 r = kvm_arm_set_reg(vcpu, ®); 1335 else 1336 r = kvm_arm_get_reg(vcpu, ®); 1337 break; 1338 } 1339 case KVM_GET_REG_LIST: { 1340 struct kvm_reg_list __user *user_list = argp; 1341 struct kvm_reg_list reg_list; 1342 unsigned n; 1343 1344 r = -ENOEXEC; 1345 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1346 break; 1347 1348 r = -EPERM; 1349 if (!kvm_arm_vcpu_is_finalized(vcpu)) 1350 break; 1351 1352 r = -EFAULT; 1353 if (copy_from_user(®_list, user_list, sizeof(reg_list))) 1354 break; 1355 n = reg_list.n; 1356 reg_list.n = kvm_arm_num_regs(vcpu); 1357 if (copy_to_user(user_list, ®_list, sizeof(reg_list))) 1358 break; 1359 r = -E2BIG; 1360 if (n < reg_list.n) 1361 break; 1362 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); 1363 break; 1364 } 1365 case KVM_SET_DEVICE_ATTR: { 1366 r = -EFAULT; 1367 if (copy_from_user(&attr, argp, sizeof(attr))) 1368 break; 1369 r = kvm_arm_vcpu_set_attr(vcpu, &attr); 1370 break; 1371 } 1372 case KVM_GET_DEVICE_ATTR: { 1373 r = -EFAULT; 1374 if (copy_from_user(&attr, argp, sizeof(attr))) 1375 break; 1376 r = kvm_arm_vcpu_get_attr(vcpu, &attr); 1377 break; 1378 } 1379 case KVM_HAS_DEVICE_ATTR: { 1380 r = -EFAULT; 1381 if (copy_from_user(&attr, argp, sizeof(attr))) 1382 break; 1383 r = kvm_arm_vcpu_has_attr(vcpu, &attr); 1384 break; 1385 } 1386 case KVM_GET_VCPU_EVENTS: { 1387 struct kvm_vcpu_events events; 1388 1389 if (kvm_arm_vcpu_get_events(vcpu, &events)) 1390 return -EINVAL; 1391 1392 if (copy_to_user(argp, &events, sizeof(events))) 1393 return -EFAULT; 1394 1395 return 0; 1396 } 1397 case KVM_SET_VCPU_EVENTS: { 1398 struct kvm_vcpu_events events; 1399 1400 if (copy_from_user(&events, argp, sizeof(events))) 1401 return -EFAULT; 1402 1403 return kvm_arm_vcpu_set_events(vcpu, &events); 1404 } 1405 case KVM_ARM_VCPU_FINALIZE: { 1406 int what; 1407 1408 if (!kvm_vcpu_initialized(vcpu)) 1409 return -ENOEXEC; 1410 1411 if (get_user(what, (const int __user *)argp)) 1412 return -EFAULT; 1413 1414 return kvm_arm_vcpu_finalize(vcpu, what); 1415 } 1416 default: 1417 r = -EINVAL; 1418 } 1419 1420 return r; 1421 } 1422 1423 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 1424 { 1425 1426 } 1427 1428 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm, 1429 const struct kvm_memory_slot *memslot) 1430 { 1431 kvm_flush_remote_tlbs(kvm); 1432 } 1433 1434 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, 1435 struct kvm_arm_device_addr *dev_addr) 1436 { 1437 switch (FIELD_GET(KVM_ARM_DEVICE_ID_MASK, dev_addr->id)) { 1438 case KVM_ARM_DEVICE_VGIC_V2: 1439 if (!vgic_present) 1440 return -ENXIO; 1441 return kvm_set_legacy_vgic_v2_addr(kvm, dev_addr); 1442 default: 1443 return -ENODEV; 1444 } 1445 } 1446 1447 long kvm_arch_vm_ioctl(struct file *filp, 1448 unsigned int ioctl, unsigned long arg) 1449 { 1450 struct kvm *kvm = filp->private_data; 1451 void __user *argp = (void __user *)arg; 1452 1453 switch (ioctl) { 1454 case KVM_CREATE_IRQCHIP: { 1455 int ret; 1456 if (!vgic_present) 1457 return -ENXIO; 1458 mutex_lock(&kvm->lock); 1459 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); 1460 mutex_unlock(&kvm->lock); 1461 return ret; 1462 } 1463 case KVM_ARM_SET_DEVICE_ADDR: { 1464 struct kvm_arm_device_addr dev_addr; 1465 1466 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) 1467 return -EFAULT; 1468 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); 1469 } 1470 case KVM_ARM_PREFERRED_TARGET: { 1471 struct kvm_vcpu_init init; 1472 1473 kvm_vcpu_preferred_target(&init); 1474 1475 if (copy_to_user(argp, &init, sizeof(init))) 1476 return -EFAULT; 1477 1478 return 0; 1479 } 1480 case KVM_ARM_MTE_COPY_TAGS: { 1481 struct kvm_arm_copy_mte_tags copy_tags; 1482 1483 if (copy_from_user(©_tags, argp, sizeof(copy_tags))) 1484 return -EFAULT; 1485 return kvm_vm_ioctl_mte_copy_tags(kvm, ©_tags); 1486 } 1487 default: 1488 return -EINVAL; 1489 } 1490 } 1491 1492 static unsigned long nvhe_percpu_size(void) 1493 { 1494 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) - 1495 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start); 1496 } 1497 1498 static unsigned long nvhe_percpu_order(void) 1499 { 1500 unsigned long size = nvhe_percpu_size(); 1501 1502 return size ? get_order(size) : 0; 1503 } 1504 1505 /* A lookup table holding the hypervisor VA for each vector slot */ 1506 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS]; 1507 1508 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot) 1509 { 1510 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot); 1511 } 1512 1513 static int kvm_init_vector_slots(void) 1514 { 1515 int err; 1516 void *base; 1517 1518 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector)); 1519 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT); 1520 1521 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs)); 1522 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT); 1523 1524 if (kvm_system_needs_idmapped_vectors() && 1525 !is_protected_kvm_enabled()) { 1526 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs), 1527 __BP_HARDEN_HYP_VECS_SZ, &base); 1528 if (err) 1529 return err; 1530 } 1531 1532 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT); 1533 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT); 1534 return 0; 1535 } 1536 1537 static void cpu_prepare_hyp_mode(int cpu, u32 hyp_va_bits) 1538 { 1539 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu); 1540 unsigned long tcr; 1541 1542 /* 1543 * Calculate the raw per-cpu offset without a translation from the 1544 * kernel's mapping to the linear mapping, and store it in tpidr_el2 1545 * so that we can use adr_l to access per-cpu variables in EL2. 1546 * Also drop the KASAN tag which gets in the way... 1547 */ 1548 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) - 1549 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start)); 1550 1551 params->mair_el2 = read_sysreg(mair_el1); 1552 1553 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1; 1554 tcr &= ~TCR_T0SZ_MASK; 1555 tcr |= TCR_T0SZ(hyp_va_bits); 1556 params->tcr_el2 = tcr; 1557 1558 params->pgd_pa = kvm_mmu_get_httbr(); 1559 if (is_protected_kvm_enabled()) 1560 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS; 1561 else 1562 params->hcr_el2 = HCR_HOST_NVHE_FLAGS; 1563 params->vttbr = params->vtcr = 0; 1564 1565 /* 1566 * Flush the init params from the data cache because the struct will 1567 * be read while the MMU is off. 1568 */ 1569 kvm_flush_dcache_to_poc(params, sizeof(*params)); 1570 } 1571 1572 static void hyp_install_host_vector(void) 1573 { 1574 struct kvm_nvhe_init_params *params; 1575 struct arm_smccc_res res; 1576 1577 /* Switch from the HYP stub to our own HYP init vector */ 1578 __hyp_set_vectors(kvm_get_idmap_vector()); 1579 1580 /* 1581 * Call initialization code, and switch to the full blown HYP code. 1582 * If the cpucaps haven't been finalized yet, something has gone very 1583 * wrong, and hyp will crash and burn when it uses any 1584 * cpus_have_const_cap() wrapper. 1585 */ 1586 BUG_ON(!system_capabilities_finalized()); 1587 params = this_cpu_ptr_nvhe_sym(kvm_init_params); 1588 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res); 1589 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); 1590 } 1591 1592 static void cpu_init_hyp_mode(void) 1593 { 1594 hyp_install_host_vector(); 1595 1596 /* 1597 * Disabling SSBD on a non-VHE system requires us to enable SSBS 1598 * at EL2. 1599 */ 1600 if (this_cpu_has_cap(ARM64_SSBS) && 1601 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) { 1602 kvm_call_hyp_nvhe(__kvm_enable_ssbs); 1603 } 1604 } 1605 1606 static void cpu_hyp_reset(void) 1607 { 1608 if (!is_kernel_in_hyp_mode()) 1609 __hyp_reset_vectors(); 1610 } 1611 1612 /* 1613 * EL2 vectors can be mapped and rerouted in a number of ways, 1614 * depending on the kernel configuration and CPU present: 1615 * 1616 * - If the CPU is affected by Spectre-v2, the hardening sequence is 1617 * placed in one of the vector slots, which is executed before jumping 1618 * to the real vectors. 1619 * 1620 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot 1621 * containing the hardening sequence is mapped next to the idmap page, 1622 * and executed before jumping to the real vectors. 1623 * 1624 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an 1625 * empty slot is selected, mapped next to the idmap page, and 1626 * executed before jumping to the real vectors. 1627 * 1628 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with 1629 * VHE, as we don't have hypervisor-specific mappings. If the system 1630 * is VHE and yet selects this capability, it will be ignored. 1631 */ 1632 static void cpu_set_hyp_vector(void) 1633 { 1634 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data); 1635 void *vector = hyp_spectre_vector_selector[data->slot]; 1636 1637 if (!is_protected_kvm_enabled()) 1638 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector; 1639 else 1640 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot); 1641 } 1642 1643 static void cpu_hyp_init_context(void) 1644 { 1645 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt); 1646 1647 if (!is_kernel_in_hyp_mode()) 1648 cpu_init_hyp_mode(); 1649 } 1650 1651 static void cpu_hyp_init_features(void) 1652 { 1653 cpu_set_hyp_vector(); 1654 kvm_arm_init_debug(); 1655 1656 if (is_kernel_in_hyp_mode()) 1657 kvm_timer_init_vhe(); 1658 1659 if (vgic_present) 1660 kvm_vgic_init_cpu_hardware(); 1661 } 1662 1663 static void cpu_hyp_reinit(void) 1664 { 1665 cpu_hyp_reset(); 1666 cpu_hyp_init_context(); 1667 cpu_hyp_init_features(); 1668 } 1669 1670 static void _kvm_arch_hardware_enable(void *discard) 1671 { 1672 if (!__this_cpu_read(kvm_arm_hardware_enabled)) { 1673 cpu_hyp_reinit(); 1674 __this_cpu_write(kvm_arm_hardware_enabled, 1); 1675 } 1676 } 1677 1678 int kvm_arch_hardware_enable(void) 1679 { 1680 int was_enabled = __this_cpu_read(kvm_arm_hardware_enabled); 1681 1682 _kvm_arch_hardware_enable(NULL); 1683 1684 if (!was_enabled) { 1685 kvm_vgic_cpu_up(); 1686 kvm_timer_cpu_up(); 1687 } 1688 1689 return 0; 1690 } 1691 1692 static void _kvm_arch_hardware_disable(void *discard) 1693 { 1694 if (__this_cpu_read(kvm_arm_hardware_enabled)) { 1695 cpu_hyp_reset(); 1696 __this_cpu_write(kvm_arm_hardware_enabled, 0); 1697 } 1698 } 1699 1700 void kvm_arch_hardware_disable(void) 1701 { 1702 if (__this_cpu_read(kvm_arm_hardware_enabled)) { 1703 kvm_timer_cpu_down(); 1704 kvm_vgic_cpu_down(); 1705 } 1706 1707 if (!is_protected_kvm_enabled()) 1708 _kvm_arch_hardware_disable(NULL); 1709 } 1710 1711 #ifdef CONFIG_CPU_PM 1712 static int hyp_init_cpu_pm_notifier(struct notifier_block *self, 1713 unsigned long cmd, 1714 void *v) 1715 { 1716 /* 1717 * kvm_arm_hardware_enabled is left with its old value over 1718 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should 1719 * re-enable hyp. 1720 */ 1721 switch (cmd) { 1722 case CPU_PM_ENTER: 1723 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1724 /* 1725 * don't update kvm_arm_hardware_enabled here 1726 * so that the hardware will be re-enabled 1727 * when we resume. See below. 1728 */ 1729 cpu_hyp_reset(); 1730 1731 return NOTIFY_OK; 1732 case CPU_PM_ENTER_FAILED: 1733 case CPU_PM_EXIT: 1734 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1735 /* The hardware was enabled before suspend. */ 1736 cpu_hyp_reinit(); 1737 1738 return NOTIFY_OK; 1739 1740 default: 1741 return NOTIFY_DONE; 1742 } 1743 } 1744 1745 static struct notifier_block hyp_init_cpu_pm_nb = { 1746 .notifier_call = hyp_init_cpu_pm_notifier, 1747 }; 1748 1749 static void hyp_cpu_pm_init(void) 1750 { 1751 if (!is_protected_kvm_enabled()) 1752 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); 1753 } 1754 static void hyp_cpu_pm_exit(void) 1755 { 1756 if (!is_protected_kvm_enabled()) 1757 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); 1758 } 1759 #else 1760 static inline void hyp_cpu_pm_init(void) 1761 { 1762 } 1763 static inline void hyp_cpu_pm_exit(void) 1764 { 1765 } 1766 #endif 1767 1768 static void init_cpu_logical_map(void) 1769 { 1770 unsigned int cpu; 1771 1772 /* 1773 * Copy the MPIDR <-> logical CPU ID mapping to hyp. 1774 * Only copy the set of online CPUs whose features have been checked 1775 * against the finalized system capabilities. The hypervisor will not 1776 * allow any other CPUs from the `possible` set to boot. 1777 */ 1778 for_each_online_cpu(cpu) 1779 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu); 1780 } 1781 1782 #define init_psci_0_1_impl_state(config, what) \ 1783 config.psci_0_1_ ## what ## _implemented = psci_ops.what 1784 1785 static bool init_psci_relay(void) 1786 { 1787 /* 1788 * If PSCI has not been initialized, protected KVM cannot install 1789 * itself on newly booted CPUs. 1790 */ 1791 if (!psci_ops.get_version) { 1792 kvm_err("Cannot initialize protected mode without PSCI\n"); 1793 return false; 1794 } 1795 1796 kvm_host_psci_config.version = psci_ops.get_version(); 1797 1798 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) { 1799 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids(); 1800 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend); 1801 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on); 1802 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off); 1803 init_psci_0_1_impl_state(kvm_host_psci_config, migrate); 1804 } 1805 return true; 1806 } 1807 1808 static int init_subsystems(void) 1809 { 1810 int err = 0; 1811 1812 /* 1813 * Enable hardware so that subsystem initialisation can access EL2. 1814 */ 1815 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); 1816 1817 /* 1818 * Register CPU lower-power notifier 1819 */ 1820 hyp_cpu_pm_init(); 1821 1822 /* 1823 * Init HYP view of VGIC 1824 */ 1825 err = kvm_vgic_hyp_init(); 1826 switch (err) { 1827 case 0: 1828 vgic_present = true; 1829 break; 1830 case -ENODEV: 1831 case -ENXIO: 1832 vgic_present = false; 1833 err = 0; 1834 break; 1835 default: 1836 goto out; 1837 } 1838 1839 /* 1840 * Init HYP architected timer support 1841 */ 1842 err = kvm_timer_hyp_init(vgic_present); 1843 if (err) 1844 goto out; 1845 1846 kvm_register_perf_callbacks(NULL); 1847 1848 out: 1849 if (err || !is_protected_kvm_enabled()) 1850 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); 1851 1852 return err; 1853 } 1854 1855 static void teardown_hyp_mode(void) 1856 { 1857 int cpu; 1858 1859 free_hyp_pgds(); 1860 for_each_possible_cpu(cpu) { 1861 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); 1862 free_pages(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu], nvhe_percpu_order()); 1863 } 1864 } 1865 1866 static int do_pkvm_init(u32 hyp_va_bits) 1867 { 1868 void *per_cpu_base = kvm_ksym_ref(kvm_nvhe_sym(kvm_arm_hyp_percpu_base)); 1869 int ret; 1870 1871 preempt_disable(); 1872 cpu_hyp_init_context(); 1873 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size, 1874 num_possible_cpus(), kern_hyp_va(per_cpu_base), 1875 hyp_va_bits); 1876 cpu_hyp_init_features(); 1877 1878 /* 1879 * The stub hypercalls are now disabled, so set our local flag to 1880 * prevent a later re-init attempt in kvm_arch_hardware_enable(). 1881 */ 1882 __this_cpu_write(kvm_arm_hardware_enabled, 1); 1883 preempt_enable(); 1884 1885 return ret; 1886 } 1887 1888 static void kvm_hyp_init_symbols(void) 1889 { 1890 kvm_nvhe_sym(id_aa64pfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR0_EL1); 1891 kvm_nvhe_sym(id_aa64pfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64PFR1_EL1); 1892 kvm_nvhe_sym(id_aa64isar0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR0_EL1); 1893 kvm_nvhe_sym(id_aa64isar1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR1_EL1); 1894 kvm_nvhe_sym(id_aa64isar2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64ISAR2_EL1); 1895 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1); 1896 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); 1897 kvm_nvhe_sym(id_aa64mmfr2_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR2_EL1); 1898 kvm_nvhe_sym(__icache_flags) = __icache_flags; 1899 kvm_nvhe_sym(kvm_arm_vmid_bits) = kvm_arm_vmid_bits; 1900 } 1901 1902 static int kvm_hyp_init_protection(u32 hyp_va_bits) 1903 { 1904 void *addr = phys_to_virt(hyp_mem_base); 1905 int ret; 1906 1907 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP); 1908 if (ret) 1909 return ret; 1910 1911 ret = do_pkvm_init(hyp_va_bits); 1912 if (ret) 1913 return ret; 1914 1915 free_hyp_pgds(); 1916 1917 return 0; 1918 } 1919 1920 /** 1921 * Inits Hyp-mode on all online CPUs 1922 */ 1923 static int init_hyp_mode(void) 1924 { 1925 u32 hyp_va_bits; 1926 int cpu; 1927 int err = -ENOMEM; 1928 1929 /* 1930 * The protected Hyp-mode cannot be initialized if the memory pool 1931 * allocation has failed. 1932 */ 1933 if (is_protected_kvm_enabled() && !hyp_mem_base) 1934 goto out_err; 1935 1936 /* 1937 * Allocate Hyp PGD and setup Hyp identity mapping 1938 */ 1939 err = kvm_mmu_init(&hyp_va_bits); 1940 if (err) 1941 goto out_err; 1942 1943 /* 1944 * Allocate stack pages for Hypervisor-mode 1945 */ 1946 for_each_possible_cpu(cpu) { 1947 unsigned long stack_page; 1948 1949 stack_page = __get_free_page(GFP_KERNEL); 1950 if (!stack_page) { 1951 err = -ENOMEM; 1952 goto out_err; 1953 } 1954 1955 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; 1956 } 1957 1958 /* 1959 * Allocate and initialize pages for Hypervisor-mode percpu regions. 1960 */ 1961 for_each_possible_cpu(cpu) { 1962 struct page *page; 1963 void *page_addr; 1964 1965 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order()); 1966 if (!page) { 1967 err = -ENOMEM; 1968 goto out_err; 1969 } 1970 1971 page_addr = page_address(page); 1972 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size()); 1973 kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu] = (unsigned long)page_addr; 1974 } 1975 1976 /* 1977 * Map the Hyp-code called directly from the host 1978 */ 1979 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), 1980 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); 1981 if (err) { 1982 kvm_err("Cannot map world-switch code\n"); 1983 goto out_err; 1984 } 1985 1986 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start), 1987 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO); 1988 if (err) { 1989 kvm_err("Cannot map .hyp.rodata section\n"); 1990 goto out_err; 1991 } 1992 1993 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), 1994 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); 1995 if (err) { 1996 kvm_err("Cannot map rodata section\n"); 1997 goto out_err; 1998 } 1999 2000 /* 2001 * .hyp.bss is guaranteed to be placed at the beginning of the .bss 2002 * section thanks to an assertion in the linker script. Map it RW and 2003 * the rest of .bss RO. 2004 */ 2005 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start), 2006 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP); 2007 if (err) { 2008 kvm_err("Cannot map hyp bss section: %d\n", err); 2009 goto out_err; 2010 } 2011 2012 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end), 2013 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); 2014 if (err) { 2015 kvm_err("Cannot map bss section\n"); 2016 goto out_err; 2017 } 2018 2019 /* 2020 * Map the Hyp stack pages 2021 */ 2022 for_each_possible_cpu(cpu) { 2023 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu); 2024 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); 2025 unsigned long hyp_addr; 2026 2027 /* 2028 * Allocate a contiguous HYP private VA range for the stack 2029 * and guard page. The allocation is also aligned based on 2030 * the order of its size. 2031 */ 2032 err = hyp_alloc_private_va_range(PAGE_SIZE * 2, &hyp_addr); 2033 if (err) { 2034 kvm_err("Cannot allocate hyp stack guard page\n"); 2035 goto out_err; 2036 } 2037 2038 /* 2039 * Since the stack grows downwards, map the stack to the page 2040 * at the higher address and leave the lower guard page 2041 * unbacked. 2042 * 2043 * Any valid stack address now has the PAGE_SHIFT bit as 1 2044 * and addresses corresponding to the guard page have the 2045 * PAGE_SHIFT bit as 0 - this is used for overflow detection. 2046 */ 2047 err = __create_hyp_mappings(hyp_addr + PAGE_SIZE, PAGE_SIZE, 2048 __pa(stack_page), PAGE_HYP); 2049 if (err) { 2050 kvm_err("Cannot map hyp stack\n"); 2051 goto out_err; 2052 } 2053 2054 /* 2055 * Save the stack PA in nvhe_init_params. This will be needed 2056 * to recreate the stack mapping in protected nVHE mode. 2057 * __hyp_pa() won't do the right thing there, since the stack 2058 * has been mapped in the flexible private VA space. 2059 */ 2060 params->stack_pa = __pa(stack_page); 2061 2062 params->stack_hyp_va = hyp_addr + (2 * PAGE_SIZE); 2063 } 2064 2065 for_each_possible_cpu(cpu) { 2066 char *percpu_begin = (char *)kvm_nvhe_sym(kvm_arm_hyp_percpu_base)[cpu]; 2067 char *percpu_end = percpu_begin + nvhe_percpu_size(); 2068 2069 /* Map Hyp percpu pages */ 2070 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP); 2071 if (err) { 2072 kvm_err("Cannot map hyp percpu region\n"); 2073 goto out_err; 2074 } 2075 2076 /* Prepare the CPU initialization parameters */ 2077 cpu_prepare_hyp_mode(cpu, hyp_va_bits); 2078 } 2079 2080 kvm_hyp_init_symbols(); 2081 2082 if (is_protected_kvm_enabled()) { 2083 init_cpu_logical_map(); 2084 2085 if (!init_psci_relay()) { 2086 err = -ENODEV; 2087 goto out_err; 2088 } 2089 2090 err = kvm_hyp_init_protection(hyp_va_bits); 2091 if (err) { 2092 kvm_err("Failed to init hyp memory protection\n"); 2093 goto out_err; 2094 } 2095 } 2096 2097 return 0; 2098 2099 out_err: 2100 teardown_hyp_mode(); 2101 kvm_err("error initializing Hyp mode: %d\n", err); 2102 return err; 2103 } 2104 2105 static void _kvm_host_prot_finalize(void *arg) 2106 { 2107 int *err = arg; 2108 2109 if (WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize))) 2110 WRITE_ONCE(*err, -EINVAL); 2111 } 2112 2113 static int pkvm_drop_host_privileges(void) 2114 { 2115 int ret = 0; 2116 2117 /* 2118 * Flip the static key upfront as that may no longer be possible 2119 * once the host stage 2 is installed. 2120 */ 2121 static_branch_enable(&kvm_protected_mode_initialized); 2122 on_each_cpu(_kvm_host_prot_finalize, &ret, 1); 2123 return ret; 2124 } 2125 2126 static int finalize_hyp_mode(void) 2127 { 2128 if (!is_protected_kvm_enabled()) 2129 return 0; 2130 2131 /* 2132 * Exclude HYP sections from kmemleak so that they don't get peeked 2133 * at, which would end badly once inaccessible. 2134 */ 2135 kmemleak_free_part(__hyp_bss_start, __hyp_bss_end - __hyp_bss_start); 2136 kmemleak_free_part_phys(hyp_mem_base, hyp_mem_size); 2137 return pkvm_drop_host_privileges(); 2138 } 2139 2140 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) 2141 { 2142 struct kvm_vcpu *vcpu; 2143 unsigned long i; 2144 2145 mpidr &= MPIDR_HWID_BITMASK; 2146 kvm_for_each_vcpu(i, vcpu, kvm) { 2147 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) 2148 return vcpu; 2149 } 2150 return NULL; 2151 } 2152 2153 bool kvm_arch_irqchip_in_kernel(struct kvm *kvm) 2154 { 2155 return irqchip_in_kernel(kvm); 2156 } 2157 2158 bool kvm_arch_has_irq_bypass(void) 2159 { 2160 return true; 2161 } 2162 2163 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 2164 struct irq_bypass_producer *prod) 2165 { 2166 struct kvm_kernel_irqfd *irqfd = 2167 container_of(cons, struct kvm_kernel_irqfd, consumer); 2168 2169 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, 2170 &irqfd->irq_entry); 2171 } 2172 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 2173 struct irq_bypass_producer *prod) 2174 { 2175 struct kvm_kernel_irqfd *irqfd = 2176 container_of(cons, struct kvm_kernel_irqfd, consumer); 2177 2178 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, 2179 &irqfd->irq_entry); 2180 } 2181 2182 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) 2183 { 2184 struct kvm_kernel_irqfd *irqfd = 2185 container_of(cons, struct kvm_kernel_irqfd, consumer); 2186 2187 kvm_arm_halt_guest(irqfd->kvm); 2188 } 2189 2190 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) 2191 { 2192 struct kvm_kernel_irqfd *irqfd = 2193 container_of(cons, struct kvm_kernel_irqfd, consumer); 2194 2195 kvm_arm_resume_guest(irqfd->kvm); 2196 } 2197 2198 /** 2199 * Initialize Hyp-mode and memory mappings on all CPUs. 2200 */ 2201 int kvm_arch_init(void *opaque) 2202 { 2203 int err; 2204 bool in_hyp_mode; 2205 2206 if (!is_hyp_mode_available()) { 2207 kvm_info("HYP mode not available\n"); 2208 return -ENODEV; 2209 } 2210 2211 if (kvm_get_mode() == KVM_MODE_NONE) { 2212 kvm_info("KVM disabled from command line\n"); 2213 return -ENODEV; 2214 } 2215 2216 err = kvm_sys_reg_table_init(); 2217 if (err) { 2218 kvm_info("Error initializing system register tables"); 2219 return err; 2220 } 2221 2222 in_hyp_mode = is_kernel_in_hyp_mode(); 2223 2224 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) || 2225 cpus_have_final_cap(ARM64_WORKAROUND_1508412)) 2226 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \ 2227 "Only trusted guests should be used on this system.\n"); 2228 2229 err = kvm_set_ipa_limit(); 2230 if (err) 2231 return err; 2232 2233 err = kvm_arm_init_sve(); 2234 if (err) 2235 return err; 2236 2237 err = kvm_arm_vmid_alloc_init(); 2238 if (err) { 2239 kvm_err("Failed to initialize VMID allocator.\n"); 2240 return err; 2241 } 2242 2243 if (!in_hyp_mode) { 2244 err = init_hyp_mode(); 2245 if (err) 2246 goto out_err; 2247 } 2248 2249 err = kvm_init_vector_slots(); 2250 if (err) { 2251 kvm_err("Cannot initialise vector slots\n"); 2252 goto out_err; 2253 } 2254 2255 err = init_subsystems(); 2256 if (err) 2257 goto out_hyp; 2258 2259 if (!in_hyp_mode) { 2260 err = finalize_hyp_mode(); 2261 if (err) { 2262 kvm_err("Failed to finalize Hyp protection\n"); 2263 goto out_hyp; 2264 } 2265 } 2266 2267 if (is_protected_kvm_enabled()) { 2268 kvm_info("Protected nVHE mode initialized successfully\n"); 2269 } else if (in_hyp_mode) { 2270 kvm_info("VHE mode initialized successfully\n"); 2271 } else { 2272 kvm_info("Hyp mode initialized successfully\n"); 2273 } 2274 2275 return 0; 2276 2277 out_hyp: 2278 hyp_cpu_pm_exit(); 2279 if (!in_hyp_mode) 2280 teardown_hyp_mode(); 2281 out_err: 2282 kvm_arm_vmid_alloc_free(); 2283 return err; 2284 } 2285 2286 /* NOP: Compiling as a module not supported */ 2287 void kvm_arch_exit(void) 2288 { 2289 kvm_unregister_perf_callbacks(); 2290 } 2291 2292 static int __init early_kvm_mode_cfg(char *arg) 2293 { 2294 if (!arg) 2295 return -EINVAL; 2296 2297 if (strcmp(arg, "none") == 0) { 2298 kvm_mode = KVM_MODE_NONE; 2299 return 0; 2300 } 2301 2302 if (!is_hyp_mode_available()) { 2303 pr_warn_once("KVM is not available. Ignoring kvm-arm.mode\n"); 2304 return 0; 2305 } 2306 2307 if (strcmp(arg, "protected") == 0) { 2308 if (!is_kernel_in_hyp_mode()) 2309 kvm_mode = KVM_MODE_PROTECTED; 2310 else 2311 pr_warn_once("Protected KVM not available with VHE\n"); 2312 2313 return 0; 2314 } 2315 2316 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) { 2317 kvm_mode = KVM_MODE_DEFAULT; 2318 return 0; 2319 } 2320 2321 return -EINVAL; 2322 } 2323 early_param("kvm-arm.mode", early_kvm_mode_cfg); 2324 2325 enum kvm_mode kvm_get_mode(void) 2326 { 2327 return kvm_mode; 2328 } 2329 2330 static int arm_init(void) 2331 { 2332 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); 2333 return rc; 2334 } 2335 2336 module_init(arm_init); 2337