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/errno.h> 10 #include <linux/err.h> 11 #include <linux/kvm_host.h> 12 #include <linux/list.h> 13 #include <linux/module.h> 14 #include <linux/vmalloc.h> 15 #include <linux/fs.h> 16 #include <linux/mman.h> 17 #include <linux/sched.h> 18 #include <linux/kvm.h> 19 #include <linux/kvm_irqfd.h> 20 #include <linux/irqbypass.h> 21 #include <linux/sched/stat.h> 22 #include <linux/psci.h> 23 #include <trace/events/kvm.h> 24 25 #define CREATE_TRACE_POINTS 26 #include "trace_arm.h" 27 28 #include <linux/uaccess.h> 29 #include <asm/ptrace.h> 30 #include <asm/mman.h> 31 #include <asm/tlbflush.h> 32 #include <asm/cacheflush.h> 33 #include <asm/cpufeature.h> 34 #include <asm/virt.h> 35 #include <asm/kvm_arm.h> 36 #include <asm/kvm_asm.h> 37 #include <asm/kvm_mmu.h> 38 #include <asm/kvm_emulate.h> 39 #include <asm/sections.h> 40 41 #include <kvm/arm_hypercalls.h> 42 #include <kvm/arm_pmu.h> 43 #include <kvm/arm_psci.h> 44 45 #ifdef REQUIRES_VIRT 46 __asm__(".arch_extension virt"); 47 #endif 48 49 static enum kvm_mode kvm_mode = KVM_MODE_DEFAULT; 50 DEFINE_STATIC_KEY_FALSE(kvm_protected_mode_initialized); 51 52 DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector); 53 54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); 55 unsigned long kvm_arm_hyp_percpu_base[NR_CPUS]; 56 DECLARE_KVM_NVHE_PER_CPU(struct kvm_nvhe_init_params, kvm_init_params); 57 58 /* The VMID used in the VTTBR */ 59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); 60 static u32 kvm_next_vmid; 61 static DEFINE_SPINLOCK(kvm_vmid_lock); 62 63 static bool vgic_present; 64 65 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); 66 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); 67 68 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 69 { 70 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; 71 } 72 73 int kvm_arch_hardware_setup(void *opaque) 74 { 75 return 0; 76 } 77 78 int kvm_arch_check_processor_compat(void *opaque) 79 { 80 return 0; 81 } 82 83 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 84 struct kvm_enable_cap *cap) 85 { 86 int r; 87 88 if (cap->flags) 89 return -EINVAL; 90 91 switch (cap->cap) { 92 case KVM_CAP_ARM_NISV_TO_USER: 93 r = 0; 94 kvm->arch.return_nisv_io_abort_to_user = true; 95 break; 96 default: 97 r = -EINVAL; 98 break; 99 } 100 101 return r; 102 } 103 104 static int kvm_arm_default_max_vcpus(void) 105 { 106 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; 107 } 108 109 static void set_default_spectre(struct kvm *kvm) 110 { 111 /* 112 * The default is to expose CSV2 == 1 if the HW isn't affected. 113 * Although this is a per-CPU feature, we make it global because 114 * asymmetric systems are just a nuisance. 115 * 116 * Userspace can override this as long as it doesn't promise 117 * the impossible. 118 */ 119 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) 120 kvm->arch.pfr0_csv2 = 1; 121 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) 122 kvm->arch.pfr0_csv3 = 1; 123 } 124 125 /** 126 * kvm_arch_init_vm - initializes a VM data structure 127 * @kvm: pointer to the KVM struct 128 */ 129 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 130 { 131 int ret; 132 133 ret = kvm_arm_setup_stage2(kvm, type); 134 if (ret) 135 return ret; 136 137 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu); 138 if (ret) 139 return ret; 140 141 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); 142 if (ret) 143 goto out_free_stage2_pgd; 144 145 kvm_vgic_early_init(kvm); 146 147 /* The maximum number of VCPUs is limited by the host's GIC model */ 148 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus(); 149 150 set_default_spectre(kvm); 151 152 return ret; 153 out_free_stage2_pgd: 154 kvm_free_stage2_pgd(&kvm->arch.mmu); 155 return ret; 156 } 157 158 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 159 { 160 return VM_FAULT_SIGBUS; 161 } 162 163 164 /** 165 * kvm_arch_destroy_vm - destroy the VM data structure 166 * @kvm: pointer to the KVM struct 167 */ 168 void kvm_arch_destroy_vm(struct kvm *kvm) 169 { 170 int i; 171 172 bitmap_free(kvm->arch.pmu_filter); 173 174 kvm_vgic_destroy(kvm); 175 176 for (i = 0; i < KVM_MAX_VCPUS; ++i) { 177 if (kvm->vcpus[i]) { 178 kvm_vcpu_destroy(kvm->vcpus[i]); 179 kvm->vcpus[i] = NULL; 180 } 181 } 182 atomic_set(&kvm->online_vcpus, 0); 183 } 184 185 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 186 { 187 int r; 188 switch (ext) { 189 case KVM_CAP_IRQCHIP: 190 r = vgic_present; 191 break; 192 case KVM_CAP_IOEVENTFD: 193 case KVM_CAP_DEVICE_CTRL: 194 case KVM_CAP_USER_MEMORY: 195 case KVM_CAP_SYNC_MMU: 196 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: 197 case KVM_CAP_ONE_REG: 198 case KVM_CAP_ARM_PSCI: 199 case KVM_CAP_ARM_PSCI_0_2: 200 case KVM_CAP_READONLY_MEM: 201 case KVM_CAP_MP_STATE: 202 case KVM_CAP_IMMEDIATE_EXIT: 203 case KVM_CAP_VCPU_EVENTS: 204 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: 205 case KVM_CAP_ARM_NISV_TO_USER: 206 case KVM_CAP_ARM_INJECT_EXT_DABT: 207 case KVM_CAP_SET_GUEST_DEBUG: 208 case KVM_CAP_VCPU_ATTRIBUTES: 209 case KVM_CAP_PTP_KVM: 210 r = 1; 211 break; 212 case KVM_CAP_SET_GUEST_DEBUG2: 213 return KVM_GUESTDBG_VALID_MASK; 214 case KVM_CAP_ARM_SET_DEVICE_ADDR: 215 r = 1; 216 break; 217 case KVM_CAP_NR_VCPUS: 218 r = num_online_cpus(); 219 break; 220 case KVM_CAP_MAX_VCPUS: 221 case KVM_CAP_MAX_VCPU_ID: 222 if (kvm) 223 r = kvm->arch.max_vcpus; 224 else 225 r = kvm_arm_default_max_vcpus(); 226 break; 227 case KVM_CAP_MSI_DEVID: 228 if (!kvm) 229 r = -EINVAL; 230 else 231 r = kvm->arch.vgic.msis_require_devid; 232 break; 233 case KVM_CAP_ARM_USER_IRQ: 234 /* 235 * 1: EL1_VTIMER, EL1_PTIMER, and PMU. 236 * (bump this number if adding more devices) 237 */ 238 r = 1; 239 break; 240 case KVM_CAP_STEAL_TIME: 241 r = kvm_arm_pvtime_supported(); 242 break; 243 case KVM_CAP_ARM_EL1_32BIT: 244 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1); 245 break; 246 case KVM_CAP_GUEST_DEBUG_HW_BPS: 247 r = get_num_brps(); 248 break; 249 case KVM_CAP_GUEST_DEBUG_HW_WPS: 250 r = get_num_wrps(); 251 break; 252 case KVM_CAP_ARM_PMU_V3: 253 r = kvm_arm_support_pmu_v3(); 254 break; 255 case KVM_CAP_ARM_INJECT_SERROR_ESR: 256 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN); 257 break; 258 case KVM_CAP_ARM_VM_IPA_SIZE: 259 r = get_kvm_ipa_limit(); 260 break; 261 case KVM_CAP_ARM_SVE: 262 r = system_supports_sve(); 263 break; 264 case KVM_CAP_ARM_PTRAUTH_ADDRESS: 265 case KVM_CAP_ARM_PTRAUTH_GENERIC: 266 r = system_has_full_ptr_auth(); 267 break; 268 default: 269 r = 0; 270 } 271 272 return r; 273 } 274 275 long kvm_arch_dev_ioctl(struct file *filp, 276 unsigned int ioctl, unsigned long arg) 277 { 278 return -EINVAL; 279 } 280 281 struct kvm *kvm_arch_alloc_vm(void) 282 { 283 if (!has_vhe()) 284 return kzalloc(sizeof(struct kvm), GFP_KERNEL); 285 286 return vzalloc(sizeof(struct kvm)); 287 } 288 289 void kvm_arch_free_vm(struct kvm *kvm) 290 { 291 if (!has_vhe()) 292 kfree(kvm); 293 else 294 vfree(kvm); 295 } 296 297 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 298 { 299 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) 300 return -EBUSY; 301 302 if (id >= kvm->arch.max_vcpus) 303 return -EINVAL; 304 305 return 0; 306 } 307 308 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 309 { 310 int err; 311 312 /* Force users to call KVM_ARM_VCPU_INIT */ 313 vcpu->arch.target = -1; 314 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 315 316 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO; 317 318 /* Set up the timer */ 319 kvm_timer_vcpu_init(vcpu); 320 321 kvm_pmu_vcpu_init(vcpu); 322 323 kvm_arm_reset_debug_ptr(vcpu); 324 325 kvm_arm_pvtime_vcpu_init(&vcpu->arch); 326 327 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu; 328 329 err = kvm_vgic_vcpu_init(vcpu); 330 if (err) 331 return err; 332 333 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); 334 } 335 336 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 337 { 338 } 339 340 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 341 { 342 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm))) 343 static_branch_dec(&userspace_irqchip_in_use); 344 345 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); 346 kvm_timer_vcpu_terminate(vcpu); 347 kvm_pmu_vcpu_destroy(vcpu); 348 349 kvm_arm_vcpu_destroy(vcpu); 350 } 351 352 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 353 { 354 return kvm_timer_is_pending(vcpu); 355 } 356 357 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) 358 { 359 /* 360 * If we're about to block (most likely because we've just hit a 361 * WFI), we need to sync back the state of the GIC CPU interface 362 * so that we have the latest PMR and group enables. This ensures 363 * that kvm_arch_vcpu_runnable has up-to-date data to decide 364 * whether we have pending interrupts. 365 * 366 * For the same reason, we want to tell GICv4 that we need 367 * doorbells to be signalled, should an interrupt become pending. 368 */ 369 preempt_disable(); 370 kvm_vgic_vmcr_sync(vcpu); 371 vgic_v4_put(vcpu, true); 372 preempt_enable(); 373 } 374 375 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) 376 { 377 preempt_disable(); 378 vgic_v4_load(vcpu); 379 preempt_enable(); 380 } 381 382 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 383 { 384 struct kvm_s2_mmu *mmu; 385 int *last_ran; 386 387 mmu = vcpu->arch.hw_mmu; 388 last_ran = this_cpu_ptr(mmu->last_vcpu_ran); 389 390 /* 391 * We guarantee that both TLBs and I-cache are private to each 392 * vcpu. If detecting that a vcpu from the same VM has 393 * previously run on the same physical CPU, call into the 394 * hypervisor code to nuke the relevant contexts. 395 * 396 * We might get preempted before the vCPU actually runs, but 397 * over-invalidation doesn't affect correctness. 398 */ 399 if (*last_ran != vcpu->vcpu_id) { 400 kvm_call_hyp(__kvm_flush_cpu_context, mmu); 401 *last_ran = vcpu->vcpu_id; 402 } 403 404 vcpu->cpu = cpu; 405 406 kvm_vgic_load(vcpu); 407 kvm_timer_vcpu_load(vcpu); 408 if (has_vhe()) 409 kvm_vcpu_load_sysregs_vhe(vcpu); 410 kvm_arch_vcpu_load_fp(vcpu); 411 kvm_vcpu_pmu_restore_guest(vcpu); 412 if (kvm_arm_is_pvtime_enabled(&vcpu->arch)) 413 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu); 414 415 if (single_task_running()) 416 vcpu_clear_wfx_traps(vcpu); 417 else 418 vcpu_set_wfx_traps(vcpu); 419 420 if (vcpu_has_ptrauth(vcpu)) 421 vcpu_ptrauth_disable(vcpu); 422 kvm_arch_vcpu_load_debug_state_flags(vcpu); 423 } 424 425 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 426 { 427 kvm_arch_vcpu_put_debug_state_flags(vcpu); 428 kvm_arch_vcpu_put_fp(vcpu); 429 if (has_vhe()) 430 kvm_vcpu_put_sysregs_vhe(vcpu); 431 kvm_timer_vcpu_put(vcpu); 432 kvm_vgic_put(vcpu); 433 kvm_vcpu_pmu_restore_host(vcpu); 434 435 vcpu->cpu = -1; 436 } 437 438 static void vcpu_power_off(struct kvm_vcpu *vcpu) 439 { 440 vcpu->arch.power_off = true; 441 kvm_make_request(KVM_REQ_SLEEP, vcpu); 442 kvm_vcpu_kick(vcpu); 443 } 444 445 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 446 struct kvm_mp_state *mp_state) 447 { 448 if (vcpu->arch.power_off) 449 mp_state->mp_state = KVM_MP_STATE_STOPPED; 450 else 451 mp_state->mp_state = KVM_MP_STATE_RUNNABLE; 452 453 return 0; 454 } 455 456 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 457 struct kvm_mp_state *mp_state) 458 { 459 int ret = 0; 460 461 switch (mp_state->mp_state) { 462 case KVM_MP_STATE_RUNNABLE: 463 vcpu->arch.power_off = false; 464 break; 465 case KVM_MP_STATE_STOPPED: 466 vcpu_power_off(vcpu); 467 break; 468 default: 469 ret = -EINVAL; 470 } 471 472 return ret; 473 } 474 475 /** 476 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled 477 * @v: The VCPU pointer 478 * 479 * If the guest CPU is not waiting for interrupts or an interrupt line is 480 * asserted, the CPU is by definition runnable. 481 */ 482 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 483 { 484 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); 485 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) 486 && !v->arch.power_off && !v->arch.pause); 487 } 488 489 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 490 { 491 return vcpu_mode_priv(vcpu); 492 } 493 494 /* Just ensure a guest exit from a particular CPU */ 495 static void exit_vm_noop(void *info) 496 { 497 } 498 499 void force_vm_exit(const cpumask_t *mask) 500 { 501 preempt_disable(); 502 smp_call_function_many(mask, exit_vm_noop, NULL, true); 503 preempt_enable(); 504 } 505 506 /** 507 * need_new_vmid_gen - check that the VMID is still valid 508 * @vmid: The VMID to check 509 * 510 * return true if there is a new generation of VMIDs being used 511 * 512 * The hardware supports a limited set of values with the value zero reserved 513 * for the host, so we check if an assigned value belongs to a previous 514 * generation, which requires us to assign a new value. If we're the first to 515 * use a VMID for the new generation, we must flush necessary caches and TLBs 516 * on all CPUs. 517 */ 518 static bool need_new_vmid_gen(struct kvm_vmid *vmid) 519 { 520 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen); 521 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */ 522 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen); 523 } 524 525 /** 526 * update_vmid - Update the vmid with a valid VMID for the current generation 527 * @vmid: The stage-2 VMID information struct 528 */ 529 static void update_vmid(struct kvm_vmid *vmid) 530 { 531 if (!need_new_vmid_gen(vmid)) 532 return; 533 534 spin_lock(&kvm_vmid_lock); 535 536 /* 537 * We need to re-check the vmid_gen here to ensure that if another vcpu 538 * already allocated a valid vmid for this vm, then this vcpu should 539 * use the same vmid. 540 */ 541 if (!need_new_vmid_gen(vmid)) { 542 spin_unlock(&kvm_vmid_lock); 543 return; 544 } 545 546 /* First user of a new VMID generation? */ 547 if (unlikely(kvm_next_vmid == 0)) { 548 atomic64_inc(&kvm_vmid_gen); 549 kvm_next_vmid = 1; 550 551 /* 552 * On SMP we know no other CPUs can use this CPU's or each 553 * other's VMID after force_vm_exit returns since the 554 * kvm_vmid_lock blocks them from reentry to the guest. 555 */ 556 force_vm_exit(cpu_all_mask); 557 /* 558 * Now broadcast TLB + ICACHE invalidation over the inner 559 * shareable domain to make sure all data structures are 560 * clean. 561 */ 562 kvm_call_hyp(__kvm_flush_vm_context); 563 } 564 565 vmid->vmid = kvm_next_vmid; 566 kvm_next_vmid++; 567 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1; 568 569 smp_wmb(); 570 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen)); 571 572 spin_unlock(&kvm_vmid_lock); 573 } 574 575 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) 576 { 577 struct kvm *kvm = vcpu->kvm; 578 int ret = 0; 579 580 if (likely(vcpu->arch.has_run_once)) 581 return 0; 582 583 if (!kvm_arm_vcpu_is_finalized(vcpu)) 584 return -EPERM; 585 586 vcpu->arch.has_run_once = true; 587 588 kvm_arm_vcpu_init_debug(vcpu); 589 590 if (likely(irqchip_in_kernel(kvm))) { 591 /* 592 * Map the VGIC hardware resources before running a vcpu the 593 * first time on this VM. 594 */ 595 ret = kvm_vgic_map_resources(kvm); 596 if (ret) 597 return ret; 598 } else { 599 /* 600 * Tell the rest of the code that there are userspace irqchip 601 * VMs in the wild. 602 */ 603 static_branch_inc(&userspace_irqchip_in_use); 604 } 605 606 ret = kvm_timer_enable(vcpu); 607 if (ret) 608 return ret; 609 610 ret = kvm_arm_pmu_v3_enable(vcpu); 611 612 return ret; 613 } 614 615 bool kvm_arch_intc_initialized(struct kvm *kvm) 616 { 617 return vgic_initialized(kvm); 618 } 619 620 void kvm_arm_halt_guest(struct kvm *kvm) 621 { 622 int i; 623 struct kvm_vcpu *vcpu; 624 625 kvm_for_each_vcpu(i, vcpu, kvm) 626 vcpu->arch.pause = true; 627 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); 628 } 629 630 void kvm_arm_resume_guest(struct kvm *kvm) 631 { 632 int i; 633 struct kvm_vcpu *vcpu; 634 635 kvm_for_each_vcpu(i, vcpu, kvm) { 636 vcpu->arch.pause = false; 637 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu)); 638 } 639 } 640 641 static void vcpu_req_sleep(struct kvm_vcpu *vcpu) 642 { 643 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 644 645 rcuwait_wait_event(wait, 646 (!vcpu->arch.power_off) &&(!vcpu->arch.pause), 647 TASK_INTERRUPTIBLE); 648 649 if (vcpu->arch.power_off || vcpu->arch.pause) { 650 /* Awaken to handle a signal, request we sleep again later. */ 651 kvm_make_request(KVM_REQ_SLEEP, vcpu); 652 } 653 654 /* 655 * Make sure we will observe a potential reset request if we've 656 * observed a change to the power state. Pairs with the smp_wmb() in 657 * kvm_psci_vcpu_on(). 658 */ 659 smp_rmb(); 660 } 661 662 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) 663 { 664 return vcpu->arch.target >= 0; 665 } 666 667 static void check_vcpu_requests(struct kvm_vcpu *vcpu) 668 { 669 if (kvm_request_pending(vcpu)) { 670 if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) 671 vcpu_req_sleep(vcpu); 672 673 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 674 kvm_reset_vcpu(vcpu); 675 676 /* 677 * Clear IRQ_PENDING requests that were made to guarantee 678 * that a VCPU sees new virtual interrupts. 679 */ 680 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); 681 682 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu)) 683 kvm_update_stolen_time(vcpu); 684 685 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) { 686 /* The distributor enable bits were changed */ 687 preempt_disable(); 688 vgic_v4_put(vcpu, false); 689 vgic_v4_load(vcpu); 690 preempt_enable(); 691 } 692 } 693 } 694 695 static bool vcpu_mode_is_bad_32bit(struct kvm_vcpu *vcpu) 696 { 697 if (likely(!vcpu_mode_is_32bit(vcpu))) 698 return false; 699 700 return !system_supports_32bit_el0() || 701 static_branch_unlikely(&arm64_mismatched_32bit_el0); 702 } 703 704 /** 705 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code 706 * @vcpu: The VCPU pointer 707 * 708 * This function is called through the VCPU_RUN ioctl called from user space. It 709 * will execute VM code in a loop until the time slice for the process is used 710 * or some emulation is needed from user space in which case the function will 711 * return with return value 0 and with the kvm_run structure filled in with the 712 * required data for the requested emulation. 713 */ 714 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 715 { 716 struct kvm_run *run = vcpu->run; 717 int ret; 718 719 if (unlikely(!kvm_vcpu_initialized(vcpu))) 720 return -ENOEXEC; 721 722 ret = kvm_vcpu_first_run_init(vcpu); 723 if (ret) 724 return ret; 725 726 if (run->exit_reason == KVM_EXIT_MMIO) { 727 ret = kvm_handle_mmio_return(vcpu); 728 if (ret) 729 return ret; 730 } 731 732 if (run->immediate_exit) 733 return -EINTR; 734 735 vcpu_load(vcpu); 736 737 kvm_sigset_activate(vcpu); 738 739 ret = 1; 740 run->exit_reason = KVM_EXIT_UNKNOWN; 741 while (ret > 0) { 742 /* 743 * Check conditions before entering the guest 744 */ 745 cond_resched(); 746 747 update_vmid(&vcpu->arch.hw_mmu->vmid); 748 749 check_vcpu_requests(vcpu); 750 751 /* 752 * Preparing the interrupts to be injected also 753 * involves poking the GIC, which must be done in a 754 * non-preemptible context. 755 */ 756 preempt_disable(); 757 758 kvm_pmu_flush_hwstate(vcpu); 759 760 local_irq_disable(); 761 762 kvm_vgic_flush_hwstate(vcpu); 763 764 /* 765 * Exit if we have a signal pending so that we can deliver the 766 * signal to user space. 767 */ 768 if (signal_pending(current)) { 769 ret = -EINTR; 770 run->exit_reason = KVM_EXIT_INTR; 771 } 772 773 /* 774 * If we're using a userspace irqchip, then check if we need 775 * to tell a userspace irqchip about timer or PMU level 776 * changes and if so, exit to userspace (the actual level 777 * state gets updated in kvm_timer_update_run and 778 * kvm_pmu_update_run below). 779 */ 780 if (static_branch_unlikely(&userspace_irqchip_in_use)) { 781 if (kvm_timer_should_notify_user(vcpu) || 782 kvm_pmu_should_notify_user(vcpu)) { 783 ret = -EINTR; 784 run->exit_reason = KVM_EXIT_INTR; 785 } 786 } 787 788 /* 789 * Ensure we set mode to IN_GUEST_MODE after we disable 790 * interrupts and before the final VCPU requests check. 791 * See the comment in kvm_vcpu_exiting_guest_mode() and 792 * Documentation/virt/kvm/vcpu-requests.rst 793 */ 794 smp_store_mb(vcpu->mode, IN_GUEST_MODE); 795 796 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) || 797 kvm_request_pending(vcpu)) { 798 vcpu->mode = OUTSIDE_GUEST_MODE; 799 isb(); /* Ensure work in x_flush_hwstate is committed */ 800 kvm_pmu_sync_hwstate(vcpu); 801 if (static_branch_unlikely(&userspace_irqchip_in_use)) 802 kvm_timer_sync_user(vcpu); 803 kvm_vgic_sync_hwstate(vcpu); 804 local_irq_enable(); 805 preempt_enable(); 806 continue; 807 } 808 809 kvm_arm_setup_debug(vcpu); 810 811 /************************************************************** 812 * Enter the guest 813 */ 814 trace_kvm_entry(*vcpu_pc(vcpu)); 815 guest_enter_irqoff(); 816 817 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu); 818 819 vcpu->mode = OUTSIDE_GUEST_MODE; 820 vcpu->stat.exits++; 821 /* 822 * Back from guest 823 *************************************************************/ 824 825 kvm_arm_clear_debug(vcpu); 826 827 /* 828 * We must sync the PMU state before the vgic state so 829 * that the vgic can properly sample the updated state of the 830 * interrupt line. 831 */ 832 kvm_pmu_sync_hwstate(vcpu); 833 834 /* 835 * Sync the vgic state before syncing the timer state because 836 * the timer code needs to know if the virtual timer 837 * interrupts are active. 838 */ 839 kvm_vgic_sync_hwstate(vcpu); 840 841 /* 842 * Sync the timer hardware state before enabling interrupts as 843 * we don't want vtimer interrupts to race with syncing the 844 * timer virtual interrupt state. 845 */ 846 if (static_branch_unlikely(&userspace_irqchip_in_use)) 847 kvm_timer_sync_user(vcpu); 848 849 kvm_arch_vcpu_ctxsync_fp(vcpu); 850 851 /* 852 * We may have taken a host interrupt in HYP mode (ie 853 * while executing the guest). This interrupt is still 854 * pending, as we haven't serviced it yet! 855 * 856 * We're now back in SVC mode, with interrupts 857 * disabled. Enabling the interrupts now will have 858 * the effect of taking the interrupt again, in SVC 859 * mode this time. 860 */ 861 local_irq_enable(); 862 863 /* 864 * We do local_irq_enable() before calling guest_exit() so 865 * that if a timer interrupt hits while running the guest we 866 * account that tick as being spent in the guest. We enable 867 * preemption after calling guest_exit() so that if we get 868 * preempted we make sure ticks after that is not counted as 869 * guest time. 870 */ 871 guest_exit(); 872 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); 873 874 /* Exit types that need handling before we can be preempted */ 875 handle_exit_early(vcpu, ret); 876 877 preempt_enable(); 878 879 /* 880 * The ARMv8 architecture doesn't give the hypervisor 881 * a mechanism to prevent a guest from dropping to AArch32 EL0 882 * if implemented by the CPU. If we spot the guest in such 883 * state and that we decided it wasn't supposed to do so (like 884 * with the asymmetric AArch32 case), return to userspace with 885 * a fatal error. 886 */ 887 if (vcpu_mode_is_bad_32bit(vcpu)) { 888 /* 889 * As we have caught the guest red-handed, decide that 890 * it isn't fit for purpose anymore by making the vcpu 891 * invalid. The VMM can try and fix it by issuing a 892 * KVM_ARM_VCPU_INIT if it really wants to. 893 */ 894 vcpu->arch.target = -1; 895 ret = ARM_EXCEPTION_IL; 896 } 897 898 ret = handle_exit(vcpu, ret); 899 } 900 901 /* Tell userspace about in-kernel device output levels */ 902 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { 903 kvm_timer_update_run(vcpu); 904 kvm_pmu_update_run(vcpu); 905 } 906 907 kvm_sigset_deactivate(vcpu); 908 909 vcpu_put(vcpu); 910 return ret; 911 } 912 913 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) 914 { 915 int bit_index; 916 bool set; 917 unsigned long *hcr; 918 919 if (number == KVM_ARM_IRQ_CPU_IRQ) 920 bit_index = __ffs(HCR_VI); 921 else /* KVM_ARM_IRQ_CPU_FIQ */ 922 bit_index = __ffs(HCR_VF); 923 924 hcr = vcpu_hcr(vcpu); 925 if (level) 926 set = test_and_set_bit(bit_index, hcr); 927 else 928 set = test_and_clear_bit(bit_index, hcr); 929 930 /* 931 * If we didn't change anything, no need to wake up or kick other CPUs 932 */ 933 if (set == level) 934 return 0; 935 936 /* 937 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and 938 * trigger a world-switch round on the running physical CPU to set the 939 * virtual IRQ/FIQ fields in the HCR appropriately. 940 */ 941 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); 942 kvm_vcpu_kick(vcpu); 943 944 return 0; 945 } 946 947 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 948 bool line_status) 949 { 950 u32 irq = irq_level->irq; 951 unsigned int irq_type, vcpu_idx, irq_num; 952 int nrcpus = atomic_read(&kvm->online_vcpus); 953 struct kvm_vcpu *vcpu = NULL; 954 bool level = irq_level->level; 955 956 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; 957 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; 958 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); 959 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; 960 961 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); 962 963 switch (irq_type) { 964 case KVM_ARM_IRQ_TYPE_CPU: 965 if (irqchip_in_kernel(kvm)) 966 return -ENXIO; 967 968 if (vcpu_idx >= nrcpus) 969 return -EINVAL; 970 971 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 972 if (!vcpu) 973 return -EINVAL; 974 975 if (irq_num > KVM_ARM_IRQ_CPU_FIQ) 976 return -EINVAL; 977 978 return vcpu_interrupt_line(vcpu, irq_num, level); 979 case KVM_ARM_IRQ_TYPE_PPI: 980 if (!irqchip_in_kernel(kvm)) 981 return -ENXIO; 982 983 if (vcpu_idx >= nrcpus) 984 return -EINVAL; 985 986 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 987 if (!vcpu) 988 return -EINVAL; 989 990 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) 991 return -EINVAL; 992 993 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); 994 case KVM_ARM_IRQ_TYPE_SPI: 995 if (!irqchip_in_kernel(kvm)) 996 return -ENXIO; 997 998 if (irq_num < VGIC_NR_PRIVATE_IRQS) 999 return -EINVAL; 1000 1001 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); 1002 } 1003 1004 return -EINVAL; 1005 } 1006 1007 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, 1008 const struct kvm_vcpu_init *init) 1009 { 1010 unsigned int i, ret; 1011 int phys_target = kvm_target_cpu(); 1012 1013 if (init->target != phys_target) 1014 return -EINVAL; 1015 1016 /* 1017 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 1018 * use the same target. 1019 */ 1020 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) 1021 return -EINVAL; 1022 1023 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ 1024 for (i = 0; i < sizeof(init->features) * 8; i++) { 1025 bool set = (init->features[i / 32] & (1 << (i % 32))); 1026 1027 if (set && i >= KVM_VCPU_MAX_FEATURES) 1028 return -ENOENT; 1029 1030 /* 1031 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 1032 * use the same feature set. 1033 */ 1034 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && 1035 test_bit(i, vcpu->arch.features) != set) 1036 return -EINVAL; 1037 1038 if (set) 1039 set_bit(i, vcpu->arch.features); 1040 } 1041 1042 vcpu->arch.target = phys_target; 1043 1044 /* Now we know what it is, we can reset it. */ 1045 ret = kvm_reset_vcpu(vcpu); 1046 if (ret) { 1047 vcpu->arch.target = -1; 1048 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 1049 } 1050 1051 return ret; 1052 } 1053 1054 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, 1055 struct kvm_vcpu_init *init) 1056 { 1057 int ret; 1058 1059 ret = kvm_vcpu_set_target(vcpu, init); 1060 if (ret) 1061 return ret; 1062 1063 /* 1064 * Ensure a rebooted VM will fault in RAM pages and detect if the 1065 * guest MMU is turned off and flush the caches as needed. 1066 * 1067 * S2FWB enforces all memory accesses to RAM being cacheable, 1068 * ensuring that the data side is always coherent. We still 1069 * need to invalidate the I-cache though, as FWB does *not* 1070 * imply CTR_EL0.DIC. 1071 */ 1072 if (vcpu->arch.has_run_once) { 1073 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB)) 1074 stage2_unmap_vm(vcpu->kvm); 1075 else 1076 __flush_icache_all(); 1077 } 1078 1079 vcpu_reset_hcr(vcpu); 1080 1081 /* 1082 * Handle the "start in power-off" case. 1083 */ 1084 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) 1085 vcpu_power_off(vcpu); 1086 else 1087 vcpu->arch.power_off = false; 1088 1089 return 0; 1090 } 1091 1092 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, 1093 struct kvm_device_attr *attr) 1094 { 1095 int ret = -ENXIO; 1096 1097 switch (attr->group) { 1098 default: 1099 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); 1100 break; 1101 } 1102 1103 return ret; 1104 } 1105 1106 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, 1107 struct kvm_device_attr *attr) 1108 { 1109 int ret = -ENXIO; 1110 1111 switch (attr->group) { 1112 default: 1113 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); 1114 break; 1115 } 1116 1117 return ret; 1118 } 1119 1120 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, 1121 struct kvm_device_attr *attr) 1122 { 1123 int ret = -ENXIO; 1124 1125 switch (attr->group) { 1126 default: 1127 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); 1128 break; 1129 } 1130 1131 return ret; 1132 } 1133 1134 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1135 struct kvm_vcpu_events *events) 1136 { 1137 memset(events, 0, sizeof(*events)); 1138 1139 return __kvm_arm_vcpu_get_events(vcpu, events); 1140 } 1141 1142 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1143 struct kvm_vcpu_events *events) 1144 { 1145 int i; 1146 1147 /* check whether the reserved field is zero */ 1148 for (i = 0; i < ARRAY_SIZE(events->reserved); i++) 1149 if (events->reserved[i]) 1150 return -EINVAL; 1151 1152 /* check whether the pad field is zero */ 1153 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) 1154 if (events->exception.pad[i]) 1155 return -EINVAL; 1156 1157 return __kvm_arm_vcpu_set_events(vcpu, events); 1158 } 1159 1160 long kvm_arch_vcpu_ioctl(struct file *filp, 1161 unsigned int ioctl, unsigned long arg) 1162 { 1163 struct kvm_vcpu *vcpu = filp->private_data; 1164 void __user *argp = (void __user *)arg; 1165 struct kvm_device_attr attr; 1166 long r; 1167 1168 switch (ioctl) { 1169 case KVM_ARM_VCPU_INIT: { 1170 struct kvm_vcpu_init init; 1171 1172 r = -EFAULT; 1173 if (copy_from_user(&init, argp, sizeof(init))) 1174 break; 1175 1176 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); 1177 break; 1178 } 1179 case KVM_SET_ONE_REG: 1180 case KVM_GET_ONE_REG: { 1181 struct kvm_one_reg reg; 1182 1183 r = -ENOEXEC; 1184 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1185 break; 1186 1187 r = -EFAULT; 1188 if (copy_from_user(®, argp, sizeof(reg))) 1189 break; 1190 1191 if (ioctl == KVM_SET_ONE_REG) 1192 r = kvm_arm_set_reg(vcpu, ®); 1193 else 1194 r = kvm_arm_get_reg(vcpu, ®); 1195 break; 1196 } 1197 case KVM_GET_REG_LIST: { 1198 struct kvm_reg_list __user *user_list = argp; 1199 struct kvm_reg_list reg_list; 1200 unsigned n; 1201 1202 r = -ENOEXEC; 1203 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1204 break; 1205 1206 r = -EPERM; 1207 if (!kvm_arm_vcpu_is_finalized(vcpu)) 1208 break; 1209 1210 r = -EFAULT; 1211 if (copy_from_user(®_list, user_list, sizeof(reg_list))) 1212 break; 1213 n = reg_list.n; 1214 reg_list.n = kvm_arm_num_regs(vcpu); 1215 if (copy_to_user(user_list, ®_list, sizeof(reg_list))) 1216 break; 1217 r = -E2BIG; 1218 if (n < reg_list.n) 1219 break; 1220 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); 1221 break; 1222 } 1223 case KVM_SET_DEVICE_ATTR: { 1224 r = -EFAULT; 1225 if (copy_from_user(&attr, argp, sizeof(attr))) 1226 break; 1227 r = kvm_arm_vcpu_set_attr(vcpu, &attr); 1228 break; 1229 } 1230 case KVM_GET_DEVICE_ATTR: { 1231 r = -EFAULT; 1232 if (copy_from_user(&attr, argp, sizeof(attr))) 1233 break; 1234 r = kvm_arm_vcpu_get_attr(vcpu, &attr); 1235 break; 1236 } 1237 case KVM_HAS_DEVICE_ATTR: { 1238 r = -EFAULT; 1239 if (copy_from_user(&attr, argp, sizeof(attr))) 1240 break; 1241 r = kvm_arm_vcpu_has_attr(vcpu, &attr); 1242 break; 1243 } 1244 case KVM_GET_VCPU_EVENTS: { 1245 struct kvm_vcpu_events events; 1246 1247 if (kvm_arm_vcpu_get_events(vcpu, &events)) 1248 return -EINVAL; 1249 1250 if (copy_to_user(argp, &events, sizeof(events))) 1251 return -EFAULT; 1252 1253 return 0; 1254 } 1255 case KVM_SET_VCPU_EVENTS: { 1256 struct kvm_vcpu_events events; 1257 1258 if (copy_from_user(&events, argp, sizeof(events))) 1259 return -EFAULT; 1260 1261 return kvm_arm_vcpu_set_events(vcpu, &events); 1262 } 1263 case KVM_ARM_VCPU_FINALIZE: { 1264 int what; 1265 1266 if (!kvm_vcpu_initialized(vcpu)) 1267 return -ENOEXEC; 1268 1269 if (get_user(what, (const int __user *)argp)) 1270 return -EFAULT; 1271 1272 return kvm_arm_vcpu_finalize(vcpu, what); 1273 } 1274 default: 1275 r = -EINVAL; 1276 } 1277 1278 return r; 1279 } 1280 1281 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 1282 { 1283 1284 } 1285 1286 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm, 1287 const struct kvm_memory_slot *memslot) 1288 { 1289 kvm_flush_remote_tlbs(kvm); 1290 } 1291 1292 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, 1293 struct kvm_arm_device_addr *dev_addr) 1294 { 1295 unsigned long dev_id, type; 1296 1297 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> 1298 KVM_ARM_DEVICE_ID_SHIFT; 1299 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> 1300 KVM_ARM_DEVICE_TYPE_SHIFT; 1301 1302 switch (dev_id) { 1303 case KVM_ARM_DEVICE_VGIC_V2: 1304 if (!vgic_present) 1305 return -ENXIO; 1306 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); 1307 default: 1308 return -ENODEV; 1309 } 1310 } 1311 1312 long kvm_arch_vm_ioctl(struct file *filp, 1313 unsigned int ioctl, unsigned long arg) 1314 { 1315 struct kvm *kvm = filp->private_data; 1316 void __user *argp = (void __user *)arg; 1317 1318 switch (ioctl) { 1319 case KVM_CREATE_IRQCHIP: { 1320 int ret; 1321 if (!vgic_present) 1322 return -ENXIO; 1323 mutex_lock(&kvm->lock); 1324 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); 1325 mutex_unlock(&kvm->lock); 1326 return ret; 1327 } 1328 case KVM_ARM_SET_DEVICE_ADDR: { 1329 struct kvm_arm_device_addr dev_addr; 1330 1331 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) 1332 return -EFAULT; 1333 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); 1334 } 1335 case KVM_ARM_PREFERRED_TARGET: { 1336 int err; 1337 struct kvm_vcpu_init init; 1338 1339 err = kvm_vcpu_preferred_target(&init); 1340 if (err) 1341 return err; 1342 1343 if (copy_to_user(argp, &init, sizeof(init))) 1344 return -EFAULT; 1345 1346 return 0; 1347 } 1348 default: 1349 return -EINVAL; 1350 } 1351 } 1352 1353 static unsigned long nvhe_percpu_size(void) 1354 { 1355 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) - 1356 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start); 1357 } 1358 1359 static unsigned long nvhe_percpu_order(void) 1360 { 1361 unsigned long size = nvhe_percpu_size(); 1362 1363 return size ? get_order(size) : 0; 1364 } 1365 1366 /* A lookup table holding the hypervisor VA for each vector slot */ 1367 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS]; 1368 1369 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot) 1370 { 1371 hyp_spectre_vector_selector[slot] = __kvm_vector_slot2addr(base, slot); 1372 } 1373 1374 static int kvm_init_vector_slots(void) 1375 { 1376 int err; 1377 void *base; 1378 1379 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector)); 1380 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT); 1381 1382 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs)); 1383 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT); 1384 1385 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A)) 1386 return 0; 1387 1388 if (!has_vhe()) { 1389 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs), 1390 __BP_HARDEN_HYP_VECS_SZ, &base); 1391 if (err) 1392 return err; 1393 } 1394 1395 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT); 1396 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT); 1397 return 0; 1398 } 1399 1400 static void cpu_prepare_hyp_mode(int cpu) 1401 { 1402 struct kvm_nvhe_init_params *params = per_cpu_ptr_nvhe_sym(kvm_init_params, cpu); 1403 unsigned long tcr; 1404 1405 /* 1406 * Calculate the raw per-cpu offset without a translation from the 1407 * kernel's mapping to the linear mapping, and store it in tpidr_el2 1408 * so that we can use adr_l to access per-cpu variables in EL2. 1409 * Also drop the KASAN tag which gets in the way... 1410 */ 1411 params->tpidr_el2 = (unsigned long)kasan_reset_tag(per_cpu_ptr_nvhe_sym(__per_cpu_start, cpu)) - 1412 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start)); 1413 1414 params->mair_el2 = read_sysreg(mair_el1); 1415 1416 /* 1417 * The ID map may be configured to use an extended virtual address 1418 * range. This is only the case if system RAM is out of range for the 1419 * currently configured page size and VA_BITS, in which case we will 1420 * also need the extended virtual range for the HYP ID map, or we won't 1421 * be able to enable the EL2 MMU. 1422 * 1423 * However, at EL2, there is only one TTBR register, and we can't switch 1424 * between translation tables *and* update TCR_EL2.T0SZ at the same 1425 * time. Bottom line: we need to use the extended range with *both* our 1426 * translation tables. 1427 * 1428 * So use the same T0SZ value we use for the ID map. 1429 */ 1430 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1; 1431 tcr &= ~TCR_T0SZ_MASK; 1432 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET; 1433 params->tcr_el2 = tcr; 1434 1435 params->stack_hyp_va = kern_hyp_va(per_cpu(kvm_arm_hyp_stack_page, cpu) + PAGE_SIZE); 1436 params->pgd_pa = kvm_mmu_get_httbr(); 1437 if (is_protected_kvm_enabled()) 1438 params->hcr_el2 = HCR_HOST_NVHE_PROTECTED_FLAGS; 1439 else 1440 params->hcr_el2 = HCR_HOST_NVHE_FLAGS; 1441 params->vttbr = params->vtcr = 0; 1442 1443 /* 1444 * Flush the init params from the data cache because the struct will 1445 * be read while the MMU is off. 1446 */ 1447 kvm_flush_dcache_to_poc(params, sizeof(*params)); 1448 } 1449 1450 static void hyp_install_host_vector(void) 1451 { 1452 struct kvm_nvhe_init_params *params; 1453 struct arm_smccc_res res; 1454 1455 /* Switch from the HYP stub to our own HYP init vector */ 1456 __hyp_set_vectors(kvm_get_idmap_vector()); 1457 1458 /* 1459 * Call initialization code, and switch to the full blown HYP code. 1460 * If the cpucaps haven't been finalized yet, something has gone very 1461 * wrong, and hyp will crash and burn when it uses any 1462 * cpus_have_const_cap() wrapper. 1463 */ 1464 BUG_ON(!system_capabilities_finalized()); 1465 params = this_cpu_ptr_nvhe_sym(kvm_init_params); 1466 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res); 1467 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); 1468 } 1469 1470 static void cpu_init_hyp_mode(void) 1471 { 1472 hyp_install_host_vector(); 1473 1474 /* 1475 * Disabling SSBD on a non-VHE system requires us to enable SSBS 1476 * at EL2. 1477 */ 1478 if (this_cpu_has_cap(ARM64_SSBS) && 1479 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) { 1480 kvm_call_hyp_nvhe(__kvm_enable_ssbs); 1481 } 1482 } 1483 1484 static void cpu_hyp_reset(void) 1485 { 1486 if (!is_kernel_in_hyp_mode()) 1487 __hyp_reset_vectors(); 1488 } 1489 1490 /* 1491 * EL2 vectors can be mapped and rerouted in a number of ways, 1492 * depending on the kernel configuration and CPU present: 1493 * 1494 * - If the CPU is affected by Spectre-v2, the hardening sequence is 1495 * placed in one of the vector slots, which is executed before jumping 1496 * to the real vectors. 1497 * 1498 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot 1499 * containing the hardening sequence is mapped next to the idmap page, 1500 * and executed before jumping to the real vectors. 1501 * 1502 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an 1503 * empty slot is selected, mapped next to the idmap page, and 1504 * executed before jumping to the real vectors. 1505 * 1506 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with 1507 * VHE, as we don't have hypervisor-specific mappings. If the system 1508 * is VHE and yet selects this capability, it will be ignored. 1509 */ 1510 static void cpu_set_hyp_vector(void) 1511 { 1512 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data); 1513 void *vector = hyp_spectre_vector_selector[data->slot]; 1514 1515 if (!is_protected_kvm_enabled()) 1516 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector; 1517 else 1518 kvm_call_hyp_nvhe(__pkvm_cpu_set_vector, data->slot); 1519 } 1520 1521 static void cpu_hyp_reinit(void) 1522 { 1523 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt); 1524 1525 cpu_hyp_reset(); 1526 1527 if (is_kernel_in_hyp_mode()) 1528 kvm_timer_init_vhe(); 1529 else 1530 cpu_init_hyp_mode(); 1531 1532 cpu_set_hyp_vector(); 1533 1534 kvm_arm_init_debug(); 1535 1536 if (vgic_present) 1537 kvm_vgic_init_cpu_hardware(); 1538 } 1539 1540 static void _kvm_arch_hardware_enable(void *discard) 1541 { 1542 if (!__this_cpu_read(kvm_arm_hardware_enabled)) { 1543 cpu_hyp_reinit(); 1544 __this_cpu_write(kvm_arm_hardware_enabled, 1); 1545 } 1546 } 1547 1548 int kvm_arch_hardware_enable(void) 1549 { 1550 _kvm_arch_hardware_enable(NULL); 1551 return 0; 1552 } 1553 1554 static void _kvm_arch_hardware_disable(void *discard) 1555 { 1556 if (__this_cpu_read(kvm_arm_hardware_enabled)) { 1557 cpu_hyp_reset(); 1558 __this_cpu_write(kvm_arm_hardware_enabled, 0); 1559 } 1560 } 1561 1562 void kvm_arch_hardware_disable(void) 1563 { 1564 if (!is_protected_kvm_enabled()) 1565 _kvm_arch_hardware_disable(NULL); 1566 } 1567 1568 #ifdef CONFIG_CPU_PM 1569 static int hyp_init_cpu_pm_notifier(struct notifier_block *self, 1570 unsigned long cmd, 1571 void *v) 1572 { 1573 /* 1574 * kvm_arm_hardware_enabled is left with its old value over 1575 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should 1576 * re-enable hyp. 1577 */ 1578 switch (cmd) { 1579 case CPU_PM_ENTER: 1580 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1581 /* 1582 * don't update kvm_arm_hardware_enabled here 1583 * so that the hardware will be re-enabled 1584 * when we resume. See below. 1585 */ 1586 cpu_hyp_reset(); 1587 1588 return NOTIFY_OK; 1589 case CPU_PM_ENTER_FAILED: 1590 case CPU_PM_EXIT: 1591 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1592 /* The hardware was enabled before suspend. */ 1593 cpu_hyp_reinit(); 1594 1595 return NOTIFY_OK; 1596 1597 default: 1598 return NOTIFY_DONE; 1599 } 1600 } 1601 1602 static struct notifier_block hyp_init_cpu_pm_nb = { 1603 .notifier_call = hyp_init_cpu_pm_notifier, 1604 }; 1605 1606 static void hyp_cpu_pm_init(void) 1607 { 1608 if (!is_protected_kvm_enabled()) 1609 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); 1610 } 1611 static void hyp_cpu_pm_exit(void) 1612 { 1613 if (!is_protected_kvm_enabled()) 1614 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); 1615 } 1616 #else 1617 static inline void hyp_cpu_pm_init(void) 1618 { 1619 } 1620 static inline void hyp_cpu_pm_exit(void) 1621 { 1622 } 1623 #endif 1624 1625 static void init_cpu_logical_map(void) 1626 { 1627 unsigned int cpu; 1628 1629 /* 1630 * Copy the MPIDR <-> logical CPU ID mapping to hyp. 1631 * Only copy the set of online CPUs whose features have been chacked 1632 * against the finalized system capabilities. The hypervisor will not 1633 * allow any other CPUs from the `possible` set to boot. 1634 */ 1635 for_each_online_cpu(cpu) 1636 hyp_cpu_logical_map[cpu] = cpu_logical_map(cpu); 1637 } 1638 1639 #define init_psci_0_1_impl_state(config, what) \ 1640 config.psci_0_1_ ## what ## _implemented = psci_ops.what 1641 1642 static bool init_psci_relay(void) 1643 { 1644 /* 1645 * If PSCI has not been initialized, protected KVM cannot install 1646 * itself on newly booted CPUs. 1647 */ 1648 if (!psci_ops.get_version) { 1649 kvm_err("Cannot initialize protected mode without PSCI\n"); 1650 return false; 1651 } 1652 1653 kvm_host_psci_config.version = psci_ops.get_version(); 1654 1655 if (kvm_host_psci_config.version == PSCI_VERSION(0, 1)) { 1656 kvm_host_psci_config.function_ids_0_1 = get_psci_0_1_function_ids(); 1657 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_suspend); 1658 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_on); 1659 init_psci_0_1_impl_state(kvm_host_psci_config, cpu_off); 1660 init_psci_0_1_impl_state(kvm_host_psci_config, migrate); 1661 } 1662 return true; 1663 } 1664 1665 static int init_common_resources(void) 1666 { 1667 return kvm_set_ipa_limit(); 1668 } 1669 1670 static int init_subsystems(void) 1671 { 1672 int err = 0; 1673 1674 /* 1675 * Enable hardware so that subsystem initialisation can access EL2. 1676 */ 1677 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); 1678 1679 /* 1680 * Register CPU lower-power notifier 1681 */ 1682 hyp_cpu_pm_init(); 1683 1684 /* 1685 * Init HYP view of VGIC 1686 */ 1687 err = kvm_vgic_hyp_init(); 1688 switch (err) { 1689 case 0: 1690 vgic_present = true; 1691 break; 1692 case -ENODEV: 1693 case -ENXIO: 1694 vgic_present = false; 1695 err = 0; 1696 break; 1697 default: 1698 goto out; 1699 } 1700 1701 /* 1702 * Init HYP architected timer support 1703 */ 1704 err = kvm_timer_hyp_init(vgic_present); 1705 if (err) 1706 goto out; 1707 1708 kvm_perf_init(); 1709 kvm_sys_reg_table_init(); 1710 1711 out: 1712 if (err || !is_protected_kvm_enabled()) 1713 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); 1714 1715 return err; 1716 } 1717 1718 static void teardown_hyp_mode(void) 1719 { 1720 int cpu; 1721 1722 free_hyp_pgds(); 1723 for_each_possible_cpu(cpu) { 1724 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); 1725 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order()); 1726 } 1727 } 1728 1729 static int do_pkvm_init(u32 hyp_va_bits) 1730 { 1731 void *per_cpu_base = kvm_ksym_ref(kvm_arm_hyp_percpu_base); 1732 int ret; 1733 1734 preempt_disable(); 1735 hyp_install_host_vector(); 1736 ret = kvm_call_hyp_nvhe(__pkvm_init, hyp_mem_base, hyp_mem_size, 1737 num_possible_cpus(), kern_hyp_va(per_cpu_base), 1738 hyp_va_bits); 1739 preempt_enable(); 1740 1741 return ret; 1742 } 1743 1744 static int kvm_hyp_init_protection(u32 hyp_va_bits) 1745 { 1746 void *addr = phys_to_virt(hyp_mem_base); 1747 int ret; 1748 1749 kvm_nvhe_sym(id_aa64mmfr0_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR0_EL1); 1750 kvm_nvhe_sym(id_aa64mmfr1_el1_sys_val) = read_sanitised_ftr_reg(SYS_ID_AA64MMFR1_EL1); 1751 1752 ret = create_hyp_mappings(addr, addr + hyp_mem_size, PAGE_HYP); 1753 if (ret) 1754 return ret; 1755 1756 ret = do_pkvm_init(hyp_va_bits); 1757 if (ret) 1758 return ret; 1759 1760 free_hyp_pgds(); 1761 1762 return 0; 1763 } 1764 1765 /** 1766 * Inits Hyp-mode on all online CPUs 1767 */ 1768 static int init_hyp_mode(void) 1769 { 1770 u32 hyp_va_bits; 1771 int cpu; 1772 int err = -ENOMEM; 1773 1774 /* 1775 * The protected Hyp-mode cannot be initialized if the memory pool 1776 * allocation has failed. 1777 */ 1778 if (is_protected_kvm_enabled() && !hyp_mem_base) 1779 goto out_err; 1780 1781 /* 1782 * Allocate Hyp PGD and setup Hyp identity mapping 1783 */ 1784 err = kvm_mmu_init(&hyp_va_bits); 1785 if (err) 1786 goto out_err; 1787 1788 /* 1789 * Allocate stack pages for Hypervisor-mode 1790 */ 1791 for_each_possible_cpu(cpu) { 1792 unsigned long stack_page; 1793 1794 stack_page = __get_free_page(GFP_KERNEL); 1795 if (!stack_page) { 1796 err = -ENOMEM; 1797 goto out_err; 1798 } 1799 1800 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; 1801 } 1802 1803 /* 1804 * Allocate and initialize pages for Hypervisor-mode percpu regions. 1805 */ 1806 for_each_possible_cpu(cpu) { 1807 struct page *page; 1808 void *page_addr; 1809 1810 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order()); 1811 if (!page) { 1812 err = -ENOMEM; 1813 goto out_err; 1814 } 1815 1816 page_addr = page_address(page); 1817 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size()); 1818 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr; 1819 } 1820 1821 /* 1822 * Map the Hyp-code called directly from the host 1823 */ 1824 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), 1825 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); 1826 if (err) { 1827 kvm_err("Cannot map world-switch code\n"); 1828 goto out_err; 1829 } 1830 1831 err = create_hyp_mappings(kvm_ksym_ref(__hyp_rodata_start), 1832 kvm_ksym_ref(__hyp_rodata_end), PAGE_HYP_RO); 1833 if (err) { 1834 kvm_err("Cannot map .hyp.rodata section\n"); 1835 goto out_err; 1836 } 1837 1838 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), 1839 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); 1840 if (err) { 1841 kvm_err("Cannot map rodata section\n"); 1842 goto out_err; 1843 } 1844 1845 /* 1846 * .hyp.bss is guaranteed to be placed at the beginning of the .bss 1847 * section thanks to an assertion in the linker script. Map it RW and 1848 * the rest of .bss RO. 1849 */ 1850 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_start), 1851 kvm_ksym_ref(__hyp_bss_end), PAGE_HYP); 1852 if (err) { 1853 kvm_err("Cannot map hyp bss section: %d\n", err); 1854 goto out_err; 1855 } 1856 1857 err = create_hyp_mappings(kvm_ksym_ref(__hyp_bss_end), 1858 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); 1859 if (err) { 1860 kvm_err("Cannot map bss section\n"); 1861 goto out_err; 1862 } 1863 1864 /* 1865 * Map the Hyp stack pages 1866 */ 1867 for_each_possible_cpu(cpu) { 1868 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); 1869 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, 1870 PAGE_HYP); 1871 1872 if (err) { 1873 kvm_err("Cannot map hyp stack\n"); 1874 goto out_err; 1875 } 1876 } 1877 1878 for_each_possible_cpu(cpu) { 1879 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu]; 1880 char *percpu_end = percpu_begin + nvhe_percpu_size(); 1881 1882 /* Map Hyp percpu pages */ 1883 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP); 1884 if (err) { 1885 kvm_err("Cannot map hyp percpu region\n"); 1886 goto out_err; 1887 } 1888 1889 /* Prepare the CPU initialization parameters */ 1890 cpu_prepare_hyp_mode(cpu); 1891 } 1892 1893 if (is_protected_kvm_enabled()) { 1894 init_cpu_logical_map(); 1895 1896 if (!init_psci_relay()) { 1897 err = -ENODEV; 1898 goto out_err; 1899 } 1900 } 1901 1902 if (is_protected_kvm_enabled()) { 1903 err = kvm_hyp_init_protection(hyp_va_bits); 1904 if (err) { 1905 kvm_err("Failed to init hyp memory protection\n"); 1906 goto out_err; 1907 } 1908 } 1909 1910 return 0; 1911 1912 out_err: 1913 teardown_hyp_mode(); 1914 kvm_err("error initializing Hyp mode: %d\n", err); 1915 return err; 1916 } 1917 1918 static void _kvm_host_prot_finalize(void *discard) 1919 { 1920 WARN_ON(kvm_call_hyp_nvhe(__pkvm_prot_finalize)); 1921 } 1922 1923 static inline int pkvm_mark_hyp(phys_addr_t start, phys_addr_t end) 1924 { 1925 return kvm_call_hyp_nvhe(__pkvm_mark_hyp, start, end); 1926 } 1927 1928 #define pkvm_mark_hyp_section(__section) \ 1929 pkvm_mark_hyp(__pa_symbol(__section##_start), \ 1930 __pa_symbol(__section##_end)) 1931 1932 static int finalize_hyp_mode(void) 1933 { 1934 int cpu, ret; 1935 1936 if (!is_protected_kvm_enabled()) 1937 return 0; 1938 1939 ret = pkvm_mark_hyp_section(__hyp_idmap_text); 1940 if (ret) 1941 return ret; 1942 1943 ret = pkvm_mark_hyp_section(__hyp_text); 1944 if (ret) 1945 return ret; 1946 1947 ret = pkvm_mark_hyp_section(__hyp_rodata); 1948 if (ret) 1949 return ret; 1950 1951 ret = pkvm_mark_hyp_section(__hyp_bss); 1952 if (ret) 1953 return ret; 1954 1955 ret = pkvm_mark_hyp(hyp_mem_base, hyp_mem_base + hyp_mem_size); 1956 if (ret) 1957 return ret; 1958 1959 for_each_possible_cpu(cpu) { 1960 phys_addr_t start = virt_to_phys((void *)kvm_arm_hyp_percpu_base[cpu]); 1961 phys_addr_t end = start + (PAGE_SIZE << nvhe_percpu_order()); 1962 1963 ret = pkvm_mark_hyp(start, end); 1964 if (ret) 1965 return ret; 1966 1967 start = virt_to_phys((void *)per_cpu(kvm_arm_hyp_stack_page, cpu)); 1968 end = start + PAGE_SIZE; 1969 ret = pkvm_mark_hyp(start, end); 1970 if (ret) 1971 return ret; 1972 } 1973 1974 /* 1975 * Flip the static key upfront as that may no longer be possible 1976 * once the host stage 2 is installed. 1977 */ 1978 static_branch_enable(&kvm_protected_mode_initialized); 1979 on_each_cpu(_kvm_host_prot_finalize, NULL, 1); 1980 1981 return 0; 1982 } 1983 1984 static void check_kvm_target_cpu(void *ret) 1985 { 1986 *(int *)ret = kvm_target_cpu(); 1987 } 1988 1989 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) 1990 { 1991 struct kvm_vcpu *vcpu; 1992 int i; 1993 1994 mpidr &= MPIDR_HWID_BITMASK; 1995 kvm_for_each_vcpu(i, vcpu, kvm) { 1996 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) 1997 return vcpu; 1998 } 1999 return NULL; 2000 } 2001 2002 bool kvm_arch_has_irq_bypass(void) 2003 { 2004 return true; 2005 } 2006 2007 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 2008 struct irq_bypass_producer *prod) 2009 { 2010 struct kvm_kernel_irqfd *irqfd = 2011 container_of(cons, struct kvm_kernel_irqfd, consumer); 2012 2013 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, 2014 &irqfd->irq_entry); 2015 } 2016 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 2017 struct irq_bypass_producer *prod) 2018 { 2019 struct kvm_kernel_irqfd *irqfd = 2020 container_of(cons, struct kvm_kernel_irqfd, consumer); 2021 2022 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, 2023 &irqfd->irq_entry); 2024 } 2025 2026 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) 2027 { 2028 struct kvm_kernel_irqfd *irqfd = 2029 container_of(cons, struct kvm_kernel_irqfd, consumer); 2030 2031 kvm_arm_halt_guest(irqfd->kvm); 2032 } 2033 2034 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) 2035 { 2036 struct kvm_kernel_irqfd *irqfd = 2037 container_of(cons, struct kvm_kernel_irqfd, consumer); 2038 2039 kvm_arm_resume_guest(irqfd->kvm); 2040 } 2041 2042 /** 2043 * Initialize Hyp-mode and memory mappings on all CPUs. 2044 */ 2045 int kvm_arch_init(void *opaque) 2046 { 2047 int err; 2048 int ret, cpu; 2049 bool in_hyp_mode; 2050 2051 if (!is_hyp_mode_available()) { 2052 kvm_info("HYP mode not available\n"); 2053 return -ENODEV; 2054 } 2055 2056 in_hyp_mode = is_kernel_in_hyp_mode(); 2057 2058 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) || 2059 cpus_have_final_cap(ARM64_WORKAROUND_1508412)) 2060 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \ 2061 "Only trusted guests should be used on this system.\n"); 2062 2063 for_each_online_cpu(cpu) { 2064 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); 2065 if (ret < 0) { 2066 kvm_err("Error, CPU %d not supported!\n", cpu); 2067 return -ENODEV; 2068 } 2069 } 2070 2071 err = init_common_resources(); 2072 if (err) 2073 return err; 2074 2075 err = kvm_arm_init_sve(); 2076 if (err) 2077 return err; 2078 2079 if (!in_hyp_mode) { 2080 err = init_hyp_mode(); 2081 if (err) 2082 goto out_err; 2083 } 2084 2085 err = kvm_init_vector_slots(); 2086 if (err) { 2087 kvm_err("Cannot initialise vector slots\n"); 2088 goto out_err; 2089 } 2090 2091 err = init_subsystems(); 2092 if (err) 2093 goto out_hyp; 2094 2095 if (!in_hyp_mode) { 2096 err = finalize_hyp_mode(); 2097 if (err) { 2098 kvm_err("Failed to finalize Hyp protection\n"); 2099 goto out_hyp; 2100 } 2101 } 2102 2103 if (is_protected_kvm_enabled()) { 2104 kvm_info("Protected nVHE mode initialized successfully\n"); 2105 } else if (in_hyp_mode) { 2106 kvm_info("VHE mode initialized successfully\n"); 2107 } else { 2108 kvm_info("Hyp mode initialized successfully\n"); 2109 } 2110 2111 return 0; 2112 2113 out_hyp: 2114 hyp_cpu_pm_exit(); 2115 if (!in_hyp_mode) 2116 teardown_hyp_mode(); 2117 out_err: 2118 return err; 2119 } 2120 2121 /* NOP: Compiling as a module not supported */ 2122 void kvm_arch_exit(void) 2123 { 2124 kvm_perf_teardown(); 2125 } 2126 2127 static int __init early_kvm_mode_cfg(char *arg) 2128 { 2129 if (!arg) 2130 return -EINVAL; 2131 2132 if (strcmp(arg, "protected") == 0) { 2133 kvm_mode = KVM_MODE_PROTECTED; 2134 return 0; 2135 } 2136 2137 if (strcmp(arg, "nvhe") == 0 && !WARN_ON(is_kernel_in_hyp_mode())) 2138 return 0; 2139 2140 return -EINVAL; 2141 } 2142 early_param("kvm-arm.mode", early_kvm_mode_cfg); 2143 2144 enum kvm_mode kvm_get_mode(void) 2145 { 2146 return kvm_mode; 2147 } 2148 2149 static int arm_init(void) 2150 { 2151 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); 2152 return rc; 2153 } 2154 2155 module_init(arm_init); 2156