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 extern u64 kvm_nvhe_sym(__cpu_logical_map)[NR_CPUS]; 69 extern u32 kvm_nvhe_sym(kvm_host_psci_version); 70 extern struct psci_0_1_function_ids kvm_nvhe_sym(kvm_host_psci_0_1_function_ids); 71 72 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 73 { 74 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; 75 } 76 77 int kvm_arch_hardware_setup(void *opaque) 78 { 79 return 0; 80 } 81 82 int kvm_arch_check_processor_compat(void *opaque) 83 { 84 return 0; 85 } 86 87 int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 88 struct kvm_enable_cap *cap) 89 { 90 int r; 91 92 if (cap->flags) 93 return -EINVAL; 94 95 switch (cap->cap) { 96 case KVM_CAP_ARM_NISV_TO_USER: 97 r = 0; 98 kvm->arch.return_nisv_io_abort_to_user = true; 99 break; 100 default: 101 r = -EINVAL; 102 break; 103 } 104 105 return r; 106 } 107 108 static int kvm_arm_default_max_vcpus(void) 109 { 110 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; 111 } 112 113 static void set_default_spectre(struct kvm *kvm) 114 { 115 /* 116 * The default is to expose CSV2 == 1 if the HW isn't affected. 117 * Although this is a per-CPU feature, we make it global because 118 * asymmetric systems are just a nuisance. 119 * 120 * Userspace can override this as long as it doesn't promise 121 * the impossible. 122 */ 123 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) 124 kvm->arch.pfr0_csv2 = 1; 125 if (arm64_get_meltdown_state() == SPECTRE_UNAFFECTED) 126 kvm->arch.pfr0_csv3 = 1; 127 } 128 129 /** 130 * kvm_arch_init_vm - initializes a VM data structure 131 * @kvm: pointer to the KVM struct 132 */ 133 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 134 { 135 int ret; 136 137 ret = kvm_arm_setup_stage2(kvm, type); 138 if (ret) 139 return ret; 140 141 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu); 142 if (ret) 143 return ret; 144 145 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); 146 if (ret) 147 goto out_free_stage2_pgd; 148 149 kvm_vgic_early_init(kvm); 150 151 /* The maximum number of VCPUs is limited by the host's GIC model */ 152 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus(); 153 154 set_default_spectre(kvm); 155 156 return ret; 157 out_free_stage2_pgd: 158 kvm_free_stage2_pgd(&kvm->arch.mmu); 159 return ret; 160 } 161 162 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 163 { 164 return VM_FAULT_SIGBUS; 165 } 166 167 168 /** 169 * kvm_arch_destroy_vm - destroy the VM data structure 170 * @kvm: pointer to the KVM struct 171 */ 172 void kvm_arch_destroy_vm(struct kvm *kvm) 173 { 174 int i; 175 176 bitmap_free(kvm->arch.pmu_filter); 177 178 kvm_vgic_destroy(kvm); 179 180 for (i = 0; i < KVM_MAX_VCPUS; ++i) { 181 if (kvm->vcpus[i]) { 182 kvm_vcpu_destroy(kvm->vcpus[i]); 183 kvm->vcpus[i] = NULL; 184 } 185 } 186 atomic_set(&kvm->online_vcpus, 0); 187 } 188 189 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 190 { 191 int r; 192 switch (ext) { 193 case KVM_CAP_IRQCHIP: 194 r = vgic_present; 195 break; 196 case KVM_CAP_IOEVENTFD: 197 case KVM_CAP_DEVICE_CTRL: 198 case KVM_CAP_USER_MEMORY: 199 case KVM_CAP_SYNC_MMU: 200 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: 201 case KVM_CAP_ONE_REG: 202 case KVM_CAP_ARM_PSCI: 203 case KVM_CAP_ARM_PSCI_0_2: 204 case KVM_CAP_READONLY_MEM: 205 case KVM_CAP_MP_STATE: 206 case KVM_CAP_IMMEDIATE_EXIT: 207 case KVM_CAP_VCPU_EVENTS: 208 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: 209 case KVM_CAP_ARM_NISV_TO_USER: 210 case KVM_CAP_ARM_INJECT_EXT_DABT: 211 case KVM_CAP_SET_GUEST_DEBUG: 212 case KVM_CAP_VCPU_ATTRIBUTES: 213 r = 1; 214 break; 215 case KVM_CAP_ARM_SET_DEVICE_ADDR: 216 r = 1; 217 break; 218 case KVM_CAP_NR_VCPUS: 219 r = num_online_cpus(); 220 break; 221 case KVM_CAP_MAX_VCPUS: 222 case KVM_CAP_MAX_VCPU_ID: 223 if (kvm) 224 r = kvm->arch.max_vcpus; 225 else 226 r = kvm_arm_default_max_vcpus(); 227 break; 228 case KVM_CAP_MSI_DEVID: 229 if (!kvm) 230 r = -EINVAL; 231 else 232 r = kvm->arch.vgic.msis_require_devid; 233 break; 234 case KVM_CAP_ARM_USER_IRQ: 235 /* 236 * 1: EL1_VTIMER, EL1_PTIMER, and PMU. 237 * (bump this number if adding more devices) 238 */ 239 r = 1; 240 break; 241 case KVM_CAP_STEAL_TIME: 242 r = kvm_arm_pvtime_supported(); 243 break; 244 case KVM_CAP_ARM_EL1_32BIT: 245 r = cpus_have_const_cap(ARM64_HAS_32BIT_EL1); 246 break; 247 case KVM_CAP_GUEST_DEBUG_HW_BPS: 248 r = get_num_brps(); 249 break; 250 case KVM_CAP_GUEST_DEBUG_HW_WPS: 251 r = get_num_wrps(); 252 break; 253 case KVM_CAP_ARM_PMU_V3: 254 r = kvm_arm_support_pmu_v3(); 255 break; 256 case KVM_CAP_ARM_INJECT_SERROR_ESR: 257 r = cpus_have_const_cap(ARM64_HAS_RAS_EXTN); 258 break; 259 case KVM_CAP_ARM_VM_IPA_SIZE: 260 r = get_kvm_ipa_limit(); 261 break; 262 case KVM_CAP_ARM_SVE: 263 r = system_supports_sve(); 264 break; 265 case KVM_CAP_ARM_PTRAUTH_ADDRESS: 266 case KVM_CAP_ARM_PTRAUTH_GENERIC: 267 r = system_has_full_ptr_auth(); 268 break; 269 default: 270 r = 0; 271 } 272 273 return r; 274 } 275 276 long kvm_arch_dev_ioctl(struct file *filp, 277 unsigned int ioctl, unsigned long arg) 278 { 279 return -EINVAL; 280 } 281 282 struct kvm *kvm_arch_alloc_vm(void) 283 { 284 if (!has_vhe()) 285 return kzalloc(sizeof(struct kvm), GFP_KERNEL); 286 287 return vzalloc(sizeof(struct kvm)); 288 } 289 290 void kvm_arch_free_vm(struct kvm *kvm) 291 { 292 if (!has_vhe()) 293 kfree(kvm); 294 else 295 vfree(kvm); 296 } 297 298 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 299 { 300 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) 301 return -EBUSY; 302 303 if (id >= kvm->arch.max_vcpus) 304 return -EINVAL; 305 306 return 0; 307 } 308 309 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 310 { 311 int err; 312 313 /* Force users to call KVM_ARM_VCPU_INIT */ 314 vcpu->arch.target = -1; 315 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 316 317 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO; 318 319 /* Set up the timer */ 320 kvm_timer_vcpu_init(vcpu); 321 322 kvm_pmu_vcpu_init(vcpu); 323 324 kvm_arm_reset_debug_ptr(vcpu); 325 326 kvm_arm_pvtime_vcpu_init(&vcpu->arch); 327 328 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu; 329 330 err = kvm_vgic_vcpu_init(vcpu); 331 if (err) 332 return err; 333 334 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); 335 } 336 337 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 338 { 339 } 340 341 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 342 { 343 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm))) 344 static_branch_dec(&userspace_irqchip_in_use); 345 346 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); 347 kvm_timer_vcpu_terminate(vcpu); 348 kvm_pmu_vcpu_destroy(vcpu); 349 350 kvm_arm_vcpu_destroy(vcpu); 351 } 352 353 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 354 { 355 return kvm_timer_is_pending(vcpu); 356 } 357 358 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) 359 { 360 /* 361 * If we're about to block (most likely because we've just hit a 362 * WFI), we need to sync back the state of the GIC CPU interface 363 * so that we have the latest PMR and group enables. This ensures 364 * that kvm_arch_vcpu_runnable has up-to-date data to decide 365 * whether we have pending interrupts. 366 * 367 * For the same reason, we want to tell GICv4 that we need 368 * doorbells to be signalled, should an interrupt become pending. 369 */ 370 preempt_disable(); 371 kvm_vgic_vmcr_sync(vcpu); 372 vgic_v4_put(vcpu, true); 373 preempt_enable(); 374 } 375 376 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) 377 { 378 preempt_disable(); 379 vgic_v4_load(vcpu); 380 preempt_enable(); 381 } 382 383 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 384 { 385 struct kvm_s2_mmu *mmu; 386 int *last_ran; 387 388 mmu = vcpu->arch.hw_mmu; 389 last_ran = this_cpu_ptr(mmu->last_vcpu_ran); 390 391 /* 392 * We might get preempted before the vCPU actually runs, but 393 * over-invalidation doesn't affect correctness. 394 */ 395 if (*last_ran != vcpu->vcpu_id) { 396 kvm_call_hyp(__kvm_tlb_flush_local_vmid, mmu); 397 *last_ran = vcpu->vcpu_id; 398 } 399 400 vcpu->cpu = cpu; 401 402 kvm_vgic_load(vcpu); 403 kvm_timer_vcpu_load(vcpu); 404 if (has_vhe()) 405 kvm_vcpu_load_sysregs_vhe(vcpu); 406 kvm_arch_vcpu_load_fp(vcpu); 407 kvm_vcpu_pmu_restore_guest(vcpu); 408 if (kvm_arm_is_pvtime_enabled(&vcpu->arch)) 409 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu); 410 411 if (single_task_running()) 412 vcpu_clear_wfx_traps(vcpu); 413 else 414 vcpu_set_wfx_traps(vcpu); 415 416 if (vcpu_has_ptrauth(vcpu)) 417 vcpu_ptrauth_disable(vcpu); 418 } 419 420 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 421 { 422 kvm_arch_vcpu_put_fp(vcpu); 423 if (has_vhe()) 424 kvm_vcpu_put_sysregs_vhe(vcpu); 425 kvm_timer_vcpu_put(vcpu); 426 kvm_vgic_put(vcpu); 427 kvm_vcpu_pmu_restore_host(vcpu); 428 429 vcpu->cpu = -1; 430 } 431 432 static void vcpu_power_off(struct kvm_vcpu *vcpu) 433 { 434 vcpu->arch.power_off = true; 435 kvm_make_request(KVM_REQ_SLEEP, vcpu); 436 kvm_vcpu_kick(vcpu); 437 } 438 439 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 440 struct kvm_mp_state *mp_state) 441 { 442 if (vcpu->arch.power_off) 443 mp_state->mp_state = KVM_MP_STATE_STOPPED; 444 else 445 mp_state->mp_state = KVM_MP_STATE_RUNNABLE; 446 447 return 0; 448 } 449 450 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 451 struct kvm_mp_state *mp_state) 452 { 453 int ret = 0; 454 455 switch (mp_state->mp_state) { 456 case KVM_MP_STATE_RUNNABLE: 457 vcpu->arch.power_off = false; 458 break; 459 case KVM_MP_STATE_STOPPED: 460 vcpu_power_off(vcpu); 461 break; 462 default: 463 ret = -EINVAL; 464 } 465 466 return ret; 467 } 468 469 /** 470 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled 471 * @v: The VCPU pointer 472 * 473 * If the guest CPU is not waiting for interrupts or an interrupt line is 474 * asserted, the CPU is by definition runnable. 475 */ 476 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 477 { 478 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); 479 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) 480 && !v->arch.power_off && !v->arch.pause); 481 } 482 483 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 484 { 485 return vcpu_mode_priv(vcpu); 486 } 487 488 /* Just ensure a guest exit from a particular CPU */ 489 static void exit_vm_noop(void *info) 490 { 491 } 492 493 void force_vm_exit(const cpumask_t *mask) 494 { 495 preempt_disable(); 496 smp_call_function_many(mask, exit_vm_noop, NULL, true); 497 preempt_enable(); 498 } 499 500 /** 501 * need_new_vmid_gen - check that the VMID is still valid 502 * @vmid: The VMID to check 503 * 504 * return true if there is a new generation of VMIDs being used 505 * 506 * The hardware supports a limited set of values with the value zero reserved 507 * for the host, so we check if an assigned value belongs to a previous 508 * generation, which requires us to assign a new value. If we're the first to 509 * use a VMID for the new generation, we must flush necessary caches and TLBs 510 * on all CPUs. 511 */ 512 static bool need_new_vmid_gen(struct kvm_vmid *vmid) 513 { 514 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen); 515 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */ 516 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen); 517 } 518 519 /** 520 * update_vmid - Update the vmid with a valid VMID for the current generation 521 * @vmid: The stage-2 VMID information struct 522 */ 523 static void update_vmid(struct kvm_vmid *vmid) 524 { 525 if (!need_new_vmid_gen(vmid)) 526 return; 527 528 spin_lock(&kvm_vmid_lock); 529 530 /* 531 * We need to re-check the vmid_gen here to ensure that if another vcpu 532 * already allocated a valid vmid for this vm, then this vcpu should 533 * use the same vmid. 534 */ 535 if (!need_new_vmid_gen(vmid)) { 536 spin_unlock(&kvm_vmid_lock); 537 return; 538 } 539 540 /* First user of a new VMID generation? */ 541 if (unlikely(kvm_next_vmid == 0)) { 542 atomic64_inc(&kvm_vmid_gen); 543 kvm_next_vmid = 1; 544 545 /* 546 * On SMP we know no other CPUs can use this CPU's or each 547 * other's VMID after force_vm_exit returns since the 548 * kvm_vmid_lock blocks them from reentry to the guest. 549 */ 550 force_vm_exit(cpu_all_mask); 551 /* 552 * Now broadcast TLB + ICACHE invalidation over the inner 553 * shareable domain to make sure all data structures are 554 * clean. 555 */ 556 kvm_call_hyp(__kvm_flush_vm_context); 557 } 558 559 vmid->vmid = kvm_next_vmid; 560 kvm_next_vmid++; 561 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1; 562 563 smp_wmb(); 564 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen)); 565 566 spin_unlock(&kvm_vmid_lock); 567 } 568 569 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) 570 { 571 struct kvm *kvm = vcpu->kvm; 572 int ret = 0; 573 574 if (likely(vcpu->arch.has_run_once)) 575 return 0; 576 577 if (!kvm_arm_vcpu_is_finalized(vcpu)) 578 return -EPERM; 579 580 vcpu->arch.has_run_once = true; 581 582 if (likely(irqchip_in_kernel(kvm))) { 583 /* 584 * Map the VGIC hardware resources before running a vcpu the 585 * first time on this VM. 586 */ 587 if (unlikely(!vgic_ready(kvm))) { 588 ret = kvm_vgic_map_resources(kvm); 589 if (ret) 590 return ret; 591 } 592 } else { 593 /* 594 * Tell the rest of the code that there are userspace irqchip 595 * VMs in the wild. 596 */ 597 static_branch_inc(&userspace_irqchip_in_use); 598 } 599 600 ret = kvm_timer_enable(vcpu); 601 if (ret) 602 return ret; 603 604 ret = kvm_arm_pmu_v3_enable(vcpu); 605 606 return ret; 607 } 608 609 bool kvm_arch_intc_initialized(struct kvm *kvm) 610 { 611 return vgic_initialized(kvm); 612 } 613 614 void kvm_arm_halt_guest(struct kvm *kvm) 615 { 616 int i; 617 struct kvm_vcpu *vcpu; 618 619 kvm_for_each_vcpu(i, vcpu, kvm) 620 vcpu->arch.pause = true; 621 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); 622 } 623 624 void kvm_arm_resume_guest(struct kvm *kvm) 625 { 626 int i; 627 struct kvm_vcpu *vcpu; 628 629 kvm_for_each_vcpu(i, vcpu, kvm) { 630 vcpu->arch.pause = false; 631 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu)); 632 } 633 } 634 635 static void vcpu_req_sleep(struct kvm_vcpu *vcpu) 636 { 637 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 638 639 rcuwait_wait_event(wait, 640 (!vcpu->arch.power_off) &&(!vcpu->arch.pause), 641 TASK_INTERRUPTIBLE); 642 643 if (vcpu->arch.power_off || vcpu->arch.pause) { 644 /* Awaken to handle a signal, request we sleep again later. */ 645 kvm_make_request(KVM_REQ_SLEEP, vcpu); 646 } 647 648 /* 649 * Make sure we will observe a potential reset request if we've 650 * observed a change to the power state. Pairs with the smp_wmb() in 651 * kvm_psci_vcpu_on(). 652 */ 653 smp_rmb(); 654 } 655 656 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) 657 { 658 return vcpu->arch.target >= 0; 659 } 660 661 static void check_vcpu_requests(struct kvm_vcpu *vcpu) 662 { 663 if (kvm_request_pending(vcpu)) { 664 if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) 665 vcpu_req_sleep(vcpu); 666 667 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 668 kvm_reset_vcpu(vcpu); 669 670 /* 671 * Clear IRQ_PENDING requests that were made to guarantee 672 * that a VCPU sees new virtual interrupts. 673 */ 674 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); 675 676 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu)) 677 kvm_update_stolen_time(vcpu); 678 679 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) { 680 /* The distributor enable bits were changed */ 681 preempt_disable(); 682 vgic_v4_put(vcpu, false); 683 vgic_v4_load(vcpu); 684 preempt_enable(); 685 } 686 } 687 } 688 689 /** 690 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code 691 * @vcpu: The VCPU pointer 692 * 693 * This function is called through the VCPU_RUN ioctl called from user space. It 694 * will execute VM code in a loop until the time slice for the process is used 695 * or some emulation is needed from user space in which case the function will 696 * return with return value 0 and with the kvm_run structure filled in with the 697 * required data for the requested emulation. 698 */ 699 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 700 { 701 struct kvm_run *run = vcpu->run; 702 int ret; 703 704 if (unlikely(!kvm_vcpu_initialized(vcpu))) 705 return -ENOEXEC; 706 707 ret = kvm_vcpu_first_run_init(vcpu); 708 if (ret) 709 return ret; 710 711 if (run->exit_reason == KVM_EXIT_MMIO) { 712 ret = kvm_handle_mmio_return(vcpu); 713 if (ret) 714 return ret; 715 } 716 717 if (run->immediate_exit) 718 return -EINTR; 719 720 vcpu_load(vcpu); 721 722 kvm_sigset_activate(vcpu); 723 724 ret = 1; 725 run->exit_reason = KVM_EXIT_UNKNOWN; 726 while (ret > 0) { 727 /* 728 * Check conditions before entering the guest 729 */ 730 cond_resched(); 731 732 update_vmid(&vcpu->arch.hw_mmu->vmid); 733 734 check_vcpu_requests(vcpu); 735 736 /* 737 * Preparing the interrupts to be injected also 738 * involves poking the GIC, which must be done in a 739 * non-preemptible context. 740 */ 741 preempt_disable(); 742 743 kvm_pmu_flush_hwstate(vcpu); 744 745 local_irq_disable(); 746 747 kvm_vgic_flush_hwstate(vcpu); 748 749 /* 750 * Exit if we have a signal pending so that we can deliver the 751 * signal to user space. 752 */ 753 if (signal_pending(current)) { 754 ret = -EINTR; 755 run->exit_reason = KVM_EXIT_INTR; 756 } 757 758 /* 759 * If we're using a userspace irqchip, then check if we need 760 * to tell a userspace irqchip about timer or PMU level 761 * changes and if so, exit to userspace (the actual level 762 * state gets updated in kvm_timer_update_run and 763 * kvm_pmu_update_run below). 764 */ 765 if (static_branch_unlikely(&userspace_irqchip_in_use)) { 766 if (kvm_timer_should_notify_user(vcpu) || 767 kvm_pmu_should_notify_user(vcpu)) { 768 ret = -EINTR; 769 run->exit_reason = KVM_EXIT_INTR; 770 } 771 } 772 773 /* 774 * Ensure we set mode to IN_GUEST_MODE after we disable 775 * interrupts and before the final VCPU requests check. 776 * See the comment in kvm_vcpu_exiting_guest_mode() and 777 * Documentation/virt/kvm/vcpu-requests.rst 778 */ 779 smp_store_mb(vcpu->mode, IN_GUEST_MODE); 780 781 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) || 782 kvm_request_pending(vcpu)) { 783 vcpu->mode = OUTSIDE_GUEST_MODE; 784 isb(); /* Ensure work in x_flush_hwstate is committed */ 785 kvm_pmu_sync_hwstate(vcpu); 786 if (static_branch_unlikely(&userspace_irqchip_in_use)) 787 kvm_timer_sync_user(vcpu); 788 kvm_vgic_sync_hwstate(vcpu); 789 local_irq_enable(); 790 preempt_enable(); 791 continue; 792 } 793 794 kvm_arm_setup_debug(vcpu); 795 796 /************************************************************** 797 * Enter the guest 798 */ 799 trace_kvm_entry(*vcpu_pc(vcpu)); 800 guest_enter_irqoff(); 801 802 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu); 803 804 vcpu->mode = OUTSIDE_GUEST_MODE; 805 vcpu->stat.exits++; 806 /* 807 * Back from guest 808 *************************************************************/ 809 810 kvm_arm_clear_debug(vcpu); 811 812 /* 813 * We must sync the PMU state before the vgic state so 814 * that the vgic can properly sample the updated state of the 815 * interrupt line. 816 */ 817 kvm_pmu_sync_hwstate(vcpu); 818 819 /* 820 * Sync the vgic state before syncing the timer state because 821 * the timer code needs to know if the virtual timer 822 * interrupts are active. 823 */ 824 kvm_vgic_sync_hwstate(vcpu); 825 826 /* 827 * Sync the timer hardware state before enabling interrupts as 828 * we don't want vtimer interrupts to race with syncing the 829 * timer virtual interrupt state. 830 */ 831 if (static_branch_unlikely(&userspace_irqchip_in_use)) 832 kvm_timer_sync_user(vcpu); 833 834 kvm_arch_vcpu_ctxsync_fp(vcpu); 835 836 /* 837 * We may have taken a host interrupt in HYP mode (ie 838 * while executing the guest). This interrupt is still 839 * pending, as we haven't serviced it yet! 840 * 841 * We're now back in SVC mode, with interrupts 842 * disabled. Enabling the interrupts now will have 843 * the effect of taking the interrupt again, in SVC 844 * mode this time. 845 */ 846 local_irq_enable(); 847 848 /* 849 * We do local_irq_enable() before calling guest_exit() so 850 * that if a timer interrupt hits while running the guest we 851 * account that tick as being spent in the guest. We enable 852 * preemption after calling guest_exit() so that if we get 853 * preempted we make sure ticks after that is not counted as 854 * guest time. 855 */ 856 guest_exit(); 857 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); 858 859 /* Exit types that need handling before we can be preempted */ 860 handle_exit_early(vcpu, ret); 861 862 preempt_enable(); 863 864 /* 865 * The ARMv8 architecture doesn't give the hypervisor 866 * a mechanism to prevent a guest from dropping to AArch32 EL0 867 * if implemented by the CPU. If we spot the guest in such 868 * state and that we decided it wasn't supposed to do so (like 869 * with the asymmetric AArch32 case), return to userspace with 870 * a fatal error. 871 */ 872 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) { 873 /* 874 * As we have caught the guest red-handed, decide that 875 * it isn't fit for purpose anymore by making the vcpu 876 * invalid. The VMM can try and fix it by issuing a 877 * KVM_ARM_VCPU_INIT if it really wants to. 878 */ 879 vcpu->arch.target = -1; 880 ret = ARM_EXCEPTION_IL; 881 } 882 883 ret = handle_exit(vcpu, ret); 884 } 885 886 /* Tell userspace about in-kernel device output levels */ 887 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { 888 kvm_timer_update_run(vcpu); 889 kvm_pmu_update_run(vcpu); 890 } 891 892 kvm_sigset_deactivate(vcpu); 893 894 vcpu_put(vcpu); 895 return ret; 896 } 897 898 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) 899 { 900 int bit_index; 901 bool set; 902 unsigned long *hcr; 903 904 if (number == KVM_ARM_IRQ_CPU_IRQ) 905 bit_index = __ffs(HCR_VI); 906 else /* KVM_ARM_IRQ_CPU_FIQ */ 907 bit_index = __ffs(HCR_VF); 908 909 hcr = vcpu_hcr(vcpu); 910 if (level) 911 set = test_and_set_bit(bit_index, hcr); 912 else 913 set = test_and_clear_bit(bit_index, hcr); 914 915 /* 916 * If we didn't change anything, no need to wake up or kick other CPUs 917 */ 918 if (set == level) 919 return 0; 920 921 /* 922 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and 923 * trigger a world-switch round on the running physical CPU to set the 924 * virtual IRQ/FIQ fields in the HCR appropriately. 925 */ 926 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); 927 kvm_vcpu_kick(vcpu); 928 929 return 0; 930 } 931 932 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 933 bool line_status) 934 { 935 u32 irq = irq_level->irq; 936 unsigned int irq_type, vcpu_idx, irq_num; 937 int nrcpus = atomic_read(&kvm->online_vcpus); 938 struct kvm_vcpu *vcpu = NULL; 939 bool level = irq_level->level; 940 941 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; 942 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; 943 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); 944 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; 945 946 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); 947 948 switch (irq_type) { 949 case KVM_ARM_IRQ_TYPE_CPU: 950 if (irqchip_in_kernel(kvm)) 951 return -ENXIO; 952 953 if (vcpu_idx >= nrcpus) 954 return -EINVAL; 955 956 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 957 if (!vcpu) 958 return -EINVAL; 959 960 if (irq_num > KVM_ARM_IRQ_CPU_FIQ) 961 return -EINVAL; 962 963 return vcpu_interrupt_line(vcpu, irq_num, level); 964 case KVM_ARM_IRQ_TYPE_PPI: 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 < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) 976 return -EINVAL; 977 978 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); 979 case KVM_ARM_IRQ_TYPE_SPI: 980 if (!irqchip_in_kernel(kvm)) 981 return -ENXIO; 982 983 if (irq_num < VGIC_NR_PRIVATE_IRQS) 984 return -EINVAL; 985 986 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); 987 } 988 989 return -EINVAL; 990 } 991 992 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, 993 const struct kvm_vcpu_init *init) 994 { 995 unsigned int i, ret; 996 int phys_target = kvm_target_cpu(); 997 998 if (init->target != phys_target) 999 return -EINVAL; 1000 1001 /* 1002 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 1003 * use the same target. 1004 */ 1005 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) 1006 return -EINVAL; 1007 1008 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ 1009 for (i = 0; i < sizeof(init->features) * 8; i++) { 1010 bool set = (init->features[i / 32] & (1 << (i % 32))); 1011 1012 if (set && i >= KVM_VCPU_MAX_FEATURES) 1013 return -ENOENT; 1014 1015 /* 1016 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 1017 * use the same feature set. 1018 */ 1019 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && 1020 test_bit(i, vcpu->arch.features) != set) 1021 return -EINVAL; 1022 1023 if (set) 1024 set_bit(i, vcpu->arch.features); 1025 } 1026 1027 vcpu->arch.target = phys_target; 1028 1029 /* Now we know what it is, we can reset it. */ 1030 ret = kvm_reset_vcpu(vcpu); 1031 if (ret) { 1032 vcpu->arch.target = -1; 1033 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 1034 } 1035 1036 return ret; 1037 } 1038 1039 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, 1040 struct kvm_vcpu_init *init) 1041 { 1042 int ret; 1043 1044 ret = kvm_vcpu_set_target(vcpu, init); 1045 if (ret) 1046 return ret; 1047 1048 /* 1049 * Ensure a rebooted VM will fault in RAM pages and detect if the 1050 * guest MMU is turned off and flush the caches as needed. 1051 * 1052 * S2FWB enforces all memory accesses to RAM being cacheable, 1053 * ensuring that the data side is always coherent. We still 1054 * need to invalidate the I-cache though, as FWB does *not* 1055 * imply CTR_EL0.DIC. 1056 */ 1057 if (vcpu->arch.has_run_once) { 1058 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB)) 1059 stage2_unmap_vm(vcpu->kvm); 1060 else 1061 __flush_icache_all(); 1062 } 1063 1064 vcpu_reset_hcr(vcpu); 1065 1066 /* 1067 * Handle the "start in power-off" case. 1068 */ 1069 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) 1070 vcpu_power_off(vcpu); 1071 else 1072 vcpu->arch.power_off = false; 1073 1074 return 0; 1075 } 1076 1077 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, 1078 struct kvm_device_attr *attr) 1079 { 1080 int ret = -ENXIO; 1081 1082 switch (attr->group) { 1083 default: 1084 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); 1085 break; 1086 } 1087 1088 return ret; 1089 } 1090 1091 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, 1092 struct kvm_device_attr *attr) 1093 { 1094 int ret = -ENXIO; 1095 1096 switch (attr->group) { 1097 default: 1098 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); 1099 break; 1100 } 1101 1102 return ret; 1103 } 1104 1105 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, 1106 struct kvm_device_attr *attr) 1107 { 1108 int ret = -ENXIO; 1109 1110 switch (attr->group) { 1111 default: 1112 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); 1113 break; 1114 } 1115 1116 return ret; 1117 } 1118 1119 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1120 struct kvm_vcpu_events *events) 1121 { 1122 memset(events, 0, sizeof(*events)); 1123 1124 return __kvm_arm_vcpu_get_events(vcpu, events); 1125 } 1126 1127 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1128 struct kvm_vcpu_events *events) 1129 { 1130 int i; 1131 1132 /* check whether the reserved field is zero */ 1133 for (i = 0; i < ARRAY_SIZE(events->reserved); i++) 1134 if (events->reserved[i]) 1135 return -EINVAL; 1136 1137 /* check whether the pad field is zero */ 1138 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) 1139 if (events->exception.pad[i]) 1140 return -EINVAL; 1141 1142 return __kvm_arm_vcpu_set_events(vcpu, events); 1143 } 1144 1145 long kvm_arch_vcpu_ioctl(struct file *filp, 1146 unsigned int ioctl, unsigned long arg) 1147 { 1148 struct kvm_vcpu *vcpu = filp->private_data; 1149 void __user *argp = (void __user *)arg; 1150 struct kvm_device_attr attr; 1151 long r; 1152 1153 switch (ioctl) { 1154 case KVM_ARM_VCPU_INIT: { 1155 struct kvm_vcpu_init init; 1156 1157 r = -EFAULT; 1158 if (copy_from_user(&init, argp, sizeof(init))) 1159 break; 1160 1161 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); 1162 break; 1163 } 1164 case KVM_SET_ONE_REG: 1165 case KVM_GET_ONE_REG: { 1166 struct kvm_one_reg reg; 1167 1168 r = -ENOEXEC; 1169 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1170 break; 1171 1172 r = -EFAULT; 1173 if (copy_from_user(®, argp, sizeof(reg))) 1174 break; 1175 1176 if (ioctl == KVM_SET_ONE_REG) 1177 r = kvm_arm_set_reg(vcpu, ®); 1178 else 1179 r = kvm_arm_get_reg(vcpu, ®); 1180 break; 1181 } 1182 case KVM_GET_REG_LIST: { 1183 struct kvm_reg_list __user *user_list = argp; 1184 struct kvm_reg_list reg_list; 1185 unsigned n; 1186 1187 r = -ENOEXEC; 1188 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1189 break; 1190 1191 r = -EPERM; 1192 if (!kvm_arm_vcpu_is_finalized(vcpu)) 1193 break; 1194 1195 r = -EFAULT; 1196 if (copy_from_user(®_list, user_list, sizeof(reg_list))) 1197 break; 1198 n = reg_list.n; 1199 reg_list.n = kvm_arm_num_regs(vcpu); 1200 if (copy_to_user(user_list, ®_list, sizeof(reg_list))) 1201 break; 1202 r = -E2BIG; 1203 if (n < reg_list.n) 1204 break; 1205 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); 1206 break; 1207 } 1208 case KVM_SET_DEVICE_ATTR: { 1209 r = -EFAULT; 1210 if (copy_from_user(&attr, argp, sizeof(attr))) 1211 break; 1212 r = kvm_arm_vcpu_set_attr(vcpu, &attr); 1213 break; 1214 } 1215 case KVM_GET_DEVICE_ATTR: { 1216 r = -EFAULT; 1217 if (copy_from_user(&attr, argp, sizeof(attr))) 1218 break; 1219 r = kvm_arm_vcpu_get_attr(vcpu, &attr); 1220 break; 1221 } 1222 case KVM_HAS_DEVICE_ATTR: { 1223 r = -EFAULT; 1224 if (copy_from_user(&attr, argp, sizeof(attr))) 1225 break; 1226 r = kvm_arm_vcpu_has_attr(vcpu, &attr); 1227 break; 1228 } 1229 case KVM_GET_VCPU_EVENTS: { 1230 struct kvm_vcpu_events events; 1231 1232 if (kvm_arm_vcpu_get_events(vcpu, &events)) 1233 return -EINVAL; 1234 1235 if (copy_to_user(argp, &events, sizeof(events))) 1236 return -EFAULT; 1237 1238 return 0; 1239 } 1240 case KVM_SET_VCPU_EVENTS: { 1241 struct kvm_vcpu_events events; 1242 1243 if (copy_from_user(&events, argp, sizeof(events))) 1244 return -EFAULT; 1245 1246 return kvm_arm_vcpu_set_events(vcpu, &events); 1247 } 1248 case KVM_ARM_VCPU_FINALIZE: { 1249 int what; 1250 1251 if (!kvm_vcpu_initialized(vcpu)) 1252 return -ENOEXEC; 1253 1254 if (get_user(what, (const int __user *)argp)) 1255 return -EFAULT; 1256 1257 return kvm_arm_vcpu_finalize(vcpu, what); 1258 } 1259 default: 1260 r = -EINVAL; 1261 } 1262 1263 return r; 1264 } 1265 1266 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 1267 { 1268 1269 } 1270 1271 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm, 1272 struct kvm_memory_slot *memslot) 1273 { 1274 kvm_flush_remote_tlbs(kvm); 1275 } 1276 1277 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, 1278 struct kvm_arm_device_addr *dev_addr) 1279 { 1280 unsigned long dev_id, type; 1281 1282 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> 1283 KVM_ARM_DEVICE_ID_SHIFT; 1284 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> 1285 KVM_ARM_DEVICE_TYPE_SHIFT; 1286 1287 switch (dev_id) { 1288 case KVM_ARM_DEVICE_VGIC_V2: 1289 if (!vgic_present) 1290 return -ENXIO; 1291 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); 1292 default: 1293 return -ENODEV; 1294 } 1295 } 1296 1297 long kvm_arch_vm_ioctl(struct file *filp, 1298 unsigned int ioctl, unsigned long arg) 1299 { 1300 struct kvm *kvm = filp->private_data; 1301 void __user *argp = (void __user *)arg; 1302 1303 switch (ioctl) { 1304 case KVM_CREATE_IRQCHIP: { 1305 int ret; 1306 if (!vgic_present) 1307 return -ENXIO; 1308 mutex_lock(&kvm->lock); 1309 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); 1310 mutex_unlock(&kvm->lock); 1311 return ret; 1312 } 1313 case KVM_ARM_SET_DEVICE_ADDR: { 1314 struct kvm_arm_device_addr dev_addr; 1315 1316 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) 1317 return -EFAULT; 1318 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); 1319 } 1320 case KVM_ARM_PREFERRED_TARGET: { 1321 int err; 1322 struct kvm_vcpu_init init; 1323 1324 err = kvm_vcpu_preferred_target(&init); 1325 if (err) 1326 return err; 1327 1328 if (copy_to_user(argp, &init, sizeof(init))) 1329 return -EFAULT; 1330 1331 return 0; 1332 } 1333 default: 1334 return -EINVAL; 1335 } 1336 } 1337 1338 static unsigned long nvhe_percpu_size(void) 1339 { 1340 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) - 1341 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start); 1342 } 1343 1344 static unsigned long nvhe_percpu_order(void) 1345 { 1346 unsigned long size = nvhe_percpu_size(); 1347 1348 return size ? get_order(size) : 0; 1349 } 1350 1351 /* A lookup table holding the hypervisor VA for each vector slot */ 1352 static void *hyp_spectre_vector_selector[BP_HARDEN_EL2_SLOTS]; 1353 1354 static int __kvm_vector_slot2idx(enum arm64_hyp_spectre_vector slot) 1355 { 1356 return slot - (slot != HYP_VECTOR_DIRECT); 1357 } 1358 1359 static void kvm_init_vector_slot(void *base, enum arm64_hyp_spectre_vector slot) 1360 { 1361 int idx = __kvm_vector_slot2idx(slot); 1362 1363 hyp_spectre_vector_selector[slot] = base + (idx * SZ_2K); 1364 } 1365 1366 static int kvm_init_vector_slots(void) 1367 { 1368 int err; 1369 void *base; 1370 1371 base = kern_hyp_va(kvm_ksym_ref(__kvm_hyp_vector)); 1372 kvm_init_vector_slot(base, HYP_VECTOR_DIRECT); 1373 1374 base = kern_hyp_va(kvm_ksym_ref(__bp_harden_hyp_vecs)); 1375 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_DIRECT); 1376 1377 if (!cpus_have_const_cap(ARM64_SPECTRE_V3A)) 1378 return 0; 1379 1380 if (!has_vhe()) { 1381 err = create_hyp_exec_mappings(__pa_symbol(__bp_harden_hyp_vecs), 1382 __BP_HARDEN_HYP_VECS_SZ, &base); 1383 if (err) 1384 return err; 1385 } 1386 1387 kvm_init_vector_slot(base, HYP_VECTOR_INDIRECT); 1388 kvm_init_vector_slot(base, HYP_VECTOR_SPECTRE_INDIRECT); 1389 return 0; 1390 } 1391 1392 static void cpu_init_hyp_mode(void) 1393 { 1394 struct kvm_nvhe_init_params *params = this_cpu_ptr_nvhe_sym(kvm_init_params); 1395 struct arm_smccc_res res; 1396 unsigned long tcr; 1397 1398 /* Switch from the HYP stub to our own HYP init vector */ 1399 __hyp_set_vectors(kvm_get_idmap_vector()); 1400 1401 /* 1402 * Calculate the raw per-cpu offset without a translation from the 1403 * kernel's mapping to the linear mapping, and store it in tpidr_el2 1404 * so that we can use adr_l to access per-cpu variables in EL2. 1405 */ 1406 params->tpidr_el2 = (unsigned long)this_cpu_ptr_nvhe_sym(__per_cpu_start) - 1407 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start)); 1408 1409 params->mair_el2 = read_sysreg(mair_el1); 1410 1411 /* 1412 * The ID map may be configured to use an extended virtual address 1413 * range. This is only the case if system RAM is out of range for the 1414 * currently configured page size and VA_BITS, in which case we will 1415 * also need the extended virtual range for the HYP ID map, or we won't 1416 * be able to enable the EL2 MMU. 1417 * 1418 * However, at EL2, there is only one TTBR register, and we can't switch 1419 * between translation tables *and* update TCR_EL2.T0SZ at the same 1420 * time. Bottom line: we need to use the extended range with *both* our 1421 * translation tables. 1422 * 1423 * So use the same T0SZ value we use for the ID map. 1424 */ 1425 tcr = (read_sysreg(tcr_el1) & TCR_EL2_MASK) | TCR_EL2_RES1; 1426 tcr &= ~TCR_T0SZ_MASK; 1427 tcr |= (idmap_t0sz & GENMASK(TCR_TxSZ_WIDTH - 1, 0)) << TCR_T0SZ_OFFSET; 1428 params->tcr_el2 = tcr; 1429 1430 params->stack_hyp_va = kern_hyp_va(__this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE); 1431 params->pgd_pa = kvm_mmu_get_httbr(); 1432 1433 /* 1434 * Flush the init params from the data cache because the struct will 1435 * be read while the MMU is off. 1436 */ 1437 kvm_flush_dcache_to_poc(params, sizeof(*params)); 1438 1439 /* 1440 * Call initialization code, and switch to the full blown HYP code. 1441 * If the cpucaps haven't been finalized yet, something has gone very 1442 * wrong, and hyp will crash and burn when it uses any 1443 * cpus_have_const_cap() wrapper. 1444 */ 1445 BUG_ON(!system_capabilities_finalized()); 1446 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), virt_to_phys(params), &res); 1447 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); 1448 1449 /* 1450 * Disabling SSBD on a non-VHE system requires us to enable SSBS 1451 * at EL2. 1452 */ 1453 if (this_cpu_has_cap(ARM64_SSBS) && 1454 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) { 1455 kvm_call_hyp_nvhe(__kvm_enable_ssbs); 1456 } 1457 } 1458 1459 static void cpu_hyp_reset(void) 1460 { 1461 if (!is_kernel_in_hyp_mode()) 1462 __hyp_reset_vectors(); 1463 } 1464 1465 /* 1466 * EL2 vectors can be mapped and rerouted in a number of ways, 1467 * depending on the kernel configuration and CPU present: 1468 * 1469 * - If the CPU is affected by Spectre-v2, the hardening sequence is 1470 * placed in one of the vector slots, which is executed before jumping 1471 * to the real vectors. 1472 * 1473 * - If the CPU also has the ARM64_SPECTRE_V3A cap, the slot 1474 * containing the hardening sequence is mapped next to the idmap page, 1475 * and executed before jumping to the real vectors. 1476 * 1477 * - If the CPU only has the ARM64_SPECTRE_V3A cap, then an 1478 * empty slot is selected, mapped next to the idmap page, and 1479 * executed before jumping to the real vectors. 1480 * 1481 * Note that ARM64_SPECTRE_V3A is somewhat incompatible with 1482 * VHE, as we don't have hypervisor-specific mappings. If the system 1483 * is VHE and yet selects this capability, it will be ignored. 1484 */ 1485 static void cpu_set_hyp_vector(void) 1486 { 1487 struct bp_hardening_data *data = this_cpu_ptr(&bp_hardening_data); 1488 void *vector = hyp_spectre_vector_selector[data->slot]; 1489 1490 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)vector; 1491 } 1492 1493 static void cpu_hyp_reinit(void) 1494 { 1495 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt); 1496 1497 cpu_hyp_reset(); 1498 cpu_set_hyp_vector(); 1499 1500 if (is_kernel_in_hyp_mode()) 1501 kvm_timer_init_vhe(); 1502 else 1503 cpu_init_hyp_mode(); 1504 1505 kvm_arm_init_debug(); 1506 1507 if (vgic_present) 1508 kvm_vgic_init_cpu_hardware(); 1509 } 1510 1511 static void _kvm_arch_hardware_enable(void *discard) 1512 { 1513 if (!__this_cpu_read(kvm_arm_hardware_enabled)) { 1514 cpu_hyp_reinit(); 1515 __this_cpu_write(kvm_arm_hardware_enabled, 1); 1516 } 1517 } 1518 1519 int kvm_arch_hardware_enable(void) 1520 { 1521 _kvm_arch_hardware_enable(NULL); 1522 return 0; 1523 } 1524 1525 static void _kvm_arch_hardware_disable(void *discard) 1526 { 1527 if (__this_cpu_read(kvm_arm_hardware_enabled)) { 1528 cpu_hyp_reset(); 1529 __this_cpu_write(kvm_arm_hardware_enabled, 0); 1530 } 1531 } 1532 1533 void kvm_arch_hardware_disable(void) 1534 { 1535 if (!is_protected_kvm_enabled()) 1536 _kvm_arch_hardware_disable(NULL); 1537 } 1538 1539 #ifdef CONFIG_CPU_PM 1540 static int hyp_init_cpu_pm_notifier(struct notifier_block *self, 1541 unsigned long cmd, 1542 void *v) 1543 { 1544 /* 1545 * kvm_arm_hardware_enabled is left with its old value over 1546 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should 1547 * re-enable hyp. 1548 */ 1549 switch (cmd) { 1550 case CPU_PM_ENTER: 1551 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1552 /* 1553 * don't update kvm_arm_hardware_enabled here 1554 * so that the hardware will be re-enabled 1555 * when we resume. See below. 1556 */ 1557 cpu_hyp_reset(); 1558 1559 return NOTIFY_OK; 1560 case CPU_PM_ENTER_FAILED: 1561 case CPU_PM_EXIT: 1562 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1563 /* The hardware was enabled before suspend. */ 1564 cpu_hyp_reinit(); 1565 1566 return NOTIFY_OK; 1567 1568 default: 1569 return NOTIFY_DONE; 1570 } 1571 } 1572 1573 static struct notifier_block hyp_init_cpu_pm_nb = { 1574 .notifier_call = hyp_init_cpu_pm_notifier, 1575 }; 1576 1577 static void __init hyp_cpu_pm_init(void) 1578 { 1579 if (!is_protected_kvm_enabled()) 1580 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); 1581 } 1582 static void __init hyp_cpu_pm_exit(void) 1583 { 1584 if (!is_protected_kvm_enabled()) 1585 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); 1586 } 1587 #else 1588 static inline void hyp_cpu_pm_init(void) 1589 { 1590 } 1591 static inline void hyp_cpu_pm_exit(void) 1592 { 1593 } 1594 #endif 1595 1596 static void init_cpu_logical_map(void) 1597 { 1598 unsigned int cpu; 1599 1600 /* 1601 * Copy the MPIDR <-> logical CPU ID mapping to hyp. 1602 * Only copy the set of online CPUs whose features have been chacked 1603 * against the finalized system capabilities. The hypervisor will not 1604 * allow any other CPUs from the `possible` set to boot. 1605 */ 1606 for_each_online_cpu(cpu) 1607 kvm_nvhe_sym(__cpu_logical_map)[cpu] = cpu_logical_map(cpu); 1608 } 1609 1610 static bool init_psci_relay(void) 1611 { 1612 /* 1613 * If PSCI has not been initialized, protected KVM cannot install 1614 * itself on newly booted CPUs. 1615 */ 1616 if (!psci_ops.get_version) { 1617 kvm_err("Cannot initialize protected mode without PSCI\n"); 1618 return false; 1619 } 1620 1621 kvm_nvhe_sym(kvm_host_psci_version) = psci_ops.get_version(); 1622 kvm_nvhe_sym(kvm_host_psci_0_1_function_ids) = get_psci_0_1_function_ids(); 1623 return true; 1624 } 1625 1626 static int init_common_resources(void) 1627 { 1628 return kvm_set_ipa_limit(); 1629 } 1630 1631 static int init_subsystems(void) 1632 { 1633 int err = 0; 1634 1635 /* 1636 * Enable hardware so that subsystem initialisation can access EL2. 1637 */ 1638 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); 1639 1640 /* 1641 * Register CPU lower-power notifier 1642 */ 1643 hyp_cpu_pm_init(); 1644 1645 /* 1646 * Init HYP view of VGIC 1647 */ 1648 err = kvm_vgic_hyp_init(); 1649 switch (err) { 1650 case 0: 1651 vgic_present = true; 1652 break; 1653 case -ENODEV: 1654 case -ENXIO: 1655 vgic_present = false; 1656 err = 0; 1657 break; 1658 default: 1659 goto out; 1660 } 1661 1662 /* 1663 * Init HYP architected timer support 1664 */ 1665 err = kvm_timer_hyp_init(vgic_present); 1666 if (err) 1667 goto out; 1668 1669 kvm_perf_init(); 1670 kvm_sys_reg_table_init(); 1671 1672 out: 1673 if (err || !is_protected_kvm_enabled()) 1674 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); 1675 1676 return err; 1677 } 1678 1679 static void teardown_hyp_mode(void) 1680 { 1681 int cpu; 1682 1683 free_hyp_pgds(); 1684 for_each_possible_cpu(cpu) { 1685 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); 1686 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order()); 1687 } 1688 } 1689 1690 /** 1691 * Inits Hyp-mode on all online CPUs 1692 */ 1693 static int init_hyp_mode(void) 1694 { 1695 int cpu; 1696 int err = 0; 1697 1698 /* 1699 * Allocate Hyp PGD and setup Hyp identity mapping 1700 */ 1701 err = kvm_mmu_init(); 1702 if (err) 1703 goto out_err; 1704 1705 /* 1706 * Allocate stack pages for Hypervisor-mode 1707 */ 1708 for_each_possible_cpu(cpu) { 1709 unsigned long stack_page; 1710 1711 stack_page = __get_free_page(GFP_KERNEL); 1712 if (!stack_page) { 1713 err = -ENOMEM; 1714 goto out_err; 1715 } 1716 1717 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; 1718 } 1719 1720 /* 1721 * Allocate and initialize pages for Hypervisor-mode percpu regions. 1722 */ 1723 for_each_possible_cpu(cpu) { 1724 struct page *page; 1725 void *page_addr; 1726 1727 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order()); 1728 if (!page) { 1729 err = -ENOMEM; 1730 goto out_err; 1731 } 1732 1733 page_addr = page_address(page); 1734 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size()); 1735 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr; 1736 } 1737 1738 /* 1739 * Map the Hyp-code called directly from the host 1740 */ 1741 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), 1742 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); 1743 if (err) { 1744 kvm_err("Cannot map world-switch code\n"); 1745 goto out_err; 1746 } 1747 1748 err = create_hyp_mappings(kvm_ksym_ref(__hyp_data_ro_after_init_start), 1749 kvm_ksym_ref(__hyp_data_ro_after_init_end), 1750 PAGE_HYP_RO); 1751 if (err) { 1752 kvm_err("Cannot map .hyp.data..ro_after_init section\n"); 1753 goto out_err; 1754 } 1755 1756 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), 1757 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); 1758 if (err) { 1759 kvm_err("Cannot map rodata section\n"); 1760 goto out_err; 1761 } 1762 1763 err = create_hyp_mappings(kvm_ksym_ref(__bss_start), 1764 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); 1765 if (err) { 1766 kvm_err("Cannot map bss section\n"); 1767 goto out_err; 1768 } 1769 1770 /* 1771 * Map the Hyp stack pages 1772 */ 1773 for_each_possible_cpu(cpu) { 1774 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); 1775 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, 1776 PAGE_HYP); 1777 1778 if (err) { 1779 kvm_err("Cannot map hyp stack\n"); 1780 goto out_err; 1781 } 1782 } 1783 1784 /* 1785 * Map Hyp percpu pages 1786 */ 1787 for_each_possible_cpu(cpu) { 1788 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu]; 1789 char *percpu_end = percpu_begin + nvhe_percpu_size(); 1790 1791 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP); 1792 1793 if (err) { 1794 kvm_err("Cannot map hyp percpu region\n"); 1795 goto out_err; 1796 } 1797 } 1798 1799 if (is_protected_kvm_enabled()) { 1800 init_cpu_logical_map(); 1801 1802 if (!init_psci_relay()) 1803 goto out_err; 1804 } 1805 1806 return 0; 1807 1808 out_err: 1809 teardown_hyp_mode(); 1810 kvm_err("error initializing Hyp mode: %d\n", err); 1811 return err; 1812 } 1813 1814 static void check_kvm_target_cpu(void *ret) 1815 { 1816 *(int *)ret = kvm_target_cpu(); 1817 } 1818 1819 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) 1820 { 1821 struct kvm_vcpu *vcpu; 1822 int i; 1823 1824 mpidr &= MPIDR_HWID_BITMASK; 1825 kvm_for_each_vcpu(i, vcpu, kvm) { 1826 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) 1827 return vcpu; 1828 } 1829 return NULL; 1830 } 1831 1832 bool kvm_arch_has_irq_bypass(void) 1833 { 1834 return true; 1835 } 1836 1837 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 1838 struct irq_bypass_producer *prod) 1839 { 1840 struct kvm_kernel_irqfd *irqfd = 1841 container_of(cons, struct kvm_kernel_irqfd, consumer); 1842 1843 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, 1844 &irqfd->irq_entry); 1845 } 1846 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 1847 struct irq_bypass_producer *prod) 1848 { 1849 struct kvm_kernel_irqfd *irqfd = 1850 container_of(cons, struct kvm_kernel_irqfd, consumer); 1851 1852 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, 1853 &irqfd->irq_entry); 1854 } 1855 1856 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) 1857 { 1858 struct kvm_kernel_irqfd *irqfd = 1859 container_of(cons, struct kvm_kernel_irqfd, consumer); 1860 1861 kvm_arm_halt_guest(irqfd->kvm); 1862 } 1863 1864 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) 1865 { 1866 struct kvm_kernel_irqfd *irqfd = 1867 container_of(cons, struct kvm_kernel_irqfd, consumer); 1868 1869 kvm_arm_resume_guest(irqfd->kvm); 1870 } 1871 1872 /** 1873 * Initialize Hyp-mode and memory mappings on all CPUs. 1874 */ 1875 int kvm_arch_init(void *opaque) 1876 { 1877 int err; 1878 int ret, cpu; 1879 bool in_hyp_mode; 1880 1881 if (!is_hyp_mode_available()) { 1882 kvm_info("HYP mode not available\n"); 1883 return -ENODEV; 1884 } 1885 1886 in_hyp_mode = is_kernel_in_hyp_mode(); 1887 1888 if (!in_hyp_mode && kvm_arch_requires_vhe()) { 1889 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n"); 1890 return -ENODEV; 1891 } 1892 1893 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) || 1894 cpus_have_final_cap(ARM64_WORKAROUND_1508412)) 1895 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \ 1896 "Only trusted guests should be used on this system.\n"); 1897 1898 for_each_online_cpu(cpu) { 1899 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); 1900 if (ret < 0) { 1901 kvm_err("Error, CPU %d not supported!\n", cpu); 1902 return -ENODEV; 1903 } 1904 } 1905 1906 err = init_common_resources(); 1907 if (err) 1908 return err; 1909 1910 err = kvm_arm_init_sve(); 1911 if (err) 1912 return err; 1913 1914 if (!in_hyp_mode) { 1915 err = init_hyp_mode(); 1916 if (err) 1917 goto out_err; 1918 } 1919 1920 err = kvm_init_vector_slots(); 1921 if (err) { 1922 kvm_err("Cannot initialise vector slots\n"); 1923 goto out_err; 1924 } 1925 1926 err = init_subsystems(); 1927 if (err) 1928 goto out_hyp; 1929 1930 if (is_protected_kvm_enabled()) { 1931 static_branch_enable(&kvm_protected_mode_initialized); 1932 kvm_info("Protected nVHE mode initialized successfully\n"); 1933 } else if (in_hyp_mode) { 1934 kvm_info("VHE mode initialized successfully\n"); 1935 } else { 1936 kvm_info("Hyp mode initialized successfully\n"); 1937 } 1938 1939 return 0; 1940 1941 out_hyp: 1942 hyp_cpu_pm_exit(); 1943 if (!in_hyp_mode) 1944 teardown_hyp_mode(); 1945 out_err: 1946 return err; 1947 } 1948 1949 /* NOP: Compiling as a module not supported */ 1950 void kvm_arch_exit(void) 1951 { 1952 kvm_perf_teardown(); 1953 } 1954 1955 static int __init early_kvm_mode_cfg(char *arg) 1956 { 1957 if (!arg) 1958 return -EINVAL; 1959 1960 if (strcmp(arg, "protected") == 0) { 1961 kvm_mode = KVM_MODE_PROTECTED; 1962 return 0; 1963 } 1964 1965 return -EINVAL; 1966 } 1967 early_param("kvm-arm.mode", early_kvm_mode_cfg); 1968 1969 enum kvm_mode kvm_get_mode(void) 1970 { 1971 return kvm_mode; 1972 } 1973 1974 static int arm_init(void) 1975 { 1976 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); 1977 return rc; 1978 } 1979 1980 module_init(arm_init); 1981