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 <trace/events/kvm.h> 23 24#define CREATE_TRACE_POINTS 25#include "trace_arm.h" 26 27#include <linux/uaccess.h> 28#include <asm/ptrace.h> 29#include <asm/mman.h> 30#include <asm/tlbflush.h> 31#include <asm/cacheflush.h> 32#include <asm/cpufeature.h> 33#include <asm/virt.h> 34#include <asm/kvm_arm.h> 35#include <asm/kvm_asm.h> 36#include <asm/kvm_mmu.h> 37#include <asm/kvm_emulate.h> 38#include <asm/kvm_coproc.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 49DECLARE_KVM_HYP_PER_CPU(unsigned long, kvm_hyp_vector); 50 51static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page); 52unsigned long kvm_arm_hyp_percpu_base[NR_CPUS]; 53 54/* The VMID used in the VTTBR */ 55static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1); 56static u32 kvm_next_vmid; 57static DEFINE_SPINLOCK(kvm_vmid_lock); 58 59static bool vgic_present; 60 61static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled); 62DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use); 63 64int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu) 65{ 66 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE; 67} 68 69int kvm_arch_hardware_setup(void *opaque) 70{ 71 return 0; 72} 73 74int kvm_arch_check_processor_compat(void *opaque) 75{ 76 return 0; 77} 78 79int kvm_vm_ioctl_enable_cap(struct kvm *kvm, 80 struct kvm_enable_cap *cap) 81{ 82 int r; 83 84 if (cap->flags) 85 return -EINVAL; 86 87 switch (cap->cap) { 88 case KVM_CAP_ARM_NISV_TO_USER: 89 r = 0; 90 kvm->arch.return_nisv_io_abort_to_user = true; 91 break; 92 default: 93 r = -EINVAL; 94 break; 95 } 96 97 return r; 98} 99 100static int kvm_arm_default_max_vcpus(void) 101{ 102 return vgic_present ? kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS; 103} 104 105static void set_default_csv2(struct kvm *kvm) 106{ 107 /* 108 * The default is to expose CSV2 == 1 if the HW isn't affected. 109 * Although this is a per-CPU feature, we make it global because 110 * asymmetric systems are just a nuisance. 111 * 112 * Userspace can override this as long as it doesn't promise 113 * the impossible. 114 */ 115 if (arm64_get_spectre_v2_state() == SPECTRE_UNAFFECTED) 116 kvm->arch.pfr0_csv2 = 1; 117} 118 119/** 120 * kvm_arch_init_vm - initializes a VM data structure 121 * @kvm: pointer to the KVM struct 122 */ 123int kvm_arch_init_vm(struct kvm *kvm, unsigned long type) 124{ 125 int ret; 126 127 ret = kvm_arm_setup_stage2(kvm, type); 128 if (ret) 129 return ret; 130 131 ret = kvm_init_stage2_mmu(kvm, &kvm->arch.mmu); 132 if (ret) 133 return ret; 134 135 ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP); 136 if (ret) 137 goto out_free_stage2_pgd; 138 139 kvm_vgic_early_init(kvm); 140 141 /* The maximum number of VCPUs is limited by the host's GIC model */ 142 kvm->arch.max_vcpus = kvm_arm_default_max_vcpus(); 143 144 set_default_csv2(kvm); 145 146 return ret; 147out_free_stage2_pgd: 148 kvm_free_stage2_pgd(&kvm->arch.mmu); 149 return ret; 150} 151 152vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf) 153{ 154 return VM_FAULT_SIGBUS; 155} 156 157 158/** 159 * kvm_arch_destroy_vm - destroy the VM data structure 160 * @kvm: pointer to the KVM struct 161 */ 162void kvm_arch_destroy_vm(struct kvm *kvm) 163{ 164 int i; 165 166 bitmap_free(kvm->arch.pmu_filter); 167 168 kvm_vgic_destroy(kvm); 169 170 for (i = 0; i < KVM_MAX_VCPUS; ++i) { 171 if (kvm->vcpus[i]) { 172 kvm_vcpu_destroy(kvm->vcpus[i]); 173 kvm->vcpus[i] = NULL; 174 } 175 } 176 atomic_set(&kvm->online_vcpus, 0); 177} 178 179int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext) 180{ 181 int r; 182 switch (ext) { 183 case KVM_CAP_IRQCHIP: 184 r = vgic_present; 185 break; 186 case KVM_CAP_IOEVENTFD: 187 case KVM_CAP_DEVICE_CTRL: 188 case KVM_CAP_USER_MEMORY: 189 case KVM_CAP_SYNC_MMU: 190 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS: 191 case KVM_CAP_ONE_REG: 192 case KVM_CAP_ARM_PSCI: 193 case KVM_CAP_ARM_PSCI_0_2: 194 case KVM_CAP_READONLY_MEM: 195 case KVM_CAP_MP_STATE: 196 case KVM_CAP_IMMEDIATE_EXIT: 197 case KVM_CAP_VCPU_EVENTS: 198 case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2: 199 case KVM_CAP_ARM_NISV_TO_USER: 200 case KVM_CAP_ARM_INJECT_EXT_DABT: 201 r = 1; 202 break; 203 case KVM_CAP_ARM_SET_DEVICE_ADDR: 204 r = 1; 205 break; 206 case KVM_CAP_NR_VCPUS: 207 r = num_online_cpus(); 208 break; 209 case KVM_CAP_MAX_VCPUS: 210 case KVM_CAP_MAX_VCPU_ID: 211 if (kvm) 212 r = kvm->arch.max_vcpus; 213 else 214 r = kvm_arm_default_max_vcpus(); 215 break; 216 case KVM_CAP_MSI_DEVID: 217 if (!kvm) 218 r = -EINVAL; 219 else 220 r = kvm->arch.vgic.msis_require_devid; 221 break; 222 case KVM_CAP_ARM_USER_IRQ: 223 /* 224 * 1: EL1_VTIMER, EL1_PTIMER, and PMU. 225 * (bump this number if adding more devices) 226 */ 227 r = 1; 228 break; 229 case KVM_CAP_STEAL_TIME: 230 r = kvm_arm_pvtime_supported(); 231 break; 232 default: 233 r = kvm_arch_vm_ioctl_check_extension(kvm, ext); 234 break; 235 } 236 return r; 237} 238 239long kvm_arch_dev_ioctl(struct file *filp, 240 unsigned int ioctl, unsigned long arg) 241{ 242 return -EINVAL; 243} 244 245struct kvm *kvm_arch_alloc_vm(void) 246{ 247 if (!has_vhe()) 248 return kzalloc(sizeof(struct kvm), GFP_KERNEL); 249 250 return vzalloc(sizeof(struct kvm)); 251} 252 253void kvm_arch_free_vm(struct kvm *kvm) 254{ 255 if (!has_vhe()) 256 kfree(kvm); 257 else 258 vfree(kvm); 259} 260 261int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id) 262{ 263 if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) 264 return -EBUSY; 265 266 if (id >= kvm->arch.max_vcpus) 267 return -EINVAL; 268 269 return 0; 270} 271 272int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu) 273{ 274 int err; 275 276 /* Force users to call KVM_ARM_VCPU_INIT */ 277 vcpu->arch.target = -1; 278 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 279 280 vcpu->arch.mmu_page_cache.gfp_zero = __GFP_ZERO; 281 282 /* Set up the timer */ 283 kvm_timer_vcpu_init(vcpu); 284 285 kvm_pmu_vcpu_init(vcpu); 286 287 kvm_arm_reset_debug_ptr(vcpu); 288 289 kvm_arm_pvtime_vcpu_init(&vcpu->arch); 290 291 vcpu->arch.hw_mmu = &vcpu->kvm->arch.mmu; 292 293 err = kvm_vgic_vcpu_init(vcpu); 294 if (err) 295 return err; 296 297 return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP); 298} 299 300void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu) 301{ 302} 303 304void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu) 305{ 306 if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm))) 307 static_branch_dec(&userspace_irqchip_in_use); 308 309 kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_cache); 310 kvm_timer_vcpu_terminate(vcpu); 311 kvm_pmu_vcpu_destroy(vcpu); 312 313 kvm_arm_vcpu_destroy(vcpu); 314} 315 316int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu) 317{ 318 return kvm_timer_is_pending(vcpu); 319} 320 321void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) 322{ 323 /* 324 * If we're about to block (most likely because we've just hit a 325 * WFI), we need to sync back the state of the GIC CPU interface 326 * so that we have the latest PMR and group enables. This ensures 327 * that kvm_arch_vcpu_runnable has up-to-date data to decide 328 * whether we have pending interrupts. 329 * 330 * For the same reason, we want to tell GICv4 that we need 331 * doorbells to be signalled, should an interrupt become pending. 332 */ 333 preempt_disable(); 334 kvm_vgic_vmcr_sync(vcpu); 335 vgic_v4_put(vcpu, true); 336 preempt_enable(); 337} 338 339void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) 340{ 341 preempt_disable(); 342 vgic_v4_load(vcpu); 343 preempt_enable(); 344} 345 346void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu) 347{ 348 struct kvm_s2_mmu *mmu; 349 int *last_ran; 350 351 mmu = vcpu->arch.hw_mmu; 352 last_ran = this_cpu_ptr(mmu->last_vcpu_ran); 353 354 /* 355 * We guarantee that both TLBs and I-cache are private to each 356 * vcpu. If detecting that a vcpu from the same VM has 357 * previously run on the same physical CPU, call into the 358 * hypervisor code to nuke the relevant contexts. 359 * 360 * We might get preempted before the vCPU actually runs, but 361 * over-invalidation doesn't affect correctness. 362 */ 363 if (*last_ran != vcpu->vcpu_id) { 364 kvm_call_hyp(__kvm_flush_cpu_context, mmu); 365 *last_ran = vcpu->vcpu_id; 366 } 367 368 vcpu->cpu = cpu; 369 370 kvm_vgic_load(vcpu); 371 kvm_timer_vcpu_load(vcpu); 372 if (has_vhe()) 373 kvm_vcpu_load_sysregs_vhe(vcpu); 374 kvm_arch_vcpu_load_fp(vcpu); 375 kvm_vcpu_pmu_restore_guest(vcpu); 376 if (kvm_arm_is_pvtime_enabled(&vcpu->arch)) 377 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu); 378 379 if (single_task_running()) 380 vcpu_clear_wfx_traps(vcpu); 381 else 382 vcpu_set_wfx_traps(vcpu); 383 384 if (vcpu_has_ptrauth(vcpu)) 385 vcpu_ptrauth_disable(vcpu); 386} 387 388void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu) 389{ 390 kvm_arch_vcpu_put_fp(vcpu); 391 if (has_vhe()) 392 kvm_vcpu_put_sysregs_vhe(vcpu); 393 kvm_timer_vcpu_put(vcpu); 394 kvm_vgic_put(vcpu); 395 kvm_vcpu_pmu_restore_host(vcpu); 396 397 vcpu->cpu = -1; 398} 399 400static void vcpu_power_off(struct kvm_vcpu *vcpu) 401{ 402 vcpu->arch.power_off = true; 403 kvm_make_request(KVM_REQ_SLEEP, vcpu); 404 kvm_vcpu_kick(vcpu); 405} 406 407int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu, 408 struct kvm_mp_state *mp_state) 409{ 410 if (vcpu->arch.power_off) 411 mp_state->mp_state = KVM_MP_STATE_STOPPED; 412 else 413 mp_state->mp_state = KVM_MP_STATE_RUNNABLE; 414 415 return 0; 416} 417 418int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu, 419 struct kvm_mp_state *mp_state) 420{ 421 int ret = 0; 422 423 switch (mp_state->mp_state) { 424 case KVM_MP_STATE_RUNNABLE: 425 vcpu->arch.power_off = false; 426 break; 427 case KVM_MP_STATE_STOPPED: 428 vcpu_power_off(vcpu); 429 break; 430 default: 431 ret = -EINVAL; 432 } 433 434 return ret; 435} 436 437/** 438 * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled 439 * @v: The VCPU pointer 440 * 441 * If the guest CPU is not waiting for interrupts or an interrupt line is 442 * asserted, the CPU is by definition runnable. 443 */ 444int kvm_arch_vcpu_runnable(struct kvm_vcpu *v) 445{ 446 bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF); 447 return ((irq_lines || kvm_vgic_vcpu_pending_irq(v)) 448 && !v->arch.power_off && !v->arch.pause); 449} 450 451bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu) 452{ 453 return vcpu_mode_priv(vcpu); 454} 455 456/* Just ensure a guest exit from a particular CPU */ 457static void exit_vm_noop(void *info) 458{ 459} 460 461void force_vm_exit(const cpumask_t *mask) 462{ 463 preempt_disable(); 464 smp_call_function_many(mask, exit_vm_noop, NULL, true); 465 preempt_enable(); 466} 467 468/** 469 * need_new_vmid_gen - check that the VMID is still valid 470 * @vmid: The VMID to check 471 * 472 * return true if there is a new generation of VMIDs being used 473 * 474 * The hardware supports a limited set of values with the value zero reserved 475 * for the host, so we check if an assigned value belongs to a previous 476 * generation, which requires us to assign a new value. If we're the first to 477 * use a VMID for the new generation, we must flush necessary caches and TLBs 478 * on all CPUs. 479 */ 480static bool need_new_vmid_gen(struct kvm_vmid *vmid) 481{ 482 u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen); 483 smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */ 484 return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen); 485} 486 487/** 488 * update_vmid - Update the vmid with a valid VMID for the current generation 489 * @vmid: The stage-2 VMID information struct 490 */ 491static void update_vmid(struct kvm_vmid *vmid) 492{ 493 if (!need_new_vmid_gen(vmid)) 494 return; 495 496 spin_lock(&kvm_vmid_lock); 497 498 /* 499 * We need to re-check the vmid_gen here to ensure that if another vcpu 500 * already allocated a valid vmid for this vm, then this vcpu should 501 * use the same vmid. 502 */ 503 if (!need_new_vmid_gen(vmid)) { 504 spin_unlock(&kvm_vmid_lock); 505 return; 506 } 507 508 /* First user of a new VMID generation? */ 509 if (unlikely(kvm_next_vmid == 0)) { 510 atomic64_inc(&kvm_vmid_gen); 511 kvm_next_vmid = 1; 512 513 /* 514 * On SMP we know no other CPUs can use this CPU's or each 515 * other's VMID after force_vm_exit returns since the 516 * kvm_vmid_lock blocks them from reentry to the guest. 517 */ 518 force_vm_exit(cpu_all_mask); 519 /* 520 * Now broadcast TLB + ICACHE invalidation over the inner 521 * shareable domain to make sure all data structures are 522 * clean. 523 */ 524 kvm_call_hyp(__kvm_flush_vm_context); 525 } 526 527 vmid->vmid = kvm_next_vmid; 528 kvm_next_vmid++; 529 kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1; 530 531 smp_wmb(); 532 WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen)); 533 534 spin_unlock(&kvm_vmid_lock); 535} 536 537static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu) 538{ 539 struct kvm *kvm = vcpu->kvm; 540 int ret = 0; 541 542 if (likely(vcpu->arch.has_run_once)) 543 return 0; 544 545 if (!kvm_arm_vcpu_is_finalized(vcpu)) 546 return -EPERM; 547 548 vcpu->arch.has_run_once = true; 549 550 kvm_arm_vcpu_init_debug(vcpu); 551 552 if (likely(irqchip_in_kernel(kvm))) { 553 /* 554 * Map the VGIC hardware resources before running a vcpu the 555 * first time on this VM. 556 */ 557 if (unlikely(!vgic_ready(kvm))) { 558 ret = kvm_vgic_map_resources(kvm); 559 if (ret) 560 return ret; 561 } 562 } else { 563 /* 564 * Tell the rest of the code that there are userspace irqchip 565 * VMs in the wild. 566 */ 567 static_branch_inc(&userspace_irqchip_in_use); 568 } 569 570 ret = kvm_timer_enable(vcpu); 571 if (ret) 572 return ret; 573 574 ret = kvm_arm_pmu_v3_enable(vcpu); 575 576 return ret; 577} 578 579bool kvm_arch_intc_initialized(struct kvm *kvm) 580{ 581 return vgic_initialized(kvm); 582} 583 584void kvm_arm_halt_guest(struct kvm *kvm) 585{ 586 int i; 587 struct kvm_vcpu *vcpu; 588 589 kvm_for_each_vcpu(i, vcpu, kvm) 590 vcpu->arch.pause = true; 591 kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP); 592} 593 594void kvm_arm_resume_guest(struct kvm *kvm) 595{ 596 int i; 597 struct kvm_vcpu *vcpu; 598 599 kvm_for_each_vcpu(i, vcpu, kvm) { 600 vcpu->arch.pause = false; 601 rcuwait_wake_up(kvm_arch_vcpu_get_wait(vcpu)); 602 } 603} 604 605static void vcpu_req_sleep(struct kvm_vcpu *vcpu) 606{ 607 struct rcuwait *wait = kvm_arch_vcpu_get_wait(vcpu); 608 609 rcuwait_wait_event(wait, 610 (!vcpu->arch.power_off) &&(!vcpu->arch.pause), 611 TASK_INTERRUPTIBLE); 612 613 if (vcpu->arch.power_off || vcpu->arch.pause) { 614 /* Awaken to handle a signal, request we sleep again later. */ 615 kvm_make_request(KVM_REQ_SLEEP, vcpu); 616 } 617 618 /* 619 * Make sure we will observe a potential reset request if we've 620 * observed a change to the power state. Pairs with the smp_wmb() in 621 * kvm_psci_vcpu_on(). 622 */ 623 smp_rmb(); 624} 625 626static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu) 627{ 628 return vcpu->arch.target >= 0; 629} 630 631static void check_vcpu_requests(struct kvm_vcpu *vcpu) 632{ 633 if (kvm_request_pending(vcpu)) { 634 if (kvm_check_request(KVM_REQ_SLEEP, vcpu)) 635 vcpu_req_sleep(vcpu); 636 637 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 638 kvm_reset_vcpu(vcpu); 639 640 /* 641 * Clear IRQ_PENDING requests that were made to guarantee 642 * that a VCPU sees new virtual interrupts. 643 */ 644 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu); 645 646 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu)) 647 kvm_update_stolen_time(vcpu); 648 649 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) { 650 /* The distributor enable bits were changed */ 651 preempt_disable(); 652 vgic_v4_put(vcpu, false); 653 vgic_v4_load(vcpu); 654 preempt_enable(); 655 } 656 } 657} 658 659/** 660 * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code 661 * @vcpu: The VCPU pointer 662 * 663 * This function is called through the VCPU_RUN ioctl called from user space. It 664 * will execute VM code in a loop until the time slice for the process is used 665 * or some emulation is needed from user space in which case the function will 666 * return with return value 0 and with the kvm_run structure filled in with the 667 * required data for the requested emulation. 668 */ 669int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu) 670{ 671 struct kvm_run *run = vcpu->run; 672 int ret; 673 674 if (unlikely(!kvm_vcpu_initialized(vcpu))) 675 return -ENOEXEC; 676 677 ret = kvm_vcpu_first_run_init(vcpu); 678 if (ret) 679 return ret; 680 681 if (run->exit_reason == KVM_EXIT_MMIO) { 682 ret = kvm_handle_mmio_return(vcpu); 683 if (ret) 684 return ret; 685 } 686 687 if (run->immediate_exit) 688 return -EINTR; 689 690 vcpu_load(vcpu); 691 692 kvm_sigset_activate(vcpu); 693 694 ret = 1; 695 run->exit_reason = KVM_EXIT_UNKNOWN; 696 while (ret > 0) { 697 /* 698 * Check conditions before entering the guest 699 */ 700 cond_resched(); 701 702 update_vmid(&vcpu->arch.hw_mmu->vmid); 703 704 check_vcpu_requests(vcpu); 705 706 /* 707 * Preparing the interrupts to be injected also 708 * involves poking the GIC, which must be done in a 709 * non-preemptible context. 710 */ 711 preempt_disable(); 712 713 kvm_pmu_flush_hwstate(vcpu); 714 715 local_irq_disable(); 716 717 kvm_vgic_flush_hwstate(vcpu); 718 719 /* 720 * Exit if we have a signal pending so that we can deliver the 721 * signal to user space. 722 */ 723 if (signal_pending(current)) { 724 ret = -EINTR; 725 run->exit_reason = KVM_EXIT_INTR; 726 } 727 728 /* 729 * If we're using a userspace irqchip, then check if we need 730 * to tell a userspace irqchip about timer or PMU level 731 * changes and if so, exit to userspace (the actual level 732 * state gets updated in kvm_timer_update_run and 733 * kvm_pmu_update_run below). 734 */ 735 if (static_branch_unlikely(&userspace_irqchip_in_use)) { 736 if (kvm_timer_should_notify_user(vcpu) || 737 kvm_pmu_should_notify_user(vcpu)) { 738 ret = -EINTR; 739 run->exit_reason = KVM_EXIT_INTR; 740 } 741 } 742 743 /* 744 * Ensure we set mode to IN_GUEST_MODE after we disable 745 * interrupts and before the final VCPU requests check. 746 * See the comment in kvm_vcpu_exiting_guest_mode() and 747 * Documentation/virt/kvm/vcpu-requests.rst 748 */ 749 smp_store_mb(vcpu->mode, IN_GUEST_MODE); 750 751 if (ret <= 0 || need_new_vmid_gen(&vcpu->arch.hw_mmu->vmid) || 752 kvm_request_pending(vcpu)) { 753 vcpu->mode = OUTSIDE_GUEST_MODE; 754 isb(); /* Ensure work in x_flush_hwstate is committed */ 755 kvm_pmu_sync_hwstate(vcpu); 756 if (static_branch_unlikely(&userspace_irqchip_in_use)) 757 kvm_timer_sync_user(vcpu); 758 kvm_vgic_sync_hwstate(vcpu); 759 local_irq_enable(); 760 preempt_enable(); 761 continue; 762 } 763 764 kvm_arm_setup_debug(vcpu); 765 766 /************************************************************** 767 * Enter the guest 768 */ 769 trace_kvm_entry(*vcpu_pc(vcpu)); 770 guest_enter_irqoff(); 771 772 ret = kvm_call_hyp_ret(__kvm_vcpu_run, vcpu); 773 774 vcpu->mode = OUTSIDE_GUEST_MODE; 775 vcpu->stat.exits++; 776 /* 777 * Back from guest 778 *************************************************************/ 779 780 kvm_arm_clear_debug(vcpu); 781 782 /* 783 * We must sync the PMU state before the vgic state so 784 * that the vgic can properly sample the updated state of the 785 * interrupt line. 786 */ 787 kvm_pmu_sync_hwstate(vcpu); 788 789 /* 790 * Sync the vgic state before syncing the timer state because 791 * the timer code needs to know if the virtual timer 792 * interrupts are active. 793 */ 794 kvm_vgic_sync_hwstate(vcpu); 795 796 /* 797 * Sync the timer hardware state before enabling interrupts as 798 * we don't want vtimer interrupts to race with syncing the 799 * timer virtual interrupt state. 800 */ 801 if (static_branch_unlikely(&userspace_irqchip_in_use)) 802 kvm_timer_sync_user(vcpu); 803 804 kvm_arch_vcpu_ctxsync_fp(vcpu); 805 806 /* 807 * We may have taken a host interrupt in HYP mode (ie 808 * while executing the guest). This interrupt is still 809 * pending, as we haven't serviced it yet! 810 * 811 * We're now back in SVC mode, with interrupts 812 * disabled. Enabling the interrupts now will have 813 * the effect of taking the interrupt again, in SVC 814 * mode this time. 815 */ 816 local_irq_enable(); 817 818 /* 819 * We do local_irq_enable() before calling guest_exit() so 820 * that if a timer interrupt hits while running the guest we 821 * account that tick as being spent in the guest. We enable 822 * preemption after calling guest_exit() so that if we get 823 * preempted we make sure ticks after that is not counted as 824 * guest time. 825 */ 826 guest_exit(); 827 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu)); 828 829 /* Exit types that need handling before we can be preempted */ 830 handle_exit_early(vcpu, ret); 831 832 preempt_enable(); 833 834 /* 835 * The ARMv8 architecture doesn't give the hypervisor 836 * a mechanism to prevent a guest from dropping to AArch32 EL0 837 * if implemented by the CPU. If we spot the guest in such 838 * state and that we decided it wasn't supposed to do so (like 839 * with the asymmetric AArch32 case), return to userspace with 840 * a fatal error. 841 */ 842 if (!system_supports_32bit_el0() && vcpu_mode_is_32bit(vcpu)) { 843 /* 844 * As we have caught the guest red-handed, decide that 845 * it isn't fit for purpose anymore by making the vcpu 846 * invalid. The VMM can try and fix it by issuing a 847 * KVM_ARM_VCPU_INIT if it really wants to. 848 */ 849 vcpu->arch.target = -1; 850 ret = ARM_EXCEPTION_IL; 851 } 852 853 ret = handle_exit(vcpu, ret); 854 } 855 856 /* Tell userspace about in-kernel device output levels */ 857 if (unlikely(!irqchip_in_kernel(vcpu->kvm))) { 858 kvm_timer_update_run(vcpu); 859 kvm_pmu_update_run(vcpu); 860 } 861 862 kvm_sigset_deactivate(vcpu); 863 864 vcpu_put(vcpu); 865 return ret; 866} 867 868static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level) 869{ 870 int bit_index; 871 bool set; 872 unsigned long *hcr; 873 874 if (number == KVM_ARM_IRQ_CPU_IRQ) 875 bit_index = __ffs(HCR_VI); 876 else /* KVM_ARM_IRQ_CPU_FIQ */ 877 bit_index = __ffs(HCR_VF); 878 879 hcr = vcpu_hcr(vcpu); 880 if (level) 881 set = test_and_set_bit(bit_index, hcr); 882 else 883 set = test_and_clear_bit(bit_index, hcr); 884 885 /* 886 * If we didn't change anything, no need to wake up or kick other CPUs 887 */ 888 if (set == level) 889 return 0; 890 891 /* 892 * The vcpu irq_lines field was updated, wake up sleeping VCPUs and 893 * trigger a world-switch round on the running physical CPU to set the 894 * virtual IRQ/FIQ fields in the HCR appropriately. 895 */ 896 kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu); 897 kvm_vcpu_kick(vcpu); 898 899 return 0; 900} 901 902int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level, 903 bool line_status) 904{ 905 u32 irq = irq_level->irq; 906 unsigned int irq_type, vcpu_idx, irq_num; 907 int nrcpus = atomic_read(&kvm->online_vcpus); 908 struct kvm_vcpu *vcpu = NULL; 909 bool level = irq_level->level; 910 911 irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK; 912 vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK; 913 vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1); 914 irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK; 915 916 trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level); 917 918 switch (irq_type) { 919 case KVM_ARM_IRQ_TYPE_CPU: 920 if (irqchip_in_kernel(kvm)) 921 return -ENXIO; 922 923 if (vcpu_idx >= nrcpus) 924 return -EINVAL; 925 926 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 927 if (!vcpu) 928 return -EINVAL; 929 930 if (irq_num > KVM_ARM_IRQ_CPU_FIQ) 931 return -EINVAL; 932 933 return vcpu_interrupt_line(vcpu, irq_num, level); 934 case KVM_ARM_IRQ_TYPE_PPI: 935 if (!irqchip_in_kernel(kvm)) 936 return -ENXIO; 937 938 if (vcpu_idx >= nrcpus) 939 return -EINVAL; 940 941 vcpu = kvm_get_vcpu(kvm, vcpu_idx); 942 if (!vcpu) 943 return -EINVAL; 944 945 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS) 946 return -EINVAL; 947 948 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL); 949 case KVM_ARM_IRQ_TYPE_SPI: 950 if (!irqchip_in_kernel(kvm)) 951 return -ENXIO; 952 953 if (irq_num < VGIC_NR_PRIVATE_IRQS) 954 return -EINVAL; 955 956 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL); 957 } 958 959 return -EINVAL; 960} 961 962static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu, 963 const struct kvm_vcpu_init *init) 964{ 965 unsigned int i, ret; 966 int phys_target = kvm_target_cpu(); 967 968 if (init->target != phys_target) 969 return -EINVAL; 970 971 /* 972 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 973 * use the same target. 974 */ 975 if (vcpu->arch.target != -1 && vcpu->arch.target != init->target) 976 return -EINVAL; 977 978 /* -ENOENT for unknown features, -EINVAL for invalid combinations. */ 979 for (i = 0; i < sizeof(init->features) * 8; i++) { 980 bool set = (init->features[i / 32] & (1 << (i % 32))); 981 982 if (set && i >= KVM_VCPU_MAX_FEATURES) 983 return -ENOENT; 984 985 /* 986 * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must 987 * use the same feature set. 988 */ 989 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES && 990 test_bit(i, vcpu->arch.features) != set) 991 return -EINVAL; 992 993 if (set) 994 set_bit(i, vcpu->arch.features); 995 } 996 997 vcpu->arch.target = phys_target; 998 999 /* Now we know what it is, we can reset it. */ 1000 ret = kvm_reset_vcpu(vcpu); 1001 if (ret) { 1002 vcpu->arch.target = -1; 1003 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES); 1004 } 1005 1006 return ret; 1007} 1008 1009static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu, 1010 struct kvm_vcpu_init *init) 1011{ 1012 int ret; 1013 1014 ret = kvm_vcpu_set_target(vcpu, init); 1015 if (ret) 1016 return ret; 1017 1018 /* 1019 * Ensure a rebooted VM will fault in RAM pages and detect if the 1020 * guest MMU is turned off and flush the caches as needed. 1021 * 1022 * S2FWB enforces all memory accesses to RAM being cacheable, 1023 * ensuring that the data side is always coherent. We still 1024 * need to invalidate the I-cache though, as FWB does *not* 1025 * imply CTR_EL0.DIC. 1026 */ 1027 if (vcpu->arch.has_run_once) { 1028 if (!cpus_have_final_cap(ARM64_HAS_STAGE2_FWB)) 1029 stage2_unmap_vm(vcpu->kvm); 1030 else 1031 __flush_icache_all(); 1032 } 1033 1034 vcpu_reset_hcr(vcpu); 1035 1036 /* 1037 * Handle the "start in power-off" case. 1038 */ 1039 if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) 1040 vcpu_power_off(vcpu); 1041 else 1042 vcpu->arch.power_off = false; 1043 1044 return 0; 1045} 1046 1047static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu, 1048 struct kvm_device_attr *attr) 1049{ 1050 int ret = -ENXIO; 1051 1052 switch (attr->group) { 1053 default: 1054 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr); 1055 break; 1056 } 1057 1058 return ret; 1059} 1060 1061static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu, 1062 struct kvm_device_attr *attr) 1063{ 1064 int ret = -ENXIO; 1065 1066 switch (attr->group) { 1067 default: 1068 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr); 1069 break; 1070 } 1071 1072 return ret; 1073} 1074 1075static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu, 1076 struct kvm_device_attr *attr) 1077{ 1078 int ret = -ENXIO; 1079 1080 switch (attr->group) { 1081 default: 1082 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr); 1083 break; 1084 } 1085 1086 return ret; 1087} 1088 1089static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu, 1090 struct kvm_vcpu_events *events) 1091{ 1092 memset(events, 0, sizeof(*events)); 1093 1094 return __kvm_arm_vcpu_get_events(vcpu, events); 1095} 1096 1097static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu, 1098 struct kvm_vcpu_events *events) 1099{ 1100 int i; 1101 1102 /* check whether the reserved field is zero */ 1103 for (i = 0; i < ARRAY_SIZE(events->reserved); i++) 1104 if (events->reserved[i]) 1105 return -EINVAL; 1106 1107 /* check whether the pad field is zero */ 1108 for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++) 1109 if (events->exception.pad[i]) 1110 return -EINVAL; 1111 1112 return __kvm_arm_vcpu_set_events(vcpu, events); 1113} 1114 1115long kvm_arch_vcpu_ioctl(struct file *filp, 1116 unsigned int ioctl, unsigned long arg) 1117{ 1118 struct kvm_vcpu *vcpu = filp->private_data; 1119 void __user *argp = (void __user *)arg; 1120 struct kvm_device_attr attr; 1121 long r; 1122 1123 switch (ioctl) { 1124 case KVM_ARM_VCPU_INIT: { 1125 struct kvm_vcpu_init init; 1126 1127 r = -EFAULT; 1128 if (copy_from_user(&init, argp, sizeof(init))) 1129 break; 1130 1131 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init); 1132 break; 1133 } 1134 case KVM_SET_ONE_REG: 1135 case KVM_GET_ONE_REG: { 1136 struct kvm_one_reg reg; 1137 1138 r = -ENOEXEC; 1139 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1140 break; 1141 1142 r = -EFAULT; 1143 if (copy_from_user(®, argp, sizeof(reg))) 1144 break; 1145 1146 /* 1147 * We could owe a reset due to PSCI. Handle the pending reset 1148 * here to ensure userspace register accesses are ordered after 1149 * the reset. 1150 */ 1151 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu)) 1152 kvm_reset_vcpu(vcpu); 1153 1154 if (ioctl == KVM_SET_ONE_REG) 1155 r = kvm_arm_set_reg(vcpu, ®); 1156 else 1157 r = kvm_arm_get_reg(vcpu, ®); 1158 break; 1159 } 1160 case KVM_GET_REG_LIST: { 1161 struct kvm_reg_list __user *user_list = argp; 1162 struct kvm_reg_list reg_list; 1163 unsigned n; 1164 1165 r = -ENOEXEC; 1166 if (unlikely(!kvm_vcpu_initialized(vcpu))) 1167 break; 1168 1169 r = -EPERM; 1170 if (!kvm_arm_vcpu_is_finalized(vcpu)) 1171 break; 1172 1173 r = -EFAULT; 1174 if (copy_from_user(®_list, user_list, sizeof(reg_list))) 1175 break; 1176 n = reg_list.n; 1177 reg_list.n = kvm_arm_num_regs(vcpu); 1178 if (copy_to_user(user_list, ®_list, sizeof(reg_list))) 1179 break; 1180 r = -E2BIG; 1181 if (n < reg_list.n) 1182 break; 1183 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg); 1184 break; 1185 } 1186 case KVM_SET_DEVICE_ATTR: { 1187 r = -EFAULT; 1188 if (copy_from_user(&attr, argp, sizeof(attr))) 1189 break; 1190 r = kvm_arm_vcpu_set_attr(vcpu, &attr); 1191 break; 1192 } 1193 case KVM_GET_DEVICE_ATTR: { 1194 r = -EFAULT; 1195 if (copy_from_user(&attr, argp, sizeof(attr))) 1196 break; 1197 r = kvm_arm_vcpu_get_attr(vcpu, &attr); 1198 break; 1199 } 1200 case KVM_HAS_DEVICE_ATTR: { 1201 r = -EFAULT; 1202 if (copy_from_user(&attr, argp, sizeof(attr))) 1203 break; 1204 r = kvm_arm_vcpu_has_attr(vcpu, &attr); 1205 break; 1206 } 1207 case KVM_GET_VCPU_EVENTS: { 1208 struct kvm_vcpu_events events; 1209 1210 if (kvm_arm_vcpu_get_events(vcpu, &events)) 1211 return -EINVAL; 1212 1213 if (copy_to_user(argp, &events, sizeof(events))) 1214 return -EFAULT; 1215 1216 return 0; 1217 } 1218 case KVM_SET_VCPU_EVENTS: { 1219 struct kvm_vcpu_events events; 1220 1221 if (copy_from_user(&events, argp, sizeof(events))) 1222 return -EFAULT; 1223 1224 return kvm_arm_vcpu_set_events(vcpu, &events); 1225 } 1226 case KVM_ARM_VCPU_FINALIZE: { 1227 int what; 1228 1229 if (!kvm_vcpu_initialized(vcpu)) 1230 return -ENOEXEC; 1231 1232 if (get_user(what, (const int __user *)argp)) 1233 return -EFAULT; 1234 1235 return kvm_arm_vcpu_finalize(vcpu, what); 1236 } 1237 default: 1238 r = -EINVAL; 1239 } 1240 1241 return r; 1242} 1243 1244void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot) 1245{ 1246 1247} 1248 1249void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm, 1250 struct kvm_memory_slot *memslot) 1251{ 1252 kvm_flush_remote_tlbs(kvm); 1253} 1254 1255static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm, 1256 struct kvm_arm_device_addr *dev_addr) 1257{ 1258 unsigned long dev_id, type; 1259 1260 dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >> 1261 KVM_ARM_DEVICE_ID_SHIFT; 1262 type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >> 1263 KVM_ARM_DEVICE_TYPE_SHIFT; 1264 1265 switch (dev_id) { 1266 case KVM_ARM_DEVICE_VGIC_V2: 1267 if (!vgic_present) 1268 return -ENXIO; 1269 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true); 1270 default: 1271 return -ENODEV; 1272 } 1273} 1274 1275long kvm_arch_vm_ioctl(struct file *filp, 1276 unsigned int ioctl, unsigned long arg) 1277{ 1278 struct kvm *kvm = filp->private_data; 1279 void __user *argp = (void __user *)arg; 1280 1281 switch (ioctl) { 1282 case KVM_CREATE_IRQCHIP: { 1283 int ret; 1284 if (!vgic_present) 1285 return -ENXIO; 1286 mutex_lock(&kvm->lock); 1287 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2); 1288 mutex_unlock(&kvm->lock); 1289 return ret; 1290 } 1291 case KVM_ARM_SET_DEVICE_ADDR: { 1292 struct kvm_arm_device_addr dev_addr; 1293 1294 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr))) 1295 return -EFAULT; 1296 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr); 1297 } 1298 case KVM_ARM_PREFERRED_TARGET: { 1299 int err; 1300 struct kvm_vcpu_init init; 1301 1302 err = kvm_vcpu_preferred_target(&init); 1303 if (err) 1304 return err; 1305 1306 if (copy_to_user(argp, &init, sizeof(init))) 1307 return -EFAULT; 1308 1309 return 0; 1310 } 1311 default: 1312 return -EINVAL; 1313 } 1314} 1315 1316static unsigned long nvhe_percpu_size(void) 1317{ 1318 return (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_end) - 1319 (unsigned long)CHOOSE_NVHE_SYM(__per_cpu_start); 1320} 1321 1322static unsigned long nvhe_percpu_order(void) 1323{ 1324 unsigned long size = nvhe_percpu_size(); 1325 1326 return size ? get_order(size) : 0; 1327} 1328 1329static int kvm_map_vectors(void) 1330{ 1331 /* 1332 * SV2 = ARM64_SPECTRE_V2 1333 * HEL2 = ARM64_HARDEN_EL2_VECTORS 1334 * 1335 * !SV2 + !HEL2 -> use direct vectors 1336 * SV2 + !HEL2 -> use hardened vectors in place 1337 * !SV2 + HEL2 -> allocate one vector slot and use exec mapping 1338 * SV2 + HEL2 -> use hardened vectors and use exec mapping 1339 */ 1340 if (cpus_have_const_cap(ARM64_SPECTRE_V2) || 1341 cpus_have_const_cap(ARM64_SPECTRE_BHB)) { 1342 __kvm_bp_vect_base = kvm_ksym_ref(__bp_harden_hyp_vecs); 1343 __kvm_bp_vect_base = kern_hyp_va(__kvm_bp_vect_base); 1344 } 1345 1346 if (cpus_have_const_cap(ARM64_HARDEN_EL2_VECTORS)) { 1347 phys_addr_t vect_pa = __pa_symbol(__bp_harden_hyp_vecs); 1348 unsigned long size = __BP_HARDEN_HYP_VECS_SZ; 1349 1350 /* 1351 * Always allocate a spare vector slot, as we don't 1352 * know yet which CPUs have a BP hardening slot that 1353 * we can reuse. 1354 */ 1355 __kvm_harden_el2_vector_slot = atomic_inc_return(&arm64_el2_vector_last_slot); 1356 BUG_ON(__kvm_harden_el2_vector_slot >= BP_HARDEN_EL2_SLOTS); 1357 return create_hyp_exec_mappings(vect_pa, size, 1358 &__kvm_bp_vect_base); 1359 } 1360 1361 return 0; 1362} 1363 1364static void cpu_init_hyp_mode(void) 1365{ 1366 phys_addr_t pgd_ptr; 1367 unsigned long hyp_stack_ptr; 1368 unsigned long vector_ptr; 1369 unsigned long tpidr_el2; 1370 struct arm_smccc_res res; 1371 1372 /* Switch from the HYP stub to our own HYP init vector */ 1373 __hyp_set_vectors(kvm_get_idmap_vector()); 1374 1375 /* 1376 * Calculate the raw per-cpu offset without a translation from the 1377 * kernel's mapping to the linear mapping, and store it in tpidr_el2 1378 * so that we can use adr_l to access per-cpu variables in EL2. 1379 */ 1380 tpidr_el2 = (unsigned long)this_cpu_ptr_nvhe_sym(__per_cpu_start) - 1381 (unsigned long)kvm_ksym_ref(CHOOSE_NVHE_SYM(__per_cpu_start)); 1382 1383 pgd_ptr = kvm_mmu_get_httbr(); 1384 hyp_stack_ptr = __this_cpu_read(kvm_arm_hyp_stack_page) + PAGE_SIZE; 1385 hyp_stack_ptr = kern_hyp_va(hyp_stack_ptr); 1386 vector_ptr = (unsigned long)kern_hyp_va(kvm_ksym_ref(__kvm_hyp_host_vector)); 1387 1388 /* 1389 * Call initialization code, and switch to the full blown HYP code. 1390 * If the cpucaps haven't been finalized yet, something has gone very 1391 * wrong, and hyp will crash and burn when it uses any 1392 * cpus_have_const_cap() wrapper. 1393 */ 1394 BUG_ON(!system_capabilities_finalized()); 1395 arm_smccc_1_1_hvc(KVM_HOST_SMCCC_FUNC(__kvm_hyp_init), 1396 pgd_ptr, tpidr_el2, hyp_stack_ptr, vector_ptr, &res); 1397 WARN_ON(res.a0 != SMCCC_RET_SUCCESS); 1398 1399 /* 1400 * Disabling SSBD on a non-VHE system requires us to enable SSBS 1401 * at EL2. 1402 */ 1403 if (this_cpu_has_cap(ARM64_SSBS) && 1404 arm64_get_spectre_v4_state() == SPECTRE_VULNERABLE) { 1405 kvm_call_hyp_nvhe(__kvm_enable_ssbs); 1406 } 1407} 1408 1409static void cpu_hyp_reset(void) 1410{ 1411 if (!is_kernel_in_hyp_mode()) 1412 __hyp_reset_vectors(); 1413} 1414 1415static void cpu_hyp_reinit(void) 1416{ 1417 kvm_init_host_cpu_context(&this_cpu_ptr_hyp_sym(kvm_host_data)->host_ctxt); 1418 1419 cpu_hyp_reset(); 1420 1421 *this_cpu_ptr_hyp_sym(kvm_hyp_vector) = (unsigned long)kvm_get_hyp_vector(); 1422 1423 if (is_kernel_in_hyp_mode()) 1424 kvm_timer_init_vhe(); 1425 else 1426 cpu_init_hyp_mode(); 1427 1428 kvm_arm_init_debug(); 1429 1430 if (vgic_present) 1431 kvm_vgic_init_cpu_hardware(); 1432} 1433 1434static void _kvm_arch_hardware_enable(void *discard) 1435{ 1436 if (!__this_cpu_read(kvm_arm_hardware_enabled)) { 1437 cpu_hyp_reinit(); 1438 __this_cpu_write(kvm_arm_hardware_enabled, 1); 1439 } 1440} 1441 1442int kvm_arch_hardware_enable(void) 1443{ 1444 _kvm_arch_hardware_enable(NULL); 1445 return 0; 1446} 1447 1448static void _kvm_arch_hardware_disable(void *discard) 1449{ 1450 if (__this_cpu_read(kvm_arm_hardware_enabled)) { 1451 cpu_hyp_reset(); 1452 __this_cpu_write(kvm_arm_hardware_enabled, 0); 1453 } 1454} 1455 1456void kvm_arch_hardware_disable(void) 1457{ 1458 _kvm_arch_hardware_disable(NULL); 1459} 1460 1461#ifdef CONFIG_CPU_PM 1462static int hyp_init_cpu_pm_notifier(struct notifier_block *self, 1463 unsigned long cmd, 1464 void *v) 1465{ 1466 /* 1467 * kvm_arm_hardware_enabled is left with its old value over 1468 * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should 1469 * re-enable hyp. 1470 */ 1471 switch (cmd) { 1472 case CPU_PM_ENTER: 1473 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1474 /* 1475 * don't update kvm_arm_hardware_enabled here 1476 * so that the hardware will be re-enabled 1477 * when we resume. See below. 1478 */ 1479 cpu_hyp_reset(); 1480 1481 return NOTIFY_OK; 1482 case CPU_PM_ENTER_FAILED: 1483 case CPU_PM_EXIT: 1484 if (__this_cpu_read(kvm_arm_hardware_enabled)) 1485 /* The hardware was enabled before suspend. */ 1486 cpu_hyp_reinit(); 1487 1488 return NOTIFY_OK; 1489 1490 default: 1491 return NOTIFY_DONE; 1492 } 1493} 1494 1495static struct notifier_block hyp_init_cpu_pm_nb = { 1496 .notifier_call = hyp_init_cpu_pm_notifier, 1497}; 1498 1499static void __init hyp_cpu_pm_init(void) 1500{ 1501 cpu_pm_register_notifier(&hyp_init_cpu_pm_nb); 1502} 1503static void __init hyp_cpu_pm_exit(void) 1504{ 1505 cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb); 1506} 1507#else 1508static inline void hyp_cpu_pm_init(void) 1509{ 1510} 1511static inline void hyp_cpu_pm_exit(void) 1512{ 1513} 1514#endif 1515 1516static int init_common_resources(void) 1517{ 1518 return kvm_set_ipa_limit(); 1519} 1520 1521static int init_subsystems(void) 1522{ 1523 int err = 0; 1524 1525 /* 1526 * Enable hardware so that subsystem initialisation can access EL2. 1527 */ 1528 on_each_cpu(_kvm_arch_hardware_enable, NULL, 1); 1529 1530 /* 1531 * Register CPU lower-power notifier 1532 */ 1533 hyp_cpu_pm_init(); 1534 1535 /* 1536 * Init HYP view of VGIC 1537 */ 1538 err = kvm_vgic_hyp_init(); 1539 switch (err) { 1540 case 0: 1541 vgic_present = true; 1542 break; 1543 case -ENODEV: 1544 case -ENXIO: 1545 vgic_present = false; 1546 err = 0; 1547 break; 1548 default: 1549 goto out; 1550 } 1551 1552 /* 1553 * Init HYP architected timer support 1554 */ 1555 err = kvm_timer_hyp_init(vgic_present); 1556 if (err) 1557 goto out; 1558 1559 kvm_perf_init(); 1560 kvm_coproc_table_init(); 1561 1562out: 1563 on_each_cpu(_kvm_arch_hardware_disable, NULL, 1); 1564 1565 return err; 1566} 1567 1568static void teardown_hyp_mode(void) 1569{ 1570 int cpu; 1571 1572 free_hyp_pgds(); 1573 for_each_possible_cpu(cpu) { 1574 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu)); 1575 free_pages(kvm_arm_hyp_percpu_base[cpu], nvhe_percpu_order()); 1576 } 1577} 1578 1579/** 1580 * Inits Hyp-mode on all online CPUs 1581 */ 1582static int init_hyp_mode(void) 1583{ 1584 int cpu; 1585 int err = 0; 1586 1587 /* 1588 * Allocate Hyp PGD and setup Hyp identity mapping 1589 */ 1590 err = kvm_mmu_init(); 1591 if (err) 1592 goto out_err; 1593 1594 /* 1595 * Allocate stack pages for Hypervisor-mode 1596 */ 1597 for_each_possible_cpu(cpu) { 1598 unsigned long stack_page; 1599 1600 stack_page = __get_free_page(GFP_KERNEL); 1601 if (!stack_page) { 1602 err = -ENOMEM; 1603 goto out_err; 1604 } 1605 1606 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page; 1607 } 1608 1609 /* 1610 * Allocate and initialize pages for Hypervisor-mode percpu regions. 1611 */ 1612 for_each_possible_cpu(cpu) { 1613 struct page *page; 1614 void *page_addr; 1615 1616 page = alloc_pages(GFP_KERNEL, nvhe_percpu_order()); 1617 if (!page) { 1618 err = -ENOMEM; 1619 goto out_err; 1620 } 1621 1622 page_addr = page_address(page); 1623 memcpy(page_addr, CHOOSE_NVHE_SYM(__per_cpu_start), nvhe_percpu_size()); 1624 kvm_arm_hyp_percpu_base[cpu] = (unsigned long)page_addr; 1625 } 1626 1627 /* 1628 * Map the Hyp-code called directly from the host 1629 */ 1630 err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start), 1631 kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC); 1632 if (err) { 1633 kvm_err("Cannot map world-switch code\n"); 1634 goto out_err; 1635 } 1636 1637 err = create_hyp_mappings(kvm_ksym_ref(__start_rodata), 1638 kvm_ksym_ref(__end_rodata), PAGE_HYP_RO); 1639 if (err) { 1640 kvm_err("Cannot map rodata section\n"); 1641 goto out_err; 1642 } 1643 1644 err = create_hyp_mappings(kvm_ksym_ref(__bss_start), 1645 kvm_ksym_ref(__bss_stop), PAGE_HYP_RO); 1646 if (err) { 1647 kvm_err("Cannot map bss section\n"); 1648 goto out_err; 1649 } 1650 1651 err = kvm_map_vectors(); 1652 if (err) { 1653 kvm_err("Cannot map vectors\n"); 1654 goto out_err; 1655 } 1656 1657 /* 1658 * Map the Hyp stack pages 1659 */ 1660 for_each_possible_cpu(cpu) { 1661 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu); 1662 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE, 1663 PAGE_HYP); 1664 1665 if (err) { 1666 kvm_err("Cannot map hyp stack\n"); 1667 goto out_err; 1668 } 1669 } 1670 1671 /* 1672 * Map Hyp percpu pages 1673 */ 1674 for_each_possible_cpu(cpu) { 1675 char *percpu_begin = (char *)kvm_arm_hyp_percpu_base[cpu]; 1676 char *percpu_end = percpu_begin + nvhe_percpu_size(); 1677 1678 err = create_hyp_mappings(percpu_begin, percpu_end, PAGE_HYP); 1679 1680 if (err) { 1681 kvm_err("Cannot map hyp percpu region\n"); 1682 goto out_err; 1683 } 1684 } 1685 1686 return 0; 1687 1688out_err: 1689 teardown_hyp_mode(); 1690 kvm_err("error initializing Hyp mode: %d\n", err); 1691 return err; 1692} 1693 1694static void check_kvm_target_cpu(void *ret) 1695{ 1696 *(int *)ret = kvm_target_cpu(); 1697} 1698 1699struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr) 1700{ 1701 struct kvm_vcpu *vcpu; 1702 int i; 1703 1704 mpidr &= MPIDR_HWID_BITMASK; 1705 kvm_for_each_vcpu(i, vcpu, kvm) { 1706 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu)) 1707 return vcpu; 1708 } 1709 return NULL; 1710} 1711 1712bool kvm_arch_has_irq_bypass(void) 1713{ 1714 return true; 1715} 1716 1717int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons, 1718 struct irq_bypass_producer *prod) 1719{ 1720 struct kvm_kernel_irqfd *irqfd = 1721 container_of(cons, struct kvm_kernel_irqfd, consumer); 1722 1723 return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq, 1724 &irqfd->irq_entry); 1725} 1726void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons, 1727 struct irq_bypass_producer *prod) 1728{ 1729 struct kvm_kernel_irqfd *irqfd = 1730 container_of(cons, struct kvm_kernel_irqfd, consumer); 1731 1732 kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq, 1733 &irqfd->irq_entry); 1734} 1735 1736void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons) 1737{ 1738 struct kvm_kernel_irqfd *irqfd = 1739 container_of(cons, struct kvm_kernel_irqfd, consumer); 1740 1741 kvm_arm_halt_guest(irqfd->kvm); 1742} 1743 1744void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons) 1745{ 1746 struct kvm_kernel_irqfd *irqfd = 1747 container_of(cons, struct kvm_kernel_irqfd, consumer); 1748 1749 kvm_arm_resume_guest(irqfd->kvm); 1750} 1751 1752/** 1753 * Initialize Hyp-mode and memory mappings on all CPUs. 1754 */ 1755int kvm_arch_init(void *opaque) 1756{ 1757 int err; 1758 int ret, cpu; 1759 bool in_hyp_mode; 1760 1761 if (!is_hyp_mode_available()) { 1762 kvm_info("HYP mode not available\n"); 1763 return -ENODEV; 1764 } 1765 1766 in_hyp_mode = is_kernel_in_hyp_mode(); 1767 1768 if (!in_hyp_mode && kvm_arch_requires_vhe()) { 1769 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n"); 1770 return -ENODEV; 1771 } 1772 1773 if (cpus_have_final_cap(ARM64_WORKAROUND_DEVICE_LOAD_ACQUIRE) || 1774 cpus_have_final_cap(ARM64_WORKAROUND_1508412)) 1775 kvm_info("Guests without required CPU erratum workarounds can deadlock system!\n" \ 1776 "Only trusted guests should be used on this system.\n"); 1777 1778 for_each_online_cpu(cpu) { 1779 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1); 1780 if (ret < 0) { 1781 kvm_err("Error, CPU %d not supported!\n", cpu); 1782 return -ENODEV; 1783 } 1784 } 1785 1786 err = init_common_resources(); 1787 if (err) 1788 return err; 1789 1790 err = kvm_arm_init_sve(); 1791 if (err) 1792 return err; 1793 1794 if (!in_hyp_mode) { 1795 err = init_hyp_mode(); 1796 if (err) 1797 goto out_err; 1798 } 1799 1800 err = init_subsystems(); 1801 if (err) 1802 goto out_hyp; 1803 1804 if (in_hyp_mode) 1805 kvm_info("VHE mode initialized successfully\n"); 1806 else 1807 kvm_info("Hyp mode initialized successfully\n"); 1808 1809 return 0; 1810 1811out_hyp: 1812 hyp_cpu_pm_exit(); 1813 if (!in_hyp_mode) 1814 teardown_hyp_mode(); 1815out_err: 1816 return err; 1817} 1818 1819/* NOP: Compiling as a module not supported */ 1820void kvm_arch_exit(void) 1821{ 1822 kvm_perf_teardown(); 1823} 1824 1825static int arm_init(void) 1826{ 1827 int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE); 1828 return rc; 1829} 1830 1831module_init(arm_init); 1832