virt/kvm/arm/arm.c

   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.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
  49 DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data);
  50 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  51
  52 /* The VMID used in the VTTBR */
  53 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  54 static u32 kvm_next_vmid;
  55 static DEFINE_SPINLOCK(kvm_vmid_lock);
  56
  57 static bool vgic_present;
  58
  59 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  60 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  61
  62 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  63 {
  64         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  65 }
  66
  67 int kvm_arch_hardware_setup(void *opaque)
  68 {
  69         return 0;
  70 }
  71
  72 int kvm_arch_check_processor_compat(void *opaque)
  73 {
  74         return 0;
  75 }
  76
  77 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
  78                             struct kvm_enable_cap *cap)
  79 {
  80         int r;
  81
  82         if (cap->flags)
  83                 return -EINVAL;
  84
  85         switch (cap->cap) {
  86         case KVM_CAP_ARM_NISV_TO_USER:
  87                 r = 0;
  88                 kvm->arch.return_nisv_io_abort_to_user = true;
  89                 break;
  90         default:
  91                 r = -EINVAL;
  92                 break;
  93         }
  94
  95         return r;
  96 }
  97
  98 /**
  99  * kvm_arch_init_vm - initializes a VM data structure
 100  * @kvm:        pointer to the KVM struct
 101  */
 102 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 103 {
 104         int ret, cpu;
 105
 106         ret = kvm_arm_setup_stage2(kvm, type);
 107         if (ret)
 108                 return ret;
 109
 110         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 111         if (!kvm->arch.last_vcpu_ran)
 112                 return -ENOMEM;
 113
 114         for_each_possible_cpu(cpu)
 115                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 116
 117         ret = kvm_alloc_stage2_pgd(kvm);
 118         if (ret)
 119                 goto out_fail_alloc;
 120
 121         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 122         if (ret)
 123                 goto out_free_stage2_pgd;
 124
 125         kvm_vgic_early_init(kvm);
 126
 127         /* Mark the initial VMID generation invalid */
 128         kvm->arch.vmid.vmid_gen = 0;
 129
 130         /* The maximum number of VCPUs is limited by the host's GIC model */
 131         kvm->arch.max_vcpus = vgic_present ?
 132                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 133
 134         return ret;
 135 out_free_stage2_pgd:
 136         kvm_free_stage2_pgd(kvm);
 137 out_fail_alloc:
 138         free_percpu(kvm->arch.last_vcpu_ran);
 139         kvm->arch.last_vcpu_ran = NULL;
 140         return ret;
 141 }
 142
 143 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 144 {
 145         return 0;
 146 }
 147
 148 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 149 {
 150         return VM_FAULT_SIGBUS;
 151 }
 152
 153
 154 /**
 155  * kvm_arch_destroy_vm - destroy the VM data structure
 156  * @kvm:        pointer to the KVM struct
 157  */
 158 void kvm_arch_destroy_vm(struct kvm *kvm)
 159 {
 160         int i;
 161
 162         kvm_vgic_destroy(kvm);
 163
 164         free_percpu(kvm->arch.last_vcpu_ran);
 165         kvm->arch.last_vcpu_ran = NULL;
 166
 167         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 168                 if (kvm->vcpus[i]) {
 169                         kvm_vcpu_destroy(kvm->vcpus[i]);
 170                         kvm->vcpus[i] = NULL;
 171                 }
 172         }
 173         atomic_set(&kvm->online_vcpus, 0);
 174 }
 175
 176 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 177 {
 178         int r;
 179         switch (ext) {
 180         case KVM_CAP_IRQCHIP:
 181                 r = vgic_present;
 182                 break;
 183         case KVM_CAP_IOEVENTFD:
 184         case KVM_CAP_DEVICE_CTRL:
 185         case KVM_CAP_USER_MEMORY:
 186         case KVM_CAP_SYNC_MMU:
 187         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 188         case KVM_CAP_ONE_REG:
 189         case KVM_CAP_ARM_PSCI:
 190         case KVM_CAP_ARM_PSCI_0_2:
 191         case KVM_CAP_READONLY_MEM:
 192         case KVM_CAP_MP_STATE:
 193         case KVM_CAP_IMMEDIATE_EXIT:
 194         case KVM_CAP_VCPU_EVENTS:
 195         case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
 196         case KVM_CAP_ARM_NISV_TO_USER:
 197         case KVM_CAP_ARM_INJECT_EXT_DABT:
 198                 r = 1;
 199                 break;
 200         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 201                 r = 1;
 202                 break;
 203         case KVM_CAP_NR_VCPUS:
 204                 r = num_online_cpus();
 205                 break;
 206         case KVM_CAP_MAX_VCPUS:
 207                 r = KVM_MAX_VCPUS;
 208                 break;
 209         case KVM_CAP_MAX_VCPU_ID:
 210                 r = KVM_MAX_VCPU_ID;
 211                 break;
 212         case KVM_CAP_MSI_DEVID:
 213                 if (!kvm)
 214                         r = -EINVAL;
 215                 else
 216                         r = kvm->arch.vgic.msis_require_devid;
 217                 break;
 218         case KVM_CAP_ARM_USER_IRQ:
 219                 /*
 220                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 221                  * (bump this number if adding more devices)
 222                  */
 223                 r = 1;
 224                 break;
 225         default:
 226                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 227                 break;
 228         }
 229         return r;
 230 }
 231
 232 long kvm_arch_dev_ioctl(struct file *filp,
 233                         unsigned int ioctl, unsigned long arg)
 234 {
 235         return -EINVAL;
 236 }
 237
 238 struct kvm *kvm_arch_alloc_vm(void)
 239 {
 240         if (!has_vhe())
 241                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 242
 243         return vzalloc(sizeof(struct kvm));
 244 }
 245
 246 void kvm_arch_free_vm(struct kvm *kvm)
 247 {
 248         if (!has_vhe())
 249                 kfree(kvm);
 250         else
 251                 vfree(kvm);
 252 }
 253
 254 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
 255 {
 256         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
 257                 return -EBUSY;
 258
 259         if (id >= kvm->arch.max_vcpus)
 260                 return -EINVAL;
 261
 262         return 0;
 263 }
 264
 265 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
 266 {
 267         int err;
 268
 269         /* Force users to call KVM_ARM_VCPU_INIT */
 270         vcpu->arch.target = -1;
 271         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 272
 273         /* Set up the timer */
 274         kvm_timer_vcpu_init(vcpu);
 275
 276         kvm_pmu_vcpu_init(vcpu);
 277
 278         kvm_arm_reset_debug_ptr(vcpu);
 279
 280         kvm_arm_pvtime_vcpu_init(&vcpu->arch);
 281
 282         err = kvm_vgic_vcpu_init(vcpu);
 283         if (err)
 284                 return err;
 285
 286         return create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 287 }
 288
 289 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 290 {
 291 }
 292
 293 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 294 {
 295         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 296                 static_branch_dec(&userspace_irqchip_in_use);
 297
 298         kvm_mmu_free_memory_caches(vcpu);
 299         kvm_timer_vcpu_terminate(vcpu);
 300         kvm_pmu_vcpu_destroy(vcpu);
 301
 302         kvm_arm_vcpu_destroy(vcpu);
 303 }
 304
 305 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 306 {
 307         return kvm_timer_is_pending(vcpu);
 308 }
 309
 310 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 311 {
 312         /*
 313          * If we're about to block (most likely because we've just hit a
 314          * WFI), we need to sync back the state of the GIC CPU interface
 315          * so that we have the latest PMR and group enables. This ensures
 316          * that kvm_arch_vcpu_runnable has up-to-date data to decide
 317          * whether we have pending interrupts.
 318          *
 319          * For the same reason, we want to tell GICv4 that we need
 320          * doorbells to be signalled, should an interrupt become pending.
 321          */
 322         preempt_disable();
 323         kvm_vgic_vmcr_sync(vcpu);
 324         vgic_v4_put(vcpu, true);
 325         preempt_enable();
 326 }
 327
 328 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 329 {
 330         preempt_disable();
 331         vgic_v4_load(vcpu);
 332         preempt_enable();
 333 }
 334
 335 #define __ptrauth_save_key(regs, key)                                           \
 336 ({                                                                              \
 337         regs[key ## KEYLO_EL1] = read_sysreg_s(SYS_ ## key ## KEYLO_EL1);       \
 338         regs[key ## KEYHI_EL1] = read_sysreg_s(SYS_ ## key ## KEYHI_EL1);       \
 339 })
 340
 341 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 342 {
 343         int *last_ran;
 344         kvm_host_data_t *cpu_data;
 345
 346         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 347         cpu_data = this_cpu_ptr(&kvm_host_data);
 348
 349         /*
 350          * We might get preempted before the vCPU actually runs, but
 351          * over-invalidation doesn't affect correctness.
 352          */
 353         if (*last_ran != vcpu->vcpu_id) {
 354                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 355                 *last_ran = vcpu->vcpu_id;
 356         }
 357
 358         vcpu->cpu = cpu;
 359         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 360
 361         kvm_vgic_load(vcpu);
 362         kvm_timer_vcpu_load(vcpu);
 363         kvm_vcpu_load_sysregs(vcpu);
 364         kvm_arch_vcpu_load_fp(vcpu);
 365         kvm_vcpu_pmu_restore_guest(vcpu);
 366         if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
 367                 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
 368
 369         if (single_task_running())
 370                 vcpu_clear_wfx_traps(vcpu);
 371         else
 372                 vcpu_set_wfx_traps(vcpu);
 373
 374         if (vcpu_has_ptrauth(vcpu)) {
 375                 struct kvm_cpu_context *ctxt = vcpu->arch.host_cpu_context;
 376
 377                 __ptrauth_save_key(ctxt->sys_regs, APIA);
 378                 __ptrauth_save_key(ctxt->sys_regs, APIB);
 379                 __ptrauth_save_key(ctxt->sys_regs, APDA);
 380                 __ptrauth_save_key(ctxt->sys_regs, APDB);
 381                 __ptrauth_save_key(ctxt->sys_regs, APGA);
 382
 383                 vcpu_ptrauth_disable(vcpu);
 384         }
 385 }
 386
 387 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 388 {
 389         kvm_arch_vcpu_put_fp(vcpu);
 390         kvm_vcpu_put_sysregs(vcpu);
 391         kvm_timer_vcpu_put(vcpu);
 392         kvm_vgic_put(vcpu);
 393         kvm_vcpu_pmu_restore_host(vcpu);
 394
 395         vcpu->cpu = -1;
 396 }
 397
 398 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 399 {
 400         vcpu->arch.power_off = true;
 401         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 402         kvm_vcpu_kick(vcpu);
 403 }
 404
 405 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 406                                     struct kvm_mp_state *mp_state)
 407 {
 408         if (vcpu->arch.power_off)
 409                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 410         else
 411                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 412
 413         return 0;
 414 }
 415
 416 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 417                                     struct kvm_mp_state *mp_state)
 418 {
 419         int ret = 0;
 420
 421         switch (mp_state->mp_state) {
 422         case KVM_MP_STATE_RUNNABLE:
 423                 vcpu->arch.power_off = false;
 424                 break;
 425         case KVM_MP_STATE_STOPPED:
 426                 vcpu_power_off(vcpu);
 427                 break;
 428         default:
 429                 ret = -EINVAL;
 430         }
 431
 432         return ret;
 433 }
 434
 435 /**
 436  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 437  * @v:          The VCPU pointer
 438  *
 439  * If the guest CPU is not waiting for interrupts or an interrupt line is
 440  * asserted, the CPU is by definition runnable.
 441  */
 442 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 443 {
 444         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 445         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 446                 && !v->arch.power_off && !v->arch.pause);
 447 }
 448
 449 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 450 {
 451         return vcpu_mode_priv(vcpu);
 452 }
 453
 454 /* Just ensure a guest exit from a particular CPU */
 455 static void exit_vm_noop(void *info)
 456 {
 457 }
 458
 459 void force_vm_exit(const cpumask_t *mask)
 460 {
 461         preempt_disable();
 462         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 463         preempt_enable();
 464 }
 465
 466 /**
 467  * need_new_vmid_gen - check that the VMID is still valid
 468  * @vmid: The VMID to check
 469  *
 470  * return true if there is a new generation of VMIDs being used
 471  *
 472  * The hardware supports a limited set of values with the value zero reserved
 473  * for the host, so we check if an assigned value belongs to a previous
 474  * generation, which which requires us to assign a new value. If we're the
 475  * first to use a VMID for the new generation, we must flush necessary caches
 476  * and TLBs on all CPUs.
 477  */
 478 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 479 {
 480         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 481         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 482         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 483 }
 484
 485 /**
 486  * update_vmid - Update the vmid with a valid VMID for the current generation
 487  * @kvm: The guest that struct vmid belongs to
 488  * @vmid: The stage-2 VMID information struct
 489  */
 490 static void update_vmid(struct kvm_vmid *vmid)
 491 {
 492         if (!need_new_vmid_gen(vmid))
 493                 return;
 494
 495         spin_lock(&kvm_vmid_lock);
 496
 497         /*
 498          * We need to re-check the vmid_gen here to ensure that if another vcpu
 499          * already allocated a valid vmid for this vm, then this vcpu should
 500          * use the same vmid.
 501          */
 502         if (!need_new_vmid_gen(vmid)) {
 503                 spin_unlock(&kvm_vmid_lock);
 504                 return;
 505         }
 506
 507         /* First user of a new VMID generation? */
 508         if (unlikely(kvm_next_vmid == 0)) {
 509                 atomic64_inc(&kvm_vmid_gen);
 510                 kvm_next_vmid = 1;
 511
 512                 /*
 513                  * On SMP we know no other CPUs can use this CPU's or each
 514                  * other's VMID after force_vm_exit returns since the
 515                  * kvm_vmid_lock blocks them from reentry to the guest.
 516                  */
 517                 force_vm_exit(cpu_all_mask);
 518                 /*
 519                  * Now broadcast TLB + ICACHE invalidation over the inner
 520                  * shareable domain to make sure all data structures are
 521                  * clean.
 522                  */
 523                 kvm_call_hyp(__kvm_flush_vm_context);
 524         }
 525
 526         vmid->vmid = kvm_next_vmid;
 527         kvm_next_vmid++;
 528         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 529
 530         smp_wmb();
 531         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 532
 533         spin_unlock(&kvm_vmid_lock);
 534 }
 535
 536 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 537 {
 538         struct kvm *kvm = vcpu->kvm;
 539         int ret = 0;
 540
 541         if (likely(vcpu->arch.has_run_once))
 542                 return 0;
 543
 544         if (!kvm_arm_vcpu_is_finalized(vcpu))
 545                 return -EPERM;
 546
 547         vcpu->arch.has_run_once = true;
 548
 549         if (likely(irqchip_in_kernel(kvm))) {
 550                 /*
 551                  * Map the VGIC hardware resources before running a vcpu the
 552                  * first time on this VM.
 553                  */
 554                 if (unlikely(!vgic_ready(kvm))) {
 555                         ret = kvm_vgic_map_resources(kvm);
 556                         if (ret)
 557                                 return ret;
 558                 }
 559         } else {
 560                 /*
 561                  * Tell the rest of the code that there are userspace irqchip
 562                  * VMs in the wild.
 563                  */
 564                 static_branch_inc(&userspace_irqchip_in_use);
 565         }
 566
 567         ret = kvm_timer_enable(vcpu);
 568         if (ret)
 569                 return ret;
 570
 571         ret = kvm_arm_pmu_v3_enable(vcpu);
 572
 573         return ret;
 574 }
 575
 576 bool kvm_arch_intc_initialized(struct kvm *kvm)
 577 {
 578         return vgic_initialized(kvm);
 579 }
 580
 581 void kvm_arm_halt_guest(struct kvm *kvm)
 582 {
 583         int i;
 584         struct kvm_vcpu *vcpu;
 585
 586         kvm_for_each_vcpu(i, vcpu, kvm)
 587                 vcpu->arch.pause = true;
 588         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 589 }
 590
 591 void kvm_arm_resume_guest(struct kvm *kvm)
 592 {
 593         int i;
 594         struct kvm_vcpu *vcpu;
 595
 596         kvm_for_each_vcpu(i, vcpu, kvm) {
 597                 vcpu->arch.pause = false;
 598                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 599         }
 600 }
 601
 602 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 603 {
 604         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 605
 606         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 607                                        (!vcpu->arch.pause)));
 608
 609         if (vcpu->arch.power_off || vcpu->arch.pause) {
 610                 /* Awaken to handle a signal, request we sleep again later. */
 611                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 612         }
 613
 614         /*
 615          * Make sure we will observe a potential reset request if we've
 616          * observed a change to the power state. Pairs with the smp_wmb() in
 617          * kvm_psci_vcpu_on().
 618          */
 619         smp_rmb();
 620 }
 621
 622 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 623 {
 624         return vcpu->arch.target >= 0;
 625 }
 626
 627 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 628 {
 629         if (kvm_request_pending(vcpu)) {
 630                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 631                         vcpu_req_sleep(vcpu);
 632
 633                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 634                         kvm_reset_vcpu(vcpu);
 635
 636                 /*
 637                  * Clear IRQ_PENDING requests that were made to guarantee
 638                  * that a VCPU sees new virtual interrupts.
 639                  */
 640                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 641
 642                 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
 643                         kvm_update_stolen_time(vcpu);
 644
 645                 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
 646                         /* The distributor enable bits were changed */
 647                         preempt_disable();
 648                         vgic_v4_put(vcpu, false);
 649                         vgic_v4_load(vcpu);
 650                         preempt_enable();
 651                 }
 652         }
 653 }
 654
 655 /**
 656  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 657  * @vcpu:       The VCPU pointer
 658  * @run:        The kvm_run structure pointer used for userspace state exchange
 659  *
 660  * This function is called through the VCPU_RUN ioctl called from user space. It
 661  * will execute VM code in a loop until the time slice for the process is used
 662  * or some emulation is needed from user space in which case the function will
 663  * return with return value 0 and with the kvm_run structure filled in with the
 664  * required data for the requested emulation.
 665  */
 666 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 667 {
 668         int ret;
 669
 670         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 671                 return -ENOEXEC;
 672
 673         ret = kvm_vcpu_first_run_init(vcpu);
 674         if (ret)
 675                 return ret;
 676
 677         if (run->exit_reason == KVM_EXIT_MMIO) {
 678                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 679                 if (ret)
 680                         return ret;
 681         }
 682
 683         if (run->immediate_exit)
 684                 return -EINTR;
 685
 686         vcpu_load(vcpu);
 687
 688         kvm_sigset_activate(vcpu);
 689
 690         ret = 1;
 691         run->exit_reason = KVM_EXIT_UNKNOWN;
 692         while (ret > 0) {
 693                 /*
 694                  * Check conditions before entering the guest
 695                  */
 696                 cond_resched();
 697
 698                 update_vmid(&vcpu->kvm->arch.vmid);
 699
 700                 check_vcpu_requests(vcpu);
 701
 702                 /*
 703                  * Preparing the interrupts to be injected also
 704                  * involves poking the GIC, which must be done in a
 705                  * non-preemptible context.
 706                  */
 707                 preempt_disable();
 708
 709                 kvm_pmu_flush_hwstate(vcpu);
 710
 711                 local_irq_disable();
 712
 713                 kvm_vgic_flush_hwstate(vcpu);
 714
 715                 /*
 716                  * Exit if we have a signal pending so that we can deliver the
 717                  * signal to user space.
 718                  */
 719                 if (signal_pending(current)) {
 720                         ret = -EINTR;
 721                         run->exit_reason = KVM_EXIT_INTR;
 722                 }
 723
 724                 /*
 725                  * If we're using a userspace irqchip, then check if we need
 726                  * to tell a userspace irqchip about timer or PMU level
 727                  * changes and if so, exit to userspace (the actual level
 728                  * state gets updated in kvm_timer_update_run and
 729                  * kvm_pmu_update_run below).
 730                  */
 731                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 732                         if (kvm_timer_should_notify_user(vcpu) ||
 733                             kvm_pmu_should_notify_user(vcpu)) {
 734                                 ret = -EINTR;
 735                                 run->exit_reason = KVM_EXIT_INTR;
 736                         }
 737                 }
 738
 739                 /*
 740                  * Ensure we set mode to IN_GUEST_MODE after we disable
 741                  * interrupts and before the final VCPU requests check.
 742                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 743                  * Documentation/virt/kvm/vcpu-requests.rst
 744                  */
 745                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 746
 747                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 748                     kvm_request_pending(vcpu)) {
 749                         vcpu->mode = OUTSIDE_GUEST_MODE;
 750                         isb(); /* Ensure work in x_flush_hwstate is committed */
 751                         kvm_pmu_sync_hwstate(vcpu);
 752                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 753                                 kvm_timer_sync_hwstate(vcpu);
 754                         kvm_vgic_sync_hwstate(vcpu);
 755                         local_irq_enable();
 756                         preempt_enable();
 757                         continue;
 758                 }
 759
 760                 kvm_arm_setup_debug(vcpu);
 761
 762                 /**************************************************************
 763                  * Enter the guest
 764                  */
 765                 trace_kvm_entry(*vcpu_pc(vcpu));
 766                 guest_enter_irqoff();
 767
 768                 if (has_vhe()) {
 769                         ret = kvm_vcpu_run_vhe(vcpu);
 770                 } else {
 771                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 772                 }
 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_hwstate(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, run, ret);
 831
 832                 preempt_enable();
 833
 834                 ret = handle_exit(vcpu, run, ret);
 835         }
 836
 837         /* Tell userspace about in-kernel device output levels */
 838         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 839                 kvm_timer_update_run(vcpu);
 840                 kvm_pmu_update_run(vcpu);
 841         }
 842
 843         kvm_sigset_deactivate(vcpu);
 844
 845         vcpu_put(vcpu);
 846         return ret;
 847 }
 848
 849 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 850 {
 851         int bit_index;
 852         bool set;
 853         unsigned long *hcr;
 854
 855         if (number == KVM_ARM_IRQ_CPU_IRQ)
 856                 bit_index = __ffs(HCR_VI);
 857         else /* KVM_ARM_IRQ_CPU_FIQ */
 858                 bit_index = __ffs(HCR_VF);
 859
 860         hcr = vcpu_hcr(vcpu);
 861         if (level)
 862                 set = test_and_set_bit(bit_index, hcr);
 863         else
 864                 set = test_and_clear_bit(bit_index, hcr);
 865
 866         /*
 867          * If we didn't change anything, no need to wake up or kick other CPUs
 868          */
 869         if (set == level)
 870                 return 0;
 871
 872         /*
 873          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 874          * trigger a world-switch round on the running physical CPU to set the
 875          * virtual IRQ/FIQ fields in the HCR appropriately.
 876          */
 877         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 878         kvm_vcpu_kick(vcpu);
 879
 880         return 0;
 881 }
 882
 883 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 884                           bool line_status)
 885 {
 886         u32 irq = irq_level->irq;
 887         unsigned int irq_type, vcpu_idx, irq_num;
 888         int nrcpus = atomic_read(&kvm->online_vcpus);
 889         struct kvm_vcpu *vcpu = NULL;
 890         bool level = irq_level->level;
 891
 892         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 893         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 894         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 895         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 896
 897         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 898
 899         switch (irq_type) {
 900         case KVM_ARM_IRQ_TYPE_CPU:
 901                 if (irqchip_in_kernel(kvm))
 902                         return -ENXIO;
 903
 904                 if (vcpu_idx >= nrcpus)
 905                         return -EINVAL;
 906
 907                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 908                 if (!vcpu)
 909                         return -EINVAL;
 910
 911                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 912                         return -EINVAL;
 913
 914                 return vcpu_interrupt_line(vcpu, irq_num, level);
 915         case KVM_ARM_IRQ_TYPE_PPI:
 916                 if (!irqchip_in_kernel(kvm))
 917                         return -ENXIO;
 918
 919                 if (vcpu_idx >= nrcpus)
 920                         return -EINVAL;
 921
 922                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 923                 if (!vcpu)
 924                         return -EINVAL;
 925
 926                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 927                         return -EINVAL;
 928
 929                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 930         case KVM_ARM_IRQ_TYPE_SPI:
 931                 if (!irqchip_in_kernel(kvm))
 932                         return -ENXIO;
 933
 934                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 935                         return -EINVAL;
 936
 937                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 938         }
 939
 940         return -EINVAL;
 941 }
 942
 943 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 944                                const struct kvm_vcpu_init *init)
 945 {
 946         unsigned int i, ret;
 947         int phys_target = kvm_target_cpu();
 948
 949         if (init->target != phys_target)
 950                 return -EINVAL;
 951
 952         /*
 953          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 954          * use the same target.
 955          */
 956         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 957                 return -EINVAL;
 958
 959         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 960         for (i = 0; i < sizeof(init->features) * 8; i++) {
 961                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 962
 963                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 964                         return -ENOENT;
 965
 966                 /*
 967                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 968                  * use the same feature set.
 969                  */
 970                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 971                     test_bit(i, vcpu->arch.features) != set)
 972                         return -EINVAL;
 973
 974                 if (set)
 975                         set_bit(i, vcpu->arch.features);
 976         }
 977
 978         vcpu->arch.target = phys_target;
 979
 980         /* Now we know what it is, we can reset it. */
 981         ret = kvm_reset_vcpu(vcpu);
 982         if (ret) {
 983                 vcpu->arch.target = -1;
 984                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 985         }
 986
 987         return ret;
 988 }
 989
 990 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 991                                          struct kvm_vcpu_init *init)
 992 {
 993         int ret;
 994
 995         ret = kvm_vcpu_set_target(vcpu, init);
 996         if (ret)
 997                 return ret;
 998
 999         /*
1000          * Ensure a rebooted VM will fault in RAM pages and detect if the
1001          * guest MMU is turned off and flush the caches as needed.
1002          */
1003         if (vcpu->arch.has_run_once)
1004                 stage2_unmap_vm(vcpu->kvm);
1005
1006         vcpu_reset_hcr(vcpu);
1007
1008         /*
1009          * Handle the "start in power-off" case.
1010          */
1011         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1012                 vcpu_power_off(vcpu);
1013         else
1014                 vcpu->arch.power_off = false;
1015
1016         return 0;
1017 }
1018
1019 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1020                                  struct kvm_device_attr *attr)
1021 {
1022         int ret = -ENXIO;
1023
1024         switch (attr->group) {
1025         default:
1026                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1027                 break;
1028         }
1029
1030         return ret;
1031 }
1032
1033 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1034                                  struct kvm_device_attr *attr)
1035 {
1036         int ret = -ENXIO;
1037
1038         switch (attr->group) {
1039         default:
1040                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1041                 break;
1042         }
1043
1044         return ret;
1045 }
1046
1047 static int kvm_arm_vcpu_has_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_has_attr(vcpu, attr);
1055                 break;
1056         }
1057
1058         return ret;
1059 }
1060
1061 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1062                                    struct kvm_vcpu_events *events)
1063 {
1064         memset(events, 0, sizeof(*events));
1065
1066         return __kvm_arm_vcpu_get_events(vcpu, events);
1067 }
1068
1069 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1070                                    struct kvm_vcpu_events *events)
1071 {
1072         int i;
1073
1074         /* check whether the reserved field is zero */
1075         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1076                 if (events->reserved[i])
1077                         return -EINVAL;
1078
1079         /* check whether the pad field is zero */
1080         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1081                 if (events->exception.pad[i])
1082                         return -EINVAL;
1083
1084         return __kvm_arm_vcpu_set_events(vcpu, events);
1085 }
1086
1087 long kvm_arch_vcpu_ioctl(struct file *filp,
1088                          unsigned int ioctl, unsigned long arg)
1089 {
1090         struct kvm_vcpu *vcpu = filp->private_data;
1091         void __user *argp = (void __user *)arg;
1092         struct kvm_device_attr attr;
1093         long r;
1094
1095         switch (ioctl) {
1096         case KVM_ARM_VCPU_INIT: {
1097                 struct kvm_vcpu_init init;
1098
1099                 r = -EFAULT;
1100                 if (copy_from_user(&init, argp, sizeof(init)))
1101                         break;
1102
1103                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1104                 break;
1105         }
1106         case KVM_SET_ONE_REG:
1107         case KVM_GET_ONE_REG: {
1108                 struct kvm_one_reg reg;
1109
1110                 r = -ENOEXEC;
1111                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1112                         break;
1113
1114                 r = -EFAULT;
1115                 if (copy_from_user(&reg, argp, sizeof(reg)))
1116                         break;
1117
1118                 if (ioctl == KVM_SET_ONE_REG)
1119                         r = kvm_arm_set_reg(vcpu, &reg);
1120                 else
1121                         r = kvm_arm_get_reg(vcpu, &reg);
1122                 break;
1123         }
1124         case KVM_GET_REG_LIST: {
1125                 struct kvm_reg_list __user *user_list = argp;
1126                 struct kvm_reg_list reg_list;
1127                 unsigned n;
1128
1129                 r = -ENOEXEC;
1130                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1131                         break;
1132
1133                 r = -EPERM;
1134                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1135                         break;
1136
1137                 r = -EFAULT;
1138                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1139                         break;
1140                 n = reg_list.n;
1141                 reg_list.n = kvm_arm_num_regs(vcpu);
1142                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1143                         break;
1144                 r = -E2BIG;
1145                 if (n < reg_list.n)
1146                         break;
1147                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1148                 break;
1149         }
1150         case KVM_SET_DEVICE_ATTR: {
1151                 r = -EFAULT;
1152                 if (copy_from_user(&attr, argp, sizeof(attr)))
1153                         break;
1154                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1155                 break;
1156         }
1157         case KVM_GET_DEVICE_ATTR: {
1158                 r = -EFAULT;
1159                 if (copy_from_user(&attr, argp, sizeof(attr)))
1160                         break;
1161                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1162                 break;
1163         }
1164         case KVM_HAS_DEVICE_ATTR: {
1165                 r = -EFAULT;
1166                 if (copy_from_user(&attr, argp, sizeof(attr)))
1167                         break;
1168                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1169                 break;
1170         }
1171         case KVM_GET_VCPU_EVENTS: {
1172                 struct kvm_vcpu_events events;
1173
1174                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1175                         return -EINVAL;
1176
1177                 if (copy_to_user(argp, &events, sizeof(events)))
1178                         return -EFAULT;
1179
1180                 return 0;
1181         }
1182         case KVM_SET_VCPU_EVENTS: {
1183                 struct kvm_vcpu_events events;
1184
1185                 if (copy_from_user(&events, argp, sizeof(events)))
1186                         return -EFAULT;
1187
1188                 return kvm_arm_vcpu_set_events(vcpu, &events);
1189         }
1190         case KVM_ARM_VCPU_FINALIZE: {
1191                 int what;
1192
1193                 if (!kvm_vcpu_initialized(vcpu))
1194                         return -ENOEXEC;
1195
1196                 if (get_user(what, (const int __user *)argp))
1197                         return -EFAULT;
1198
1199                 return kvm_arm_vcpu_finalize(vcpu, what);
1200         }
1201         default:
1202                 r = -EINVAL;
1203         }
1204
1205         return r;
1206 }
1207
1208 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1209 {
1210
1211 }
1212
1213 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1214                                         struct kvm_memory_slot *memslot)
1215 {
1216         kvm_flush_remote_tlbs(kvm);
1217 }
1218
1219 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1220                                         struct kvm_arm_device_addr *dev_addr)
1221 {
1222         unsigned long dev_id, type;
1223
1224         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1225                 KVM_ARM_DEVICE_ID_SHIFT;
1226         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1227                 KVM_ARM_DEVICE_TYPE_SHIFT;
1228
1229         switch (dev_id) {
1230         case KVM_ARM_DEVICE_VGIC_V2:
1231                 if (!vgic_present)
1232                         return -ENXIO;
1233                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1234         default:
1235                 return -ENODEV;
1236         }
1237 }
1238
1239 long kvm_arch_vm_ioctl(struct file *filp,
1240                        unsigned int ioctl, unsigned long arg)
1241 {
1242         struct kvm *kvm = filp->private_data;
1243         void __user *argp = (void __user *)arg;
1244
1245         switch (ioctl) {
1246         case KVM_CREATE_IRQCHIP: {
1247                 int ret;
1248                 if (!vgic_present)
1249                         return -ENXIO;
1250                 mutex_lock(&kvm->lock);
1251                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1252                 mutex_unlock(&kvm->lock);
1253                 return ret;
1254         }
1255         case KVM_ARM_SET_DEVICE_ADDR: {
1256                 struct kvm_arm_device_addr dev_addr;
1257
1258                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1259                         return -EFAULT;
1260                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1261         }
1262         case KVM_ARM_PREFERRED_TARGET: {
1263                 int err;
1264                 struct kvm_vcpu_init init;
1265
1266                 err = kvm_vcpu_preferred_target(&init);
1267                 if (err)
1268                         return err;
1269
1270                 if (copy_to_user(argp, &init, sizeof(init)))
1271                         return -EFAULT;
1272
1273                 return 0;
1274         }
1275         default:
1276                 return -EINVAL;
1277         }
1278 }
1279
1280 static void cpu_init_hyp_mode(void)
1281 {
1282         phys_addr_t pgd_ptr;
1283         unsigned long hyp_stack_ptr;
1284         unsigned long stack_page;
1285         unsigned long vector_ptr;
1286
1287         /* Switch from the HYP stub to our own HYP init vector */
1288         __hyp_set_vectors(kvm_get_idmap_vector());
1289
1290         pgd_ptr = kvm_mmu_get_httbr();
1291         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1292         hyp_stack_ptr = stack_page + PAGE_SIZE;
1293         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1294
1295         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1296         __cpu_init_stage2();
1297 }
1298
1299 static void cpu_hyp_reset(void)
1300 {
1301         if (!is_kernel_in_hyp_mode())
1302                 __hyp_reset_vectors();
1303 }
1304
1305 static void cpu_hyp_reinit(void)
1306 {
1307         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1308
1309         cpu_hyp_reset();
1310
1311         if (is_kernel_in_hyp_mode())
1312                 kvm_timer_init_vhe();
1313         else
1314                 cpu_init_hyp_mode();
1315
1316         kvm_arm_init_debug();
1317
1318         if (vgic_present)
1319                 kvm_vgic_init_cpu_hardware();
1320 }
1321
1322 static void _kvm_arch_hardware_enable(void *discard)
1323 {
1324         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1325                 cpu_hyp_reinit();
1326                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1327         }
1328 }
1329
1330 int kvm_arch_hardware_enable(void)
1331 {
1332         _kvm_arch_hardware_enable(NULL);
1333         return 0;
1334 }
1335
1336 static void _kvm_arch_hardware_disable(void *discard)
1337 {
1338         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1339                 cpu_hyp_reset();
1340                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1341         }
1342 }
1343
1344 void kvm_arch_hardware_disable(void)
1345 {
1346         _kvm_arch_hardware_disable(NULL);
1347 }
1348
1349 #ifdef CONFIG_CPU_PM
1350 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1351                                     unsigned long cmd,
1352                                     void *v)
1353 {
1354         /*
1355          * kvm_arm_hardware_enabled is left with its old value over
1356          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1357          * re-enable hyp.
1358          */
1359         switch (cmd) {
1360         case CPU_PM_ENTER:
1361                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1362                         /*
1363                          * don't update kvm_arm_hardware_enabled here
1364                          * so that the hardware will be re-enabled
1365                          * when we resume. See below.
1366                          */
1367                         cpu_hyp_reset();
1368
1369                 return NOTIFY_OK;
1370         case CPU_PM_ENTER_FAILED:
1371         case CPU_PM_EXIT:
1372                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1373                         /* The hardware was enabled before suspend. */
1374                         cpu_hyp_reinit();
1375
1376                 return NOTIFY_OK;
1377
1378         default:
1379                 return NOTIFY_DONE;
1380         }
1381 }
1382
1383 static struct notifier_block hyp_init_cpu_pm_nb = {
1384         .notifier_call = hyp_init_cpu_pm_notifier,
1385 };
1386
1387 static void __init hyp_cpu_pm_init(void)
1388 {
1389         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1390 }
1391 static void __init hyp_cpu_pm_exit(void)
1392 {
1393         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1394 }
1395 #else
1396 static inline void hyp_cpu_pm_init(void)
1397 {
1398 }
1399 static inline void hyp_cpu_pm_exit(void)
1400 {
1401 }
1402 #endif
1403
1404 static int init_common_resources(void)
1405 {
1406         kvm_set_ipa_limit();
1407
1408         return 0;
1409 }
1410
1411 static int init_subsystems(void)
1412 {
1413         int err = 0;
1414
1415         /*
1416          * Enable hardware so that subsystem initialisation can access EL2.
1417          */
1418         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1419
1420         /*
1421          * Register CPU lower-power notifier
1422          */
1423         hyp_cpu_pm_init();
1424
1425         /*
1426          * Init HYP view of VGIC
1427          */
1428         err = kvm_vgic_hyp_init();
1429         switch (err) {
1430         case 0:
1431                 vgic_present = true;
1432                 break;
1433         case -ENODEV:
1434         case -ENXIO:
1435                 vgic_present = false;
1436                 err = 0;
1437                 break;
1438         default:
1439                 goto out;
1440         }
1441
1442         /*
1443          * Init HYP architected timer support
1444          */
1445         err = kvm_timer_hyp_init(vgic_present);
1446         if (err)
1447                 goto out;
1448
1449         kvm_perf_init();
1450         kvm_coproc_table_init();
1451
1452 out:
1453         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1454
1455         return err;
1456 }
1457
1458 static void teardown_hyp_mode(void)
1459 {
1460         int cpu;
1461
1462         free_hyp_pgds();
1463         for_each_possible_cpu(cpu)
1464                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1465 }
1466
1467 /**
1468  * Inits Hyp-mode on all online CPUs
1469  */
1470 static int init_hyp_mode(void)
1471 {
1472         int cpu;
1473         int err = 0;
1474
1475         /*
1476          * Allocate Hyp PGD and setup Hyp identity mapping
1477          */
1478         err = kvm_mmu_init();
1479         if (err)
1480                 goto out_err;
1481
1482         /*
1483          * Allocate stack pages for Hypervisor-mode
1484          */
1485         for_each_possible_cpu(cpu) {
1486                 unsigned long stack_page;
1487
1488                 stack_page = __get_free_page(GFP_KERNEL);
1489                 if (!stack_page) {
1490                         err = -ENOMEM;
1491                         goto out_err;
1492                 }
1493
1494                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1495         }
1496
1497         /*
1498          * Map the Hyp-code called directly from the host
1499          */
1500         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1501                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1502         if (err) {
1503                 kvm_err("Cannot map world-switch code\n");
1504                 goto out_err;
1505         }
1506
1507         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1508                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1509         if (err) {
1510                 kvm_err("Cannot map rodata section\n");
1511                 goto out_err;
1512         }
1513
1514         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1515                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1516         if (err) {
1517                 kvm_err("Cannot map bss section\n");
1518                 goto out_err;
1519         }
1520
1521         err = kvm_map_vectors();
1522         if (err) {
1523                 kvm_err("Cannot map vectors\n");
1524                 goto out_err;
1525         }
1526
1527         /*
1528          * Map the Hyp stack pages
1529          */
1530         for_each_possible_cpu(cpu) {
1531                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1532                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1533                                           PAGE_HYP);
1534
1535                 if (err) {
1536                         kvm_err("Cannot map hyp stack\n");
1537                         goto out_err;
1538                 }
1539         }
1540
1541         for_each_possible_cpu(cpu) {
1542                 kvm_host_data_t *cpu_data;
1543
1544                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1545                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1546
1547                 if (err) {
1548                         kvm_err("Cannot map host CPU state: %d\n", err);
1549                         goto out_err;
1550                 }
1551         }
1552
1553         err = hyp_map_aux_data();
1554         if (err)
1555                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1556
1557         return 0;
1558
1559 out_err:
1560         teardown_hyp_mode();
1561         kvm_err("error initializing Hyp mode: %d\n", err);
1562         return err;
1563 }
1564
1565 static void check_kvm_target_cpu(void *ret)
1566 {
1567         *(int *)ret = kvm_target_cpu();
1568 }
1569
1570 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1571 {
1572         struct kvm_vcpu *vcpu;
1573         int i;
1574
1575         mpidr &= MPIDR_HWID_BITMASK;
1576         kvm_for_each_vcpu(i, vcpu, kvm) {
1577                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1578                         return vcpu;
1579         }
1580         return NULL;
1581 }
1582
1583 bool kvm_arch_has_irq_bypass(void)
1584 {
1585         return true;
1586 }
1587
1588 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1589                                       struct irq_bypass_producer *prod)
1590 {
1591         struct kvm_kernel_irqfd *irqfd =
1592                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1593
1594         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1595                                           &irqfd->irq_entry);
1596 }
1597 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1598                                       struct irq_bypass_producer *prod)
1599 {
1600         struct kvm_kernel_irqfd *irqfd =
1601                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1602
1603         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1604                                      &irqfd->irq_entry);
1605 }
1606
1607 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1608 {
1609         struct kvm_kernel_irqfd *irqfd =
1610                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1611
1612         kvm_arm_halt_guest(irqfd->kvm);
1613 }
1614
1615 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1616 {
1617         struct kvm_kernel_irqfd *irqfd =
1618                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1619
1620         kvm_arm_resume_guest(irqfd->kvm);
1621 }
1622
1623 /**
1624  * Initialize Hyp-mode and memory mappings on all CPUs.
1625  */
1626 int kvm_arch_init(void *opaque)
1627 {
1628         int err;
1629         int ret, cpu;
1630         bool in_hyp_mode;
1631
1632         if (!is_hyp_mode_available()) {
1633                 kvm_info("HYP mode not available\n");
1634                 return -ENODEV;
1635         }
1636
1637         in_hyp_mode = is_kernel_in_hyp_mode();
1638
1639         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1640                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1641                 return -ENODEV;
1642         }
1643
1644         for_each_online_cpu(cpu) {
1645                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1646                 if (ret < 0) {
1647                         kvm_err("Error, CPU %d not supported!\n", cpu);
1648                         return -ENODEV;
1649                 }
1650         }
1651
1652         err = init_common_resources();
1653         if (err)
1654                 return err;
1655
1656         err = kvm_arm_init_sve();
1657         if (err)
1658                 return err;
1659
1660         if (!in_hyp_mode) {
1661                 err = init_hyp_mode();
1662                 if (err)
1663                         goto out_err;
1664         }
1665
1666         err = init_subsystems();
1667         if (err)
1668                 goto out_hyp;
1669
1670         if (in_hyp_mode)
1671                 kvm_info("VHE mode initialized successfully\n");
1672         else
1673                 kvm_info("Hyp mode initialized successfully\n");
1674
1675         return 0;
1676
1677 out_hyp:
1678         hyp_cpu_pm_exit();
1679         if (!in_hyp_mode)
1680                 teardown_hyp_mode();
1681 out_err:
1682         return err;
1683 }
1684
1685 /* NOP: Compiling as a module not supported */
1686 void kvm_arch_exit(void)
1687 {
1688         kvm_perf_teardown();
1689 }
1690
1691 static int arm_init(void)
1692 {
1693         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1694         return rc;
1695 }
1696
1697 module_init(arm_init);