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 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 336 {
 337         int *last_ran;
 338         kvm_host_data_t *cpu_data;
 339
 340         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 341         cpu_data = this_cpu_ptr(&kvm_host_data);
 342
 343         /*
 344          * We might get preempted before the vCPU actually runs, but
 345          * over-invalidation doesn't affect correctness.
 346          */
 347         if (*last_ran != vcpu->vcpu_id) {
 348                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 349                 *last_ran = vcpu->vcpu_id;
 350         }
 351
 352         vcpu->cpu = cpu;
 353         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 354
 355         kvm_vgic_load(vcpu);
 356         kvm_timer_vcpu_load(vcpu);
 357         kvm_vcpu_load_sysregs(vcpu);
 358         kvm_arch_vcpu_load_fp(vcpu);
 359         kvm_vcpu_pmu_restore_guest(vcpu);
 360         if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
 361                 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
 362
 363         if (single_task_running())
 364                 vcpu_clear_wfx_traps(vcpu);
 365         else
 366                 vcpu_set_wfx_traps(vcpu);
 367
 368         vcpu_ptrauth_setup_lazy(vcpu);
 369 }
 370
 371 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 372 {
 373         kvm_arch_vcpu_put_fp(vcpu);
 374         kvm_vcpu_put_sysregs(vcpu);
 375         kvm_timer_vcpu_put(vcpu);
 376         kvm_vgic_put(vcpu);
 377         kvm_vcpu_pmu_restore_host(vcpu);
 378
 379         vcpu->cpu = -1;
 380 }
 381
 382 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 383 {
 384         vcpu->arch.power_off = true;
 385         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 386         kvm_vcpu_kick(vcpu);
 387 }
 388
 389 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 390                                     struct kvm_mp_state *mp_state)
 391 {
 392         if (vcpu->arch.power_off)
 393                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 394         else
 395                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 396
 397         return 0;
 398 }
 399
 400 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 401                                     struct kvm_mp_state *mp_state)
 402 {
 403         int ret = 0;
 404
 405         switch (mp_state->mp_state) {
 406         case KVM_MP_STATE_RUNNABLE:
 407                 vcpu->arch.power_off = false;
 408                 break;
 409         case KVM_MP_STATE_STOPPED:
 410                 vcpu_power_off(vcpu);
 411                 break;
 412         default:
 413                 ret = -EINVAL;
 414         }
 415
 416         return ret;
 417 }
 418
 419 /**
 420  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 421  * @v:          The VCPU pointer
 422  *
 423  * If the guest CPU is not waiting for interrupts or an interrupt line is
 424  * asserted, the CPU is by definition runnable.
 425  */
 426 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 427 {
 428         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 429         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 430                 && !v->arch.power_off && !v->arch.pause);
 431 }
 432
 433 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 434 {
 435         return vcpu_mode_priv(vcpu);
 436 }
 437
 438 /* Just ensure a guest exit from a particular CPU */
 439 static void exit_vm_noop(void *info)
 440 {
 441 }
 442
 443 void force_vm_exit(const cpumask_t *mask)
 444 {
 445         preempt_disable();
 446         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 447         preempt_enable();
 448 }
 449
 450 /**
 451  * need_new_vmid_gen - check that the VMID is still valid
 452  * @vmid: The VMID to check
 453  *
 454  * return true if there is a new generation of VMIDs being used
 455  *
 456  * The hardware supports a limited set of values with the value zero reserved
 457  * for the host, so we check if an assigned value belongs to a previous
 458  * generation, which which requires us to assign a new value. If we're the
 459  * first to use a VMID for the new generation, we must flush necessary caches
 460  * and TLBs on all CPUs.
 461  */
 462 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 463 {
 464         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 465         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 466         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 467 }
 468
 469 /**
 470  * update_vmid - Update the vmid with a valid VMID for the current generation
 471  * @kvm: The guest that struct vmid belongs to
 472  * @vmid: The stage-2 VMID information struct
 473  */
 474 static void update_vmid(struct kvm_vmid *vmid)
 475 {
 476         if (!need_new_vmid_gen(vmid))
 477                 return;
 478
 479         spin_lock(&kvm_vmid_lock);
 480
 481         /*
 482          * We need to re-check the vmid_gen here to ensure that if another vcpu
 483          * already allocated a valid vmid for this vm, then this vcpu should
 484          * use the same vmid.
 485          */
 486         if (!need_new_vmid_gen(vmid)) {
 487                 spin_unlock(&kvm_vmid_lock);
 488                 return;
 489         }
 490
 491         /* First user of a new VMID generation? */
 492         if (unlikely(kvm_next_vmid == 0)) {
 493                 atomic64_inc(&kvm_vmid_gen);
 494                 kvm_next_vmid = 1;
 495
 496                 /*
 497                  * On SMP we know no other CPUs can use this CPU's or each
 498                  * other's VMID after force_vm_exit returns since the
 499                  * kvm_vmid_lock blocks them from reentry to the guest.
 500                  */
 501                 force_vm_exit(cpu_all_mask);
 502                 /*
 503                  * Now broadcast TLB + ICACHE invalidation over the inner
 504                  * shareable domain to make sure all data structures are
 505                  * clean.
 506                  */
 507                 kvm_call_hyp(__kvm_flush_vm_context);
 508         }
 509
 510         vmid->vmid = kvm_next_vmid;
 511         kvm_next_vmid++;
 512         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 513
 514         smp_wmb();
 515         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 516
 517         spin_unlock(&kvm_vmid_lock);
 518 }
 519
 520 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 521 {
 522         struct kvm *kvm = vcpu->kvm;
 523         int ret = 0;
 524
 525         if (likely(vcpu->arch.has_run_once))
 526                 return 0;
 527
 528         if (!kvm_arm_vcpu_is_finalized(vcpu))
 529                 return -EPERM;
 530
 531         vcpu->arch.has_run_once = true;
 532
 533         if (likely(irqchip_in_kernel(kvm))) {
 534                 /*
 535                  * Map the VGIC hardware resources before running a vcpu the
 536                  * first time on this VM.
 537                  */
 538                 if (unlikely(!vgic_ready(kvm))) {
 539                         ret = kvm_vgic_map_resources(kvm);
 540                         if (ret)
 541                                 return ret;
 542                 }
 543         } else {
 544                 /*
 545                  * Tell the rest of the code that there are userspace irqchip
 546                  * VMs in the wild.
 547                  */
 548                 static_branch_inc(&userspace_irqchip_in_use);
 549         }
 550
 551         ret = kvm_timer_enable(vcpu);
 552         if (ret)
 553                 return ret;
 554
 555         ret = kvm_arm_pmu_v3_enable(vcpu);
 556
 557         return ret;
 558 }
 559
 560 bool kvm_arch_intc_initialized(struct kvm *kvm)
 561 {
 562         return vgic_initialized(kvm);
 563 }
 564
 565 void kvm_arm_halt_guest(struct kvm *kvm)
 566 {
 567         int i;
 568         struct kvm_vcpu *vcpu;
 569
 570         kvm_for_each_vcpu(i, vcpu, kvm)
 571                 vcpu->arch.pause = true;
 572         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 573 }
 574
 575 void kvm_arm_resume_guest(struct kvm *kvm)
 576 {
 577         int i;
 578         struct kvm_vcpu *vcpu;
 579
 580         kvm_for_each_vcpu(i, vcpu, kvm) {
 581                 vcpu->arch.pause = false;
 582                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 583         }
 584 }
 585
 586 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 587 {
 588         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 589
 590         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 591                                        (!vcpu->arch.pause)));
 592
 593         if (vcpu->arch.power_off || vcpu->arch.pause) {
 594                 /* Awaken to handle a signal, request we sleep again later. */
 595                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 596         }
 597
 598         /*
 599          * Make sure we will observe a potential reset request if we've
 600          * observed a change to the power state. Pairs with the smp_wmb() in
 601          * kvm_psci_vcpu_on().
 602          */
 603         smp_rmb();
 604 }
 605
 606 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 607 {
 608         return vcpu->arch.target >= 0;
 609 }
 610
 611 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 612 {
 613         if (kvm_request_pending(vcpu)) {
 614                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 615                         vcpu_req_sleep(vcpu);
 616
 617                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 618                         kvm_reset_vcpu(vcpu);
 619
 620                 /*
 621                  * Clear IRQ_PENDING requests that were made to guarantee
 622                  * that a VCPU sees new virtual interrupts.
 623                  */
 624                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 625
 626                 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
 627                         kvm_update_stolen_time(vcpu);
 628
 629                 if (kvm_check_request(KVM_REQ_RELOAD_GICv4, vcpu)) {
 630                         /* The distributor enable bits were changed */
 631                         preempt_disable();
 632                         vgic_v4_put(vcpu, false);
 633                         vgic_v4_load(vcpu);
 634                         preempt_enable();
 635                 }
 636         }
 637 }
 638
 639 /**
 640  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 641  * @vcpu:       The VCPU pointer
 642  *
 643  * This function is called through the VCPU_RUN ioctl called from user space. It
 644  * will execute VM code in a loop until the time slice for the process is used
 645  * or some emulation is needed from user space in which case the function will
 646  * return with return value 0 and with the kvm_run structure filled in with the
 647  * required data for the requested emulation.
 648  */
 649 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
 650 {
 651         struct kvm_run *run = vcpu->run;
 652         int ret;
 653
 654         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 655                 return -ENOEXEC;
 656
 657         ret = kvm_vcpu_first_run_init(vcpu);
 658         if (ret)
 659                 return ret;
 660
 661         if (run->exit_reason == KVM_EXIT_MMIO) {
 662                 ret = kvm_handle_mmio_return(vcpu, run);
 663                 if (ret)
 664                         return ret;
 665         }
 666
 667         if (run->immediate_exit)
 668                 return -EINTR;
 669
 670         vcpu_load(vcpu);
 671
 672         kvm_sigset_activate(vcpu);
 673
 674         ret = 1;
 675         run->exit_reason = KVM_EXIT_UNKNOWN;
 676         while (ret > 0) {
 677                 /*
 678                  * Check conditions before entering the guest
 679                  */
 680                 cond_resched();
 681
 682                 update_vmid(&vcpu->kvm->arch.vmid);
 683
 684                 check_vcpu_requests(vcpu);
 685
 686                 /*
 687                  * Preparing the interrupts to be injected also
 688                  * involves poking the GIC, which must be done in a
 689                  * non-preemptible context.
 690                  */
 691                 preempt_disable();
 692
 693                 kvm_pmu_flush_hwstate(vcpu);
 694
 695                 local_irq_disable();
 696
 697                 kvm_vgic_flush_hwstate(vcpu);
 698
 699                 /*
 700                  * Exit if we have a signal pending so that we can deliver the
 701                  * signal to user space.
 702                  */
 703                 if (signal_pending(current)) {
 704                         ret = -EINTR;
 705                         run->exit_reason = KVM_EXIT_INTR;
 706                 }
 707
 708                 /*
 709                  * If we're using a userspace irqchip, then check if we need
 710                  * to tell a userspace irqchip about timer or PMU level
 711                  * changes and if so, exit to userspace (the actual level
 712                  * state gets updated in kvm_timer_update_run and
 713                  * kvm_pmu_update_run below).
 714                  */
 715                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 716                         if (kvm_timer_should_notify_user(vcpu) ||
 717                             kvm_pmu_should_notify_user(vcpu)) {
 718                                 ret = -EINTR;
 719                                 run->exit_reason = KVM_EXIT_INTR;
 720                         }
 721                 }
 722
 723                 /*
 724                  * Ensure we set mode to IN_GUEST_MODE after we disable
 725                  * interrupts and before the final VCPU requests check.
 726                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 727                  * Documentation/virt/kvm/vcpu-requests.rst
 728                  */
 729                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 730
 731                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 732                     kvm_request_pending(vcpu)) {
 733                         vcpu->mode = OUTSIDE_GUEST_MODE;
 734                         isb(); /* Ensure work in x_flush_hwstate is committed */
 735                         kvm_pmu_sync_hwstate(vcpu);
 736                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 737                                 kvm_timer_sync_hwstate(vcpu);
 738                         kvm_vgic_sync_hwstate(vcpu);
 739                         local_irq_enable();
 740                         preempt_enable();
 741                         continue;
 742                 }
 743
 744                 kvm_arm_setup_debug(vcpu);
 745
 746                 /**************************************************************
 747                  * Enter the guest
 748                  */
 749                 trace_kvm_entry(*vcpu_pc(vcpu));
 750                 guest_enter_irqoff();
 751
 752                 if (has_vhe()) {
 753                         ret = kvm_vcpu_run_vhe(vcpu);
 754                 } else {
 755                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 756                 }
 757
 758                 vcpu->mode = OUTSIDE_GUEST_MODE;
 759                 vcpu->stat.exits++;
 760                 /*
 761                  * Back from guest
 762                  *************************************************************/
 763
 764                 kvm_arm_clear_debug(vcpu);
 765
 766                 /*
 767                  * We must sync the PMU state before the vgic state so
 768                  * that the vgic can properly sample the updated state of the
 769                  * interrupt line.
 770                  */
 771                 kvm_pmu_sync_hwstate(vcpu);
 772
 773                 /*
 774                  * Sync the vgic state before syncing the timer state because
 775                  * the timer code needs to know if the virtual timer
 776                  * interrupts are active.
 777                  */
 778                 kvm_vgic_sync_hwstate(vcpu);
 779
 780                 /*
 781                  * Sync the timer hardware state before enabling interrupts as
 782                  * we don't want vtimer interrupts to race with syncing the
 783                  * timer virtual interrupt state.
 784                  */
 785                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 786                         kvm_timer_sync_hwstate(vcpu);
 787
 788                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 789
 790                 /*
 791                  * We may have taken a host interrupt in HYP mode (ie
 792                  * while executing the guest). This interrupt is still
 793                  * pending, as we haven't serviced it yet!
 794                  *
 795                  * We're now back in SVC mode, with interrupts
 796                  * disabled.  Enabling the interrupts now will have
 797                  * the effect of taking the interrupt again, in SVC
 798                  * mode this time.
 799                  */
 800                 local_irq_enable();
 801
 802                 /*
 803                  * We do local_irq_enable() before calling guest_exit() so
 804                  * that if a timer interrupt hits while running the guest we
 805                  * account that tick as being spent in the guest.  We enable
 806                  * preemption after calling guest_exit() so that if we get
 807                  * preempted we make sure ticks after that is not counted as
 808                  * guest time.
 809                  */
 810                 guest_exit();
 811                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 812
 813                 /* Exit types that need handling before we can be preempted */
 814                 handle_exit_early(vcpu, run, ret);
 815
 816                 preempt_enable();
 817
 818                 ret = handle_exit(vcpu, run, ret);
 819         }
 820
 821         /* Tell userspace about in-kernel device output levels */
 822         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 823                 kvm_timer_update_run(vcpu);
 824                 kvm_pmu_update_run(vcpu);
 825         }
 826
 827         kvm_sigset_deactivate(vcpu);
 828
 829         vcpu_put(vcpu);
 830         return ret;
 831 }
 832
 833 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 834 {
 835         int bit_index;
 836         bool set;
 837         unsigned long *hcr;
 838
 839         if (number == KVM_ARM_IRQ_CPU_IRQ)
 840                 bit_index = __ffs(HCR_VI);
 841         else /* KVM_ARM_IRQ_CPU_FIQ */
 842                 bit_index = __ffs(HCR_VF);
 843
 844         hcr = vcpu_hcr(vcpu);
 845         if (level)
 846                 set = test_and_set_bit(bit_index, hcr);
 847         else
 848                 set = test_and_clear_bit(bit_index, hcr);
 849
 850         /*
 851          * If we didn't change anything, no need to wake up or kick other CPUs
 852          */
 853         if (set == level)
 854                 return 0;
 855
 856         /*
 857          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 858          * trigger a world-switch round on the running physical CPU to set the
 859          * virtual IRQ/FIQ fields in the HCR appropriately.
 860          */
 861         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 862         kvm_vcpu_kick(vcpu);
 863
 864         return 0;
 865 }
 866
 867 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 868                           bool line_status)
 869 {
 870         u32 irq = irq_level->irq;
 871         unsigned int irq_type, vcpu_idx, irq_num;
 872         int nrcpus = atomic_read(&kvm->online_vcpus);
 873         struct kvm_vcpu *vcpu = NULL;
 874         bool level = irq_level->level;
 875
 876         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 877         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 878         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 879         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 880
 881         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 882
 883         switch (irq_type) {
 884         case KVM_ARM_IRQ_TYPE_CPU:
 885                 if (irqchip_in_kernel(kvm))
 886                         return -ENXIO;
 887
 888                 if (vcpu_idx >= nrcpus)
 889                         return -EINVAL;
 890
 891                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 892                 if (!vcpu)
 893                         return -EINVAL;
 894
 895                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 896                         return -EINVAL;
 897
 898                 return vcpu_interrupt_line(vcpu, irq_num, level);
 899         case KVM_ARM_IRQ_TYPE_PPI:
 900                 if (!irqchip_in_kernel(kvm))
 901                         return -ENXIO;
 902
 903                 if (vcpu_idx >= nrcpus)
 904                         return -EINVAL;
 905
 906                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 907                 if (!vcpu)
 908                         return -EINVAL;
 909
 910                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 911                         return -EINVAL;
 912
 913                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 914         case KVM_ARM_IRQ_TYPE_SPI:
 915                 if (!irqchip_in_kernel(kvm))
 916                         return -ENXIO;
 917
 918                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 919                         return -EINVAL;
 920
 921                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 922         }
 923
 924         return -EINVAL;
 925 }
 926
 927 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 928                                const struct kvm_vcpu_init *init)
 929 {
 930         unsigned int i, ret;
 931         int phys_target = kvm_target_cpu();
 932
 933         if (init->target != phys_target)
 934                 return -EINVAL;
 935
 936         /*
 937          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 938          * use the same target.
 939          */
 940         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 941                 return -EINVAL;
 942
 943         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 944         for (i = 0; i < sizeof(init->features) * 8; i++) {
 945                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 946
 947                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 948                         return -ENOENT;
 949
 950                 /*
 951                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 952                  * use the same feature set.
 953                  */
 954                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 955                     test_bit(i, vcpu->arch.features) != set)
 956                         return -EINVAL;
 957
 958                 if (set)
 959                         set_bit(i, vcpu->arch.features);
 960         }
 961
 962         vcpu->arch.target = phys_target;
 963
 964         /* Now we know what it is, we can reset it. */
 965         ret = kvm_reset_vcpu(vcpu);
 966         if (ret) {
 967                 vcpu->arch.target = -1;
 968                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 969         }
 970
 971         return ret;
 972 }
 973
 974 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 975                                          struct kvm_vcpu_init *init)
 976 {
 977         int ret;
 978
 979         ret = kvm_vcpu_set_target(vcpu, init);
 980         if (ret)
 981                 return ret;
 982
 983         /*
 984          * Ensure a rebooted VM will fault in RAM pages and detect if the
 985          * guest MMU is turned off and flush the caches as needed.
 986          */
 987         if (vcpu->arch.has_run_once)
 988                 stage2_unmap_vm(vcpu->kvm);
 989
 990         vcpu_reset_hcr(vcpu);
 991
 992         /*
 993          * Handle the "start in power-off" case.
 994          */
 995         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 996                 vcpu_power_off(vcpu);
 997         else
 998                 vcpu->arch.power_off = false;
 999
1000         return 0;
1001 }
1002
1003 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1004                                  struct kvm_device_attr *attr)
1005 {
1006         int ret = -ENXIO;
1007
1008         switch (attr->group) {
1009         default:
1010                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1011                 break;
1012         }
1013
1014         return ret;
1015 }
1016
1017 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1018                                  struct kvm_device_attr *attr)
1019 {
1020         int ret = -ENXIO;
1021
1022         switch (attr->group) {
1023         default:
1024                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1025                 break;
1026         }
1027
1028         return ret;
1029 }
1030
1031 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1032                                  struct kvm_device_attr *attr)
1033 {
1034         int ret = -ENXIO;
1035
1036         switch (attr->group) {
1037         default:
1038                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1039                 break;
1040         }
1041
1042         return ret;
1043 }
1044
1045 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1046                                    struct kvm_vcpu_events *events)
1047 {
1048         memset(events, 0, sizeof(*events));
1049
1050         return __kvm_arm_vcpu_get_events(vcpu, events);
1051 }
1052
1053 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1054                                    struct kvm_vcpu_events *events)
1055 {
1056         int i;
1057
1058         /* check whether the reserved field is zero */
1059         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1060                 if (events->reserved[i])
1061                         return -EINVAL;
1062
1063         /* check whether the pad field is zero */
1064         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1065                 if (events->exception.pad[i])
1066                         return -EINVAL;
1067
1068         return __kvm_arm_vcpu_set_events(vcpu, events);
1069 }
1070
1071 long kvm_arch_vcpu_ioctl(struct file *filp,
1072                          unsigned int ioctl, unsigned long arg)
1073 {
1074         struct kvm_vcpu *vcpu = filp->private_data;
1075         void __user *argp = (void __user *)arg;
1076         struct kvm_device_attr attr;
1077         long r;
1078
1079         switch (ioctl) {
1080         case KVM_ARM_VCPU_INIT: {
1081                 struct kvm_vcpu_init init;
1082
1083                 r = -EFAULT;
1084                 if (copy_from_user(&init, argp, sizeof(init)))
1085                         break;
1086
1087                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1088                 break;
1089         }
1090         case KVM_SET_ONE_REG:
1091         case KVM_GET_ONE_REG: {
1092                 struct kvm_one_reg reg;
1093
1094                 r = -ENOEXEC;
1095                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1096                         break;
1097
1098                 r = -EFAULT;
1099                 if (copy_from_user(&reg, argp, sizeof(reg)))
1100                         break;
1101
1102                 if (ioctl == KVM_SET_ONE_REG)
1103                         r = kvm_arm_set_reg(vcpu, &reg);
1104                 else
1105                         r = kvm_arm_get_reg(vcpu, &reg);
1106                 break;
1107         }
1108         case KVM_GET_REG_LIST: {
1109                 struct kvm_reg_list __user *user_list = argp;
1110                 struct kvm_reg_list reg_list;
1111                 unsigned n;
1112
1113                 r = -ENOEXEC;
1114                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1115                         break;
1116
1117                 r = -EPERM;
1118                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1119                         break;
1120
1121                 r = -EFAULT;
1122                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1123                         break;
1124                 n = reg_list.n;
1125                 reg_list.n = kvm_arm_num_regs(vcpu);
1126                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1127                         break;
1128                 r = -E2BIG;
1129                 if (n < reg_list.n)
1130                         break;
1131                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1132                 break;
1133         }
1134         case KVM_SET_DEVICE_ATTR: {
1135                 r = -EFAULT;
1136                 if (copy_from_user(&attr, argp, sizeof(attr)))
1137                         break;
1138                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1139                 break;
1140         }
1141         case KVM_GET_DEVICE_ATTR: {
1142                 r = -EFAULT;
1143                 if (copy_from_user(&attr, argp, sizeof(attr)))
1144                         break;
1145                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1146                 break;
1147         }
1148         case KVM_HAS_DEVICE_ATTR: {
1149                 r = -EFAULT;
1150                 if (copy_from_user(&attr, argp, sizeof(attr)))
1151                         break;
1152                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1153                 break;
1154         }
1155         case KVM_GET_VCPU_EVENTS: {
1156                 struct kvm_vcpu_events events;
1157
1158                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1159                         return -EINVAL;
1160
1161                 if (copy_to_user(argp, &events, sizeof(events)))
1162                         return -EFAULT;
1163
1164                 return 0;
1165         }
1166         case KVM_SET_VCPU_EVENTS: {
1167                 struct kvm_vcpu_events events;
1168
1169                 if (copy_from_user(&events, argp, sizeof(events)))
1170                         return -EFAULT;
1171
1172                 return kvm_arm_vcpu_set_events(vcpu, &events);
1173         }
1174         case KVM_ARM_VCPU_FINALIZE: {
1175                 int what;
1176
1177                 if (!kvm_vcpu_initialized(vcpu))
1178                         return -ENOEXEC;
1179
1180                 if (get_user(what, (const int __user *)argp))
1181                         return -EFAULT;
1182
1183                 return kvm_arm_vcpu_finalize(vcpu, what);
1184         }
1185         default:
1186                 r = -EINVAL;
1187         }
1188
1189         return r;
1190 }
1191
1192 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
1193 {
1194
1195 }
1196
1197 void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
1198                                         struct kvm_memory_slot *memslot)
1199 {
1200         kvm_flush_remote_tlbs(kvm);
1201 }
1202
1203 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1204                                         struct kvm_arm_device_addr *dev_addr)
1205 {
1206         unsigned long dev_id, type;
1207
1208         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1209                 KVM_ARM_DEVICE_ID_SHIFT;
1210         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1211                 KVM_ARM_DEVICE_TYPE_SHIFT;
1212
1213         switch (dev_id) {
1214         case KVM_ARM_DEVICE_VGIC_V2:
1215                 if (!vgic_present)
1216                         return -ENXIO;
1217                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1218         default:
1219                 return -ENODEV;
1220         }
1221 }
1222
1223 long kvm_arch_vm_ioctl(struct file *filp,
1224                        unsigned int ioctl, unsigned long arg)
1225 {
1226         struct kvm *kvm = filp->private_data;
1227         void __user *argp = (void __user *)arg;
1228
1229         switch (ioctl) {
1230         case KVM_CREATE_IRQCHIP: {
1231                 int ret;
1232                 if (!vgic_present)
1233                         return -ENXIO;
1234                 mutex_lock(&kvm->lock);
1235                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1236                 mutex_unlock(&kvm->lock);
1237                 return ret;
1238         }
1239         case KVM_ARM_SET_DEVICE_ADDR: {
1240                 struct kvm_arm_device_addr dev_addr;
1241
1242                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1243                         return -EFAULT;
1244                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1245         }
1246         case KVM_ARM_PREFERRED_TARGET: {
1247                 int err;
1248                 struct kvm_vcpu_init init;
1249
1250                 err = kvm_vcpu_preferred_target(&init);
1251                 if (err)
1252                         return err;
1253
1254                 if (copy_to_user(argp, &init, sizeof(init)))
1255                         return -EFAULT;
1256
1257                 return 0;
1258         }
1259         default:
1260                 return -EINVAL;
1261         }
1262 }
1263
1264 static void cpu_init_hyp_mode(void)
1265 {
1266         phys_addr_t pgd_ptr;
1267         unsigned long hyp_stack_ptr;
1268         unsigned long stack_page;
1269         unsigned long vector_ptr;
1270
1271         /* Switch from the HYP stub to our own HYP init vector */
1272         __hyp_set_vectors(kvm_get_idmap_vector());
1273
1274         pgd_ptr = kvm_mmu_get_httbr();
1275         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1276         hyp_stack_ptr = stack_page + PAGE_SIZE;
1277         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1278
1279         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1280         __cpu_init_stage2();
1281 }
1282
1283 static void cpu_hyp_reset(void)
1284 {
1285         if (!is_kernel_in_hyp_mode())
1286                 __hyp_reset_vectors();
1287 }
1288
1289 static void cpu_hyp_reinit(void)
1290 {
1291         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1292
1293         cpu_hyp_reset();
1294
1295         if (is_kernel_in_hyp_mode())
1296                 kvm_timer_init_vhe();
1297         else
1298                 cpu_init_hyp_mode();
1299
1300         kvm_arm_init_debug();
1301
1302         if (vgic_present)
1303                 kvm_vgic_init_cpu_hardware();
1304 }
1305
1306 static void _kvm_arch_hardware_enable(void *discard)
1307 {
1308         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1309                 cpu_hyp_reinit();
1310                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1311         }
1312 }
1313
1314 int kvm_arch_hardware_enable(void)
1315 {
1316         _kvm_arch_hardware_enable(NULL);
1317         return 0;
1318 }
1319
1320 static void _kvm_arch_hardware_disable(void *discard)
1321 {
1322         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1323                 cpu_hyp_reset();
1324                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1325         }
1326 }
1327
1328 void kvm_arch_hardware_disable(void)
1329 {
1330         _kvm_arch_hardware_disable(NULL);
1331 }
1332
1333 #ifdef CONFIG_CPU_PM
1334 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1335                                     unsigned long cmd,
1336                                     void *v)
1337 {
1338         /*
1339          * kvm_arm_hardware_enabled is left with its old value over
1340          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1341          * re-enable hyp.
1342          */
1343         switch (cmd) {
1344         case CPU_PM_ENTER:
1345                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1346                         /*
1347                          * don't update kvm_arm_hardware_enabled here
1348                          * so that the hardware will be re-enabled
1349                          * when we resume. See below.
1350                          */
1351                         cpu_hyp_reset();
1352
1353                 return NOTIFY_OK;
1354         case CPU_PM_ENTER_FAILED:
1355         case CPU_PM_EXIT:
1356                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1357                         /* The hardware was enabled before suspend. */
1358                         cpu_hyp_reinit();
1359
1360                 return NOTIFY_OK;
1361
1362         default:
1363                 return NOTIFY_DONE;
1364         }
1365 }
1366
1367 static struct notifier_block hyp_init_cpu_pm_nb = {
1368         .notifier_call = hyp_init_cpu_pm_notifier,
1369 };
1370
1371 static void __init hyp_cpu_pm_init(void)
1372 {
1373         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1374 }
1375 static void __init hyp_cpu_pm_exit(void)
1376 {
1377         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1378 }
1379 #else
1380 static inline void hyp_cpu_pm_init(void)
1381 {
1382 }
1383 static inline void hyp_cpu_pm_exit(void)
1384 {
1385 }
1386 #endif
1387
1388 static int init_common_resources(void)
1389 {
1390         kvm_set_ipa_limit();
1391
1392         return 0;
1393 }
1394
1395 static int init_subsystems(void)
1396 {
1397         int err = 0;
1398
1399         /*
1400          * Enable hardware so that subsystem initialisation can access EL2.
1401          */
1402         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1403
1404         /*
1405          * Register CPU lower-power notifier
1406          */
1407         hyp_cpu_pm_init();
1408
1409         /*
1410          * Init HYP view of VGIC
1411          */
1412         err = kvm_vgic_hyp_init();
1413         switch (err) {
1414         case 0:
1415                 vgic_present = true;
1416                 break;
1417         case -ENODEV:
1418         case -ENXIO:
1419                 vgic_present = false;
1420                 err = 0;
1421                 break;
1422         default:
1423                 goto out;
1424         }
1425
1426         /*
1427          * Init HYP architected timer support
1428          */
1429         err = kvm_timer_hyp_init(vgic_present);
1430         if (err)
1431                 goto out;
1432
1433         kvm_perf_init();
1434         kvm_coproc_table_init();
1435
1436 out:
1437         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1438
1439         return err;
1440 }
1441
1442 static void teardown_hyp_mode(void)
1443 {
1444         int cpu;
1445
1446         free_hyp_pgds();
1447         for_each_possible_cpu(cpu)
1448                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1449 }
1450
1451 /**
1452  * Inits Hyp-mode on all online CPUs
1453  */
1454 static int init_hyp_mode(void)
1455 {
1456         int cpu;
1457         int err = 0;
1458
1459         /*
1460          * Allocate Hyp PGD and setup Hyp identity mapping
1461          */
1462         err = kvm_mmu_init();
1463         if (err)
1464                 goto out_err;
1465
1466         /*
1467          * Allocate stack pages for Hypervisor-mode
1468          */
1469         for_each_possible_cpu(cpu) {
1470                 unsigned long stack_page;
1471
1472                 stack_page = __get_free_page(GFP_KERNEL);
1473                 if (!stack_page) {
1474                         err = -ENOMEM;
1475                         goto out_err;
1476                 }
1477
1478                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1479         }
1480
1481         /*
1482          * Map the Hyp-code called directly from the host
1483          */
1484         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1485                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1486         if (err) {
1487                 kvm_err("Cannot map world-switch code\n");
1488                 goto out_err;
1489         }
1490
1491         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1492                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1493         if (err) {
1494                 kvm_err("Cannot map rodata section\n");
1495                 goto out_err;
1496         }
1497
1498         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1499                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1500         if (err) {
1501                 kvm_err("Cannot map bss section\n");
1502                 goto out_err;
1503         }
1504
1505         err = kvm_map_vectors();
1506         if (err) {
1507                 kvm_err("Cannot map vectors\n");
1508                 goto out_err;
1509         }
1510
1511         /*
1512          * Map the Hyp stack pages
1513          */
1514         for_each_possible_cpu(cpu) {
1515                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1516                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1517                                           PAGE_HYP);
1518
1519                 if (err) {
1520                         kvm_err("Cannot map hyp stack\n");
1521                         goto out_err;
1522                 }
1523         }
1524
1525         for_each_possible_cpu(cpu) {
1526                 kvm_host_data_t *cpu_data;
1527
1528                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1529                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1530
1531                 if (err) {
1532                         kvm_err("Cannot map host CPU state: %d\n", err);
1533                         goto out_err;
1534                 }
1535         }
1536
1537         err = hyp_map_aux_data();
1538         if (err)
1539                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1540
1541         return 0;
1542
1543 out_err:
1544         teardown_hyp_mode();
1545         kvm_err("error initializing Hyp mode: %d\n", err);
1546         return err;
1547 }
1548
1549 static void check_kvm_target_cpu(void *ret)
1550 {
1551         *(int *)ret = kvm_target_cpu();
1552 }
1553
1554 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1555 {
1556         struct kvm_vcpu *vcpu;
1557         int i;
1558
1559         mpidr &= MPIDR_HWID_BITMASK;
1560         kvm_for_each_vcpu(i, vcpu, kvm) {
1561                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1562                         return vcpu;
1563         }
1564         return NULL;
1565 }
1566
1567 bool kvm_arch_has_irq_bypass(void)
1568 {
1569         return true;
1570 }
1571
1572 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1573                                       struct irq_bypass_producer *prod)
1574 {
1575         struct kvm_kernel_irqfd *irqfd =
1576                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1577
1578         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1579                                           &irqfd->irq_entry);
1580 }
1581 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1582                                       struct irq_bypass_producer *prod)
1583 {
1584         struct kvm_kernel_irqfd *irqfd =
1585                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1586
1587         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1588                                      &irqfd->irq_entry);
1589 }
1590
1591 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1592 {
1593         struct kvm_kernel_irqfd *irqfd =
1594                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1595
1596         kvm_arm_halt_guest(irqfd->kvm);
1597 }
1598
1599 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1600 {
1601         struct kvm_kernel_irqfd *irqfd =
1602                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1603
1604         kvm_arm_resume_guest(irqfd->kvm);
1605 }
1606
1607 /**
1608  * Initialize Hyp-mode and memory mappings on all CPUs.
1609  */
1610 int kvm_arch_init(void *opaque)
1611 {
1612         int err;
1613         int ret, cpu;
1614         bool in_hyp_mode;
1615
1616         if (!is_hyp_mode_available()) {
1617                 kvm_info("HYP mode not available\n");
1618                 return -ENODEV;
1619         }
1620
1621         in_hyp_mode = is_kernel_in_hyp_mode();
1622
1623         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1624                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1625                 return -ENODEV;
1626         }
1627
1628         for_each_online_cpu(cpu) {
1629                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1630                 if (ret < 0) {
1631                         kvm_err("Error, CPU %d not supported!\n", cpu);
1632                         return -ENODEV;
1633                 }
1634         }
1635
1636         err = init_common_resources();
1637         if (err)
1638                 return err;
1639
1640         err = kvm_arm_init_sve();
1641         if (err)
1642                 return err;
1643
1644         if (!in_hyp_mode) {
1645                 err = init_hyp_mode();
1646                 if (err)
1647                         goto out_err;
1648         }
1649
1650         err = init_subsystems();
1651         if (err)
1652                 goto out_hyp;
1653
1654         if (in_hyp_mode)
1655                 kvm_info("VHE mode initialized successfully\n");
1656         else
1657                 kvm_info("Hyp mode initialized successfully\n");
1658
1659         return 0;
1660
1661 out_hyp:
1662         hyp_cpu_pm_exit();
1663         if (!in_hyp_mode)
1664                 teardown_hyp_mode();
1665 out_err:
1666         return err;
1667 }
1668
1669 /* NOP: Compiling as a module not supported */
1670 void kvm_arch_exit(void)
1671 {
1672         kvm_perf_teardown();
1673 }
1674
1675 static int arm_init(void)
1676 {
1677         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1678         return rc;
1679 }
1680
1681 module_init(arm_init);