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)
  68 {
  69         return 0;
  70 }
  71
  72 int kvm_arch_check_processor_compat(void)
  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 }
 630
 631 /**
 632  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 633  * @vcpu:       The VCPU pointer
 634  * @run:        The kvm_run structure pointer used for userspace state exchange
 635  *
 636  * This function is called through the VCPU_RUN ioctl called from user space. It
 637  * will execute VM code in a loop until the time slice for the process is used
 638  * or some emulation is needed from user space in which case the function will
 639  * return with return value 0 and with the kvm_run structure filled in with the
 640  * required data for the requested emulation.
 641  */
 642 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 643 {
 644         int ret;
 645
 646         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 647                 return -ENOEXEC;
 648
 649         ret = kvm_vcpu_first_run_init(vcpu);
 650         if (ret)
 651                 return ret;
 652
 653         if (run->exit_reason == KVM_EXIT_MMIO) {
 654                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 655                 if (ret)
 656                         return ret;
 657         }
 658
 659         if (run->immediate_exit)
 660                 return -EINTR;
 661
 662         vcpu_load(vcpu);
 663
 664         kvm_sigset_activate(vcpu);
 665
 666         ret = 1;
 667         run->exit_reason = KVM_EXIT_UNKNOWN;
 668         while (ret > 0) {
 669                 /*
 670                  * Check conditions before entering the guest
 671                  */
 672                 cond_resched();
 673
 674                 update_vmid(&vcpu->kvm->arch.vmid);
 675
 676                 check_vcpu_requests(vcpu);
 677
 678                 /*
 679                  * Preparing the interrupts to be injected also
 680                  * involves poking the GIC, which must be done in a
 681                  * non-preemptible context.
 682                  */
 683                 preempt_disable();
 684
 685                 kvm_pmu_flush_hwstate(vcpu);
 686
 687                 local_irq_disable();
 688
 689                 kvm_vgic_flush_hwstate(vcpu);
 690
 691                 /*
 692                  * Exit if we have a signal pending so that we can deliver the
 693                  * signal to user space.
 694                  */
 695                 if (signal_pending(current)) {
 696                         ret = -EINTR;
 697                         run->exit_reason = KVM_EXIT_INTR;
 698                 }
 699
 700                 /*
 701                  * If we're using a userspace irqchip, then check if we need
 702                  * to tell a userspace irqchip about timer or PMU level
 703                  * changes and if so, exit to userspace (the actual level
 704                  * state gets updated in kvm_timer_update_run and
 705                  * kvm_pmu_update_run below).
 706                  */
 707                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 708                         if (kvm_timer_should_notify_user(vcpu) ||
 709                             kvm_pmu_should_notify_user(vcpu)) {
 710                                 ret = -EINTR;
 711                                 run->exit_reason = KVM_EXIT_INTR;
 712                         }
 713                 }
 714
 715                 /*
 716                  * Ensure we set mode to IN_GUEST_MODE after we disable
 717                  * interrupts and before the final VCPU requests check.
 718                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 719                  * Documentation/virt/kvm/vcpu-requests.rst
 720                  */
 721                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 722
 723                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 724                     kvm_request_pending(vcpu)) {
 725                         vcpu->mode = OUTSIDE_GUEST_MODE;
 726                         isb(); /* Ensure work in x_flush_hwstate is committed */
 727                         kvm_pmu_sync_hwstate(vcpu);
 728                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 729                                 kvm_timer_sync_hwstate(vcpu);
 730                         kvm_vgic_sync_hwstate(vcpu);
 731                         local_irq_enable();
 732                         preempt_enable();
 733                         continue;
 734                 }
 735
 736                 kvm_arm_setup_debug(vcpu);
 737
 738                 /**************************************************************
 739                  * Enter the guest
 740                  */
 741                 trace_kvm_entry(*vcpu_pc(vcpu));
 742                 guest_enter_irqoff();
 743
 744                 if (has_vhe()) {
 745                         ret = kvm_vcpu_run_vhe(vcpu);
 746                 } else {
 747                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 748                 }
 749
 750                 vcpu->mode = OUTSIDE_GUEST_MODE;
 751                 vcpu->stat.exits++;
 752                 /*
 753                  * Back from guest
 754                  *************************************************************/
 755
 756                 kvm_arm_clear_debug(vcpu);
 757
 758                 /*
 759                  * We must sync the PMU state before the vgic state so
 760                  * that the vgic can properly sample the updated state of the
 761                  * interrupt line.
 762                  */
 763                 kvm_pmu_sync_hwstate(vcpu);
 764
 765                 /*
 766                  * Sync the vgic state before syncing the timer state because
 767                  * the timer code needs to know if the virtual timer
 768                  * interrupts are active.
 769                  */
 770                 kvm_vgic_sync_hwstate(vcpu);
 771
 772                 /*
 773                  * Sync the timer hardware state before enabling interrupts as
 774                  * we don't want vtimer interrupts to race with syncing the
 775                  * timer virtual interrupt state.
 776                  */
 777                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 778                         kvm_timer_sync_hwstate(vcpu);
 779
 780                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 781
 782                 /*
 783                  * We may have taken a host interrupt in HYP mode (ie
 784                  * while executing the guest). This interrupt is still
 785                  * pending, as we haven't serviced it yet!
 786                  *
 787                  * We're now back in SVC mode, with interrupts
 788                  * disabled.  Enabling the interrupts now will have
 789                  * the effect of taking the interrupt again, in SVC
 790                  * mode this time.
 791                  */
 792                 local_irq_enable();
 793
 794                 /*
 795                  * We do local_irq_enable() before calling guest_exit() so
 796                  * that if a timer interrupt hits while running the guest we
 797                  * account that tick as being spent in the guest.  We enable
 798                  * preemption after calling guest_exit() so that if we get
 799                  * preempted we make sure ticks after that is not counted as
 800                  * guest time.
 801                  */
 802                 guest_exit();
 803                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 804
 805                 /* Exit types that need handling before we can be preempted */
 806                 handle_exit_early(vcpu, run, ret);
 807
 808                 preempt_enable();
 809
 810                 ret = handle_exit(vcpu, run, ret);
 811         }
 812
 813         /* Tell userspace about in-kernel device output levels */
 814         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 815                 kvm_timer_update_run(vcpu);
 816                 kvm_pmu_update_run(vcpu);
 817         }
 818
 819         kvm_sigset_deactivate(vcpu);
 820
 821         vcpu_put(vcpu);
 822         return ret;
 823 }
 824
 825 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 826 {
 827         int bit_index;
 828         bool set;
 829         unsigned long *hcr;
 830
 831         if (number == KVM_ARM_IRQ_CPU_IRQ)
 832                 bit_index = __ffs(HCR_VI);
 833         else /* KVM_ARM_IRQ_CPU_FIQ */
 834                 bit_index = __ffs(HCR_VF);
 835
 836         hcr = vcpu_hcr(vcpu);
 837         if (level)
 838                 set = test_and_set_bit(bit_index, hcr);
 839         else
 840                 set = test_and_clear_bit(bit_index, hcr);
 841
 842         /*
 843          * If we didn't change anything, no need to wake up or kick other CPUs
 844          */
 845         if (set == level)
 846                 return 0;
 847
 848         /*
 849          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 850          * trigger a world-switch round on the running physical CPU to set the
 851          * virtual IRQ/FIQ fields in the HCR appropriately.
 852          */
 853         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 854         kvm_vcpu_kick(vcpu);
 855
 856         return 0;
 857 }
 858
 859 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 860                           bool line_status)
 861 {
 862         u32 irq = irq_level->irq;
 863         unsigned int irq_type, vcpu_idx, irq_num;
 864         int nrcpus = atomic_read(&kvm->online_vcpus);
 865         struct kvm_vcpu *vcpu = NULL;
 866         bool level = irq_level->level;
 867
 868         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 869         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 870         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 871         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 872
 873         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 874
 875         switch (irq_type) {
 876         case KVM_ARM_IRQ_TYPE_CPU:
 877                 if (irqchip_in_kernel(kvm))
 878                         return -ENXIO;
 879
 880                 if (vcpu_idx >= nrcpus)
 881                         return -EINVAL;
 882
 883                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 884                 if (!vcpu)
 885                         return -EINVAL;
 886
 887                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 888                         return -EINVAL;
 889
 890                 return vcpu_interrupt_line(vcpu, irq_num, level);
 891         case KVM_ARM_IRQ_TYPE_PPI:
 892                 if (!irqchip_in_kernel(kvm))
 893                         return -ENXIO;
 894
 895                 if (vcpu_idx >= nrcpus)
 896                         return -EINVAL;
 897
 898                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 899                 if (!vcpu)
 900                         return -EINVAL;
 901
 902                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 903                         return -EINVAL;
 904
 905                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 906         case KVM_ARM_IRQ_TYPE_SPI:
 907                 if (!irqchip_in_kernel(kvm))
 908                         return -ENXIO;
 909
 910                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 911                         return -EINVAL;
 912
 913                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 914         }
 915
 916         return -EINVAL;
 917 }
 918
 919 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 920                                const struct kvm_vcpu_init *init)
 921 {
 922         unsigned int i, ret;
 923         int phys_target = kvm_target_cpu();
 924
 925         if (init->target != phys_target)
 926                 return -EINVAL;
 927
 928         /*
 929          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 930          * use the same target.
 931          */
 932         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 933                 return -EINVAL;
 934
 935         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 936         for (i = 0; i < sizeof(init->features) * 8; i++) {
 937                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 938
 939                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 940                         return -ENOENT;
 941
 942                 /*
 943                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 944                  * use the same feature set.
 945                  */
 946                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 947                     test_bit(i, vcpu->arch.features) != set)
 948                         return -EINVAL;
 949
 950                 if (set)
 951                         set_bit(i, vcpu->arch.features);
 952         }
 953
 954         vcpu->arch.target = phys_target;
 955
 956         /* Now we know what it is, we can reset it. */
 957         ret = kvm_reset_vcpu(vcpu);
 958         if (ret) {
 959                 vcpu->arch.target = -1;
 960                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 961         }
 962
 963         return ret;
 964 }
 965
 966 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 967                                          struct kvm_vcpu_init *init)
 968 {
 969         int ret;
 970
 971         ret = kvm_vcpu_set_target(vcpu, init);
 972         if (ret)
 973                 return ret;
 974
 975         /*
 976          * Ensure a rebooted VM will fault in RAM pages and detect if the
 977          * guest MMU is turned off and flush the caches as needed.
 978          */
 979         if (vcpu->arch.has_run_once)
 980                 stage2_unmap_vm(vcpu->kvm);
 981
 982         vcpu_reset_hcr(vcpu);
 983
 984         /*
 985          * Handle the "start in power-off" case.
 986          */
 987         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 988                 vcpu_power_off(vcpu);
 989         else
 990                 vcpu->arch.power_off = false;
 991
 992         return 0;
 993 }
 994
 995 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 996                                  struct kvm_device_attr *attr)
 997 {
 998         int ret = -ENXIO;
 999
1000         switch (attr->group) {
1001         default:
1002                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1003                 break;
1004         }
1005
1006         return ret;
1007 }
1008
1009 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1010                                  struct kvm_device_attr *attr)
1011 {
1012         int ret = -ENXIO;
1013
1014         switch (attr->group) {
1015         default:
1016                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1017                 break;
1018         }
1019
1020         return ret;
1021 }
1022
1023 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1024                                  struct kvm_device_attr *attr)
1025 {
1026         int ret = -ENXIO;
1027
1028         switch (attr->group) {
1029         default:
1030                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1031                 break;
1032         }
1033
1034         return ret;
1035 }
1036
1037 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1038                                    struct kvm_vcpu_events *events)
1039 {
1040         memset(events, 0, sizeof(*events));
1041
1042         return __kvm_arm_vcpu_get_events(vcpu, events);
1043 }
1044
1045 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1046                                    struct kvm_vcpu_events *events)
1047 {
1048         int i;
1049
1050         /* check whether the reserved field is zero */
1051         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1052                 if (events->reserved[i])
1053                         return -EINVAL;
1054
1055         /* check whether the pad field is zero */
1056         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1057                 if (events->exception.pad[i])
1058                         return -EINVAL;
1059
1060         return __kvm_arm_vcpu_set_events(vcpu, events);
1061 }
1062
1063 long kvm_arch_vcpu_ioctl(struct file *filp,
1064                          unsigned int ioctl, unsigned long arg)
1065 {
1066         struct kvm_vcpu *vcpu = filp->private_data;
1067         void __user *argp = (void __user *)arg;
1068         struct kvm_device_attr attr;
1069         long r;
1070
1071         switch (ioctl) {
1072         case KVM_ARM_VCPU_INIT: {
1073                 struct kvm_vcpu_init init;
1074
1075                 r = -EFAULT;
1076                 if (copy_from_user(&init, argp, sizeof(init)))
1077                         break;
1078
1079                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1080                 break;
1081         }
1082         case KVM_SET_ONE_REG:
1083         case KVM_GET_ONE_REG: {
1084                 struct kvm_one_reg reg;
1085
1086                 r = -ENOEXEC;
1087                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1088                         break;
1089
1090                 r = -EFAULT;
1091                 if (copy_from_user(&reg, argp, sizeof(reg)))
1092                         break;
1093
1094                 if (ioctl == KVM_SET_ONE_REG)
1095                         r = kvm_arm_set_reg(vcpu, &reg);
1096                 else
1097                         r = kvm_arm_get_reg(vcpu, &reg);
1098                 break;
1099         }
1100         case KVM_GET_REG_LIST: {
1101                 struct kvm_reg_list __user *user_list = argp;
1102                 struct kvm_reg_list reg_list;
1103                 unsigned n;
1104
1105                 r = -ENOEXEC;
1106                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1107                         break;
1108
1109                 r = -EPERM;
1110                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1111                         break;
1112
1113                 r = -EFAULT;
1114                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1115                         break;
1116                 n = reg_list.n;
1117                 reg_list.n = kvm_arm_num_regs(vcpu);
1118                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1119                         break;
1120                 r = -E2BIG;
1121                 if (n < reg_list.n)
1122                         break;
1123                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1124                 break;
1125         }
1126         case KVM_SET_DEVICE_ATTR: {
1127                 r = -EFAULT;
1128                 if (copy_from_user(&attr, argp, sizeof(attr)))
1129                         break;
1130                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1131                 break;
1132         }
1133         case KVM_GET_DEVICE_ATTR: {
1134                 r = -EFAULT;
1135                 if (copy_from_user(&attr, argp, sizeof(attr)))
1136                         break;
1137                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1138                 break;
1139         }
1140         case KVM_HAS_DEVICE_ATTR: {
1141                 r = -EFAULT;
1142                 if (copy_from_user(&attr, argp, sizeof(attr)))
1143                         break;
1144                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1145                 break;
1146         }
1147         case KVM_GET_VCPU_EVENTS: {
1148                 struct kvm_vcpu_events events;
1149
1150                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1151                         return -EINVAL;
1152
1153                 if (copy_to_user(argp, &events, sizeof(events)))
1154                         return -EFAULT;
1155
1156                 return 0;
1157         }
1158         case KVM_SET_VCPU_EVENTS: {
1159                 struct kvm_vcpu_events events;
1160
1161                 if (copy_from_user(&events, argp, sizeof(events)))
1162                         return -EFAULT;
1163
1164                 return kvm_arm_vcpu_set_events(vcpu, &events);
1165         }
1166         case KVM_ARM_VCPU_FINALIZE: {
1167                 int what;
1168
1169                 if (!kvm_vcpu_initialized(vcpu))
1170                         return -ENOEXEC;
1171
1172                 if (get_user(what, (const int __user *)argp))
1173                         return -EFAULT;
1174
1175                 return kvm_arm_vcpu_finalize(vcpu, what);
1176         }
1177         default:
1178                 r = -EINVAL;
1179         }
1180
1181         return r;
1182 }
1183
1184 /**
1185  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1186  * @kvm: kvm instance
1187  * @log: slot id and address to which we copy the log
1188  *
1189  * Steps 1-4 below provide general overview of dirty page logging. See
1190  * kvm_get_dirty_log_protect() function description for additional details.
1191  *
1192  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1193  * always flush the TLB (step 4) even if previous step failed  and the dirty
1194  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1195  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1196  * writes will be marked dirty for next log read.
1197  *
1198  *   1. Take a snapshot of the bit and clear it if needed.
1199  *   2. Write protect the corresponding page.
1200  *   3. Copy the snapshot to the userspace.
1201  *   4. Flush TLB's if needed.
1202  */
1203 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1204 {
1205         bool flush = false;
1206         int r;
1207
1208         mutex_lock(&kvm->slots_lock);
1209
1210         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1211
1212         if (flush)
1213                 kvm_flush_remote_tlbs(kvm);
1214
1215         mutex_unlock(&kvm->slots_lock);
1216         return r;
1217 }
1218
1219 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1220 {
1221         bool flush = false;
1222         int r;
1223
1224         mutex_lock(&kvm->slots_lock);
1225
1226         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1227
1228         if (flush)
1229                 kvm_flush_remote_tlbs(kvm);
1230
1231         mutex_unlock(&kvm->slots_lock);
1232         return r;
1233 }
1234
1235 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1236                                         struct kvm_arm_device_addr *dev_addr)
1237 {
1238         unsigned long dev_id, type;
1239
1240         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1241                 KVM_ARM_DEVICE_ID_SHIFT;
1242         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1243                 KVM_ARM_DEVICE_TYPE_SHIFT;
1244
1245         switch (dev_id) {
1246         case KVM_ARM_DEVICE_VGIC_V2:
1247                 if (!vgic_present)
1248                         return -ENXIO;
1249                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1250         default:
1251                 return -ENODEV;
1252         }
1253 }
1254
1255 long kvm_arch_vm_ioctl(struct file *filp,
1256                        unsigned int ioctl, unsigned long arg)
1257 {
1258         struct kvm *kvm = filp->private_data;
1259         void __user *argp = (void __user *)arg;
1260
1261         switch (ioctl) {
1262         case KVM_CREATE_IRQCHIP: {
1263                 int ret;
1264                 if (!vgic_present)
1265                         return -ENXIO;
1266                 mutex_lock(&kvm->lock);
1267                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1268                 mutex_unlock(&kvm->lock);
1269                 return ret;
1270         }
1271         case KVM_ARM_SET_DEVICE_ADDR: {
1272                 struct kvm_arm_device_addr dev_addr;
1273
1274                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1275                         return -EFAULT;
1276                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1277         }
1278         case KVM_ARM_PREFERRED_TARGET: {
1279                 int err;
1280                 struct kvm_vcpu_init init;
1281
1282                 err = kvm_vcpu_preferred_target(&init);
1283                 if (err)
1284                         return err;
1285
1286                 if (copy_to_user(argp, &init, sizeof(init)))
1287                         return -EFAULT;
1288
1289                 return 0;
1290         }
1291         default:
1292                 return -EINVAL;
1293         }
1294 }
1295
1296 static void cpu_init_hyp_mode(void)
1297 {
1298         phys_addr_t pgd_ptr;
1299         unsigned long hyp_stack_ptr;
1300         unsigned long stack_page;
1301         unsigned long vector_ptr;
1302
1303         /* Switch from the HYP stub to our own HYP init vector */
1304         __hyp_set_vectors(kvm_get_idmap_vector());
1305
1306         pgd_ptr = kvm_mmu_get_httbr();
1307         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1308         hyp_stack_ptr = stack_page + PAGE_SIZE;
1309         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1310
1311         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1312         __cpu_init_stage2();
1313 }
1314
1315 static void cpu_hyp_reset(void)
1316 {
1317         if (!is_kernel_in_hyp_mode())
1318                 __hyp_reset_vectors();
1319 }
1320
1321 static void cpu_hyp_reinit(void)
1322 {
1323         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1324
1325         cpu_hyp_reset();
1326
1327         if (is_kernel_in_hyp_mode())
1328                 kvm_timer_init_vhe();
1329         else
1330                 cpu_init_hyp_mode();
1331
1332         kvm_arm_init_debug();
1333
1334         if (vgic_present)
1335                 kvm_vgic_init_cpu_hardware();
1336 }
1337
1338 static void _kvm_arch_hardware_enable(void *discard)
1339 {
1340         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1341                 cpu_hyp_reinit();
1342                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1343         }
1344 }
1345
1346 int kvm_arch_hardware_enable(void)
1347 {
1348         _kvm_arch_hardware_enable(NULL);
1349         return 0;
1350 }
1351
1352 static void _kvm_arch_hardware_disable(void *discard)
1353 {
1354         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1355                 cpu_hyp_reset();
1356                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1357         }
1358 }
1359
1360 void kvm_arch_hardware_disable(void)
1361 {
1362         _kvm_arch_hardware_disable(NULL);
1363 }
1364
1365 #ifdef CONFIG_CPU_PM
1366 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1367                                     unsigned long cmd,
1368                                     void *v)
1369 {
1370         /*
1371          * kvm_arm_hardware_enabled is left with its old value over
1372          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1373          * re-enable hyp.
1374          */
1375         switch (cmd) {
1376         case CPU_PM_ENTER:
1377                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1378                         /*
1379                          * don't update kvm_arm_hardware_enabled here
1380                          * so that the hardware will be re-enabled
1381                          * when we resume. See below.
1382                          */
1383                         cpu_hyp_reset();
1384
1385                 return NOTIFY_OK;
1386         case CPU_PM_ENTER_FAILED:
1387         case CPU_PM_EXIT:
1388                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1389                         /* The hardware was enabled before suspend. */
1390                         cpu_hyp_reinit();
1391
1392                 return NOTIFY_OK;
1393
1394         default:
1395                 return NOTIFY_DONE;
1396         }
1397 }
1398
1399 static struct notifier_block hyp_init_cpu_pm_nb = {
1400         .notifier_call = hyp_init_cpu_pm_notifier,
1401 };
1402
1403 static void __init hyp_cpu_pm_init(void)
1404 {
1405         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1406 }
1407 static void __init hyp_cpu_pm_exit(void)
1408 {
1409         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1410 }
1411 #else
1412 static inline void hyp_cpu_pm_init(void)
1413 {
1414 }
1415 static inline void hyp_cpu_pm_exit(void)
1416 {
1417 }
1418 #endif
1419
1420 static int init_common_resources(void)
1421 {
1422         kvm_set_ipa_limit();
1423
1424         return 0;
1425 }
1426
1427 static int init_subsystems(void)
1428 {
1429         int err = 0;
1430
1431         /*
1432          * Enable hardware so that subsystem initialisation can access EL2.
1433          */
1434         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1435
1436         /*
1437          * Register CPU lower-power notifier
1438          */
1439         hyp_cpu_pm_init();
1440
1441         /*
1442          * Init HYP view of VGIC
1443          */
1444         err = kvm_vgic_hyp_init();
1445         switch (err) {
1446         case 0:
1447                 vgic_present = true;
1448                 break;
1449         case -ENODEV:
1450         case -ENXIO:
1451                 vgic_present = false;
1452                 err = 0;
1453                 break;
1454         default:
1455                 goto out;
1456         }
1457
1458         /*
1459          * Init HYP architected timer support
1460          */
1461         err = kvm_timer_hyp_init(vgic_present);
1462         if (err)
1463                 goto out;
1464
1465         kvm_perf_init();
1466         kvm_coproc_table_init();
1467
1468 out:
1469         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1470
1471         return err;
1472 }
1473
1474 static void teardown_hyp_mode(void)
1475 {
1476         int cpu;
1477
1478         free_hyp_pgds();
1479         for_each_possible_cpu(cpu)
1480                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1481 }
1482
1483 /**
1484  * Inits Hyp-mode on all online CPUs
1485  */
1486 static int init_hyp_mode(void)
1487 {
1488         int cpu;
1489         int err = 0;
1490
1491         /*
1492          * Allocate Hyp PGD and setup Hyp identity mapping
1493          */
1494         err = kvm_mmu_init();
1495         if (err)
1496                 goto out_err;
1497
1498         /*
1499          * Allocate stack pages for Hypervisor-mode
1500          */
1501         for_each_possible_cpu(cpu) {
1502                 unsigned long stack_page;
1503
1504                 stack_page = __get_free_page(GFP_KERNEL);
1505                 if (!stack_page) {
1506                         err = -ENOMEM;
1507                         goto out_err;
1508                 }
1509
1510                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1511         }
1512
1513         /*
1514          * Map the Hyp-code called directly from the host
1515          */
1516         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1517                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1518         if (err) {
1519                 kvm_err("Cannot map world-switch code\n");
1520                 goto out_err;
1521         }
1522
1523         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1524                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1525         if (err) {
1526                 kvm_err("Cannot map rodata section\n");
1527                 goto out_err;
1528         }
1529
1530         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1531                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1532         if (err) {
1533                 kvm_err("Cannot map bss section\n");
1534                 goto out_err;
1535         }
1536
1537         err = kvm_map_vectors();
1538         if (err) {
1539                 kvm_err("Cannot map vectors\n");
1540                 goto out_err;
1541         }
1542
1543         /*
1544          * Map the Hyp stack pages
1545          */
1546         for_each_possible_cpu(cpu) {
1547                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1548                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1549                                           PAGE_HYP);
1550
1551                 if (err) {
1552                         kvm_err("Cannot map hyp stack\n");
1553                         goto out_err;
1554                 }
1555         }
1556
1557         for_each_possible_cpu(cpu) {
1558                 kvm_host_data_t *cpu_data;
1559
1560                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1561                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1562
1563                 if (err) {
1564                         kvm_err("Cannot map host CPU state: %d\n", err);
1565                         goto out_err;
1566                 }
1567         }
1568
1569         err = hyp_map_aux_data();
1570         if (err)
1571                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1572
1573         return 0;
1574
1575 out_err:
1576         teardown_hyp_mode();
1577         kvm_err("error initializing Hyp mode: %d\n", err);
1578         return err;
1579 }
1580
1581 static void check_kvm_target_cpu(void *ret)
1582 {
1583         *(int *)ret = kvm_target_cpu();
1584 }
1585
1586 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1587 {
1588         struct kvm_vcpu *vcpu;
1589         int i;
1590
1591         mpidr &= MPIDR_HWID_BITMASK;
1592         kvm_for_each_vcpu(i, vcpu, kvm) {
1593                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1594                         return vcpu;
1595         }
1596         return NULL;
1597 }
1598
1599 bool kvm_arch_has_irq_bypass(void)
1600 {
1601         return true;
1602 }
1603
1604 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1605                                       struct irq_bypass_producer *prod)
1606 {
1607         struct kvm_kernel_irqfd *irqfd =
1608                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1609
1610         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1611                                           &irqfd->irq_entry);
1612 }
1613 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1614                                       struct irq_bypass_producer *prod)
1615 {
1616         struct kvm_kernel_irqfd *irqfd =
1617                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1618
1619         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1620                                      &irqfd->irq_entry);
1621 }
1622
1623 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1624 {
1625         struct kvm_kernel_irqfd *irqfd =
1626                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1627
1628         kvm_arm_halt_guest(irqfd->kvm);
1629 }
1630
1631 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1632 {
1633         struct kvm_kernel_irqfd *irqfd =
1634                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1635
1636         kvm_arm_resume_guest(irqfd->kvm);
1637 }
1638
1639 /**
1640  * Initialize Hyp-mode and memory mappings on all CPUs.
1641  */
1642 int kvm_arch_init(void *opaque)
1643 {
1644         int err;
1645         int ret, cpu;
1646         bool in_hyp_mode;
1647
1648         if (!is_hyp_mode_available()) {
1649                 kvm_info("HYP mode not available\n");
1650                 return -ENODEV;
1651         }
1652
1653         in_hyp_mode = is_kernel_in_hyp_mode();
1654
1655         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1656                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1657                 return -ENODEV;
1658         }
1659
1660         for_each_online_cpu(cpu) {
1661                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1662                 if (ret < 0) {
1663                         kvm_err("Error, CPU %d not supported!\n", cpu);
1664                         return -ENODEV;
1665                 }
1666         }
1667
1668         err = init_common_resources();
1669         if (err)
1670                 return err;
1671
1672         err = kvm_arm_init_sve();
1673         if (err)
1674                 return err;
1675
1676         if (!in_hyp_mode) {
1677                 err = init_hyp_mode();
1678                 if (err)
1679                         goto out_err;
1680         }
1681
1682         err = init_subsystems();
1683         if (err)
1684                 goto out_hyp;
1685
1686         if (in_hyp_mode)
1687                 kvm_info("VHE mode initialized successfully\n");
1688         else
1689                 kvm_info("Hyp mode initialized successfully\n");
1690
1691         return 0;
1692
1693 out_hyp:
1694         hyp_cpu_pm_exit();
1695         if (!in_hyp_mode)
1696                 teardown_hyp_mode();
1697 out_err:
1698         return err;
1699 }
1700
1701 /* NOP: Compiling as a module not supported */
1702 void kvm_arch_exit(void)
1703 {
1704         kvm_perf_teardown();
1705 }
1706
1707 static int arm_init(void)
1708 {
1709         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1710         return rc;
1711 }
1712
1713 module_init(arm_init);