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