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 #include <kvm/arm_pmu.h>
  24 #include <kvm/arm_psci.h>
  25
  26 #define CREATE_TRACE_POINTS
  27 #include "trace.h"
  28
  29 #include <linux/uaccess.h>
  30 #include <asm/ptrace.h>
  31 #include <asm/mman.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/cacheflush.h>
  34 #include <asm/cpufeature.h>
  35 #include <asm/virt.h>
  36 #include <asm/kvm_arm.h>
  37 #include <asm/kvm_asm.h>
  38 #include <asm/kvm_mmu.h>
  39 #include <asm/kvm_emulate.h>
  40 #include <asm/kvm_coproc.h>
  41 #include <asm/sections.h>
  42
  43 #include <kvm/arm_hypercalls.h>
  44 #include <kvm/arm_pmu.h>
  45 #include <kvm/arm_psci.h>
  46
  47 #ifdef REQUIRES_VIRT
  48 __asm__(".arch_extension        virt");
  49 #endif
  50
  51 DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data);
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53
  54 /* Per-CPU variable containing the currently running vcpu. */
  55 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  56
  57 /* The VMID used in the VTTBR */
  58 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  59 static u32 kvm_next_vmid;
  60 static DEFINE_SPINLOCK(kvm_vmid_lock);
  61
  62 static bool vgic_present;
  63
  64 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  69 }
  70
  71 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  72
  73 /**
  74  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  75  * Must be called from non-preemptible context
  76  */
  77 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  78 {
  79         return __this_cpu_read(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  91 {
  92         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  93 }
  94
  95 int kvm_arch_hardware_setup(void)
  96 {
  97         return 0;
  98 }
  99
 100 int kvm_arch_check_processor_compat(void)
 101 {
 102         return 0;
 103 }
 104
 105 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 106                             struct kvm_enable_cap *cap)
 107 {
 108         int r;
 109
 110         if (cap->flags)
 111                 return -EINVAL;
 112
 113         switch (cap->cap) {
 114         case KVM_CAP_ARM_NISV_TO_USER:
 115                 r = 0;
 116                 kvm->arch.return_nisv_io_abort_to_user = true;
 117                 break;
 118         default:
 119                 r = -EINVAL;
 120                 break;
 121         }
 122
 123         return r;
 124 }
 125
 126 /**
 127  * kvm_arch_init_vm - initializes a VM data structure
 128  * @kvm:        pointer to the KVM struct
 129  */
 130 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 131 {
 132         int ret, cpu;
 133
 134         ret = kvm_arm_setup_stage2(kvm, type);
 135         if (ret)
 136                 return ret;
 137
 138         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 139         if (!kvm->arch.last_vcpu_ran)
 140                 return -ENOMEM;
 141
 142         for_each_possible_cpu(cpu)
 143                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 144
 145         ret = kvm_alloc_stage2_pgd(kvm);
 146         if (ret)
 147                 goto out_fail_alloc;
 148
 149         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 150         if (ret)
 151                 goto out_free_stage2_pgd;
 152
 153         kvm_vgic_early_init(kvm);
 154
 155         /* Mark the initial VMID generation invalid */
 156         kvm->arch.vmid.vmid_gen = 0;
 157
 158         /* The maximum number of VCPUs is limited by the host's GIC model */
 159         kvm->arch.max_vcpus = vgic_present ?
 160                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 161
 162         return ret;
 163 out_free_stage2_pgd:
 164         kvm_free_stage2_pgd(kvm);
 165 out_fail_alloc:
 166         free_percpu(kvm->arch.last_vcpu_ran);
 167         kvm->arch.last_vcpu_ran = NULL;
 168         return ret;
 169 }
 170
 171 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 172 {
 173         return 0;
 174 }
 175
 176 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 177 {
 178         return VM_FAULT_SIGBUS;
 179 }
 180
 181
 182 /**
 183  * kvm_arch_destroy_vm - destroy the VM data structure
 184  * @kvm:        pointer to the KVM struct
 185  */
 186 void kvm_arch_destroy_vm(struct kvm *kvm)
 187 {
 188         int i;
 189
 190         kvm_vgic_destroy(kvm);
 191
 192         free_percpu(kvm->arch.last_vcpu_ran);
 193         kvm->arch.last_vcpu_ran = NULL;
 194
 195         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 196                 if (kvm->vcpus[i]) {
 197                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 198                         kvm->vcpus[i] = NULL;
 199                 }
 200         }
 201         atomic_set(&kvm->online_vcpus, 0);
 202 }
 203
 204 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 205 {
 206         int r;
 207         switch (ext) {
 208         case KVM_CAP_IRQCHIP:
 209                 r = vgic_present;
 210                 break;
 211         case KVM_CAP_IOEVENTFD:
 212         case KVM_CAP_DEVICE_CTRL:
 213         case KVM_CAP_USER_MEMORY:
 214         case KVM_CAP_SYNC_MMU:
 215         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 216         case KVM_CAP_ONE_REG:
 217         case KVM_CAP_ARM_PSCI:
 218         case KVM_CAP_ARM_PSCI_0_2:
 219         case KVM_CAP_READONLY_MEM:
 220         case KVM_CAP_MP_STATE:
 221         case KVM_CAP_IMMEDIATE_EXIT:
 222         case KVM_CAP_VCPU_EVENTS:
 223         case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
 224         case KVM_CAP_ARM_NISV_TO_USER:
 225         case KVM_CAP_ARM_INJECT_EXT_DABT:
 226                 r = 1;
 227                 break;
 228         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 229                 r = 1;
 230                 break;
 231         case KVM_CAP_NR_VCPUS:
 232                 r = num_online_cpus();
 233                 break;
 234         case KVM_CAP_MAX_VCPUS:
 235                 r = KVM_MAX_VCPUS;
 236                 break;
 237         case KVM_CAP_MAX_VCPU_ID:
 238                 r = KVM_MAX_VCPU_ID;
 239                 break;
 240         case KVM_CAP_MSI_DEVID:
 241                 if (!kvm)
 242                         r = -EINVAL;
 243                 else
 244                         r = kvm->arch.vgic.msis_require_devid;
 245                 break;
 246         case KVM_CAP_ARM_USER_IRQ:
 247                 /*
 248                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 249                  * (bump this number if adding more devices)
 250                  */
 251                 r = 1;
 252                 break;
 253         default:
 254                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 255                 break;
 256         }
 257         return r;
 258 }
 259
 260 long kvm_arch_dev_ioctl(struct file *filp,
 261                         unsigned int ioctl, unsigned long arg)
 262 {
 263         return -EINVAL;
 264 }
 265
 266 struct kvm *kvm_arch_alloc_vm(void)
 267 {
 268         if (!has_vhe())
 269                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 270
 271         return vzalloc(sizeof(struct kvm));
 272 }
 273
 274 void kvm_arch_free_vm(struct kvm *kvm)
 275 {
 276         if (!has_vhe())
 277                 kfree(kvm);
 278         else
 279                 vfree(kvm);
 280 }
 281
 282 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 283 {
 284         int err;
 285         struct kvm_vcpu *vcpu;
 286
 287         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 288                 err = -EBUSY;
 289                 goto out;
 290         }
 291
 292         if (id >= kvm->arch.max_vcpus) {
 293                 err = -EINVAL;
 294                 goto out;
 295         }
 296
 297         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 298         if (!vcpu) {
 299                 err = -ENOMEM;
 300                 goto out;
 301         }
 302
 303         err = kvm_vcpu_init(vcpu, kvm, id);
 304         if (err)
 305                 goto free_vcpu;
 306
 307         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 308         if (err)
 309                 goto vcpu_uninit;
 310
 311         return vcpu;
 312 vcpu_uninit:
 313         kvm_vcpu_uninit(vcpu);
 314 free_vcpu:
 315         kmem_cache_free(kvm_vcpu_cache, vcpu);
 316 out:
 317         return ERR_PTR(err);
 318 }
 319
 320 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 321 {
 322 }
 323
 324 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 325 {
 326         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 327                 static_branch_dec(&userspace_irqchip_in_use);
 328
 329         kvm_mmu_free_memory_caches(vcpu);
 330         kvm_timer_vcpu_terminate(vcpu);
 331         kvm_pmu_vcpu_destroy(vcpu);
 332         kvm_vcpu_uninit(vcpu);
 333         kmem_cache_free(kvm_vcpu_cache, vcpu);
 334 }
 335
 336 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 337 {
 338         kvm_arch_vcpu_free(vcpu);
 339 }
 340
 341 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 342 {
 343         return kvm_timer_is_pending(vcpu);
 344 }
 345
 346 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 347 {
 348         /*
 349          * If we're about to block (most likely because we've just hit a
 350          * WFI), we need to sync back the state of the GIC CPU interface
 351          * so that we have the latest PMR and group enables. This ensures
 352          * that kvm_arch_vcpu_runnable has up-to-date data to decide
 353          * whether we have pending interrupts.
 354          *
 355          * For the same reason, we want to tell GICv4 that we need
 356          * doorbells to be signalled, should an interrupt become pending.
 357          */
 358         preempt_disable();
 359         kvm_vgic_vmcr_sync(vcpu);
 360         vgic_v4_put(vcpu, true);
 361         preempt_enable();
 362 }
 363
 364 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 365 {
 366         preempt_disable();
 367         vgic_v4_load(vcpu);
 368         preempt_enable();
 369 }
 370
 371 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 372 {
 373         /* Force users to call KVM_ARM_VCPU_INIT */
 374         vcpu->arch.target = -1;
 375         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 376
 377         /* Set up the timer */
 378         kvm_timer_vcpu_init(vcpu);
 379
 380         kvm_pmu_vcpu_init(vcpu);
 381
 382         kvm_arm_reset_debug_ptr(vcpu);
 383
 384         kvm_arm_pvtime_vcpu_init(&vcpu->arch);
 385
 386         return kvm_vgic_vcpu_init(vcpu);
 387 }
 388
 389 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 390 {
 391         int *last_ran;
 392         kvm_host_data_t *cpu_data;
 393
 394         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 395         cpu_data = this_cpu_ptr(&kvm_host_data);
 396
 397         /*
 398          * We might get preempted before the vCPU actually runs, but
 399          * over-invalidation doesn't affect correctness.
 400          */
 401         if (*last_ran != vcpu->vcpu_id) {
 402                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 403                 *last_ran = vcpu->vcpu_id;
 404         }
 405
 406         vcpu->cpu = cpu;
 407         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 408
 409         kvm_arm_set_running_vcpu(vcpu);
 410         kvm_vgic_load(vcpu);
 411         kvm_timer_vcpu_load(vcpu);
 412         kvm_vcpu_load_sysregs(vcpu);
 413         kvm_arch_vcpu_load_fp(vcpu);
 414         kvm_vcpu_pmu_restore_guest(vcpu);
 415         if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
 416                 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
 417
 418         if (single_task_running())
 419                 vcpu_clear_wfx_traps(vcpu);
 420         else
 421                 vcpu_set_wfx_traps(vcpu);
 422
 423         vcpu_ptrauth_setup_lazy(vcpu);
 424 }
 425
 426 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 427 {
 428         kvm_arch_vcpu_put_fp(vcpu);
 429         kvm_vcpu_put_sysregs(vcpu);
 430         kvm_timer_vcpu_put(vcpu);
 431         kvm_vgic_put(vcpu);
 432         kvm_vcpu_pmu_restore_host(vcpu);
 433
 434         vcpu->cpu = -1;
 435
 436         kvm_arm_set_running_vcpu(NULL);
 437 }
 438
 439 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 440 {
 441         vcpu->arch.power_off = true;
 442         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 443         kvm_vcpu_kick(vcpu);
 444 }
 445
 446 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 447                                     struct kvm_mp_state *mp_state)
 448 {
 449         if (vcpu->arch.power_off)
 450                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 451         else
 452                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 453
 454         return 0;
 455 }
 456
 457 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 458                                     struct kvm_mp_state *mp_state)
 459 {
 460         int ret = 0;
 461
 462         switch (mp_state->mp_state) {
 463         case KVM_MP_STATE_RUNNABLE:
 464                 vcpu->arch.power_off = false;
 465                 break;
 466         case KVM_MP_STATE_STOPPED:
 467                 vcpu_power_off(vcpu);
 468                 break;
 469         default:
 470                 ret = -EINVAL;
 471         }
 472
 473         return ret;
 474 }
 475
 476 /**
 477  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 478  * @v:          The VCPU pointer
 479  *
 480  * If the guest CPU is not waiting for interrupts or an interrupt line is
 481  * asserted, the CPU is by definition runnable.
 482  */
 483 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 484 {
 485         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 486         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 487                 && !v->arch.power_off && !v->arch.pause);
 488 }
 489
 490 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 491 {
 492         return vcpu_mode_priv(vcpu);
 493 }
 494
 495 /* Just ensure a guest exit from a particular CPU */
 496 static void exit_vm_noop(void *info)
 497 {
 498 }
 499
 500 void force_vm_exit(const cpumask_t *mask)
 501 {
 502         preempt_disable();
 503         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 504         preempt_enable();
 505 }
 506
 507 /**
 508  * need_new_vmid_gen - check that the VMID is still valid
 509  * @vmid: The VMID to check
 510  *
 511  * return true if there is a new generation of VMIDs being used
 512  *
 513  * The hardware supports a limited set of values with the value zero reserved
 514  * for the host, so we check if an assigned value belongs to a previous
 515  * generation, which which requires us to assign a new value. If we're the
 516  * first to use a VMID for the new generation, we must flush necessary caches
 517  * and TLBs on all CPUs.
 518  */
 519 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 520 {
 521         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 522         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 523         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 524 }
 525
 526 /**
 527  * update_vmid - Update the vmid with a valid VMID for the current generation
 528  * @kvm: The guest that struct vmid belongs to
 529  * @vmid: The stage-2 VMID information struct
 530  */
 531 static void update_vmid(struct kvm_vmid *vmid)
 532 {
 533         if (!need_new_vmid_gen(vmid))
 534                 return;
 535
 536         spin_lock(&kvm_vmid_lock);
 537
 538         /*
 539          * We need to re-check the vmid_gen here to ensure that if another vcpu
 540          * already allocated a valid vmid for this vm, then this vcpu should
 541          * use the same vmid.
 542          */
 543         if (!need_new_vmid_gen(vmid)) {
 544                 spin_unlock(&kvm_vmid_lock);
 545                 return;
 546         }
 547
 548         /* First user of a new VMID generation? */
 549         if (unlikely(kvm_next_vmid == 0)) {
 550                 atomic64_inc(&kvm_vmid_gen);
 551                 kvm_next_vmid = 1;
 552
 553                 /*
 554                  * On SMP we know no other CPUs can use this CPU's or each
 555                  * other's VMID after force_vm_exit returns since the
 556                  * kvm_vmid_lock blocks them from reentry to the guest.
 557                  */
 558                 force_vm_exit(cpu_all_mask);
 559                 /*
 560                  * Now broadcast TLB + ICACHE invalidation over the inner
 561                  * shareable domain to make sure all data structures are
 562                  * clean.
 563                  */
 564                 kvm_call_hyp(__kvm_flush_vm_context);
 565         }
 566
 567         vmid->vmid = kvm_next_vmid;
 568         kvm_next_vmid++;
 569         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 570
 571         smp_wmb();
 572         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 573
 574         spin_unlock(&kvm_vmid_lock);
 575 }
 576
 577 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 578 {
 579         struct kvm *kvm = vcpu->kvm;
 580         int ret = 0;
 581
 582         if (likely(vcpu->arch.has_run_once))
 583                 return 0;
 584
 585         if (!kvm_arm_vcpu_is_finalized(vcpu))
 586                 return -EPERM;
 587
 588         vcpu->arch.has_run_once = true;
 589
 590         if (likely(irqchip_in_kernel(kvm))) {
 591                 /*
 592                  * Map the VGIC hardware resources before running a vcpu the
 593                  * first time on this VM.
 594                  */
 595                 if (unlikely(!vgic_ready(kvm))) {
 596                         ret = kvm_vgic_map_resources(kvm);
 597                         if (ret)
 598                                 return ret;
 599                 }
 600         } else {
 601                 /*
 602                  * Tell the rest of the code that there are userspace irqchip
 603                  * VMs in the wild.
 604                  */
 605                 static_branch_inc(&userspace_irqchip_in_use);
 606         }
 607
 608         ret = kvm_timer_enable(vcpu);
 609         if (ret)
 610                 return ret;
 611
 612         ret = kvm_arm_pmu_v3_enable(vcpu);
 613
 614         return ret;
 615 }
 616
 617 bool kvm_arch_intc_initialized(struct kvm *kvm)
 618 {
 619         return vgic_initialized(kvm);
 620 }
 621
 622 void kvm_arm_halt_guest(struct kvm *kvm)
 623 {
 624         int i;
 625         struct kvm_vcpu *vcpu;
 626
 627         kvm_for_each_vcpu(i, vcpu, kvm)
 628                 vcpu->arch.pause = true;
 629         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 630 }
 631
 632 void kvm_arm_resume_guest(struct kvm *kvm)
 633 {
 634         int i;
 635         struct kvm_vcpu *vcpu;
 636
 637         kvm_for_each_vcpu(i, vcpu, kvm) {
 638                 vcpu->arch.pause = false;
 639                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 640         }
 641 }
 642
 643 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 644 {
 645         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 646
 647         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 648                                        (!vcpu->arch.pause)));
 649
 650         if (vcpu->arch.power_off || vcpu->arch.pause) {
 651                 /* Awaken to handle a signal, request we sleep again later. */
 652                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 653         }
 654
 655         /*
 656          * Make sure we will observe a potential reset request if we've
 657          * observed a change to the power state. Pairs with the smp_wmb() in
 658          * kvm_psci_vcpu_on().
 659          */
 660         smp_rmb();
 661 }
 662
 663 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 664 {
 665         return vcpu->arch.target >= 0;
 666 }
 667
 668 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 669 {
 670         if (kvm_request_pending(vcpu)) {
 671                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 672                         vcpu_req_sleep(vcpu);
 673
 674                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 675                         kvm_reset_vcpu(vcpu);
 676
 677                 /*
 678                  * Clear IRQ_PENDING requests that were made to guarantee
 679                  * that a VCPU sees new virtual interrupts.
 680                  */
 681                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 682
 683                 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
 684                         kvm_update_stolen_time(vcpu);
 685         }
 686 }
 687
 688 /**
 689  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 690  * @vcpu:       The VCPU pointer
 691  * @run:        The kvm_run structure pointer used for userspace state exchange
 692  *
 693  * This function is called through the VCPU_RUN ioctl called from user space. It
 694  * will execute VM code in a loop until the time slice for the process is used
 695  * or some emulation is needed from user space in which case the function will
 696  * return with return value 0 and with the kvm_run structure filled in with the
 697  * required data for the requested emulation.
 698  */
 699 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 700 {
 701         int ret;
 702
 703         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 704                 return -ENOEXEC;
 705
 706         ret = kvm_vcpu_first_run_init(vcpu);
 707         if (ret)
 708                 return ret;
 709
 710         if (run->exit_reason == KVM_EXIT_MMIO) {
 711                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 712                 if (ret)
 713                         return ret;
 714         }
 715
 716         if (run->immediate_exit)
 717                 return -EINTR;
 718
 719         vcpu_load(vcpu);
 720
 721         kvm_sigset_activate(vcpu);
 722
 723         ret = 1;
 724         run->exit_reason = KVM_EXIT_UNKNOWN;
 725         while (ret > 0) {
 726                 /*
 727                  * Check conditions before entering the guest
 728                  */
 729                 cond_resched();
 730
 731                 update_vmid(&vcpu->kvm->arch.vmid);
 732
 733                 check_vcpu_requests(vcpu);
 734
 735                 /*
 736                  * Preparing the interrupts to be injected also
 737                  * involves poking the GIC, which must be done in a
 738                  * non-preemptible context.
 739                  */
 740                 preempt_disable();
 741
 742                 kvm_pmu_flush_hwstate(vcpu);
 743
 744                 local_irq_disable();
 745
 746                 kvm_vgic_flush_hwstate(vcpu);
 747
 748                 /*
 749                  * Exit if we have a signal pending so that we can deliver the
 750                  * signal to user space.
 751                  */
 752                 if (signal_pending(current)) {
 753                         ret = -EINTR;
 754                         run->exit_reason = KVM_EXIT_INTR;
 755                 }
 756
 757                 /*
 758                  * If we're using a userspace irqchip, then check if we need
 759                  * to tell a userspace irqchip about timer or PMU level
 760                  * changes and if so, exit to userspace (the actual level
 761                  * state gets updated in kvm_timer_update_run and
 762                  * kvm_pmu_update_run below).
 763                  */
 764                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 765                         if (kvm_timer_should_notify_user(vcpu) ||
 766                             kvm_pmu_should_notify_user(vcpu)) {
 767                                 ret = -EINTR;
 768                                 run->exit_reason = KVM_EXIT_INTR;
 769                         }
 770                 }
 771
 772                 /*
 773                  * Ensure we set mode to IN_GUEST_MODE after we disable
 774                  * interrupts and before the final VCPU requests check.
 775                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 776                  * Documentation/virt/kvm/vcpu-requests.rst
 777                  */
 778                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 779
 780                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 781                     kvm_request_pending(vcpu)) {
 782                         vcpu->mode = OUTSIDE_GUEST_MODE;
 783                         isb(); /* Ensure work in x_flush_hwstate is committed */
 784                         kvm_pmu_sync_hwstate(vcpu);
 785                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 786                                 kvm_timer_sync_hwstate(vcpu);
 787                         kvm_vgic_sync_hwstate(vcpu);
 788                         local_irq_enable();
 789                         preempt_enable();
 790                         continue;
 791                 }
 792
 793                 kvm_arm_setup_debug(vcpu);
 794
 795                 /**************************************************************
 796                  * Enter the guest
 797                  */
 798                 trace_kvm_entry(*vcpu_pc(vcpu));
 799                 guest_enter_irqoff();
 800
 801                 if (has_vhe()) {
 802                         kvm_arm_vhe_guest_enter();
 803                         ret = kvm_vcpu_run_vhe(vcpu);
 804                         kvm_arm_vhe_guest_exit();
 805                 } else {
 806                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 807                 }
 808
 809                 vcpu->mode = OUTSIDE_GUEST_MODE;
 810                 vcpu->stat.exits++;
 811                 /*
 812                  * Back from guest
 813                  *************************************************************/
 814
 815                 kvm_arm_clear_debug(vcpu);
 816
 817                 /*
 818                  * We must sync the PMU state before the vgic state so
 819                  * that the vgic can properly sample the updated state of the
 820                  * interrupt line.
 821                  */
 822                 kvm_pmu_sync_hwstate(vcpu);
 823
 824                 /*
 825                  * Sync the vgic state before syncing the timer state because
 826                  * the timer code needs to know if the virtual timer
 827                  * interrupts are active.
 828                  */
 829                 kvm_vgic_sync_hwstate(vcpu);
 830
 831                 /*
 832                  * Sync the timer hardware state before enabling interrupts as
 833                  * we don't want vtimer interrupts to race with syncing the
 834                  * timer virtual interrupt state.
 835                  */
 836                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 837                         kvm_timer_sync_hwstate(vcpu);
 838
 839                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 840
 841                 /*
 842                  * We may have taken a host interrupt in HYP mode (ie
 843                  * while executing the guest). This interrupt is still
 844                  * pending, as we haven't serviced it yet!
 845                  *
 846                  * We're now back in SVC mode, with interrupts
 847                  * disabled.  Enabling the interrupts now will have
 848                  * the effect of taking the interrupt again, in SVC
 849                  * mode this time.
 850                  */
 851                 local_irq_enable();
 852
 853                 /*
 854                  * We do local_irq_enable() before calling guest_exit() so
 855                  * that if a timer interrupt hits while running the guest we
 856                  * account that tick as being spent in the guest.  We enable
 857                  * preemption after calling guest_exit() so that if we get
 858                  * preempted we make sure ticks after that is not counted as
 859                  * guest time.
 860                  */
 861                 guest_exit();
 862                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 863
 864                 /* Exit types that need handling before we can be preempted */
 865                 handle_exit_early(vcpu, run, ret);
 866
 867                 preempt_enable();
 868
 869                 ret = handle_exit(vcpu, run, ret);
 870         }
 871
 872         /* Tell userspace about in-kernel device output levels */
 873         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 874                 kvm_timer_update_run(vcpu);
 875                 kvm_pmu_update_run(vcpu);
 876         }
 877
 878         kvm_sigset_deactivate(vcpu);
 879
 880         vcpu_put(vcpu);
 881         return ret;
 882 }
 883
 884 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 885 {
 886         int bit_index;
 887         bool set;
 888         unsigned long *hcr;
 889
 890         if (number == KVM_ARM_IRQ_CPU_IRQ)
 891                 bit_index = __ffs(HCR_VI);
 892         else /* KVM_ARM_IRQ_CPU_FIQ */
 893                 bit_index = __ffs(HCR_VF);
 894
 895         hcr = vcpu_hcr(vcpu);
 896         if (level)
 897                 set = test_and_set_bit(bit_index, hcr);
 898         else
 899                 set = test_and_clear_bit(bit_index, hcr);
 900
 901         /*
 902          * If we didn't change anything, no need to wake up or kick other CPUs
 903          */
 904         if (set == level)
 905                 return 0;
 906
 907         /*
 908          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 909          * trigger a world-switch round on the running physical CPU to set the
 910          * virtual IRQ/FIQ fields in the HCR appropriately.
 911          */
 912         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 913         kvm_vcpu_kick(vcpu);
 914
 915         return 0;
 916 }
 917
 918 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 919                           bool line_status)
 920 {
 921         u32 irq = irq_level->irq;
 922         unsigned int irq_type, vcpu_idx, irq_num;
 923         int nrcpus = atomic_read(&kvm->online_vcpus);
 924         struct kvm_vcpu *vcpu = NULL;
 925         bool level = irq_level->level;
 926
 927         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 928         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 929         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 930         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 931
 932         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 933
 934         switch (irq_type) {
 935         case KVM_ARM_IRQ_TYPE_CPU:
 936                 if (irqchip_in_kernel(kvm))
 937                         return -ENXIO;
 938
 939                 if (vcpu_idx >= nrcpus)
 940                         return -EINVAL;
 941
 942                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 943                 if (!vcpu)
 944                         return -EINVAL;
 945
 946                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 947                         return -EINVAL;
 948
 949                 return vcpu_interrupt_line(vcpu, irq_num, level);
 950         case KVM_ARM_IRQ_TYPE_PPI:
 951                 if (!irqchip_in_kernel(kvm))
 952                         return -ENXIO;
 953
 954                 if (vcpu_idx >= nrcpus)
 955                         return -EINVAL;
 956
 957                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 958                 if (!vcpu)
 959                         return -EINVAL;
 960
 961                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 962                         return -EINVAL;
 963
 964                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 965         case KVM_ARM_IRQ_TYPE_SPI:
 966                 if (!irqchip_in_kernel(kvm))
 967                         return -ENXIO;
 968
 969                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 970                         return -EINVAL;
 971
 972                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 973         }
 974
 975         return -EINVAL;
 976 }
 977
 978 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 979                                const struct kvm_vcpu_init *init)
 980 {
 981         unsigned int i, ret;
 982         int phys_target = kvm_target_cpu();
 983
 984         if (init->target != phys_target)
 985                 return -EINVAL;
 986
 987         /*
 988          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 989          * use the same target.
 990          */
 991         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 992                 return -EINVAL;
 993
 994         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 995         for (i = 0; i < sizeof(init->features) * 8; i++) {
 996                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 997
 998                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 999                         return -ENOENT;
1000
1001                 /*
1002                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
1003                  * use the same feature set.
1004                  */
1005                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1006                     test_bit(i, vcpu->arch.features) != set)
1007                         return -EINVAL;
1008
1009                 if (set)
1010                         set_bit(i, vcpu->arch.features);
1011         }
1012
1013         vcpu->arch.target = phys_target;
1014
1015         /* Now we know what it is, we can reset it. */
1016         ret = kvm_reset_vcpu(vcpu);
1017         if (ret) {
1018                 vcpu->arch.target = -1;
1019                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1020         }
1021
1022         return ret;
1023 }
1024
1025 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1026                                          struct kvm_vcpu_init *init)
1027 {
1028         int ret;
1029
1030         ret = kvm_vcpu_set_target(vcpu, init);
1031         if (ret)
1032                 return ret;
1033
1034         /*
1035          * Ensure a rebooted VM will fault in RAM pages and detect if the
1036          * guest MMU is turned off and flush the caches as needed.
1037          */
1038         if (vcpu->arch.has_run_once)
1039                 stage2_unmap_vm(vcpu->kvm);
1040
1041         vcpu_reset_hcr(vcpu);
1042
1043         /*
1044          * Handle the "start in power-off" case.
1045          */
1046         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1047                 vcpu_power_off(vcpu);
1048         else
1049                 vcpu->arch.power_off = false;
1050
1051         return 0;
1052 }
1053
1054 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1055                                  struct kvm_device_attr *attr)
1056 {
1057         int ret = -ENXIO;
1058
1059         switch (attr->group) {
1060         default:
1061                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1062                 break;
1063         }
1064
1065         return ret;
1066 }
1067
1068 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1069                                  struct kvm_device_attr *attr)
1070 {
1071         int ret = -ENXIO;
1072
1073         switch (attr->group) {
1074         default:
1075                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1076                 break;
1077         }
1078
1079         return ret;
1080 }
1081
1082 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1083                                  struct kvm_device_attr *attr)
1084 {
1085         int ret = -ENXIO;
1086
1087         switch (attr->group) {
1088         default:
1089                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1090                 break;
1091         }
1092
1093         return ret;
1094 }
1095
1096 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1097                                    struct kvm_vcpu_events *events)
1098 {
1099         memset(events, 0, sizeof(*events));
1100
1101         return __kvm_arm_vcpu_get_events(vcpu, events);
1102 }
1103
1104 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1105                                    struct kvm_vcpu_events *events)
1106 {
1107         int i;
1108
1109         /* check whether the reserved field is zero */
1110         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1111                 if (events->reserved[i])
1112                         return -EINVAL;
1113
1114         /* check whether the pad field is zero */
1115         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1116                 if (events->exception.pad[i])
1117                         return -EINVAL;
1118
1119         return __kvm_arm_vcpu_set_events(vcpu, events);
1120 }
1121
1122 long kvm_arch_vcpu_ioctl(struct file *filp,
1123                          unsigned int ioctl, unsigned long arg)
1124 {
1125         struct kvm_vcpu *vcpu = filp->private_data;
1126         void __user *argp = (void __user *)arg;
1127         struct kvm_device_attr attr;
1128         long r;
1129
1130         switch (ioctl) {
1131         case KVM_ARM_VCPU_INIT: {
1132                 struct kvm_vcpu_init init;
1133
1134                 r = -EFAULT;
1135                 if (copy_from_user(&init, argp, sizeof(init)))
1136                         break;
1137
1138                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1139                 break;
1140         }
1141         case KVM_SET_ONE_REG:
1142         case KVM_GET_ONE_REG: {
1143                 struct kvm_one_reg reg;
1144
1145                 r = -ENOEXEC;
1146                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1147                         break;
1148
1149                 r = -EFAULT;
1150                 if (copy_from_user(&reg, argp, sizeof(reg)))
1151                         break;
1152
1153                 if (ioctl == KVM_SET_ONE_REG)
1154                         r = kvm_arm_set_reg(vcpu, &reg);
1155                 else
1156                         r = kvm_arm_get_reg(vcpu, &reg);
1157                 break;
1158         }
1159         case KVM_GET_REG_LIST: {
1160                 struct kvm_reg_list __user *user_list = argp;
1161                 struct kvm_reg_list reg_list;
1162                 unsigned n;
1163
1164                 r = -ENOEXEC;
1165                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1166                         break;
1167
1168                 r = -EPERM;
1169                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1170                         break;
1171
1172                 r = -EFAULT;
1173                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1174                         break;
1175                 n = reg_list.n;
1176                 reg_list.n = kvm_arm_num_regs(vcpu);
1177                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1178                         break;
1179                 r = -E2BIG;
1180                 if (n < reg_list.n)
1181                         break;
1182                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1183                 break;
1184         }
1185         case KVM_SET_DEVICE_ATTR: {
1186                 r = -EFAULT;
1187                 if (copy_from_user(&attr, argp, sizeof(attr)))
1188                         break;
1189                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1190                 break;
1191         }
1192         case KVM_GET_DEVICE_ATTR: {
1193                 r = -EFAULT;
1194                 if (copy_from_user(&attr, argp, sizeof(attr)))
1195                         break;
1196                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1197                 break;
1198         }
1199         case KVM_HAS_DEVICE_ATTR: {
1200                 r = -EFAULT;
1201                 if (copy_from_user(&attr, argp, sizeof(attr)))
1202                         break;
1203                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1204                 break;
1205         }
1206         case KVM_GET_VCPU_EVENTS: {
1207                 struct kvm_vcpu_events events;
1208
1209                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1210                         return -EINVAL;
1211
1212                 if (copy_to_user(argp, &events, sizeof(events)))
1213                         return -EFAULT;
1214
1215                 return 0;
1216         }
1217         case KVM_SET_VCPU_EVENTS: {
1218                 struct kvm_vcpu_events events;
1219
1220                 if (copy_from_user(&events, argp, sizeof(events)))
1221                         return -EFAULT;
1222
1223                 return kvm_arm_vcpu_set_events(vcpu, &events);
1224         }
1225         case KVM_ARM_VCPU_FINALIZE: {
1226                 int what;
1227
1228                 if (!kvm_vcpu_initialized(vcpu))
1229                         return -ENOEXEC;
1230
1231                 if (get_user(what, (const int __user *)argp))
1232                         return -EFAULT;
1233
1234                 return kvm_arm_vcpu_finalize(vcpu, what);
1235         }
1236         default:
1237                 r = -EINVAL;
1238         }
1239
1240         return r;
1241 }
1242
1243 /**
1244  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1245  * @kvm: kvm instance
1246  * @log: slot id and address to which we copy the log
1247  *
1248  * Steps 1-4 below provide general overview of dirty page logging. See
1249  * kvm_get_dirty_log_protect() function description for additional details.
1250  *
1251  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1252  * always flush the TLB (step 4) even if previous step failed  and the dirty
1253  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1254  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1255  * writes will be marked dirty for next log read.
1256  *
1257  *   1. Take a snapshot of the bit and clear it if needed.
1258  *   2. Write protect the corresponding page.
1259  *   3. Copy the snapshot to the userspace.
1260  *   4. Flush TLB's if needed.
1261  */
1262 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1263 {
1264         bool flush = false;
1265         int r;
1266
1267         mutex_lock(&kvm->slots_lock);
1268
1269         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1270
1271         if (flush)
1272                 kvm_flush_remote_tlbs(kvm);
1273
1274         mutex_unlock(&kvm->slots_lock);
1275         return r;
1276 }
1277
1278 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1279 {
1280         bool flush = false;
1281         int r;
1282
1283         mutex_lock(&kvm->slots_lock);
1284
1285         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1286
1287         if (flush)
1288                 kvm_flush_remote_tlbs(kvm);
1289
1290         mutex_unlock(&kvm->slots_lock);
1291         return r;
1292 }
1293
1294 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1295                                         struct kvm_arm_device_addr *dev_addr)
1296 {
1297         unsigned long dev_id, type;
1298
1299         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1300                 KVM_ARM_DEVICE_ID_SHIFT;
1301         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1302                 KVM_ARM_DEVICE_TYPE_SHIFT;
1303
1304         switch (dev_id) {
1305         case KVM_ARM_DEVICE_VGIC_V2:
1306                 if (!vgic_present)
1307                         return -ENXIO;
1308                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1309         default:
1310                 return -ENODEV;
1311         }
1312 }
1313
1314 long kvm_arch_vm_ioctl(struct file *filp,
1315                        unsigned int ioctl, unsigned long arg)
1316 {
1317         struct kvm *kvm = filp->private_data;
1318         void __user *argp = (void __user *)arg;
1319
1320         switch (ioctl) {
1321         case KVM_CREATE_IRQCHIP: {
1322                 int ret;
1323                 if (!vgic_present)
1324                         return -ENXIO;
1325                 mutex_lock(&kvm->lock);
1326                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1327                 mutex_unlock(&kvm->lock);
1328                 return ret;
1329         }
1330         case KVM_ARM_SET_DEVICE_ADDR: {
1331                 struct kvm_arm_device_addr dev_addr;
1332
1333                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1334                         return -EFAULT;
1335                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1336         }
1337         case KVM_ARM_PREFERRED_TARGET: {
1338                 int err;
1339                 struct kvm_vcpu_init init;
1340
1341                 err = kvm_vcpu_preferred_target(&init);
1342                 if (err)
1343                         return err;
1344
1345                 if (copy_to_user(argp, &init, sizeof(init)))
1346                         return -EFAULT;
1347
1348                 return 0;
1349         }
1350         default:
1351                 return -EINVAL;
1352         }
1353 }
1354
1355 static void cpu_init_hyp_mode(void *dummy)
1356 {
1357         phys_addr_t pgd_ptr;
1358         unsigned long hyp_stack_ptr;
1359         unsigned long stack_page;
1360         unsigned long vector_ptr;
1361
1362         /* Switch from the HYP stub to our own HYP init vector */
1363         __hyp_set_vectors(kvm_get_idmap_vector());
1364
1365         pgd_ptr = kvm_mmu_get_httbr();
1366         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1367         hyp_stack_ptr = stack_page + PAGE_SIZE;
1368         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1369
1370         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1371         __cpu_init_stage2();
1372 }
1373
1374 static void cpu_hyp_reset(void)
1375 {
1376         if (!is_kernel_in_hyp_mode())
1377                 __hyp_reset_vectors();
1378 }
1379
1380 static void cpu_hyp_reinit(void)
1381 {
1382         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1383
1384         cpu_hyp_reset();
1385
1386         if (is_kernel_in_hyp_mode())
1387                 kvm_timer_init_vhe();
1388         else
1389                 cpu_init_hyp_mode(NULL);
1390
1391         kvm_arm_init_debug();
1392
1393         if (vgic_present)
1394                 kvm_vgic_init_cpu_hardware();
1395 }
1396
1397 static void _kvm_arch_hardware_enable(void *discard)
1398 {
1399         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1400                 cpu_hyp_reinit();
1401                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1402         }
1403 }
1404
1405 int kvm_arch_hardware_enable(void)
1406 {
1407         _kvm_arch_hardware_enable(NULL);
1408         return 0;
1409 }
1410
1411 static void _kvm_arch_hardware_disable(void *discard)
1412 {
1413         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1414                 cpu_hyp_reset();
1415                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1416         }
1417 }
1418
1419 void kvm_arch_hardware_disable(void)
1420 {
1421         _kvm_arch_hardware_disable(NULL);
1422 }
1423
1424 #ifdef CONFIG_CPU_PM
1425 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1426                                     unsigned long cmd,
1427                                     void *v)
1428 {
1429         /*
1430          * kvm_arm_hardware_enabled is left with its old value over
1431          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1432          * re-enable hyp.
1433          */
1434         switch (cmd) {
1435         case CPU_PM_ENTER:
1436                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1437                         /*
1438                          * don't update kvm_arm_hardware_enabled here
1439                          * so that the hardware will be re-enabled
1440                          * when we resume. See below.
1441                          */
1442                         cpu_hyp_reset();
1443
1444                 return NOTIFY_OK;
1445         case CPU_PM_ENTER_FAILED:
1446         case CPU_PM_EXIT:
1447                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1448                         /* The hardware was enabled before suspend. */
1449                         cpu_hyp_reinit();
1450
1451                 return NOTIFY_OK;
1452
1453         default:
1454                 return NOTIFY_DONE;
1455         }
1456 }
1457
1458 static struct notifier_block hyp_init_cpu_pm_nb = {
1459         .notifier_call = hyp_init_cpu_pm_notifier,
1460 };
1461
1462 static void __init hyp_cpu_pm_init(void)
1463 {
1464         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1465 }
1466 static void __init hyp_cpu_pm_exit(void)
1467 {
1468         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1469 }
1470 #else
1471 static inline void hyp_cpu_pm_init(void)
1472 {
1473 }
1474 static inline void hyp_cpu_pm_exit(void)
1475 {
1476 }
1477 #endif
1478
1479 static int init_common_resources(void)
1480 {
1481         kvm_set_ipa_limit();
1482
1483         return 0;
1484 }
1485
1486 static int init_subsystems(void)
1487 {
1488         int err = 0;
1489
1490         /*
1491          * Enable hardware so that subsystem initialisation can access EL2.
1492          */
1493         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1494
1495         /*
1496          * Register CPU lower-power notifier
1497          */
1498         hyp_cpu_pm_init();
1499
1500         /*
1501          * Init HYP view of VGIC
1502          */
1503         err = kvm_vgic_hyp_init();
1504         switch (err) {
1505         case 0:
1506                 vgic_present = true;
1507                 break;
1508         case -ENODEV:
1509         case -ENXIO:
1510                 vgic_present = false;
1511                 err = 0;
1512                 break;
1513         default:
1514                 goto out;
1515         }
1516
1517         /*
1518          * Init HYP architected timer support
1519          */
1520         err = kvm_timer_hyp_init(vgic_present);
1521         if (err)
1522                 goto out;
1523
1524         kvm_perf_init();
1525         kvm_coproc_table_init();
1526
1527 out:
1528         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1529
1530         return err;
1531 }
1532
1533 static void teardown_hyp_mode(void)
1534 {
1535         int cpu;
1536
1537         free_hyp_pgds();
1538         for_each_possible_cpu(cpu)
1539                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1540         hyp_cpu_pm_exit();
1541 }
1542
1543 /**
1544  * Inits Hyp-mode on all online CPUs
1545  */
1546 static int init_hyp_mode(void)
1547 {
1548         int cpu;
1549         int err = 0;
1550
1551         /*
1552          * Allocate Hyp PGD and setup Hyp identity mapping
1553          */
1554         err = kvm_mmu_init();
1555         if (err)
1556                 goto out_err;
1557
1558         /*
1559          * Allocate stack pages for Hypervisor-mode
1560          */
1561         for_each_possible_cpu(cpu) {
1562                 unsigned long stack_page;
1563
1564                 stack_page = __get_free_page(GFP_KERNEL);
1565                 if (!stack_page) {
1566                         err = -ENOMEM;
1567                         goto out_err;
1568                 }
1569
1570                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1571         }
1572
1573         /*
1574          * Map the Hyp-code called directly from the host
1575          */
1576         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1577                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1578         if (err) {
1579                 kvm_err("Cannot map world-switch code\n");
1580                 goto out_err;
1581         }
1582
1583         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1584                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1585         if (err) {
1586                 kvm_err("Cannot map rodata section\n");
1587                 goto out_err;
1588         }
1589
1590         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1591                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1592         if (err) {
1593                 kvm_err("Cannot map bss section\n");
1594                 goto out_err;
1595         }
1596
1597         err = kvm_map_vectors();
1598         if (err) {
1599                 kvm_err("Cannot map vectors\n");
1600                 goto out_err;
1601         }
1602
1603         /*
1604          * Map the Hyp stack pages
1605          */
1606         for_each_possible_cpu(cpu) {
1607                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1608                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1609                                           PAGE_HYP);
1610
1611                 if (err) {
1612                         kvm_err("Cannot map hyp stack\n");
1613                         goto out_err;
1614                 }
1615         }
1616
1617         for_each_possible_cpu(cpu) {
1618                 kvm_host_data_t *cpu_data;
1619
1620                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1621                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1622
1623                 if (err) {
1624                         kvm_err("Cannot map host CPU state: %d\n", err);
1625                         goto out_err;
1626                 }
1627         }
1628
1629         err = hyp_map_aux_data();
1630         if (err)
1631                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1632
1633         return 0;
1634
1635 out_err:
1636         teardown_hyp_mode();
1637         kvm_err("error initializing Hyp mode: %d\n", err);
1638         return err;
1639 }
1640
1641 static void check_kvm_target_cpu(void *ret)
1642 {
1643         *(int *)ret = kvm_target_cpu();
1644 }
1645
1646 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1647 {
1648         struct kvm_vcpu *vcpu;
1649         int i;
1650
1651         mpidr &= MPIDR_HWID_BITMASK;
1652         kvm_for_each_vcpu(i, vcpu, kvm) {
1653                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1654                         return vcpu;
1655         }
1656         return NULL;
1657 }
1658
1659 bool kvm_arch_has_irq_bypass(void)
1660 {
1661         return true;
1662 }
1663
1664 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1665                                       struct irq_bypass_producer *prod)
1666 {
1667         struct kvm_kernel_irqfd *irqfd =
1668                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1669
1670         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1671                                           &irqfd->irq_entry);
1672 }
1673 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1674                                       struct irq_bypass_producer *prod)
1675 {
1676         struct kvm_kernel_irqfd *irqfd =
1677                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1678
1679         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1680                                      &irqfd->irq_entry);
1681 }
1682
1683 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1684 {
1685         struct kvm_kernel_irqfd *irqfd =
1686                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1687
1688         kvm_arm_halt_guest(irqfd->kvm);
1689 }
1690
1691 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1692 {
1693         struct kvm_kernel_irqfd *irqfd =
1694                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1695
1696         kvm_arm_resume_guest(irqfd->kvm);
1697 }
1698
1699 /**
1700  * Initialize Hyp-mode and memory mappings on all CPUs.
1701  */
1702 int kvm_arch_init(void *opaque)
1703 {
1704         int err;
1705         int ret, cpu;
1706         bool in_hyp_mode;
1707
1708         if (!is_hyp_mode_available()) {
1709                 kvm_info("HYP mode not available\n");
1710                 return -ENODEV;
1711         }
1712
1713         in_hyp_mode = is_kernel_in_hyp_mode();
1714
1715         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1716                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1717                 return -ENODEV;
1718         }
1719
1720         for_each_online_cpu(cpu) {
1721                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1722                 if (ret < 0) {
1723                         kvm_err("Error, CPU %d not supported!\n", cpu);
1724                         return -ENODEV;
1725                 }
1726         }
1727
1728         err = init_common_resources();
1729         if (err)
1730                 return err;
1731
1732         err = kvm_arm_init_sve();
1733         if (err)
1734                 return err;
1735
1736         if (!in_hyp_mode) {
1737                 err = init_hyp_mode();
1738                 if (err)
1739                         goto out_err;
1740         }
1741
1742         err = init_subsystems();
1743         if (err)
1744                 goto out_hyp;
1745
1746         if (in_hyp_mode)
1747                 kvm_info("VHE mode initialized successfully\n");
1748         else
1749                 kvm_info("Hyp mode initialized successfully\n");
1750
1751         return 0;
1752
1753 out_hyp:
1754         if (!in_hyp_mode)
1755                 teardown_hyp_mode();
1756 out_err:
1757         return err;
1758 }
1759
1760 /* NOP: Compiling as a module not supported */
1761 void kvm_arch_exit(void)
1762 {
1763         kvm_perf_teardown();
1764 }
1765
1766 static int arm_init(void)
1767 {
1768         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1769         return rc;
1770 }
1771
1772 module_init(arm_init);