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