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 lastest 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         preempt_disable();
 330         kvm_vgic_vmcr_sync(vcpu);
 331         preempt_enable();
 332
 333         kvm_vgic_v4_enable_doorbell(vcpu);
 334 }
 335
 336 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 337 {
 338         kvm_vgic_v4_disable_doorbell(vcpu);
 339 }
 340
 341 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 342 {
 343         /* Force users to call KVM_ARM_VCPU_INIT */
 344         vcpu->arch.target = -1;
 345         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 346
 347         /* Set up the timer */
 348         kvm_timer_vcpu_init(vcpu);
 349
 350         kvm_pmu_vcpu_init(vcpu);
 351
 352         kvm_arm_reset_debug_ptr(vcpu);
 353
 354         return kvm_vgic_vcpu_init(vcpu);
 355 }
 356
 357 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 358 {
 359         int *last_ran;
 360         kvm_host_data_t *cpu_data;
 361
 362         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 363         cpu_data = this_cpu_ptr(&kvm_host_data);
 364
 365         /*
 366          * We might get preempted before the vCPU actually runs, but
 367          * over-invalidation doesn't affect correctness.
 368          */
 369         if (*last_ran != vcpu->vcpu_id) {
 370                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 371                 *last_ran = vcpu->vcpu_id;
 372         }
 373
 374         vcpu->cpu = cpu;
 375         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 376
 377         kvm_arm_set_running_vcpu(vcpu);
 378         kvm_vgic_load(vcpu);
 379         kvm_timer_vcpu_load(vcpu);
 380         kvm_vcpu_load_sysregs(vcpu);
 381         kvm_arch_vcpu_load_fp(vcpu);
 382         kvm_vcpu_pmu_restore_guest(vcpu);
 383
 384         if (single_task_running())
 385                 vcpu_clear_wfe_traps(vcpu);
 386         else
 387                 vcpu_set_wfe_traps(vcpu);
 388
 389         vcpu_ptrauth_setup_lazy(vcpu);
 390 }
 391
 392 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 393 {
 394         kvm_arch_vcpu_put_fp(vcpu);
 395         kvm_vcpu_put_sysregs(vcpu);
 396         kvm_timer_vcpu_put(vcpu);
 397         kvm_vgic_put(vcpu);
 398         kvm_vcpu_pmu_restore_host(vcpu);
 399
 400         vcpu->cpu = -1;
 401
 402         kvm_arm_set_running_vcpu(NULL);
 403 }
 404
 405 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 406 {
 407         vcpu->arch.power_off = true;
 408         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 409         kvm_vcpu_kick(vcpu);
 410 }
 411
 412 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 413                                     struct kvm_mp_state *mp_state)
 414 {
 415         if (vcpu->arch.power_off)
 416                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 417         else
 418                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 419
 420         return 0;
 421 }
 422
 423 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 424                                     struct kvm_mp_state *mp_state)
 425 {
 426         int ret = 0;
 427
 428         switch (mp_state->mp_state) {
 429         case KVM_MP_STATE_RUNNABLE:
 430                 vcpu->arch.power_off = false;
 431                 break;
 432         case KVM_MP_STATE_STOPPED:
 433                 vcpu_power_off(vcpu);
 434                 break;
 435         default:
 436                 ret = -EINVAL;
 437         }
 438
 439         return ret;
 440 }
 441
 442 /**
 443  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 444  * @v:          The VCPU pointer
 445  *
 446  * If the guest CPU is not waiting for interrupts or an interrupt line is
 447  * asserted, the CPU is by definition runnable.
 448  */
 449 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 450 {
 451         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 452         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 453                 && !v->arch.power_off && !v->arch.pause);
 454 }
 455
 456 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 457 {
 458         return vcpu_mode_priv(vcpu);
 459 }
 460
 461 /* Just ensure a guest exit from a particular CPU */
 462 static void exit_vm_noop(void *info)
 463 {
 464 }
 465
 466 void force_vm_exit(const cpumask_t *mask)
 467 {
 468         preempt_disable();
 469         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 470         preempt_enable();
 471 }
 472
 473 /**
 474  * need_new_vmid_gen - check that the VMID is still valid
 475  * @vmid: The VMID to check
 476  *
 477  * return true if there is a new generation of VMIDs being used
 478  *
 479  * The hardware supports a limited set of values with the value zero reserved
 480  * for the host, so we check if an assigned value belongs to a previous
 481  * generation, which which requires us to assign a new value. If we're the
 482  * first to use a VMID for the new generation, we must flush necessary caches
 483  * and TLBs on all CPUs.
 484  */
 485 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 486 {
 487         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 488         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 489         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 490 }
 491
 492 /**
 493  * update_vmid - Update the vmid with a valid VMID for the current generation
 494  * @kvm: The guest that struct vmid belongs to
 495  * @vmid: The stage-2 VMID information struct
 496  */
 497 static void update_vmid(struct kvm_vmid *vmid)
 498 {
 499         if (!need_new_vmid_gen(vmid))
 500                 return;
 501
 502         spin_lock(&kvm_vmid_lock);
 503
 504         /*
 505          * We need to re-check the vmid_gen here to ensure that if another vcpu
 506          * already allocated a valid vmid for this vm, then this vcpu should
 507          * use the same vmid.
 508          */
 509         if (!need_new_vmid_gen(vmid)) {
 510                 spin_unlock(&kvm_vmid_lock);
 511                 return;
 512         }
 513
 514         /* First user of a new VMID generation? */
 515         if (unlikely(kvm_next_vmid == 0)) {
 516                 atomic64_inc(&kvm_vmid_gen);
 517                 kvm_next_vmid = 1;
 518
 519                 /*
 520                  * On SMP we know no other CPUs can use this CPU's or each
 521                  * other's VMID after force_vm_exit returns since the
 522                  * kvm_vmid_lock blocks them from reentry to the guest.
 523                  */
 524                 force_vm_exit(cpu_all_mask);
 525                 /*
 526                  * Now broadcast TLB + ICACHE invalidation over the inner
 527                  * shareable domain to make sure all data structures are
 528                  * clean.
 529                  */
 530                 kvm_call_hyp(__kvm_flush_vm_context);
 531         }
 532
 533         vmid->vmid = kvm_next_vmid;
 534         kvm_next_vmid++;
 535         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 536
 537         smp_wmb();
 538         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 539
 540         spin_unlock(&kvm_vmid_lock);
 541 }
 542
 543 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 544 {
 545         struct kvm *kvm = vcpu->kvm;
 546         int ret = 0;
 547
 548         if (likely(vcpu->arch.has_run_once))
 549                 return 0;
 550
 551         if (!kvm_arm_vcpu_is_finalized(vcpu))
 552                 return -EPERM;
 553
 554         vcpu->arch.has_run_once = true;
 555
 556         if (likely(irqchip_in_kernel(kvm))) {
 557                 /*
 558                  * Map the VGIC hardware resources before running a vcpu the
 559                  * first time on this VM.
 560                  */
 561                 if (unlikely(!vgic_ready(kvm))) {
 562                         ret = kvm_vgic_map_resources(kvm);
 563                         if (ret)
 564                                 return ret;
 565                 }
 566         } else {
 567                 /*
 568                  * Tell the rest of the code that there are userspace irqchip
 569                  * VMs in the wild.
 570                  */
 571                 static_branch_inc(&userspace_irqchip_in_use);
 572         }
 573
 574         ret = kvm_timer_enable(vcpu);
 575         if (ret)
 576                 return ret;
 577
 578         ret = kvm_arm_pmu_v3_enable(vcpu);
 579
 580         return ret;
 581 }
 582
 583 bool kvm_arch_intc_initialized(struct kvm *kvm)
 584 {
 585         return vgic_initialized(kvm);
 586 }
 587
 588 void kvm_arm_halt_guest(struct kvm *kvm)
 589 {
 590         int i;
 591         struct kvm_vcpu *vcpu;
 592
 593         kvm_for_each_vcpu(i, vcpu, kvm)
 594                 vcpu->arch.pause = true;
 595         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 596 }
 597
 598 void kvm_arm_resume_guest(struct kvm *kvm)
 599 {
 600         int i;
 601         struct kvm_vcpu *vcpu;
 602
 603         kvm_for_each_vcpu(i, vcpu, kvm) {
 604                 vcpu->arch.pause = false;
 605                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 606         }
 607 }
 608
 609 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 610 {
 611         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 612
 613         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 614                                        (!vcpu->arch.pause)));
 615
 616         if (vcpu->arch.power_off || vcpu->arch.pause) {
 617                 /* Awaken to handle a signal, request we sleep again later. */
 618                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 619         }
 620
 621         /*
 622          * Make sure we will observe a potential reset request if we've
 623          * observed a change to the power state. Pairs with the smp_wmb() in
 624          * kvm_psci_vcpu_on().
 625          */
 626         smp_rmb();
 627 }
 628
 629 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 630 {
 631         return vcpu->arch.target >= 0;
 632 }
 633
 634 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 635 {
 636         if (kvm_request_pending(vcpu)) {
 637                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 638                         vcpu_req_sleep(vcpu);
 639
 640                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 641                         kvm_reset_vcpu(vcpu);
 642
 643                 /*
 644                  * Clear IRQ_PENDING requests that were made to guarantee
 645                  * that a VCPU sees new virtual interrupts.
 646                  */
 647                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 648         }
 649 }
 650
 651 /**
 652  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 653  * @vcpu:       The VCPU pointer
 654  * @run:        The kvm_run structure pointer used for userspace state exchange
 655  *
 656  * This function is called through the VCPU_RUN ioctl called from user space. It
 657  * will execute VM code in a loop until the time slice for the process is used
 658  * or some emulation is needed from user space in which case the function will
 659  * return with return value 0 and with the kvm_run structure filled in with the
 660  * required data for the requested emulation.
 661  */
 662 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 663 {
 664         int ret;
 665
 666         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 667                 return -ENOEXEC;
 668
 669         ret = kvm_vcpu_first_run_init(vcpu);
 670         if (ret)
 671                 return ret;
 672
 673         if (run->exit_reason == KVM_EXIT_MMIO) {
 674                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 675                 if (ret)
 676                         return ret;
 677         }
 678
 679         if (run->immediate_exit)
 680                 return -EINTR;
 681
 682         vcpu_load(vcpu);
 683
 684         kvm_sigset_activate(vcpu);
 685
 686         ret = 1;
 687         run->exit_reason = KVM_EXIT_UNKNOWN;
 688         while (ret > 0) {
 689                 /*
 690                  * Check conditions before entering the guest
 691                  */
 692                 cond_resched();
 693
 694                 update_vmid(&vcpu->kvm->arch.vmid);
 695
 696                 check_vcpu_requests(vcpu);
 697
 698                 /*
 699                  * Preparing the interrupts to be injected also
 700                  * involves poking the GIC, which must be done in a
 701                  * non-preemptible context.
 702                  */
 703                 preempt_disable();
 704
 705                 kvm_pmu_flush_hwstate(vcpu);
 706
 707                 local_irq_disable();
 708
 709                 kvm_vgic_flush_hwstate(vcpu);
 710
 711                 /*
 712                  * Exit if we have a signal pending so that we can deliver the
 713                  * signal to user space.
 714                  */
 715                 if (signal_pending(current)) {
 716                         ret = -EINTR;
 717                         run->exit_reason = KVM_EXIT_INTR;
 718                 }
 719
 720                 /*
 721                  * If we're using a userspace irqchip, then check if we need
 722                  * to tell a userspace irqchip about timer or PMU level
 723                  * changes and if so, exit to userspace (the actual level
 724                  * state gets updated in kvm_timer_update_run and
 725                  * kvm_pmu_update_run below).
 726                  */
 727                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 728                         if (kvm_timer_should_notify_user(vcpu) ||
 729                             kvm_pmu_should_notify_user(vcpu)) {
 730                                 ret = -EINTR;
 731                                 run->exit_reason = KVM_EXIT_INTR;
 732                         }
 733                 }
 734
 735                 /*
 736                  * Ensure we set mode to IN_GUEST_MODE after we disable
 737                  * interrupts and before the final VCPU requests check.
 738                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 739                  * Documentation/virt/kvm/vcpu-requests.rst
 740                  */
 741                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 742
 743                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 744                     kvm_request_pending(vcpu)) {
 745                         vcpu->mode = OUTSIDE_GUEST_MODE;
 746                         isb(); /* Ensure work in x_flush_hwstate is committed */
 747                         kvm_pmu_sync_hwstate(vcpu);
 748                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 749                                 kvm_timer_sync_hwstate(vcpu);
 750                         kvm_vgic_sync_hwstate(vcpu);
 751                         local_irq_enable();
 752                         preempt_enable();
 753                         continue;
 754                 }
 755
 756                 kvm_arm_setup_debug(vcpu);
 757
 758                 /**************************************************************
 759                  * Enter the guest
 760                  */
 761                 trace_kvm_entry(*vcpu_pc(vcpu));
 762                 guest_enter_irqoff();
 763
 764                 if (has_vhe()) {
 765                         kvm_arm_vhe_guest_enter();
 766                         ret = kvm_vcpu_run_vhe(vcpu);
 767                         kvm_arm_vhe_guest_exit();
 768                 } else {
 769                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 770                 }
 771
 772                 vcpu->mode = OUTSIDE_GUEST_MODE;
 773                 vcpu->stat.exits++;
 774                 /*
 775                  * Back from guest
 776                  *************************************************************/
 777
 778                 kvm_arm_clear_debug(vcpu);
 779
 780                 /*
 781                  * We must sync the PMU state before the vgic state so
 782                  * that the vgic can properly sample the updated state of the
 783                  * interrupt line.
 784                  */
 785                 kvm_pmu_sync_hwstate(vcpu);
 786
 787                 /*
 788                  * Sync the vgic state before syncing the timer state because
 789                  * the timer code needs to know if the virtual timer
 790                  * interrupts are active.
 791                  */
 792                 kvm_vgic_sync_hwstate(vcpu);
 793
 794                 /*
 795                  * Sync the timer hardware state before enabling interrupts as
 796                  * we don't want vtimer interrupts to race with syncing the
 797                  * timer virtual interrupt state.
 798                  */
 799                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 800                         kvm_timer_sync_hwstate(vcpu);
 801
 802                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 803
 804                 /*
 805                  * We may have taken a host interrupt in HYP mode (ie
 806                  * while executing the guest). This interrupt is still
 807                  * pending, as we haven't serviced it yet!
 808                  *
 809                  * We're now back in SVC mode, with interrupts
 810                  * disabled.  Enabling the interrupts now will have
 811                  * the effect of taking the interrupt again, in SVC
 812                  * mode this time.
 813                  */
 814                 local_irq_enable();
 815
 816                 /*
 817                  * We do local_irq_enable() before calling guest_exit() so
 818                  * that if a timer interrupt hits while running the guest we
 819                  * account that tick as being spent in the guest.  We enable
 820                  * preemption after calling guest_exit() so that if we get
 821                  * preempted we make sure ticks after that is not counted as
 822                  * guest time.
 823                  */
 824                 guest_exit();
 825                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 826
 827                 /* Exit types that need handling before we can be preempted */
 828                 handle_exit_early(vcpu, run, ret);
 829
 830                 preempt_enable();
 831
 832                 ret = handle_exit(vcpu, run, ret);
 833         }
 834
 835         /* Tell userspace about in-kernel device output levels */
 836         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 837                 kvm_timer_update_run(vcpu);
 838                 kvm_pmu_update_run(vcpu);
 839         }
 840
 841         kvm_sigset_deactivate(vcpu);
 842
 843         vcpu_put(vcpu);
 844         return ret;
 845 }
 846
 847 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 848 {
 849         int bit_index;
 850         bool set;
 851         unsigned long *hcr;
 852
 853         if (number == KVM_ARM_IRQ_CPU_IRQ)
 854                 bit_index = __ffs(HCR_VI);
 855         else /* KVM_ARM_IRQ_CPU_FIQ */
 856                 bit_index = __ffs(HCR_VF);
 857
 858         hcr = vcpu_hcr(vcpu);
 859         if (level)
 860                 set = test_and_set_bit(bit_index, hcr);
 861         else
 862                 set = test_and_clear_bit(bit_index, hcr);
 863
 864         /*
 865          * If we didn't change anything, no need to wake up or kick other CPUs
 866          */
 867         if (set == level)
 868                 return 0;
 869
 870         /*
 871          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 872          * trigger a world-switch round on the running physical CPU to set the
 873          * virtual IRQ/FIQ fields in the HCR appropriately.
 874          */
 875         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 876         kvm_vcpu_kick(vcpu);
 877
 878         return 0;
 879 }
 880
 881 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 882                           bool line_status)
 883 {
 884         u32 irq = irq_level->irq;
 885         unsigned int irq_type, vcpu_idx, irq_num;
 886         int nrcpus = atomic_read(&kvm->online_vcpus);
 887         struct kvm_vcpu *vcpu = NULL;
 888         bool level = irq_level->level;
 889
 890         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 891         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 892         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 893         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 894
 895         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 896
 897         switch (irq_type) {
 898         case KVM_ARM_IRQ_TYPE_CPU:
 899                 if (irqchip_in_kernel(kvm))
 900                         return -ENXIO;
 901
 902                 if (vcpu_idx >= nrcpus)
 903                         return -EINVAL;
 904
 905                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 906                 if (!vcpu)
 907                         return -EINVAL;
 908
 909                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 910                         return -EINVAL;
 911
 912                 return vcpu_interrupt_line(vcpu, irq_num, level);
 913         case KVM_ARM_IRQ_TYPE_PPI:
 914                 if (!irqchip_in_kernel(kvm))
 915                         return -ENXIO;
 916
 917                 if (vcpu_idx >= nrcpus)
 918                         return -EINVAL;
 919
 920                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 921                 if (!vcpu)
 922                         return -EINVAL;
 923
 924                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 925                         return -EINVAL;
 926
 927                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 928         case KVM_ARM_IRQ_TYPE_SPI:
 929                 if (!irqchip_in_kernel(kvm))
 930                         return -ENXIO;
 931
 932                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 933                         return -EINVAL;
 934
 935                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 936         }
 937
 938         return -EINVAL;
 939 }
 940
 941 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 942                                const struct kvm_vcpu_init *init)
 943 {
 944         unsigned int i, ret;
 945         int phys_target = kvm_target_cpu();
 946
 947         if (init->target != phys_target)
 948                 return -EINVAL;
 949
 950         /*
 951          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 952          * use the same target.
 953          */
 954         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 955                 return -EINVAL;
 956
 957         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 958         for (i = 0; i < sizeof(init->features) * 8; i++) {
 959                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 960
 961                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 962                         return -ENOENT;
 963
 964                 /*
 965                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 966                  * use the same feature set.
 967                  */
 968                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 969                     test_bit(i, vcpu->arch.features) != set)
 970                         return -EINVAL;
 971
 972                 if (set)
 973                         set_bit(i, vcpu->arch.features);
 974         }
 975
 976         vcpu->arch.target = phys_target;
 977
 978         /* Now we know what it is, we can reset it. */
 979         ret = kvm_reset_vcpu(vcpu);
 980         if (ret) {
 981                 vcpu->arch.target = -1;
 982                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 983         }
 984
 985         return ret;
 986 }
 987
 988 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 989                                          struct kvm_vcpu_init *init)
 990 {
 991         int ret;
 992
 993         ret = kvm_vcpu_set_target(vcpu, init);
 994         if (ret)
 995                 return ret;
 996
 997         /*
 998          * Ensure a rebooted VM will fault in RAM pages and detect if the
 999          * guest MMU is turned off and flush the caches as needed.
1000          */
1001         if (vcpu->arch.has_run_once)
1002                 stage2_unmap_vm(vcpu->kvm);
1003
1004         vcpu_reset_hcr(vcpu);
1005
1006         /*
1007          * Handle the "start in power-off" case.
1008          */
1009         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1010                 vcpu_power_off(vcpu);
1011         else
1012                 vcpu->arch.power_off = false;
1013
1014         return 0;
1015 }
1016
1017 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1018                                  struct kvm_device_attr *attr)
1019 {
1020         int ret = -ENXIO;
1021
1022         switch (attr->group) {
1023         default:
1024                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1025                 break;
1026         }
1027
1028         return ret;
1029 }
1030
1031 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1032                                  struct kvm_device_attr *attr)
1033 {
1034         int ret = -ENXIO;
1035
1036         switch (attr->group) {
1037         default:
1038                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1039                 break;
1040         }
1041
1042         return ret;
1043 }
1044
1045 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1046                                  struct kvm_device_attr *attr)
1047 {
1048         int ret = -ENXIO;
1049
1050         switch (attr->group) {
1051         default:
1052                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1053                 break;
1054         }
1055
1056         return ret;
1057 }
1058
1059 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1060                                    struct kvm_vcpu_events *events)
1061 {
1062         memset(events, 0, sizeof(*events));
1063
1064         return __kvm_arm_vcpu_get_events(vcpu, events);
1065 }
1066
1067 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1068                                    struct kvm_vcpu_events *events)
1069 {
1070         int i;
1071
1072         /* check whether the reserved field is zero */
1073         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1074                 if (events->reserved[i])
1075                         return -EINVAL;
1076
1077         /* check whether the pad field is zero */
1078         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1079                 if (events->exception.pad[i])
1080                         return -EINVAL;
1081
1082         return __kvm_arm_vcpu_set_events(vcpu, events);
1083 }
1084
1085 long kvm_arch_vcpu_ioctl(struct file *filp,
1086                          unsigned int ioctl, unsigned long arg)
1087 {
1088         struct kvm_vcpu *vcpu = filp->private_data;
1089         void __user *argp = (void __user *)arg;
1090         struct kvm_device_attr attr;
1091         long r;
1092
1093         switch (ioctl) {
1094         case KVM_ARM_VCPU_INIT: {
1095                 struct kvm_vcpu_init init;
1096
1097                 r = -EFAULT;
1098                 if (copy_from_user(&init, argp, sizeof(init)))
1099                         break;
1100
1101                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1102                 break;
1103         }
1104         case KVM_SET_ONE_REG:
1105         case KVM_GET_ONE_REG: {
1106                 struct kvm_one_reg reg;
1107
1108                 r = -ENOEXEC;
1109                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1110                         break;
1111
1112                 r = -EFAULT;
1113                 if (copy_from_user(&reg, argp, sizeof(reg)))
1114                         break;
1115
1116                 if (ioctl == KVM_SET_ONE_REG)
1117                         r = kvm_arm_set_reg(vcpu, &reg);
1118                 else
1119                         r = kvm_arm_get_reg(vcpu, &reg);
1120                 break;
1121         }
1122         case KVM_GET_REG_LIST: {
1123                 struct kvm_reg_list __user *user_list = argp;
1124                 struct kvm_reg_list reg_list;
1125                 unsigned n;
1126
1127                 r = -ENOEXEC;
1128                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1129                         break;
1130
1131                 r = -EPERM;
1132                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1133                         break;
1134
1135                 r = -EFAULT;
1136                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1137                         break;
1138                 n = reg_list.n;
1139                 reg_list.n = kvm_arm_num_regs(vcpu);
1140                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1141                         break;
1142                 r = -E2BIG;
1143                 if (n < reg_list.n)
1144                         break;
1145                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1146                 break;
1147         }
1148         case KVM_SET_DEVICE_ATTR: {
1149                 r = -EFAULT;
1150                 if (copy_from_user(&attr, argp, sizeof(attr)))
1151                         break;
1152                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1153                 break;
1154         }
1155         case KVM_GET_DEVICE_ATTR: {
1156                 r = -EFAULT;
1157                 if (copy_from_user(&attr, argp, sizeof(attr)))
1158                         break;
1159                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1160                 break;
1161         }
1162         case KVM_HAS_DEVICE_ATTR: {
1163                 r = -EFAULT;
1164                 if (copy_from_user(&attr, argp, sizeof(attr)))
1165                         break;
1166                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1167                 break;
1168         }
1169         case KVM_GET_VCPU_EVENTS: {
1170                 struct kvm_vcpu_events events;
1171
1172                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1173                         return -EINVAL;
1174
1175                 if (copy_to_user(argp, &events, sizeof(events)))
1176                         return -EFAULT;
1177
1178                 return 0;
1179         }
1180         case KVM_SET_VCPU_EVENTS: {
1181                 struct kvm_vcpu_events events;
1182
1183                 if (copy_from_user(&events, argp, sizeof(events)))
1184                         return -EFAULT;
1185
1186                 return kvm_arm_vcpu_set_events(vcpu, &events);
1187         }
1188         case KVM_ARM_VCPU_FINALIZE: {
1189                 int what;
1190
1191                 if (!kvm_vcpu_initialized(vcpu))
1192                         return -ENOEXEC;
1193
1194                 if (get_user(what, (const int __user *)argp))
1195                         return -EFAULT;
1196
1197                 return kvm_arm_vcpu_finalize(vcpu, what);
1198         }
1199         default:
1200                 r = -EINVAL;
1201         }
1202
1203         return r;
1204 }
1205
1206 /**
1207  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1208  * @kvm: kvm instance
1209  * @log: slot id and address to which we copy the log
1210  *
1211  * Steps 1-4 below provide general overview of dirty page logging. See
1212  * kvm_get_dirty_log_protect() function description for additional details.
1213  *
1214  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1215  * always flush the TLB (step 4) even if previous step failed  and the dirty
1216  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1217  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1218  * writes will be marked dirty for next log read.
1219  *
1220  *   1. Take a snapshot of the bit and clear it if needed.
1221  *   2. Write protect the corresponding page.
1222  *   3. Copy the snapshot to the userspace.
1223  *   4. Flush TLB's if needed.
1224  */
1225 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1226 {
1227         bool flush = false;
1228         int r;
1229
1230         mutex_lock(&kvm->slots_lock);
1231
1232         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1233
1234         if (flush)
1235                 kvm_flush_remote_tlbs(kvm);
1236
1237         mutex_unlock(&kvm->slots_lock);
1238         return r;
1239 }
1240
1241 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1242 {
1243         bool flush = false;
1244         int r;
1245
1246         mutex_lock(&kvm->slots_lock);
1247
1248         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1249
1250         if (flush)
1251                 kvm_flush_remote_tlbs(kvm);
1252
1253         mutex_unlock(&kvm->slots_lock);
1254         return r;
1255 }
1256
1257 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1258                                         struct kvm_arm_device_addr *dev_addr)
1259 {
1260         unsigned long dev_id, type;
1261
1262         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1263                 KVM_ARM_DEVICE_ID_SHIFT;
1264         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1265                 KVM_ARM_DEVICE_TYPE_SHIFT;
1266
1267         switch (dev_id) {
1268         case KVM_ARM_DEVICE_VGIC_V2:
1269                 if (!vgic_present)
1270                         return -ENXIO;
1271                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1272         default:
1273                 return -ENODEV;
1274         }
1275 }
1276
1277 long kvm_arch_vm_ioctl(struct file *filp,
1278                        unsigned int ioctl, unsigned long arg)
1279 {
1280         struct kvm *kvm = filp->private_data;
1281         void __user *argp = (void __user *)arg;
1282
1283         switch (ioctl) {
1284         case KVM_CREATE_IRQCHIP: {
1285                 int ret;
1286                 if (!vgic_present)
1287                         return -ENXIO;
1288                 mutex_lock(&kvm->lock);
1289                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1290                 mutex_unlock(&kvm->lock);
1291                 return ret;
1292         }
1293         case KVM_ARM_SET_DEVICE_ADDR: {
1294                 struct kvm_arm_device_addr dev_addr;
1295
1296                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1297                         return -EFAULT;
1298                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1299         }
1300         case KVM_ARM_PREFERRED_TARGET: {
1301                 int err;
1302                 struct kvm_vcpu_init init;
1303
1304                 err = kvm_vcpu_preferred_target(&init);
1305                 if (err)
1306                         return err;
1307
1308                 if (copy_to_user(argp, &init, sizeof(init)))
1309                         return -EFAULT;
1310
1311                 return 0;
1312         }
1313         default:
1314                 return -EINVAL;
1315         }
1316 }
1317
1318 static void cpu_init_hyp_mode(void *dummy)
1319 {
1320         phys_addr_t pgd_ptr;
1321         unsigned long hyp_stack_ptr;
1322         unsigned long stack_page;
1323         unsigned long vector_ptr;
1324
1325         /* Switch from the HYP stub to our own HYP init vector */
1326         __hyp_set_vectors(kvm_get_idmap_vector());
1327
1328         pgd_ptr = kvm_mmu_get_httbr();
1329         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1330         hyp_stack_ptr = stack_page + PAGE_SIZE;
1331         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1332
1333         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1334         __cpu_init_stage2();
1335 }
1336
1337 static void cpu_hyp_reset(void)
1338 {
1339         if (!is_kernel_in_hyp_mode())
1340                 __hyp_reset_vectors();
1341 }
1342
1343 static void cpu_hyp_reinit(void)
1344 {
1345         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1346
1347         cpu_hyp_reset();
1348
1349         if (is_kernel_in_hyp_mode())
1350                 kvm_timer_init_vhe();
1351         else
1352                 cpu_init_hyp_mode(NULL);
1353
1354         kvm_arm_init_debug();
1355
1356         if (vgic_present)
1357                 kvm_vgic_init_cpu_hardware();
1358 }
1359
1360 static void _kvm_arch_hardware_enable(void *discard)
1361 {
1362         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1363                 cpu_hyp_reinit();
1364                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1365         }
1366 }
1367
1368 int kvm_arch_hardware_enable(void)
1369 {
1370         _kvm_arch_hardware_enable(NULL);
1371         return 0;
1372 }
1373
1374 static void _kvm_arch_hardware_disable(void *discard)
1375 {
1376         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1377                 cpu_hyp_reset();
1378                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1379         }
1380 }
1381
1382 void kvm_arch_hardware_disable(void)
1383 {
1384         _kvm_arch_hardware_disable(NULL);
1385 }
1386
1387 #ifdef CONFIG_CPU_PM
1388 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1389                                     unsigned long cmd,
1390                                     void *v)
1391 {
1392         /*
1393          * kvm_arm_hardware_enabled is left with its old value over
1394          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1395          * re-enable hyp.
1396          */
1397         switch (cmd) {
1398         case CPU_PM_ENTER:
1399                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1400                         /*
1401                          * don't update kvm_arm_hardware_enabled here
1402                          * so that the hardware will be re-enabled
1403                          * when we resume. See below.
1404                          */
1405                         cpu_hyp_reset();
1406
1407                 return NOTIFY_OK;
1408         case CPU_PM_ENTER_FAILED:
1409         case CPU_PM_EXIT:
1410                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1411                         /* The hardware was enabled before suspend. */
1412                         cpu_hyp_reinit();
1413
1414                 return NOTIFY_OK;
1415
1416         default:
1417                 return NOTIFY_DONE;
1418         }
1419 }
1420
1421 static struct notifier_block hyp_init_cpu_pm_nb = {
1422         .notifier_call = hyp_init_cpu_pm_notifier,
1423 };
1424
1425 static void __init hyp_cpu_pm_init(void)
1426 {
1427         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1428 }
1429 static void __init hyp_cpu_pm_exit(void)
1430 {
1431         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1432 }
1433 #else
1434 static inline void hyp_cpu_pm_init(void)
1435 {
1436 }
1437 static inline void hyp_cpu_pm_exit(void)
1438 {
1439 }
1440 #endif
1441
1442 static int init_common_resources(void)
1443 {
1444         kvm_set_ipa_limit();
1445
1446         return 0;
1447 }
1448
1449 static int init_subsystems(void)
1450 {
1451         int err = 0;
1452
1453         /*
1454          * Enable hardware so that subsystem initialisation can access EL2.
1455          */
1456         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1457
1458         /*
1459          * Register CPU lower-power notifier
1460          */
1461         hyp_cpu_pm_init();
1462
1463         /*
1464          * Init HYP view of VGIC
1465          */
1466         err = kvm_vgic_hyp_init();
1467         switch (err) {
1468         case 0:
1469                 vgic_present = true;
1470                 break;
1471         case -ENODEV:
1472         case -ENXIO:
1473                 vgic_present = false;
1474                 err = 0;
1475                 break;
1476         default:
1477                 goto out;
1478         }
1479
1480         /*
1481          * Init HYP architected timer support
1482          */
1483         err = kvm_timer_hyp_init(vgic_present);
1484         if (err)
1485                 goto out;
1486
1487         kvm_perf_init();
1488         kvm_coproc_table_init();
1489
1490 out:
1491         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1492
1493         return err;
1494 }
1495
1496 static void teardown_hyp_mode(void)
1497 {
1498         int cpu;
1499
1500         free_hyp_pgds();
1501         for_each_possible_cpu(cpu)
1502                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1503         hyp_cpu_pm_exit();
1504 }
1505
1506 /**
1507  * Inits Hyp-mode on all online CPUs
1508  */
1509 static int init_hyp_mode(void)
1510 {
1511         int cpu;
1512         int err = 0;
1513
1514         /*
1515          * Allocate Hyp PGD and setup Hyp identity mapping
1516          */
1517         err = kvm_mmu_init();
1518         if (err)
1519                 goto out_err;
1520
1521         /*
1522          * Allocate stack pages for Hypervisor-mode
1523          */
1524         for_each_possible_cpu(cpu) {
1525                 unsigned long stack_page;
1526
1527                 stack_page = __get_free_page(GFP_KERNEL);
1528                 if (!stack_page) {
1529                         err = -ENOMEM;
1530                         goto out_err;
1531                 }
1532
1533                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1534         }
1535
1536         /*
1537          * Map the Hyp-code called directly from the host
1538          */
1539         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1540                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1541         if (err) {
1542                 kvm_err("Cannot map world-switch code\n");
1543                 goto out_err;
1544         }
1545
1546         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1547                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1548         if (err) {
1549                 kvm_err("Cannot map rodata section\n");
1550                 goto out_err;
1551         }
1552
1553         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1554                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1555         if (err) {
1556                 kvm_err("Cannot map bss section\n");
1557                 goto out_err;
1558         }
1559
1560         err = kvm_map_vectors();
1561         if (err) {
1562                 kvm_err("Cannot map vectors\n");
1563                 goto out_err;
1564         }
1565
1566         /*
1567          * Map the Hyp stack pages
1568          */
1569         for_each_possible_cpu(cpu) {
1570                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1571                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1572                                           PAGE_HYP);
1573
1574                 if (err) {
1575                         kvm_err("Cannot map hyp stack\n");
1576                         goto out_err;
1577                 }
1578         }
1579
1580         for_each_possible_cpu(cpu) {
1581                 kvm_host_data_t *cpu_data;
1582
1583                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1584                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1585
1586                 if (err) {
1587                         kvm_err("Cannot map host CPU state: %d\n", err);
1588                         goto out_err;
1589                 }
1590         }
1591
1592         err = hyp_map_aux_data();
1593         if (err)
1594                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1595
1596         return 0;
1597
1598 out_err:
1599         teardown_hyp_mode();
1600         kvm_err("error initializing Hyp mode: %d\n", err);
1601         return err;
1602 }
1603
1604 static void check_kvm_target_cpu(void *ret)
1605 {
1606         *(int *)ret = kvm_target_cpu();
1607 }
1608
1609 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1610 {
1611         struct kvm_vcpu *vcpu;
1612         int i;
1613
1614         mpidr &= MPIDR_HWID_BITMASK;
1615         kvm_for_each_vcpu(i, vcpu, kvm) {
1616                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1617                         return vcpu;
1618         }
1619         return NULL;
1620 }
1621
1622 bool kvm_arch_has_irq_bypass(void)
1623 {
1624         return true;
1625 }
1626
1627 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1628                                       struct irq_bypass_producer *prod)
1629 {
1630         struct kvm_kernel_irqfd *irqfd =
1631                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1632
1633         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1634                                           &irqfd->irq_entry);
1635 }
1636 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1637                                       struct irq_bypass_producer *prod)
1638 {
1639         struct kvm_kernel_irqfd *irqfd =
1640                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1641
1642         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1643                                      &irqfd->irq_entry);
1644 }
1645
1646 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1647 {
1648         struct kvm_kernel_irqfd *irqfd =
1649                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1650
1651         kvm_arm_halt_guest(irqfd->kvm);
1652 }
1653
1654 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1655 {
1656         struct kvm_kernel_irqfd *irqfd =
1657                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1658
1659         kvm_arm_resume_guest(irqfd->kvm);
1660 }
1661
1662 /**
1663  * Initialize Hyp-mode and memory mappings on all CPUs.
1664  */
1665 int kvm_arch_init(void *opaque)
1666 {
1667         int err;
1668         int ret, cpu;
1669         bool in_hyp_mode;
1670
1671         if (!is_hyp_mode_available()) {
1672                 kvm_info("HYP mode not available\n");
1673                 return -ENODEV;
1674         }
1675
1676         in_hyp_mode = is_kernel_in_hyp_mode();
1677
1678         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1679                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1680                 return -ENODEV;
1681         }
1682
1683         for_each_online_cpu(cpu) {
1684                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1685                 if (ret < 0) {
1686                         kvm_err("Error, CPU %d not supported!\n", cpu);
1687                         return -ENODEV;
1688                 }
1689         }
1690
1691         err = init_common_resources();
1692         if (err)
1693                 return err;
1694
1695         err = kvm_arm_init_sve();
1696         if (err)
1697                 return err;
1698
1699         if (!in_hyp_mode) {
1700                 err = init_hyp_mode();
1701                 if (err)
1702                         goto out_err;
1703         }
1704
1705         err = init_subsystems();
1706         if (err)
1707                 goto out_hyp;
1708
1709         if (in_hyp_mode)
1710                 kvm_info("VHE mode initialized successfully\n");
1711         else
1712                 kvm_info("Hyp mode initialized successfully\n");
1713
1714         return 0;
1715
1716 out_hyp:
1717         if (!in_hyp_mode)
1718                 teardown_hyp_mode();
1719 out_err:
1720         return err;
1721 }
1722
1723 /* NOP: Compiling as a module not supported */
1724 void kvm_arch_exit(void)
1725 {
1726         kvm_perf_teardown();
1727 }
1728
1729 static int arm_init(void)
1730 {
1731         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1732         return rc;
1733 }
1734
1735 module_init(arm_init);