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 bool kvm_arch_has_vcpu_debugfs(void)
 148 {
 149         return false;
 150 }
 151
 152 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 153 {
 154         return 0;
 155 }
 156
 157 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 158 {
 159         return VM_FAULT_SIGBUS;
 160 }
 161
 162
 163 /**
 164  * kvm_arch_destroy_vm - destroy the VM data structure
 165  * @kvm:        pointer to the KVM struct
 166  */
 167 void kvm_arch_destroy_vm(struct kvm *kvm)
 168 {
 169         int i;
 170
 171         kvm_vgic_destroy(kvm);
 172
 173         free_percpu(kvm->arch.last_vcpu_ran);
 174         kvm->arch.last_vcpu_ran = NULL;
 175
 176         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 177                 if (kvm->vcpus[i]) {
 178                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 179                         kvm->vcpus[i] = NULL;
 180                 }
 181         }
 182         atomic_set(&kvm->online_vcpus, 0);
 183 }
 184
 185 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 186 {
 187         int r;
 188         switch (ext) {
 189         case KVM_CAP_IRQCHIP:
 190                 r = vgic_present;
 191                 break;
 192         case KVM_CAP_IOEVENTFD:
 193         case KVM_CAP_DEVICE_CTRL:
 194         case KVM_CAP_USER_MEMORY:
 195         case KVM_CAP_SYNC_MMU:
 196         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 197         case KVM_CAP_ONE_REG:
 198         case KVM_CAP_ARM_PSCI:
 199         case KVM_CAP_ARM_PSCI_0_2:
 200         case KVM_CAP_READONLY_MEM:
 201         case KVM_CAP_MP_STATE:
 202         case KVM_CAP_IMMEDIATE_EXIT:
 203         case KVM_CAP_VCPU_EVENTS:
 204                 r = 1;
 205                 break;
 206         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 207                 r = 1;
 208                 break;
 209         case KVM_CAP_NR_VCPUS:
 210                 r = num_online_cpus();
 211                 break;
 212         case KVM_CAP_MAX_VCPUS:
 213                 r = KVM_MAX_VCPUS;
 214                 break;
 215         case KVM_CAP_MAX_VCPU_ID:
 216                 r = KVM_MAX_VCPU_ID;
 217                 break;
 218         case KVM_CAP_MSI_DEVID:
 219                 if (!kvm)
 220                         r = -EINVAL;
 221                 else
 222                         r = kvm->arch.vgic.msis_require_devid;
 223                 break;
 224         case KVM_CAP_ARM_USER_IRQ:
 225                 /*
 226                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 227                  * (bump this number if adding more devices)
 228                  */
 229                 r = 1;
 230                 break;
 231         default:
 232                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 233                 break;
 234         }
 235         return r;
 236 }
 237
 238 long kvm_arch_dev_ioctl(struct file *filp,
 239                         unsigned int ioctl, unsigned long arg)
 240 {
 241         return -EINVAL;
 242 }
 243
 244 struct kvm *kvm_arch_alloc_vm(void)
 245 {
 246         if (!has_vhe())
 247                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 248
 249         return vzalloc(sizeof(struct kvm));
 250 }
 251
 252 void kvm_arch_free_vm(struct kvm *kvm)
 253 {
 254         if (!has_vhe())
 255                 kfree(kvm);
 256         else
 257                 vfree(kvm);
 258 }
 259
 260 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 261 {
 262         int err;
 263         struct kvm_vcpu *vcpu;
 264
 265         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 266                 err = -EBUSY;
 267                 goto out;
 268         }
 269
 270         if (id >= kvm->arch.max_vcpus) {
 271                 err = -EINVAL;
 272                 goto out;
 273         }
 274
 275         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 276         if (!vcpu) {
 277                 err = -ENOMEM;
 278                 goto out;
 279         }
 280
 281         err = kvm_vcpu_init(vcpu, kvm, id);
 282         if (err)
 283                 goto free_vcpu;
 284
 285         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 286         if (err)
 287                 goto vcpu_uninit;
 288
 289         return vcpu;
 290 vcpu_uninit:
 291         kvm_vcpu_uninit(vcpu);
 292 free_vcpu:
 293         kmem_cache_free(kvm_vcpu_cache, vcpu);
 294 out:
 295         return ERR_PTR(err);
 296 }
 297
 298 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 299 {
 300 }
 301
 302 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 303 {
 304         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 305                 static_branch_dec(&userspace_irqchip_in_use);
 306
 307         kvm_mmu_free_memory_caches(vcpu);
 308         kvm_timer_vcpu_terminate(vcpu);
 309         kvm_pmu_vcpu_destroy(vcpu);
 310         kvm_vcpu_uninit(vcpu);
 311         kmem_cache_free(kvm_vcpu_cache, vcpu);
 312 }
 313
 314 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 315 {
 316         kvm_arch_vcpu_free(vcpu);
 317 }
 318
 319 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 320 {
 321         return kvm_timer_is_pending(vcpu);
 322 }
 323
 324 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 325 {
 326         kvm_vgic_v4_enable_doorbell(vcpu);
 327 }
 328
 329 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 330 {
 331         kvm_vgic_v4_disable_doorbell(vcpu);
 332 }
 333
 334 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 335 {
 336         /* Force users to call KVM_ARM_VCPU_INIT */
 337         vcpu->arch.target = -1;
 338         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 339
 340         /* Set up the timer */
 341         kvm_timer_vcpu_init(vcpu);
 342
 343         kvm_arm_reset_debug_ptr(vcpu);
 344
 345         return kvm_vgic_vcpu_init(vcpu);
 346 }
 347
 348 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 349 {
 350         int *last_ran;
 351         kvm_host_data_t *cpu_data;
 352
 353         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 354         cpu_data = this_cpu_ptr(&kvm_host_data);
 355
 356         /*
 357          * We might get preempted before the vCPU actually runs, but
 358          * over-invalidation doesn't affect correctness.
 359          */
 360         if (*last_ran != vcpu->vcpu_id) {
 361                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 362                 *last_ran = vcpu->vcpu_id;
 363         }
 364
 365         vcpu->cpu = cpu;
 366         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 367
 368         kvm_arm_set_running_vcpu(vcpu);
 369         kvm_vgic_load(vcpu);
 370         kvm_timer_vcpu_load(vcpu);
 371         kvm_vcpu_load_sysregs(vcpu);
 372         kvm_arch_vcpu_load_fp(vcpu);
 373         kvm_vcpu_pmu_restore_guest(vcpu);
 374
 375         if (single_task_running())
 376                 vcpu_clear_wfe_traps(vcpu);
 377         else
 378                 vcpu_set_wfe_traps(vcpu);
 379
 380         vcpu_ptrauth_setup_lazy(vcpu);
 381 }
 382
 383 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 384 {
 385         kvm_arch_vcpu_put_fp(vcpu);
 386         kvm_vcpu_put_sysregs(vcpu);
 387         kvm_timer_vcpu_put(vcpu);
 388         kvm_vgic_put(vcpu);
 389         kvm_vcpu_pmu_restore_host(vcpu);
 390
 391         vcpu->cpu = -1;
 392
 393         kvm_arm_set_running_vcpu(NULL);
 394 }
 395
 396 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 397 {
 398         vcpu->arch.power_off = true;
 399         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 400         kvm_vcpu_kick(vcpu);
 401 }
 402
 403 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 404                                     struct kvm_mp_state *mp_state)
 405 {
 406         if (vcpu->arch.power_off)
 407                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 408         else
 409                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 410
 411         return 0;
 412 }
 413
 414 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 415                                     struct kvm_mp_state *mp_state)
 416 {
 417         int ret = 0;
 418
 419         switch (mp_state->mp_state) {
 420         case KVM_MP_STATE_RUNNABLE:
 421                 vcpu->arch.power_off = false;
 422                 break;
 423         case KVM_MP_STATE_STOPPED:
 424                 vcpu_power_off(vcpu);
 425                 break;
 426         default:
 427                 ret = -EINVAL;
 428         }
 429
 430         return ret;
 431 }
 432
 433 /**
 434  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 435  * @v:          The VCPU pointer
 436  *
 437  * If the guest CPU is not waiting for interrupts or an interrupt line is
 438  * asserted, the CPU is by definition runnable.
 439  */
 440 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 441 {
 442         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 443         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 444                 && !v->arch.power_off && !v->arch.pause);
 445 }
 446
 447 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 448 {
 449         return vcpu_mode_priv(vcpu);
 450 }
 451
 452 /* Just ensure a guest exit from a particular CPU */
 453 static void exit_vm_noop(void *info)
 454 {
 455 }
 456
 457 void force_vm_exit(const cpumask_t *mask)
 458 {
 459         preempt_disable();
 460         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 461         preempt_enable();
 462 }
 463
 464 /**
 465  * need_new_vmid_gen - check that the VMID is still valid
 466  * @vmid: The VMID to check
 467  *
 468  * return true if there is a new generation of VMIDs being used
 469  *
 470  * The hardware supports a limited set of values with the value zero reserved
 471  * for the host, so we check if an assigned value belongs to a previous
 472  * generation, which which requires us to assign a new value. If we're the
 473  * first to use a VMID for the new generation, we must flush necessary caches
 474  * and TLBs on all CPUs.
 475  */
 476 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 477 {
 478         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 479         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 480         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 481 }
 482
 483 /**
 484  * update_vmid - Update the vmid with a valid VMID for the current generation
 485  * @kvm: The guest that struct vmid belongs to
 486  * @vmid: The stage-2 VMID information struct
 487  */
 488 static void update_vmid(struct kvm_vmid *vmid)
 489 {
 490         if (!need_new_vmid_gen(vmid))
 491                 return;
 492
 493         spin_lock(&kvm_vmid_lock);
 494
 495         /*
 496          * We need to re-check the vmid_gen here to ensure that if another vcpu
 497          * already allocated a valid vmid for this vm, then this vcpu should
 498          * use the same vmid.
 499          */
 500         if (!need_new_vmid_gen(vmid)) {
 501                 spin_unlock(&kvm_vmid_lock);
 502                 return;
 503         }
 504
 505         /* First user of a new VMID generation? */
 506         if (unlikely(kvm_next_vmid == 0)) {
 507                 atomic64_inc(&kvm_vmid_gen);
 508                 kvm_next_vmid = 1;
 509
 510                 /*
 511                  * On SMP we know no other CPUs can use this CPU's or each
 512                  * other's VMID after force_vm_exit returns since the
 513                  * kvm_vmid_lock blocks them from reentry to the guest.
 514                  */
 515                 force_vm_exit(cpu_all_mask);
 516                 /*
 517                  * Now broadcast TLB + ICACHE invalidation over the inner
 518                  * shareable domain to make sure all data structures are
 519                  * clean.
 520                  */
 521                 kvm_call_hyp(__kvm_flush_vm_context);
 522         }
 523
 524         vmid->vmid = kvm_next_vmid;
 525         kvm_next_vmid++;
 526         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 527
 528         smp_wmb();
 529         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 530
 531         spin_unlock(&kvm_vmid_lock);
 532 }
 533
 534 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 535 {
 536         struct kvm *kvm = vcpu->kvm;
 537         int ret = 0;
 538
 539         if (likely(vcpu->arch.has_run_once))
 540                 return 0;
 541
 542         if (!kvm_arm_vcpu_is_finalized(vcpu))
 543                 return -EPERM;
 544
 545         vcpu->arch.has_run_once = true;
 546
 547         if (likely(irqchip_in_kernel(kvm))) {
 548                 /*
 549                  * Map the VGIC hardware resources before running a vcpu the
 550                  * first time on this VM.
 551                  */
 552                 if (unlikely(!vgic_ready(kvm))) {
 553                         ret = kvm_vgic_map_resources(kvm);
 554                         if (ret)
 555                                 return ret;
 556                 }
 557         } else {
 558                 /*
 559                  * Tell the rest of the code that there are userspace irqchip
 560                  * VMs in the wild.
 561                  */
 562                 static_branch_inc(&userspace_irqchip_in_use);
 563         }
 564
 565         ret = kvm_timer_enable(vcpu);
 566         if (ret)
 567                 return ret;
 568
 569         ret = kvm_arm_pmu_v3_enable(vcpu);
 570
 571         return ret;
 572 }
 573
 574 bool kvm_arch_intc_initialized(struct kvm *kvm)
 575 {
 576         return vgic_initialized(kvm);
 577 }
 578
 579 void kvm_arm_halt_guest(struct kvm *kvm)
 580 {
 581         int i;
 582         struct kvm_vcpu *vcpu;
 583
 584         kvm_for_each_vcpu(i, vcpu, kvm)
 585                 vcpu->arch.pause = true;
 586         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 587 }
 588
 589 void kvm_arm_resume_guest(struct kvm *kvm)
 590 {
 591         int i;
 592         struct kvm_vcpu *vcpu;
 593
 594         kvm_for_each_vcpu(i, vcpu, kvm) {
 595                 vcpu->arch.pause = false;
 596                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 597         }
 598 }
 599
 600 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 601 {
 602         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 603
 604         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 605                                        (!vcpu->arch.pause)));
 606
 607         if (vcpu->arch.power_off || vcpu->arch.pause) {
 608                 /* Awaken to handle a signal, request we sleep again later. */
 609                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 610         }
 611
 612         /*
 613          * Make sure we will observe a potential reset request if we've
 614          * observed a change to the power state. Pairs with the smp_wmb() in
 615          * kvm_psci_vcpu_on().
 616          */
 617         smp_rmb();
 618 }
 619
 620 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 621 {
 622         return vcpu->arch.target >= 0;
 623 }
 624
 625 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 626 {
 627         if (kvm_request_pending(vcpu)) {
 628                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 629                         vcpu_req_sleep(vcpu);
 630
 631                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 632                         kvm_reset_vcpu(vcpu);
 633
 634                 /*
 635                  * Clear IRQ_PENDING requests that were made to guarantee
 636                  * that a VCPU sees new virtual interrupts.
 637                  */
 638                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 639         }
 640 }
 641
 642 /**
 643  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 644  * @vcpu:       The VCPU pointer
 645  * @run:        The kvm_run structure pointer used for userspace state exchange
 646  *
 647  * This function is called through the VCPU_RUN ioctl called from user space. It
 648  * will execute VM code in a loop until the time slice for the process is used
 649  * or some emulation is needed from user space in which case the function will
 650  * return with return value 0 and with the kvm_run structure filled in with the
 651  * required data for the requested emulation.
 652  */
 653 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 654 {
 655         int ret;
 656
 657         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 658                 return -ENOEXEC;
 659
 660         ret = kvm_vcpu_first_run_init(vcpu);
 661         if (ret)
 662                 return ret;
 663
 664         if (run->exit_reason == KVM_EXIT_MMIO) {
 665                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 666                 if (ret)
 667                         return ret;
 668         }
 669
 670         if (run->immediate_exit)
 671                 return -EINTR;
 672
 673         vcpu_load(vcpu);
 674
 675         kvm_sigset_activate(vcpu);
 676
 677         ret = 1;
 678         run->exit_reason = KVM_EXIT_UNKNOWN;
 679         while (ret > 0) {
 680                 /*
 681                  * Check conditions before entering the guest
 682                  */
 683                 cond_resched();
 684
 685                 update_vmid(&vcpu->kvm->arch.vmid);
 686
 687                 check_vcpu_requests(vcpu);
 688
 689                 /*
 690                  * Preparing the interrupts to be injected also
 691                  * involves poking the GIC, which must be done in a
 692                  * non-preemptible context.
 693                  */
 694                 preempt_disable();
 695
 696                 kvm_pmu_flush_hwstate(vcpu);
 697
 698                 local_irq_disable();
 699
 700                 kvm_vgic_flush_hwstate(vcpu);
 701
 702                 /*
 703                  * Exit if we have a signal pending so that we can deliver the
 704                  * signal to user space.
 705                  */
 706                 if (signal_pending(current)) {
 707                         ret = -EINTR;
 708                         run->exit_reason = KVM_EXIT_INTR;
 709                 }
 710
 711                 /*
 712                  * If we're using a userspace irqchip, then check if we need
 713                  * to tell a userspace irqchip about timer or PMU level
 714                  * changes and if so, exit to userspace (the actual level
 715                  * state gets updated in kvm_timer_update_run and
 716                  * kvm_pmu_update_run below).
 717                  */
 718                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 719                         if (kvm_timer_should_notify_user(vcpu) ||
 720                             kvm_pmu_should_notify_user(vcpu)) {
 721                                 ret = -EINTR;
 722                                 run->exit_reason = KVM_EXIT_INTR;
 723                         }
 724                 }
 725
 726                 /*
 727                  * Ensure we set mode to IN_GUEST_MODE after we disable
 728                  * interrupts and before the final VCPU requests check.
 729                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 730                  * Documentation/virtual/kvm/vcpu-requests.rst
 731                  */
 732                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 733
 734                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 735                     kvm_request_pending(vcpu)) {
 736                         vcpu->mode = OUTSIDE_GUEST_MODE;
 737                         isb(); /* Ensure work in x_flush_hwstate is committed */
 738                         kvm_pmu_sync_hwstate(vcpu);
 739                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 740                                 kvm_timer_sync_hwstate(vcpu);
 741                         kvm_vgic_sync_hwstate(vcpu);
 742                         local_irq_enable();
 743                         preempt_enable();
 744                         continue;
 745                 }
 746
 747                 kvm_arm_setup_debug(vcpu);
 748
 749                 /**************************************************************
 750                  * Enter the guest
 751                  */
 752                 trace_kvm_entry(*vcpu_pc(vcpu));
 753                 guest_enter_irqoff();
 754
 755                 if (has_vhe()) {
 756                         kvm_arm_vhe_guest_enter();
 757                         ret = kvm_vcpu_run_vhe(vcpu);
 758                         kvm_arm_vhe_guest_exit();
 759                 } else {
 760                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 761                 }
 762
 763                 vcpu->mode = OUTSIDE_GUEST_MODE;
 764                 vcpu->stat.exits++;
 765                 /*
 766                  * Back from guest
 767                  *************************************************************/
 768
 769                 kvm_arm_clear_debug(vcpu);
 770
 771                 /*
 772                  * We must sync the PMU state before the vgic state so
 773                  * that the vgic can properly sample the updated state of the
 774                  * interrupt line.
 775                  */
 776                 kvm_pmu_sync_hwstate(vcpu);
 777
 778                 /*
 779                  * Sync the vgic state before syncing the timer state because
 780                  * the timer code needs to know if the virtual timer
 781                  * interrupts are active.
 782                  */
 783                 kvm_vgic_sync_hwstate(vcpu);
 784
 785                 /*
 786                  * Sync the timer hardware state before enabling interrupts as
 787                  * we don't want vtimer interrupts to race with syncing the
 788                  * timer virtual interrupt state.
 789                  */
 790                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 791                         kvm_timer_sync_hwstate(vcpu);
 792
 793                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 794
 795                 /*
 796                  * We may have taken a host interrupt in HYP mode (ie
 797                  * while executing the guest). This interrupt is still
 798                  * pending, as we haven't serviced it yet!
 799                  *
 800                  * We're now back in SVC mode, with interrupts
 801                  * disabled.  Enabling the interrupts now will have
 802                  * the effect of taking the interrupt again, in SVC
 803                  * mode this time.
 804                  */
 805                 local_irq_enable();
 806
 807                 /*
 808                  * We do local_irq_enable() before calling guest_exit() so
 809                  * that if a timer interrupt hits while running the guest we
 810                  * account that tick as being spent in the guest.  We enable
 811                  * preemption after calling guest_exit() so that if we get
 812                  * preempted we make sure ticks after that is not counted as
 813                  * guest time.
 814                  */
 815                 guest_exit();
 816                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 817
 818                 /* Exit types that need handling before we can be preempted */
 819                 handle_exit_early(vcpu, run, ret);
 820
 821                 preempt_enable();
 822
 823                 ret = handle_exit(vcpu, run, ret);
 824         }
 825
 826         /* Tell userspace about in-kernel device output levels */
 827         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 828                 kvm_timer_update_run(vcpu);
 829                 kvm_pmu_update_run(vcpu);
 830         }
 831
 832         kvm_sigset_deactivate(vcpu);
 833
 834         vcpu_put(vcpu);
 835         return ret;
 836 }
 837
 838 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 839 {
 840         int bit_index;
 841         bool set;
 842         unsigned long *hcr;
 843
 844         if (number == KVM_ARM_IRQ_CPU_IRQ)
 845                 bit_index = __ffs(HCR_VI);
 846         else /* KVM_ARM_IRQ_CPU_FIQ */
 847                 bit_index = __ffs(HCR_VF);
 848
 849         hcr = vcpu_hcr(vcpu);
 850         if (level)
 851                 set = test_and_set_bit(bit_index, hcr);
 852         else
 853                 set = test_and_clear_bit(bit_index, hcr);
 854
 855         /*
 856          * If we didn't change anything, no need to wake up or kick other CPUs
 857          */
 858         if (set == level)
 859                 return 0;
 860
 861         /*
 862          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 863          * trigger a world-switch round on the running physical CPU to set the
 864          * virtual IRQ/FIQ fields in the HCR appropriately.
 865          */
 866         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 867         kvm_vcpu_kick(vcpu);
 868
 869         return 0;
 870 }
 871
 872 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 873                           bool line_status)
 874 {
 875         u32 irq = irq_level->irq;
 876         unsigned int irq_type, vcpu_idx, irq_num;
 877         int nrcpus = atomic_read(&kvm->online_vcpus);
 878         struct kvm_vcpu *vcpu = NULL;
 879         bool level = irq_level->level;
 880
 881         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 882         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 883         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 884
 885         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 886
 887         switch (irq_type) {
 888         case KVM_ARM_IRQ_TYPE_CPU:
 889                 if (irqchip_in_kernel(kvm))
 890                         return -ENXIO;
 891
 892                 if (vcpu_idx >= nrcpus)
 893                         return -EINVAL;
 894
 895                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 896                 if (!vcpu)
 897                         return -EINVAL;
 898
 899                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 900                         return -EINVAL;
 901
 902                 return vcpu_interrupt_line(vcpu, irq_num, level);
 903         case KVM_ARM_IRQ_TYPE_PPI:
 904                 if (!irqchip_in_kernel(kvm))
 905                         return -ENXIO;
 906
 907                 if (vcpu_idx >= nrcpus)
 908                         return -EINVAL;
 909
 910                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 911                 if (!vcpu)
 912                         return -EINVAL;
 913
 914                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 915                         return -EINVAL;
 916
 917                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 918         case KVM_ARM_IRQ_TYPE_SPI:
 919                 if (!irqchip_in_kernel(kvm))
 920                         return -ENXIO;
 921
 922                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 923                         return -EINVAL;
 924
 925                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 926         }
 927
 928         return -EINVAL;
 929 }
 930
 931 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 932                                const struct kvm_vcpu_init *init)
 933 {
 934         unsigned int i, ret;
 935         int phys_target = kvm_target_cpu();
 936
 937         if (init->target != phys_target)
 938                 return -EINVAL;
 939
 940         /*
 941          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 942          * use the same target.
 943          */
 944         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 945                 return -EINVAL;
 946
 947         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 948         for (i = 0; i < sizeof(init->features) * 8; i++) {
 949                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 950
 951                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 952                         return -ENOENT;
 953
 954                 /*
 955                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 956                  * use the same feature set.
 957                  */
 958                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 959                     test_bit(i, vcpu->arch.features) != set)
 960                         return -EINVAL;
 961
 962                 if (set)
 963                         set_bit(i, vcpu->arch.features);
 964         }
 965
 966         vcpu->arch.target = phys_target;
 967
 968         /* Now we know what it is, we can reset it. */
 969         ret = kvm_reset_vcpu(vcpu);
 970         if (ret) {
 971                 vcpu->arch.target = -1;
 972                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 973         }
 974
 975         return ret;
 976 }
 977
 978 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 979                                          struct kvm_vcpu_init *init)
 980 {
 981         int ret;
 982
 983         ret = kvm_vcpu_set_target(vcpu, init);
 984         if (ret)
 985                 return ret;
 986
 987         /*
 988          * Ensure a rebooted VM will fault in RAM pages and detect if the
 989          * guest MMU is turned off and flush the caches as needed.
 990          */
 991         if (vcpu->arch.has_run_once)
 992                 stage2_unmap_vm(vcpu->kvm);
 993
 994         vcpu_reset_hcr(vcpu);
 995
 996         /*
 997          * Handle the "start in power-off" case.
 998          */
 999         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1000                 vcpu_power_off(vcpu);
1001         else
1002                 vcpu->arch.power_off = false;
1003
1004         return 0;
1005 }
1006
1007 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1008                                  struct kvm_device_attr *attr)
1009 {
1010         int ret = -ENXIO;
1011
1012         switch (attr->group) {
1013         default:
1014                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1015                 break;
1016         }
1017
1018         return ret;
1019 }
1020
1021 static int kvm_arm_vcpu_get_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_get_attr(vcpu, attr);
1029                 break;
1030         }
1031
1032         return ret;
1033 }
1034
1035 static int kvm_arm_vcpu_has_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_has_attr(vcpu, attr);
1043                 break;
1044         }
1045
1046         return ret;
1047 }
1048
1049 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1050                                    struct kvm_vcpu_events *events)
1051 {
1052         memset(events, 0, sizeof(*events));
1053
1054         return __kvm_arm_vcpu_get_events(vcpu, events);
1055 }
1056
1057 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1058                                    struct kvm_vcpu_events *events)
1059 {
1060         int i;
1061
1062         /* check whether the reserved field is zero */
1063         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1064                 if (events->reserved[i])
1065                         return -EINVAL;
1066
1067         /* check whether the pad field is zero */
1068         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1069                 if (events->exception.pad[i])
1070                         return -EINVAL;
1071
1072         return __kvm_arm_vcpu_set_events(vcpu, events);
1073 }
1074
1075 long kvm_arch_vcpu_ioctl(struct file *filp,
1076                          unsigned int ioctl, unsigned long arg)
1077 {
1078         struct kvm_vcpu *vcpu = filp->private_data;
1079         void __user *argp = (void __user *)arg;
1080         struct kvm_device_attr attr;
1081         long r;
1082
1083         switch (ioctl) {
1084         case KVM_ARM_VCPU_INIT: {
1085                 struct kvm_vcpu_init init;
1086
1087                 r = -EFAULT;
1088                 if (copy_from_user(&init, argp, sizeof(init)))
1089                         break;
1090
1091                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1092                 break;
1093         }
1094         case KVM_SET_ONE_REG:
1095         case KVM_GET_ONE_REG: {
1096                 struct kvm_one_reg reg;
1097
1098                 r = -ENOEXEC;
1099                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1100                         break;
1101
1102                 r = -EFAULT;
1103                 if (copy_from_user(&reg, argp, sizeof(reg)))
1104                         break;
1105
1106                 if (ioctl == KVM_SET_ONE_REG)
1107                         r = kvm_arm_set_reg(vcpu, &reg);
1108                 else
1109                         r = kvm_arm_get_reg(vcpu, &reg);
1110                 break;
1111         }
1112         case KVM_GET_REG_LIST: {
1113                 struct kvm_reg_list __user *user_list = argp;
1114                 struct kvm_reg_list reg_list;
1115                 unsigned n;
1116
1117                 r = -ENOEXEC;
1118                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1119                         break;
1120
1121                 r = -EPERM;
1122                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1123                         break;
1124
1125                 r = -EFAULT;
1126                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1127                         break;
1128                 n = reg_list.n;
1129                 reg_list.n = kvm_arm_num_regs(vcpu);
1130                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1131                         break;
1132                 r = -E2BIG;
1133                 if (n < reg_list.n)
1134                         break;
1135                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1136                 break;
1137         }
1138         case KVM_SET_DEVICE_ATTR: {
1139                 r = -EFAULT;
1140                 if (copy_from_user(&attr, argp, sizeof(attr)))
1141                         break;
1142                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1143                 break;
1144         }
1145         case KVM_GET_DEVICE_ATTR: {
1146                 r = -EFAULT;
1147                 if (copy_from_user(&attr, argp, sizeof(attr)))
1148                         break;
1149                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1150                 break;
1151         }
1152         case KVM_HAS_DEVICE_ATTR: {
1153                 r = -EFAULT;
1154                 if (copy_from_user(&attr, argp, sizeof(attr)))
1155                         break;
1156                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1157                 break;
1158         }
1159         case KVM_GET_VCPU_EVENTS: {
1160                 struct kvm_vcpu_events events;
1161
1162                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1163                         return -EINVAL;
1164
1165                 if (copy_to_user(argp, &events, sizeof(events)))
1166                         return -EFAULT;
1167
1168                 return 0;
1169         }
1170         case KVM_SET_VCPU_EVENTS: {
1171                 struct kvm_vcpu_events events;
1172
1173                 if (copy_from_user(&events, argp, sizeof(events)))
1174                         return -EFAULT;
1175
1176                 return kvm_arm_vcpu_set_events(vcpu, &events);
1177         }
1178         case KVM_ARM_VCPU_FINALIZE: {
1179                 int what;
1180
1181                 if (!kvm_vcpu_initialized(vcpu))
1182                         return -ENOEXEC;
1183
1184                 if (get_user(what, (const int __user *)argp))
1185                         return -EFAULT;
1186
1187                 return kvm_arm_vcpu_finalize(vcpu, what);
1188         }
1189         default:
1190                 r = -EINVAL;
1191         }
1192
1193         return r;
1194 }
1195
1196 /**
1197  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1198  * @kvm: kvm instance
1199  * @log: slot id and address to which we copy the log
1200  *
1201  * Steps 1-4 below provide general overview of dirty page logging. See
1202  * kvm_get_dirty_log_protect() function description for additional details.
1203  *
1204  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1205  * always flush the TLB (step 4) even if previous step failed  and the dirty
1206  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1207  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1208  * writes will be marked dirty for next log read.
1209  *
1210  *   1. Take a snapshot of the bit and clear it if needed.
1211  *   2. Write protect the corresponding page.
1212  *   3. Copy the snapshot to the userspace.
1213  *   4. Flush TLB's if needed.
1214  */
1215 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1216 {
1217         bool flush = false;
1218         int r;
1219
1220         mutex_lock(&kvm->slots_lock);
1221
1222         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1223
1224         if (flush)
1225                 kvm_flush_remote_tlbs(kvm);
1226
1227         mutex_unlock(&kvm->slots_lock);
1228         return r;
1229 }
1230
1231 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1232 {
1233         bool flush = false;
1234         int r;
1235
1236         mutex_lock(&kvm->slots_lock);
1237
1238         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1239
1240         if (flush)
1241                 kvm_flush_remote_tlbs(kvm);
1242
1243         mutex_unlock(&kvm->slots_lock);
1244         return r;
1245 }
1246
1247 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1248                                         struct kvm_arm_device_addr *dev_addr)
1249 {
1250         unsigned long dev_id, type;
1251
1252         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1253                 KVM_ARM_DEVICE_ID_SHIFT;
1254         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1255                 KVM_ARM_DEVICE_TYPE_SHIFT;
1256
1257         switch (dev_id) {
1258         case KVM_ARM_DEVICE_VGIC_V2:
1259                 if (!vgic_present)
1260                         return -ENXIO;
1261                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1262         default:
1263                 return -ENODEV;
1264         }
1265 }
1266
1267 long kvm_arch_vm_ioctl(struct file *filp,
1268                        unsigned int ioctl, unsigned long arg)
1269 {
1270         struct kvm *kvm = filp->private_data;
1271         void __user *argp = (void __user *)arg;
1272
1273         switch (ioctl) {
1274         case KVM_CREATE_IRQCHIP: {
1275                 int ret;
1276                 if (!vgic_present)
1277                         return -ENXIO;
1278                 mutex_lock(&kvm->lock);
1279                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1280                 mutex_unlock(&kvm->lock);
1281                 return ret;
1282         }
1283         case KVM_ARM_SET_DEVICE_ADDR: {
1284                 struct kvm_arm_device_addr dev_addr;
1285
1286                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1287                         return -EFAULT;
1288                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1289         }
1290         case KVM_ARM_PREFERRED_TARGET: {
1291                 int err;
1292                 struct kvm_vcpu_init init;
1293
1294                 err = kvm_vcpu_preferred_target(&init);
1295                 if (err)
1296                         return err;
1297
1298                 if (copy_to_user(argp, &init, sizeof(init)))
1299                         return -EFAULT;
1300
1301                 return 0;
1302         }
1303         default:
1304                 return -EINVAL;
1305         }
1306 }
1307
1308 static void cpu_init_hyp_mode(void *dummy)
1309 {
1310         phys_addr_t pgd_ptr;
1311         unsigned long hyp_stack_ptr;
1312         unsigned long stack_page;
1313         unsigned long vector_ptr;
1314
1315         /* Switch from the HYP stub to our own HYP init vector */
1316         __hyp_set_vectors(kvm_get_idmap_vector());
1317
1318         pgd_ptr = kvm_mmu_get_httbr();
1319         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1320         hyp_stack_ptr = stack_page + PAGE_SIZE;
1321         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1322
1323         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1324         __cpu_init_stage2();
1325 }
1326
1327 static void cpu_hyp_reset(void)
1328 {
1329         if (!is_kernel_in_hyp_mode())
1330                 __hyp_reset_vectors();
1331 }
1332
1333 static void cpu_hyp_reinit(void)
1334 {
1335         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1336
1337         cpu_hyp_reset();
1338
1339         if (is_kernel_in_hyp_mode())
1340                 kvm_timer_init_vhe();
1341         else
1342                 cpu_init_hyp_mode(NULL);
1343
1344         kvm_arm_init_debug();
1345
1346         if (vgic_present)
1347                 kvm_vgic_init_cpu_hardware();
1348 }
1349
1350 static void _kvm_arch_hardware_enable(void *discard)
1351 {
1352         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1353                 cpu_hyp_reinit();
1354                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1355         }
1356 }
1357
1358 int kvm_arch_hardware_enable(void)
1359 {
1360         _kvm_arch_hardware_enable(NULL);
1361         return 0;
1362 }
1363
1364 static void _kvm_arch_hardware_disable(void *discard)
1365 {
1366         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1367                 cpu_hyp_reset();
1368                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1369         }
1370 }
1371
1372 void kvm_arch_hardware_disable(void)
1373 {
1374         _kvm_arch_hardware_disable(NULL);
1375 }
1376
1377 #ifdef CONFIG_CPU_PM
1378 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1379                                     unsigned long cmd,
1380                                     void *v)
1381 {
1382         /*
1383          * kvm_arm_hardware_enabled is left with its old value over
1384          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1385          * re-enable hyp.
1386          */
1387         switch (cmd) {
1388         case CPU_PM_ENTER:
1389                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1390                         /*
1391                          * don't update kvm_arm_hardware_enabled here
1392                          * so that the hardware will be re-enabled
1393                          * when we resume. See below.
1394                          */
1395                         cpu_hyp_reset();
1396
1397                 return NOTIFY_OK;
1398         case CPU_PM_ENTER_FAILED:
1399         case CPU_PM_EXIT:
1400                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1401                         /* The hardware was enabled before suspend. */
1402                         cpu_hyp_reinit();
1403
1404                 return NOTIFY_OK;
1405
1406         default:
1407                 return NOTIFY_DONE;
1408         }
1409 }
1410
1411 static struct notifier_block hyp_init_cpu_pm_nb = {
1412         .notifier_call = hyp_init_cpu_pm_notifier,
1413 };
1414
1415 static void __init hyp_cpu_pm_init(void)
1416 {
1417         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1418 }
1419 static void __init hyp_cpu_pm_exit(void)
1420 {
1421         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1422 }
1423 #else
1424 static inline void hyp_cpu_pm_init(void)
1425 {
1426 }
1427 static inline void hyp_cpu_pm_exit(void)
1428 {
1429 }
1430 #endif
1431
1432 static int init_common_resources(void)
1433 {
1434         kvm_set_ipa_limit();
1435
1436         return 0;
1437 }
1438
1439 static int init_subsystems(void)
1440 {
1441         int err = 0;
1442
1443         /*
1444          * Enable hardware so that subsystem initialisation can access EL2.
1445          */
1446         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1447
1448         /*
1449          * Register CPU lower-power notifier
1450          */
1451         hyp_cpu_pm_init();
1452
1453         /*
1454          * Init HYP view of VGIC
1455          */
1456         err = kvm_vgic_hyp_init();
1457         switch (err) {
1458         case 0:
1459                 vgic_present = true;
1460                 break;
1461         case -ENODEV:
1462         case -ENXIO:
1463                 vgic_present = false;
1464                 err = 0;
1465                 break;
1466         default:
1467                 goto out;
1468         }
1469
1470         /*
1471          * Init HYP architected timer support
1472          */
1473         err = kvm_timer_hyp_init(vgic_present);
1474         if (err)
1475                 goto out;
1476
1477         kvm_perf_init();
1478         kvm_coproc_table_init();
1479
1480 out:
1481         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1482
1483         return err;
1484 }
1485
1486 static void teardown_hyp_mode(void)
1487 {
1488         int cpu;
1489
1490         free_hyp_pgds();
1491         for_each_possible_cpu(cpu)
1492                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1493         hyp_cpu_pm_exit();
1494 }
1495
1496 /**
1497  * Inits Hyp-mode on all online CPUs
1498  */
1499 static int init_hyp_mode(void)
1500 {
1501         int cpu;
1502         int err = 0;
1503
1504         /*
1505          * Allocate Hyp PGD and setup Hyp identity mapping
1506          */
1507         err = kvm_mmu_init();
1508         if (err)
1509                 goto out_err;
1510
1511         /*
1512          * Allocate stack pages for Hypervisor-mode
1513          */
1514         for_each_possible_cpu(cpu) {
1515                 unsigned long stack_page;
1516
1517                 stack_page = __get_free_page(GFP_KERNEL);
1518                 if (!stack_page) {
1519                         err = -ENOMEM;
1520                         goto out_err;
1521                 }
1522
1523                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1524         }
1525
1526         /*
1527          * Map the Hyp-code called directly from the host
1528          */
1529         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1530                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1531         if (err) {
1532                 kvm_err("Cannot map world-switch code\n");
1533                 goto out_err;
1534         }
1535
1536         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1537                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1538         if (err) {
1539                 kvm_err("Cannot map rodata section\n");
1540                 goto out_err;
1541         }
1542
1543         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1544                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1545         if (err) {
1546                 kvm_err("Cannot map bss section\n");
1547                 goto out_err;
1548         }
1549
1550         err = kvm_map_vectors();
1551         if (err) {
1552                 kvm_err("Cannot map vectors\n");
1553                 goto out_err;
1554         }
1555
1556         /*
1557          * Map the Hyp stack pages
1558          */
1559         for_each_possible_cpu(cpu) {
1560                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1561                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1562                                           PAGE_HYP);
1563
1564                 if (err) {
1565                         kvm_err("Cannot map hyp stack\n");
1566                         goto out_err;
1567                 }
1568         }
1569
1570         for_each_possible_cpu(cpu) {
1571                 kvm_host_data_t *cpu_data;
1572
1573                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1574                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1575
1576                 if (err) {
1577                         kvm_err("Cannot map host CPU state: %d\n", err);
1578                         goto out_err;
1579                 }
1580         }
1581
1582         err = hyp_map_aux_data();
1583         if (err)
1584                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1585
1586         return 0;
1587
1588 out_err:
1589         teardown_hyp_mode();
1590         kvm_err("error initializing Hyp mode: %d\n", err);
1591         return err;
1592 }
1593
1594 static void check_kvm_target_cpu(void *ret)
1595 {
1596         *(int *)ret = kvm_target_cpu();
1597 }
1598
1599 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1600 {
1601         struct kvm_vcpu *vcpu;
1602         int i;
1603
1604         mpidr &= MPIDR_HWID_BITMASK;
1605         kvm_for_each_vcpu(i, vcpu, kvm) {
1606                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1607                         return vcpu;
1608         }
1609         return NULL;
1610 }
1611
1612 bool kvm_arch_has_irq_bypass(void)
1613 {
1614         return true;
1615 }
1616
1617 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1618                                       struct irq_bypass_producer *prod)
1619 {
1620         struct kvm_kernel_irqfd *irqfd =
1621                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1622
1623         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1624                                           &irqfd->irq_entry);
1625 }
1626 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1627                                       struct irq_bypass_producer *prod)
1628 {
1629         struct kvm_kernel_irqfd *irqfd =
1630                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1631
1632         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1633                                      &irqfd->irq_entry);
1634 }
1635
1636 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1637 {
1638         struct kvm_kernel_irqfd *irqfd =
1639                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1640
1641         kvm_arm_halt_guest(irqfd->kvm);
1642 }
1643
1644 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1645 {
1646         struct kvm_kernel_irqfd *irqfd =
1647                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1648
1649         kvm_arm_resume_guest(irqfd->kvm);
1650 }
1651
1652 /**
1653  * Initialize Hyp-mode and memory mappings on all CPUs.
1654  */
1655 int kvm_arch_init(void *opaque)
1656 {
1657         int err;
1658         int ret, cpu;
1659         bool in_hyp_mode;
1660
1661         if (!is_hyp_mode_available()) {
1662                 kvm_info("HYP mode not available\n");
1663                 return -ENODEV;
1664         }
1665
1666         in_hyp_mode = is_kernel_in_hyp_mode();
1667
1668         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1669                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1670                 return -ENODEV;
1671         }
1672
1673         for_each_online_cpu(cpu) {
1674                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1675                 if (ret < 0) {
1676                         kvm_err("Error, CPU %d not supported!\n", cpu);
1677                         return -ENODEV;
1678                 }
1679         }
1680
1681         err = init_common_resources();
1682         if (err)
1683                 return err;
1684
1685         err = kvm_arm_init_sve();
1686         if (err)
1687                 return err;
1688
1689         if (!in_hyp_mode) {
1690                 err = init_hyp_mode();
1691                 if (err)
1692                         goto out_err;
1693         }
1694
1695         err = init_subsystems();
1696         if (err)
1697                 goto out_hyp;
1698
1699         if (in_hyp_mode)
1700                 kvm_info("VHE mode initialized successfully\n");
1701         else
1702                 kvm_info("Hyp mode initialized successfully\n");
1703
1704         return 0;
1705
1706 out_hyp:
1707         if (!in_hyp_mode)
1708                 teardown_hyp_mode();
1709 out_err:
1710         return err;
1711 }
1712
1713 /* NOP: Compiling as a module not supported */
1714 void kvm_arch_exit(void)
1715 {
1716         kvm_perf_teardown();
1717 }
1718
1719 static int arm_init(void)
1720 {
1721         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1722         return rc;
1723 }
1724
1725 module_init(arm_init);