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 #ifdef CONFIG_ARM64
 358 #define __ptrauth_save_key(regs, key)                                           \
 359 ({                                                                              \
 360         regs[key ## KEYLO_EL1] = read_sysreg_s(SYS_ ## key ## KEYLO_EL1);       \
 361         regs[key ## KEYHI_EL1] = read_sysreg_s(SYS_ ## key ## KEYHI_EL1);       \
 362 })
 363 #else
 364 #define  __ptrauth_save_key(regs, key)  do { } while (0)
 365 #endif
 366
 367 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 368 {
 369         int *last_ran;
 370         kvm_host_data_t *cpu_data;
 371
 372         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 373         cpu_data = this_cpu_ptr(&kvm_host_data);
 374
 375         /*
 376          * We might get preempted before the vCPU actually runs, but
 377          * over-invalidation doesn't affect correctness.
 378          */
 379         if (*last_ran != vcpu->vcpu_id) {
 380                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 381                 *last_ran = vcpu->vcpu_id;
 382         }
 383
 384         vcpu->cpu = cpu;
 385         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 386
 387         kvm_arm_set_running_vcpu(vcpu);
 388         kvm_vgic_load(vcpu);
 389         kvm_timer_vcpu_load(vcpu);
 390         kvm_vcpu_load_sysregs(vcpu);
 391         kvm_arch_vcpu_load_fp(vcpu);
 392         kvm_vcpu_pmu_restore_guest(vcpu);
 393
 394         if (single_task_running())
 395                 vcpu_clear_wfe_traps(vcpu);
 396         else
 397                 vcpu_set_wfe_traps(vcpu);
 398
 399         if (vcpu_has_ptrauth(vcpu)) {
 400                 struct kvm_cpu_context __maybe_unused *ctxt = vcpu->arch.host_cpu_context;
 401
 402                 __ptrauth_save_key(ctxt->sys_regs, APIA);
 403                 __ptrauth_save_key(ctxt->sys_regs, APIB);
 404                 __ptrauth_save_key(ctxt->sys_regs, APDA);
 405                 __ptrauth_save_key(ctxt->sys_regs, APDB);
 406                 __ptrauth_save_key(ctxt->sys_regs, APGA);
 407
 408                 vcpu_ptrauth_disable(vcpu);
 409         }
 410 }
 411
 412 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 413 {
 414         kvm_arch_vcpu_put_fp(vcpu);
 415         kvm_vcpu_put_sysregs(vcpu);
 416         kvm_timer_vcpu_put(vcpu);
 417         kvm_vgic_put(vcpu);
 418         kvm_vcpu_pmu_restore_host(vcpu);
 419
 420         vcpu->cpu = -1;
 421
 422         kvm_arm_set_running_vcpu(NULL);
 423 }
 424
 425 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 426 {
 427         vcpu->arch.power_off = true;
 428         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 429         kvm_vcpu_kick(vcpu);
 430 }
 431
 432 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 433                                     struct kvm_mp_state *mp_state)
 434 {
 435         if (vcpu->arch.power_off)
 436                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 437         else
 438                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 439
 440         return 0;
 441 }
 442
 443 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 444                                     struct kvm_mp_state *mp_state)
 445 {
 446         int ret = 0;
 447
 448         switch (mp_state->mp_state) {
 449         case KVM_MP_STATE_RUNNABLE:
 450                 vcpu->arch.power_off = false;
 451                 break;
 452         case KVM_MP_STATE_STOPPED:
 453                 vcpu_power_off(vcpu);
 454                 break;
 455         default:
 456                 ret = -EINVAL;
 457         }
 458
 459         return ret;
 460 }
 461
 462 /**
 463  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 464  * @v:          The VCPU pointer
 465  *
 466  * If the guest CPU is not waiting for interrupts or an interrupt line is
 467  * asserted, the CPU is by definition runnable.
 468  */
 469 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 470 {
 471         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 472         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 473                 && !v->arch.power_off && !v->arch.pause);
 474 }
 475
 476 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 477 {
 478         return vcpu_mode_priv(vcpu);
 479 }
 480
 481 /* Just ensure a guest exit from a particular CPU */
 482 static void exit_vm_noop(void *info)
 483 {
 484 }
 485
 486 void force_vm_exit(const cpumask_t *mask)
 487 {
 488         preempt_disable();
 489         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 490         preempt_enable();
 491 }
 492
 493 /**
 494  * need_new_vmid_gen - check that the VMID is still valid
 495  * @vmid: The VMID to check
 496  *
 497  * return true if there is a new generation of VMIDs being used
 498  *
 499  * The hardware supports a limited set of values with the value zero reserved
 500  * for the host, so we check if an assigned value belongs to a previous
 501  * generation, which which requires us to assign a new value. If we're the
 502  * first to use a VMID for the new generation, we must flush necessary caches
 503  * and TLBs on all CPUs.
 504  */
 505 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 506 {
 507         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 508         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 509         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 510 }
 511
 512 /**
 513  * update_vmid - Update the vmid with a valid VMID for the current generation
 514  * @kvm: The guest that struct vmid belongs to
 515  * @vmid: The stage-2 VMID information struct
 516  */
 517 static void update_vmid(struct kvm_vmid *vmid)
 518 {
 519         if (!need_new_vmid_gen(vmid))
 520                 return;
 521
 522         spin_lock(&kvm_vmid_lock);
 523
 524         /*
 525          * We need to re-check the vmid_gen here to ensure that if another vcpu
 526          * already allocated a valid vmid for this vm, then this vcpu should
 527          * use the same vmid.
 528          */
 529         if (!need_new_vmid_gen(vmid)) {
 530                 spin_unlock(&kvm_vmid_lock);
 531                 return;
 532         }
 533
 534         /* First user of a new VMID generation? */
 535         if (unlikely(kvm_next_vmid == 0)) {
 536                 atomic64_inc(&kvm_vmid_gen);
 537                 kvm_next_vmid = 1;
 538
 539                 /*
 540                  * On SMP we know no other CPUs can use this CPU's or each
 541                  * other's VMID after force_vm_exit returns since the
 542                  * kvm_vmid_lock blocks them from reentry to the guest.
 543                  */
 544                 force_vm_exit(cpu_all_mask);
 545                 /*
 546                  * Now broadcast TLB + ICACHE invalidation over the inner
 547                  * shareable domain to make sure all data structures are
 548                  * clean.
 549                  */
 550                 kvm_call_hyp(__kvm_flush_vm_context);
 551         }
 552
 553         vmid->vmid = kvm_next_vmid;
 554         kvm_next_vmid++;
 555         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 556
 557         smp_wmb();
 558         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 559
 560         spin_unlock(&kvm_vmid_lock);
 561 }
 562
 563 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 564 {
 565         struct kvm *kvm = vcpu->kvm;
 566         int ret = 0;
 567
 568         if (likely(vcpu->arch.has_run_once))
 569                 return 0;
 570
 571         if (!kvm_arm_vcpu_is_finalized(vcpu))
 572                 return -EPERM;
 573
 574         vcpu->arch.has_run_once = true;
 575
 576         if (likely(irqchip_in_kernel(kvm))) {
 577                 /*
 578                  * Map the VGIC hardware resources before running a vcpu the
 579                  * first time on this VM.
 580                  */
 581                 if (unlikely(!vgic_ready(kvm))) {
 582                         ret = kvm_vgic_map_resources(kvm);
 583                         if (ret)
 584                                 return ret;
 585                 }
 586         } else {
 587                 /*
 588                  * Tell the rest of the code that there are userspace irqchip
 589                  * VMs in the wild.
 590                  */
 591                 static_branch_inc(&userspace_irqchip_in_use);
 592         }
 593
 594         ret = kvm_timer_enable(vcpu);
 595         if (ret)
 596                 return ret;
 597
 598         ret = kvm_arm_pmu_v3_enable(vcpu);
 599
 600         return ret;
 601 }
 602
 603 bool kvm_arch_intc_initialized(struct kvm *kvm)
 604 {
 605         return vgic_initialized(kvm);
 606 }
 607
 608 void kvm_arm_halt_guest(struct kvm *kvm)
 609 {
 610         int i;
 611         struct kvm_vcpu *vcpu;
 612
 613         kvm_for_each_vcpu(i, vcpu, kvm)
 614                 vcpu->arch.pause = true;
 615         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 616 }
 617
 618 void kvm_arm_resume_guest(struct kvm *kvm)
 619 {
 620         int i;
 621         struct kvm_vcpu *vcpu;
 622
 623         kvm_for_each_vcpu(i, vcpu, kvm) {
 624                 vcpu->arch.pause = false;
 625                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 626         }
 627 }
 628
 629 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 630 {
 631         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 632
 633         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 634                                        (!vcpu->arch.pause)));
 635
 636         if (vcpu->arch.power_off || vcpu->arch.pause) {
 637                 /* Awaken to handle a signal, request we sleep again later. */
 638                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 639         }
 640
 641         /*
 642          * Make sure we will observe a potential reset request if we've
 643          * observed a change to the power state. Pairs with the smp_wmb() in
 644          * kvm_psci_vcpu_on().
 645          */
 646         smp_rmb();
 647 }
 648
 649 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 650 {
 651         return vcpu->arch.target >= 0;
 652 }
 653
 654 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 655 {
 656         if (kvm_request_pending(vcpu)) {
 657                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 658                         vcpu_req_sleep(vcpu);
 659
 660                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 661                         kvm_reset_vcpu(vcpu);
 662
 663                 /*
 664                  * Clear IRQ_PENDING requests that were made to guarantee
 665                  * that a VCPU sees new virtual interrupts.
 666                  */
 667                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 668         }
 669 }
 670
 671 /**
 672  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 673  * @vcpu:       The VCPU pointer
 674  * @run:        The kvm_run structure pointer used for userspace state exchange
 675  *
 676  * This function is called through the VCPU_RUN ioctl called from user space. It
 677  * will execute VM code in a loop until the time slice for the process is used
 678  * or some emulation is needed from user space in which case the function will
 679  * return with return value 0 and with the kvm_run structure filled in with the
 680  * required data for the requested emulation.
 681  */
 682 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 683 {
 684         int ret;
 685
 686         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 687                 return -ENOEXEC;
 688
 689         ret = kvm_vcpu_first_run_init(vcpu);
 690         if (ret)
 691                 return ret;
 692
 693         if (run->exit_reason == KVM_EXIT_MMIO) {
 694                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 695                 if (ret)
 696                         return ret;
 697         }
 698
 699         if (run->immediate_exit)
 700                 return -EINTR;
 701
 702         vcpu_load(vcpu);
 703
 704         kvm_sigset_activate(vcpu);
 705
 706         ret = 1;
 707         run->exit_reason = KVM_EXIT_UNKNOWN;
 708         while (ret > 0) {
 709                 /*
 710                  * Check conditions before entering the guest
 711                  */
 712                 cond_resched();
 713
 714                 update_vmid(&vcpu->kvm->arch.vmid);
 715
 716                 check_vcpu_requests(vcpu);
 717
 718                 /*
 719                  * Preparing the interrupts to be injected also
 720                  * involves poking the GIC, which must be done in a
 721                  * non-preemptible context.
 722                  */
 723                 preempt_disable();
 724
 725                 kvm_pmu_flush_hwstate(vcpu);
 726
 727                 local_irq_disable();
 728
 729                 kvm_vgic_flush_hwstate(vcpu);
 730
 731                 /*
 732                  * Exit if we have a signal pending so that we can deliver the
 733                  * signal to user space.
 734                  */
 735                 if (signal_pending(current)) {
 736                         ret = -EINTR;
 737                         run->exit_reason = KVM_EXIT_INTR;
 738                 }
 739
 740                 /*
 741                  * If we're using a userspace irqchip, then check if we need
 742                  * to tell a userspace irqchip about timer or PMU level
 743                  * changes and if so, exit to userspace (the actual level
 744                  * state gets updated in kvm_timer_update_run and
 745                  * kvm_pmu_update_run below).
 746                  */
 747                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 748                         if (kvm_timer_should_notify_user(vcpu) ||
 749                             kvm_pmu_should_notify_user(vcpu)) {
 750                                 ret = -EINTR;
 751                                 run->exit_reason = KVM_EXIT_INTR;
 752                         }
 753                 }
 754
 755                 /*
 756                  * Ensure we set mode to IN_GUEST_MODE after we disable
 757                  * interrupts and before the final VCPU requests check.
 758                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 759                  * Documentation/virt/kvm/vcpu-requests.rst
 760                  */
 761                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 762
 763                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 764                     kvm_request_pending(vcpu)) {
 765                         vcpu->mode = OUTSIDE_GUEST_MODE;
 766                         isb(); /* Ensure work in x_flush_hwstate is committed */
 767                         kvm_pmu_sync_hwstate(vcpu);
 768                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 769                                 kvm_timer_sync_hwstate(vcpu);
 770                         kvm_vgic_sync_hwstate(vcpu);
 771                         local_irq_enable();
 772                         preempt_enable();
 773                         continue;
 774                 }
 775
 776                 kvm_arm_setup_debug(vcpu);
 777
 778                 /**************************************************************
 779                  * Enter the guest
 780                  */
 781                 trace_kvm_entry(*vcpu_pc(vcpu));
 782                 guest_enter_irqoff();
 783
 784                 if (has_vhe()) {
 785                         kvm_arm_vhe_guest_enter();
 786                         ret = kvm_vcpu_run_vhe(vcpu);
 787                         kvm_arm_vhe_guest_exit();
 788                 } else {
 789                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 790                 }
 791
 792                 vcpu->mode = OUTSIDE_GUEST_MODE;
 793                 vcpu->stat.exits++;
 794                 /*
 795                  * Back from guest
 796                  *************************************************************/
 797
 798                 kvm_arm_clear_debug(vcpu);
 799
 800                 /*
 801                  * We must sync the PMU state before the vgic state so
 802                  * that the vgic can properly sample the updated state of the
 803                  * interrupt line.
 804                  */
 805                 kvm_pmu_sync_hwstate(vcpu);
 806
 807                 /*
 808                  * Sync the vgic state before syncing the timer state because
 809                  * the timer code needs to know if the virtual timer
 810                  * interrupts are active.
 811                  */
 812                 kvm_vgic_sync_hwstate(vcpu);
 813
 814                 /*
 815                  * Sync the timer hardware state before enabling interrupts as
 816                  * we don't want vtimer interrupts to race with syncing the
 817                  * timer virtual interrupt state.
 818                  */
 819                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 820                         kvm_timer_sync_hwstate(vcpu);
 821
 822                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 823
 824                 /*
 825                  * We may have taken a host interrupt in HYP mode (ie
 826                  * while executing the guest). This interrupt is still
 827                  * pending, as we haven't serviced it yet!
 828                  *
 829                  * We're now back in SVC mode, with interrupts
 830                  * disabled.  Enabling the interrupts now will have
 831                  * the effect of taking the interrupt again, in SVC
 832                  * mode this time.
 833                  */
 834                 local_irq_enable();
 835
 836                 /*
 837                  * We do local_irq_enable() before calling guest_exit() so
 838                  * that if a timer interrupt hits while running the guest we
 839                  * account that tick as being spent in the guest.  We enable
 840                  * preemption after calling guest_exit() so that if we get
 841                  * preempted we make sure ticks after that is not counted as
 842                  * guest time.
 843                  */
 844                 guest_exit();
 845                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 846
 847                 /* Exit types that need handling before we can be preempted */
 848                 handle_exit_early(vcpu, run, ret);
 849
 850                 preempt_enable();
 851
 852                 ret = handle_exit(vcpu, run, ret);
 853         }
 854
 855         /* Tell userspace about in-kernel device output levels */
 856         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 857                 kvm_timer_update_run(vcpu);
 858                 kvm_pmu_update_run(vcpu);
 859         }
 860
 861         kvm_sigset_deactivate(vcpu);
 862
 863         vcpu_put(vcpu);
 864         return ret;
 865 }
 866
 867 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 868 {
 869         int bit_index;
 870         bool set;
 871         unsigned long *hcr;
 872
 873         if (number == KVM_ARM_IRQ_CPU_IRQ)
 874                 bit_index = __ffs(HCR_VI);
 875         else /* KVM_ARM_IRQ_CPU_FIQ */
 876                 bit_index = __ffs(HCR_VF);
 877
 878         hcr = vcpu_hcr(vcpu);
 879         if (level)
 880                 set = test_and_set_bit(bit_index, hcr);
 881         else
 882                 set = test_and_clear_bit(bit_index, hcr);
 883
 884         /*
 885          * If we didn't change anything, no need to wake up or kick other CPUs
 886          */
 887         if (set == level)
 888                 return 0;
 889
 890         /*
 891          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 892          * trigger a world-switch round on the running physical CPU to set the
 893          * virtual IRQ/FIQ fields in the HCR appropriately.
 894          */
 895         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 896         kvm_vcpu_kick(vcpu);
 897
 898         return 0;
 899 }
 900
 901 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 902                           bool line_status)
 903 {
 904         u32 irq = irq_level->irq;
 905         unsigned int irq_type, vcpu_idx, irq_num;
 906         int nrcpus = atomic_read(&kvm->online_vcpus);
 907         struct kvm_vcpu *vcpu = NULL;
 908         bool level = irq_level->level;
 909
 910         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 911         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 912         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 913         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 914
 915         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 916
 917         switch (irq_type) {
 918         case KVM_ARM_IRQ_TYPE_CPU:
 919                 if (irqchip_in_kernel(kvm))
 920                         return -ENXIO;
 921
 922                 if (vcpu_idx >= nrcpus)
 923                         return -EINVAL;
 924
 925                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 926                 if (!vcpu)
 927                         return -EINVAL;
 928
 929                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 930                         return -EINVAL;
 931
 932                 return vcpu_interrupt_line(vcpu, irq_num, level);
 933         case KVM_ARM_IRQ_TYPE_PPI:
 934                 if (!irqchip_in_kernel(kvm))
 935                         return -ENXIO;
 936
 937                 if (vcpu_idx >= nrcpus)
 938                         return -EINVAL;
 939
 940                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 941                 if (!vcpu)
 942                         return -EINVAL;
 943
 944                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 945                         return -EINVAL;
 946
 947                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 948         case KVM_ARM_IRQ_TYPE_SPI:
 949                 if (!irqchip_in_kernel(kvm))
 950                         return -ENXIO;
 951
 952                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 953                         return -EINVAL;
 954
 955                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 956         }
 957
 958         return -EINVAL;
 959 }
 960
 961 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 962                                const struct kvm_vcpu_init *init)
 963 {
 964         unsigned int i, ret;
 965         int phys_target = kvm_target_cpu();
 966
 967         if (init->target != phys_target)
 968                 return -EINVAL;
 969
 970         /*
 971          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 972          * use the same target.
 973          */
 974         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 975                 return -EINVAL;
 976
 977         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 978         for (i = 0; i < sizeof(init->features) * 8; i++) {
 979                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 980
 981                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 982                         return -ENOENT;
 983
 984                 /*
 985                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 986                  * use the same feature set.
 987                  */
 988                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 989                     test_bit(i, vcpu->arch.features) != set)
 990                         return -EINVAL;
 991
 992                 if (set)
 993                         set_bit(i, vcpu->arch.features);
 994         }
 995
 996         vcpu->arch.target = phys_target;
 997
 998         /* Now we know what it is, we can reset it. */
 999         ret = kvm_reset_vcpu(vcpu);
1000         if (ret) {
1001                 vcpu->arch.target = -1;
1002                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1003         }
1004
1005         return ret;
1006 }
1007
1008 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1009                                          struct kvm_vcpu_init *init)
1010 {
1011         int ret;
1012
1013         ret = kvm_vcpu_set_target(vcpu, init);
1014         if (ret)
1015                 return ret;
1016
1017         /*
1018          * Ensure a rebooted VM will fault in RAM pages and detect if the
1019          * guest MMU is turned off and flush the caches as needed.
1020          */
1021         if (vcpu->arch.has_run_once)
1022                 stage2_unmap_vm(vcpu->kvm);
1023
1024         vcpu_reset_hcr(vcpu);
1025
1026         /*
1027          * Handle the "start in power-off" case.
1028          */
1029         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1030                 vcpu_power_off(vcpu);
1031         else
1032                 vcpu->arch.power_off = false;
1033
1034         return 0;
1035 }
1036
1037 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1038                                  struct kvm_device_attr *attr)
1039 {
1040         int ret = -ENXIO;
1041
1042         switch (attr->group) {
1043         default:
1044                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1045                 break;
1046         }
1047
1048         return ret;
1049 }
1050
1051 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1052                                  struct kvm_device_attr *attr)
1053 {
1054         int ret = -ENXIO;
1055
1056         switch (attr->group) {
1057         default:
1058                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1059                 break;
1060         }
1061
1062         return ret;
1063 }
1064
1065 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1066                                  struct kvm_device_attr *attr)
1067 {
1068         int ret = -ENXIO;
1069
1070         switch (attr->group) {
1071         default:
1072                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1073                 break;
1074         }
1075
1076         return ret;
1077 }
1078
1079 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1080                                    struct kvm_vcpu_events *events)
1081 {
1082         memset(events, 0, sizeof(*events));
1083
1084         return __kvm_arm_vcpu_get_events(vcpu, events);
1085 }
1086
1087 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1088                                    struct kvm_vcpu_events *events)
1089 {
1090         int i;
1091
1092         /* check whether the reserved field is zero */
1093         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1094                 if (events->reserved[i])
1095                         return -EINVAL;
1096
1097         /* check whether the pad field is zero */
1098         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1099                 if (events->exception.pad[i])
1100                         return -EINVAL;
1101
1102         return __kvm_arm_vcpu_set_events(vcpu, events);
1103 }
1104
1105 long kvm_arch_vcpu_ioctl(struct file *filp,
1106                          unsigned int ioctl, unsigned long arg)
1107 {
1108         struct kvm_vcpu *vcpu = filp->private_data;
1109         void __user *argp = (void __user *)arg;
1110         struct kvm_device_attr attr;
1111         long r;
1112
1113         switch (ioctl) {
1114         case KVM_ARM_VCPU_INIT: {
1115                 struct kvm_vcpu_init init;
1116
1117                 r = -EFAULT;
1118                 if (copy_from_user(&init, argp, sizeof(init)))
1119                         break;
1120
1121                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1122                 break;
1123         }
1124         case KVM_SET_ONE_REG:
1125         case KVM_GET_ONE_REG: {
1126                 struct kvm_one_reg reg;
1127
1128                 r = -ENOEXEC;
1129                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1130                         break;
1131
1132                 r = -EFAULT;
1133                 if (copy_from_user(&reg, argp, sizeof(reg)))
1134                         break;
1135
1136                 if (ioctl == KVM_SET_ONE_REG)
1137                         r = kvm_arm_set_reg(vcpu, &reg);
1138                 else
1139                         r = kvm_arm_get_reg(vcpu, &reg);
1140                 break;
1141         }
1142         case KVM_GET_REG_LIST: {
1143                 struct kvm_reg_list __user *user_list = argp;
1144                 struct kvm_reg_list reg_list;
1145                 unsigned n;
1146
1147                 r = -ENOEXEC;
1148                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1149                         break;
1150
1151                 r = -EPERM;
1152                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1153                         break;
1154
1155                 r = -EFAULT;
1156                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1157                         break;
1158                 n = reg_list.n;
1159                 reg_list.n = kvm_arm_num_regs(vcpu);
1160                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1161                         break;
1162                 r = -E2BIG;
1163                 if (n < reg_list.n)
1164                         break;
1165                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1166                 break;
1167         }
1168         case KVM_SET_DEVICE_ATTR: {
1169                 r = -EFAULT;
1170                 if (copy_from_user(&attr, argp, sizeof(attr)))
1171                         break;
1172                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1173                 break;
1174         }
1175         case KVM_GET_DEVICE_ATTR: {
1176                 r = -EFAULT;
1177                 if (copy_from_user(&attr, argp, sizeof(attr)))
1178                         break;
1179                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1180                 break;
1181         }
1182         case KVM_HAS_DEVICE_ATTR: {
1183                 r = -EFAULT;
1184                 if (copy_from_user(&attr, argp, sizeof(attr)))
1185                         break;
1186                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1187                 break;
1188         }
1189         case KVM_GET_VCPU_EVENTS: {
1190                 struct kvm_vcpu_events events;
1191
1192                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1193                         return -EINVAL;
1194
1195                 if (copy_to_user(argp, &events, sizeof(events)))
1196                         return -EFAULT;
1197
1198                 return 0;
1199         }
1200         case KVM_SET_VCPU_EVENTS: {
1201                 struct kvm_vcpu_events events;
1202
1203                 if (copy_from_user(&events, argp, sizeof(events)))
1204                         return -EFAULT;
1205
1206                 return kvm_arm_vcpu_set_events(vcpu, &events);
1207         }
1208         case KVM_ARM_VCPU_FINALIZE: {
1209                 int what;
1210
1211                 if (!kvm_vcpu_initialized(vcpu))
1212                         return -ENOEXEC;
1213
1214                 if (get_user(what, (const int __user *)argp))
1215                         return -EFAULT;
1216
1217                 return kvm_arm_vcpu_finalize(vcpu, what);
1218         }
1219         default:
1220                 r = -EINVAL;
1221         }
1222
1223         return r;
1224 }
1225
1226 /**
1227  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1228  * @kvm: kvm instance
1229  * @log: slot id and address to which we copy the log
1230  *
1231  * Steps 1-4 below provide general overview of dirty page logging. See
1232  * kvm_get_dirty_log_protect() function description for additional details.
1233  *
1234  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1235  * always flush the TLB (step 4) even if previous step failed  and the dirty
1236  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1237  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1238  * writes will be marked dirty for next log read.
1239  *
1240  *   1. Take a snapshot of the bit and clear it if needed.
1241  *   2. Write protect the corresponding page.
1242  *   3. Copy the snapshot to the userspace.
1243  *   4. Flush TLB's if needed.
1244  */
1245 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1246 {
1247         bool flush = false;
1248         int r;
1249
1250         mutex_lock(&kvm->slots_lock);
1251
1252         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1253
1254         if (flush)
1255                 kvm_flush_remote_tlbs(kvm);
1256
1257         mutex_unlock(&kvm->slots_lock);
1258         return r;
1259 }
1260
1261 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1262 {
1263         bool flush = false;
1264         int r;
1265
1266         mutex_lock(&kvm->slots_lock);
1267
1268         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1269
1270         if (flush)
1271                 kvm_flush_remote_tlbs(kvm);
1272
1273         mutex_unlock(&kvm->slots_lock);
1274         return r;
1275 }
1276
1277 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1278                                         struct kvm_arm_device_addr *dev_addr)
1279 {
1280         unsigned long dev_id, type;
1281
1282         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1283                 KVM_ARM_DEVICE_ID_SHIFT;
1284         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1285                 KVM_ARM_DEVICE_TYPE_SHIFT;
1286
1287         switch (dev_id) {
1288         case KVM_ARM_DEVICE_VGIC_V2:
1289                 if (!vgic_present)
1290                         return -ENXIO;
1291                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1292         default:
1293                 return -ENODEV;
1294         }
1295 }
1296
1297 long kvm_arch_vm_ioctl(struct file *filp,
1298                        unsigned int ioctl, unsigned long arg)
1299 {
1300         struct kvm *kvm = filp->private_data;
1301         void __user *argp = (void __user *)arg;
1302
1303         switch (ioctl) {
1304         case KVM_CREATE_IRQCHIP: {
1305                 int ret;
1306                 if (!vgic_present)
1307                         return -ENXIO;
1308                 mutex_lock(&kvm->lock);
1309                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1310                 mutex_unlock(&kvm->lock);
1311                 return ret;
1312         }
1313         case KVM_ARM_SET_DEVICE_ADDR: {
1314                 struct kvm_arm_device_addr dev_addr;
1315
1316                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1317                         return -EFAULT;
1318                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1319         }
1320         case KVM_ARM_PREFERRED_TARGET: {
1321                 int err;
1322                 struct kvm_vcpu_init init;
1323
1324                 err = kvm_vcpu_preferred_target(&init);
1325                 if (err)
1326                         return err;
1327
1328                 if (copy_to_user(argp, &init, sizeof(init)))
1329                         return -EFAULT;
1330
1331                 return 0;
1332         }
1333         default:
1334                 return -EINVAL;
1335         }
1336 }
1337
1338 static void cpu_init_hyp_mode(void *dummy)
1339 {
1340         phys_addr_t pgd_ptr;
1341         unsigned long hyp_stack_ptr;
1342         unsigned long stack_page;
1343         unsigned long vector_ptr;
1344
1345         /* Switch from the HYP stub to our own HYP init vector */
1346         __hyp_set_vectors(kvm_get_idmap_vector());
1347
1348         pgd_ptr = kvm_mmu_get_httbr();
1349         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1350         hyp_stack_ptr = stack_page + PAGE_SIZE;
1351         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1352
1353         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1354         __cpu_init_stage2();
1355 }
1356
1357 static void cpu_hyp_reset(void)
1358 {
1359         if (!is_kernel_in_hyp_mode())
1360                 __hyp_reset_vectors();
1361 }
1362
1363 static void cpu_hyp_reinit(void)
1364 {
1365         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1366
1367         cpu_hyp_reset();
1368
1369         if (is_kernel_in_hyp_mode())
1370                 kvm_timer_init_vhe();
1371         else
1372                 cpu_init_hyp_mode(NULL);
1373
1374         kvm_arm_init_debug();
1375
1376         if (vgic_present)
1377                 kvm_vgic_init_cpu_hardware();
1378 }
1379
1380 static void _kvm_arch_hardware_enable(void *discard)
1381 {
1382         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1383                 cpu_hyp_reinit();
1384                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1385         }
1386 }
1387
1388 int kvm_arch_hardware_enable(void)
1389 {
1390         _kvm_arch_hardware_enable(NULL);
1391         return 0;
1392 }
1393
1394 static void _kvm_arch_hardware_disable(void *discard)
1395 {
1396         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1397                 cpu_hyp_reset();
1398                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1399         }
1400 }
1401
1402 void kvm_arch_hardware_disable(void)
1403 {
1404         _kvm_arch_hardware_disable(NULL);
1405 }
1406
1407 #ifdef CONFIG_CPU_PM
1408 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1409                                     unsigned long cmd,
1410                                     void *v)
1411 {
1412         /*
1413          * kvm_arm_hardware_enabled is left with its old value over
1414          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1415          * re-enable hyp.
1416          */
1417         switch (cmd) {
1418         case CPU_PM_ENTER:
1419                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1420                         /*
1421                          * don't update kvm_arm_hardware_enabled here
1422                          * so that the hardware will be re-enabled
1423                          * when we resume. See below.
1424                          */
1425                         cpu_hyp_reset();
1426
1427                 return NOTIFY_OK;
1428         case CPU_PM_ENTER_FAILED:
1429         case CPU_PM_EXIT:
1430                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1431                         /* The hardware was enabled before suspend. */
1432                         cpu_hyp_reinit();
1433
1434                 return NOTIFY_OK;
1435
1436         default:
1437                 return NOTIFY_DONE;
1438         }
1439 }
1440
1441 static struct notifier_block hyp_init_cpu_pm_nb = {
1442         .notifier_call = hyp_init_cpu_pm_notifier,
1443 };
1444
1445 static void __init hyp_cpu_pm_init(void)
1446 {
1447         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1448 }
1449 static void __init hyp_cpu_pm_exit(void)
1450 {
1451         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1452 }
1453 #else
1454 static inline void hyp_cpu_pm_init(void)
1455 {
1456 }
1457 static inline void hyp_cpu_pm_exit(void)
1458 {
1459 }
1460 #endif
1461
1462 static int init_common_resources(void)
1463 {
1464         kvm_set_ipa_limit();
1465
1466         return 0;
1467 }
1468
1469 static int init_subsystems(void)
1470 {
1471         int err = 0;
1472
1473         /*
1474          * Enable hardware so that subsystem initialisation can access EL2.
1475          */
1476         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1477
1478         /*
1479          * Register CPU lower-power notifier
1480          */
1481         hyp_cpu_pm_init();
1482
1483         /*
1484          * Init HYP view of VGIC
1485          */
1486         err = kvm_vgic_hyp_init();
1487         switch (err) {
1488         case 0:
1489                 vgic_present = true;
1490                 break;
1491         case -ENODEV:
1492         case -ENXIO:
1493                 vgic_present = false;
1494                 err = 0;
1495                 break;
1496         default:
1497                 goto out;
1498         }
1499
1500         /*
1501          * Init HYP architected timer support
1502          */
1503         err = kvm_timer_hyp_init(vgic_present);
1504         if (err)
1505                 goto out;
1506
1507         kvm_perf_init();
1508         kvm_coproc_table_init();
1509
1510 out:
1511         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1512
1513         return err;
1514 }
1515
1516 static void teardown_hyp_mode(void)
1517 {
1518         int cpu;
1519
1520         free_hyp_pgds();
1521         for_each_possible_cpu(cpu)
1522                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1523         hyp_cpu_pm_exit();
1524 }
1525
1526 /**
1527  * Inits Hyp-mode on all online CPUs
1528  */
1529 static int init_hyp_mode(void)
1530 {
1531         int cpu;
1532         int err = 0;
1533
1534         /*
1535          * Allocate Hyp PGD and setup Hyp identity mapping
1536          */
1537         err = kvm_mmu_init();
1538         if (err)
1539                 goto out_err;
1540
1541         /*
1542          * Allocate stack pages for Hypervisor-mode
1543          */
1544         for_each_possible_cpu(cpu) {
1545                 unsigned long stack_page;
1546
1547                 stack_page = __get_free_page(GFP_KERNEL);
1548                 if (!stack_page) {
1549                         err = -ENOMEM;
1550                         goto out_err;
1551                 }
1552
1553                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1554         }
1555
1556         /*
1557          * Map the Hyp-code called directly from the host
1558          */
1559         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1560                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1561         if (err) {
1562                 kvm_err("Cannot map world-switch code\n");
1563                 goto out_err;
1564         }
1565
1566         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1567                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1568         if (err) {
1569                 kvm_err("Cannot map rodata section\n");
1570                 goto out_err;
1571         }
1572
1573         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1574                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1575         if (err) {
1576                 kvm_err("Cannot map bss section\n");
1577                 goto out_err;
1578         }
1579
1580         err = kvm_map_vectors();
1581         if (err) {
1582                 kvm_err("Cannot map vectors\n");
1583                 goto out_err;
1584         }
1585
1586         /*
1587          * Map the Hyp stack pages
1588          */
1589         for_each_possible_cpu(cpu) {
1590                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1591                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1592                                           PAGE_HYP);
1593
1594                 if (err) {
1595                         kvm_err("Cannot map hyp stack\n");
1596                         goto out_err;
1597                 }
1598         }
1599
1600         for_each_possible_cpu(cpu) {
1601                 kvm_host_data_t *cpu_data;
1602
1603                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1604                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1605
1606                 if (err) {
1607                         kvm_err("Cannot map host CPU state: %d\n", err);
1608                         goto out_err;
1609                 }
1610         }
1611
1612         err = hyp_map_aux_data();
1613         if (err)
1614                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1615
1616         return 0;
1617
1618 out_err:
1619         teardown_hyp_mode();
1620         kvm_err("error initializing Hyp mode: %d\n", err);
1621         return err;
1622 }
1623
1624 static void check_kvm_target_cpu(void *ret)
1625 {
1626         *(int *)ret = kvm_target_cpu();
1627 }
1628
1629 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1630 {
1631         struct kvm_vcpu *vcpu;
1632         int i;
1633
1634         mpidr &= MPIDR_HWID_BITMASK;
1635         kvm_for_each_vcpu(i, vcpu, kvm) {
1636                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1637                         return vcpu;
1638         }
1639         return NULL;
1640 }
1641
1642 bool kvm_arch_has_irq_bypass(void)
1643 {
1644         return true;
1645 }
1646
1647 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1648                                       struct irq_bypass_producer *prod)
1649 {
1650         struct kvm_kernel_irqfd *irqfd =
1651                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1652
1653         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1654                                           &irqfd->irq_entry);
1655 }
1656 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1657                                       struct irq_bypass_producer *prod)
1658 {
1659         struct kvm_kernel_irqfd *irqfd =
1660                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1661
1662         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1663                                      &irqfd->irq_entry);
1664 }
1665
1666 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1667 {
1668         struct kvm_kernel_irqfd *irqfd =
1669                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1670
1671         kvm_arm_halt_guest(irqfd->kvm);
1672 }
1673
1674 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1675 {
1676         struct kvm_kernel_irqfd *irqfd =
1677                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1678
1679         kvm_arm_resume_guest(irqfd->kvm);
1680 }
1681
1682 /**
1683  * Initialize Hyp-mode and memory mappings on all CPUs.
1684  */
1685 int kvm_arch_init(void *opaque)
1686 {
1687         int err;
1688         int ret, cpu;
1689         bool in_hyp_mode;
1690
1691         if (!is_hyp_mode_available()) {
1692                 kvm_info("HYP mode not available\n");
1693                 return -ENODEV;
1694         }
1695
1696         in_hyp_mode = is_kernel_in_hyp_mode();
1697
1698         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1699                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1700                 return -ENODEV;
1701         }
1702
1703         for_each_online_cpu(cpu) {
1704                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1705                 if (ret < 0) {
1706                         kvm_err("Error, CPU %d not supported!\n", cpu);
1707                         return -ENODEV;
1708                 }
1709         }
1710
1711         err = init_common_resources();
1712         if (err)
1713                 return err;
1714
1715         err = kvm_arm_init_sve();
1716         if (err)
1717                 return err;
1718
1719         if (!in_hyp_mode) {
1720                 err = init_hyp_mode();
1721                 if (err)
1722                         goto out_err;
1723         }
1724
1725         err = init_subsystems();
1726         if (err)
1727                 goto out_hyp;
1728
1729         if (in_hyp_mode)
1730                 kvm_info("VHE mode initialized successfully\n");
1731         else
1732                 kvm_info("Hyp mode initialized successfully\n");
1733
1734         return 0;
1735
1736 out_hyp:
1737         if (!in_hyp_mode)
1738                 teardown_hyp_mode();
1739 out_err:
1740         return err;
1741 }
1742
1743 /* NOP: Compiling as a module not supported */
1744 void kvm_arch_exit(void)
1745 {
1746         kvm_perf_teardown();
1747 }
1748
1749 static int arm_init(void)
1750 {
1751         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1752         return rc;
1753 }
1754
1755 module_init(arm_init);