virt/kvm/arm/arm.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   4  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   5  */
   6
   7 #include <linux/bug.h>
   8 #include <linux/cpu_pm.h>
   9 #include <linux/errno.h>
  10 #include <linux/err.h>
  11 #include <linux/kvm_host.h>
  12 #include <linux/list.h>
  13 #include <linux/module.h>
  14 #include <linux/vmalloc.h>
  15 #include <linux/fs.h>
  16 #include <linux/mman.h>
  17 #include <linux/sched.h>
  18 #include <linux/kvm.h>
  19 #include <linux/kvm_irqfd.h>
  20 #include <linux/irqbypass.h>
  21 #include <linux/sched/stat.h>
  22 #include <trace/events/kvm.h>
  23 #include <kvm/arm_pmu.h>
  24 #include <kvm/arm_psci.h>
  25
  26 #define CREATE_TRACE_POINTS
  27 #include "trace.h"
  28
  29 #include <linux/uaccess.h>
  30 #include <asm/ptrace.h>
  31 #include <asm/mman.h>
  32 #include <asm/tlbflush.h>
  33 #include <asm/cacheflush.h>
  34 #include <asm/cpufeature.h>
  35 #include <asm/virt.h>
  36 #include <asm/kvm_arm.h>
  37 #include <asm/kvm_asm.h>
  38 #include <asm/kvm_mmu.h>
  39 #include <asm/kvm_emulate.h>
  40 #include <asm/kvm_coproc.h>
  41 #include <asm/sections.h>
  42
  43 #include <kvm/arm_hypercalls.h>
  44 #include <kvm/arm_pmu.h>
  45 #include <kvm/arm_psci.h>
  46
  47 #ifdef REQUIRES_VIRT
  48 __asm__(".arch_extension        virt");
  49 #endif
  50
  51 DEFINE_PER_CPU(kvm_host_data_t, kvm_host_data);
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53
  54 /* Per-CPU variable containing the currently running vcpu. */
  55 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  56
  57 /* The VMID used in the VTTBR */
  58 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  59 static u32 kvm_next_vmid;
  60 static DEFINE_SPINLOCK(kvm_vmid_lock);
  61
  62 static bool vgic_present;
  63
  64 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  69 }
  70
  71 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  72
  73 /**
  74  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  75  * Must be called from non-preemptible context
  76  */
  77 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  78 {
  79         return __this_cpu_read(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  91 {
  92         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  93 }
  94
  95 int kvm_arch_hardware_setup(void)
  96 {
  97         return 0;
  98 }
  99
 100 int kvm_arch_check_processor_compat(void)
 101 {
 102         return 0;
 103 }
 104
 105 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
 106                             struct kvm_enable_cap *cap)
 107 {
 108         int r;
 109
 110         if (cap->flags)
 111                 return -EINVAL;
 112
 113         switch (cap->cap) {
 114         case KVM_CAP_ARM_NISV_TO_USER:
 115                 r = 0;
 116                 kvm->arch.return_nisv_io_abort_to_user = true;
 117                 break;
 118         default:
 119                 r = -EINVAL;
 120                 break;
 121         }
 122
 123         return r;
 124 }
 125
 126 /**
 127  * kvm_arch_init_vm - initializes a VM data structure
 128  * @kvm:        pointer to the KVM struct
 129  */
 130 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 131 {
 132         int ret, cpu;
 133
 134         ret = kvm_arm_setup_stage2(kvm, type);
 135         if (ret)
 136                 return ret;
 137
 138         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 139         if (!kvm->arch.last_vcpu_ran)
 140                 return -ENOMEM;
 141
 142         for_each_possible_cpu(cpu)
 143                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 144
 145         ret = kvm_alloc_stage2_pgd(kvm);
 146         if (ret)
 147                 goto out_fail_alloc;
 148
 149         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 150         if (ret)
 151                 goto out_free_stage2_pgd;
 152
 153         kvm_vgic_early_init(kvm);
 154
 155         /* Mark the initial VMID generation invalid */
 156         kvm->arch.vmid.vmid_gen = 0;
 157
 158         /* The maximum number of VCPUs is limited by the host's GIC model */
 159         kvm->arch.max_vcpus = vgic_present ?
 160                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 161
 162         return ret;
 163 out_free_stage2_pgd:
 164         kvm_free_stage2_pgd(kvm);
 165 out_fail_alloc:
 166         free_percpu(kvm->arch.last_vcpu_ran);
 167         kvm->arch.last_vcpu_ran = NULL;
 168         return ret;
 169 }
 170
 171 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 172 {
 173         return 0;
 174 }
 175
 176 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 177 {
 178         return VM_FAULT_SIGBUS;
 179 }
 180
 181
 182 /**
 183  * kvm_arch_destroy_vm - destroy the VM data structure
 184  * @kvm:        pointer to the KVM struct
 185  */
 186 void kvm_arch_destroy_vm(struct kvm *kvm)
 187 {
 188         int i;
 189
 190         kvm_vgic_destroy(kvm);
 191
 192         free_percpu(kvm->arch.last_vcpu_ran);
 193         kvm->arch.last_vcpu_ran = NULL;
 194
 195         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 196                 if (kvm->vcpus[i]) {
 197                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 198                         kvm->vcpus[i] = NULL;
 199                 }
 200         }
 201         atomic_set(&kvm->online_vcpus, 0);
 202 }
 203
 204 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 205 {
 206         int r;
 207         switch (ext) {
 208         case KVM_CAP_IRQCHIP:
 209                 r = vgic_present;
 210                 break;
 211         case KVM_CAP_IOEVENTFD:
 212         case KVM_CAP_DEVICE_CTRL:
 213         case KVM_CAP_USER_MEMORY:
 214         case KVM_CAP_SYNC_MMU:
 215         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 216         case KVM_CAP_ONE_REG:
 217         case KVM_CAP_ARM_PSCI:
 218         case KVM_CAP_ARM_PSCI_0_2:
 219         case KVM_CAP_READONLY_MEM:
 220         case KVM_CAP_MP_STATE:
 221         case KVM_CAP_IMMEDIATE_EXIT:
 222         case KVM_CAP_VCPU_EVENTS:
 223         case KVM_CAP_ARM_IRQ_LINE_LAYOUT_2:
 224         case KVM_CAP_ARM_NISV_TO_USER:
 225         case KVM_CAP_ARM_INJECT_EXT_DABT:
 226                 r = 1;
 227                 break;
 228         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 229                 r = 1;
 230                 break;
 231         case KVM_CAP_NR_VCPUS:
 232                 r = num_online_cpus();
 233                 break;
 234         case KVM_CAP_MAX_VCPUS:
 235                 r = KVM_MAX_VCPUS;
 236                 break;
 237         case KVM_CAP_MAX_VCPU_ID:
 238                 r = KVM_MAX_VCPU_ID;
 239                 break;
 240         case KVM_CAP_MSI_DEVID:
 241                 if (!kvm)
 242                         r = -EINVAL;
 243                 else
 244                         r = kvm->arch.vgic.msis_require_devid;
 245                 break;
 246         case KVM_CAP_ARM_USER_IRQ:
 247                 /*
 248                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 249                  * (bump this number if adding more devices)
 250                  */
 251                 r = 1;
 252                 break;
 253         default:
 254                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 255                 break;
 256         }
 257         return r;
 258 }
 259
 260 long kvm_arch_dev_ioctl(struct file *filp,
 261                         unsigned int ioctl, unsigned long arg)
 262 {
 263         return -EINVAL;
 264 }
 265
 266 struct kvm *kvm_arch_alloc_vm(void)
 267 {
 268         if (!has_vhe())
 269                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 270
 271         return vzalloc(sizeof(struct kvm));
 272 }
 273
 274 void kvm_arch_free_vm(struct kvm *kvm)
 275 {
 276         if (!has_vhe())
 277                 kfree(kvm);
 278         else
 279                 vfree(kvm);
 280 }
 281
 282 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 283 {
 284         int err;
 285         struct kvm_vcpu *vcpu;
 286
 287         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 288                 err = -EBUSY;
 289                 goto out;
 290         }
 291
 292         if (id >= kvm->arch.max_vcpus) {
 293                 err = -EINVAL;
 294                 goto out;
 295         }
 296
 297         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 298         if (!vcpu) {
 299                 err = -ENOMEM;
 300                 goto out;
 301         }
 302
 303         err = kvm_vcpu_init(vcpu, kvm, id);
 304         if (err)
 305                 goto free_vcpu;
 306
 307         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 308         if (err)
 309                 goto vcpu_uninit;
 310
 311         return vcpu;
 312 vcpu_uninit:
 313         kvm_vcpu_uninit(vcpu);
 314 free_vcpu:
 315         kmem_cache_free(kvm_vcpu_cache, vcpu);
 316 out:
 317         return ERR_PTR(err);
 318 }
 319
 320 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 321 {
 322 }
 323
 324 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 325 {
 326         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 327                 static_branch_dec(&userspace_irqchip_in_use);
 328
 329         kvm_mmu_free_memory_caches(vcpu);
 330         kvm_timer_vcpu_terminate(vcpu);
 331         kvm_pmu_vcpu_destroy(vcpu);
 332         kvm_vcpu_uninit(vcpu);
 333         kmem_cache_free(kvm_vcpu_cache, vcpu);
 334 }
 335
 336 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 337 {
 338         kvm_arch_vcpu_free(vcpu);
 339 }
 340
 341 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 342 {
 343         return kvm_timer_is_pending(vcpu);
 344 }
 345
 346 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 347 {
 348         /*
 349          * If we're about to block (most likely because we've just hit a
 350          * WFI), we need to sync back the state of the GIC CPU interface
 351          * so that we have the lastest PMR and group enables. This ensures
 352          * that kvm_arch_vcpu_runnable has up-to-date data to decide
 353          * whether we have pending interrupts.
 354          */
 355         preempt_disable();
 356         kvm_vgic_vmcr_sync(vcpu);
 357         preempt_enable();
 358
 359         kvm_vgic_v4_enable_doorbell(vcpu);
 360 }
 361
 362 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 363 {
 364         kvm_vgic_v4_disable_doorbell(vcpu);
 365 }
 366
 367 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 368 {
 369         /* Force users to call KVM_ARM_VCPU_INIT */
 370         vcpu->arch.target = -1;
 371         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 372
 373         /* Set up the timer */
 374         kvm_timer_vcpu_init(vcpu);
 375
 376         kvm_pmu_vcpu_init(vcpu);
 377
 378         kvm_arm_reset_debug_ptr(vcpu);
 379
 380         kvm_arm_pvtime_vcpu_init(&vcpu->arch);
 381
 382         return kvm_vgic_vcpu_init(vcpu);
 383 }
 384
 385 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 386 {
 387         int *last_ran;
 388         kvm_host_data_t *cpu_data;
 389
 390         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 391         cpu_data = this_cpu_ptr(&kvm_host_data);
 392
 393         /*
 394          * We might get preempted before the vCPU actually runs, but
 395          * over-invalidation doesn't affect correctness.
 396          */
 397         if (*last_ran != vcpu->vcpu_id) {
 398                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 399                 *last_ran = vcpu->vcpu_id;
 400         }
 401
 402         vcpu->cpu = cpu;
 403         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 404
 405         kvm_arm_set_running_vcpu(vcpu);
 406         kvm_vgic_load(vcpu);
 407         kvm_timer_vcpu_load(vcpu);
 408         kvm_vcpu_load_sysregs(vcpu);
 409         kvm_arch_vcpu_load_fp(vcpu);
 410         kvm_vcpu_pmu_restore_guest(vcpu);
 411         if (kvm_arm_is_pvtime_enabled(&vcpu->arch))
 412                 kvm_make_request(KVM_REQ_RECORD_STEAL, vcpu);
 413
 414         if (single_task_running())
 415                 vcpu_clear_wfe_traps(vcpu);
 416         else
 417                 vcpu_set_wfe_traps(vcpu);
 418
 419         vcpu_ptrauth_setup_lazy(vcpu);
 420 }
 421
 422 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 423 {
 424         kvm_arch_vcpu_put_fp(vcpu);
 425         kvm_vcpu_put_sysregs(vcpu);
 426         kvm_timer_vcpu_put(vcpu);
 427         kvm_vgic_put(vcpu);
 428         kvm_vcpu_pmu_restore_host(vcpu);
 429
 430         vcpu->cpu = -1;
 431
 432         kvm_arm_set_running_vcpu(NULL);
 433 }
 434
 435 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 436 {
 437         vcpu->arch.power_off = true;
 438         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 439         kvm_vcpu_kick(vcpu);
 440 }
 441
 442 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 443                                     struct kvm_mp_state *mp_state)
 444 {
 445         if (vcpu->arch.power_off)
 446                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 447         else
 448                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 449
 450         return 0;
 451 }
 452
 453 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 454                                     struct kvm_mp_state *mp_state)
 455 {
 456         int ret = 0;
 457
 458         switch (mp_state->mp_state) {
 459         case KVM_MP_STATE_RUNNABLE:
 460                 vcpu->arch.power_off = false;
 461                 break;
 462         case KVM_MP_STATE_STOPPED:
 463                 vcpu_power_off(vcpu);
 464                 break;
 465         default:
 466                 ret = -EINVAL;
 467         }
 468
 469         return ret;
 470 }
 471
 472 /**
 473  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 474  * @v:          The VCPU pointer
 475  *
 476  * If the guest CPU is not waiting for interrupts or an interrupt line is
 477  * asserted, the CPU is by definition runnable.
 478  */
 479 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 480 {
 481         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 482         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 483                 && !v->arch.power_off && !v->arch.pause);
 484 }
 485
 486 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 487 {
 488         return vcpu_mode_priv(vcpu);
 489 }
 490
 491 /* Just ensure a guest exit from a particular CPU */
 492 static void exit_vm_noop(void *info)
 493 {
 494 }
 495
 496 void force_vm_exit(const cpumask_t *mask)
 497 {
 498         preempt_disable();
 499         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 500         preempt_enable();
 501 }
 502
 503 /**
 504  * need_new_vmid_gen - check that the VMID is still valid
 505  * @vmid: The VMID to check
 506  *
 507  * return true if there is a new generation of VMIDs being used
 508  *
 509  * The hardware supports a limited set of values with the value zero reserved
 510  * for the host, so we check if an assigned value belongs to a previous
 511  * generation, which which requires us to assign a new value. If we're the
 512  * first to use a VMID for the new generation, we must flush necessary caches
 513  * and TLBs on all CPUs.
 514  */
 515 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 516 {
 517         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 518         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 519         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 520 }
 521
 522 /**
 523  * update_vmid - Update the vmid with a valid VMID for the current generation
 524  * @kvm: The guest that struct vmid belongs to
 525  * @vmid: The stage-2 VMID information struct
 526  */
 527 static void update_vmid(struct kvm_vmid *vmid)
 528 {
 529         if (!need_new_vmid_gen(vmid))
 530                 return;
 531
 532         spin_lock(&kvm_vmid_lock);
 533
 534         /*
 535          * We need to re-check the vmid_gen here to ensure that if another vcpu
 536          * already allocated a valid vmid for this vm, then this vcpu should
 537          * use the same vmid.
 538          */
 539         if (!need_new_vmid_gen(vmid)) {
 540                 spin_unlock(&kvm_vmid_lock);
 541                 return;
 542         }
 543
 544         /* First user of a new VMID generation? */
 545         if (unlikely(kvm_next_vmid == 0)) {
 546                 atomic64_inc(&kvm_vmid_gen);
 547                 kvm_next_vmid = 1;
 548
 549                 /*
 550                  * On SMP we know no other CPUs can use this CPU's or each
 551                  * other's VMID after force_vm_exit returns since the
 552                  * kvm_vmid_lock blocks them from reentry to the guest.
 553                  */
 554                 force_vm_exit(cpu_all_mask);
 555                 /*
 556                  * Now broadcast TLB + ICACHE invalidation over the inner
 557                  * shareable domain to make sure all data structures are
 558                  * clean.
 559                  */
 560                 kvm_call_hyp(__kvm_flush_vm_context);
 561         }
 562
 563         vmid->vmid = kvm_next_vmid;
 564         kvm_next_vmid++;
 565         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 566
 567         smp_wmb();
 568         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 569
 570         spin_unlock(&kvm_vmid_lock);
 571 }
 572
 573 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 574 {
 575         struct kvm *kvm = vcpu->kvm;
 576         int ret = 0;
 577
 578         if (likely(vcpu->arch.has_run_once))
 579                 return 0;
 580
 581         if (!kvm_arm_vcpu_is_finalized(vcpu))
 582                 return -EPERM;
 583
 584         vcpu->arch.has_run_once = true;
 585
 586         if (likely(irqchip_in_kernel(kvm))) {
 587                 /*
 588                  * Map the VGIC hardware resources before running a vcpu the
 589                  * first time on this VM.
 590                  */
 591                 if (unlikely(!vgic_ready(kvm))) {
 592                         ret = kvm_vgic_map_resources(kvm);
 593                         if (ret)
 594                                 return ret;
 595                 }
 596         } else {
 597                 /*
 598                  * Tell the rest of the code that there are userspace irqchip
 599                  * VMs in the wild.
 600                  */
 601                 static_branch_inc(&userspace_irqchip_in_use);
 602         }
 603
 604         ret = kvm_timer_enable(vcpu);
 605         if (ret)
 606                 return ret;
 607
 608         ret = kvm_arm_pmu_v3_enable(vcpu);
 609
 610         return ret;
 611 }
 612
 613 bool kvm_arch_intc_initialized(struct kvm *kvm)
 614 {
 615         return vgic_initialized(kvm);
 616 }
 617
 618 void kvm_arm_halt_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 = true;
 625         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 626 }
 627
 628 void kvm_arm_resume_guest(struct kvm *kvm)
 629 {
 630         int i;
 631         struct kvm_vcpu *vcpu;
 632
 633         kvm_for_each_vcpu(i, vcpu, kvm) {
 634                 vcpu->arch.pause = false;
 635                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 636         }
 637 }
 638
 639 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 640 {
 641         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 642
 643         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 644                                        (!vcpu->arch.pause)));
 645
 646         if (vcpu->arch.power_off || vcpu->arch.pause) {
 647                 /* Awaken to handle a signal, request we sleep again later. */
 648                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 649         }
 650
 651         /*
 652          * Make sure we will observe a potential reset request if we've
 653          * observed a change to the power state. Pairs with the smp_wmb() in
 654          * kvm_psci_vcpu_on().
 655          */
 656         smp_rmb();
 657 }
 658
 659 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 660 {
 661         return vcpu->arch.target >= 0;
 662 }
 663
 664 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 665 {
 666         if (kvm_request_pending(vcpu)) {
 667                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 668                         vcpu_req_sleep(vcpu);
 669
 670                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 671                         kvm_reset_vcpu(vcpu);
 672
 673                 /*
 674                  * Clear IRQ_PENDING requests that were made to guarantee
 675                  * that a VCPU sees new virtual interrupts.
 676                  */
 677                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 678
 679                 if (kvm_check_request(KVM_REQ_RECORD_STEAL, vcpu))
 680                         kvm_update_stolen_time(vcpu);
 681         }
 682 }
 683
 684 /**
 685  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 686  * @vcpu:       The VCPU pointer
 687  * @run:        The kvm_run structure pointer used for userspace state exchange
 688  *
 689  * This function is called through the VCPU_RUN ioctl called from user space. It
 690  * will execute VM code in a loop until the time slice for the process is used
 691  * or some emulation is needed from user space in which case the function will
 692  * return with return value 0 and with the kvm_run structure filled in with the
 693  * required data for the requested emulation.
 694  */
 695 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 696 {
 697         int ret;
 698
 699         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 700                 return -ENOEXEC;
 701
 702         ret = kvm_vcpu_first_run_init(vcpu);
 703         if (ret)
 704                 return ret;
 705
 706         if (run->exit_reason == KVM_EXIT_MMIO) {
 707                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 708                 if (ret)
 709                         return ret;
 710         }
 711
 712         if (run->immediate_exit)
 713                 return -EINTR;
 714
 715         vcpu_load(vcpu);
 716
 717         kvm_sigset_activate(vcpu);
 718
 719         ret = 1;
 720         run->exit_reason = KVM_EXIT_UNKNOWN;
 721         while (ret > 0) {
 722                 /*
 723                  * Check conditions before entering the guest
 724                  */
 725                 cond_resched();
 726
 727                 update_vmid(&vcpu->kvm->arch.vmid);
 728
 729                 check_vcpu_requests(vcpu);
 730
 731                 /*
 732                  * Preparing the interrupts to be injected also
 733                  * involves poking the GIC, which must be done in a
 734                  * non-preemptible context.
 735                  */
 736                 preempt_disable();
 737
 738                 kvm_pmu_flush_hwstate(vcpu);
 739
 740                 local_irq_disable();
 741
 742                 kvm_vgic_flush_hwstate(vcpu);
 743
 744                 /*
 745                  * Exit if we have a signal pending so that we can deliver the
 746                  * signal to user space.
 747                  */
 748                 if (signal_pending(current)) {
 749                         ret = -EINTR;
 750                         run->exit_reason = KVM_EXIT_INTR;
 751                 }
 752
 753                 /*
 754                  * If we're using a userspace irqchip, then check if we need
 755                  * to tell a userspace irqchip about timer or PMU level
 756                  * changes and if so, exit to userspace (the actual level
 757                  * state gets updated in kvm_timer_update_run and
 758                  * kvm_pmu_update_run below).
 759                  */
 760                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 761                         if (kvm_timer_should_notify_user(vcpu) ||
 762                             kvm_pmu_should_notify_user(vcpu)) {
 763                                 ret = -EINTR;
 764                                 run->exit_reason = KVM_EXIT_INTR;
 765                         }
 766                 }
 767
 768                 /*
 769                  * Ensure we set mode to IN_GUEST_MODE after we disable
 770                  * interrupts and before the final VCPU requests check.
 771                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 772                  * Documentation/virt/kvm/vcpu-requests.rst
 773                  */
 774                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 775
 776                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 777                     kvm_request_pending(vcpu)) {
 778                         vcpu->mode = OUTSIDE_GUEST_MODE;
 779                         isb(); /* Ensure work in x_flush_hwstate is committed */
 780                         kvm_pmu_sync_hwstate(vcpu);
 781                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 782                                 kvm_timer_sync_hwstate(vcpu);
 783                         kvm_vgic_sync_hwstate(vcpu);
 784                         local_irq_enable();
 785                         preempt_enable();
 786                         continue;
 787                 }
 788
 789                 kvm_arm_setup_debug(vcpu);
 790
 791                 /**************************************************************
 792                  * Enter the guest
 793                  */
 794                 trace_kvm_entry(*vcpu_pc(vcpu));
 795                 guest_enter_irqoff();
 796
 797                 if (has_vhe()) {
 798                         kvm_arm_vhe_guest_enter();
 799                         ret = kvm_vcpu_run_vhe(vcpu);
 800                         kvm_arm_vhe_guest_exit();
 801                 } else {
 802                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 803                 }
 804
 805                 vcpu->mode = OUTSIDE_GUEST_MODE;
 806                 vcpu->stat.exits++;
 807                 /*
 808                  * Back from guest
 809                  *************************************************************/
 810
 811                 kvm_arm_clear_debug(vcpu);
 812
 813                 /*
 814                  * We must sync the PMU state before the vgic state so
 815                  * that the vgic can properly sample the updated state of the
 816                  * interrupt line.
 817                  */
 818                 kvm_pmu_sync_hwstate(vcpu);
 819
 820                 /*
 821                  * Sync the vgic state before syncing the timer state because
 822                  * the timer code needs to know if the virtual timer
 823                  * interrupts are active.
 824                  */
 825                 kvm_vgic_sync_hwstate(vcpu);
 826
 827                 /*
 828                  * Sync the timer hardware state before enabling interrupts as
 829                  * we don't want vtimer interrupts to race with syncing the
 830                  * timer virtual interrupt state.
 831                  */
 832                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 833                         kvm_timer_sync_hwstate(vcpu);
 834
 835                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 836
 837                 /*
 838                  * We may have taken a host interrupt in HYP mode (ie
 839                  * while executing the guest). This interrupt is still
 840                  * pending, as we haven't serviced it yet!
 841                  *
 842                  * We're now back in SVC mode, with interrupts
 843                  * disabled.  Enabling the interrupts now will have
 844                  * the effect of taking the interrupt again, in SVC
 845                  * mode this time.
 846                  */
 847                 local_irq_enable();
 848
 849                 /*
 850                  * We do local_irq_enable() before calling guest_exit() so
 851                  * that if a timer interrupt hits while running the guest we
 852                  * account that tick as being spent in the guest.  We enable
 853                  * preemption after calling guest_exit() so that if we get
 854                  * preempted we make sure ticks after that is not counted as
 855                  * guest time.
 856                  */
 857                 guest_exit();
 858                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 859
 860                 /* Exit types that need handling before we can be preempted */
 861                 handle_exit_early(vcpu, run, ret);
 862
 863                 preempt_enable();
 864
 865                 ret = handle_exit(vcpu, run, ret);
 866         }
 867
 868         /* Tell userspace about in-kernel device output levels */
 869         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 870                 kvm_timer_update_run(vcpu);
 871                 kvm_pmu_update_run(vcpu);
 872         }
 873
 874         kvm_sigset_deactivate(vcpu);
 875
 876         vcpu_put(vcpu);
 877         return ret;
 878 }
 879
 880 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 881 {
 882         int bit_index;
 883         bool set;
 884         unsigned long *hcr;
 885
 886         if (number == KVM_ARM_IRQ_CPU_IRQ)
 887                 bit_index = __ffs(HCR_VI);
 888         else /* KVM_ARM_IRQ_CPU_FIQ */
 889                 bit_index = __ffs(HCR_VF);
 890
 891         hcr = vcpu_hcr(vcpu);
 892         if (level)
 893                 set = test_and_set_bit(bit_index, hcr);
 894         else
 895                 set = test_and_clear_bit(bit_index, hcr);
 896
 897         /*
 898          * If we didn't change anything, no need to wake up or kick other CPUs
 899          */
 900         if (set == level)
 901                 return 0;
 902
 903         /*
 904          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 905          * trigger a world-switch round on the running physical CPU to set the
 906          * virtual IRQ/FIQ fields in the HCR appropriately.
 907          */
 908         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 909         kvm_vcpu_kick(vcpu);
 910
 911         return 0;
 912 }
 913
 914 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 915                           bool line_status)
 916 {
 917         u32 irq = irq_level->irq;
 918         unsigned int irq_type, vcpu_idx, irq_num;
 919         int nrcpus = atomic_read(&kvm->online_vcpus);
 920         struct kvm_vcpu *vcpu = NULL;
 921         bool level = irq_level->level;
 922
 923         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 924         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 925         vcpu_idx += ((irq >> KVM_ARM_IRQ_VCPU2_SHIFT) & KVM_ARM_IRQ_VCPU2_MASK) * (KVM_ARM_IRQ_VCPU_MASK + 1);
 926         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 927
 928         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 929
 930         switch (irq_type) {
 931         case KVM_ARM_IRQ_TYPE_CPU:
 932                 if (irqchip_in_kernel(kvm))
 933                         return -ENXIO;
 934
 935                 if (vcpu_idx >= nrcpus)
 936                         return -EINVAL;
 937
 938                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 939                 if (!vcpu)
 940                         return -EINVAL;
 941
 942                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 943                         return -EINVAL;
 944
 945                 return vcpu_interrupt_line(vcpu, irq_num, level);
 946         case KVM_ARM_IRQ_TYPE_PPI:
 947                 if (!irqchip_in_kernel(kvm))
 948                         return -ENXIO;
 949
 950                 if (vcpu_idx >= nrcpus)
 951                         return -EINVAL;
 952
 953                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 954                 if (!vcpu)
 955                         return -EINVAL;
 956
 957                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 958                         return -EINVAL;
 959
 960                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 961         case KVM_ARM_IRQ_TYPE_SPI:
 962                 if (!irqchip_in_kernel(kvm))
 963                         return -ENXIO;
 964
 965                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 966                         return -EINVAL;
 967
 968                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 969         }
 970
 971         return -EINVAL;
 972 }
 973
 974 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 975                                const struct kvm_vcpu_init *init)
 976 {
 977         unsigned int i, ret;
 978         int phys_target = kvm_target_cpu();
 979
 980         if (init->target != phys_target)
 981                 return -EINVAL;
 982
 983         /*
 984          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 985          * use the same target.
 986          */
 987         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 988                 return -EINVAL;
 989
 990         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 991         for (i = 0; i < sizeof(init->features) * 8; i++) {
 992                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 993
 994                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 995                         return -ENOENT;
 996
 997                 /*
 998                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 999                  * use the same feature set.
1000                  */
1001                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
1002                     test_bit(i, vcpu->arch.features) != set)
1003                         return -EINVAL;
1004
1005                 if (set)
1006                         set_bit(i, vcpu->arch.features);
1007         }
1008
1009         vcpu->arch.target = phys_target;
1010
1011         /* Now we know what it is, we can reset it. */
1012         ret = kvm_reset_vcpu(vcpu);
1013         if (ret) {
1014                 vcpu->arch.target = -1;
1015                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
1016         }
1017
1018         return ret;
1019 }
1020
1021 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
1022                                          struct kvm_vcpu_init *init)
1023 {
1024         int ret;
1025
1026         ret = kvm_vcpu_set_target(vcpu, init);
1027         if (ret)
1028                 return ret;
1029
1030         /*
1031          * Ensure a rebooted VM will fault in RAM pages and detect if the
1032          * guest MMU is turned off and flush the caches as needed.
1033          */
1034         if (vcpu->arch.has_run_once)
1035                 stage2_unmap_vm(vcpu->kvm);
1036
1037         vcpu_reset_hcr(vcpu);
1038
1039         /*
1040          * Handle the "start in power-off" case.
1041          */
1042         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1043                 vcpu_power_off(vcpu);
1044         else
1045                 vcpu->arch.power_off = false;
1046
1047         return 0;
1048 }
1049
1050 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1051                                  struct kvm_device_attr *attr)
1052 {
1053         int ret = -ENXIO;
1054
1055         switch (attr->group) {
1056         default:
1057                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1058                 break;
1059         }
1060
1061         return ret;
1062 }
1063
1064 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1065                                  struct kvm_device_attr *attr)
1066 {
1067         int ret = -ENXIO;
1068
1069         switch (attr->group) {
1070         default:
1071                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1072                 break;
1073         }
1074
1075         return ret;
1076 }
1077
1078 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1079                                  struct kvm_device_attr *attr)
1080 {
1081         int ret = -ENXIO;
1082
1083         switch (attr->group) {
1084         default:
1085                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1086                 break;
1087         }
1088
1089         return ret;
1090 }
1091
1092 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1093                                    struct kvm_vcpu_events *events)
1094 {
1095         memset(events, 0, sizeof(*events));
1096
1097         return __kvm_arm_vcpu_get_events(vcpu, events);
1098 }
1099
1100 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1101                                    struct kvm_vcpu_events *events)
1102 {
1103         int i;
1104
1105         /* check whether the reserved field is zero */
1106         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1107                 if (events->reserved[i])
1108                         return -EINVAL;
1109
1110         /* check whether the pad field is zero */
1111         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1112                 if (events->exception.pad[i])
1113                         return -EINVAL;
1114
1115         return __kvm_arm_vcpu_set_events(vcpu, events);
1116 }
1117
1118 long kvm_arch_vcpu_ioctl(struct file *filp,
1119                          unsigned int ioctl, unsigned long arg)
1120 {
1121         struct kvm_vcpu *vcpu = filp->private_data;
1122         void __user *argp = (void __user *)arg;
1123         struct kvm_device_attr attr;
1124         long r;
1125
1126         switch (ioctl) {
1127         case KVM_ARM_VCPU_INIT: {
1128                 struct kvm_vcpu_init init;
1129
1130                 r = -EFAULT;
1131                 if (copy_from_user(&init, argp, sizeof(init)))
1132                         break;
1133
1134                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1135                 break;
1136         }
1137         case KVM_SET_ONE_REG:
1138         case KVM_GET_ONE_REG: {
1139                 struct kvm_one_reg reg;
1140
1141                 r = -ENOEXEC;
1142                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1143                         break;
1144
1145                 r = -EFAULT;
1146                 if (copy_from_user(&reg, argp, sizeof(reg)))
1147                         break;
1148
1149                 if (ioctl == KVM_SET_ONE_REG)
1150                         r = kvm_arm_set_reg(vcpu, &reg);
1151                 else
1152                         r = kvm_arm_get_reg(vcpu, &reg);
1153                 break;
1154         }
1155         case KVM_GET_REG_LIST: {
1156                 struct kvm_reg_list __user *user_list = argp;
1157                 struct kvm_reg_list reg_list;
1158                 unsigned n;
1159
1160                 r = -ENOEXEC;
1161                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1162                         break;
1163
1164                 r = -EPERM;
1165                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1166                         break;
1167
1168                 r = -EFAULT;
1169                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1170                         break;
1171                 n = reg_list.n;
1172                 reg_list.n = kvm_arm_num_regs(vcpu);
1173                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1174                         break;
1175                 r = -E2BIG;
1176                 if (n < reg_list.n)
1177                         break;
1178                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1179                 break;
1180         }
1181         case KVM_SET_DEVICE_ATTR: {
1182                 r = -EFAULT;
1183                 if (copy_from_user(&attr, argp, sizeof(attr)))
1184                         break;
1185                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1186                 break;
1187         }
1188         case KVM_GET_DEVICE_ATTR: {
1189                 r = -EFAULT;
1190                 if (copy_from_user(&attr, argp, sizeof(attr)))
1191                         break;
1192                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1193                 break;
1194         }
1195         case KVM_HAS_DEVICE_ATTR: {
1196                 r = -EFAULT;
1197                 if (copy_from_user(&attr, argp, sizeof(attr)))
1198                         break;
1199                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1200                 break;
1201         }
1202         case KVM_GET_VCPU_EVENTS: {
1203                 struct kvm_vcpu_events events;
1204
1205                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1206                         return -EINVAL;
1207
1208                 if (copy_to_user(argp, &events, sizeof(events)))
1209                         return -EFAULT;
1210
1211                 return 0;
1212         }
1213         case KVM_SET_VCPU_EVENTS: {
1214                 struct kvm_vcpu_events events;
1215
1216                 if (copy_from_user(&events, argp, sizeof(events)))
1217                         return -EFAULT;
1218
1219                 return kvm_arm_vcpu_set_events(vcpu, &events);
1220         }
1221         case KVM_ARM_VCPU_FINALIZE: {
1222                 int what;
1223
1224                 if (!kvm_vcpu_initialized(vcpu))
1225                         return -ENOEXEC;
1226
1227                 if (get_user(what, (const int __user *)argp))
1228                         return -EFAULT;
1229
1230                 return kvm_arm_vcpu_finalize(vcpu, what);
1231         }
1232         default:
1233                 r = -EINVAL;
1234         }
1235
1236         return r;
1237 }
1238
1239 /**
1240  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1241  * @kvm: kvm instance
1242  * @log: slot id and address to which we copy the log
1243  *
1244  * Steps 1-4 below provide general overview of dirty page logging. See
1245  * kvm_get_dirty_log_protect() function description for additional details.
1246  *
1247  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1248  * always flush the TLB (step 4) even if previous step failed  and the dirty
1249  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1250  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1251  * writes will be marked dirty for next log read.
1252  *
1253  *   1. Take a snapshot of the bit and clear it if needed.
1254  *   2. Write protect the corresponding page.
1255  *   3. Copy the snapshot to the userspace.
1256  *   4. Flush TLB's if needed.
1257  */
1258 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1259 {
1260         bool flush = false;
1261         int r;
1262
1263         mutex_lock(&kvm->slots_lock);
1264
1265         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1266
1267         if (flush)
1268                 kvm_flush_remote_tlbs(kvm);
1269
1270         mutex_unlock(&kvm->slots_lock);
1271         return r;
1272 }
1273
1274 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1275 {
1276         bool flush = false;
1277         int r;
1278
1279         mutex_lock(&kvm->slots_lock);
1280
1281         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1282
1283         if (flush)
1284                 kvm_flush_remote_tlbs(kvm);
1285
1286         mutex_unlock(&kvm->slots_lock);
1287         return r;
1288 }
1289
1290 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1291                                         struct kvm_arm_device_addr *dev_addr)
1292 {
1293         unsigned long dev_id, type;
1294
1295         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1296                 KVM_ARM_DEVICE_ID_SHIFT;
1297         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1298                 KVM_ARM_DEVICE_TYPE_SHIFT;
1299
1300         switch (dev_id) {
1301         case KVM_ARM_DEVICE_VGIC_V2:
1302                 if (!vgic_present)
1303                         return -ENXIO;
1304                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1305         default:
1306                 return -ENODEV;
1307         }
1308 }
1309
1310 long kvm_arch_vm_ioctl(struct file *filp,
1311                        unsigned int ioctl, unsigned long arg)
1312 {
1313         struct kvm *kvm = filp->private_data;
1314         void __user *argp = (void __user *)arg;
1315
1316         switch (ioctl) {
1317         case KVM_CREATE_IRQCHIP: {
1318                 int ret;
1319                 if (!vgic_present)
1320                         return -ENXIO;
1321                 mutex_lock(&kvm->lock);
1322                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1323                 mutex_unlock(&kvm->lock);
1324                 return ret;
1325         }
1326         case KVM_ARM_SET_DEVICE_ADDR: {
1327                 struct kvm_arm_device_addr dev_addr;
1328
1329                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1330                         return -EFAULT;
1331                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1332         }
1333         case KVM_ARM_PREFERRED_TARGET: {
1334                 int err;
1335                 struct kvm_vcpu_init init;
1336
1337                 err = kvm_vcpu_preferred_target(&init);
1338                 if (err)
1339                         return err;
1340
1341                 if (copy_to_user(argp, &init, sizeof(init)))
1342                         return -EFAULT;
1343
1344                 return 0;
1345         }
1346         default:
1347                 return -EINVAL;
1348         }
1349 }
1350
1351 static void cpu_init_hyp_mode(void *dummy)
1352 {
1353         phys_addr_t pgd_ptr;
1354         unsigned long hyp_stack_ptr;
1355         unsigned long stack_page;
1356         unsigned long vector_ptr;
1357
1358         /* Switch from the HYP stub to our own HYP init vector */
1359         __hyp_set_vectors(kvm_get_idmap_vector());
1360
1361         pgd_ptr = kvm_mmu_get_httbr();
1362         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1363         hyp_stack_ptr = stack_page + PAGE_SIZE;
1364         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1365
1366         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1367         __cpu_init_stage2();
1368 }
1369
1370 static void cpu_hyp_reset(void)
1371 {
1372         if (!is_kernel_in_hyp_mode())
1373                 __hyp_reset_vectors();
1374 }
1375
1376 static void cpu_hyp_reinit(void)
1377 {
1378         kvm_init_host_cpu_context(&this_cpu_ptr(&kvm_host_data)->host_ctxt);
1379
1380         cpu_hyp_reset();
1381
1382         if (is_kernel_in_hyp_mode())
1383                 kvm_timer_init_vhe();
1384         else
1385                 cpu_init_hyp_mode(NULL);
1386
1387         kvm_arm_init_debug();
1388
1389         if (vgic_present)
1390                 kvm_vgic_init_cpu_hardware();
1391 }
1392
1393 static void _kvm_arch_hardware_enable(void *discard)
1394 {
1395         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1396                 cpu_hyp_reinit();
1397                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1398         }
1399 }
1400
1401 int kvm_arch_hardware_enable(void)
1402 {
1403         _kvm_arch_hardware_enable(NULL);
1404         return 0;
1405 }
1406
1407 static void _kvm_arch_hardware_disable(void *discard)
1408 {
1409         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1410                 cpu_hyp_reset();
1411                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1412         }
1413 }
1414
1415 void kvm_arch_hardware_disable(void)
1416 {
1417         _kvm_arch_hardware_disable(NULL);
1418 }
1419
1420 #ifdef CONFIG_CPU_PM
1421 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1422                                     unsigned long cmd,
1423                                     void *v)
1424 {
1425         /*
1426          * kvm_arm_hardware_enabled is left with its old value over
1427          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1428          * re-enable hyp.
1429          */
1430         switch (cmd) {
1431         case CPU_PM_ENTER:
1432                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1433                         /*
1434                          * don't update kvm_arm_hardware_enabled here
1435                          * so that the hardware will be re-enabled
1436                          * when we resume. See below.
1437                          */
1438                         cpu_hyp_reset();
1439
1440                 return NOTIFY_OK;
1441         case CPU_PM_ENTER_FAILED:
1442         case CPU_PM_EXIT:
1443                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1444                         /* The hardware was enabled before suspend. */
1445                         cpu_hyp_reinit();
1446
1447                 return NOTIFY_OK;
1448
1449         default:
1450                 return NOTIFY_DONE;
1451         }
1452 }
1453
1454 static struct notifier_block hyp_init_cpu_pm_nb = {
1455         .notifier_call = hyp_init_cpu_pm_notifier,
1456 };
1457
1458 static void __init hyp_cpu_pm_init(void)
1459 {
1460         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1461 }
1462 static void __init hyp_cpu_pm_exit(void)
1463 {
1464         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1465 }
1466 #else
1467 static inline void hyp_cpu_pm_init(void)
1468 {
1469 }
1470 static inline void hyp_cpu_pm_exit(void)
1471 {
1472 }
1473 #endif
1474
1475 static int init_common_resources(void)
1476 {
1477         kvm_set_ipa_limit();
1478
1479         return 0;
1480 }
1481
1482 static int init_subsystems(void)
1483 {
1484         int err = 0;
1485
1486         /*
1487          * Enable hardware so that subsystem initialisation can access EL2.
1488          */
1489         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1490
1491         /*
1492          * Register CPU lower-power notifier
1493          */
1494         hyp_cpu_pm_init();
1495
1496         /*
1497          * Init HYP view of VGIC
1498          */
1499         err = kvm_vgic_hyp_init();
1500         switch (err) {
1501         case 0:
1502                 vgic_present = true;
1503                 break;
1504         case -ENODEV:
1505         case -ENXIO:
1506                 vgic_present = false;
1507                 err = 0;
1508                 break;
1509         default:
1510                 goto out;
1511         }
1512
1513         /*
1514          * Init HYP architected timer support
1515          */
1516         err = kvm_timer_hyp_init(vgic_present);
1517         if (err)
1518                 goto out;
1519
1520         kvm_perf_init();
1521         kvm_coproc_table_init();
1522
1523 out:
1524         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1525
1526         return err;
1527 }
1528
1529 static void teardown_hyp_mode(void)
1530 {
1531         int cpu;
1532
1533         free_hyp_pgds();
1534         for_each_possible_cpu(cpu)
1535                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1536         hyp_cpu_pm_exit();
1537 }
1538
1539 /**
1540  * Inits Hyp-mode on all online CPUs
1541  */
1542 static int init_hyp_mode(void)
1543 {
1544         int cpu;
1545         int err = 0;
1546
1547         /*
1548          * Allocate Hyp PGD and setup Hyp identity mapping
1549          */
1550         err = kvm_mmu_init();
1551         if (err)
1552                 goto out_err;
1553
1554         /*
1555          * Allocate stack pages for Hypervisor-mode
1556          */
1557         for_each_possible_cpu(cpu) {
1558                 unsigned long stack_page;
1559
1560                 stack_page = __get_free_page(GFP_KERNEL);
1561                 if (!stack_page) {
1562                         err = -ENOMEM;
1563                         goto out_err;
1564                 }
1565
1566                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1567         }
1568
1569         /*
1570          * Map the Hyp-code called directly from the host
1571          */
1572         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1573                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1574         if (err) {
1575                 kvm_err("Cannot map world-switch code\n");
1576                 goto out_err;
1577         }
1578
1579         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1580                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1581         if (err) {
1582                 kvm_err("Cannot map rodata section\n");
1583                 goto out_err;
1584         }
1585
1586         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1587                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1588         if (err) {
1589                 kvm_err("Cannot map bss section\n");
1590                 goto out_err;
1591         }
1592
1593         err = kvm_map_vectors();
1594         if (err) {
1595                 kvm_err("Cannot map vectors\n");
1596                 goto out_err;
1597         }
1598
1599         /*
1600          * Map the Hyp stack pages
1601          */
1602         for_each_possible_cpu(cpu) {
1603                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1604                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1605                                           PAGE_HYP);
1606
1607                 if (err) {
1608                         kvm_err("Cannot map hyp stack\n");
1609                         goto out_err;
1610                 }
1611         }
1612
1613         for_each_possible_cpu(cpu) {
1614                 kvm_host_data_t *cpu_data;
1615
1616                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1617                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1618
1619                 if (err) {
1620                         kvm_err("Cannot map host CPU state: %d\n", err);
1621                         goto out_err;
1622                 }
1623         }
1624
1625         err = hyp_map_aux_data();
1626         if (err)
1627                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1628
1629         return 0;
1630
1631 out_err:
1632         teardown_hyp_mode();
1633         kvm_err("error initializing Hyp mode: %d\n", err);
1634         return err;
1635 }
1636
1637 static void check_kvm_target_cpu(void *ret)
1638 {
1639         *(int *)ret = kvm_target_cpu();
1640 }
1641
1642 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1643 {
1644         struct kvm_vcpu *vcpu;
1645         int i;
1646
1647         mpidr &= MPIDR_HWID_BITMASK;
1648         kvm_for_each_vcpu(i, vcpu, kvm) {
1649                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1650                         return vcpu;
1651         }
1652         return NULL;
1653 }
1654
1655 bool kvm_arch_has_irq_bypass(void)
1656 {
1657         return true;
1658 }
1659
1660 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1661                                       struct irq_bypass_producer *prod)
1662 {
1663         struct kvm_kernel_irqfd *irqfd =
1664                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1665
1666         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1667                                           &irqfd->irq_entry);
1668 }
1669 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1670                                       struct irq_bypass_producer *prod)
1671 {
1672         struct kvm_kernel_irqfd *irqfd =
1673                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1674
1675         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1676                                      &irqfd->irq_entry);
1677 }
1678
1679 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1680 {
1681         struct kvm_kernel_irqfd *irqfd =
1682                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1683
1684         kvm_arm_halt_guest(irqfd->kvm);
1685 }
1686
1687 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1688 {
1689         struct kvm_kernel_irqfd *irqfd =
1690                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1691
1692         kvm_arm_resume_guest(irqfd->kvm);
1693 }
1694
1695 /**
1696  * Initialize Hyp-mode and memory mappings on all CPUs.
1697  */
1698 int kvm_arch_init(void *opaque)
1699 {
1700         int err;
1701         int ret, cpu;
1702         bool in_hyp_mode;
1703
1704         if (!is_hyp_mode_available()) {
1705                 kvm_info("HYP mode not available\n");
1706                 return -ENODEV;
1707         }
1708
1709         in_hyp_mode = is_kernel_in_hyp_mode();
1710
1711         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1712                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1713                 return -ENODEV;
1714         }
1715
1716         for_each_online_cpu(cpu) {
1717                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1718                 if (ret < 0) {
1719                         kvm_err("Error, CPU %d not supported!\n", cpu);
1720                         return -ENODEV;
1721                 }
1722         }
1723
1724         err = init_common_resources();
1725         if (err)
1726                 return err;
1727
1728         err = kvm_arm_init_sve();
1729         if (err)
1730                 return err;
1731
1732         if (!in_hyp_mode) {
1733                 err = init_hyp_mode();
1734                 if (err)
1735                         goto out_err;
1736         }
1737
1738         err = init_subsystems();
1739         if (err)
1740                 goto out_hyp;
1741
1742         if (in_hyp_mode)
1743                 kvm_info("VHE mode initialized successfully\n");
1744         else
1745                 kvm_info("Hyp mode initialized successfully\n");
1746
1747         return 0;
1748
1749 out_hyp:
1750         if (!in_hyp_mode)
1751                 teardown_hyp_mode();
1752 out_err:
1753         return err;
1754 }
1755
1756 /* NOP: Compiling as a module not supported */
1757 void kvm_arch_exit(void)
1758 {
1759         kvm_perf_teardown();
1760 }
1761
1762 static int arm_init(void)
1763 {
1764         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1765         return rc;
1766 }
1767
1768 module_init(arm_init);