arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu_pm.h>
  20 #include <linux/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31 #include <kvm/arm_pmu.h>
  32
  33 #define CREATE_TRACE_POINTS
  34 #include "trace.h"
  35
  36 #include <linux/uaccess.h>
  37 #include <asm/ptrace.h>
  38 #include <asm/mman.h>
  39 #include <asm/tlbflush.h>
  40 #include <asm/cacheflush.h>
  41 #include <asm/virt.h>
  42 #include <asm/kvm_arm.h>
  43 #include <asm/kvm_asm.h>
  44 #include <asm/kvm_mmu.h>
  45 #include <asm/kvm_emulate.h>
  46 #include <asm/kvm_coproc.h>
  47 #include <asm/kvm_psci.h>
  48 #include <asm/sections.h>
  49
  50 #ifdef REQUIRES_VIRT
  51 __asm__(".arch_extension        virt");
  52 #endif
  53
  54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  55 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  56
  57 /* Per-CPU variable containing the currently running vcpu. */
  58 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  59
  60 /* The VMID used in the VTTBR */
  61 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  62 static u32 kvm_next_vmid;
  63 static unsigned int kvm_vmid_bits __read_mostly;
  64 static DEFINE_SPINLOCK(kvm_vmid_lock);
  65
  66 static bool vgic_present;
  67
  68 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  69
  70 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  71 {
  72         BUG_ON(preemptible());
  73         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  74 }
  75
  76 /**
  77  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  78  * Must be called from non-preemptible context
  79  */
  80 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  81 {
  82         BUG_ON(preemptible());
  83         return __this_cpu_read(kvm_arm_running_vcpu);
  84 }
  85
  86 /**
  87  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  88  */
  89 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  90 {
  91         return &kvm_arm_running_vcpu;
  92 }
  93
  94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  95 {
  96         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  97 }
  98
  99 int kvm_arch_hardware_setup(void)
 100 {
 101         return 0;
 102 }
 103
 104 void kvm_arch_check_processor_compat(void *rtn)
 105 {
 106         *(int *)rtn = 0;
 107 }
 108
 109
 110 /**
 111  * kvm_arch_init_vm - initializes a VM data structure
 112  * @kvm:        pointer to the KVM struct
 113  */
 114 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 115 {
 116         int ret, cpu;
 117
 118         if (type)
 119                 return -EINVAL;
 120
 121         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 122         if (!kvm->arch.last_vcpu_ran)
 123                 return -ENOMEM;
 124
 125         for_each_possible_cpu(cpu)
 126                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 127
 128         ret = kvm_alloc_stage2_pgd(kvm);
 129         if (ret)
 130                 goto out_fail_alloc;
 131
 132         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 133         if (ret)
 134                 goto out_free_stage2_pgd;
 135
 136         kvm_vgic_early_init(kvm);
 137
 138         /* Mark the initial VMID generation invalid */
 139         kvm->arch.vmid_gen = 0;
 140
 141         /* The maximum number of VCPUs is limited by the host's GIC model */
 142         kvm->arch.max_vcpus = vgic_present ?
 143                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 144
 145         return ret;
 146 out_free_stage2_pgd:
 147         kvm_free_stage2_pgd(kvm);
 148 out_fail_alloc:
 149         free_percpu(kvm->arch.last_vcpu_ran);
 150         kvm->arch.last_vcpu_ran = NULL;
 151         return ret;
 152 }
 153
 154 bool kvm_arch_has_vcpu_debugfs(void)
 155 {
 156         return false;
 157 }
 158
 159 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 160 {
 161         return 0;
 162 }
 163
 164 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 165 {
 166         return VM_FAULT_SIGBUS;
 167 }
 168
 169
 170 /**
 171  * kvm_arch_destroy_vm - destroy the VM data structure
 172  * @kvm:        pointer to the KVM struct
 173  */
 174 void kvm_arch_destroy_vm(struct kvm *kvm)
 175 {
 176         int i;
 177
 178         free_percpu(kvm->arch.last_vcpu_ran);
 179         kvm->arch.last_vcpu_ran = NULL;
 180
 181         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 182                 if (kvm->vcpus[i]) {
 183                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 184                         kvm->vcpus[i] = NULL;
 185                 }
 186         }
 187
 188         kvm_vgic_destroy(kvm);
 189 }
 190
 191 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 192 {
 193         int r;
 194         switch (ext) {
 195         case KVM_CAP_IRQCHIP:
 196                 r = vgic_present;
 197                 break;
 198         case KVM_CAP_IOEVENTFD:
 199         case KVM_CAP_DEVICE_CTRL:
 200         case KVM_CAP_USER_MEMORY:
 201         case KVM_CAP_SYNC_MMU:
 202         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 203         case KVM_CAP_ONE_REG:
 204         case KVM_CAP_ARM_PSCI:
 205         case KVM_CAP_ARM_PSCI_0_2:
 206         case KVM_CAP_READONLY_MEM:
 207         case KVM_CAP_MP_STATE:
 208         case KVM_CAP_IMMEDIATE_EXIT:
 209                 r = 1;
 210                 break;
 211         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 212                 r = 1;
 213                 break;
 214         case KVM_CAP_NR_VCPUS:
 215                 r = num_online_cpus();
 216                 break;
 217         case KVM_CAP_MAX_VCPUS:
 218                 r = KVM_MAX_VCPUS;
 219                 break;
 220         case KVM_CAP_NR_MEMSLOTS:
 221                 r = KVM_USER_MEM_SLOTS;
 222                 break;
 223         case KVM_CAP_MSI_DEVID:
 224                 if (!kvm)
 225                         r = -EINVAL;
 226                 else
 227                         r = kvm->arch.vgic.msis_require_devid;
 228                 break;
 229         case KVM_CAP_ARM_USER_IRQ:
 230                 /*
 231                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 232                  * (bump this number if adding more devices)
 233                  */
 234                 r = 1;
 235                 break;
 236         default:
 237                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 238                 break;
 239         }
 240         return r;
 241 }
 242
 243 long kvm_arch_dev_ioctl(struct file *filp,
 244                         unsigned int ioctl, unsigned long arg)
 245 {
 246         return -EINVAL;
 247 }
 248
 249
 250 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 251 {
 252         int err;
 253         struct kvm_vcpu *vcpu;
 254
 255         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 256                 err = -EBUSY;
 257                 goto out;
 258         }
 259
 260         if (id >= kvm->arch.max_vcpus) {
 261                 err = -EINVAL;
 262                 goto out;
 263         }
 264
 265         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 266         if (!vcpu) {
 267                 err = -ENOMEM;
 268                 goto out;
 269         }
 270
 271         err = kvm_vcpu_init(vcpu, kvm, id);
 272         if (err)
 273                 goto free_vcpu;
 274
 275         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 276         if (err)
 277                 goto vcpu_uninit;
 278
 279         return vcpu;
 280 vcpu_uninit:
 281         kvm_vcpu_uninit(vcpu);
 282 free_vcpu:
 283         kmem_cache_free(kvm_vcpu_cache, vcpu);
 284 out:
 285         return ERR_PTR(err);
 286 }
 287
 288 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 289 {
 290         kvm_vgic_vcpu_early_init(vcpu);
 291 }
 292
 293 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 294 {
 295         kvm_mmu_free_memory_caches(vcpu);
 296         kvm_timer_vcpu_terminate(vcpu);
 297         kvm_vgic_vcpu_destroy(vcpu);
 298         kvm_pmu_vcpu_destroy(vcpu);
 299         kvm_vcpu_uninit(vcpu);
 300         kmem_cache_free(kvm_vcpu_cache, vcpu);
 301 }
 302
 303 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 304 {
 305         kvm_arch_vcpu_free(vcpu);
 306 }
 307
 308 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 309 {
 310         return kvm_timer_should_fire(vcpu_vtimer(vcpu)) ||
 311                kvm_timer_should_fire(vcpu_ptimer(vcpu));
 312 }
 313
 314 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 315 {
 316         kvm_timer_schedule(vcpu);
 317 }
 318
 319 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 320 {
 321         kvm_timer_unschedule(vcpu);
 322 }
 323
 324 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 325 {
 326         /* Force users to call KVM_ARM_VCPU_INIT */
 327         vcpu->arch.target = -1;
 328         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 329
 330         /* Set up the timer */
 331         kvm_timer_vcpu_init(vcpu);
 332
 333         kvm_arm_reset_debug_ptr(vcpu);
 334
 335         return 0;
 336 }
 337
 338 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 339 {
 340         int *last_ran;
 341
 342         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 343
 344         /*
 345          * We might get preempted before the vCPU actually runs, but
 346          * over-invalidation doesn't affect correctness.
 347          */
 348         if (*last_ran != vcpu->vcpu_id) {
 349                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 350                 *last_ran = vcpu->vcpu_id;
 351         }
 352
 353         vcpu->cpu = cpu;
 354         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 355
 356         kvm_arm_set_running_vcpu(vcpu);
 357
 358         kvm_vgic_load(vcpu);
 359 }
 360
 361 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 362 {
 363         kvm_vgic_put(vcpu);
 364
 365         vcpu->cpu = -1;
 366
 367         kvm_arm_set_running_vcpu(NULL);
 368         kvm_timer_vcpu_put(vcpu);
 369 }
 370
 371 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 372                                     struct kvm_mp_state *mp_state)
 373 {
 374         if (vcpu->arch.power_off)
 375                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 376         else
 377                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 378
 379         return 0;
 380 }
 381
 382 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 383                                     struct kvm_mp_state *mp_state)
 384 {
 385         switch (mp_state->mp_state) {
 386         case KVM_MP_STATE_RUNNABLE:
 387                 vcpu->arch.power_off = false;
 388                 break;
 389         case KVM_MP_STATE_STOPPED:
 390                 vcpu->arch.power_off = true;
 391                 break;
 392         default:
 393                 return -EINVAL;
 394         }
 395
 396         return 0;
 397 }
 398
 399 /**
 400  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 401  * @v:          The VCPU pointer
 402  *
 403  * If the guest CPU is not waiting for interrupts or an interrupt line is
 404  * asserted, the CPU is by definition runnable.
 405  */
 406 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 407 {
 408         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
 409                 && !v->arch.power_off && !v->arch.pause);
 410 }
 411
 412 /* Just ensure a guest exit from a particular CPU */
 413 static void exit_vm_noop(void *info)
 414 {
 415 }
 416
 417 void force_vm_exit(const cpumask_t *mask)
 418 {
 419         preempt_disable();
 420         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 421         preempt_enable();
 422 }
 423
 424 /**
 425  * need_new_vmid_gen - check that the VMID is still valid
 426  * @kvm: The VM's VMID to check
 427  *
 428  * return true if there is a new generation of VMIDs being used
 429  *
 430  * The hardware supports only 256 values with the value zero reserved for the
 431  * host, so we check if an assigned value belongs to a previous generation,
 432  * which which requires us to assign a new value. If we're the first to use a
 433  * VMID for the new generation, we must flush necessary caches and TLBs on all
 434  * CPUs.
 435  */
 436 static bool need_new_vmid_gen(struct kvm *kvm)
 437 {
 438         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 439 }
 440
 441 /**
 442  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 443  * @kvm The guest that we are about to run
 444  *
 445  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 446  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 447  * caches and TLBs.
 448  */
 449 static void update_vttbr(struct kvm *kvm)
 450 {
 451         phys_addr_t pgd_phys;
 452         u64 vmid;
 453
 454         if (!need_new_vmid_gen(kvm))
 455                 return;
 456
 457         spin_lock(&kvm_vmid_lock);
 458
 459         /*
 460          * We need to re-check the vmid_gen here to ensure that if another vcpu
 461          * already allocated a valid vmid for this vm, then this vcpu should
 462          * use the same vmid.
 463          */
 464         if (!need_new_vmid_gen(kvm)) {
 465                 spin_unlock(&kvm_vmid_lock);
 466                 return;
 467         }
 468
 469         /* First user of a new VMID generation? */
 470         if (unlikely(kvm_next_vmid == 0)) {
 471                 atomic64_inc(&kvm_vmid_gen);
 472                 kvm_next_vmid = 1;
 473
 474                 /*
 475                  * On SMP we know no other CPUs can use this CPU's or each
 476                  * other's VMID after force_vm_exit returns since the
 477                  * kvm_vmid_lock blocks them from reentry to the guest.
 478                  */
 479                 force_vm_exit(cpu_all_mask);
 480                 /*
 481                  * Now broadcast TLB + ICACHE invalidation over the inner
 482                  * shareable domain to make sure all data structures are
 483                  * clean.
 484                  */
 485                 kvm_call_hyp(__kvm_flush_vm_context);
 486         }
 487
 488         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 489         kvm->arch.vmid = kvm_next_vmid;
 490         kvm_next_vmid++;
 491         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 492
 493         /* update vttbr to be used with the new vmid */
 494         pgd_phys = virt_to_phys(kvm->arch.pgd);
 495         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 496         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 497         kvm->arch.vttbr = pgd_phys | vmid;
 498
 499         spin_unlock(&kvm_vmid_lock);
 500 }
 501
 502 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 503 {
 504         struct kvm *kvm = vcpu->kvm;
 505         int ret = 0;
 506
 507         if (likely(vcpu->arch.has_run_once))
 508                 return 0;
 509
 510         vcpu->arch.has_run_once = true;
 511
 512         /*
 513          * Map the VGIC hardware resources before running a vcpu the first
 514          * time on this VM.
 515          */
 516         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
 517                 ret = kvm_vgic_map_resources(kvm);
 518                 if (ret)
 519                         return ret;
 520         }
 521
 522         ret = kvm_timer_enable(vcpu);
 523
 524         return ret;
 525 }
 526
 527 bool kvm_arch_intc_initialized(struct kvm *kvm)
 528 {
 529         return vgic_initialized(kvm);
 530 }
 531
 532 void kvm_arm_halt_guest(struct kvm *kvm)
 533 {
 534         int i;
 535         struct kvm_vcpu *vcpu;
 536
 537         kvm_for_each_vcpu(i, vcpu, kvm)
 538                 vcpu->arch.pause = true;
 539         kvm_make_all_cpus_request(kvm, KVM_REQ_VCPU_EXIT);
 540 }
 541
 542 void kvm_arm_halt_vcpu(struct kvm_vcpu *vcpu)
 543 {
 544         vcpu->arch.pause = true;
 545         kvm_vcpu_kick(vcpu);
 546 }
 547
 548 void kvm_arm_resume_vcpu(struct kvm_vcpu *vcpu)
 549 {
 550         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 551
 552         vcpu->arch.pause = false;
 553         swake_up(wq);
 554 }
 555
 556 void kvm_arm_resume_guest(struct kvm *kvm)
 557 {
 558         int i;
 559         struct kvm_vcpu *vcpu;
 560
 561         kvm_for_each_vcpu(i, vcpu, kvm)
 562                 kvm_arm_resume_vcpu(vcpu);
 563 }
 564
 565 static void vcpu_sleep(struct kvm_vcpu *vcpu)
 566 {
 567         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 568
 569         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 570                                        (!vcpu->arch.pause)));
 571 }
 572
 573 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 574 {
 575         return vcpu->arch.target >= 0;
 576 }
 577
 578 /**
 579  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 580  * @vcpu:       The VCPU pointer
 581  * @run:        The kvm_run structure pointer used for userspace state exchange
 582  *
 583  * This function is called through the VCPU_RUN ioctl called from user space. It
 584  * will execute VM code in a loop until the time slice for the process is used
 585  * or some emulation is needed from user space in which case the function will
 586  * return with return value 0 and with the kvm_run structure filled in with the
 587  * required data for the requested emulation.
 588  */
 589 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 590 {
 591         int ret;
 592         sigset_t sigsaved;
 593
 594         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 595                 return -ENOEXEC;
 596
 597         ret = kvm_vcpu_first_run_init(vcpu);
 598         if (ret)
 599                 return ret;
 600
 601         if (run->exit_reason == KVM_EXIT_MMIO) {
 602                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 603                 if (ret)
 604                         return ret;
 605         }
 606
 607         if (run->immediate_exit)
 608                 return -EINTR;
 609
 610         if (vcpu->sigset_active)
 611                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 612
 613         ret = 1;
 614         run->exit_reason = KVM_EXIT_UNKNOWN;
 615         while (ret > 0) {
 616                 /*
 617                  * Check conditions before entering the guest
 618                  */
 619                 cond_resched();
 620
 621                 update_vttbr(vcpu->kvm);
 622
 623                 if (vcpu->arch.power_off || vcpu->arch.pause)
 624                         vcpu_sleep(vcpu);
 625
 626                 /*
 627                  * Preparing the interrupts to be injected also
 628                  * involves poking the GIC, which must be done in a
 629                  * non-preemptible context.
 630                  */
 631                 preempt_disable();
 632
 633                 kvm_pmu_flush_hwstate(vcpu);
 634
 635                 kvm_timer_flush_hwstate(vcpu);
 636                 kvm_vgic_flush_hwstate(vcpu);
 637
 638                 local_irq_disable();
 639
 640                 /*
 641                  * If we have a singal pending, or need to notify a userspace
 642                  * irqchip about timer or PMU level changes, then we exit (and
 643                  * update the timer level state in kvm_timer_update_run
 644                  * below).
 645                  */
 646                 if (signal_pending(current) ||
 647                     kvm_timer_should_notify_user(vcpu) ||
 648                     kvm_pmu_should_notify_user(vcpu)) {
 649                         ret = -EINTR;
 650                         run->exit_reason = KVM_EXIT_INTR;
 651                 }
 652
 653                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 654                         vcpu->arch.power_off || vcpu->arch.pause) {
 655                         local_irq_enable();
 656                         kvm_pmu_sync_hwstate(vcpu);
 657                         kvm_timer_sync_hwstate(vcpu);
 658                         kvm_vgic_sync_hwstate(vcpu);
 659                         preempt_enable();
 660                         continue;
 661                 }
 662
 663                 kvm_arm_setup_debug(vcpu);
 664
 665                 /**************************************************************
 666                  * Enter the guest
 667                  */
 668                 trace_kvm_entry(*vcpu_pc(vcpu));
 669                 guest_enter_irqoff();
 670                 vcpu->mode = IN_GUEST_MODE;
 671
 672                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 673
 674                 vcpu->mode = OUTSIDE_GUEST_MODE;
 675                 vcpu->stat.exits++;
 676                 /*
 677                  * Back from guest
 678                  *************************************************************/
 679
 680                 kvm_arm_clear_debug(vcpu);
 681
 682                 /*
 683                  * We may have taken a host interrupt in HYP mode (ie
 684                  * while executing the guest). This interrupt is still
 685                  * pending, as we haven't serviced it yet!
 686                  *
 687                  * We're now back in SVC mode, with interrupts
 688                  * disabled.  Enabling the interrupts now will have
 689                  * the effect of taking the interrupt again, in SVC
 690                  * mode this time.
 691                  */
 692                 local_irq_enable();
 693
 694                 /*
 695                  * We do local_irq_enable() before calling guest_exit() so
 696                  * that if a timer interrupt hits while running the guest we
 697                  * account that tick as being spent in the guest.  We enable
 698                  * preemption after calling guest_exit() so that if we get
 699                  * preempted we make sure ticks after that is not counted as
 700                  * guest time.
 701                  */
 702                 guest_exit();
 703                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 704
 705                 /*
 706                  * We must sync the PMU and timer state before the vgic state so
 707                  * that the vgic can properly sample the updated state of the
 708                  * interrupt line.
 709                  */
 710                 kvm_pmu_sync_hwstate(vcpu);
 711                 kvm_timer_sync_hwstate(vcpu);
 712
 713                 kvm_vgic_sync_hwstate(vcpu);
 714
 715                 preempt_enable();
 716
 717                 ret = handle_exit(vcpu, run, ret);
 718         }
 719
 720         /* Tell userspace about in-kernel device output levels */
 721         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 722                 kvm_timer_update_run(vcpu);
 723                 kvm_pmu_update_run(vcpu);
 724         }
 725
 726         if (vcpu->sigset_active)
 727                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 728         return ret;
 729 }
 730
 731 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 732 {
 733         int bit_index;
 734         bool set;
 735         unsigned long *ptr;
 736
 737         if (number == KVM_ARM_IRQ_CPU_IRQ)
 738                 bit_index = __ffs(HCR_VI);
 739         else /* KVM_ARM_IRQ_CPU_FIQ */
 740                 bit_index = __ffs(HCR_VF);
 741
 742         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 743         if (level)
 744                 set = test_and_set_bit(bit_index, ptr);
 745         else
 746                 set = test_and_clear_bit(bit_index, ptr);
 747
 748         /*
 749          * If we didn't change anything, no need to wake up or kick other CPUs
 750          */
 751         if (set == level)
 752                 return 0;
 753
 754         /*
 755          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 756          * trigger a world-switch round on the running physical CPU to set the
 757          * virtual IRQ/FIQ fields in the HCR appropriately.
 758          */
 759         kvm_vcpu_kick(vcpu);
 760
 761         return 0;
 762 }
 763
 764 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 765                           bool line_status)
 766 {
 767         u32 irq = irq_level->irq;
 768         unsigned int irq_type, vcpu_idx, irq_num;
 769         int nrcpus = atomic_read(&kvm->online_vcpus);
 770         struct kvm_vcpu *vcpu = NULL;
 771         bool level = irq_level->level;
 772
 773         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 774         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 775         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 776
 777         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 778
 779         switch (irq_type) {
 780         case KVM_ARM_IRQ_TYPE_CPU:
 781                 if (irqchip_in_kernel(kvm))
 782                         return -ENXIO;
 783
 784                 if (vcpu_idx >= nrcpus)
 785                         return -EINVAL;
 786
 787                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 788                 if (!vcpu)
 789                         return -EINVAL;
 790
 791                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 792                         return -EINVAL;
 793
 794                 return vcpu_interrupt_line(vcpu, irq_num, level);
 795         case KVM_ARM_IRQ_TYPE_PPI:
 796                 if (!irqchip_in_kernel(kvm))
 797                         return -ENXIO;
 798
 799                 if (vcpu_idx >= nrcpus)
 800                         return -EINVAL;
 801
 802                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 803                 if (!vcpu)
 804                         return -EINVAL;
 805
 806                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 807                         return -EINVAL;
 808
 809                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 810         case KVM_ARM_IRQ_TYPE_SPI:
 811                 if (!irqchip_in_kernel(kvm))
 812                         return -ENXIO;
 813
 814                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 815                         return -EINVAL;
 816
 817                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 818         }
 819
 820         return -EINVAL;
 821 }
 822
 823 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 824                                const struct kvm_vcpu_init *init)
 825 {
 826         unsigned int i;
 827         int phys_target = kvm_target_cpu();
 828
 829         if (init->target != phys_target)
 830                 return -EINVAL;
 831
 832         /*
 833          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 834          * use the same target.
 835          */
 836         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 837                 return -EINVAL;
 838
 839         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 840         for (i = 0; i < sizeof(init->features) * 8; i++) {
 841                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 842
 843                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 844                         return -ENOENT;
 845
 846                 /*
 847                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 848                  * use the same feature set.
 849                  */
 850                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 851                     test_bit(i, vcpu->arch.features) != set)
 852                         return -EINVAL;
 853
 854                 if (set)
 855                         set_bit(i, vcpu->arch.features);
 856         }
 857
 858         vcpu->arch.target = phys_target;
 859
 860         /* Now we know what it is, we can reset it. */
 861         return kvm_reset_vcpu(vcpu);
 862 }
 863
 864
 865 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 866                                          struct kvm_vcpu_init *init)
 867 {
 868         int ret;
 869
 870         ret = kvm_vcpu_set_target(vcpu, init);
 871         if (ret)
 872                 return ret;
 873
 874         /*
 875          * Ensure a rebooted VM will fault in RAM pages and detect if the
 876          * guest MMU is turned off and flush the caches as needed.
 877          */
 878         if (vcpu->arch.has_run_once)
 879                 stage2_unmap_vm(vcpu->kvm);
 880
 881         vcpu_reset_hcr(vcpu);
 882
 883         /*
 884          * Handle the "start in power-off" case.
 885          */
 886         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 887                 vcpu->arch.power_off = true;
 888         else
 889                 vcpu->arch.power_off = false;
 890
 891         return 0;
 892 }
 893
 894 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 895                                  struct kvm_device_attr *attr)
 896 {
 897         int ret = -ENXIO;
 898
 899         switch (attr->group) {
 900         default:
 901                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 902                 break;
 903         }
 904
 905         return ret;
 906 }
 907
 908 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 909                                  struct kvm_device_attr *attr)
 910 {
 911         int ret = -ENXIO;
 912
 913         switch (attr->group) {
 914         default:
 915                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
 916                 break;
 917         }
 918
 919         return ret;
 920 }
 921
 922 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
 923                                  struct kvm_device_attr *attr)
 924 {
 925         int ret = -ENXIO;
 926
 927         switch (attr->group) {
 928         default:
 929                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
 930                 break;
 931         }
 932
 933         return ret;
 934 }
 935
 936 long kvm_arch_vcpu_ioctl(struct file *filp,
 937                          unsigned int ioctl, unsigned long arg)
 938 {
 939         struct kvm_vcpu *vcpu = filp->private_data;
 940         void __user *argp = (void __user *)arg;
 941         struct kvm_device_attr attr;
 942
 943         switch (ioctl) {
 944         case KVM_ARM_VCPU_INIT: {
 945                 struct kvm_vcpu_init init;
 946
 947                 if (copy_from_user(&init, argp, sizeof(init)))
 948                         return -EFAULT;
 949
 950                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 951         }
 952         case KVM_SET_ONE_REG:
 953         case KVM_GET_ONE_REG: {
 954                 struct kvm_one_reg reg;
 955
 956                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 957                         return -ENOEXEC;
 958
 959                 if (copy_from_user(&reg, argp, sizeof(reg)))
 960                         return -EFAULT;
 961                 if (ioctl == KVM_SET_ONE_REG)
 962                         return kvm_arm_set_reg(vcpu, &reg);
 963                 else
 964                         return kvm_arm_get_reg(vcpu, &reg);
 965         }
 966         case KVM_GET_REG_LIST: {
 967                 struct kvm_reg_list __user *user_list = argp;
 968                 struct kvm_reg_list reg_list;
 969                 unsigned n;
 970
 971                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 972                         return -ENOEXEC;
 973
 974                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 975                         return -EFAULT;
 976                 n = reg_list.n;
 977                 reg_list.n = kvm_arm_num_regs(vcpu);
 978                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 979                         return -EFAULT;
 980                 if (n < reg_list.n)
 981                         return -E2BIG;
 982                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 983         }
 984         case KVM_SET_DEVICE_ATTR: {
 985                 if (copy_from_user(&attr, argp, sizeof(attr)))
 986                         return -EFAULT;
 987                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
 988         }
 989         case KVM_GET_DEVICE_ATTR: {
 990                 if (copy_from_user(&attr, argp, sizeof(attr)))
 991                         return -EFAULT;
 992                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
 993         }
 994         case KVM_HAS_DEVICE_ATTR: {
 995                 if (copy_from_user(&attr, argp, sizeof(attr)))
 996                         return -EFAULT;
 997                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
 998         }
 999         default:
1000                 return -EINVAL;
1001         }
1002 }
1003
1004 /**
1005  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1006  * @kvm: kvm instance
1007  * @log: slot id and address to which we copy the log
1008  *
1009  * Steps 1-4 below provide general overview of dirty page logging. See
1010  * kvm_get_dirty_log_protect() function description for additional details.
1011  *
1012  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1013  * always flush the TLB (step 4) even if previous step failed  and the dirty
1014  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1015  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1016  * writes will be marked dirty for next log read.
1017  *
1018  *   1. Take a snapshot of the bit and clear it if needed.
1019  *   2. Write protect the corresponding page.
1020  *   3. Copy the snapshot to the userspace.
1021  *   4. Flush TLB's if needed.
1022  */
1023 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1024 {
1025         bool is_dirty = false;
1026         int r;
1027
1028         mutex_lock(&kvm->slots_lock);
1029
1030         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1031
1032         if (is_dirty)
1033                 kvm_flush_remote_tlbs(kvm);
1034
1035         mutex_unlock(&kvm->slots_lock);
1036         return r;
1037 }
1038
1039 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1040                                         struct kvm_arm_device_addr *dev_addr)
1041 {
1042         unsigned long dev_id, type;
1043
1044         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1045                 KVM_ARM_DEVICE_ID_SHIFT;
1046         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1047                 KVM_ARM_DEVICE_TYPE_SHIFT;
1048
1049         switch (dev_id) {
1050         case KVM_ARM_DEVICE_VGIC_V2:
1051                 if (!vgic_present)
1052                         return -ENXIO;
1053                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1054         default:
1055                 return -ENODEV;
1056         }
1057 }
1058
1059 long kvm_arch_vm_ioctl(struct file *filp,
1060                        unsigned int ioctl, unsigned long arg)
1061 {
1062         struct kvm *kvm = filp->private_data;
1063         void __user *argp = (void __user *)arg;
1064
1065         switch (ioctl) {
1066         case KVM_CREATE_IRQCHIP: {
1067                 int ret;
1068                 if (!vgic_present)
1069                         return -ENXIO;
1070                 mutex_lock(&kvm->lock);
1071                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1072                 mutex_unlock(&kvm->lock);
1073                 return ret;
1074         }
1075         case KVM_ARM_SET_DEVICE_ADDR: {
1076                 struct kvm_arm_device_addr dev_addr;
1077
1078                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1079                         return -EFAULT;
1080                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1081         }
1082         case KVM_ARM_PREFERRED_TARGET: {
1083                 int err;
1084                 struct kvm_vcpu_init init;
1085
1086                 err = kvm_vcpu_preferred_target(&init);
1087                 if (err)
1088                         return err;
1089
1090                 if (copy_to_user(argp, &init, sizeof(init)))
1091                         return -EFAULT;
1092
1093                 return 0;
1094         }
1095         default:
1096                 return -EINVAL;
1097         }
1098 }
1099
1100 static void cpu_init_hyp_mode(void *dummy)
1101 {
1102         phys_addr_t pgd_ptr;
1103         unsigned long hyp_stack_ptr;
1104         unsigned long stack_page;
1105         unsigned long vector_ptr;
1106
1107         /* Switch from the HYP stub to our own HYP init vector */
1108         __hyp_set_vectors(kvm_get_idmap_vector());
1109
1110         pgd_ptr = kvm_mmu_get_httbr();
1111         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1112         hyp_stack_ptr = stack_page + PAGE_SIZE;
1113         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1114
1115         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1116         __cpu_init_stage2();
1117
1118         if (is_kernel_in_hyp_mode())
1119                 kvm_timer_init_vhe();
1120
1121         kvm_arm_init_debug();
1122 }
1123
1124 static void cpu_hyp_reset(void)
1125 {
1126         if (!is_kernel_in_hyp_mode())
1127                 __hyp_reset_vectors();
1128 }
1129
1130 static void cpu_hyp_reinit(void)
1131 {
1132         cpu_hyp_reset();
1133
1134         if (is_kernel_in_hyp_mode()) {
1135                 /*
1136                  * __cpu_init_stage2() is safe to call even if the PM
1137                  * event was cancelled before the CPU was reset.
1138                  */
1139                 __cpu_init_stage2();
1140         } else {
1141                 cpu_init_hyp_mode(NULL);
1142         }
1143
1144         if (vgic_present)
1145                 kvm_vgic_init_cpu_hardware();
1146 }
1147
1148 static void _kvm_arch_hardware_enable(void *discard)
1149 {
1150         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1151                 cpu_hyp_reinit();
1152                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1153         }
1154 }
1155
1156 int kvm_arch_hardware_enable(void)
1157 {
1158         _kvm_arch_hardware_enable(NULL);
1159         return 0;
1160 }
1161
1162 static void _kvm_arch_hardware_disable(void *discard)
1163 {
1164         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1165                 cpu_hyp_reset();
1166                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1167         }
1168 }
1169
1170 void kvm_arch_hardware_disable(void)
1171 {
1172         _kvm_arch_hardware_disable(NULL);
1173 }
1174
1175 #ifdef CONFIG_CPU_PM
1176 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1177                                     unsigned long cmd,
1178                                     void *v)
1179 {
1180         /*
1181          * kvm_arm_hardware_enabled is left with its old value over
1182          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1183          * re-enable hyp.
1184          */
1185         switch (cmd) {
1186         case CPU_PM_ENTER:
1187                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1188                         /*
1189                          * don't update kvm_arm_hardware_enabled here
1190                          * so that the hardware will be re-enabled
1191                          * when we resume. See below.
1192                          */
1193                         cpu_hyp_reset();
1194
1195                 return NOTIFY_OK;
1196         case CPU_PM_EXIT:
1197                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1198                         /* The hardware was enabled before suspend. */
1199                         cpu_hyp_reinit();
1200
1201                 return NOTIFY_OK;
1202
1203         default:
1204                 return NOTIFY_DONE;
1205         }
1206 }
1207
1208 static struct notifier_block hyp_init_cpu_pm_nb = {
1209         .notifier_call = hyp_init_cpu_pm_notifier,
1210 };
1211
1212 static void __init hyp_cpu_pm_init(void)
1213 {
1214         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1215 }
1216 static void __init hyp_cpu_pm_exit(void)
1217 {
1218         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1219 }
1220 #else
1221 static inline void hyp_cpu_pm_init(void)
1222 {
1223 }
1224 static inline void hyp_cpu_pm_exit(void)
1225 {
1226 }
1227 #endif
1228
1229 static void teardown_common_resources(void)
1230 {
1231         free_percpu(kvm_host_cpu_state);
1232 }
1233
1234 static int init_common_resources(void)
1235 {
1236         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1237         if (!kvm_host_cpu_state) {
1238                 kvm_err("Cannot allocate host CPU state\n");
1239                 return -ENOMEM;
1240         }
1241
1242         /* set size of VMID supported by CPU */
1243         kvm_vmid_bits = kvm_get_vmid_bits();
1244         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1245
1246         return 0;
1247 }
1248
1249 static int init_subsystems(void)
1250 {
1251         int err = 0;
1252
1253         /*
1254          * Enable hardware so that subsystem initialisation can access EL2.
1255          */
1256         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1257
1258         /*
1259          * Register CPU lower-power notifier
1260          */
1261         hyp_cpu_pm_init();
1262
1263         /*
1264          * Init HYP view of VGIC
1265          */
1266         err = kvm_vgic_hyp_init();
1267         switch (err) {
1268         case 0:
1269                 vgic_present = true;
1270                 break;
1271         case -ENODEV:
1272         case -ENXIO:
1273                 vgic_present = false;
1274                 err = 0;
1275                 break;
1276         default:
1277                 goto out;
1278         }
1279
1280         /*
1281          * Init HYP architected timer support
1282          */
1283         err = kvm_timer_hyp_init();
1284         if (err)
1285                 goto out;
1286
1287         kvm_perf_init();
1288         kvm_coproc_table_init();
1289
1290 out:
1291         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1292
1293         return err;
1294 }
1295
1296 static void teardown_hyp_mode(void)
1297 {
1298         int cpu;
1299
1300         if (is_kernel_in_hyp_mode())
1301                 return;
1302
1303         free_hyp_pgds();
1304         for_each_possible_cpu(cpu)
1305                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1306         hyp_cpu_pm_exit();
1307 }
1308
1309 static int init_vhe_mode(void)
1310 {
1311         kvm_info("VHE mode initialized successfully\n");
1312         return 0;
1313 }
1314
1315 /**
1316  * Inits Hyp-mode on all online CPUs
1317  */
1318 static int init_hyp_mode(void)
1319 {
1320         int cpu;
1321         int err = 0;
1322
1323         /*
1324          * Allocate Hyp PGD and setup Hyp identity mapping
1325          */
1326         err = kvm_mmu_init();
1327         if (err)
1328                 goto out_err;
1329
1330         /*
1331          * Allocate stack pages for Hypervisor-mode
1332          */
1333         for_each_possible_cpu(cpu) {
1334                 unsigned long stack_page;
1335
1336                 stack_page = __get_free_page(GFP_KERNEL);
1337                 if (!stack_page) {
1338                         err = -ENOMEM;
1339                         goto out_err;
1340                 }
1341
1342                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1343         }
1344
1345         /*
1346          * Map the Hyp-code called directly from the host
1347          */
1348         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1349                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1350         if (err) {
1351                 kvm_err("Cannot map world-switch code\n");
1352                 goto out_err;
1353         }
1354
1355         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1356                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1357         if (err) {
1358                 kvm_err("Cannot map rodata section\n");
1359                 goto out_err;
1360         }
1361
1362         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1363                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1364         if (err) {
1365                 kvm_err("Cannot map bss section\n");
1366                 goto out_err;
1367         }
1368
1369         /*
1370          * Map the Hyp stack pages
1371          */
1372         for_each_possible_cpu(cpu) {
1373                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1374                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1375                                           PAGE_HYP);
1376
1377                 if (err) {
1378                         kvm_err("Cannot map hyp stack\n");
1379                         goto out_err;
1380                 }
1381         }
1382
1383         for_each_possible_cpu(cpu) {
1384                 kvm_cpu_context_t *cpu_ctxt;
1385
1386                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1387                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1388
1389                 if (err) {
1390                         kvm_err("Cannot map host CPU state: %d\n", err);
1391                         goto out_err;
1392                 }
1393         }
1394
1395         kvm_info("Hyp mode initialized successfully\n");
1396
1397         return 0;
1398
1399 out_err:
1400         teardown_hyp_mode();
1401         kvm_err("error initializing Hyp mode: %d\n", err);
1402         return err;
1403 }
1404
1405 static void check_kvm_target_cpu(void *ret)
1406 {
1407         *(int *)ret = kvm_target_cpu();
1408 }
1409
1410 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1411 {
1412         struct kvm_vcpu *vcpu;
1413         int i;
1414
1415         mpidr &= MPIDR_HWID_BITMASK;
1416         kvm_for_each_vcpu(i, vcpu, kvm) {
1417                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1418                         return vcpu;
1419         }
1420         return NULL;
1421 }
1422
1423 /**
1424  * Initialize Hyp-mode and memory mappings on all CPUs.
1425  */
1426 int kvm_arch_init(void *opaque)
1427 {
1428         int err;
1429         int ret, cpu;
1430
1431         if (!is_hyp_mode_available()) {
1432                 kvm_err("HYP mode not available\n");
1433                 return -ENODEV;
1434         }
1435
1436         for_each_online_cpu(cpu) {
1437                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1438                 if (ret < 0) {
1439                         kvm_err("Error, CPU %d not supported!\n", cpu);
1440                         return -ENODEV;
1441                 }
1442         }
1443
1444         err = init_common_resources();
1445         if (err)
1446                 return err;
1447
1448         if (is_kernel_in_hyp_mode())
1449                 err = init_vhe_mode();
1450         else
1451                 err = init_hyp_mode();
1452         if (err)
1453                 goto out_err;
1454
1455         err = init_subsystems();
1456         if (err)
1457                 goto out_hyp;
1458
1459         return 0;
1460
1461 out_hyp:
1462         teardown_hyp_mode();
1463 out_err:
1464         teardown_common_resources();
1465         return err;
1466 }
1467
1468 /* NOP: Compiling as a module not supported */
1469 void kvm_arch_exit(void)
1470 {
1471         kvm_perf_teardown();
1472 }
1473
1474 static int arm_init(void)
1475 {
1476         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1477         return rc;
1478 }
1479
1480 module_init(arm_init);