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