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