virt/kvm/arm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <[email protected]>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu_pm.h>
  20 #include <linux/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31 #include <kvm/arm_pmu.h>
  32
  33 #define CREATE_TRACE_POINTS
  34 #include "trace.h"
  35
  36 #include <linux/uaccess.h>
  37 #include <asm/ptrace.h>
  38 #include <asm/mman.h>
  39 #include <asm/tlbflush.h>
  40 #include <asm/cacheflush.h>
  41 #include <asm/virt.h>
  42 #include <asm/kvm_arm.h>
  43 #include <asm/kvm_asm.h>
  44 #include <asm/kvm_mmu.h>
  45 #include <asm/kvm_emulate.h>
  46 #include <asm/kvm_coproc.h>
  47 #include <asm/kvm_psci.h>
  48 #include <asm/sections.h>
  49
  50 #ifdef REQUIRES_VIRT
  51 __asm__(".arch_extension        virt");
  52 #endif
  53
  54 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  55 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  56
  57 /* Per-CPU variable containing the currently running vcpu. */
  58 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  59
  60 /* The VMID used in the VTTBR */
  61 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  62 static u32 kvm_next_vmid;
  63 static unsigned int kvm_vmid_bits __read_mostly;
  64 static DEFINE_SPINLOCK(kvm_vmid_lock);
  65
  66 static bool vgic_present;
  67
  68 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  69
  70 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  71 {
  72         BUG_ON(preemptible());
  73         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  74 }
  75
  76 /**
  77  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  78  * Must be called from non-preemptible context
  79  */
  80 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  81 {
  82         BUG_ON(preemptible());
  83         return __this_cpu_read(kvm_arm_running_vcpu);
  84 }
  85
  86 /**
  87  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  88  */
  89 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  90 {
  91         return &kvm_arm_running_vcpu;
  92 }
  93
  94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  95 {
  96         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  97 }
  98
  99 int kvm_arch_hardware_setup(void)
 100 {
 101         return 0;
 102 }
 103
 104 void kvm_arch_check_processor_compat(void *rtn)
 105 {
 106         *(int *)rtn = 0;
 107 }
 108
 109
 110 /**
 111  * kvm_arch_init_vm - initializes a VM data structure
 112  * @kvm:        pointer to the KVM struct
 113  */
 114 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 115 {
 116         int ret, cpu;
 117
 118         if (type)
 119                 return -EINVAL;
 120
 121         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 122         if (!kvm->arch.last_vcpu_ran)
 123                 return -ENOMEM;
 124
 125         for_each_possible_cpu(cpu)
 126                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 127
 128         ret = kvm_alloc_stage2_pgd(kvm);
 129         if (ret)
 130                 goto out_fail_alloc;
 131
 132         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 133         if (ret)
 134                 goto out_free_stage2_pgd;
 135
 136         kvm_vgic_early_init(kvm);
 137
 138         /* Mark the initial VMID generation invalid */
 139         kvm->arch.vmid_gen = 0;
 140
 141         /* The maximum number of VCPUs is limited by the host's GIC model */
 142         kvm->arch.max_vcpus = vgic_present ?
 143                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 144
 145         return ret;
 146 out_free_stage2_pgd:
 147         kvm_free_stage2_pgd(kvm);
 148 out_fail_alloc:
 149         free_percpu(kvm->arch.last_vcpu_ran);
 150         kvm->arch.last_vcpu_ran = NULL;
 151         return ret;
 152 }
 153
 154 bool kvm_arch_has_vcpu_debugfs(void)
 155 {
 156         return false;
 157 }
 158
 159 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 160 {
 161         return 0;
 162 }
 163
 164 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 165 {
 166         return VM_FAULT_SIGBUS;
 167 }
 168
 169
 170 /**
 171  * kvm_arch_destroy_vm - destroy the VM data structure
 172  * @kvm:        pointer to the KVM struct
 173  */
 174 void kvm_arch_destroy_vm(struct kvm *kvm)
 175 {
 176         int i;
 177
 178         free_percpu(kvm->arch.last_vcpu_ran);
 179         kvm->arch.last_vcpu_ran = NULL;
 180
 181         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 182                 if (kvm->vcpus[i]) {
 183                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 184                         kvm->vcpus[i] = NULL;
 185                 }
 186         }
 187
 188         kvm_vgic_destroy(kvm);
 189 }
 190
 191 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 192 {
 193         int r;
 194         switch (ext) {
 195         case KVM_CAP_IRQCHIP:
 196                 r = vgic_present;
 197                 break;
 198         case KVM_CAP_IOEVENTFD:
 199         case KVM_CAP_DEVICE_CTRL:
 200         case KVM_CAP_USER_MEMORY:
 201         case KVM_CAP_SYNC_MMU:
 202         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 203         case KVM_CAP_ONE_REG:
 204         case KVM_CAP_ARM_PSCI:
 205         case KVM_CAP_ARM_PSCI_0_2:
 206         case KVM_CAP_READONLY_MEM:
 207         case KVM_CAP_MP_STATE:
 208         case KVM_CAP_IMMEDIATE_EXIT:
 209                 r = 1;
 210                 break;
 211         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 212                 r = 1;
 213                 break;
 214         case KVM_CAP_NR_VCPUS:
 215                 r = num_online_cpus();
 216                 break;
 217         case KVM_CAP_MAX_VCPUS:
 218                 r = KVM_MAX_VCPUS;
 219                 break;
 220         case KVM_CAP_NR_MEMSLOTS:
 221                 r = KVM_USER_MEM_SLOTS;
 222                 break;
 223         case KVM_CAP_MSI_DEVID:
 224                 if (!kvm)
 225                         r = -EINVAL;
 226                 else
 227                         r = kvm->arch.vgic.msis_require_devid;
 228                 break;
 229         case KVM_CAP_ARM_USER_IRQ:
 230                 /*
 231                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 232                  * (bump this number if adding more devices)
 233                  */
 234                 r = 1;
 235                 break;
 236         default:
 237                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 238                 break;
 239         }
 240         return r;
 241 }
 242
 243 long kvm_arch_dev_ioctl(struct file *filp,
 244                         unsigned int ioctl, unsigned long arg)
 245 {
 246         return -EINVAL;
 247 }
 248
 249
 250 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 251 {
 252         int err;
 253         struct kvm_vcpu *vcpu;
 254
 255         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 256                 err = -EBUSY;
 257                 goto out;
 258         }
 259
 260         if (id >= kvm->arch.max_vcpus) {
 261                 err = -EINVAL;
 262                 goto out;
 263         }
 264
 265         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 266         if (!vcpu) {
 267                 err = -ENOMEM;
 268                 goto out;
 269         }
 270
 271         err = kvm_vcpu_init(vcpu, kvm, id);
 272         if (err)
 273                 goto free_vcpu;
 274
 275         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 276         if (err)
 277                 goto vcpu_uninit;
 278
 279         return vcpu;
 280 vcpu_uninit:
 281         kvm_vcpu_uninit(vcpu);
 282 free_vcpu:
 283         kmem_cache_free(kvm_vcpu_cache, vcpu);
 284 out:
 285         return ERR_PTR(err);
 286 }
 287
 288 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 289 {
 290         kvm_vgic_vcpu_early_init(vcpu);
 291 }
 292
 293 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 294 {
 295         kvm_mmu_free_memory_caches(vcpu);
 296         kvm_timer_vcpu_terminate(vcpu);
 297         kvm_vgic_vcpu_destroy(vcpu);
 298         kvm_pmu_vcpu_destroy(vcpu);
 299         kvm_vcpu_uninit(vcpu);
 300         kmem_cache_free(kvm_vcpu_cache, vcpu);
 301 }
 302
 303 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 304 {
 305         kvm_arch_vcpu_free(vcpu);
 306 }
 307
 308 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 309 {
 310         return kvm_timer_should_fire(vcpu_vtimer(vcpu)) ||
 311                kvm_timer_should_fire(vcpu_ptimer(vcpu));
 312 }
 313
 314 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 315 {
 316         kvm_timer_schedule(vcpu);
 317 }
 318
 319 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 320 {
 321         kvm_timer_unschedule(vcpu);
 322 }
 323
 324 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 325 {
 326         /* Force users to call KVM_ARM_VCPU_INIT */
 327         vcpu->arch.target = -1;
 328         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 329
 330         /* Set up the timer */
 331         kvm_timer_vcpu_init(vcpu);
 332
 333         kvm_arm_reset_debug_ptr(vcpu);
 334
 335         return kvm_vgic_vcpu_init(vcpu);
 336 }
 337
 338 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 339 {
 340         int *last_ran;
 341
 342         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 343
 344         /*
 345          * We might get preempted before the vCPU actually runs, but
 346          * over-invalidation doesn't affect correctness.
 347          */
 348         if (*last_ran != vcpu->vcpu_id) {
 349                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 350                 *last_ran = vcpu->vcpu_id;
 351         }
 352
 353         vcpu->cpu = cpu;
 354         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 355
 356         kvm_arm_set_running_vcpu(vcpu);
 357
 358         kvm_vgic_load(vcpu);
 359 }
 360
 361 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 362 {
 363         kvm_vgic_put(vcpu);
 364
 365         vcpu->cpu = -1;
 366
 367         kvm_arm_set_running_vcpu(NULL);
 368         kvm_timer_vcpu_put(vcpu);
 369 }
 370
 371 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 372 {
 373         vcpu->arch.power_off = true;
 374         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 375         kvm_vcpu_kick(vcpu);
 376 }
 377
 378 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 379                                     struct kvm_mp_state *mp_state)
 380 {
 381         if (vcpu->arch.power_off)
 382                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 383         else
 384                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 385
 386         return 0;
 387 }
 388
 389 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 390                                     struct kvm_mp_state *mp_state)
 391 {
 392         switch (mp_state->mp_state) {
 393         case KVM_MP_STATE_RUNNABLE:
 394                 vcpu->arch.power_off = false;
 395                 break;
 396         case KVM_MP_STATE_STOPPED:
 397                 vcpu_power_off(vcpu);
 398                 break;
 399         default:
 400                 return -EINVAL;
 401         }
 402
 403         return 0;
 404 }
 405
 406 /**
 407  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 408  * @v:          The VCPU pointer
 409  *
 410  * If the guest CPU is not waiting for interrupts or an interrupt line is
 411  * asserted, the CPU is by definition runnable.
 412  */
 413 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 414 {
 415         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
 416                 && !v->arch.power_off && !v->arch.pause);
 417 }
 418
 419 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 420 {
 421         return vcpu_mode_priv(vcpu);
 422 }
 423
 424 /* Just ensure a guest exit from a particular CPU */
 425 static void exit_vm_noop(void *info)
 426 {
 427 }
 428
 429 void force_vm_exit(const cpumask_t *mask)
 430 {
 431         preempt_disable();
 432         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 433         preempt_enable();
 434 }
 435
 436 /**
 437  * need_new_vmid_gen - check that the VMID is still valid
 438  * @kvm: The VM's VMID to check
 439  *
 440  * return true if there is a new generation of VMIDs being used
 441  *
 442  * The hardware supports only 256 values with the value zero reserved for the
 443  * host, so we check if an assigned value belongs to a previous generation,
 444  * which which requires us to assign a new value. If we're the first to use a
 445  * VMID for the new generation, we must flush necessary caches and TLBs on all
 446  * CPUs.
 447  */
 448 static bool need_new_vmid_gen(struct kvm *kvm)
 449 {
 450         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 451 }
 452
 453 /**
 454  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 455  * @kvm The guest that we are about to run
 456  *
 457  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 458  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 459  * caches and TLBs.
 460  */
 461 static void update_vttbr(struct kvm *kvm)
 462 {
 463         phys_addr_t pgd_phys;
 464         u64 vmid;
 465
 466         if (!need_new_vmid_gen(kvm))
 467                 return;
 468
 469         spin_lock(&kvm_vmid_lock);
 470
 471         /*
 472          * We need to re-check the vmid_gen here to ensure that if another vcpu
 473          * already allocated a valid vmid for this vm, then this vcpu should
 474          * use the same vmid.
 475          */
 476         if (!need_new_vmid_gen(kvm)) {
 477                 spin_unlock(&kvm_vmid_lock);
 478                 return;
 479         }
 480
 481         /* First user of a new VMID generation? */
 482         if (unlikely(kvm_next_vmid == 0)) {
 483                 atomic64_inc(&kvm_vmid_gen);
 484                 kvm_next_vmid = 1;
 485
 486                 /*
 487                  * On SMP we know no other CPUs can use this CPU's or each
 488                  * other's VMID after force_vm_exit returns since the
 489                  * kvm_vmid_lock blocks them from reentry to the guest.
 490                  */
 491                 force_vm_exit(cpu_all_mask);
 492                 /*
 493                  * Now broadcast TLB + ICACHE invalidation over the inner
 494                  * shareable domain to make sure all data structures are
 495                  * clean.
 496                  */
 497                 kvm_call_hyp(__kvm_flush_vm_context);
 498         }
 499
 500         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 501         kvm->arch.vmid = kvm_next_vmid;
 502         kvm_next_vmid++;
 503         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 504
 505         /* update vttbr to be used with the new vmid */
 506         pgd_phys = virt_to_phys(kvm->arch.pgd);
 507         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 508         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 509         kvm->arch.vttbr = pgd_phys | vmid;
 510
 511         spin_unlock(&kvm_vmid_lock);
 512 }
 513
 514 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 515 {
 516         struct kvm *kvm = vcpu->kvm;
 517         int ret = 0;
 518
 519         if (likely(vcpu->arch.has_run_once))
 520                 return 0;
 521
 522         vcpu->arch.has_run_once = true;
 523
 524         /*
 525          * Map the VGIC hardware resources before running a vcpu the first
 526          * time on this VM.
 527          */
 528         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
 529                 ret = kvm_vgic_map_resources(kvm);
 530                 if (ret)
 531                         return ret;
 532         }
 533
 534         ret = kvm_timer_enable(vcpu);
 535         if (ret)
 536                 return ret;
 537
 538         ret = kvm_arm_pmu_v3_enable(vcpu);
 539
 540         return ret;
 541 }
 542
 543 bool kvm_arch_intc_initialized(struct kvm *kvm)
 544 {
 545         return vgic_initialized(kvm);
 546 }
 547
 548 void kvm_arm_halt_guest(struct kvm *kvm)
 549 {
 550         int i;
 551         struct kvm_vcpu *vcpu;
 552
 553         kvm_for_each_vcpu(i, vcpu, kvm)
 554                 vcpu->arch.pause = true;
 555         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 556 }
 557
 558 void kvm_arm_resume_guest(struct kvm *kvm)
 559 {
 560         int i;
 561         struct kvm_vcpu *vcpu;
 562
 563         kvm_for_each_vcpu(i, vcpu, kvm) {
 564                 vcpu->arch.pause = false;
 565                 swake_up(kvm_arch_vcpu_wq(vcpu));
 566         }
 567 }
 568
 569 static void vcpu_req_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         if (vcpu->arch.power_off || vcpu->arch.pause) {
 577                 /* Awaken to handle a signal, request we sleep again later. */
 578                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 579         }
 580 }
 581
 582 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 583 {
 584         return vcpu->arch.target >= 0;
 585 }
 586
 587 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 588 {
 589         if (kvm_request_pending(vcpu)) {
 590                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 591                         vcpu_req_sleep(vcpu);
 592
 593                 /*
 594                  * Clear IRQ_PENDING requests that were made to guarantee
 595                  * that a VCPU sees new virtual interrupts.
 596                  */
 597                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 598         }
 599 }
 600
 601 /**
 602  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 603  * @vcpu:       The VCPU pointer
 604  * @run:        The kvm_run structure pointer used for userspace state exchange
 605  *
 606  * This function is called through the VCPU_RUN ioctl called from user space. It
 607  * will execute VM code in a loop until the time slice for the process is used
 608  * or some emulation is needed from user space in which case the function will
 609  * return with return value 0 and with the kvm_run structure filled in with the
 610  * required data for the requested emulation.
 611  */
 612 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 613 {
 614         int ret;
 615         sigset_t sigsaved;
 616
 617         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 618                 return -ENOEXEC;
 619
 620         ret = kvm_vcpu_first_run_init(vcpu);
 621         if (ret)
 622                 return ret;
 623
 624         if (run->exit_reason == KVM_EXIT_MMIO) {
 625                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 626                 if (ret)
 627                         return ret;
 628         }
 629
 630         if (run->immediate_exit)
 631                 return -EINTR;
 632
 633         if (vcpu->sigset_active)
 634                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 635
 636         ret = 1;
 637         run->exit_reason = KVM_EXIT_UNKNOWN;
 638         while (ret > 0) {
 639                 /*
 640                  * Check conditions before entering the guest
 641                  */
 642                 cond_resched();
 643
 644                 update_vttbr(vcpu->kvm);
 645
 646                 check_vcpu_requests(vcpu);
 647
 648                 /*
 649                  * Preparing the interrupts to be injected also
 650                  * involves poking the GIC, which must be done in a
 651                  * non-preemptible context.
 652                  */
 653                 preempt_disable();
 654
 655                 kvm_pmu_flush_hwstate(vcpu);
 656
 657                 kvm_timer_flush_hwstate(vcpu);
 658                 kvm_vgic_flush_hwstate(vcpu);
 659
 660                 local_irq_disable();
 661
 662                 /*
 663                  * If we have a singal pending, or need to notify a userspace
 664                  * irqchip about timer or PMU level changes, then we exit (and
 665                  * update the timer level state in kvm_timer_update_run
 666                  * below).
 667                  */
 668                 if (signal_pending(current) ||
 669                     kvm_timer_should_notify_user(vcpu) ||
 670                     kvm_pmu_should_notify_user(vcpu)) {
 671                         ret = -EINTR;
 672                         run->exit_reason = KVM_EXIT_INTR;
 673                 }
 674
 675                 /*
 676                  * Ensure we set mode to IN_GUEST_MODE after we disable
 677                  * interrupts and before the final VCPU requests check.
 678                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 679                  * Documentation/virtual/kvm/vcpu-requests.rst
 680                  */
 681                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 682
 683                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 684                     kvm_request_pending(vcpu)) {
 685                         vcpu->mode = OUTSIDE_GUEST_MODE;
 686                         local_irq_enable();
 687                         kvm_pmu_sync_hwstate(vcpu);
 688                         kvm_timer_sync_hwstate(vcpu);
 689                         kvm_vgic_sync_hwstate(vcpu);
 690                         preempt_enable();
 691                         continue;
 692                 }
 693
 694                 kvm_arm_setup_debug(vcpu);
 695
 696                 /**************************************************************
 697                  * Enter the guest
 698                  */
 699                 trace_kvm_entry(*vcpu_pc(vcpu));
 700                 guest_enter_irqoff();
 701
 702                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 703
 704                 vcpu->mode = OUTSIDE_GUEST_MODE;
 705                 vcpu->stat.exits++;
 706                 /*
 707                  * Back from guest
 708                  *************************************************************/
 709
 710                 kvm_arm_clear_debug(vcpu);
 711
 712                 /*
 713                  * We may have taken a host interrupt in HYP mode (ie
 714                  * while executing the guest). This interrupt is still
 715                  * pending, as we haven't serviced it yet!
 716                  *
 717                  * We're now back in SVC mode, with interrupts
 718                  * disabled.  Enabling the interrupts now will have
 719                  * the effect of taking the interrupt again, in SVC
 720                  * mode this time.
 721                  */
 722                 local_irq_enable();
 723
 724                 /*
 725                  * We do local_irq_enable() before calling guest_exit() so
 726                  * that if a timer interrupt hits while running the guest we
 727                  * account that tick as being spent in the guest.  We enable
 728                  * preemption after calling guest_exit() so that if we get
 729                  * preempted we make sure ticks after that is not counted as
 730                  * guest time.
 731                  */
 732                 guest_exit();
 733                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 734
 735                 /*
 736                  * We must sync the PMU and timer state before the vgic state so
 737                  * that the vgic can properly sample the updated state of the
 738                  * interrupt line.
 739                  */
 740                 kvm_pmu_sync_hwstate(vcpu);
 741                 kvm_timer_sync_hwstate(vcpu);
 742
 743                 kvm_vgic_sync_hwstate(vcpu);
 744
 745                 preempt_enable();
 746
 747                 ret = handle_exit(vcpu, run, ret);
 748         }
 749
 750         /* Tell userspace about in-kernel device output levels */
 751         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 752                 kvm_timer_update_run(vcpu);
 753                 kvm_pmu_update_run(vcpu);
 754         }
 755
 756         if (vcpu->sigset_active)
 757                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 758         return ret;
 759 }
 760
 761 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 762 {
 763         int bit_index;
 764         bool set;
 765         unsigned long *ptr;
 766
 767         if (number == KVM_ARM_IRQ_CPU_IRQ)
 768                 bit_index = __ffs(HCR_VI);
 769         else /* KVM_ARM_IRQ_CPU_FIQ */
 770                 bit_index = __ffs(HCR_VF);
 771
 772         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 773         if (level)
 774                 set = test_and_set_bit(bit_index, ptr);
 775         else
 776                 set = test_and_clear_bit(bit_index, ptr);
 777
 778         /*
 779          * If we didn't change anything, no need to wake up or kick other CPUs
 780          */
 781         if (set == level)
 782                 return 0;
 783
 784         /*
 785          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 786          * trigger a world-switch round on the running physical CPU to set the
 787          * virtual IRQ/FIQ fields in the HCR appropriately.
 788          */
 789         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 790         kvm_vcpu_kick(vcpu);
 791
 792         return 0;
 793 }
 794
 795 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 796                           bool line_status)
 797 {
 798         u32 irq = irq_level->irq;
 799         unsigned int irq_type, vcpu_idx, irq_num;
 800         int nrcpus = atomic_read(&kvm->online_vcpus);
 801         struct kvm_vcpu *vcpu = NULL;
 802         bool level = irq_level->level;
 803
 804         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 805         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 806         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 807
 808         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 809
 810         switch (irq_type) {
 811         case KVM_ARM_IRQ_TYPE_CPU:
 812                 if (irqchip_in_kernel(kvm))
 813                         return -ENXIO;
 814
 815                 if (vcpu_idx >= nrcpus)
 816                         return -EINVAL;
 817
 818                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 819                 if (!vcpu)
 820                         return -EINVAL;
 821
 822                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 823                         return -EINVAL;
 824
 825                 return vcpu_interrupt_line(vcpu, irq_num, level);
 826         case KVM_ARM_IRQ_TYPE_PPI:
 827                 if (!irqchip_in_kernel(kvm))
 828                         return -ENXIO;
 829
 830                 if (vcpu_idx >= nrcpus)
 831                         return -EINVAL;
 832
 833                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 834                 if (!vcpu)
 835                         return -EINVAL;
 836
 837                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 838                         return -EINVAL;
 839
 840                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 841         case KVM_ARM_IRQ_TYPE_SPI:
 842                 if (!irqchip_in_kernel(kvm))
 843                         return -ENXIO;
 844
 845                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 846                         return -EINVAL;
 847
 848                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 849         }
 850
 851         return -EINVAL;
 852 }
 853
 854 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 855                                const struct kvm_vcpu_init *init)
 856 {
 857         unsigned int i;
 858         int phys_target = kvm_target_cpu();
 859
 860         if (init->target != phys_target)
 861                 return -EINVAL;
 862
 863         /*
 864          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 865          * use the same target.
 866          */
 867         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 868                 return -EINVAL;
 869
 870         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 871         for (i = 0; i < sizeof(init->features) * 8; i++) {
 872                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 873
 874                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 875                         return -ENOENT;
 876
 877                 /*
 878                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 879                  * use the same feature set.
 880                  */
 881                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 882                     test_bit(i, vcpu->arch.features) != set)
 883                         return -EINVAL;
 884
 885                 if (set)
 886                         set_bit(i, vcpu->arch.features);
 887         }
 888
 889         vcpu->arch.target = phys_target;
 890
 891         /* Now we know what it is, we can reset it. */
 892         return kvm_reset_vcpu(vcpu);
 893 }
 894
 895
 896 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 897                                          struct kvm_vcpu_init *init)
 898 {
 899         int ret;
 900
 901         ret = kvm_vcpu_set_target(vcpu, init);
 902         if (ret)
 903                 return ret;
 904
 905         /*
 906          * Ensure a rebooted VM will fault in RAM pages and detect if the
 907          * guest MMU is turned off and flush the caches as needed.
 908          */
 909         if (vcpu->arch.has_run_once)
 910                 stage2_unmap_vm(vcpu->kvm);
 911
 912         vcpu_reset_hcr(vcpu);
 913
 914         /*
 915          * Handle the "start in power-off" case.
 916          */
 917         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 918                 vcpu_power_off(vcpu);
 919         else
 920                 vcpu->arch.power_off = false;
 921
 922         return 0;
 923 }
 924
 925 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 926                                  struct kvm_device_attr *attr)
 927 {
 928         int ret = -ENXIO;
 929
 930         switch (attr->group) {
 931         default:
 932                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 933                 break;
 934         }
 935
 936         return ret;
 937 }
 938
 939 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 940                                  struct kvm_device_attr *attr)
 941 {
 942         int ret = -ENXIO;
 943
 944         switch (attr->group) {
 945         default:
 946                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
 947                 break;
 948         }
 949
 950         return ret;
 951 }
 952
 953 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
 954                                  struct kvm_device_attr *attr)
 955 {
 956         int ret = -ENXIO;
 957
 958         switch (attr->group) {
 959         default:
 960                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
 961                 break;
 962         }
 963
 964         return ret;
 965 }
 966
 967 long kvm_arch_vcpu_ioctl(struct file *filp,
 968                          unsigned int ioctl, unsigned long arg)
 969 {
 970         struct kvm_vcpu *vcpu = filp->private_data;
 971         void __user *argp = (void __user *)arg;
 972         struct kvm_device_attr attr;
 973
 974         switch (ioctl) {
 975         case KVM_ARM_VCPU_INIT: {
 976                 struct kvm_vcpu_init init;
 977
 978                 if (copy_from_user(&init, argp, sizeof(init)))
 979                         return -EFAULT;
 980
 981                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 982         }
 983         case KVM_SET_ONE_REG:
 984         case KVM_GET_ONE_REG: {
 985                 struct kvm_one_reg reg;
 986
 987                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 988                         return -ENOEXEC;
 989
 990                 if (copy_from_user(&reg, argp, sizeof(reg)))
 991                         return -EFAULT;
 992                 if (ioctl == KVM_SET_ONE_REG)
 993                         return kvm_arm_set_reg(vcpu, &reg);
 994                 else
 995                         return kvm_arm_get_reg(vcpu, &reg);
 996         }
 997         case KVM_GET_REG_LIST: {
 998                 struct kvm_reg_list __user *user_list = argp;
 999                 struct kvm_reg_list reg_list;
1000                 unsigned n;
1001
1002                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1003                         return -ENOEXEC;
1004
1005                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1006                         return -EFAULT;
1007                 n = reg_list.n;
1008                 reg_list.n = kvm_arm_num_regs(vcpu);
1009                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1010                         return -EFAULT;
1011                 if (n < reg_list.n)
1012                         return -E2BIG;
1013                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1014         }
1015         case KVM_SET_DEVICE_ATTR: {
1016                 if (copy_from_user(&attr, argp, sizeof(attr)))
1017                         return -EFAULT;
1018                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1019         }
1020         case KVM_GET_DEVICE_ATTR: {
1021                 if (copy_from_user(&attr, argp, sizeof(attr)))
1022                         return -EFAULT;
1023                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1024         }
1025         case KVM_HAS_DEVICE_ATTR: {
1026                 if (copy_from_user(&attr, argp, sizeof(attr)))
1027                         return -EFAULT;
1028                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1029         }
1030         default:
1031                 return -EINVAL;
1032         }
1033 }
1034
1035 /**
1036  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1037  * @kvm: kvm instance
1038  * @log: slot id and address to which we copy the log
1039  *
1040  * Steps 1-4 below provide general overview of dirty page logging. See
1041  * kvm_get_dirty_log_protect() function description for additional details.
1042  *
1043  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1044  * always flush the TLB (step 4) even if previous step failed  and the dirty
1045  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1046  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1047  * writes will be marked dirty for next log read.
1048  *
1049  *   1. Take a snapshot of the bit and clear it if needed.
1050  *   2. Write protect the corresponding page.
1051  *   3. Copy the snapshot to the userspace.
1052  *   4. Flush TLB's if needed.
1053  */
1054 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1055 {
1056         bool is_dirty = false;
1057         int r;
1058
1059         mutex_lock(&kvm->slots_lock);
1060
1061         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1062
1063         if (is_dirty)
1064                 kvm_flush_remote_tlbs(kvm);
1065
1066         mutex_unlock(&kvm->slots_lock);
1067         return r;
1068 }
1069
1070 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1071                                         struct kvm_arm_device_addr *dev_addr)
1072 {
1073         unsigned long dev_id, type;
1074
1075         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1076                 KVM_ARM_DEVICE_ID_SHIFT;
1077         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1078                 KVM_ARM_DEVICE_TYPE_SHIFT;
1079
1080         switch (dev_id) {
1081         case KVM_ARM_DEVICE_VGIC_V2:
1082                 if (!vgic_present)
1083                         return -ENXIO;
1084                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1085         default:
1086                 return -ENODEV;
1087         }
1088 }
1089
1090 long kvm_arch_vm_ioctl(struct file *filp,
1091                        unsigned int ioctl, unsigned long arg)
1092 {
1093         struct kvm *kvm = filp->private_data;
1094         void __user *argp = (void __user *)arg;
1095
1096         switch (ioctl) {
1097         case KVM_CREATE_IRQCHIP: {
1098                 int ret;
1099                 if (!vgic_present)
1100                         return -ENXIO;
1101                 mutex_lock(&kvm->lock);
1102                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1103                 mutex_unlock(&kvm->lock);
1104                 return ret;
1105         }
1106         case KVM_ARM_SET_DEVICE_ADDR: {
1107                 struct kvm_arm_device_addr dev_addr;
1108
1109                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1110                         return -EFAULT;
1111                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1112         }
1113         case KVM_ARM_PREFERRED_TARGET: {
1114                 int err;
1115                 struct kvm_vcpu_init init;
1116
1117                 err = kvm_vcpu_preferred_target(&init);
1118                 if (err)
1119                         return err;
1120
1121                 if (copy_to_user(argp, &init, sizeof(init)))
1122                         return -EFAULT;
1123
1124                 return 0;
1125         }
1126         default:
1127                 return -EINVAL;
1128         }
1129 }
1130
1131 static void cpu_init_hyp_mode(void *dummy)
1132 {
1133         phys_addr_t pgd_ptr;
1134         unsigned long hyp_stack_ptr;
1135         unsigned long stack_page;
1136         unsigned long vector_ptr;
1137
1138         /* Switch from the HYP stub to our own HYP init vector */
1139         __hyp_set_vectors(kvm_get_idmap_vector());
1140
1141         pgd_ptr = kvm_mmu_get_httbr();
1142         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1143         hyp_stack_ptr = stack_page + PAGE_SIZE;
1144         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1145
1146         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1147         __cpu_init_stage2();
1148
1149         kvm_arm_init_debug();
1150 }
1151
1152 static void cpu_hyp_reset(void)
1153 {
1154         if (!is_kernel_in_hyp_mode())
1155                 __hyp_reset_vectors();
1156 }
1157
1158 static void cpu_hyp_reinit(void)
1159 {
1160         cpu_hyp_reset();
1161
1162         if (is_kernel_in_hyp_mode()) {
1163                 /*
1164                  * __cpu_init_stage2() is safe to call even if the PM
1165                  * event was cancelled before the CPU was reset.
1166                  */
1167                 __cpu_init_stage2();
1168                 kvm_timer_init_vhe();
1169         } else {
1170                 cpu_init_hyp_mode(NULL);
1171         }
1172
1173         if (vgic_present)
1174                 kvm_vgic_init_cpu_hardware();
1175 }
1176
1177 static void _kvm_arch_hardware_enable(void *discard)
1178 {
1179         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1180                 cpu_hyp_reinit();
1181                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1182         }
1183 }
1184
1185 int kvm_arch_hardware_enable(void)
1186 {
1187         _kvm_arch_hardware_enable(NULL);
1188         return 0;
1189 }
1190
1191 static void _kvm_arch_hardware_disable(void *discard)
1192 {
1193         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1194                 cpu_hyp_reset();
1195                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1196         }
1197 }
1198
1199 void kvm_arch_hardware_disable(void)
1200 {
1201         _kvm_arch_hardware_disable(NULL);
1202 }
1203
1204 #ifdef CONFIG_CPU_PM
1205 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1206                                     unsigned long cmd,
1207                                     void *v)
1208 {
1209         /*
1210          * kvm_arm_hardware_enabled is left with its old value over
1211          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1212          * re-enable hyp.
1213          */
1214         switch (cmd) {
1215         case CPU_PM_ENTER:
1216                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1217                         /*
1218                          * don't update kvm_arm_hardware_enabled here
1219                          * so that the hardware will be re-enabled
1220                          * when we resume. See below.
1221                          */
1222                         cpu_hyp_reset();
1223
1224                 return NOTIFY_OK;
1225         case CPU_PM_EXIT:
1226                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1227                         /* The hardware was enabled before suspend. */
1228                         cpu_hyp_reinit();
1229
1230                 return NOTIFY_OK;
1231
1232         default:
1233                 return NOTIFY_DONE;
1234         }
1235 }
1236
1237 static struct notifier_block hyp_init_cpu_pm_nb = {
1238         .notifier_call = hyp_init_cpu_pm_notifier,
1239 };
1240
1241 static void __init hyp_cpu_pm_init(void)
1242 {
1243         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1244 }
1245 static void __init hyp_cpu_pm_exit(void)
1246 {
1247         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1248 }
1249 #else
1250 static inline void hyp_cpu_pm_init(void)
1251 {
1252 }
1253 static inline void hyp_cpu_pm_exit(void)
1254 {
1255 }
1256 #endif
1257
1258 static void teardown_common_resources(void)
1259 {
1260         free_percpu(kvm_host_cpu_state);
1261 }
1262
1263 static int init_common_resources(void)
1264 {
1265         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1266         if (!kvm_host_cpu_state) {
1267                 kvm_err("Cannot allocate host CPU state\n");
1268                 return -ENOMEM;
1269         }
1270
1271         /* set size of VMID supported by CPU */
1272         kvm_vmid_bits = kvm_get_vmid_bits();
1273         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1274
1275         return 0;
1276 }
1277
1278 static int init_subsystems(void)
1279 {
1280         int err = 0;
1281
1282         /*
1283          * Enable hardware so that subsystem initialisation can access EL2.
1284          */
1285         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1286
1287         /*
1288          * Register CPU lower-power notifier
1289          */
1290         hyp_cpu_pm_init();
1291
1292         /*
1293          * Init HYP view of VGIC
1294          */
1295         err = kvm_vgic_hyp_init();
1296         switch (err) {
1297         case 0:
1298                 vgic_present = true;
1299                 break;
1300         case -ENODEV:
1301         case -ENXIO:
1302                 vgic_present = false;
1303                 err = 0;
1304                 break;
1305         default:
1306                 goto out;
1307         }
1308
1309         /*
1310          * Init HYP architected timer support
1311          */
1312         err = kvm_timer_hyp_init();
1313         if (err)
1314                 goto out;
1315
1316         kvm_perf_init();
1317         kvm_coproc_table_init();
1318
1319 out:
1320         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1321
1322         return err;
1323 }
1324
1325 static void teardown_hyp_mode(void)
1326 {
1327         int cpu;
1328
1329         if (is_kernel_in_hyp_mode())
1330                 return;
1331
1332         free_hyp_pgds();
1333         for_each_possible_cpu(cpu)
1334                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1335         hyp_cpu_pm_exit();
1336 }
1337
1338 static int init_vhe_mode(void)
1339 {
1340         kvm_info("VHE mode initialized successfully\n");
1341         return 0;
1342 }
1343
1344 /**
1345  * Inits Hyp-mode on all online CPUs
1346  */
1347 static int init_hyp_mode(void)
1348 {
1349         int cpu;
1350         int err = 0;
1351
1352         /*
1353          * Allocate Hyp PGD and setup Hyp identity mapping
1354          */
1355         err = kvm_mmu_init();
1356         if (err)
1357                 goto out_err;
1358
1359         /*
1360          * Allocate stack pages for Hypervisor-mode
1361          */
1362         for_each_possible_cpu(cpu) {
1363                 unsigned long stack_page;
1364
1365                 stack_page = __get_free_page(GFP_KERNEL);
1366                 if (!stack_page) {
1367                         err = -ENOMEM;
1368                         goto out_err;
1369                 }
1370
1371                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1372         }
1373
1374         /*
1375          * Map the Hyp-code called directly from the host
1376          */
1377         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1378                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1379         if (err) {
1380                 kvm_err("Cannot map world-switch code\n");
1381                 goto out_err;
1382         }
1383
1384         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1385                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1386         if (err) {
1387                 kvm_err("Cannot map rodata section\n");
1388                 goto out_err;
1389         }
1390
1391         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1392                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1393         if (err) {
1394                 kvm_err("Cannot map bss section\n");
1395                 goto out_err;
1396         }
1397
1398         /*
1399          * Map the Hyp stack pages
1400          */
1401         for_each_possible_cpu(cpu) {
1402                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1403                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1404                                           PAGE_HYP);
1405
1406                 if (err) {
1407                         kvm_err("Cannot map hyp stack\n");
1408                         goto out_err;
1409                 }
1410         }
1411
1412         for_each_possible_cpu(cpu) {
1413                 kvm_cpu_context_t *cpu_ctxt;
1414
1415                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1416                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1417
1418                 if (err) {
1419                         kvm_err("Cannot map host CPU state: %d\n", err);
1420                         goto out_err;
1421                 }
1422         }
1423
1424         kvm_info("Hyp mode initialized successfully\n");
1425
1426         return 0;
1427
1428 out_err:
1429         teardown_hyp_mode();
1430         kvm_err("error initializing Hyp mode: %d\n", err);
1431         return err;
1432 }
1433
1434 static void check_kvm_target_cpu(void *ret)
1435 {
1436         *(int *)ret = kvm_target_cpu();
1437 }
1438
1439 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1440 {
1441         struct kvm_vcpu *vcpu;
1442         int i;
1443
1444         mpidr &= MPIDR_HWID_BITMASK;
1445         kvm_for_each_vcpu(i, vcpu, kvm) {
1446                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1447                         return vcpu;
1448         }
1449         return NULL;
1450 }
1451
1452 /**
1453  * Initialize Hyp-mode and memory mappings on all CPUs.
1454  */
1455 int kvm_arch_init(void *opaque)
1456 {
1457         int err;
1458         int ret, cpu;
1459
1460         if (!is_hyp_mode_available()) {
1461                 kvm_err("HYP mode not available\n");
1462                 return -ENODEV;
1463         }
1464
1465         for_each_online_cpu(cpu) {
1466                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1467                 if (ret < 0) {
1468                         kvm_err("Error, CPU %d not supported!\n", cpu);
1469                         return -ENODEV;
1470                 }
1471         }
1472
1473         err = init_common_resources();
1474         if (err)
1475                 return err;
1476
1477         if (is_kernel_in_hyp_mode())
1478                 err = init_vhe_mode();
1479         else
1480                 err = init_hyp_mode();
1481         if (err)
1482                 goto out_err;
1483
1484         err = init_subsystems();
1485         if (err)
1486                 goto out_hyp;
1487
1488         return 0;
1489
1490 out_hyp:
1491         teardown_hyp_mode();
1492 out_err:
1493         teardown_common_resources();
1494         return err;
1495 }
1496
1497 /* NOP: Compiling as a module not supported */
1498 void kvm_arch_exit(void)
1499 {
1500         kvm_perf_teardown();
1501 }
1502
1503 static int arm_init(void)
1504 {
1505         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1506         return rc;
1507 }
1508
1509 module_init(arm_init);