virt/kvm/arm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu_pm.h>
  20 #include <linux/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31 #include <kvm/arm_pmu.h>
  32 #include <kvm/arm_psci.h>
  33
  34 #define CREATE_TRACE_POINTS
  35 #include "trace.h"
  36
  37 #include <linux/uaccess.h>
  38 #include <asm/ptrace.h>
  39 #include <asm/mman.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/cacheflush.h>
  42 #include <asm/virt.h>
  43 #include <asm/kvm_arm.h>
  44 #include <asm/kvm_asm.h>
  45 #include <asm/kvm_mmu.h>
  46 #include <asm/kvm_emulate.h>
  47 #include <asm/kvm_coproc.h>
  48 #include <asm/sections.h>
  49
  50 #ifdef REQUIRES_VIRT
  51 __asm__(".arch_extension        virt");
  52 #endif
  53
  54 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
 616         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 617                 return -ENOEXEC;
 618
 619         ret = kvm_vcpu_first_run_init(vcpu);
 620         if (ret)
 621                 return ret;
 622
 623         if (run->exit_reason == KVM_EXIT_MMIO) {
 624                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 625                 if (ret)
 626                         return ret;
 627         }
 628
 629         if (run->immediate_exit)
 630                 return -EINTR;
 631
 632         kvm_sigset_activate(vcpu);
 633
 634         ret = 1;
 635         run->exit_reason = KVM_EXIT_UNKNOWN;
 636         while (ret > 0) {
 637                 /*
 638                  * Check conditions before entering the guest
 639                  */
 640                 cond_resched();
 641
 642                 update_vttbr(vcpu->kvm);
 643
 644                 check_vcpu_requests(vcpu);
 645
 646                 /*
 647                  * Preparing the interrupts to be injected also
 648                  * involves poking the GIC, which must be done in a
 649                  * non-preemptible context.
 650                  */
 651                 preempt_disable();
 652
 653                 kvm_pmu_flush_hwstate(vcpu);
 654
 655                 kvm_timer_flush_hwstate(vcpu);
 656                 kvm_vgic_flush_hwstate(vcpu);
 657
 658                 local_irq_disable();
 659
 660                 /*
 661                  * If we have a singal pending, or need to notify a userspace
 662                  * irqchip about timer or PMU level changes, then we exit (and
 663                  * update the timer level state in kvm_timer_update_run
 664                  * below).
 665                  */
 666                 if (signal_pending(current) ||
 667                     kvm_timer_should_notify_user(vcpu) ||
 668                     kvm_pmu_should_notify_user(vcpu)) {
 669                         ret = -EINTR;
 670                         run->exit_reason = KVM_EXIT_INTR;
 671                 }
 672
 673                 /*
 674                  * Ensure we set mode to IN_GUEST_MODE after we disable
 675                  * interrupts and before the final VCPU requests check.
 676                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 677                  * Documentation/virtual/kvm/vcpu-requests.rst
 678                  */
 679                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 680
 681                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 682                     kvm_request_pending(vcpu)) {
 683                         vcpu->mode = OUTSIDE_GUEST_MODE;
 684                         local_irq_enable();
 685                         kvm_pmu_sync_hwstate(vcpu);
 686                         kvm_timer_sync_hwstate(vcpu);
 687                         kvm_vgic_sync_hwstate(vcpu);
 688                         preempt_enable();
 689                         continue;
 690                 }
 691
 692                 kvm_arm_setup_debug(vcpu);
 693
 694                 /**************************************************************
 695                  * Enter the guest
 696                  */
 697                 trace_kvm_entry(*vcpu_pc(vcpu));
 698                 guest_enter_irqoff();
 699
 700                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 701
 702                 vcpu->mode = OUTSIDE_GUEST_MODE;
 703                 vcpu->stat.exits++;
 704                 /*
 705                  * Back from guest
 706                  *************************************************************/
 707
 708                 kvm_arm_clear_debug(vcpu);
 709
 710                 /*
 711                  * We may have taken a host interrupt in HYP mode (ie
 712                  * while executing the guest). This interrupt is still
 713                  * pending, as we haven't serviced it yet!
 714                  *
 715                  * We're now back in SVC mode, with interrupts
 716                  * disabled.  Enabling the interrupts now will have
 717                  * the effect of taking the interrupt again, in SVC
 718                  * mode this time.
 719                  */
 720                 local_irq_enable();
 721
 722                 /*
 723                  * We do local_irq_enable() before calling guest_exit() so
 724                  * that if a timer interrupt hits while running the guest we
 725                  * account that tick as being spent in the guest.  We enable
 726                  * preemption after calling guest_exit() so that if we get
 727                  * preempted we make sure ticks after that is not counted as
 728                  * guest time.
 729                  */
 730                 guest_exit();
 731                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 732
 733                 /*
 734                  * We must sync the PMU and timer state before the vgic state so
 735                  * that the vgic can properly sample the updated state of the
 736                  * interrupt line.
 737                  */
 738                 kvm_pmu_sync_hwstate(vcpu);
 739                 kvm_timer_sync_hwstate(vcpu);
 740
 741                 kvm_vgic_sync_hwstate(vcpu);
 742
 743                 preempt_enable();
 744
 745                 ret = handle_exit(vcpu, run, ret);
 746         }
 747
 748         /* Tell userspace about in-kernel device output levels */
 749         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 750                 kvm_timer_update_run(vcpu);
 751                 kvm_pmu_update_run(vcpu);
 752         }
 753
 754         kvm_sigset_deactivate(vcpu);
 755
 756         return ret;
 757 }
 758
 759 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 760 {
 761         int bit_index;
 762         bool set;
 763         unsigned long *ptr;
 764
 765         if (number == KVM_ARM_IRQ_CPU_IRQ)
 766                 bit_index = __ffs(HCR_VI);
 767         else /* KVM_ARM_IRQ_CPU_FIQ */
 768                 bit_index = __ffs(HCR_VF);
 769
 770         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 771         if (level)
 772                 set = test_and_set_bit(bit_index, ptr);
 773         else
 774                 set = test_and_clear_bit(bit_index, ptr);
 775
 776         /*
 777          * If we didn't change anything, no need to wake up or kick other CPUs
 778          */
 779         if (set == level)
 780                 return 0;
 781
 782         /*
 783          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 784          * trigger a world-switch round on the running physical CPU to set the
 785          * virtual IRQ/FIQ fields in the HCR appropriately.
 786          */
 787         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 788         kvm_vcpu_kick(vcpu);
 789
 790         return 0;
 791 }
 792
 793 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 794                           bool line_status)
 795 {
 796         u32 irq = irq_level->irq;
 797         unsigned int irq_type, vcpu_idx, irq_num;
 798         int nrcpus = atomic_read(&kvm->online_vcpus);
 799         struct kvm_vcpu *vcpu = NULL;
 800         bool level = irq_level->level;
 801
 802         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 803         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 804         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 805
 806         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 807
 808         switch (irq_type) {
 809         case KVM_ARM_IRQ_TYPE_CPU:
 810                 if (irqchip_in_kernel(kvm))
 811                         return -ENXIO;
 812
 813                 if (vcpu_idx >= nrcpus)
 814                         return -EINVAL;
 815
 816                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 817                 if (!vcpu)
 818                         return -EINVAL;
 819
 820                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 821                         return -EINVAL;
 822
 823                 return vcpu_interrupt_line(vcpu, irq_num, level);
 824         case KVM_ARM_IRQ_TYPE_PPI:
 825                 if (!irqchip_in_kernel(kvm))
 826                         return -ENXIO;
 827
 828                 if (vcpu_idx >= nrcpus)
 829                         return -EINVAL;
 830
 831                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 832                 if (!vcpu)
 833                         return -EINVAL;
 834
 835                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 836                         return -EINVAL;
 837
 838                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 839         case KVM_ARM_IRQ_TYPE_SPI:
 840                 if (!irqchip_in_kernel(kvm))
 841                         return -ENXIO;
 842
 843                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 844                         return -EINVAL;
 845
 846                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 847         }
 848
 849         return -EINVAL;
 850 }
 851
 852 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 853                                const struct kvm_vcpu_init *init)
 854 {
 855         unsigned int i;
 856         int phys_target = kvm_target_cpu();
 857
 858         if (init->target != phys_target)
 859                 return -EINVAL;
 860
 861         /*
 862          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 863          * use the same target.
 864          */
 865         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 866                 return -EINVAL;
 867
 868         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 869         for (i = 0; i < sizeof(init->features) * 8; i++) {
 870                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 871
 872                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 873                         return -ENOENT;
 874
 875                 /*
 876                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 877                  * use the same feature set.
 878                  */
 879                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 880                     test_bit(i, vcpu->arch.features) != set)
 881                         return -EINVAL;
 882
 883                 if (set)
 884                         set_bit(i, vcpu->arch.features);
 885         }
 886
 887         vcpu->arch.target = phys_target;
 888
 889         /* Now we know what it is, we can reset it. */
 890         return kvm_reset_vcpu(vcpu);
 891 }
 892
 893
 894 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 895                                          struct kvm_vcpu_init *init)
 896 {
 897         int ret;
 898
 899         ret = kvm_vcpu_set_target(vcpu, init);
 900         if (ret)
 901                 return ret;
 902
 903         /*
 904          * Ensure a rebooted VM will fault in RAM pages and detect if the
 905          * guest MMU is turned off and flush the caches as needed.
 906          */
 907         if (vcpu->arch.has_run_once)
 908                 stage2_unmap_vm(vcpu->kvm);
 909
 910         vcpu_reset_hcr(vcpu);
 911
 912         /*
 913          * Handle the "start in power-off" case.
 914          */
 915         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 916                 vcpu_power_off(vcpu);
 917         else
 918                 vcpu->arch.power_off = false;
 919
 920         return 0;
 921 }
 922
 923 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 924                                  struct kvm_device_attr *attr)
 925 {
 926         int ret = -ENXIO;
 927
 928         switch (attr->group) {
 929         default:
 930                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 931                 break;
 932         }
 933
 934         return ret;
 935 }
 936
 937 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 938                                  struct kvm_device_attr *attr)
 939 {
 940         int ret = -ENXIO;
 941
 942         switch (attr->group) {
 943         default:
 944                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
 945                 break;
 946         }
 947
 948         return ret;
 949 }
 950
 951 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
 952                                  struct kvm_device_attr *attr)
 953 {
 954         int ret = -ENXIO;
 955
 956         switch (attr->group) {
 957         default:
 958                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
 959                 break;
 960         }
 961
 962         return ret;
 963 }
 964
 965 long kvm_arch_vcpu_ioctl(struct file *filp,
 966                          unsigned int ioctl, unsigned long arg)
 967 {
 968         struct kvm_vcpu *vcpu = filp->private_data;
 969         void __user *argp = (void __user *)arg;
 970         struct kvm_device_attr attr;
 971
 972         switch (ioctl) {
 973         case KVM_ARM_VCPU_INIT: {
 974                 struct kvm_vcpu_init init;
 975
 976                 if (copy_from_user(&init, argp, sizeof(init)))
 977                         return -EFAULT;
 978
 979                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 980         }
 981         case KVM_SET_ONE_REG:
 982         case KVM_GET_ONE_REG: {
 983                 struct kvm_one_reg reg;
 984
 985                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 986                         return -ENOEXEC;
 987
 988                 if (copy_from_user(&reg, argp, sizeof(reg)))
 989                         return -EFAULT;
 990                 if (ioctl == KVM_SET_ONE_REG)
 991                         return kvm_arm_set_reg(vcpu, &reg);
 992                 else
 993                         return kvm_arm_get_reg(vcpu, &reg);
 994         }
 995         case KVM_GET_REG_LIST: {
 996                 struct kvm_reg_list __user *user_list = argp;
 997                 struct kvm_reg_list reg_list;
 998                 unsigned n;
 999
1000                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1001                         return -ENOEXEC;
1002
1003                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1004                         return -EFAULT;
1005                 n = reg_list.n;
1006                 reg_list.n = kvm_arm_num_regs(vcpu);
1007                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1008                         return -EFAULT;
1009                 if (n < reg_list.n)
1010                         return -E2BIG;
1011                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1012         }
1013         case KVM_SET_DEVICE_ATTR: {
1014                 if (copy_from_user(&attr, argp, sizeof(attr)))
1015                         return -EFAULT;
1016                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1017         }
1018         case KVM_GET_DEVICE_ATTR: {
1019                 if (copy_from_user(&attr, argp, sizeof(attr)))
1020                         return -EFAULT;
1021                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1022         }
1023         case KVM_HAS_DEVICE_ATTR: {
1024                 if (copy_from_user(&attr, argp, sizeof(attr)))
1025                         return -EFAULT;
1026                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1027         }
1028         default:
1029                 return -EINVAL;
1030         }
1031 }
1032
1033 /**
1034  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1035  * @kvm: kvm instance
1036  * @log: slot id and address to which we copy the log
1037  *
1038  * Steps 1-4 below provide general overview of dirty page logging. See
1039  * kvm_get_dirty_log_protect() function description for additional details.
1040  *
1041  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1042  * always flush the TLB (step 4) even if previous step failed  and the dirty
1043  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1044  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1045  * writes will be marked dirty for next log read.
1046  *
1047  *   1. Take a snapshot of the bit and clear it if needed.
1048  *   2. Write protect the corresponding page.
1049  *   3. Copy the snapshot to the userspace.
1050  *   4. Flush TLB's if needed.
1051  */
1052 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1053 {
1054         bool is_dirty = false;
1055         int r;
1056
1057         mutex_lock(&kvm->slots_lock);
1058
1059         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1060
1061         if (is_dirty)
1062                 kvm_flush_remote_tlbs(kvm);
1063
1064         mutex_unlock(&kvm->slots_lock);
1065         return r;
1066 }
1067
1068 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1069                                         struct kvm_arm_device_addr *dev_addr)
1070 {
1071         unsigned long dev_id, type;
1072
1073         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1074                 KVM_ARM_DEVICE_ID_SHIFT;
1075         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1076                 KVM_ARM_DEVICE_TYPE_SHIFT;
1077
1078         switch (dev_id) {
1079         case KVM_ARM_DEVICE_VGIC_V2:
1080                 if (!vgic_present)
1081                         return -ENXIO;
1082                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1083         default:
1084                 return -ENODEV;
1085         }
1086 }
1087
1088 long kvm_arch_vm_ioctl(struct file *filp,
1089                        unsigned int ioctl, unsigned long arg)
1090 {
1091         struct kvm *kvm = filp->private_data;
1092         void __user *argp = (void __user *)arg;
1093
1094         switch (ioctl) {
1095         case KVM_CREATE_IRQCHIP: {
1096                 int ret;
1097                 if (!vgic_present)
1098                         return -ENXIO;
1099                 mutex_lock(&kvm->lock);
1100                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1101                 mutex_unlock(&kvm->lock);
1102                 return ret;
1103         }
1104         case KVM_ARM_SET_DEVICE_ADDR: {
1105                 struct kvm_arm_device_addr dev_addr;
1106
1107                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1108                         return -EFAULT;
1109                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1110         }
1111         case KVM_ARM_PREFERRED_TARGET: {
1112                 int err;
1113                 struct kvm_vcpu_init init;
1114
1115                 err = kvm_vcpu_preferred_target(&init);
1116                 if (err)
1117                         return err;
1118
1119                 if (copy_to_user(argp, &init, sizeof(init)))
1120                         return -EFAULT;
1121
1122                 return 0;
1123         }
1124         default:
1125                 return -EINVAL;
1126         }
1127 }
1128
1129 static void cpu_init_hyp_mode(void *dummy)
1130 {
1131         phys_addr_t pgd_ptr;
1132         unsigned long hyp_stack_ptr;
1133         unsigned long stack_page;
1134         unsigned long vector_ptr;
1135
1136         /* Switch from the HYP stub to our own HYP init vector */
1137         __hyp_set_vectors(kvm_get_idmap_vector());
1138
1139         pgd_ptr = kvm_mmu_get_httbr();
1140         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1141         hyp_stack_ptr = stack_page + PAGE_SIZE;
1142         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1143
1144         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1145         __cpu_init_stage2();
1146
1147         kvm_arm_init_debug();
1148 }
1149
1150 static void cpu_hyp_reset(void)
1151 {
1152         if (!is_kernel_in_hyp_mode())
1153                 __hyp_reset_vectors();
1154 }
1155
1156 static void cpu_hyp_reinit(void)
1157 {
1158         cpu_hyp_reset();
1159
1160         if (is_kernel_in_hyp_mode()) {
1161                 /*
1162                  * __cpu_init_stage2() is safe to call even if the PM
1163                  * event was cancelled before the CPU was reset.
1164                  */
1165                 __cpu_init_stage2();
1166                 kvm_timer_init_vhe();
1167         } else {
1168                 cpu_init_hyp_mode(NULL);
1169         }
1170
1171         if (vgic_present)
1172                 kvm_vgic_init_cpu_hardware();
1173 }
1174
1175 static void _kvm_arch_hardware_enable(void *discard)
1176 {
1177         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1178                 cpu_hyp_reinit();
1179                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1180         }
1181 }
1182
1183 int kvm_arch_hardware_enable(void)
1184 {
1185         _kvm_arch_hardware_enable(NULL);
1186         return 0;
1187 }
1188
1189 static void _kvm_arch_hardware_disable(void *discard)
1190 {
1191         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1192                 cpu_hyp_reset();
1193                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1194         }
1195 }
1196
1197 void kvm_arch_hardware_disable(void)
1198 {
1199         _kvm_arch_hardware_disable(NULL);
1200 }
1201
1202 #ifdef CONFIG_CPU_PM
1203 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1204                                     unsigned long cmd,
1205                                     void *v)
1206 {
1207         /*
1208          * kvm_arm_hardware_enabled is left with its old value over
1209          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1210          * re-enable hyp.
1211          */
1212         switch (cmd) {
1213         case CPU_PM_ENTER:
1214                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1215                         /*
1216                          * don't update kvm_arm_hardware_enabled here
1217                          * so that the hardware will be re-enabled
1218                          * when we resume. See below.
1219                          */
1220                         cpu_hyp_reset();
1221
1222                 return NOTIFY_OK;
1223         case CPU_PM_EXIT:
1224                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1225                         /* The hardware was enabled before suspend. */
1226                         cpu_hyp_reinit();
1227
1228                 return NOTIFY_OK;
1229
1230         default:
1231                 return NOTIFY_DONE;
1232         }
1233 }
1234
1235 static struct notifier_block hyp_init_cpu_pm_nb = {
1236         .notifier_call = hyp_init_cpu_pm_notifier,
1237 };
1238
1239 static void __init hyp_cpu_pm_init(void)
1240 {
1241         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1242 }
1243 static void __init hyp_cpu_pm_exit(void)
1244 {
1245         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1246 }
1247 #else
1248 static inline void hyp_cpu_pm_init(void)
1249 {
1250 }
1251 static inline void hyp_cpu_pm_exit(void)
1252 {
1253 }
1254 #endif
1255
1256 static void teardown_common_resources(void)
1257 {
1258         free_percpu(kvm_host_cpu_state);
1259 }
1260
1261 static int init_common_resources(void)
1262 {
1263         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1264         if (!kvm_host_cpu_state) {
1265                 kvm_err("Cannot allocate host CPU state\n");
1266                 return -ENOMEM;
1267         }
1268
1269         /* set size of VMID supported by CPU */
1270         kvm_vmid_bits = kvm_get_vmid_bits();
1271         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1272
1273         return 0;
1274 }
1275
1276 static int init_subsystems(void)
1277 {
1278         int err = 0;
1279
1280         /*
1281          * Enable hardware so that subsystem initialisation can access EL2.
1282          */
1283         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1284
1285         /*
1286          * Register CPU lower-power notifier
1287          */
1288         hyp_cpu_pm_init();
1289
1290         /*
1291          * Init HYP view of VGIC
1292          */
1293         err = kvm_vgic_hyp_init();
1294         switch (err) {
1295         case 0:
1296                 vgic_present = true;
1297                 break;
1298         case -ENODEV:
1299         case -ENXIO:
1300                 vgic_present = false;
1301                 err = 0;
1302                 break;
1303         default:
1304                 goto out;
1305         }
1306
1307         /*
1308          * Init HYP architected timer support
1309          */
1310         err = kvm_timer_hyp_init();
1311         if (err)
1312                 goto out;
1313
1314         kvm_perf_init();
1315         kvm_coproc_table_init();
1316
1317 out:
1318         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1319
1320         return err;
1321 }
1322
1323 static void teardown_hyp_mode(void)
1324 {
1325         int cpu;
1326
1327         free_hyp_pgds();
1328         for_each_possible_cpu(cpu)
1329                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1330         hyp_cpu_pm_exit();
1331 }
1332
1333 /**
1334  * Inits Hyp-mode on all online CPUs
1335  */
1336 static int init_hyp_mode(void)
1337 {
1338         int cpu;
1339         int err = 0;
1340
1341         /*
1342          * Allocate Hyp PGD and setup Hyp identity mapping
1343          */
1344         err = kvm_mmu_init();
1345         if (err)
1346                 goto out_err;
1347
1348         /*
1349          * Allocate stack pages for Hypervisor-mode
1350          */
1351         for_each_possible_cpu(cpu) {
1352                 unsigned long stack_page;
1353
1354                 stack_page = __get_free_page(GFP_KERNEL);
1355                 if (!stack_page) {
1356                         err = -ENOMEM;
1357                         goto out_err;
1358                 }
1359
1360                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1361         }
1362
1363         /*
1364          * Map the Hyp-code called directly from the host
1365          */
1366         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1367                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1368         if (err) {
1369                 kvm_err("Cannot map world-switch code\n");
1370                 goto out_err;
1371         }
1372
1373         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1374                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1375         if (err) {
1376                 kvm_err("Cannot map rodata section\n");
1377                 goto out_err;
1378         }
1379
1380         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1381                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1382         if (err) {
1383                 kvm_err("Cannot map bss section\n");
1384                 goto out_err;
1385         }
1386
1387         err = kvm_map_vectors();
1388         if (err) {
1389                 kvm_err("Cannot map vectors\n");
1390                 goto out_err;
1391         }
1392
1393         /*
1394          * Map the Hyp stack pages
1395          */
1396         for_each_possible_cpu(cpu) {
1397                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1398                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1399                                           PAGE_HYP);
1400
1401                 if (err) {
1402                         kvm_err("Cannot map hyp stack\n");
1403                         goto out_err;
1404                 }
1405         }
1406
1407         for_each_possible_cpu(cpu) {
1408                 kvm_cpu_context_t *cpu_ctxt;
1409
1410                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1411                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1412
1413                 if (err) {
1414                         kvm_err("Cannot map host CPU state: %d\n", err);
1415                         goto out_err;
1416                 }
1417         }
1418
1419         return 0;
1420
1421 out_err:
1422         teardown_hyp_mode();
1423         kvm_err("error initializing Hyp mode: %d\n", err);
1424         return err;
1425 }
1426
1427 static void check_kvm_target_cpu(void *ret)
1428 {
1429         *(int *)ret = kvm_target_cpu();
1430 }
1431
1432 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1433 {
1434         struct kvm_vcpu *vcpu;
1435         int i;
1436
1437         mpidr &= MPIDR_HWID_BITMASK;
1438         kvm_for_each_vcpu(i, vcpu, kvm) {
1439                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1440                         return vcpu;
1441         }
1442         return NULL;
1443 }
1444
1445 /**
1446  * Initialize Hyp-mode and memory mappings on all CPUs.
1447  */
1448 int kvm_arch_init(void *opaque)
1449 {
1450         int err;
1451         int ret, cpu;
1452         bool in_hyp_mode;
1453
1454         if (!is_hyp_mode_available()) {
1455                 kvm_err("HYP mode not available\n");
1456                 return -ENODEV;
1457         }
1458
1459         for_each_online_cpu(cpu) {
1460                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1461                 if (ret < 0) {
1462                         kvm_err("Error, CPU %d not supported!\n", cpu);
1463                         return -ENODEV;
1464                 }
1465         }
1466
1467         err = init_common_resources();
1468         if (err)
1469                 return err;
1470
1471         in_hyp_mode = is_kernel_in_hyp_mode();
1472
1473         if (!in_hyp_mode) {
1474                 err = init_hyp_mode();
1475                 if (err)
1476                         goto out_err;
1477         }
1478
1479         err = init_subsystems();
1480         if (err)
1481                 goto out_hyp;
1482
1483         if (in_hyp_mode)
1484                 kvm_info("VHE mode initialized successfully\n");
1485         else
1486                 kvm_info("Hyp mode initialized successfully\n");
1487
1488         return 0;
1489
1490 out_hyp:
1491         if (!in_hyp_mode)
1492                 teardown_hyp_mode();
1493 out_err:
1494         teardown_common_resources();
1495         return err;
1496 }
1497
1498 /* NOP: Compiling as a module not supported */
1499 void kvm_arch_exit(void)
1500 {
1501         kvm_perf_teardown();
1502 }
1503
1504 static int arm_init(void)
1505 {
1506         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1507         return rc;
1508 }
1509
1510 module_init(arm_init);