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 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  55 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  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_MAX_VCPU_ID:
 221                 r = KVM_MAX_VCPU_ID;
 222                 break;
 223         case KVM_CAP_NR_MEMSLOTS:
 224                 r = KVM_USER_MEM_SLOTS;
 225                 break;
 226         case KVM_CAP_MSI_DEVID:
 227                 if (!kvm)
 228                         r = -EINVAL;
 229                 else
 230                         r = kvm->arch.vgic.msis_require_devid;
 231                 break;
 232         case KVM_CAP_ARM_USER_IRQ:
 233                 /*
 234                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 235                  * (bump this number if adding more devices)
 236                  */
 237                 r = 1;
 238                 break;
 239         default:
 240                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 241                 break;
 242         }
 243         return r;
 244 }
 245
 246 long kvm_arch_dev_ioctl(struct file *filp,
 247                         unsigned int ioctl, unsigned long arg)
 248 {
 249         return -EINVAL;
 250 }
 251
 252
 253 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 254 {
 255         int err;
 256         struct kvm_vcpu *vcpu;
 257
 258         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 259                 err = -EBUSY;
 260                 goto out;
 261         }
 262
 263         if (id >= kvm->arch.max_vcpus) {
 264                 err = -EINVAL;
 265                 goto out;
 266         }
 267
 268         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 269         if (!vcpu) {
 270                 err = -ENOMEM;
 271                 goto out;
 272         }
 273
 274         err = kvm_vcpu_init(vcpu, kvm, id);
 275         if (err)
 276                 goto free_vcpu;
 277
 278         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 279         if (err)
 280                 goto vcpu_uninit;
 281
 282         return vcpu;
 283 vcpu_uninit:
 284         kvm_vcpu_uninit(vcpu);
 285 free_vcpu:
 286         kmem_cache_free(kvm_vcpu_cache, vcpu);
 287 out:
 288         return ERR_PTR(err);
 289 }
 290
 291 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 292 {
 293         kvm_vgic_vcpu_early_init(vcpu);
 294 }
 295
 296 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 297 {
 298         kvm_mmu_free_memory_caches(vcpu);
 299         kvm_timer_vcpu_terminate(vcpu);
 300         kvm_vgic_vcpu_destroy(vcpu);
 301         kvm_pmu_vcpu_destroy(vcpu);
 302         kvm_vcpu_uninit(vcpu);
 303         kmem_cache_free(kvm_vcpu_cache, vcpu);
 304 }
 305
 306 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 307 {
 308         kvm_arch_vcpu_free(vcpu);
 309 }
 310
 311 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 312 {
 313         return kvm_timer_should_fire(vcpu_vtimer(vcpu)) ||
 314                kvm_timer_should_fire(vcpu_ptimer(vcpu));
 315 }
 316
 317 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 318 {
 319         kvm_timer_schedule(vcpu);
 320         /*
 321          * If we're about to block (most likely because we've just hit a
 322          * WFI), we need to sync back the state of the GIC CPU interface
 323          * so that we have the lastest PMR and group enables. This ensures
 324          * that kvm_arch_vcpu_runnable has up-to-date data to decide
 325          * whether we have pending interrupts.
 326          */
 327         preempt_disable();
 328         kvm_vgic_vmcr_sync(vcpu);
 329         preempt_enable();
 330 }
 331
 332 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 333 {
 334         kvm_timer_unschedule(vcpu);
 335 }
 336
 337 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 338 {
 339         /* Force users to call KVM_ARM_VCPU_INIT */
 340         vcpu->arch.target = -1;
 341         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 342
 343         /* Set up the timer */
 344         kvm_timer_vcpu_init(vcpu);
 345
 346         kvm_arm_reset_debug_ptr(vcpu);
 347
 348         return kvm_vgic_vcpu_init(vcpu);
 349 }
 350
 351 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 352 {
 353         int *last_ran;
 354
 355         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 356
 357         /*
 358          * We might get preempted before the vCPU actually runs, but
 359          * over-invalidation doesn't affect correctness.
 360          */
 361         if (*last_ran != vcpu->vcpu_id) {
 362                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 363                 *last_ran = vcpu->vcpu_id;
 364         }
 365
 366         vcpu->cpu = cpu;
 367         vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
 368
 369         kvm_arm_set_running_vcpu(vcpu);
 370
 371         kvm_vgic_load(vcpu);
 372 }
 373
 374 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 375 {
 376         kvm_vgic_put(vcpu);
 377
 378         vcpu->cpu = -1;
 379
 380         kvm_arm_set_running_vcpu(NULL);
 381         kvm_timer_vcpu_put(vcpu);
 382 }
 383
 384 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 385 {
 386         vcpu->arch.power_off = true;
 387         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 388         kvm_vcpu_kick(vcpu);
 389 }
 390
 391 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 392                                     struct kvm_mp_state *mp_state)
 393 {
 394         if (vcpu->arch.power_off)
 395                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 396         else
 397                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 398
 399         return 0;
 400 }
 401
 402 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 403                                     struct kvm_mp_state *mp_state)
 404 {
 405         switch (mp_state->mp_state) {
 406         case KVM_MP_STATE_RUNNABLE:
 407                 vcpu->arch.power_off = false;
 408                 break;
 409         case KVM_MP_STATE_STOPPED:
 410                 vcpu_power_off(vcpu);
 411                 break;
 412         default:
 413                 return -EINVAL;
 414         }
 415
 416         return 0;
 417 }
 418
 419 /**
 420  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 421  * @v:          The VCPU pointer
 422  *
 423  * If the guest CPU is not waiting for interrupts or an interrupt line is
 424  * asserted, the CPU is by definition runnable.
 425  */
 426 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 427 {
 428         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
 429                 && !v->arch.power_off && !v->arch.pause);
 430 }
 431
 432 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 433 {
 434         return vcpu_mode_priv(vcpu);
 435 }
 436
 437 /* Just ensure a guest exit from a particular CPU */
 438 static void exit_vm_noop(void *info)
 439 {
 440 }
 441
 442 void force_vm_exit(const cpumask_t *mask)
 443 {
 444         preempt_disable();
 445         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 446         preempt_enable();
 447 }
 448
 449 /**
 450  * need_new_vmid_gen - check that the VMID is still valid
 451  * @kvm: The VM's VMID to check
 452  *
 453  * return true if there is a new generation of VMIDs being used
 454  *
 455  * The hardware supports only 256 values with the value zero reserved for the
 456  * host, so we check if an assigned value belongs to a previous generation,
 457  * which which requires us to assign a new value. If we're the first to use a
 458  * VMID for the new generation, we must flush necessary caches and TLBs on all
 459  * CPUs.
 460  */
 461 static bool need_new_vmid_gen(struct kvm *kvm)
 462 {
 463         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 464         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 465         return unlikely(READ_ONCE(kvm->arch.vmid_gen) != current_vmid_gen);
 466 }
 467
 468 /**
 469  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 470  * @kvm The guest that we are about to run
 471  *
 472  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 473  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 474  * caches and TLBs.
 475  */
 476 static void update_vttbr(struct kvm *kvm)
 477 {
 478         phys_addr_t pgd_phys;
 479         u64 vmid;
 480
 481         if (!need_new_vmid_gen(kvm))
 482                 return;
 483
 484         spin_lock(&kvm_vmid_lock);
 485
 486         /*
 487          * We need to re-check the vmid_gen here to ensure that if another vcpu
 488          * already allocated a valid vmid for this vm, then this vcpu should
 489          * use the same vmid.
 490          */
 491         if (!need_new_vmid_gen(kvm)) {
 492                 spin_unlock(&kvm_vmid_lock);
 493                 return;
 494         }
 495
 496         /* First user of a new VMID generation? */
 497         if (unlikely(kvm_next_vmid == 0)) {
 498                 atomic64_inc(&kvm_vmid_gen);
 499                 kvm_next_vmid = 1;
 500
 501                 /*
 502                  * On SMP we know no other CPUs can use this CPU's or each
 503                  * other's VMID after force_vm_exit returns since the
 504                  * kvm_vmid_lock blocks them from reentry to the guest.
 505                  */
 506                 force_vm_exit(cpu_all_mask);
 507                 /*
 508                  * Now broadcast TLB + ICACHE invalidation over the inner
 509                  * shareable domain to make sure all data structures are
 510                  * clean.
 511                  */
 512                 kvm_call_hyp(__kvm_flush_vm_context);
 513         }
 514
 515         kvm->arch.vmid = kvm_next_vmid;
 516         kvm_next_vmid++;
 517         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 518
 519         /* update vttbr to be used with the new vmid */
 520         pgd_phys = virt_to_phys(kvm->arch.pgd);
 521         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 522         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 523         kvm->arch.vttbr = pgd_phys | vmid;
 524
 525         smp_wmb();
 526         WRITE_ONCE(kvm->arch.vmid_gen, atomic64_read(&kvm_vmid_gen));
 527
 528         spin_unlock(&kvm_vmid_lock);
 529 }
 530
 531 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 532 {
 533         struct kvm *kvm = vcpu->kvm;
 534         int ret = 0;
 535
 536         if (likely(vcpu->arch.has_run_once))
 537                 return 0;
 538
 539         vcpu->arch.has_run_once = true;
 540
 541         /*
 542          * Map the VGIC hardware resources before running a vcpu the first
 543          * time on this VM.
 544          */
 545         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
 546                 ret = kvm_vgic_map_resources(kvm);
 547                 if (ret)
 548                         return ret;
 549         }
 550
 551         ret = kvm_timer_enable(vcpu);
 552         if (ret)
 553                 return ret;
 554
 555         ret = kvm_arm_pmu_v3_enable(vcpu);
 556
 557         return ret;
 558 }
 559
 560 bool kvm_arch_intc_initialized(struct kvm *kvm)
 561 {
 562         return vgic_initialized(kvm);
 563 }
 564
 565 void kvm_arm_halt_guest(struct kvm *kvm)
 566 {
 567         int i;
 568         struct kvm_vcpu *vcpu;
 569
 570         kvm_for_each_vcpu(i, vcpu, kvm)
 571                 vcpu->arch.pause = true;
 572         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 573 }
 574
 575 void kvm_arm_resume_guest(struct kvm *kvm)
 576 {
 577         int i;
 578         struct kvm_vcpu *vcpu;
 579
 580         kvm_for_each_vcpu(i, vcpu, kvm) {
 581                 vcpu->arch.pause = false;
 582                 swake_up(kvm_arch_vcpu_wq(vcpu));
 583         }
 584 }
 585
 586 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 587 {
 588         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 589
 590         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 591                                        (!vcpu->arch.pause)));
 592
 593         if (vcpu->arch.power_off || vcpu->arch.pause) {
 594                 /* Awaken to handle a signal, request we sleep again later. */
 595                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 596         }
 597 }
 598
 599 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 600 {
 601         return vcpu->arch.target >= 0;
 602 }
 603
 604 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 605 {
 606         if (kvm_request_pending(vcpu)) {
 607                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 608                         vcpu_req_sleep(vcpu);
 609
 610                 /*
 611                  * Clear IRQ_PENDING requests that were made to guarantee
 612                  * that a VCPU sees new virtual interrupts.
 613                  */
 614                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 615         }
 616 }
 617
 618 /**
 619  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 620  * @vcpu:       The VCPU pointer
 621  * @run:        The kvm_run structure pointer used for userspace state exchange
 622  *
 623  * This function is called through the VCPU_RUN ioctl called from user space. It
 624  * will execute VM code in a loop until the time slice for the process is used
 625  * or some emulation is needed from user space in which case the function will
 626  * return with return value 0 and with the kvm_run structure filled in with the
 627  * required data for the requested emulation.
 628  */
 629 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 630 {
 631         int ret;
 632
 633         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 634                 return -ENOEXEC;
 635
 636         ret = kvm_vcpu_first_run_init(vcpu);
 637         if (ret)
 638                 return ret;
 639
 640         if (run->exit_reason == KVM_EXIT_MMIO) {
 641                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 642                 if (ret)
 643                         return ret;
 644         }
 645
 646         if (run->immediate_exit)
 647                 return -EINTR;
 648
 649         kvm_sigset_activate(vcpu);
 650
 651         ret = 1;
 652         run->exit_reason = KVM_EXIT_UNKNOWN;
 653         while (ret > 0) {
 654                 /*
 655                  * Check conditions before entering the guest
 656                  */
 657                 cond_resched();
 658
 659                 update_vttbr(vcpu->kvm);
 660
 661                 check_vcpu_requests(vcpu);
 662
 663                 /*
 664                  * Preparing the interrupts to be injected also
 665                  * involves poking the GIC, which must be done in a
 666                  * non-preemptible context.
 667                  */
 668                 preempt_disable();
 669
 670                 kvm_pmu_flush_hwstate(vcpu);
 671
 672                 kvm_timer_flush_hwstate(vcpu);
 673                 kvm_vgic_flush_hwstate(vcpu);
 674
 675                 local_irq_disable();
 676
 677                 /*
 678                  * If we have a singal pending, or need to notify a userspace
 679                  * irqchip about timer or PMU level changes, then we exit (and
 680                  * update the timer level state in kvm_timer_update_run
 681                  * below).
 682                  */
 683                 if (signal_pending(current) ||
 684                     kvm_timer_should_notify_user(vcpu) ||
 685                     kvm_pmu_should_notify_user(vcpu)) {
 686                         ret = -EINTR;
 687                         run->exit_reason = KVM_EXIT_INTR;
 688                 }
 689
 690                 /*
 691                  * Ensure we set mode to IN_GUEST_MODE after we disable
 692                  * interrupts and before the final VCPU requests check.
 693                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 694                  * Documentation/virtual/kvm/vcpu-requests.rst
 695                  */
 696                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 697
 698                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 699                     kvm_request_pending(vcpu)) {
 700                         vcpu->mode = OUTSIDE_GUEST_MODE;
 701                         local_irq_enable();
 702                         kvm_pmu_sync_hwstate(vcpu);
 703                         kvm_timer_sync_hwstate(vcpu);
 704                         kvm_vgic_sync_hwstate(vcpu);
 705                         preempt_enable();
 706                         continue;
 707                 }
 708
 709                 kvm_arm_setup_debug(vcpu);
 710
 711                 /**************************************************************
 712                  * Enter the guest
 713                  */
 714                 trace_kvm_entry(*vcpu_pc(vcpu));
 715                 guest_enter_irqoff();
 716
 717                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 718
 719                 vcpu->mode = OUTSIDE_GUEST_MODE;
 720                 vcpu->stat.exits++;
 721                 /*
 722                  * Back from guest
 723                  *************************************************************/
 724
 725                 kvm_arm_clear_debug(vcpu);
 726
 727                 /*
 728                  * We may have taken a host interrupt in HYP mode (ie
 729                  * while executing the guest). This interrupt is still
 730                  * pending, as we haven't serviced it yet!
 731                  *
 732                  * We're now back in SVC mode, with interrupts
 733                  * disabled.  Enabling the interrupts now will have
 734                  * the effect of taking the interrupt again, in SVC
 735                  * mode this time.
 736                  */
 737                 local_irq_enable();
 738
 739                 /*
 740                  * We do local_irq_enable() before calling guest_exit() so
 741                  * that if a timer interrupt hits while running the guest we
 742                  * account that tick as being spent in the guest.  We enable
 743                  * preemption after calling guest_exit() so that if we get
 744                  * preempted we make sure ticks after that is not counted as
 745                  * guest time.
 746                  */
 747                 guest_exit();
 748                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 749
 750                 /*
 751                  * We must sync the PMU and timer state before the vgic state so
 752                  * that the vgic can properly sample the updated state of the
 753                  * interrupt line.
 754                  */
 755                 kvm_pmu_sync_hwstate(vcpu);
 756                 kvm_timer_sync_hwstate(vcpu);
 757
 758                 kvm_vgic_sync_hwstate(vcpu);
 759
 760                 preempt_enable();
 761
 762                 ret = handle_exit(vcpu, run, ret);
 763         }
 764
 765         /* Tell userspace about in-kernel device output levels */
 766         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 767                 kvm_timer_update_run(vcpu);
 768                 kvm_pmu_update_run(vcpu);
 769         }
 770
 771         kvm_sigset_deactivate(vcpu);
 772
 773         return ret;
 774 }
 775
 776 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 777 {
 778         int bit_index;
 779         bool set;
 780         unsigned long *ptr;
 781
 782         if (number == KVM_ARM_IRQ_CPU_IRQ)
 783                 bit_index = __ffs(HCR_VI);
 784         else /* KVM_ARM_IRQ_CPU_FIQ */
 785                 bit_index = __ffs(HCR_VF);
 786
 787         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 788         if (level)
 789                 set = test_and_set_bit(bit_index, ptr);
 790         else
 791                 set = test_and_clear_bit(bit_index, ptr);
 792
 793         /*
 794          * If we didn't change anything, no need to wake up or kick other CPUs
 795          */
 796         if (set == level)
 797                 return 0;
 798
 799         /*
 800          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 801          * trigger a world-switch round on the running physical CPU to set the
 802          * virtual IRQ/FIQ fields in the HCR appropriately.
 803          */
 804         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 805         kvm_vcpu_kick(vcpu);
 806
 807         return 0;
 808 }
 809
 810 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 811                           bool line_status)
 812 {
 813         u32 irq = irq_level->irq;
 814         unsigned int irq_type, vcpu_idx, irq_num;
 815         int nrcpus = atomic_read(&kvm->online_vcpus);
 816         struct kvm_vcpu *vcpu = NULL;
 817         bool level = irq_level->level;
 818
 819         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 820         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 821         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 822
 823         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 824
 825         switch (irq_type) {
 826         case KVM_ARM_IRQ_TYPE_CPU:
 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 > KVM_ARM_IRQ_CPU_FIQ)
 838                         return -EINVAL;
 839
 840                 return vcpu_interrupt_line(vcpu, irq_num, level);
 841         case KVM_ARM_IRQ_TYPE_PPI:
 842                 if (!irqchip_in_kernel(kvm))
 843                         return -ENXIO;
 844
 845                 if (vcpu_idx >= nrcpus)
 846                         return -EINVAL;
 847
 848                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 849                 if (!vcpu)
 850                         return -EINVAL;
 851
 852                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 853                         return -EINVAL;
 854
 855                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 856         case KVM_ARM_IRQ_TYPE_SPI:
 857                 if (!irqchip_in_kernel(kvm))
 858                         return -ENXIO;
 859
 860                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 861                         return -EINVAL;
 862
 863                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 864         }
 865
 866         return -EINVAL;
 867 }
 868
 869 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 870                                const struct kvm_vcpu_init *init)
 871 {
 872         unsigned int i, ret;
 873         int phys_target = kvm_target_cpu();
 874
 875         if (init->target != phys_target)
 876                 return -EINVAL;
 877
 878         /*
 879          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 880          * use the same target.
 881          */
 882         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 883                 return -EINVAL;
 884
 885         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 886         for (i = 0; i < sizeof(init->features) * 8; i++) {
 887                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 888
 889                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 890                         return -ENOENT;
 891
 892                 /*
 893                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 894                  * use the same feature set.
 895                  */
 896                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 897                     test_bit(i, vcpu->arch.features) != set)
 898                         return -EINVAL;
 899
 900                 if (set)
 901                         set_bit(i, vcpu->arch.features);
 902         }
 903
 904         vcpu->arch.target = phys_target;
 905
 906         /* Now we know what it is, we can reset it. */
 907         ret = kvm_reset_vcpu(vcpu);
 908         if (ret) {
 909                 vcpu->arch.target = -1;
 910                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 911         }
 912
 913         return ret;
 914 }
 915
 916 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 917                                          struct kvm_vcpu_init *init)
 918 {
 919         int ret;
 920
 921         ret = kvm_vcpu_set_target(vcpu, init);
 922         if (ret)
 923                 return ret;
 924
 925         /*
 926          * Ensure a rebooted VM will fault in RAM pages and detect if the
 927          * guest MMU is turned off and flush the caches as needed.
 928          */
 929         if (vcpu->arch.has_run_once)
 930                 stage2_unmap_vm(vcpu->kvm);
 931
 932         vcpu_reset_hcr(vcpu);
 933
 934         /*
 935          * Handle the "start in power-off" case.
 936          */
 937         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 938                 vcpu_power_off(vcpu);
 939         else
 940                 vcpu->arch.power_off = false;
 941
 942         return 0;
 943 }
 944
 945 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 946                                  struct kvm_device_attr *attr)
 947 {
 948         int ret = -ENXIO;
 949
 950         switch (attr->group) {
 951         default:
 952                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 953                 break;
 954         }
 955
 956         return ret;
 957 }
 958
 959 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 960                                  struct kvm_device_attr *attr)
 961 {
 962         int ret = -ENXIO;
 963
 964         switch (attr->group) {
 965         default:
 966                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
 967                 break;
 968         }
 969
 970         return ret;
 971 }
 972
 973 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
 974                                  struct kvm_device_attr *attr)
 975 {
 976         int ret = -ENXIO;
 977
 978         switch (attr->group) {
 979         default:
 980                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
 981                 break;
 982         }
 983
 984         return ret;
 985 }
 986
 987 long kvm_arch_vcpu_ioctl(struct file *filp,
 988                          unsigned int ioctl, unsigned long arg)
 989 {
 990         struct kvm_vcpu *vcpu = filp->private_data;
 991         void __user *argp = (void __user *)arg;
 992         struct kvm_device_attr attr;
 993
 994         switch (ioctl) {
 995         case KVM_ARM_VCPU_INIT: {
 996                 struct kvm_vcpu_init init;
 997
 998                 if (copy_from_user(&init, argp, sizeof(init)))
 999                         return -EFAULT;
1000
1001                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1002         }
1003         case KVM_SET_ONE_REG:
1004         case KVM_GET_ONE_REG: {
1005                 struct kvm_one_reg reg;
1006
1007                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1008                         return -ENOEXEC;
1009
1010                 if (copy_from_user(&reg, argp, sizeof(reg)))
1011                         return -EFAULT;
1012                 if (ioctl == KVM_SET_ONE_REG)
1013                         return kvm_arm_set_reg(vcpu, &reg);
1014                 else
1015                         return kvm_arm_get_reg(vcpu, &reg);
1016         }
1017         case KVM_GET_REG_LIST: {
1018                 struct kvm_reg_list __user *user_list = argp;
1019                 struct kvm_reg_list reg_list;
1020                 unsigned n;
1021
1022                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1023                         return -ENOEXEC;
1024
1025                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1026                         return -EFAULT;
1027                 n = reg_list.n;
1028                 reg_list.n = kvm_arm_num_regs(vcpu);
1029                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1030                         return -EFAULT;
1031                 if (n < reg_list.n)
1032                         return -E2BIG;
1033                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1034         }
1035         case KVM_SET_DEVICE_ATTR: {
1036                 if (copy_from_user(&attr, argp, sizeof(attr)))
1037                         return -EFAULT;
1038                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1039         }
1040         case KVM_GET_DEVICE_ATTR: {
1041                 if (copy_from_user(&attr, argp, sizeof(attr)))
1042                         return -EFAULT;
1043                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1044         }
1045         case KVM_HAS_DEVICE_ATTR: {
1046                 if (copy_from_user(&attr, argp, sizeof(attr)))
1047                         return -EFAULT;
1048                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1049         }
1050         default:
1051                 return -EINVAL;
1052         }
1053 }
1054
1055 /**
1056  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1057  * @kvm: kvm instance
1058  * @log: slot id and address to which we copy the log
1059  *
1060  * Steps 1-4 below provide general overview of dirty page logging. See
1061  * kvm_get_dirty_log_protect() function description for additional details.
1062  *
1063  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1064  * always flush the TLB (step 4) even if previous step failed  and the dirty
1065  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1066  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1067  * writes will be marked dirty for next log read.
1068  *
1069  *   1. Take a snapshot of the bit and clear it if needed.
1070  *   2. Write protect the corresponding page.
1071  *   3. Copy the snapshot to the userspace.
1072  *   4. Flush TLB's if needed.
1073  */
1074 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1075 {
1076         bool is_dirty = false;
1077         int r;
1078
1079         mutex_lock(&kvm->slots_lock);
1080
1081         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1082
1083         if (is_dirty)
1084                 kvm_flush_remote_tlbs(kvm);
1085
1086         mutex_unlock(&kvm->slots_lock);
1087         return r;
1088 }
1089
1090 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1091                                         struct kvm_arm_device_addr *dev_addr)
1092 {
1093         unsigned long dev_id, type;
1094
1095         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1096                 KVM_ARM_DEVICE_ID_SHIFT;
1097         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1098                 KVM_ARM_DEVICE_TYPE_SHIFT;
1099
1100         switch (dev_id) {
1101         case KVM_ARM_DEVICE_VGIC_V2:
1102                 if (!vgic_present)
1103                         return -ENXIO;
1104                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1105         default:
1106                 return -ENODEV;
1107         }
1108 }
1109
1110 long kvm_arch_vm_ioctl(struct file *filp,
1111                        unsigned int ioctl, unsigned long arg)
1112 {
1113         struct kvm *kvm = filp->private_data;
1114         void __user *argp = (void __user *)arg;
1115
1116         switch (ioctl) {
1117         case KVM_CREATE_IRQCHIP: {
1118                 int ret;
1119                 if (!vgic_present)
1120                         return -ENXIO;
1121                 mutex_lock(&kvm->lock);
1122                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1123                 mutex_unlock(&kvm->lock);
1124                 return ret;
1125         }
1126         case KVM_ARM_SET_DEVICE_ADDR: {
1127                 struct kvm_arm_device_addr dev_addr;
1128
1129                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1130                         return -EFAULT;
1131                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1132         }
1133         case KVM_ARM_PREFERRED_TARGET: {
1134                 int err;
1135                 struct kvm_vcpu_init init;
1136
1137                 err = kvm_vcpu_preferred_target(&init);
1138                 if (err)
1139                         return err;
1140
1141                 if (copy_to_user(argp, &init, sizeof(init)))
1142                         return -EFAULT;
1143
1144                 return 0;
1145         }
1146         default:
1147                 return -EINVAL;
1148         }
1149 }
1150
1151 static void cpu_init_hyp_mode(void *dummy)
1152 {
1153         phys_addr_t pgd_ptr;
1154         unsigned long hyp_stack_ptr;
1155         unsigned long stack_page;
1156         unsigned long vector_ptr;
1157
1158         /* Switch from the HYP stub to our own HYP init vector */
1159         __hyp_set_vectors(kvm_get_idmap_vector());
1160
1161         pgd_ptr = kvm_mmu_get_httbr();
1162         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1163         hyp_stack_ptr = stack_page + PAGE_SIZE;
1164         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1165
1166         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1167         __cpu_init_stage2();
1168 }
1169
1170 static void cpu_hyp_reset(void)
1171 {
1172         if (!is_kernel_in_hyp_mode())
1173                 __hyp_reset_vectors();
1174 }
1175
1176 static void cpu_hyp_reinit(void)
1177 {
1178         cpu_hyp_reset();
1179
1180         if (is_kernel_in_hyp_mode()) {
1181                 /*
1182                  * __cpu_init_stage2() is safe to call even if the PM
1183                  * event was cancelled before the CPU was reset.
1184                  */
1185                 __cpu_init_stage2();
1186                 kvm_timer_init_vhe();
1187         } else {
1188                 cpu_init_hyp_mode(NULL);
1189         }
1190
1191         kvm_arm_init_debug();
1192
1193         if (vgic_present)
1194                 kvm_vgic_init_cpu_hardware();
1195 }
1196
1197 static void _kvm_arch_hardware_enable(void *discard)
1198 {
1199         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1200                 cpu_hyp_reinit();
1201                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1202         }
1203 }
1204
1205 int kvm_arch_hardware_enable(void)
1206 {
1207         _kvm_arch_hardware_enable(NULL);
1208         return 0;
1209 }
1210
1211 static void _kvm_arch_hardware_disable(void *discard)
1212 {
1213         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1214                 cpu_hyp_reset();
1215                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1216         }
1217 }
1218
1219 void kvm_arch_hardware_disable(void)
1220 {
1221         _kvm_arch_hardware_disable(NULL);
1222 }
1223
1224 #ifdef CONFIG_CPU_PM
1225 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1226                                     unsigned long cmd,
1227                                     void *v)
1228 {
1229         /*
1230          * kvm_arm_hardware_enabled is left with its old value over
1231          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1232          * re-enable hyp.
1233          */
1234         switch (cmd) {
1235         case CPU_PM_ENTER:
1236                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1237                         /*
1238                          * don't update kvm_arm_hardware_enabled here
1239                          * so that the hardware will be re-enabled
1240                          * when we resume. See below.
1241                          */
1242                         cpu_hyp_reset();
1243
1244                 return NOTIFY_OK;
1245         case CPU_PM_ENTER_FAILED:
1246         case CPU_PM_EXIT:
1247                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1248                         /* The hardware was enabled before suspend. */
1249                         cpu_hyp_reinit();
1250
1251                 return NOTIFY_OK;
1252
1253         default:
1254                 return NOTIFY_DONE;
1255         }
1256 }
1257
1258 static struct notifier_block hyp_init_cpu_pm_nb = {
1259         .notifier_call = hyp_init_cpu_pm_notifier,
1260 };
1261
1262 static void __init hyp_cpu_pm_init(void)
1263 {
1264         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1265 }
1266 static void __init hyp_cpu_pm_exit(void)
1267 {
1268         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1269 }
1270 #else
1271 static inline void hyp_cpu_pm_init(void)
1272 {
1273 }
1274 static inline void hyp_cpu_pm_exit(void)
1275 {
1276 }
1277 #endif
1278
1279 static int init_common_resources(void)
1280 {
1281         /* set size of VMID supported by CPU */
1282         kvm_vmid_bits = kvm_get_vmid_bits();
1283         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1284
1285         return 0;
1286 }
1287
1288 static int init_subsystems(void)
1289 {
1290         int err = 0;
1291
1292         /*
1293          * Enable hardware so that subsystem initialisation can access EL2.
1294          */
1295         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1296
1297         /*
1298          * Register CPU lower-power notifier
1299          */
1300         hyp_cpu_pm_init();
1301
1302         /*
1303          * Init HYP view of VGIC
1304          */
1305         err = kvm_vgic_hyp_init();
1306         switch (err) {
1307         case 0:
1308                 vgic_present = true;
1309                 break;
1310         case -ENODEV:
1311         case -ENXIO:
1312                 vgic_present = false;
1313                 err = 0;
1314                 break;
1315         default:
1316                 goto out;
1317         }
1318
1319         /*
1320          * Init HYP architected timer support
1321          */
1322         err = kvm_timer_hyp_init();
1323         if (err)
1324                 goto out;
1325
1326         kvm_perf_init();
1327         kvm_coproc_table_init();
1328
1329 out:
1330         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1331
1332         return err;
1333 }
1334
1335 static void teardown_hyp_mode(void)
1336 {
1337         int cpu;
1338
1339         free_hyp_pgds();
1340         for_each_possible_cpu(cpu)
1341                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1342         hyp_cpu_pm_exit();
1343 }
1344
1345 /**
1346  * Inits Hyp-mode on all online CPUs
1347  */
1348 static int init_hyp_mode(void)
1349 {
1350         int cpu;
1351         int err = 0;
1352
1353         /*
1354          * Allocate Hyp PGD and setup Hyp identity mapping
1355          */
1356         err = kvm_mmu_init();
1357         if (err)
1358                 goto out_err;
1359
1360         /*
1361          * Allocate stack pages for Hypervisor-mode
1362          */
1363         for_each_possible_cpu(cpu) {
1364                 unsigned long stack_page;
1365
1366                 stack_page = __get_free_page(GFP_KERNEL);
1367                 if (!stack_page) {
1368                         err = -ENOMEM;
1369                         goto out_err;
1370                 }
1371
1372                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1373         }
1374
1375         /*
1376          * Map the Hyp-code called directly from the host
1377          */
1378         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1379                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1380         if (err) {
1381                 kvm_err("Cannot map world-switch code\n");
1382                 goto out_err;
1383         }
1384
1385         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1386                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1387         if (err) {
1388                 kvm_err("Cannot map rodata section\n");
1389                 goto out_err;
1390         }
1391
1392         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1393                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1394         if (err) {
1395                 kvm_err("Cannot map bss section\n");
1396                 goto out_err;
1397         }
1398
1399         err = kvm_map_vectors();
1400         if (err) {
1401                 kvm_err("Cannot map vectors\n");
1402                 goto out_err;
1403         }
1404
1405         /*
1406          * Map the Hyp stack pages
1407          */
1408         for_each_possible_cpu(cpu) {
1409                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1410                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1411                                           PAGE_HYP);
1412
1413                 if (err) {
1414                         kvm_err("Cannot map hyp stack\n");
1415                         goto out_err;
1416                 }
1417         }
1418
1419         for_each_possible_cpu(cpu) {
1420                 kvm_cpu_context_t *cpu_ctxt;
1421
1422                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1423                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1424
1425                 if (err) {
1426                         kvm_err("Cannot map host CPU state: %d\n", err);
1427                         goto out_err;
1428                 }
1429         }
1430
1431         err = hyp_map_aux_data();
1432         if (err)
1433                 kvm_err("Cannot map host auxilary data: %d\n", err);
1434
1435         return 0;
1436
1437 out_err:
1438         teardown_hyp_mode();
1439         kvm_err("error initializing Hyp mode: %d\n", err);
1440         return err;
1441 }
1442
1443 static void check_kvm_target_cpu(void *ret)
1444 {
1445         *(int *)ret = kvm_target_cpu();
1446 }
1447
1448 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1449 {
1450         struct kvm_vcpu *vcpu;
1451         int i;
1452
1453         mpidr &= MPIDR_HWID_BITMASK;
1454         kvm_for_each_vcpu(i, vcpu, kvm) {
1455                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1456                         return vcpu;
1457         }
1458         return NULL;
1459 }
1460
1461 /**
1462  * Initialize Hyp-mode and memory mappings on all CPUs.
1463  */
1464 int kvm_arch_init(void *opaque)
1465 {
1466         int err;
1467         int ret, cpu;
1468         bool in_hyp_mode;
1469
1470         if (!is_hyp_mode_available()) {
1471                 kvm_info("HYP mode not available\n");
1472                 return -ENODEV;
1473         }
1474
1475         for_each_online_cpu(cpu) {
1476                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1477                 if (ret < 0) {
1478                         kvm_err("Error, CPU %d not supported!\n", cpu);
1479                         return -ENODEV;
1480                 }
1481         }
1482
1483         err = init_common_resources();
1484         if (err)
1485                 return err;
1486
1487         in_hyp_mode = is_kernel_in_hyp_mode();
1488
1489         if (!in_hyp_mode) {
1490                 err = init_hyp_mode();
1491                 if (err)
1492                         goto out_err;
1493         }
1494
1495         err = init_subsystems();
1496         if (err)
1497                 goto out_hyp;
1498
1499         if (in_hyp_mode)
1500                 kvm_info("VHE mode initialized successfully\n");
1501         else
1502                 kvm_info("Hyp mode initialized successfully\n");
1503
1504         return 0;
1505
1506 out_hyp:
1507         if (!in_hyp_mode)
1508                 teardown_hyp_mode();
1509 out_err:
1510         return err;
1511 }
1512
1513 /* NOP: Compiling as a module not supported */
1514 void kvm_arch_exit(void)
1515 {
1516         kvm_perf_teardown();
1517 }
1518
1519 static int arm_init(void)
1520 {
1521         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1522         return rc;
1523 }
1524
1525 module_init(arm_init);