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 <linux/kvm_irqfd.h>
  31 #include <linux/irqbypass.h>
  32 #include <trace/events/kvm.h>
  33 #include <kvm/arm_pmu.h>
  34 #include <kvm/arm_psci.h>
  35
  36 #define CREATE_TRACE_POINTS
  37 #include "trace.h"
  38
  39 #include <linux/uaccess.h>
  40 #include <asm/ptrace.h>
  41 #include <asm/mman.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/cacheflush.h>
  44 #include <asm/virt.h>
  45 #include <asm/kvm_arm.h>
  46 #include <asm/kvm_asm.h>
  47 #include <asm/kvm_mmu.h>
  48 #include <asm/kvm_emulate.h>
  49 #include <asm/kvm_coproc.h>
  50 #include <asm/sections.h>
  51
  52 #ifdef REQUIRES_VIRT
  53 __asm__(".arch_extension        virt");
  54 #endif
  55
  56 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  57 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  58
  59 /* Per-CPU variable containing the currently running vcpu. */
  60 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  61
  62 /* The VMID used in the VTTBR */
  63 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  64 static u32 kvm_next_vmid;
  65 static unsigned int kvm_vmid_bits __read_mostly;
  66 static DEFINE_SPINLOCK(kvm_vmid_lock);
  67
  68 static bool vgic_present;
  69
  70 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  71
  72 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  73 {
  74         BUG_ON(preemptible());
  75         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  76 }
  77
  78 /**
  79  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  80  * Must be called from non-preemptible context
  81  */
  82 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  83 {
  84         BUG_ON(preemptible());
  85         return __this_cpu_read(kvm_arm_running_vcpu);
  86 }
  87
  88 /**
  89  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  90  */
  91 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  92 {
  93         return &kvm_arm_running_vcpu;
  94 }
  95
  96 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  97 {
  98         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  99 }
 100
 101 int kvm_arch_hardware_setup(void)
 102 {
 103         return 0;
 104 }
 105
 106 void kvm_arch_check_processor_compat(void *rtn)
 107 {
 108         *(int *)rtn = 0;
 109 }
 110
 111
 112 /**
 113  * kvm_arch_init_vm - initializes a VM data structure
 114  * @kvm:        pointer to the KVM struct
 115  */
 116 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 117 {
 118         int ret, cpu;
 119
 120         if (type)
 121                 return -EINVAL;
 122
 123         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 124         if (!kvm->arch.last_vcpu_ran)
 125                 return -ENOMEM;
 126
 127         for_each_possible_cpu(cpu)
 128                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 129
 130         ret = kvm_alloc_stage2_pgd(kvm);
 131         if (ret)
 132                 goto out_fail_alloc;
 133
 134         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 135         if (ret)
 136                 goto out_free_stage2_pgd;
 137
 138         kvm_vgic_early_init(kvm);
 139
 140         /* Mark the initial VMID generation invalid */
 141         kvm->arch.vmid_gen = 0;
 142
 143         /* The maximum number of VCPUs is limited by the host's GIC model */
 144         kvm->arch.max_vcpus = vgic_present ?
 145                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 146
 147         return ret;
 148 out_free_stage2_pgd:
 149         kvm_free_stage2_pgd(kvm);
 150 out_fail_alloc:
 151         free_percpu(kvm->arch.last_vcpu_ran);
 152         kvm->arch.last_vcpu_ran = NULL;
 153         return ret;
 154 }
 155
 156 bool kvm_arch_has_vcpu_debugfs(void)
 157 {
 158         return false;
 159 }
 160
 161 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 162 {
 163         return 0;
 164 }
 165
 166 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 167 {
 168         return VM_FAULT_SIGBUS;
 169 }
 170
 171
 172 /**
 173  * kvm_arch_destroy_vm - destroy the VM data structure
 174  * @kvm:        pointer to the KVM struct
 175  */
 176 void kvm_arch_destroy_vm(struct kvm *kvm)
 177 {
 178         int i;
 179
 180         kvm_vgic_destroy(kvm);
 181
 182         free_percpu(kvm->arch.last_vcpu_ran);
 183         kvm->arch.last_vcpu_ran = NULL;
 184
 185         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 186                 if (kvm->vcpus[i]) {
 187                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 188                         kvm->vcpus[i] = NULL;
 189                 }
 190         }
 191         atomic_set(&kvm->online_vcpus, 0);
 192 }
 193
 194 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 195 {
 196         int r;
 197         switch (ext) {
 198         case KVM_CAP_IRQCHIP:
 199                 r = vgic_present;
 200                 break;
 201         case KVM_CAP_IOEVENTFD:
 202         case KVM_CAP_DEVICE_CTRL:
 203         case KVM_CAP_USER_MEMORY:
 204         case KVM_CAP_SYNC_MMU:
 205         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 206         case KVM_CAP_ONE_REG:
 207         case KVM_CAP_ARM_PSCI:
 208         case KVM_CAP_ARM_PSCI_0_2:
 209         case KVM_CAP_READONLY_MEM:
 210         case KVM_CAP_MP_STATE:
 211         case KVM_CAP_IMMEDIATE_EXIT:
 212                 r = 1;
 213                 break;
 214         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 215                 r = 1;
 216                 break;
 217         case KVM_CAP_NR_VCPUS:
 218                 r = num_online_cpus();
 219                 break;
 220         case KVM_CAP_MAX_VCPUS:
 221                 r = KVM_MAX_VCPUS;
 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_pmu_vcpu_destroy(vcpu);
 301         kvm_vcpu_uninit(vcpu);
 302         kmem_cache_free(kvm_vcpu_cache, vcpu);
 303 }
 304
 305 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 306 {
 307         kvm_arch_vcpu_free(vcpu);
 308 }
 309
 310 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 311 {
 312         return kvm_timer_is_pending(vcpu);
 313 }
 314
 315 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 316 {
 317         kvm_timer_schedule(vcpu);
 318         kvm_vgic_v4_enable_doorbell(vcpu);
 319 }
 320
 321 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 322 {
 323         kvm_timer_unschedule(vcpu);
 324         kvm_vgic_v4_disable_doorbell(vcpu);
 325 }
 326
 327 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 328 {
 329         /* Force users to call KVM_ARM_VCPU_INIT */
 330         vcpu->arch.target = -1;
 331         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 332
 333         /* Set up the timer */
 334         kvm_timer_vcpu_init(vcpu);
 335
 336         kvm_arm_reset_debug_ptr(vcpu);
 337
 338         return kvm_vgic_vcpu_init(vcpu);
 339 }
 340
 341 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 342 {
 343         int *last_ran;
 344
 345         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 346
 347         /*
 348          * We might get preempted before the vCPU actually runs, but
 349          * over-invalidation doesn't affect correctness.
 350          */
 351         if (*last_ran != vcpu->vcpu_id) {
 352                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 353                 *last_ran = vcpu->vcpu_id;
 354         }
 355
 356         vcpu->cpu = cpu;
 357         vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
 358
 359         kvm_arm_set_running_vcpu(vcpu);
 360         kvm_vgic_load(vcpu);
 361         kvm_timer_vcpu_load(vcpu);
 362 }
 363
 364 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 365 {
 366         kvm_timer_vcpu_put(vcpu);
 367         kvm_vgic_put(vcpu);
 368
 369         vcpu->cpu = -1;
 370
 371         kvm_arm_set_running_vcpu(NULL);
 372 }
 373
 374 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 375 {
 376         vcpu->arch.power_off = true;
 377         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 378         kvm_vcpu_kick(vcpu);
 379 }
 380
 381 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 382                                     struct kvm_mp_state *mp_state)
 383 {
 384         if (vcpu->arch.power_off)
 385                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 386         else
 387                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 388
 389         return 0;
 390 }
 391
 392 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 393                                     struct kvm_mp_state *mp_state)
 394 {
 395         switch (mp_state->mp_state) {
 396         case KVM_MP_STATE_RUNNABLE:
 397                 vcpu->arch.power_off = false;
 398                 break;
 399         case KVM_MP_STATE_STOPPED:
 400                 vcpu_power_off(vcpu);
 401                 break;
 402         default:
 403                 return -EINVAL;
 404         }
 405
 406         return 0;
 407 }
 408
 409 /**
 410  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 411  * @v:          The VCPU pointer
 412  *
 413  * If the guest CPU is not waiting for interrupts or an interrupt line is
 414  * asserted, the CPU is by definition runnable.
 415  */
 416 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 417 {
 418         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
 419                 && !v->arch.power_off && !v->arch.pause);
 420 }
 421
 422 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 423 {
 424         return vcpu_mode_priv(vcpu);
 425 }
 426
 427 /* Just ensure a guest exit from a particular CPU */
 428 static void exit_vm_noop(void *info)
 429 {
 430 }
 431
 432 void force_vm_exit(const cpumask_t *mask)
 433 {
 434         preempt_disable();
 435         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 436         preempt_enable();
 437 }
 438
 439 /**
 440  * need_new_vmid_gen - check that the VMID is still valid
 441  * @kvm: The VM's VMID to check
 442  *
 443  * return true if there is a new generation of VMIDs being used
 444  *
 445  * The hardware supports only 256 values with the value zero reserved for the
 446  * host, so we check if an assigned value belongs to a previous generation,
 447  * which which requires us to assign a new value. If we're the first to use a
 448  * VMID for the new generation, we must flush necessary caches and TLBs on all
 449  * CPUs.
 450  */
 451 static bool need_new_vmid_gen(struct kvm *kvm)
 452 {
 453         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 454 }
 455
 456 /**
 457  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 458  * @kvm The guest that we are about to run
 459  *
 460  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 461  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 462  * caches and TLBs.
 463  */
 464 static void update_vttbr(struct kvm *kvm)
 465 {
 466         phys_addr_t pgd_phys;
 467         u64 vmid;
 468
 469         if (!need_new_vmid_gen(kvm))
 470                 return;
 471
 472         spin_lock(&kvm_vmid_lock);
 473
 474         /*
 475          * We need to re-check the vmid_gen here to ensure that if another vcpu
 476          * already allocated a valid vmid for this vm, then this vcpu should
 477          * use the same vmid.
 478          */
 479         if (!need_new_vmid_gen(kvm)) {
 480                 spin_unlock(&kvm_vmid_lock);
 481                 return;
 482         }
 483
 484         /* First user of a new VMID generation? */
 485         if (unlikely(kvm_next_vmid == 0)) {
 486                 atomic64_inc(&kvm_vmid_gen);
 487                 kvm_next_vmid = 1;
 488
 489                 /*
 490                  * On SMP we know no other CPUs can use this CPU's or each
 491                  * other's VMID after force_vm_exit returns since the
 492                  * kvm_vmid_lock blocks them from reentry to the guest.
 493                  */
 494                 force_vm_exit(cpu_all_mask);
 495                 /*
 496                  * Now broadcast TLB + ICACHE invalidation over the inner
 497                  * shareable domain to make sure all data structures are
 498                  * clean.
 499                  */
 500                 kvm_call_hyp(__kvm_flush_vm_context);
 501         }
 502
 503         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 504         kvm->arch.vmid = kvm_next_vmid;
 505         kvm_next_vmid++;
 506         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 507
 508         /* update vttbr to be used with the new vmid */
 509         pgd_phys = virt_to_phys(kvm->arch.pgd);
 510         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 511         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 512         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid;
 513
 514         spin_unlock(&kvm_vmid_lock);
 515 }
 516
 517 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 518 {
 519         struct kvm *kvm = vcpu->kvm;
 520         int ret = 0;
 521
 522         if (likely(vcpu->arch.has_run_once))
 523                 return 0;
 524
 525         vcpu->arch.has_run_once = true;
 526
 527         /*
 528          * Map the VGIC hardware resources before running a vcpu the first
 529          * time on this VM.
 530          */
 531         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
 532                 ret = kvm_vgic_map_resources(kvm);
 533                 if (ret)
 534                         return ret;
 535         }
 536
 537         ret = kvm_timer_enable(vcpu);
 538         if (ret)
 539                 return ret;
 540
 541         ret = kvm_arm_pmu_v3_enable(vcpu);
 542
 543         return ret;
 544 }
 545
 546 bool kvm_arch_intc_initialized(struct kvm *kvm)
 547 {
 548         return vgic_initialized(kvm);
 549 }
 550
 551 void kvm_arm_halt_guest(struct kvm *kvm)
 552 {
 553         int i;
 554         struct kvm_vcpu *vcpu;
 555
 556         kvm_for_each_vcpu(i, vcpu, kvm)
 557                 vcpu->arch.pause = true;
 558         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 559 }
 560
 561 void kvm_arm_resume_guest(struct kvm *kvm)
 562 {
 563         int i;
 564         struct kvm_vcpu *vcpu;
 565
 566         kvm_for_each_vcpu(i, vcpu, kvm) {
 567                 vcpu->arch.pause = false;
 568                 swake_up(kvm_arch_vcpu_wq(vcpu));
 569         }
 570 }
 571
 572 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 573 {
 574         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 575
 576         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 577                                        (!vcpu->arch.pause)));
 578
 579         if (vcpu->arch.power_off || vcpu->arch.pause) {
 580                 /* Awaken to handle a signal, request we sleep again later. */
 581                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 582         }
 583 }
 584
 585 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 586 {
 587         return vcpu->arch.target >= 0;
 588 }
 589
 590 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 591 {
 592         if (kvm_request_pending(vcpu)) {
 593                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 594                         vcpu_req_sleep(vcpu);
 595
 596                 /*
 597                  * Clear IRQ_PENDING requests that were made to guarantee
 598                  * that a VCPU sees new virtual interrupts.
 599                  */
 600                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 601         }
 602 }
 603
 604 /**
 605  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 606  * @vcpu:       The VCPU pointer
 607  * @run:        The kvm_run structure pointer used for userspace state exchange
 608  *
 609  * This function is called through the VCPU_RUN ioctl called from user space. It
 610  * will execute VM code in a loop until the time slice for the process is used
 611  * or some emulation is needed from user space in which case the function will
 612  * return with return value 0 and with the kvm_run structure filled in with the
 613  * required data for the requested emulation.
 614  */
 615 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 616 {
 617         int ret;
 618
 619         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 620                 return -ENOEXEC;
 621
 622         ret = kvm_vcpu_first_run_init(vcpu);
 623         if (ret)
 624                 return ret;
 625
 626         if (run->exit_reason == KVM_EXIT_MMIO) {
 627                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 628                 if (ret)
 629                         return ret;
 630                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
 631                         return 0;
 632
 633         }
 634
 635         if (run->immediate_exit)
 636                 return -EINTR;
 637
 638         kvm_sigset_activate(vcpu);
 639
 640         ret = 1;
 641         run->exit_reason = KVM_EXIT_UNKNOWN;
 642         while (ret > 0) {
 643                 /*
 644                  * Check conditions before entering the guest
 645                  */
 646                 cond_resched();
 647
 648                 update_vttbr(vcpu->kvm);
 649
 650                 check_vcpu_requests(vcpu);
 651
 652                 /*
 653                  * Preparing the interrupts to be injected also
 654                  * involves poking the GIC, which must be done in a
 655                  * non-preemptible context.
 656                  */
 657                 preempt_disable();
 658
 659                 /* Flush FP/SIMD state that can't survive guest entry/exit */
 660                 kvm_fpsimd_flush_cpu_state();
 661
 662                 kvm_pmu_flush_hwstate(vcpu);
 663
 664                 local_irq_disable();
 665
 666                 kvm_vgic_flush_hwstate(vcpu);
 667
 668                 /*
 669                  * If we have a singal pending, or need to notify a userspace
 670                  * irqchip about timer or PMU level changes, then we exit (and
 671                  * update the timer level state in kvm_timer_update_run
 672                  * below).
 673                  */
 674                 if (signal_pending(current) ||
 675                     kvm_timer_should_notify_user(vcpu) ||
 676                     kvm_pmu_should_notify_user(vcpu)) {
 677                         ret = -EINTR;
 678                         run->exit_reason = KVM_EXIT_INTR;
 679                 }
 680
 681                 /*
 682                  * Ensure we set mode to IN_GUEST_MODE after we disable
 683                  * interrupts and before the final VCPU requests check.
 684                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 685                  * Documentation/virtual/kvm/vcpu-requests.rst
 686                  */
 687                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 688
 689                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 690                     kvm_request_pending(vcpu)) {
 691                         vcpu->mode = OUTSIDE_GUEST_MODE;
 692                         kvm_pmu_sync_hwstate(vcpu);
 693                         kvm_timer_sync_hwstate(vcpu);
 694                         kvm_vgic_sync_hwstate(vcpu);
 695                         local_irq_enable();
 696                         preempt_enable();
 697                         continue;
 698                 }
 699
 700                 kvm_arm_setup_debug(vcpu);
 701
 702                 /**************************************************************
 703                  * Enter the guest
 704                  */
 705                 trace_kvm_entry(*vcpu_pc(vcpu));
 706                 guest_enter_irqoff();
 707                 if (has_vhe())
 708                         kvm_arm_vhe_guest_enter();
 709
 710                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 711
 712                 if (has_vhe())
 713                         kvm_arm_vhe_guest_exit();
 714                 vcpu->mode = OUTSIDE_GUEST_MODE;
 715                 vcpu->stat.exits++;
 716                 /*
 717                  * Back from guest
 718                  *************************************************************/
 719
 720                 kvm_arm_clear_debug(vcpu);
 721
 722                 /*
 723                  * We must sync the PMU state before the vgic state so
 724                  * that the vgic can properly sample the updated state of the
 725                  * interrupt line.
 726                  */
 727                 kvm_pmu_sync_hwstate(vcpu);
 728
 729                 /*
 730                  * Sync the vgic state before syncing the timer state because
 731                  * the timer code needs to know if the virtual timer
 732                  * interrupts are active.
 733                  */
 734                 kvm_vgic_sync_hwstate(vcpu);
 735
 736                 /*
 737                  * Sync the timer hardware state before enabling interrupts as
 738                  * we don't want vtimer interrupts to race with syncing the
 739                  * timer virtual interrupt state.
 740                  */
 741                 kvm_timer_sync_hwstate(vcpu);
 742
 743                 /*
 744                  * We may have taken a host interrupt in HYP mode (ie
 745                  * while executing the guest). This interrupt is still
 746                  * pending, as we haven't serviced it yet!
 747                  *
 748                  * We're now back in SVC mode, with interrupts
 749                  * disabled.  Enabling the interrupts now will have
 750                  * the effect of taking the interrupt again, in SVC
 751                  * mode this time.
 752                  */
 753                 local_irq_enable();
 754
 755                 /*
 756                  * We do local_irq_enable() before calling guest_exit() so
 757                  * that if a timer interrupt hits while running the guest we
 758                  * account that tick as being spent in the guest.  We enable
 759                  * preemption after calling guest_exit() so that if we get
 760                  * preempted we make sure ticks after that is not counted as
 761                  * guest time.
 762                  */
 763                 guest_exit();
 764                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 765
 766                 /* Exit types that need handling before we can be preempted */
 767                 handle_exit_early(vcpu, run, ret);
 768
 769                 preempt_enable();
 770
 771                 ret = handle_exit(vcpu, run, ret);
 772         }
 773
 774         /* Tell userspace about in-kernel device output levels */
 775         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 776                 kvm_timer_update_run(vcpu);
 777                 kvm_pmu_update_run(vcpu);
 778         }
 779
 780         kvm_sigset_deactivate(vcpu);
 781
 782         return ret;
 783 }
 784
 785 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 786 {
 787         int bit_index;
 788         bool set;
 789         unsigned long *ptr;
 790
 791         if (number == KVM_ARM_IRQ_CPU_IRQ)
 792                 bit_index = __ffs(HCR_VI);
 793         else /* KVM_ARM_IRQ_CPU_FIQ */
 794                 bit_index = __ffs(HCR_VF);
 795
 796         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 797         if (level)
 798                 set = test_and_set_bit(bit_index, ptr);
 799         else
 800                 set = test_and_clear_bit(bit_index, ptr);
 801
 802         /*
 803          * If we didn't change anything, no need to wake up or kick other CPUs
 804          */
 805         if (set == level)
 806                 return 0;
 807
 808         /*
 809          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 810          * trigger a world-switch round on the running physical CPU to set the
 811          * virtual IRQ/FIQ fields in the HCR appropriately.
 812          */
 813         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 814         kvm_vcpu_kick(vcpu);
 815
 816         return 0;
 817 }
 818
 819 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 820                           bool line_status)
 821 {
 822         u32 irq = irq_level->irq;
 823         unsigned int irq_type, vcpu_idx, irq_num;
 824         int nrcpus = atomic_read(&kvm->online_vcpus);
 825         struct kvm_vcpu *vcpu = NULL;
 826         bool level = irq_level->level;
 827
 828         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 829         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 830         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 831
 832         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 833
 834         switch (irq_type) {
 835         case KVM_ARM_IRQ_TYPE_CPU:
 836                 if (irqchip_in_kernel(kvm))
 837                         return -ENXIO;
 838
 839                 if (vcpu_idx >= nrcpus)
 840                         return -EINVAL;
 841
 842                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 843                 if (!vcpu)
 844                         return -EINVAL;
 845
 846                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 847                         return -EINVAL;
 848
 849                 return vcpu_interrupt_line(vcpu, irq_num, level);
 850         case KVM_ARM_IRQ_TYPE_PPI:
 851                 if (!irqchip_in_kernel(kvm))
 852                         return -ENXIO;
 853
 854                 if (vcpu_idx >= nrcpus)
 855                         return -EINVAL;
 856
 857                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 858                 if (!vcpu)
 859                         return -EINVAL;
 860
 861                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 862                         return -EINVAL;
 863
 864                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 865         case KVM_ARM_IRQ_TYPE_SPI:
 866                 if (!irqchip_in_kernel(kvm))
 867                         return -ENXIO;
 868
 869                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 870                         return -EINVAL;
 871
 872                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 873         }
 874
 875         return -EINVAL;
 876 }
 877
 878 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 879                                const struct kvm_vcpu_init *init)
 880 {
 881         unsigned int i;
 882         int phys_target = kvm_target_cpu();
 883
 884         if (init->target != phys_target)
 885                 return -EINVAL;
 886
 887         /*
 888          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 889          * use the same target.
 890          */
 891         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 892                 return -EINVAL;
 893
 894         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 895         for (i = 0; i < sizeof(init->features) * 8; i++) {
 896                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 897
 898                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 899                         return -ENOENT;
 900
 901                 /*
 902                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 903                  * use the same feature set.
 904                  */
 905                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 906                     test_bit(i, vcpu->arch.features) != set)
 907                         return -EINVAL;
 908
 909                 if (set)
 910                         set_bit(i, vcpu->arch.features);
 911         }
 912
 913         vcpu->arch.target = phys_target;
 914
 915         /* Now we know what it is, we can reset it. */
 916         return kvm_reset_vcpu(vcpu);
 917 }
 918
 919
 920 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 921                                          struct kvm_vcpu_init *init)
 922 {
 923         int ret;
 924
 925         ret = kvm_vcpu_set_target(vcpu, init);
 926         if (ret)
 927                 return ret;
 928
 929         /*
 930          * Ensure a rebooted VM will fault in RAM pages and detect if the
 931          * guest MMU is turned off and flush the caches as needed.
 932          */
 933         if (vcpu->arch.has_run_once)
 934                 stage2_unmap_vm(vcpu->kvm);
 935
 936         vcpu_reset_hcr(vcpu);
 937
 938         /*
 939          * Handle the "start in power-off" case.
 940          */
 941         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 942                 vcpu_power_off(vcpu);
 943         else
 944                 vcpu->arch.power_off = false;
 945
 946         return 0;
 947 }
 948
 949 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 950                                  struct kvm_device_attr *attr)
 951 {
 952         int ret = -ENXIO;
 953
 954         switch (attr->group) {
 955         default:
 956                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 957                 break;
 958         }
 959
 960         return ret;
 961 }
 962
 963 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 964                                  struct kvm_device_attr *attr)
 965 {
 966         int ret = -ENXIO;
 967
 968         switch (attr->group) {
 969         default:
 970                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
 971                 break;
 972         }
 973
 974         return ret;
 975 }
 976
 977 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
 978                                  struct kvm_device_attr *attr)
 979 {
 980         int ret = -ENXIO;
 981
 982         switch (attr->group) {
 983         default:
 984                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
 985                 break;
 986         }
 987
 988         return ret;
 989 }
 990
 991 long kvm_arch_vcpu_ioctl(struct file *filp,
 992                          unsigned int ioctl, unsigned long arg)
 993 {
 994         struct kvm_vcpu *vcpu = filp->private_data;
 995         void __user *argp = (void __user *)arg;
 996         struct kvm_device_attr attr;
 997
 998         switch (ioctl) {
 999         case KVM_ARM_VCPU_INIT: {
1000                 struct kvm_vcpu_init init;
1001
1002                 if (copy_from_user(&init, argp, sizeof(init)))
1003                         return -EFAULT;
1004
1005                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1006         }
1007         case KVM_SET_ONE_REG:
1008         case KVM_GET_ONE_REG: {
1009                 struct kvm_one_reg reg;
1010
1011                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1012                         return -ENOEXEC;
1013
1014                 if (copy_from_user(&reg, argp, sizeof(reg)))
1015                         return -EFAULT;
1016                 if (ioctl == KVM_SET_ONE_REG)
1017                         return kvm_arm_set_reg(vcpu, &reg);
1018                 else
1019                         return kvm_arm_get_reg(vcpu, &reg);
1020         }
1021         case KVM_GET_REG_LIST: {
1022                 struct kvm_reg_list __user *user_list = argp;
1023                 struct kvm_reg_list reg_list;
1024                 unsigned n;
1025
1026                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1027                         return -ENOEXEC;
1028
1029                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1030                         return -EFAULT;
1031                 n = reg_list.n;
1032                 reg_list.n = kvm_arm_num_regs(vcpu);
1033                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1034                         return -EFAULT;
1035                 if (n < reg_list.n)
1036                         return -E2BIG;
1037                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1038         }
1039         case KVM_SET_DEVICE_ATTR: {
1040                 if (copy_from_user(&attr, argp, sizeof(attr)))
1041                         return -EFAULT;
1042                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1043         }
1044         case KVM_GET_DEVICE_ATTR: {
1045                 if (copy_from_user(&attr, argp, sizeof(attr)))
1046                         return -EFAULT;
1047                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1048         }
1049         case KVM_HAS_DEVICE_ATTR: {
1050                 if (copy_from_user(&attr, argp, sizeof(attr)))
1051                         return -EFAULT;
1052                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1053         }
1054         default:
1055                 return -EINVAL;
1056         }
1057 }
1058
1059 /**
1060  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1061  * @kvm: kvm instance
1062  * @log: slot id and address to which we copy the log
1063  *
1064  * Steps 1-4 below provide general overview of dirty page logging. See
1065  * kvm_get_dirty_log_protect() function description for additional details.
1066  *
1067  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1068  * always flush the TLB (step 4) even if previous step failed  and the dirty
1069  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1070  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1071  * writes will be marked dirty for next log read.
1072  *
1073  *   1. Take a snapshot of the bit and clear it if needed.
1074  *   2. Write protect the corresponding page.
1075  *   3. Copy the snapshot to the userspace.
1076  *   4. Flush TLB's if needed.
1077  */
1078 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1079 {
1080         bool is_dirty = false;
1081         int r;
1082
1083         mutex_lock(&kvm->slots_lock);
1084
1085         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1086
1087         if (is_dirty)
1088                 kvm_flush_remote_tlbs(kvm);
1089
1090         mutex_unlock(&kvm->slots_lock);
1091         return r;
1092 }
1093
1094 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1095                                         struct kvm_arm_device_addr *dev_addr)
1096 {
1097         unsigned long dev_id, type;
1098
1099         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1100                 KVM_ARM_DEVICE_ID_SHIFT;
1101         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1102                 KVM_ARM_DEVICE_TYPE_SHIFT;
1103
1104         switch (dev_id) {
1105         case KVM_ARM_DEVICE_VGIC_V2:
1106                 if (!vgic_present)
1107                         return -ENXIO;
1108                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1109         default:
1110                 return -ENODEV;
1111         }
1112 }
1113
1114 long kvm_arch_vm_ioctl(struct file *filp,
1115                        unsigned int ioctl, unsigned long arg)
1116 {
1117         struct kvm *kvm = filp->private_data;
1118         void __user *argp = (void __user *)arg;
1119
1120         switch (ioctl) {
1121         case KVM_CREATE_IRQCHIP: {
1122                 int ret;
1123                 if (!vgic_present)
1124                         return -ENXIO;
1125                 mutex_lock(&kvm->lock);
1126                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1127                 mutex_unlock(&kvm->lock);
1128                 return ret;
1129         }
1130         case KVM_ARM_SET_DEVICE_ADDR: {
1131                 struct kvm_arm_device_addr dev_addr;
1132
1133                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1134                         return -EFAULT;
1135                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1136         }
1137         case KVM_ARM_PREFERRED_TARGET: {
1138                 int err;
1139                 struct kvm_vcpu_init init;
1140
1141                 err = kvm_vcpu_preferred_target(&init);
1142                 if (err)
1143                         return err;
1144
1145                 if (copy_to_user(argp, &init, sizeof(init)))
1146                         return -EFAULT;
1147
1148                 return 0;
1149         }
1150         default:
1151                 return -EINVAL;
1152         }
1153 }
1154
1155 static void cpu_init_hyp_mode(void *dummy)
1156 {
1157         phys_addr_t pgd_ptr;
1158         unsigned long hyp_stack_ptr;
1159         unsigned long stack_page;
1160         unsigned long vector_ptr;
1161
1162         /* Switch from the HYP stub to our own HYP init vector */
1163         __hyp_set_vectors(kvm_get_idmap_vector());
1164
1165         pgd_ptr = kvm_mmu_get_httbr();
1166         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1167         hyp_stack_ptr = stack_page + PAGE_SIZE;
1168         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1169
1170         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1171         __cpu_init_stage2();
1172
1173         kvm_arm_init_debug();
1174 }
1175
1176 static void cpu_hyp_reset(void)
1177 {
1178         if (!is_kernel_in_hyp_mode())
1179                 __hyp_reset_vectors();
1180 }
1181
1182 static void cpu_hyp_reinit(void)
1183 {
1184         cpu_hyp_reset();
1185
1186         if (is_kernel_in_hyp_mode()) {
1187                 /*
1188                  * __cpu_init_stage2() is safe to call even if the PM
1189                  * event was cancelled before the CPU was reset.
1190                  */
1191                 __cpu_init_stage2();
1192                 kvm_timer_init_vhe();
1193         } else {
1194                 cpu_init_hyp_mode(NULL);
1195         }
1196
1197         if (vgic_present)
1198                 kvm_vgic_init_cpu_hardware();
1199 }
1200
1201 static void _kvm_arch_hardware_enable(void *discard)
1202 {
1203         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1204                 cpu_hyp_reinit();
1205                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1206         }
1207 }
1208
1209 int kvm_arch_hardware_enable(void)
1210 {
1211         _kvm_arch_hardware_enable(NULL);
1212         return 0;
1213 }
1214
1215 static void _kvm_arch_hardware_disable(void *discard)
1216 {
1217         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1218                 cpu_hyp_reset();
1219                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1220         }
1221 }
1222
1223 void kvm_arch_hardware_disable(void)
1224 {
1225         _kvm_arch_hardware_disable(NULL);
1226 }
1227
1228 #ifdef CONFIG_CPU_PM
1229 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1230                                     unsigned long cmd,
1231                                     void *v)
1232 {
1233         /*
1234          * kvm_arm_hardware_enabled is left with its old value over
1235          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1236          * re-enable hyp.
1237          */
1238         switch (cmd) {
1239         case CPU_PM_ENTER:
1240                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1241                         /*
1242                          * don't update kvm_arm_hardware_enabled here
1243                          * so that the hardware will be re-enabled
1244                          * when we resume. See below.
1245                          */
1246                         cpu_hyp_reset();
1247
1248                 return NOTIFY_OK;
1249         case CPU_PM_EXIT:
1250                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1251                         /* The hardware was enabled before suspend. */
1252                         cpu_hyp_reinit();
1253
1254                 return NOTIFY_OK;
1255
1256         default:
1257                 return NOTIFY_DONE;
1258         }
1259 }
1260
1261 static struct notifier_block hyp_init_cpu_pm_nb = {
1262         .notifier_call = hyp_init_cpu_pm_notifier,
1263 };
1264
1265 static void __init hyp_cpu_pm_init(void)
1266 {
1267         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1268 }
1269 static void __init hyp_cpu_pm_exit(void)
1270 {
1271         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1272 }
1273 #else
1274 static inline void hyp_cpu_pm_init(void)
1275 {
1276 }
1277 static inline void hyp_cpu_pm_exit(void)
1278 {
1279 }
1280 #endif
1281
1282 static int init_common_resources(void)
1283 {
1284         /* set size of VMID supported by CPU */
1285         kvm_vmid_bits = kvm_get_vmid_bits();
1286         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1287
1288         return 0;
1289 }
1290
1291 static int init_subsystems(void)
1292 {
1293         int err = 0;
1294
1295         /*
1296          * Enable hardware so that subsystem initialisation can access EL2.
1297          */
1298         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1299
1300         /*
1301          * Register CPU lower-power notifier
1302          */
1303         hyp_cpu_pm_init();
1304
1305         /*
1306          * Init HYP view of VGIC
1307          */
1308         err = kvm_vgic_hyp_init();
1309         switch (err) {
1310         case 0:
1311                 vgic_present = true;
1312                 break;
1313         case -ENODEV:
1314         case -ENXIO:
1315                 vgic_present = false;
1316                 err = 0;
1317                 break;
1318         default:
1319                 goto out;
1320         }
1321
1322         /*
1323          * Init HYP architected timer support
1324          */
1325         err = kvm_timer_hyp_init(vgic_present);
1326         if (err)
1327                 goto out;
1328
1329         kvm_perf_init();
1330         kvm_coproc_table_init();
1331
1332 out:
1333         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1334
1335         return err;
1336 }
1337
1338 static void teardown_hyp_mode(void)
1339 {
1340         int cpu;
1341
1342         free_hyp_pgds();
1343         for_each_possible_cpu(cpu)
1344                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1345         hyp_cpu_pm_exit();
1346 }
1347
1348 /**
1349  * Inits Hyp-mode on all online CPUs
1350  */
1351 static int init_hyp_mode(void)
1352 {
1353         int cpu;
1354         int err = 0;
1355
1356         /*
1357          * Allocate Hyp PGD and setup Hyp identity mapping
1358          */
1359         err = kvm_mmu_init();
1360         if (err)
1361                 goto out_err;
1362
1363         /*
1364          * Allocate stack pages for Hypervisor-mode
1365          */
1366         for_each_possible_cpu(cpu) {
1367                 unsigned long stack_page;
1368
1369                 stack_page = __get_free_page(GFP_KERNEL);
1370                 if (!stack_page) {
1371                         err = -ENOMEM;
1372                         goto out_err;
1373                 }
1374
1375                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1376         }
1377
1378         /*
1379          * Map the Hyp-code called directly from the host
1380          */
1381         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1382                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1383         if (err) {
1384                 kvm_err("Cannot map world-switch code\n");
1385                 goto out_err;
1386         }
1387
1388         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1389                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1390         if (err) {
1391                 kvm_err("Cannot map rodata section\n");
1392                 goto out_err;
1393         }
1394
1395         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1396                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1397         if (err) {
1398                 kvm_err("Cannot map bss section\n");
1399                 goto out_err;
1400         }
1401
1402         err = kvm_map_vectors();
1403         if (err) {
1404                 kvm_err("Cannot map vectors\n");
1405                 goto out_err;
1406         }
1407
1408         /*
1409          * Map the Hyp stack pages
1410          */
1411         for_each_possible_cpu(cpu) {
1412                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1413                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1414                                           PAGE_HYP);
1415
1416                 if (err) {
1417                         kvm_err("Cannot map hyp stack\n");
1418                         goto out_err;
1419                 }
1420         }
1421
1422         for_each_possible_cpu(cpu) {
1423                 kvm_cpu_context_t *cpu_ctxt;
1424
1425                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1426                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1427
1428                 if (err) {
1429                         kvm_err("Cannot map host CPU state: %d\n", err);
1430                         goto out_err;
1431                 }
1432         }
1433
1434         return 0;
1435
1436 out_err:
1437         teardown_hyp_mode();
1438         kvm_err("error initializing Hyp mode: %d\n", err);
1439         return err;
1440 }
1441
1442 static void check_kvm_target_cpu(void *ret)
1443 {
1444         *(int *)ret = kvm_target_cpu();
1445 }
1446
1447 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1448 {
1449         struct kvm_vcpu *vcpu;
1450         int i;
1451
1452         mpidr &= MPIDR_HWID_BITMASK;
1453         kvm_for_each_vcpu(i, vcpu, kvm) {
1454                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1455                         return vcpu;
1456         }
1457         return NULL;
1458 }
1459
1460 bool kvm_arch_has_irq_bypass(void)
1461 {
1462         return true;
1463 }
1464
1465 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1466                                       struct irq_bypass_producer *prod)
1467 {
1468         struct kvm_kernel_irqfd *irqfd =
1469                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1470
1471         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1472                                           &irqfd->irq_entry);
1473 }
1474 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1475                                       struct irq_bypass_producer *prod)
1476 {
1477         struct kvm_kernel_irqfd *irqfd =
1478                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1479
1480         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1481                                      &irqfd->irq_entry);
1482 }
1483
1484 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1485 {
1486         struct kvm_kernel_irqfd *irqfd =
1487                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1488
1489         kvm_arm_halt_guest(irqfd->kvm);
1490 }
1491
1492 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1493 {
1494         struct kvm_kernel_irqfd *irqfd =
1495                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1496
1497         kvm_arm_resume_guest(irqfd->kvm);
1498 }
1499
1500 /**
1501  * Initialize Hyp-mode and memory mappings on all CPUs.
1502  */
1503 int kvm_arch_init(void *opaque)
1504 {
1505         int err;
1506         int ret, cpu;
1507         bool in_hyp_mode;
1508
1509         if (!is_hyp_mode_available()) {
1510                 kvm_info("HYP mode not available\n");
1511                 return -ENODEV;
1512         }
1513
1514         for_each_online_cpu(cpu) {
1515                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1516                 if (ret < 0) {
1517                         kvm_err("Error, CPU %d not supported!\n", cpu);
1518                         return -ENODEV;
1519                 }
1520         }
1521
1522         err = init_common_resources();
1523         if (err)
1524                 return err;
1525
1526         in_hyp_mode = is_kernel_in_hyp_mode();
1527
1528         if (!in_hyp_mode) {
1529                 err = init_hyp_mode();
1530                 if (err)
1531                         goto out_err;
1532         }
1533
1534         err = init_subsystems();
1535         if (err)
1536                 goto out_hyp;
1537
1538         if (in_hyp_mode)
1539                 kvm_info("VHE mode initialized successfully\n");
1540         else
1541                 kvm_info("Hyp mode initialized successfully\n");
1542
1543         return 0;
1544
1545 out_hyp:
1546         if (!in_hyp_mode)
1547                 teardown_hyp_mode();
1548 out_err:
1549         return err;
1550 }
1551
1552 /* NOP: Compiling as a module not supported */
1553 void kvm_arch_exit(void)
1554 {
1555         kvm_perf_teardown();
1556 }
1557
1558 static int arm_init(void)
1559 {
1560         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1561         return rc;
1562 }
1563
1564 module_init(arm_init);