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/bug.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/list.h>
  25 #include <linux/module.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/fs.h>
  28 #include <linux/mman.h>
  29 #include <linux/sched.h>
  30 #include <linux/kvm.h>
  31 #include <linux/kvm_irqfd.h>
  32 #include <linux/irqbypass.h>
  33 #include <trace/events/kvm.h>
  34 #include <kvm/arm_pmu.h>
  35 #include <kvm/arm_psci.h>
  36
  37 #define CREATE_TRACE_POINTS
  38 #include "trace.h"
  39
  40 #include <linux/uaccess.h>
  41 #include <asm/ptrace.h>
  42 #include <asm/mman.h>
  43 #include <asm/tlbflush.h>
  44 #include <asm/cacheflush.h>
  45 #include <asm/cpufeature.h>
  46 #include <asm/virt.h>
  47 #include <asm/kvm_arm.h>
  48 #include <asm/kvm_asm.h>
  49 #include <asm/kvm_mmu.h>
  50 #include <asm/kvm_emulate.h>
  51 #include <asm/kvm_coproc.h>
  52 #include <asm/sections.h>
  53
  54 #ifdef REQUIRES_VIRT
  55 __asm__(".arch_extension        virt");
  56 #endif
  57
  58 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  59 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  60
  61 /* Per-CPU variable containing the currently running vcpu. */
  62 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  63
  64 /* The VMID used in the VTTBR */
  65 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  66 static u32 kvm_next_vmid;
  67 static unsigned int kvm_vmid_bits __read_mostly;
  68 static DEFINE_RWLOCK(kvm_vmid_lock);
  69
  70 static bool vgic_present;
  71
  72 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  73
  74 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  75 {
  76         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  77 }
  78
  79 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  80
  81 /**
  82  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  83  * Must be called from non-preemptible context
  84  */
  85 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  86 {
  87         return __this_cpu_read(kvm_arm_running_vcpu);
  88 }
  89
  90 /**
  91  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  92  */
  93 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  94 {
  95         return &kvm_arm_running_vcpu;
  96 }
  97
  98 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  99 {
 100         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
 101 }
 102
 103 int kvm_arch_hardware_setup(void)
 104 {
 105         return 0;
 106 }
 107
 108 void kvm_arch_check_processor_compat(void *rtn)
 109 {
 110         *(int *)rtn = 0;
 111 }
 112
 113
 114 /**
 115  * kvm_arch_init_vm - initializes a VM data structure
 116  * @kvm:        pointer to the KVM struct
 117  */
 118 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 119 {
 120         int ret, cpu;
 121
 122         if (type)
 123                 return -EINVAL;
 124
 125         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 126         if (!kvm->arch.last_vcpu_ran)
 127                 return -ENOMEM;
 128
 129         for_each_possible_cpu(cpu)
 130                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 131
 132         ret = kvm_alloc_stage2_pgd(kvm);
 133         if (ret)
 134                 goto out_fail_alloc;
 135
 136         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 137         if (ret)
 138                 goto out_free_stage2_pgd;
 139
 140         kvm_vgic_early_init(kvm);
 141
 142         /* Mark the initial VMID generation invalid */
 143         kvm->arch.vmid_gen = 0;
 144
 145         /* The maximum number of VCPUs is limited by the host's GIC model */
 146         kvm->arch.max_vcpus = vgic_present ?
 147                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 148
 149         return ret;
 150 out_free_stage2_pgd:
 151         kvm_free_stage2_pgd(kvm);
 152 out_fail_alloc:
 153         free_percpu(kvm->arch.last_vcpu_ran);
 154         kvm->arch.last_vcpu_ran = NULL;
 155         return ret;
 156 }
 157
 158 bool kvm_arch_has_vcpu_debugfs(void)
 159 {
 160         return false;
 161 }
 162
 163 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 164 {
 165         return 0;
 166 }
 167
 168 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 169 {
 170         return VM_FAULT_SIGBUS;
 171 }
 172
 173
 174 /**
 175  * kvm_arch_destroy_vm - destroy the VM data structure
 176  * @kvm:        pointer to the KVM struct
 177  */
 178 void kvm_arch_destroy_vm(struct kvm *kvm)
 179 {
 180         int i;
 181
 182         kvm_vgic_destroy(kvm);
 183
 184         free_percpu(kvm->arch.last_vcpu_ran);
 185         kvm->arch.last_vcpu_ran = NULL;
 186
 187         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 188                 if (kvm->vcpus[i]) {
 189                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 190                         kvm->vcpus[i] = NULL;
 191                 }
 192         }
 193         atomic_set(&kvm->online_vcpus, 0);
 194 }
 195
 196 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 197 {
 198         int r;
 199         switch (ext) {
 200         case KVM_CAP_IRQCHIP:
 201                 r = vgic_present;
 202                 break;
 203         case KVM_CAP_IOEVENTFD:
 204         case KVM_CAP_DEVICE_CTRL:
 205         case KVM_CAP_USER_MEMORY:
 206         case KVM_CAP_SYNC_MMU:
 207         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 208         case KVM_CAP_ONE_REG:
 209         case KVM_CAP_ARM_PSCI:
 210         case KVM_CAP_ARM_PSCI_0_2:
 211         case KVM_CAP_READONLY_MEM:
 212         case KVM_CAP_MP_STATE:
 213         case KVM_CAP_IMMEDIATE_EXIT:
 214                 r = 1;
 215                 break;
 216         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 217                 r = 1;
 218                 break;
 219         case KVM_CAP_NR_VCPUS:
 220                 r = num_online_cpus();
 221                 break;
 222         case KVM_CAP_MAX_VCPUS:
 223                 r = KVM_MAX_VCPUS;
 224                 break;
 225         case KVM_CAP_NR_MEMSLOTS:
 226                 r = KVM_USER_MEM_SLOTS;
 227                 break;
 228         case KVM_CAP_MSI_DEVID:
 229                 if (!kvm)
 230                         r = -EINVAL;
 231                 else
 232                         r = kvm->arch.vgic.msis_require_devid;
 233                 break;
 234         case KVM_CAP_ARM_USER_IRQ:
 235                 /*
 236                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 237                  * (bump this number if adding more devices)
 238                  */
 239                 r = 1;
 240                 break;
 241         default:
 242                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 243                 break;
 244         }
 245         return r;
 246 }
 247
 248 long kvm_arch_dev_ioctl(struct file *filp,
 249                         unsigned int ioctl, unsigned long arg)
 250 {
 251         return -EINVAL;
 252 }
 253
 254 struct kvm *kvm_arch_alloc_vm(void)
 255 {
 256         if (!has_vhe())
 257                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 258
 259         return vzalloc(sizeof(struct kvm));
 260 }
 261
 262 void kvm_arch_free_vm(struct kvm *kvm)
 263 {
 264         if (!has_vhe())
 265                 kfree(kvm);
 266         else
 267                 vfree(kvm);
 268 }
 269
 270 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 271 {
 272         int err;
 273         struct kvm_vcpu *vcpu;
 274
 275         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 276                 err = -EBUSY;
 277                 goto out;
 278         }
 279
 280         if (id >= kvm->arch.max_vcpus) {
 281                 err = -EINVAL;
 282                 goto out;
 283         }
 284
 285         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 286         if (!vcpu) {
 287                 err = -ENOMEM;
 288                 goto out;
 289         }
 290
 291         err = kvm_vcpu_init(vcpu, kvm, id);
 292         if (err)
 293                 goto free_vcpu;
 294
 295         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 296         if (err)
 297                 goto vcpu_uninit;
 298
 299         return vcpu;
 300 vcpu_uninit:
 301         kvm_vcpu_uninit(vcpu);
 302 free_vcpu:
 303         kmem_cache_free(kvm_vcpu_cache, vcpu);
 304 out:
 305         return ERR_PTR(err);
 306 }
 307
 308 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 309 {
 310 }
 311
 312 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 313 {
 314         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 315                 static_branch_dec(&userspace_irqchip_in_use);
 316
 317         kvm_mmu_free_memory_caches(vcpu);
 318         kvm_timer_vcpu_terminate(vcpu);
 319         kvm_pmu_vcpu_destroy(vcpu);
 320         kvm_vcpu_uninit(vcpu);
 321         kmem_cache_free(kvm_vcpu_cache, vcpu);
 322 }
 323
 324 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 325 {
 326         kvm_arch_vcpu_free(vcpu);
 327 }
 328
 329 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 330 {
 331         return kvm_timer_is_pending(vcpu);
 332 }
 333
 334 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 335 {
 336         kvm_timer_schedule(vcpu);
 337         kvm_vgic_v4_enable_doorbell(vcpu);
 338 }
 339
 340 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 341 {
 342         kvm_timer_unschedule(vcpu);
 343         kvm_vgic_v4_disable_doorbell(vcpu);
 344 }
 345
 346 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 347 {
 348         /* Force users to call KVM_ARM_VCPU_INIT */
 349         vcpu->arch.target = -1;
 350         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 351
 352         /* Set up the timer */
 353         kvm_timer_vcpu_init(vcpu);
 354
 355         kvm_arm_reset_debug_ptr(vcpu);
 356
 357         return kvm_vgic_vcpu_init(vcpu);
 358 }
 359
 360 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 361 {
 362         int *last_ran;
 363
 364         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 365
 366         /*
 367          * We might get preempted before the vCPU actually runs, but
 368          * over-invalidation doesn't affect correctness.
 369          */
 370         if (*last_ran != vcpu->vcpu_id) {
 371                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 372                 *last_ran = vcpu->vcpu_id;
 373         }
 374
 375         vcpu->cpu = cpu;
 376         vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
 377
 378         kvm_arm_set_running_vcpu(vcpu);
 379         kvm_vgic_load(vcpu);
 380         kvm_timer_vcpu_load(vcpu);
 381         kvm_vcpu_load_sysregs(vcpu);
 382         kvm_arch_vcpu_load_fp(vcpu);
 383 }
 384
 385 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 386 {
 387         kvm_arch_vcpu_put_fp(vcpu);
 388         kvm_vcpu_put_sysregs(vcpu);
 389         kvm_timer_vcpu_put(vcpu);
 390         kvm_vgic_put(vcpu);
 391
 392         vcpu->cpu = -1;
 393
 394         kvm_arm_set_running_vcpu(NULL);
 395 }
 396
 397 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 398 {
 399         vcpu->arch.power_off = true;
 400         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 401         kvm_vcpu_kick(vcpu);
 402 }
 403
 404 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 405                                     struct kvm_mp_state *mp_state)
 406 {
 407         if (vcpu->arch.power_off)
 408                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 409         else
 410                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 411
 412         return 0;
 413 }
 414
 415 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 416                                     struct kvm_mp_state *mp_state)
 417 {
 418         int ret = 0;
 419
 420         switch (mp_state->mp_state) {
 421         case KVM_MP_STATE_RUNNABLE:
 422                 vcpu->arch.power_off = false;
 423                 break;
 424         case KVM_MP_STATE_STOPPED:
 425                 vcpu_power_off(vcpu);
 426                 break;
 427         default:
 428                 ret = -EINVAL;
 429         }
 430
 431         return ret;
 432 }
 433
 434 /**
 435  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 436  * @v:          The VCPU pointer
 437  *
 438  * If the guest CPU is not waiting for interrupts or an interrupt line is
 439  * asserted, the CPU is by definition runnable.
 440  */
 441 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 442 {
 443         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 444         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 445                 && !v->arch.power_off && !v->arch.pause);
 446 }
 447
 448 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 449 {
 450         return vcpu_mode_priv(vcpu);
 451 }
 452
 453 /* Just ensure a guest exit from a particular CPU */
 454 static void exit_vm_noop(void *info)
 455 {
 456 }
 457
 458 void force_vm_exit(const cpumask_t *mask)
 459 {
 460         preempt_disable();
 461         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 462         preempt_enable();
 463 }
 464
 465 /**
 466  * need_new_vmid_gen - check that the VMID is still valid
 467  * @kvm: The VM's VMID to check
 468  *
 469  * return true if there is a new generation of VMIDs being used
 470  *
 471  * The hardware supports only 256 values with the value zero reserved for the
 472  * host, so we check if an assigned value belongs to a previous generation,
 473  * which which requires us to assign a new value. If we're the first to use a
 474  * VMID for the new generation, we must flush necessary caches and TLBs on all
 475  * CPUs.
 476  */
 477 static bool need_new_vmid_gen(struct kvm *kvm)
 478 {
 479         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 480 }
 481
 482 /**
 483  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 484  * @kvm The guest that we are about to run
 485  *
 486  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 487  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 488  * caches and TLBs.
 489  */
 490 static void update_vttbr(struct kvm *kvm)
 491 {
 492         phys_addr_t pgd_phys;
 493         u64 vmid;
 494         bool new_gen;
 495
 496         read_lock(&kvm_vmid_lock);
 497         new_gen = need_new_vmid_gen(kvm);
 498         read_unlock(&kvm_vmid_lock);
 499
 500         if (!new_gen)
 501                 return;
 502
 503         write_lock(&kvm_vmid_lock);
 504
 505         /*
 506          * We need to re-check the vmid_gen here to ensure that if another vcpu
 507          * already allocated a valid vmid for this vm, then this vcpu should
 508          * use the same vmid.
 509          */
 510         if (!need_new_vmid_gen(kvm)) {
 511                 write_unlock(&kvm_vmid_lock);
 512                 return;
 513         }
 514
 515         /* First user of a new VMID generation? */
 516         if (unlikely(kvm_next_vmid == 0)) {
 517                 atomic64_inc(&kvm_vmid_gen);
 518                 kvm_next_vmid = 1;
 519
 520                 /*
 521                  * On SMP we know no other CPUs can use this CPU's or each
 522                  * other's VMID after force_vm_exit returns since the
 523                  * kvm_vmid_lock blocks them from reentry to the guest.
 524                  */
 525                 force_vm_exit(cpu_all_mask);
 526                 /*
 527                  * Now broadcast TLB + ICACHE invalidation over the inner
 528                  * shareable domain to make sure all data structures are
 529                  * clean.
 530                  */
 531                 kvm_call_hyp(__kvm_flush_vm_context);
 532         }
 533
 534         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 535         kvm->arch.vmid = kvm_next_vmid;
 536         kvm_next_vmid++;
 537         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 538
 539         /* update vttbr to be used with the new vmid */
 540         pgd_phys = virt_to_phys(kvm->arch.pgd);
 541         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 542         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 543         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid;
 544
 545         write_unlock(&kvm_vmid_lock);
 546 }
 547
 548 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 549 {
 550         struct kvm *kvm = vcpu->kvm;
 551         int ret = 0;
 552
 553         if (likely(vcpu->arch.has_run_once))
 554                 return 0;
 555
 556         vcpu->arch.has_run_once = true;
 557
 558         if (likely(irqchip_in_kernel(kvm))) {
 559                 /*
 560                  * Map the VGIC hardware resources before running a vcpu the
 561                  * first time on this VM.
 562                  */
 563                 if (unlikely(!vgic_ready(kvm))) {
 564                         ret = kvm_vgic_map_resources(kvm);
 565                         if (ret)
 566                                 return ret;
 567                 }
 568         } else {
 569                 /*
 570                  * Tell the rest of the code that there are userspace irqchip
 571                  * VMs in the wild.
 572                  */
 573                 static_branch_inc(&userspace_irqchip_in_use);
 574         }
 575
 576         ret = kvm_timer_enable(vcpu);
 577         if (ret)
 578                 return ret;
 579
 580         ret = kvm_arm_pmu_v3_enable(vcpu);
 581
 582         return ret;
 583 }
 584
 585 bool kvm_arch_intc_initialized(struct kvm *kvm)
 586 {
 587         return vgic_initialized(kvm);
 588 }
 589
 590 void kvm_arm_halt_guest(struct kvm *kvm)
 591 {
 592         int i;
 593         struct kvm_vcpu *vcpu;
 594
 595         kvm_for_each_vcpu(i, vcpu, kvm)
 596                 vcpu->arch.pause = true;
 597         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 598 }
 599
 600 void kvm_arm_resume_guest(struct kvm *kvm)
 601 {
 602         int i;
 603         struct kvm_vcpu *vcpu;
 604
 605         kvm_for_each_vcpu(i, vcpu, kvm) {
 606                 vcpu->arch.pause = false;
 607                 swake_up(kvm_arch_vcpu_wq(vcpu));
 608         }
 609 }
 610
 611 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 612 {
 613         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 614
 615         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 616                                        (!vcpu->arch.pause)));
 617
 618         if (vcpu->arch.power_off || vcpu->arch.pause) {
 619                 /* Awaken to handle a signal, request we sleep again later. */
 620                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 621         }
 622 }
 623
 624 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 625 {
 626         return vcpu->arch.target >= 0;
 627 }
 628
 629 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 630 {
 631         if (kvm_request_pending(vcpu)) {
 632                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 633                         vcpu_req_sleep(vcpu);
 634
 635                 /*
 636                  * Clear IRQ_PENDING requests that were made to guarantee
 637                  * that a VCPU sees new virtual interrupts.
 638                  */
 639                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 640         }
 641 }
 642
 643 /**
 644  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 645  * @vcpu:       The VCPU pointer
 646  * @run:        The kvm_run structure pointer used for userspace state exchange
 647  *
 648  * This function is called through the VCPU_RUN ioctl called from user space. It
 649  * will execute VM code in a loop until the time slice for the process is used
 650  * or some emulation is needed from user space in which case the function will
 651  * return with return value 0 and with the kvm_run structure filled in with the
 652  * required data for the requested emulation.
 653  */
 654 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 655 {
 656         int ret;
 657
 658         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 659                 return -ENOEXEC;
 660
 661         ret = kvm_vcpu_first_run_init(vcpu);
 662         if (ret)
 663                 return ret;
 664
 665         if (run->exit_reason == KVM_EXIT_MMIO) {
 666                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 667                 if (ret)
 668                         return ret;
 669                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
 670                         return 0;
 671         }
 672
 673         if (run->immediate_exit)
 674                 return -EINTR;
 675
 676         vcpu_load(vcpu);
 677
 678         kvm_sigset_activate(vcpu);
 679
 680         ret = 1;
 681         run->exit_reason = KVM_EXIT_UNKNOWN;
 682         while (ret > 0) {
 683                 /*
 684                  * Check conditions before entering the guest
 685                  */
 686                 cond_resched();
 687
 688                 update_vttbr(vcpu->kvm);
 689
 690                 check_vcpu_requests(vcpu);
 691
 692                 /*
 693                  * Preparing the interrupts to be injected also
 694                  * involves poking the GIC, which must be done in a
 695                  * non-preemptible context.
 696                  */
 697                 preempt_disable();
 698
 699                 kvm_pmu_flush_hwstate(vcpu);
 700
 701                 local_irq_disable();
 702
 703                 kvm_vgic_flush_hwstate(vcpu);
 704
 705                 /*
 706                  * Exit if we have a signal pending so that we can deliver the
 707                  * signal to user space.
 708                  */
 709                 if (signal_pending(current)) {
 710                         ret = -EINTR;
 711                         run->exit_reason = KVM_EXIT_INTR;
 712                 }
 713
 714                 /*
 715                  * If we're using a userspace irqchip, then check if we need
 716                  * to tell a userspace irqchip about timer or PMU level
 717                  * changes and if so, exit to userspace (the actual level
 718                  * state gets updated in kvm_timer_update_run and
 719                  * kvm_pmu_update_run below).
 720                  */
 721                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 722                         if (kvm_timer_should_notify_user(vcpu) ||
 723                             kvm_pmu_should_notify_user(vcpu)) {
 724                                 ret = -EINTR;
 725                                 run->exit_reason = KVM_EXIT_INTR;
 726                         }
 727                 }
 728
 729                 /*
 730                  * Ensure we set mode to IN_GUEST_MODE after we disable
 731                  * interrupts and before the final VCPU requests check.
 732                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 733                  * Documentation/virtual/kvm/vcpu-requests.rst
 734                  */
 735                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 736
 737                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 738                     kvm_request_pending(vcpu)) {
 739                         vcpu->mode = OUTSIDE_GUEST_MODE;
 740                         isb(); /* Ensure work in x_flush_hwstate is committed */
 741                         kvm_pmu_sync_hwstate(vcpu);
 742                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 743                                 kvm_timer_sync_hwstate(vcpu);
 744                         kvm_vgic_sync_hwstate(vcpu);
 745                         local_irq_enable();
 746                         preempt_enable();
 747                         continue;
 748                 }
 749
 750                 kvm_arm_setup_debug(vcpu);
 751
 752                 /**************************************************************
 753                  * Enter the guest
 754                  */
 755                 trace_kvm_entry(*vcpu_pc(vcpu));
 756                 guest_enter_irqoff();
 757
 758                 if (has_vhe()) {
 759                         kvm_arm_vhe_guest_enter();
 760                         ret = kvm_vcpu_run_vhe(vcpu);
 761                         kvm_arm_vhe_guest_exit();
 762                 } else {
 763                         ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
 764                 }
 765
 766                 vcpu->mode = OUTSIDE_GUEST_MODE;
 767                 vcpu->stat.exits++;
 768                 /*
 769                  * Back from guest
 770                  *************************************************************/
 771
 772                 kvm_arm_clear_debug(vcpu);
 773
 774                 /*
 775                  * We must sync the PMU state before the vgic state so
 776                  * that the vgic can properly sample the updated state of the
 777                  * interrupt line.
 778                  */
 779                 kvm_pmu_sync_hwstate(vcpu);
 780
 781                 /*
 782                  * Sync the vgic state before syncing the timer state because
 783                  * the timer code needs to know if the virtual timer
 784                  * interrupts are active.
 785                  */
 786                 kvm_vgic_sync_hwstate(vcpu);
 787
 788                 /*
 789                  * Sync the timer hardware state before enabling interrupts as
 790                  * we don't want vtimer interrupts to race with syncing the
 791                  * timer virtual interrupt state.
 792                  */
 793                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 794                         kvm_timer_sync_hwstate(vcpu);
 795
 796                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 797
 798                 /*
 799                  * We may have taken a host interrupt in HYP mode (ie
 800                  * while executing the guest). This interrupt is still
 801                  * pending, as we haven't serviced it yet!
 802                  *
 803                  * We're now back in SVC mode, with interrupts
 804                  * disabled.  Enabling the interrupts now will have
 805                  * the effect of taking the interrupt again, in SVC
 806                  * mode this time.
 807                  */
 808                 local_irq_enable();
 809
 810                 /*
 811                  * We do local_irq_enable() before calling guest_exit() so
 812                  * that if a timer interrupt hits while running the guest we
 813                  * account that tick as being spent in the guest.  We enable
 814                  * preemption after calling guest_exit() so that if we get
 815                  * preempted we make sure ticks after that is not counted as
 816                  * guest time.
 817                  */
 818                 guest_exit();
 819                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 820
 821                 /* Exit types that need handling before we can be preempted */
 822                 handle_exit_early(vcpu, run, ret);
 823
 824                 preempt_enable();
 825
 826                 ret = handle_exit(vcpu, run, ret);
 827         }
 828
 829         /* Tell userspace about in-kernel device output levels */
 830         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 831                 kvm_timer_update_run(vcpu);
 832                 kvm_pmu_update_run(vcpu);
 833         }
 834
 835         kvm_sigset_deactivate(vcpu);
 836
 837         vcpu_put(vcpu);
 838         return ret;
 839 }
 840
 841 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 842 {
 843         int bit_index;
 844         bool set;
 845         unsigned long *hcr;
 846
 847         if (number == KVM_ARM_IRQ_CPU_IRQ)
 848                 bit_index = __ffs(HCR_VI);
 849         else /* KVM_ARM_IRQ_CPU_FIQ */
 850                 bit_index = __ffs(HCR_VF);
 851
 852         hcr = vcpu_hcr(vcpu);
 853         if (level)
 854                 set = test_and_set_bit(bit_index, hcr);
 855         else
 856                 set = test_and_clear_bit(bit_index, hcr);
 857
 858         /*
 859          * If we didn't change anything, no need to wake up or kick other CPUs
 860          */
 861         if (set == level)
 862                 return 0;
 863
 864         /*
 865          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 866          * trigger a world-switch round on the running physical CPU to set the
 867          * virtual IRQ/FIQ fields in the HCR appropriately.
 868          */
 869         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 870         kvm_vcpu_kick(vcpu);
 871
 872         return 0;
 873 }
 874
 875 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 876                           bool line_status)
 877 {
 878         u32 irq = irq_level->irq;
 879         unsigned int irq_type, vcpu_idx, irq_num;
 880         int nrcpus = atomic_read(&kvm->online_vcpus);
 881         struct kvm_vcpu *vcpu = NULL;
 882         bool level = irq_level->level;
 883
 884         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 885         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 886         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 887
 888         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 889
 890         switch (irq_type) {
 891         case KVM_ARM_IRQ_TYPE_CPU:
 892                 if (irqchip_in_kernel(kvm))
 893                         return -ENXIO;
 894
 895                 if (vcpu_idx >= nrcpus)
 896                         return -EINVAL;
 897
 898                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 899                 if (!vcpu)
 900                         return -EINVAL;
 901
 902                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 903                         return -EINVAL;
 904
 905                 return vcpu_interrupt_line(vcpu, irq_num, level);
 906         case KVM_ARM_IRQ_TYPE_PPI:
 907                 if (!irqchip_in_kernel(kvm))
 908                         return -ENXIO;
 909
 910                 if (vcpu_idx >= nrcpus)
 911                         return -EINVAL;
 912
 913                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 914                 if (!vcpu)
 915                         return -EINVAL;
 916
 917                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 918                         return -EINVAL;
 919
 920                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 921         case KVM_ARM_IRQ_TYPE_SPI:
 922                 if (!irqchip_in_kernel(kvm))
 923                         return -ENXIO;
 924
 925                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 926                         return -EINVAL;
 927
 928                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 929         }
 930
 931         return -EINVAL;
 932 }
 933
 934 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 935                                const struct kvm_vcpu_init *init)
 936 {
 937         unsigned int i;
 938         int phys_target = kvm_target_cpu();
 939
 940         if (init->target != phys_target)
 941                 return -EINVAL;
 942
 943         /*
 944          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 945          * use the same target.
 946          */
 947         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 948                 return -EINVAL;
 949
 950         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 951         for (i = 0; i < sizeof(init->features) * 8; i++) {
 952                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 953
 954                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 955                         return -ENOENT;
 956
 957                 /*
 958                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 959                  * use the same feature set.
 960                  */
 961                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 962                     test_bit(i, vcpu->arch.features) != set)
 963                         return -EINVAL;
 964
 965                 if (set)
 966                         set_bit(i, vcpu->arch.features);
 967         }
 968
 969         vcpu->arch.target = phys_target;
 970
 971         /* Now we know what it is, we can reset it. */
 972         return kvm_reset_vcpu(vcpu);
 973 }
 974
 975
 976 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 977                                          struct kvm_vcpu_init *init)
 978 {
 979         int ret;
 980
 981         ret = kvm_vcpu_set_target(vcpu, init);
 982         if (ret)
 983                 return ret;
 984
 985         /*
 986          * Ensure a rebooted VM will fault in RAM pages and detect if the
 987          * guest MMU is turned off and flush the caches as needed.
 988          */
 989         if (vcpu->arch.has_run_once)
 990                 stage2_unmap_vm(vcpu->kvm);
 991
 992         vcpu_reset_hcr(vcpu);
 993
 994         /*
 995          * Handle the "start in power-off" case.
 996          */
 997         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 998                 vcpu_power_off(vcpu);
 999         else
1000                 vcpu->arch.power_off = false;
1001
1002         return 0;
1003 }
1004
1005 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1006                                  struct kvm_device_attr *attr)
1007 {
1008         int ret = -ENXIO;
1009
1010         switch (attr->group) {
1011         default:
1012                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1013                 break;
1014         }
1015
1016         return ret;
1017 }
1018
1019 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1020                                  struct kvm_device_attr *attr)
1021 {
1022         int ret = -ENXIO;
1023
1024         switch (attr->group) {
1025         default:
1026                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1027                 break;
1028         }
1029
1030         return ret;
1031 }
1032
1033 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1034                                  struct kvm_device_attr *attr)
1035 {
1036         int ret = -ENXIO;
1037
1038         switch (attr->group) {
1039         default:
1040                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1041                 break;
1042         }
1043
1044         return ret;
1045 }
1046
1047 long kvm_arch_vcpu_ioctl(struct file *filp,
1048                          unsigned int ioctl, unsigned long arg)
1049 {
1050         struct kvm_vcpu *vcpu = filp->private_data;
1051         void __user *argp = (void __user *)arg;
1052         struct kvm_device_attr attr;
1053         long r;
1054
1055         switch (ioctl) {
1056         case KVM_ARM_VCPU_INIT: {
1057                 struct kvm_vcpu_init init;
1058
1059                 r = -EFAULT;
1060                 if (copy_from_user(&init, argp, sizeof(init)))
1061                         break;
1062
1063                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1064                 break;
1065         }
1066         case KVM_SET_ONE_REG:
1067         case KVM_GET_ONE_REG: {
1068                 struct kvm_one_reg reg;
1069
1070                 r = -ENOEXEC;
1071                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1072                         break;
1073
1074                 r = -EFAULT;
1075                 if (copy_from_user(&reg, argp, sizeof(reg)))
1076                         break;
1077
1078                 if (ioctl == KVM_SET_ONE_REG)
1079                         r = kvm_arm_set_reg(vcpu, &reg);
1080                 else
1081                         r = kvm_arm_get_reg(vcpu, &reg);
1082                 break;
1083         }
1084         case KVM_GET_REG_LIST: {
1085                 struct kvm_reg_list __user *user_list = argp;
1086                 struct kvm_reg_list reg_list;
1087                 unsigned n;
1088
1089                 r = -ENOEXEC;
1090                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1091                         break;
1092
1093                 r = -EFAULT;
1094                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1095                         break;
1096                 n = reg_list.n;
1097                 reg_list.n = kvm_arm_num_regs(vcpu);
1098                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1099                         break;
1100                 r = -E2BIG;
1101                 if (n < reg_list.n)
1102                         break;
1103                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1104                 break;
1105         }
1106         case KVM_SET_DEVICE_ATTR: {
1107                 r = -EFAULT;
1108                 if (copy_from_user(&attr, argp, sizeof(attr)))
1109                         break;
1110                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1111                 break;
1112         }
1113         case KVM_GET_DEVICE_ATTR: {
1114                 r = -EFAULT;
1115                 if (copy_from_user(&attr, argp, sizeof(attr)))
1116                         break;
1117                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1118                 break;
1119         }
1120         case KVM_HAS_DEVICE_ATTR: {
1121                 r = -EFAULT;
1122                 if (copy_from_user(&attr, argp, sizeof(attr)))
1123                         break;
1124                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1125                 break;
1126         }
1127         default:
1128                 r = -EINVAL;
1129         }
1130
1131         return r;
1132 }
1133
1134 /**
1135  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1136  * @kvm: kvm instance
1137  * @log: slot id and address to which we copy the log
1138  *
1139  * Steps 1-4 below provide general overview of dirty page logging. See
1140  * kvm_get_dirty_log_protect() function description for additional details.
1141  *
1142  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1143  * always flush the TLB (step 4) even if previous step failed  and the dirty
1144  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1145  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1146  * writes will be marked dirty for next log read.
1147  *
1148  *   1. Take a snapshot of the bit and clear it if needed.
1149  *   2. Write protect the corresponding page.
1150  *   3. Copy the snapshot to the userspace.
1151  *   4. Flush TLB's if needed.
1152  */
1153 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1154 {
1155         bool is_dirty = false;
1156         int r;
1157
1158         mutex_lock(&kvm->slots_lock);
1159
1160         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1161
1162         if (is_dirty)
1163                 kvm_flush_remote_tlbs(kvm);
1164
1165         mutex_unlock(&kvm->slots_lock);
1166         return r;
1167 }
1168
1169 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1170                                         struct kvm_arm_device_addr *dev_addr)
1171 {
1172         unsigned long dev_id, type;
1173
1174         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1175                 KVM_ARM_DEVICE_ID_SHIFT;
1176         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1177                 KVM_ARM_DEVICE_TYPE_SHIFT;
1178
1179         switch (dev_id) {
1180         case KVM_ARM_DEVICE_VGIC_V2:
1181                 if (!vgic_present)
1182                         return -ENXIO;
1183                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1184         default:
1185                 return -ENODEV;
1186         }
1187 }
1188
1189 long kvm_arch_vm_ioctl(struct file *filp,
1190                        unsigned int ioctl, unsigned long arg)
1191 {
1192         struct kvm *kvm = filp->private_data;
1193         void __user *argp = (void __user *)arg;
1194
1195         switch (ioctl) {
1196         case KVM_CREATE_IRQCHIP: {
1197                 int ret;
1198                 if (!vgic_present)
1199                         return -ENXIO;
1200                 mutex_lock(&kvm->lock);
1201                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1202                 mutex_unlock(&kvm->lock);
1203                 return ret;
1204         }
1205         case KVM_ARM_SET_DEVICE_ADDR: {
1206                 struct kvm_arm_device_addr dev_addr;
1207
1208                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1209                         return -EFAULT;
1210                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1211         }
1212         case KVM_ARM_PREFERRED_TARGET: {
1213                 int err;
1214                 struct kvm_vcpu_init init;
1215
1216                 err = kvm_vcpu_preferred_target(&init);
1217                 if (err)
1218                         return err;
1219
1220                 if (copy_to_user(argp, &init, sizeof(init)))
1221                         return -EFAULT;
1222
1223                 return 0;
1224         }
1225         default:
1226                 return -EINVAL;
1227         }
1228 }
1229
1230 static void cpu_init_hyp_mode(void *dummy)
1231 {
1232         phys_addr_t pgd_ptr;
1233         unsigned long hyp_stack_ptr;
1234         unsigned long stack_page;
1235         unsigned long vector_ptr;
1236
1237         /* Switch from the HYP stub to our own HYP init vector */
1238         __hyp_set_vectors(kvm_get_idmap_vector());
1239
1240         pgd_ptr = kvm_mmu_get_httbr();
1241         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1242         hyp_stack_ptr = stack_page + PAGE_SIZE;
1243         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1244
1245         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1246         __cpu_init_stage2();
1247
1248         kvm_arm_init_debug();
1249 }
1250
1251 static void cpu_hyp_reset(void)
1252 {
1253         if (!is_kernel_in_hyp_mode())
1254                 __hyp_reset_vectors();
1255 }
1256
1257 static void cpu_hyp_reinit(void)
1258 {
1259         cpu_hyp_reset();
1260
1261         if (is_kernel_in_hyp_mode()) {
1262                 /*
1263                  * __cpu_init_stage2() is safe to call even if the PM
1264                  * event was cancelled before the CPU was reset.
1265                  */
1266                 __cpu_init_stage2();
1267                 kvm_timer_init_vhe();
1268         } else {
1269                 cpu_init_hyp_mode(NULL);
1270         }
1271
1272         if (vgic_present)
1273                 kvm_vgic_init_cpu_hardware();
1274 }
1275
1276 static void _kvm_arch_hardware_enable(void *discard)
1277 {
1278         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1279                 cpu_hyp_reinit();
1280                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1281         }
1282 }
1283
1284 int kvm_arch_hardware_enable(void)
1285 {
1286         _kvm_arch_hardware_enable(NULL);
1287         return 0;
1288 }
1289
1290 static void _kvm_arch_hardware_disable(void *discard)
1291 {
1292         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1293                 cpu_hyp_reset();
1294                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1295         }
1296 }
1297
1298 void kvm_arch_hardware_disable(void)
1299 {
1300         _kvm_arch_hardware_disable(NULL);
1301 }
1302
1303 #ifdef CONFIG_CPU_PM
1304 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1305                                     unsigned long cmd,
1306                                     void *v)
1307 {
1308         /*
1309          * kvm_arm_hardware_enabled is left with its old value over
1310          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1311          * re-enable hyp.
1312          */
1313         switch (cmd) {
1314         case CPU_PM_ENTER:
1315                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1316                         /*
1317                          * don't update kvm_arm_hardware_enabled here
1318                          * so that the hardware will be re-enabled
1319                          * when we resume. See below.
1320                          */
1321                         cpu_hyp_reset();
1322
1323                 return NOTIFY_OK;
1324         case CPU_PM_ENTER_FAILED:
1325         case CPU_PM_EXIT:
1326                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1327                         /* The hardware was enabled before suspend. */
1328                         cpu_hyp_reinit();
1329
1330                 return NOTIFY_OK;
1331
1332         default:
1333                 return NOTIFY_DONE;
1334         }
1335 }
1336
1337 static struct notifier_block hyp_init_cpu_pm_nb = {
1338         .notifier_call = hyp_init_cpu_pm_notifier,
1339 };
1340
1341 static void __init hyp_cpu_pm_init(void)
1342 {
1343         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1344 }
1345 static void __init hyp_cpu_pm_exit(void)
1346 {
1347         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1348 }
1349 #else
1350 static inline void hyp_cpu_pm_init(void)
1351 {
1352 }
1353 static inline void hyp_cpu_pm_exit(void)
1354 {
1355 }
1356 #endif
1357
1358 static int init_common_resources(void)
1359 {
1360         /* set size of VMID supported by CPU */
1361         kvm_vmid_bits = kvm_get_vmid_bits();
1362         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1363
1364         return 0;
1365 }
1366
1367 static int init_subsystems(void)
1368 {
1369         int err = 0;
1370
1371         /*
1372          * Enable hardware so that subsystem initialisation can access EL2.
1373          */
1374         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1375
1376         /*
1377          * Register CPU lower-power notifier
1378          */
1379         hyp_cpu_pm_init();
1380
1381         /*
1382          * Init HYP view of VGIC
1383          */
1384         err = kvm_vgic_hyp_init();
1385         switch (err) {
1386         case 0:
1387                 vgic_present = true;
1388                 break;
1389         case -ENODEV:
1390         case -ENXIO:
1391                 vgic_present = false;
1392                 err = 0;
1393                 break;
1394         default:
1395                 goto out;
1396         }
1397
1398         /*
1399          * Init HYP architected timer support
1400          */
1401         err = kvm_timer_hyp_init(vgic_present);
1402         if (err)
1403                 goto out;
1404
1405         kvm_perf_init();
1406         kvm_coproc_table_init();
1407
1408 out:
1409         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1410
1411         return err;
1412 }
1413
1414 static void teardown_hyp_mode(void)
1415 {
1416         int cpu;
1417
1418         free_hyp_pgds();
1419         for_each_possible_cpu(cpu)
1420                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1421         hyp_cpu_pm_exit();
1422 }
1423
1424 /**
1425  * Inits Hyp-mode on all online CPUs
1426  */
1427 static int init_hyp_mode(void)
1428 {
1429         int cpu;
1430         int err = 0;
1431
1432         /*
1433          * Allocate Hyp PGD and setup Hyp identity mapping
1434          */
1435         err = kvm_mmu_init();
1436         if (err)
1437                 goto out_err;
1438
1439         /*
1440          * Allocate stack pages for Hypervisor-mode
1441          */
1442         for_each_possible_cpu(cpu) {
1443                 unsigned long stack_page;
1444
1445                 stack_page = __get_free_page(GFP_KERNEL);
1446                 if (!stack_page) {
1447                         err = -ENOMEM;
1448                         goto out_err;
1449                 }
1450
1451                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1452         }
1453
1454         /*
1455          * Map the Hyp-code called directly from the host
1456          */
1457         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1458                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1459         if (err) {
1460                 kvm_err("Cannot map world-switch code\n");
1461                 goto out_err;
1462         }
1463
1464         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1465                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1466         if (err) {
1467                 kvm_err("Cannot map rodata section\n");
1468                 goto out_err;
1469         }
1470
1471         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1472                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1473         if (err) {
1474                 kvm_err("Cannot map bss section\n");
1475                 goto out_err;
1476         }
1477
1478         err = kvm_map_vectors();
1479         if (err) {
1480                 kvm_err("Cannot map vectors\n");
1481                 goto out_err;
1482         }
1483
1484         /*
1485          * Map the Hyp stack pages
1486          */
1487         for_each_possible_cpu(cpu) {
1488                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1489                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1490                                           PAGE_HYP);
1491
1492                 if (err) {
1493                         kvm_err("Cannot map hyp stack\n");
1494                         goto out_err;
1495                 }
1496         }
1497
1498         for_each_possible_cpu(cpu) {
1499                 kvm_cpu_context_t *cpu_ctxt;
1500
1501                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1502                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1503
1504                 if (err) {
1505                         kvm_err("Cannot map host CPU state: %d\n", err);
1506                         goto out_err;
1507                 }
1508         }
1509
1510         err = hyp_map_aux_data();
1511         if (err)
1512                 kvm_err("Cannot map host auxilary data: %d\n", err);
1513
1514         return 0;
1515
1516 out_err:
1517         teardown_hyp_mode();
1518         kvm_err("error initializing Hyp mode: %d\n", err);
1519         return err;
1520 }
1521
1522 static void check_kvm_target_cpu(void *ret)
1523 {
1524         *(int *)ret = kvm_target_cpu();
1525 }
1526
1527 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1528 {
1529         struct kvm_vcpu *vcpu;
1530         int i;
1531
1532         mpidr &= MPIDR_HWID_BITMASK;
1533         kvm_for_each_vcpu(i, vcpu, kvm) {
1534                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1535                         return vcpu;
1536         }
1537         return NULL;
1538 }
1539
1540 bool kvm_arch_has_irq_bypass(void)
1541 {
1542         return true;
1543 }
1544
1545 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1546                                       struct irq_bypass_producer *prod)
1547 {
1548         struct kvm_kernel_irqfd *irqfd =
1549                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1550
1551         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1552                                           &irqfd->irq_entry);
1553 }
1554 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1555                                       struct irq_bypass_producer *prod)
1556 {
1557         struct kvm_kernel_irqfd *irqfd =
1558                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1559
1560         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1561                                      &irqfd->irq_entry);
1562 }
1563
1564 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1565 {
1566         struct kvm_kernel_irqfd *irqfd =
1567                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1568
1569         kvm_arm_halt_guest(irqfd->kvm);
1570 }
1571
1572 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1573 {
1574         struct kvm_kernel_irqfd *irqfd =
1575                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1576
1577         kvm_arm_resume_guest(irqfd->kvm);
1578 }
1579
1580 /**
1581  * Initialize Hyp-mode and memory mappings on all CPUs.
1582  */
1583 int kvm_arch_init(void *opaque)
1584 {
1585         int err;
1586         int ret, cpu;
1587         bool in_hyp_mode;
1588
1589         if (!is_hyp_mode_available()) {
1590                 kvm_info("HYP mode not available\n");
1591                 return -ENODEV;
1592         }
1593
1594         if (!kvm_arch_check_sve_has_vhe()) {
1595                 kvm_pr_unimpl("SVE system without VHE unsupported.  Broken cpu?");
1596                 return -ENODEV;
1597         }
1598
1599         for_each_online_cpu(cpu) {
1600                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1601                 if (ret < 0) {
1602                         kvm_err("Error, CPU %d not supported!\n", cpu);
1603                         return -ENODEV;
1604                 }
1605         }
1606
1607         err = init_common_resources();
1608         if (err)
1609                 return err;
1610
1611         in_hyp_mode = is_kernel_in_hyp_mode();
1612
1613         if (!in_hyp_mode) {
1614                 err = init_hyp_mode();
1615                 if (err)
1616                         goto out_err;
1617         }
1618
1619         err = init_subsystems();
1620         if (err)
1621                 goto out_hyp;
1622
1623         if (in_hyp_mode)
1624                 kvm_info("VHE mode initialized successfully\n");
1625         else
1626                 kvm_info("Hyp mode initialized successfully\n");
1627
1628         return 0;
1629
1630 out_hyp:
1631         if (!in_hyp_mode)
1632                 teardown_hyp_mode();
1633 out_err:
1634         return err;
1635 }
1636
1637 /* NOP: Compiling as a module not supported */
1638 void kvm_arch_exit(void)
1639 {
1640         kvm_perf_teardown();
1641 }
1642
1643 static int arm_init(void)
1644 {
1645         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1646         return rc;
1647 }
1648
1649 module_init(arm_init);