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 <linux/sched/stat.h>
  34 #include <trace/events/kvm.h>
  35 #include <kvm/arm_pmu.h>
  36 #include <kvm/arm_psci.h>
  37
  38 #define CREATE_TRACE_POINTS
  39 #include "trace.h"
  40
  41 #include <linux/uaccess.h>
  42 #include <asm/ptrace.h>
  43 #include <asm/mman.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/cacheflush.h>
  46 #include <asm/cpufeature.h>
  47 #include <asm/virt.h>
  48 #include <asm/kvm_arm.h>
  49 #include <asm/kvm_asm.h>
  50 #include <asm/kvm_mmu.h>
  51 #include <asm/kvm_emulate.h>
  52 #include <asm/kvm_coproc.h>
  53 #include <asm/sections.h>
  54
  55 #ifdef REQUIRES_VIRT
  56 __asm__(".arch_extension        virt");
  57 #endif
  58
  59 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  60 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  61
  62 /* Per-CPU variable containing the currently running vcpu. */
  63 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  64
  65 /* The VMID used in the VTTBR */
  66 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  67 static u32 kvm_next_vmid;
  68 static DEFINE_SPINLOCK(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         ret = kvm_arm_setup_stage2(kvm, type);
 123         if (ret)
 124                 return ret;
 125
 126         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 127         if (!kvm->arch.last_vcpu_ran)
 128                 return -ENOMEM;
 129
 130         for_each_possible_cpu(cpu)
 131                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 132
 133         ret = kvm_alloc_stage2_pgd(kvm);
 134         if (ret)
 135                 goto out_fail_alloc;
 136
 137         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 138         if (ret)
 139                 goto out_free_stage2_pgd;
 140
 141         kvm_vgic_early_init(kvm);
 142
 143         /* Mark the initial VMID generation invalid */
 144         kvm->arch.vmid.vmid_gen = 0;
 145
 146         /* The maximum number of VCPUs is limited by the host's GIC model */
 147         kvm->arch.max_vcpus = vgic_present ?
 148                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 149
 150         return ret;
 151 out_free_stage2_pgd:
 152         kvm_free_stage2_pgd(kvm);
 153 out_fail_alloc:
 154         free_percpu(kvm->arch.last_vcpu_ran);
 155         kvm->arch.last_vcpu_ran = NULL;
 156         return ret;
 157 }
 158
 159 bool kvm_arch_has_vcpu_debugfs(void)
 160 {
 161         return false;
 162 }
 163
 164 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 165 {
 166         return 0;
 167 }
 168
 169 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 170 {
 171         return VM_FAULT_SIGBUS;
 172 }
 173
 174
 175 /**
 176  * kvm_arch_destroy_vm - destroy the VM data structure
 177  * @kvm:        pointer to the KVM struct
 178  */
 179 void kvm_arch_destroy_vm(struct kvm *kvm)
 180 {
 181         int i;
 182
 183         kvm_vgic_destroy(kvm);
 184
 185         free_percpu(kvm->arch.last_vcpu_ran);
 186         kvm->arch.last_vcpu_ran = NULL;
 187
 188         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 189                 if (kvm->vcpus[i]) {
 190                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 191                         kvm->vcpus[i] = NULL;
 192                 }
 193         }
 194         atomic_set(&kvm->online_vcpus, 0);
 195 }
 196
 197 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 198 {
 199         int r;
 200         switch (ext) {
 201         case KVM_CAP_IRQCHIP:
 202                 r = vgic_present;
 203                 break;
 204         case KVM_CAP_IOEVENTFD:
 205         case KVM_CAP_DEVICE_CTRL:
 206         case KVM_CAP_USER_MEMORY:
 207         case KVM_CAP_SYNC_MMU:
 208         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 209         case KVM_CAP_ONE_REG:
 210         case KVM_CAP_ARM_PSCI:
 211         case KVM_CAP_ARM_PSCI_0_2:
 212         case KVM_CAP_READONLY_MEM:
 213         case KVM_CAP_MP_STATE:
 214         case KVM_CAP_IMMEDIATE_EXIT:
 215         case KVM_CAP_VCPU_EVENTS:
 216                 r = 1;
 217                 break;
 218         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 219                 r = 1;
 220                 break;
 221         case KVM_CAP_NR_VCPUS:
 222                 r = num_online_cpus();
 223                 break;
 224         case KVM_CAP_MAX_VCPUS:
 225                 r = KVM_MAX_VCPUS;
 226                 break;
 227         case KVM_CAP_NR_MEMSLOTS:
 228                 r = KVM_USER_MEM_SLOTS;
 229                 break;
 230         case KVM_CAP_MSI_DEVID:
 231                 if (!kvm)
 232                         r = -EINVAL;
 233                 else
 234                         r = kvm->arch.vgic.msis_require_devid;
 235                 break;
 236         case KVM_CAP_ARM_USER_IRQ:
 237                 /*
 238                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 239                  * (bump this number if adding more devices)
 240                  */
 241                 r = 1;
 242                 break;
 243         default:
 244                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 245                 break;
 246         }
 247         return r;
 248 }
 249
 250 long kvm_arch_dev_ioctl(struct file *filp,
 251                         unsigned int ioctl, unsigned long arg)
 252 {
 253         return -EINVAL;
 254 }
 255
 256 struct kvm *kvm_arch_alloc_vm(void)
 257 {
 258         if (!has_vhe())
 259                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 260
 261         return vzalloc(sizeof(struct kvm));
 262 }
 263
 264 void kvm_arch_free_vm(struct kvm *kvm)
 265 {
 266         if (!has_vhe())
 267                 kfree(kvm);
 268         else
 269                 vfree(kvm);
 270 }
 271
 272 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 273 {
 274         int err;
 275         struct kvm_vcpu *vcpu;
 276
 277         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 278                 err = -EBUSY;
 279                 goto out;
 280         }
 281
 282         if (id >= kvm->arch.max_vcpus) {
 283                 err = -EINVAL;
 284                 goto out;
 285         }
 286
 287         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 288         if (!vcpu) {
 289                 err = -ENOMEM;
 290                 goto out;
 291         }
 292
 293         err = kvm_vcpu_init(vcpu, kvm, id);
 294         if (err)
 295                 goto free_vcpu;
 296
 297         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 298         if (err)
 299                 goto vcpu_uninit;
 300
 301         return vcpu;
 302 vcpu_uninit:
 303         kvm_vcpu_uninit(vcpu);
 304 free_vcpu:
 305         kmem_cache_free(kvm_vcpu_cache, vcpu);
 306 out:
 307         return ERR_PTR(err);
 308 }
 309
 310 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 311 {
 312 }
 313
 314 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 315 {
 316         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 317                 static_branch_dec(&userspace_irqchip_in_use);
 318
 319         kvm_mmu_free_memory_caches(vcpu);
 320         kvm_timer_vcpu_terminate(vcpu);
 321         kvm_pmu_vcpu_destroy(vcpu);
 322         kvm_vcpu_uninit(vcpu);
 323         kmem_cache_free(kvm_vcpu_cache, vcpu);
 324 }
 325
 326 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 327 {
 328         kvm_arch_vcpu_free(vcpu);
 329 }
 330
 331 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 332 {
 333         return kvm_timer_is_pending(vcpu);
 334 }
 335
 336 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 337 {
 338         kvm_vgic_v4_enable_doorbell(vcpu);
 339 }
 340
 341 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 342 {
 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         if (single_task_running())
 385                 vcpu_clear_wfe_traps(vcpu);
 386         else
 387                 vcpu_set_wfe_traps(vcpu);
 388 }
 389
 390 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 391 {
 392         kvm_arch_vcpu_put_fp(vcpu);
 393         kvm_vcpu_put_sysregs(vcpu);
 394         kvm_timer_vcpu_put(vcpu);
 395         kvm_vgic_put(vcpu);
 396
 397         vcpu->cpu = -1;
 398
 399         kvm_arm_set_running_vcpu(NULL);
 400 }
 401
 402 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 403 {
 404         vcpu->arch.power_off = true;
 405         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 406         kvm_vcpu_kick(vcpu);
 407 }
 408
 409 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 410                                     struct kvm_mp_state *mp_state)
 411 {
 412         if (vcpu->arch.power_off)
 413                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 414         else
 415                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 416
 417         return 0;
 418 }
 419
 420 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 421                                     struct kvm_mp_state *mp_state)
 422 {
 423         int ret = 0;
 424
 425         switch (mp_state->mp_state) {
 426         case KVM_MP_STATE_RUNNABLE:
 427                 vcpu->arch.power_off = false;
 428                 break;
 429         case KVM_MP_STATE_STOPPED:
 430                 vcpu_power_off(vcpu);
 431                 break;
 432         default:
 433                 ret = -EINVAL;
 434         }
 435
 436         return ret;
 437 }
 438
 439 /**
 440  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 441  * @v:          The VCPU pointer
 442  *
 443  * If the guest CPU is not waiting for interrupts or an interrupt line is
 444  * asserted, the CPU is by definition runnable.
 445  */
 446 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 447 {
 448         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 449         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 450                 && !v->arch.power_off && !v->arch.pause);
 451 }
 452
 453 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 454 {
 455         return vcpu_mode_priv(vcpu);
 456 }
 457
 458 /* Just ensure a guest exit from a particular CPU */
 459 static void exit_vm_noop(void *info)
 460 {
 461 }
 462
 463 void force_vm_exit(const cpumask_t *mask)
 464 {
 465         preempt_disable();
 466         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 467         preempt_enable();
 468 }
 469
 470 /**
 471  * need_new_vmid_gen - check that the VMID is still valid
 472  * @vmid: The VMID to check
 473  *
 474  * return true if there is a new generation of VMIDs being used
 475  *
 476  * The hardware supports a limited set of values with the value zero reserved
 477  * for the host, so we check if an assigned value belongs to a previous
 478  * generation, which which requires us to assign a new value. If we're the
 479  * first to use a VMID for the new generation, we must flush necessary caches
 480  * and TLBs on all CPUs.
 481  */
 482 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 483 {
 484         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 485         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 486         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 487 }
 488
 489 /**
 490  * update_vmid - Update the vmid with a valid VMID for the current generation
 491  * @kvm: The guest that struct vmid belongs to
 492  * @vmid: The stage-2 VMID information struct
 493  */
 494 static void update_vmid(struct kvm_vmid *vmid)
 495 {
 496         if (!need_new_vmid_gen(vmid))
 497                 return;
 498
 499         spin_lock(&kvm_vmid_lock);
 500
 501         /*
 502          * We need to re-check the vmid_gen here to ensure that if another vcpu
 503          * already allocated a valid vmid for this vm, then this vcpu should
 504          * use the same vmid.
 505          */
 506         if (!need_new_vmid_gen(vmid)) {
 507                 spin_unlock(&kvm_vmid_lock);
 508                 return;
 509         }
 510
 511         /* First user of a new VMID generation? */
 512         if (unlikely(kvm_next_vmid == 0)) {
 513                 atomic64_inc(&kvm_vmid_gen);
 514                 kvm_next_vmid = 1;
 515
 516                 /*
 517                  * On SMP we know no other CPUs can use this CPU's or each
 518                  * other's VMID after force_vm_exit returns since the
 519                  * kvm_vmid_lock blocks them from reentry to the guest.
 520                  */
 521                 force_vm_exit(cpu_all_mask);
 522                 /*
 523                  * Now broadcast TLB + ICACHE invalidation over the inner
 524                  * shareable domain to make sure all data structures are
 525                  * clean.
 526                  */
 527                 kvm_call_hyp(__kvm_flush_vm_context);
 528         }
 529
 530         vmid->vmid = kvm_next_vmid;
 531         kvm_next_vmid++;
 532         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 533
 534         smp_wmb();
 535         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 536
 537         spin_unlock(&kvm_vmid_lock);
 538 }
 539
 540 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 541 {
 542         struct kvm *kvm = vcpu->kvm;
 543         int ret = 0;
 544
 545         if (likely(vcpu->arch.has_run_once))
 546                 return 0;
 547
 548         vcpu->arch.has_run_once = true;
 549
 550         if (likely(irqchip_in_kernel(kvm))) {
 551                 /*
 552                  * Map the VGIC hardware resources before running a vcpu the
 553                  * first time on this VM.
 554                  */
 555                 if (unlikely(!vgic_ready(kvm))) {
 556                         ret = kvm_vgic_map_resources(kvm);
 557                         if (ret)
 558                                 return ret;
 559                 }
 560         } else {
 561                 /*
 562                  * Tell the rest of the code that there are userspace irqchip
 563                  * VMs in the wild.
 564                  */
 565                 static_branch_inc(&userspace_irqchip_in_use);
 566         }
 567
 568         ret = kvm_timer_enable(vcpu);
 569         if (ret)
 570                 return ret;
 571
 572         ret = kvm_arm_pmu_v3_enable(vcpu);
 573
 574         return ret;
 575 }
 576
 577 bool kvm_arch_intc_initialized(struct kvm *kvm)
 578 {
 579         return vgic_initialized(kvm);
 580 }
 581
 582 void kvm_arm_halt_guest(struct kvm *kvm)
 583 {
 584         int i;
 585         struct kvm_vcpu *vcpu;
 586
 587         kvm_for_each_vcpu(i, vcpu, kvm)
 588                 vcpu->arch.pause = true;
 589         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 590 }
 591
 592 void kvm_arm_resume_guest(struct kvm *kvm)
 593 {
 594         int i;
 595         struct kvm_vcpu *vcpu;
 596
 597         kvm_for_each_vcpu(i, vcpu, kvm) {
 598                 vcpu->arch.pause = false;
 599                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 600         }
 601 }
 602
 603 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 604 {
 605         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 606
 607         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 608                                        (!vcpu->arch.pause)));
 609
 610         if (vcpu->arch.power_off || vcpu->arch.pause) {
 611                 /* Awaken to handle a signal, request we sleep again later. */
 612                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 613         }
 614
 615         /*
 616          * Make sure we will observe a potential reset request if we've
 617          * observed a change to the power state. Pairs with the smp_wmb() in
 618          * kvm_psci_vcpu_on().
 619          */
 620         smp_rmb();
 621 }
 622
 623 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 624 {
 625         return vcpu->arch.target >= 0;
 626 }
 627
 628 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 629 {
 630         if (kvm_request_pending(vcpu)) {
 631                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 632                         vcpu_req_sleep(vcpu);
 633
 634                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 635                         kvm_reset_vcpu(vcpu);
 636
 637                 /*
 638                  * Clear IRQ_PENDING requests that were made to guarantee
 639                  * that a VCPU sees new virtual interrupts.
 640                  */
 641                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 642         }
 643 }
 644
 645 /**
 646  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 647  * @vcpu:       The VCPU pointer
 648  * @run:        The kvm_run structure pointer used for userspace state exchange
 649  *
 650  * This function is called through the VCPU_RUN ioctl called from user space. It
 651  * will execute VM code in a loop until the time slice for the process is used
 652  * or some emulation is needed from user space in which case the function will
 653  * return with return value 0 and with the kvm_run structure filled in with the
 654  * required data for the requested emulation.
 655  */
 656 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 657 {
 658         int ret;
 659
 660         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 661                 return -ENOEXEC;
 662
 663         ret = kvm_vcpu_first_run_init(vcpu);
 664         if (ret)
 665                 return ret;
 666
 667         if (run->exit_reason == KVM_EXIT_MMIO) {
 668                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 669                 if (ret)
 670                         return ret;
 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_vmid(&vcpu->kvm->arch.vmid);
 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->arch.vmid) ||
 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_ret(__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, ret;
 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         ret = kvm_reset_vcpu(vcpu);
 973         if (ret) {
 974                 vcpu->arch.target = -1;
 975                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 976         }
 977
 978         return ret;
 979 }
 980
 981 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 982                                          struct kvm_vcpu_init *init)
 983 {
 984         int ret;
 985
 986         ret = kvm_vcpu_set_target(vcpu, init);
 987         if (ret)
 988                 return ret;
 989
 990         /*
 991          * Ensure a rebooted VM will fault in RAM pages and detect if the
 992          * guest MMU is turned off and flush the caches as needed.
 993          */
 994         if (vcpu->arch.has_run_once)
 995                 stage2_unmap_vm(vcpu->kvm);
 996
 997         vcpu_reset_hcr(vcpu);
 998
 999         /*
1000          * Handle the "start in power-off" case.
1001          */
1002         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1003                 vcpu_power_off(vcpu);
1004         else
1005                 vcpu->arch.power_off = false;
1006
1007         return 0;
1008 }
1009
1010 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1011                                  struct kvm_device_attr *attr)
1012 {
1013         int ret = -ENXIO;
1014
1015         switch (attr->group) {
1016         default:
1017                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1018                 break;
1019         }
1020
1021         return ret;
1022 }
1023
1024 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1025                                  struct kvm_device_attr *attr)
1026 {
1027         int ret = -ENXIO;
1028
1029         switch (attr->group) {
1030         default:
1031                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1032                 break;
1033         }
1034
1035         return ret;
1036 }
1037
1038 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1039                                  struct kvm_device_attr *attr)
1040 {
1041         int ret = -ENXIO;
1042
1043         switch (attr->group) {
1044         default:
1045                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1046                 break;
1047         }
1048
1049         return ret;
1050 }
1051
1052 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1053                                    struct kvm_vcpu_events *events)
1054 {
1055         memset(events, 0, sizeof(*events));
1056
1057         return __kvm_arm_vcpu_get_events(vcpu, events);
1058 }
1059
1060 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1061                                    struct kvm_vcpu_events *events)
1062 {
1063         int i;
1064
1065         /* check whether the reserved field is zero */
1066         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1067                 if (events->reserved[i])
1068                         return -EINVAL;
1069
1070         /* check whether the pad field is zero */
1071         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1072                 if (events->exception.pad[i])
1073                         return -EINVAL;
1074
1075         return __kvm_arm_vcpu_set_events(vcpu, events);
1076 }
1077
1078 long kvm_arch_vcpu_ioctl(struct file *filp,
1079                          unsigned int ioctl, unsigned long arg)
1080 {
1081         struct kvm_vcpu *vcpu = filp->private_data;
1082         void __user *argp = (void __user *)arg;
1083         struct kvm_device_attr attr;
1084         long r;
1085
1086         switch (ioctl) {
1087         case KVM_ARM_VCPU_INIT: {
1088                 struct kvm_vcpu_init init;
1089
1090                 r = -EFAULT;
1091                 if (copy_from_user(&init, argp, sizeof(init)))
1092                         break;
1093
1094                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1095                 break;
1096         }
1097         case KVM_SET_ONE_REG:
1098         case KVM_GET_ONE_REG: {
1099                 struct kvm_one_reg reg;
1100
1101                 r = -ENOEXEC;
1102                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1103                         break;
1104
1105                 r = -EFAULT;
1106                 if (copy_from_user(&reg, argp, sizeof(reg)))
1107                         break;
1108
1109                 if (ioctl == KVM_SET_ONE_REG)
1110                         r = kvm_arm_set_reg(vcpu, &reg);
1111                 else
1112                         r = kvm_arm_get_reg(vcpu, &reg);
1113                 break;
1114         }
1115         case KVM_GET_REG_LIST: {
1116                 struct kvm_reg_list __user *user_list = argp;
1117                 struct kvm_reg_list reg_list;
1118                 unsigned n;
1119
1120                 r = -ENOEXEC;
1121                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1122                         break;
1123
1124                 r = -EFAULT;
1125                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1126                         break;
1127                 n = reg_list.n;
1128                 reg_list.n = kvm_arm_num_regs(vcpu);
1129                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1130                         break;
1131                 r = -E2BIG;
1132                 if (n < reg_list.n)
1133                         break;
1134                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1135                 break;
1136         }
1137         case KVM_SET_DEVICE_ATTR: {
1138                 r = -EFAULT;
1139                 if (copy_from_user(&attr, argp, sizeof(attr)))
1140                         break;
1141                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1142                 break;
1143         }
1144         case KVM_GET_DEVICE_ATTR: {
1145                 r = -EFAULT;
1146                 if (copy_from_user(&attr, argp, sizeof(attr)))
1147                         break;
1148                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1149                 break;
1150         }
1151         case KVM_HAS_DEVICE_ATTR: {
1152                 r = -EFAULT;
1153                 if (copy_from_user(&attr, argp, sizeof(attr)))
1154                         break;
1155                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1156                 break;
1157         }
1158         case KVM_GET_VCPU_EVENTS: {
1159                 struct kvm_vcpu_events events;
1160
1161                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1162                         return -EINVAL;
1163
1164                 if (copy_to_user(argp, &events, sizeof(events)))
1165                         return -EFAULT;
1166
1167                 return 0;
1168         }
1169         case KVM_SET_VCPU_EVENTS: {
1170                 struct kvm_vcpu_events events;
1171
1172                 if (copy_from_user(&events, argp, sizeof(events)))
1173                         return -EFAULT;
1174
1175                 return kvm_arm_vcpu_set_events(vcpu, &events);
1176         }
1177         default:
1178                 r = -EINVAL;
1179         }
1180
1181         return r;
1182 }
1183
1184 /**
1185  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1186  * @kvm: kvm instance
1187  * @log: slot id and address to which we copy the log
1188  *
1189  * Steps 1-4 below provide general overview of dirty page logging. See
1190  * kvm_get_dirty_log_protect() function description for additional details.
1191  *
1192  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1193  * always flush the TLB (step 4) even if previous step failed  and the dirty
1194  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1195  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1196  * writes will be marked dirty for next log read.
1197  *
1198  *   1. Take a snapshot of the bit and clear it if needed.
1199  *   2. Write protect the corresponding page.
1200  *   3. Copy the snapshot to the userspace.
1201  *   4. Flush TLB's if needed.
1202  */
1203 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1204 {
1205         bool flush = false;
1206         int r;
1207
1208         mutex_lock(&kvm->slots_lock);
1209
1210         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1211
1212         if (flush)
1213                 kvm_flush_remote_tlbs(kvm);
1214
1215         mutex_unlock(&kvm->slots_lock);
1216         return r;
1217 }
1218
1219 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1220 {
1221         bool flush = false;
1222         int r;
1223
1224         mutex_lock(&kvm->slots_lock);
1225
1226         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1227
1228         if (flush)
1229                 kvm_flush_remote_tlbs(kvm);
1230
1231         mutex_unlock(&kvm->slots_lock);
1232         return r;
1233 }
1234
1235 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1236                                         struct kvm_arm_device_addr *dev_addr)
1237 {
1238         unsigned long dev_id, type;
1239
1240         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1241                 KVM_ARM_DEVICE_ID_SHIFT;
1242         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1243                 KVM_ARM_DEVICE_TYPE_SHIFT;
1244
1245         switch (dev_id) {
1246         case KVM_ARM_DEVICE_VGIC_V2:
1247                 if (!vgic_present)
1248                         return -ENXIO;
1249                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1250         default:
1251                 return -ENODEV;
1252         }
1253 }
1254
1255 long kvm_arch_vm_ioctl(struct file *filp,
1256                        unsigned int ioctl, unsigned long arg)
1257 {
1258         struct kvm *kvm = filp->private_data;
1259         void __user *argp = (void __user *)arg;
1260
1261         switch (ioctl) {
1262         case KVM_CREATE_IRQCHIP: {
1263                 int ret;
1264                 if (!vgic_present)
1265                         return -ENXIO;
1266                 mutex_lock(&kvm->lock);
1267                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1268                 mutex_unlock(&kvm->lock);
1269                 return ret;
1270         }
1271         case KVM_ARM_SET_DEVICE_ADDR: {
1272                 struct kvm_arm_device_addr dev_addr;
1273
1274                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1275                         return -EFAULT;
1276                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1277         }
1278         case KVM_ARM_PREFERRED_TARGET: {
1279                 int err;
1280                 struct kvm_vcpu_init init;
1281
1282                 err = kvm_vcpu_preferred_target(&init);
1283                 if (err)
1284                         return err;
1285
1286                 if (copy_to_user(argp, &init, sizeof(init)))
1287                         return -EFAULT;
1288
1289                 return 0;
1290         }
1291         default:
1292                 return -EINVAL;
1293         }
1294 }
1295
1296 static void cpu_init_hyp_mode(void *dummy)
1297 {
1298         phys_addr_t pgd_ptr;
1299         unsigned long hyp_stack_ptr;
1300         unsigned long stack_page;
1301         unsigned long vector_ptr;
1302
1303         /* Switch from the HYP stub to our own HYP init vector */
1304         __hyp_set_vectors(kvm_get_idmap_vector());
1305
1306         pgd_ptr = kvm_mmu_get_httbr();
1307         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1308         hyp_stack_ptr = stack_page + PAGE_SIZE;
1309         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1310
1311         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1312         __cpu_init_stage2();
1313 }
1314
1315 static void cpu_hyp_reset(void)
1316 {
1317         if (!is_kernel_in_hyp_mode())
1318                 __hyp_reset_vectors();
1319 }
1320
1321 static void cpu_hyp_reinit(void)
1322 {
1323         cpu_hyp_reset();
1324
1325         if (is_kernel_in_hyp_mode())
1326                 kvm_timer_init_vhe();
1327         else
1328                 cpu_init_hyp_mode(NULL);
1329
1330         kvm_arm_init_debug();
1331
1332         if (vgic_present)
1333                 kvm_vgic_init_cpu_hardware();
1334 }
1335
1336 static void _kvm_arch_hardware_enable(void *discard)
1337 {
1338         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1339                 cpu_hyp_reinit();
1340                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1341         }
1342 }
1343
1344 int kvm_arch_hardware_enable(void)
1345 {
1346         _kvm_arch_hardware_enable(NULL);
1347         return 0;
1348 }
1349
1350 static void _kvm_arch_hardware_disable(void *discard)
1351 {
1352         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1353                 cpu_hyp_reset();
1354                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1355         }
1356 }
1357
1358 void kvm_arch_hardware_disable(void)
1359 {
1360         _kvm_arch_hardware_disable(NULL);
1361 }
1362
1363 #ifdef CONFIG_CPU_PM
1364 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1365                                     unsigned long cmd,
1366                                     void *v)
1367 {
1368         /*
1369          * kvm_arm_hardware_enabled is left with its old value over
1370          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1371          * re-enable hyp.
1372          */
1373         switch (cmd) {
1374         case CPU_PM_ENTER:
1375                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1376                         /*
1377                          * don't update kvm_arm_hardware_enabled here
1378                          * so that the hardware will be re-enabled
1379                          * when we resume. See below.
1380                          */
1381                         cpu_hyp_reset();
1382
1383                 return NOTIFY_OK;
1384         case CPU_PM_ENTER_FAILED:
1385         case CPU_PM_EXIT:
1386                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1387                         /* The hardware was enabled before suspend. */
1388                         cpu_hyp_reinit();
1389
1390                 return NOTIFY_OK;
1391
1392         default:
1393                 return NOTIFY_DONE;
1394         }
1395 }
1396
1397 static struct notifier_block hyp_init_cpu_pm_nb = {
1398         .notifier_call = hyp_init_cpu_pm_notifier,
1399 };
1400
1401 static void __init hyp_cpu_pm_init(void)
1402 {
1403         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1404 }
1405 static void __init hyp_cpu_pm_exit(void)
1406 {
1407         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1408 }
1409 #else
1410 static inline void hyp_cpu_pm_init(void)
1411 {
1412 }
1413 static inline void hyp_cpu_pm_exit(void)
1414 {
1415 }
1416 #endif
1417
1418 static int init_common_resources(void)
1419 {
1420         kvm_set_ipa_limit();
1421
1422         return 0;
1423 }
1424
1425 static int init_subsystems(void)
1426 {
1427         int err = 0;
1428
1429         /*
1430          * Enable hardware so that subsystem initialisation can access EL2.
1431          */
1432         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1433
1434         /*
1435          * Register CPU lower-power notifier
1436          */
1437         hyp_cpu_pm_init();
1438
1439         /*
1440          * Init HYP view of VGIC
1441          */
1442         err = kvm_vgic_hyp_init();
1443         switch (err) {
1444         case 0:
1445                 vgic_present = true;
1446                 break;
1447         case -ENODEV:
1448         case -ENXIO:
1449                 vgic_present = false;
1450                 err = 0;
1451                 break;
1452         default:
1453                 goto out;
1454         }
1455
1456         /*
1457          * Init HYP architected timer support
1458          */
1459         err = kvm_timer_hyp_init(vgic_present);
1460         if (err)
1461                 goto out;
1462
1463         kvm_perf_init();
1464         kvm_coproc_table_init();
1465
1466 out:
1467         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1468
1469         return err;
1470 }
1471
1472 static void teardown_hyp_mode(void)
1473 {
1474         int cpu;
1475
1476         free_hyp_pgds();
1477         for_each_possible_cpu(cpu)
1478                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1479         hyp_cpu_pm_exit();
1480 }
1481
1482 /**
1483  * Inits Hyp-mode on all online CPUs
1484  */
1485 static int init_hyp_mode(void)
1486 {
1487         int cpu;
1488         int err = 0;
1489
1490         /*
1491          * Allocate Hyp PGD and setup Hyp identity mapping
1492          */
1493         err = kvm_mmu_init();
1494         if (err)
1495                 goto out_err;
1496
1497         /*
1498          * Allocate stack pages for Hypervisor-mode
1499          */
1500         for_each_possible_cpu(cpu) {
1501                 unsigned long stack_page;
1502
1503                 stack_page = __get_free_page(GFP_KERNEL);
1504                 if (!stack_page) {
1505                         err = -ENOMEM;
1506                         goto out_err;
1507                 }
1508
1509                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1510         }
1511
1512         /*
1513          * Map the Hyp-code called directly from the host
1514          */
1515         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1516                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1517         if (err) {
1518                 kvm_err("Cannot map world-switch code\n");
1519                 goto out_err;
1520         }
1521
1522         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1523                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1524         if (err) {
1525                 kvm_err("Cannot map rodata section\n");
1526                 goto out_err;
1527         }
1528
1529         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1530                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1531         if (err) {
1532                 kvm_err("Cannot map bss section\n");
1533                 goto out_err;
1534         }
1535
1536         err = kvm_map_vectors();
1537         if (err) {
1538                 kvm_err("Cannot map vectors\n");
1539                 goto out_err;
1540         }
1541
1542         /*
1543          * Map the Hyp stack pages
1544          */
1545         for_each_possible_cpu(cpu) {
1546                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1547                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1548                                           PAGE_HYP);
1549
1550                 if (err) {
1551                         kvm_err("Cannot map hyp stack\n");
1552                         goto out_err;
1553                 }
1554         }
1555
1556         for_each_possible_cpu(cpu) {
1557                 kvm_cpu_context_t *cpu_ctxt;
1558
1559                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1560                 kvm_init_host_cpu_context(cpu_ctxt, cpu);
1561                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1562
1563                 if (err) {
1564                         kvm_err("Cannot map host CPU state: %d\n", err);
1565                         goto out_err;
1566                 }
1567         }
1568
1569         err = hyp_map_aux_data();
1570         if (err)
1571                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1572
1573         return 0;
1574
1575 out_err:
1576         teardown_hyp_mode();
1577         kvm_err("error initializing Hyp mode: %d\n", err);
1578         return err;
1579 }
1580
1581 static void check_kvm_target_cpu(void *ret)
1582 {
1583         *(int *)ret = kvm_target_cpu();
1584 }
1585
1586 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1587 {
1588         struct kvm_vcpu *vcpu;
1589         int i;
1590
1591         mpidr &= MPIDR_HWID_BITMASK;
1592         kvm_for_each_vcpu(i, vcpu, kvm) {
1593                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1594                         return vcpu;
1595         }
1596         return NULL;
1597 }
1598
1599 bool kvm_arch_has_irq_bypass(void)
1600 {
1601         return true;
1602 }
1603
1604 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1605                                       struct irq_bypass_producer *prod)
1606 {
1607         struct kvm_kernel_irqfd *irqfd =
1608                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1609
1610         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1611                                           &irqfd->irq_entry);
1612 }
1613 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1614                                       struct irq_bypass_producer *prod)
1615 {
1616         struct kvm_kernel_irqfd *irqfd =
1617                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1618
1619         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1620                                      &irqfd->irq_entry);
1621 }
1622
1623 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1624 {
1625         struct kvm_kernel_irqfd *irqfd =
1626                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1627
1628         kvm_arm_halt_guest(irqfd->kvm);
1629 }
1630
1631 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1632 {
1633         struct kvm_kernel_irqfd *irqfd =
1634                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1635
1636         kvm_arm_resume_guest(irqfd->kvm);
1637 }
1638
1639 /**
1640  * Initialize Hyp-mode and memory mappings on all CPUs.
1641  */
1642 int kvm_arch_init(void *opaque)
1643 {
1644         int err;
1645         int ret, cpu;
1646         bool in_hyp_mode;
1647
1648         if (!is_hyp_mode_available()) {
1649                 kvm_info("HYP mode not available\n");
1650                 return -ENODEV;
1651         }
1652
1653         in_hyp_mode = is_kernel_in_hyp_mode();
1654
1655         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1656                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1657                 return -ENODEV;
1658         }
1659
1660         for_each_online_cpu(cpu) {
1661                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1662                 if (ret < 0) {
1663                         kvm_err("Error, CPU %d not supported!\n", cpu);
1664                         return -ENODEV;
1665                 }
1666         }
1667
1668         err = init_common_resources();
1669         if (err)
1670                 return err;
1671
1672         if (!in_hyp_mode) {
1673                 err = init_hyp_mode();
1674                 if (err)
1675                         goto out_err;
1676         }
1677
1678         err = init_subsystems();
1679         if (err)
1680                 goto out_hyp;
1681
1682         if (in_hyp_mode)
1683                 kvm_info("VHE mode initialized successfully\n");
1684         else
1685                 kvm_info("Hyp mode initialized successfully\n");
1686
1687         return 0;
1688
1689 out_hyp:
1690         if (!in_hyp_mode)
1691                 teardown_hyp_mode();
1692 out_err:
1693         return err;
1694 }
1695
1696 /* NOP: Compiling as a module not supported */
1697 void kvm_arch_exit(void)
1698 {
1699         kvm_perf_teardown();
1700 }
1701
1702 static int arm_init(void)
1703 {
1704         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1705         return rc;
1706 }
1707
1708 module_init(arm_init);