virt/kvm/arm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <[email protected]>
   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 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 617 {
 618         return vcpu->arch.target >= 0;
 619 }
 620
 621 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 622 {
 623         if (kvm_request_pending(vcpu)) {
 624                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 625                         vcpu_req_sleep(vcpu);
 626
 627                 /*
 628                  * Clear IRQ_PENDING requests that were made to guarantee
 629                  * that a VCPU sees new virtual interrupts.
 630                  */
 631                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 632         }
 633 }
 634
 635 /**
 636  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 637  * @vcpu:       The VCPU pointer
 638  * @run:        The kvm_run structure pointer used for userspace state exchange
 639  *
 640  * This function is called through the VCPU_RUN ioctl called from user space. It
 641  * will execute VM code in a loop until the time slice for the process is used
 642  * or some emulation is needed from user space in which case the function will
 643  * return with return value 0 and with the kvm_run structure filled in with the
 644  * required data for the requested emulation.
 645  */
 646 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 647 {
 648         int ret;
 649
 650         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 651                 return -ENOEXEC;
 652
 653         ret = kvm_vcpu_first_run_init(vcpu);
 654         if (ret)
 655                 return ret;
 656
 657         if (run->exit_reason == KVM_EXIT_MMIO) {
 658                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 659                 if (ret)
 660                         return ret;
 661         }
 662
 663         if (run->immediate_exit)
 664                 return -EINTR;
 665
 666         vcpu_load(vcpu);
 667
 668         kvm_sigset_activate(vcpu);
 669
 670         ret = 1;
 671         run->exit_reason = KVM_EXIT_UNKNOWN;
 672         while (ret > 0) {
 673                 /*
 674                  * Check conditions before entering the guest
 675                  */
 676                 cond_resched();
 677
 678                 update_vmid(&vcpu->kvm->arch.vmid);
 679
 680                 check_vcpu_requests(vcpu);
 681
 682                 /*
 683                  * Preparing the interrupts to be injected also
 684                  * involves poking the GIC, which must be done in a
 685                  * non-preemptible context.
 686                  */
 687                 preempt_disable();
 688
 689                 kvm_pmu_flush_hwstate(vcpu);
 690
 691                 local_irq_disable();
 692
 693                 kvm_vgic_flush_hwstate(vcpu);
 694
 695                 /*
 696                  * Exit if we have a signal pending so that we can deliver the
 697                  * signal to user space.
 698                  */
 699                 if (signal_pending(current)) {
 700                         ret = -EINTR;
 701                         run->exit_reason = KVM_EXIT_INTR;
 702                 }
 703
 704                 /*
 705                  * If we're using a userspace irqchip, then check if we need
 706                  * to tell a userspace irqchip about timer or PMU level
 707                  * changes and if so, exit to userspace (the actual level
 708                  * state gets updated in kvm_timer_update_run and
 709                  * kvm_pmu_update_run below).
 710                  */
 711                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 712                         if (kvm_timer_should_notify_user(vcpu) ||
 713                             kvm_pmu_should_notify_user(vcpu)) {
 714                                 ret = -EINTR;
 715                                 run->exit_reason = KVM_EXIT_INTR;
 716                         }
 717                 }
 718
 719                 /*
 720                  * Ensure we set mode to IN_GUEST_MODE after we disable
 721                  * interrupts and before the final VCPU requests check.
 722                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 723                  * Documentation/virtual/kvm/vcpu-requests.rst
 724                  */
 725                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 726
 727                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 728                     kvm_request_pending(vcpu)) {
 729                         vcpu->mode = OUTSIDE_GUEST_MODE;
 730                         isb(); /* Ensure work in x_flush_hwstate is committed */
 731                         kvm_pmu_sync_hwstate(vcpu);
 732                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 733                                 kvm_timer_sync_hwstate(vcpu);
 734                         kvm_vgic_sync_hwstate(vcpu);
 735                         local_irq_enable();
 736                         preempt_enable();
 737                         continue;
 738                 }
 739
 740                 kvm_arm_setup_debug(vcpu);
 741
 742                 /**************************************************************
 743                  * Enter the guest
 744                  */
 745                 trace_kvm_entry(*vcpu_pc(vcpu));
 746                 guest_enter_irqoff();
 747
 748                 if (has_vhe()) {
 749                         kvm_arm_vhe_guest_enter();
 750                         ret = kvm_vcpu_run_vhe(vcpu);
 751                         kvm_arm_vhe_guest_exit();
 752                 } else {
 753                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 754                 }
 755
 756                 vcpu->mode = OUTSIDE_GUEST_MODE;
 757                 vcpu->stat.exits++;
 758                 /*
 759                  * Back from guest
 760                  *************************************************************/
 761
 762                 kvm_arm_clear_debug(vcpu);
 763
 764                 /*
 765                  * We must sync the PMU state before the vgic state so
 766                  * that the vgic can properly sample the updated state of the
 767                  * interrupt line.
 768                  */
 769                 kvm_pmu_sync_hwstate(vcpu);
 770
 771                 /*
 772                  * Sync the vgic state before syncing the timer state because
 773                  * the timer code needs to know if the virtual timer
 774                  * interrupts are active.
 775                  */
 776                 kvm_vgic_sync_hwstate(vcpu);
 777
 778                 /*
 779                  * Sync the timer hardware state before enabling interrupts as
 780                  * we don't want vtimer interrupts to race with syncing the
 781                  * timer virtual interrupt state.
 782                  */
 783                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 784                         kvm_timer_sync_hwstate(vcpu);
 785
 786                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 787
 788                 /*
 789                  * We may have taken a host interrupt in HYP mode (ie
 790                  * while executing the guest). This interrupt is still
 791                  * pending, as we haven't serviced it yet!
 792                  *
 793                  * We're now back in SVC mode, with interrupts
 794                  * disabled.  Enabling the interrupts now will have
 795                  * the effect of taking the interrupt again, in SVC
 796                  * mode this time.
 797                  */
 798                 local_irq_enable();
 799
 800                 /*
 801                  * We do local_irq_enable() before calling guest_exit() so
 802                  * that if a timer interrupt hits while running the guest we
 803                  * account that tick as being spent in the guest.  We enable
 804                  * preemption after calling guest_exit() so that if we get
 805                  * preempted we make sure ticks after that is not counted as
 806                  * guest time.
 807                  */
 808                 guest_exit();
 809                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 810
 811                 /* Exit types that need handling before we can be preempted */
 812                 handle_exit_early(vcpu, run, ret);
 813
 814                 preempt_enable();
 815
 816                 ret = handle_exit(vcpu, run, ret);
 817         }
 818
 819         /* Tell userspace about in-kernel device output levels */
 820         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 821                 kvm_timer_update_run(vcpu);
 822                 kvm_pmu_update_run(vcpu);
 823         }
 824
 825         kvm_sigset_deactivate(vcpu);
 826
 827         vcpu_put(vcpu);
 828         return ret;
 829 }
 830
 831 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 832 {
 833         int bit_index;
 834         bool set;
 835         unsigned long *hcr;
 836
 837         if (number == KVM_ARM_IRQ_CPU_IRQ)
 838                 bit_index = __ffs(HCR_VI);
 839         else /* KVM_ARM_IRQ_CPU_FIQ */
 840                 bit_index = __ffs(HCR_VF);
 841
 842         hcr = vcpu_hcr(vcpu);
 843         if (level)
 844                 set = test_and_set_bit(bit_index, hcr);
 845         else
 846                 set = test_and_clear_bit(bit_index, hcr);
 847
 848         /*
 849          * If we didn't change anything, no need to wake up or kick other CPUs
 850          */
 851         if (set == level)
 852                 return 0;
 853
 854         /*
 855          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 856          * trigger a world-switch round on the running physical CPU to set the
 857          * virtual IRQ/FIQ fields in the HCR appropriately.
 858          */
 859         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 860         kvm_vcpu_kick(vcpu);
 861
 862         return 0;
 863 }
 864
 865 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 866                           bool line_status)
 867 {
 868         u32 irq = irq_level->irq;
 869         unsigned int irq_type, vcpu_idx, irq_num;
 870         int nrcpus = atomic_read(&kvm->online_vcpus);
 871         struct kvm_vcpu *vcpu = NULL;
 872         bool level = irq_level->level;
 873
 874         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 875         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 876         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 877
 878         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 879
 880         switch (irq_type) {
 881         case KVM_ARM_IRQ_TYPE_CPU:
 882                 if (irqchip_in_kernel(kvm))
 883                         return -ENXIO;
 884
 885                 if (vcpu_idx >= nrcpus)
 886                         return -EINVAL;
 887
 888                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 889                 if (!vcpu)
 890                         return -EINVAL;
 891
 892                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 893                         return -EINVAL;
 894
 895                 return vcpu_interrupt_line(vcpu, irq_num, level);
 896         case KVM_ARM_IRQ_TYPE_PPI:
 897                 if (!irqchip_in_kernel(kvm))
 898                         return -ENXIO;
 899
 900                 if (vcpu_idx >= nrcpus)
 901                         return -EINVAL;
 902
 903                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 904                 if (!vcpu)
 905                         return -EINVAL;
 906
 907                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 908                         return -EINVAL;
 909
 910                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 911         case KVM_ARM_IRQ_TYPE_SPI:
 912                 if (!irqchip_in_kernel(kvm))
 913                         return -ENXIO;
 914
 915                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 916                         return -EINVAL;
 917
 918                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 919         }
 920
 921         return -EINVAL;
 922 }
 923
 924 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 925                                const struct kvm_vcpu_init *init)
 926 {
 927         unsigned int i;
 928         int phys_target = kvm_target_cpu();
 929
 930         if (init->target != phys_target)
 931                 return -EINVAL;
 932
 933         /*
 934          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 935          * use the same target.
 936          */
 937         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 938                 return -EINVAL;
 939
 940         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 941         for (i = 0; i < sizeof(init->features) * 8; i++) {
 942                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 943
 944                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 945                         return -ENOENT;
 946
 947                 /*
 948                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 949                  * use the same feature set.
 950                  */
 951                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 952                     test_bit(i, vcpu->arch.features) != set)
 953                         return -EINVAL;
 954
 955                 if (set)
 956                         set_bit(i, vcpu->arch.features);
 957         }
 958
 959         vcpu->arch.target = phys_target;
 960
 961         /* Now we know what it is, we can reset it. */
 962         return kvm_reset_vcpu(vcpu);
 963 }
 964
 965
 966 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 967                                          struct kvm_vcpu_init *init)
 968 {
 969         int ret;
 970
 971         ret = kvm_vcpu_set_target(vcpu, init);
 972         if (ret)
 973                 return ret;
 974
 975         /*
 976          * Ensure a rebooted VM will fault in RAM pages and detect if the
 977          * guest MMU is turned off and flush the caches as needed.
 978          */
 979         if (vcpu->arch.has_run_once)
 980                 stage2_unmap_vm(vcpu->kvm);
 981
 982         vcpu_reset_hcr(vcpu);
 983
 984         /*
 985          * Handle the "start in power-off" case.
 986          */
 987         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 988                 vcpu_power_off(vcpu);
 989         else
 990                 vcpu->arch.power_off = false;
 991
 992         return 0;
 993 }
 994
 995 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 996                                  struct kvm_device_attr *attr)
 997 {
 998         int ret = -ENXIO;
 999
1000         switch (attr->group) {
1001         default:
1002                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1003                 break;
1004         }
1005
1006         return ret;
1007 }
1008
1009 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1010                                  struct kvm_device_attr *attr)
1011 {
1012         int ret = -ENXIO;
1013
1014         switch (attr->group) {
1015         default:
1016                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1017                 break;
1018         }
1019
1020         return ret;
1021 }
1022
1023 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1024                                  struct kvm_device_attr *attr)
1025 {
1026         int ret = -ENXIO;
1027
1028         switch (attr->group) {
1029         default:
1030                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1031                 break;
1032         }
1033
1034         return ret;
1035 }
1036
1037 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1038                                    struct kvm_vcpu_events *events)
1039 {
1040         memset(events, 0, sizeof(*events));
1041
1042         return __kvm_arm_vcpu_get_events(vcpu, events);
1043 }
1044
1045 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1046                                    struct kvm_vcpu_events *events)
1047 {
1048         int i;
1049
1050         /* check whether the reserved field is zero */
1051         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1052                 if (events->reserved[i])
1053                         return -EINVAL;
1054
1055         /* check whether the pad field is zero */
1056         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1057                 if (events->exception.pad[i])
1058                         return -EINVAL;
1059
1060         return __kvm_arm_vcpu_set_events(vcpu, events);
1061 }
1062
1063 long kvm_arch_vcpu_ioctl(struct file *filp,
1064                          unsigned int ioctl, unsigned long arg)
1065 {
1066         struct kvm_vcpu *vcpu = filp->private_data;
1067         void __user *argp = (void __user *)arg;
1068         struct kvm_device_attr attr;
1069         long r;
1070
1071         switch (ioctl) {
1072         case KVM_ARM_VCPU_INIT: {
1073                 struct kvm_vcpu_init init;
1074
1075                 r = -EFAULT;
1076                 if (copy_from_user(&init, argp, sizeof(init)))
1077                         break;
1078
1079                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1080                 break;
1081         }
1082         case KVM_SET_ONE_REG:
1083         case KVM_GET_ONE_REG: {
1084                 struct kvm_one_reg reg;
1085
1086                 r = -ENOEXEC;
1087                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1088                         break;
1089
1090                 r = -EFAULT;
1091                 if (copy_from_user(&reg, argp, sizeof(reg)))
1092                         break;
1093
1094                 if (ioctl == KVM_SET_ONE_REG)
1095                         r = kvm_arm_set_reg(vcpu, &reg);
1096                 else
1097                         r = kvm_arm_get_reg(vcpu, &reg);
1098                 break;
1099         }
1100         case KVM_GET_REG_LIST: {
1101                 struct kvm_reg_list __user *user_list = argp;
1102                 struct kvm_reg_list reg_list;
1103                 unsigned n;
1104
1105                 r = -ENOEXEC;
1106                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1107                         break;
1108
1109                 r = -EFAULT;
1110                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1111                         break;
1112                 n = reg_list.n;
1113                 reg_list.n = kvm_arm_num_regs(vcpu);
1114                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1115                         break;
1116                 r = -E2BIG;
1117                 if (n < reg_list.n)
1118                         break;
1119                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1120                 break;
1121         }
1122         case KVM_SET_DEVICE_ATTR: {
1123                 r = -EFAULT;
1124                 if (copy_from_user(&attr, argp, sizeof(attr)))
1125                         break;
1126                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1127                 break;
1128         }
1129         case KVM_GET_DEVICE_ATTR: {
1130                 r = -EFAULT;
1131                 if (copy_from_user(&attr, argp, sizeof(attr)))
1132                         break;
1133                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1134                 break;
1135         }
1136         case KVM_HAS_DEVICE_ATTR: {
1137                 r = -EFAULT;
1138                 if (copy_from_user(&attr, argp, sizeof(attr)))
1139                         break;
1140                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1141                 break;
1142         }
1143         case KVM_GET_VCPU_EVENTS: {
1144                 struct kvm_vcpu_events events;
1145
1146                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1147                         return -EINVAL;
1148
1149                 if (copy_to_user(argp, &events, sizeof(events)))
1150                         return -EFAULT;
1151
1152                 return 0;
1153         }
1154         case KVM_SET_VCPU_EVENTS: {
1155                 struct kvm_vcpu_events events;
1156
1157                 if (copy_from_user(&events, argp, sizeof(events)))
1158                         return -EFAULT;
1159
1160                 return kvm_arm_vcpu_set_events(vcpu, &events);
1161         }
1162         default:
1163                 r = -EINVAL;
1164         }
1165
1166         return r;
1167 }
1168
1169 /**
1170  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1171  * @kvm: kvm instance
1172  * @log: slot id and address to which we copy the log
1173  *
1174  * Steps 1-4 below provide general overview of dirty page logging. See
1175  * kvm_get_dirty_log_protect() function description for additional details.
1176  *
1177  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1178  * always flush the TLB (step 4) even if previous step failed  and the dirty
1179  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1180  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1181  * writes will be marked dirty for next log read.
1182  *
1183  *   1. Take a snapshot of the bit and clear it if needed.
1184  *   2. Write protect the corresponding page.
1185  *   3. Copy the snapshot to the userspace.
1186  *   4. Flush TLB's if needed.
1187  */
1188 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1189 {
1190         bool flush = false;
1191         int r;
1192
1193         mutex_lock(&kvm->slots_lock);
1194
1195         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1196
1197         if (flush)
1198                 kvm_flush_remote_tlbs(kvm);
1199
1200         mutex_unlock(&kvm->slots_lock);
1201         return r;
1202 }
1203
1204 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1205 {
1206         bool flush = false;
1207         int r;
1208
1209         mutex_lock(&kvm->slots_lock);
1210
1211         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1212
1213         if (flush)
1214                 kvm_flush_remote_tlbs(kvm);
1215
1216         mutex_unlock(&kvm->slots_lock);
1217         return r;
1218 }
1219
1220 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1221                                         struct kvm_arm_device_addr *dev_addr)
1222 {
1223         unsigned long dev_id, type;
1224
1225         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1226                 KVM_ARM_DEVICE_ID_SHIFT;
1227         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1228                 KVM_ARM_DEVICE_TYPE_SHIFT;
1229
1230         switch (dev_id) {
1231         case KVM_ARM_DEVICE_VGIC_V2:
1232                 if (!vgic_present)
1233                         return -ENXIO;
1234                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1235         default:
1236                 return -ENODEV;
1237         }
1238 }
1239
1240 long kvm_arch_vm_ioctl(struct file *filp,
1241                        unsigned int ioctl, unsigned long arg)
1242 {
1243         struct kvm *kvm = filp->private_data;
1244         void __user *argp = (void __user *)arg;
1245
1246         switch (ioctl) {
1247         case KVM_CREATE_IRQCHIP: {
1248                 int ret;
1249                 if (!vgic_present)
1250                         return -ENXIO;
1251                 mutex_lock(&kvm->lock);
1252                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1253                 mutex_unlock(&kvm->lock);
1254                 return ret;
1255         }
1256         case KVM_ARM_SET_DEVICE_ADDR: {
1257                 struct kvm_arm_device_addr dev_addr;
1258
1259                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1260                         return -EFAULT;
1261                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1262         }
1263         case KVM_ARM_PREFERRED_TARGET: {
1264                 int err;
1265                 struct kvm_vcpu_init init;
1266
1267                 err = kvm_vcpu_preferred_target(&init);
1268                 if (err)
1269                         return err;
1270
1271                 if (copy_to_user(argp, &init, sizeof(init)))
1272                         return -EFAULT;
1273
1274                 return 0;
1275         }
1276         default:
1277                 return -EINVAL;
1278         }
1279 }
1280
1281 static void cpu_init_hyp_mode(void *dummy)
1282 {
1283         phys_addr_t pgd_ptr;
1284         unsigned long hyp_stack_ptr;
1285         unsigned long stack_page;
1286         unsigned long vector_ptr;
1287
1288         /* Switch from the HYP stub to our own HYP init vector */
1289         __hyp_set_vectors(kvm_get_idmap_vector());
1290
1291         pgd_ptr = kvm_mmu_get_httbr();
1292         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1293         hyp_stack_ptr = stack_page + PAGE_SIZE;
1294         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1295
1296         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1297         __cpu_init_stage2();
1298 }
1299
1300 static void cpu_hyp_reset(void)
1301 {
1302         if (!is_kernel_in_hyp_mode())
1303                 __hyp_reset_vectors();
1304 }
1305
1306 static void cpu_hyp_reinit(void)
1307 {
1308         cpu_hyp_reset();
1309
1310         if (is_kernel_in_hyp_mode())
1311                 kvm_timer_init_vhe();
1312         else
1313                 cpu_init_hyp_mode(NULL);
1314
1315         kvm_arm_init_debug();
1316
1317         if (vgic_present)
1318                 kvm_vgic_init_cpu_hardware();
1319 }
1320
1321 static void _kvm_arch_hardware_enable(void *discard)
1322 {
1323         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1324                 cpu_hyp_reinit();
1325                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1326         }
1327 }
1328
1329 int kvm_arch_hardware_enable(void)
1330 {
1331         _kvm_arch_hardware_enable(NULL);
1332         return 0;
1333 }
1334
1335 static void _kvm_arch_hardware_disable(void *discard)
1336 {
1337         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1338                 cpu_hyp_reset();
1339                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1340         }
1341 }
1342
1343 void kvm_arch_hardware_disable(void)
1344 {
1345         _kvm_arch_hardware_disable(NULL);
1346 }
1347
1348 #ifdef CONFIG_CPU_PM
1349 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1350                                     unsigned long cmd,
1351                                     void *v)
1352 {
1353         /*
1354          * kvm_arm_hardware_enabled is left with its old value over
1355          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1356          * re-enable hyp.
1357          */
1358         switch (cmd) {
1359         case CPU_PM_ENTER:
1360                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1361                         /*
1362                          * don't update kvm_arm_hardware_enabled here
1363                          * so that the hardware will be re-enabled
1364                          * when we resume. See below.
1365                          */
1366                         cpu_hyp_reset();
1367
1368                 return NOTIFY_OK;
1369         case CPU_PM_ENTER_FAILED:
1370         case CPU_PM_EXIT:
1371                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1372                         /* The hardware was enabled before suspend. */
1373                         cpu_hyp_reinit();
1374
1375                 return NOTIFY_OK;
1376
1377         default:
1378                 return NOTIFY_DONE;
1379         }
1380 }
1381
1382 static struct notifier_block hyp_init_cpu_pm_nb = {
1383         .notifier_call = hyp_init_cpu_pm_notifier,
1384 };
1385
1386 static void __init hyp_cpu_pm_init(void)
1387 {
1388         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1389 }
1390 static void __init hyp_cpu_pm_exit(void)
1391 {
1392         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1393 }
1394 #else
1395 static inline void hyp_cpu_pm_init(void)
1396 {
1397 }
1398 static inline void hyp_cpu_pm_exit(void)
1399 {
1400 }
1401 #endif
1402
1403 static int init_common_resources(void)
1404 {
1405         kvm_set_ipa_limit();
1406
1407         return 0;
1408 }
1409
1410 static int init_subsystems(void)
1411 {
1412         int err = 0;
1413
1414         /*
1415          * Enable hardware so that subsystem initialisation can access EL2.
1416          */
1417         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1418
1419         /*
1420          * Register CPU lower-power notifier
1421          */
1422         hyp_cpu_pm_init();
1423
1424         /*
1425          * Init HYP view of VGIC
1426          */
1427         err = kvm_vgic_hyp_init();
1428         switch (err) {
1429         case 0:
1430                 vgic_present = true;
1431                 break;
1432         case -ENODEV:
1433         case -ENXIO:
1434                 vgic_present = false;
1435                 err = 0;
1436                 break;
1437         default:
1438                 goto out;
1439         }
1440
1441         /*
1442          * Init HYP architected timer support
1443          */
1444         err = kvm_timer_hyp_init(vgic_present);
1445         if (err)
1446                 goto out;
1447
1448         kvm_perf_init();
1449         kvm_coproc_table_init();
1450
1451 out:
1452         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1453
1454         return err;
1455 }
1456
1457 static void teardown_hyp_mode(void)
1458 {
1459         int cpu;
1460
1461         free_hyp_pgds();
1462         for_each_possible_cpu(cpu)
1463                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1464         hyp_cpu_pm_exit();
1465 }
1466
1467 /**
1468  * Inits Hyp-mode on all online CPUs
1469  */
1470 static int init_hyp_mode(void)
1471 {
1472         int cpu;
1473         int err = 0;
1474
1475         /*
1476          * Allocate Hyp PGD and setup Hyp identity mapping
1477          */
1478         err = kvm_mmu_init();
1479         if (err)
1480                 goto out_err;
1481
1482         /*
1483          * Allocate stack pages for Hypervisor-mode
1484          */
1485         for_each_possible_cpu(cpu) {
1486                 unsigned long stack_page;
1487
1488                 stack_page = __get_free_page(GFP_KERNEL);
1489                 if (!stack_page) {
1490                         err = -ENOMEM;
1491                         goto out_err;
1492                 }
1493
1494                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1495         }
1496
1497         /*
1498          * Map the Hyp-code called directly from the host
1499          */
1500         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1501                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1502         if (err) {
1503                 kvm_err("Cannot map world-switch code\n");
1504                 goto out_err;
1505         }
1506
1507         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1508                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1509         if (err) {
1510                 kvm_err("Cannot map rodata section\n");
1511                 goto out_err;
1512         }
1513
1514         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1515                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1516         if (err) {
1517                 kvm_err("Cannot map bss section\n");
1518                 goto out_err;
1519         }
1520
1521         err = kvm_map_vectors();
1522         if (err) {
1523                 kvm_err("Cannot map vectors\n");
1524                 goto out_err;
1525         }
1526
1527         /*
1528          * Map the Hyp stack pages
1529          */
1530         for_each_possible_cpu(cpu) {
1531                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1532                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1533                                           PAGE_HYP);
1534
1535                 if (err) {
1536                         kvm_err("Cannot map hyp stack\n");
1537                         goto out_err;
1538                 }
1539         }
1540
1541         for_each_possible_cpu(cpu) {
1542                 kvm_cpu_context_t *cpu_ctxt;
1543
1544                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1545                 kvm_init_host_cpu_context(cpu_ctxt, cpu);
1546                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1547
1548                 if (err) {
1549                         kvm_err("Cannot map host CPU state: %d\n", err);
1550                         goto out_err;
1551                 }
1552         }
1553
1554         err = hyp_map_aux_data();
1555         if (err)
1556                 kvm_err("Cannot map host auxilary data: %d\n", err);
1557
1558         return 0;
1559
1560 out_err:
1561         teardown_hyp_mode();
1562         kvm_err("error initializing Hyp mode: %d\n", err);
1563         return err;
1564 }
1565
1566 static void check_kvm_target_cpu(void *ret)
1567 {
1568         *(int *)ret = kvm_target_cpu();
1569 }
1570
1571 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1572 {
1573         struct kvm_vcpu *vcpu;
1574         int i;
1575
1576         mpidr &= MPIDR_HWID_BITMASK;
1577         kvm_for_each_vcpu(i, vcpu, kvm) {
1578                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1579                         return vcpu;
1580         }
1581         return NULL;
1582 }
1583
1584 bool kvm_arch_has_irq_bypass(void)
1585 {
1586         return true;
1587 }
1588
1589 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1590                                       struct irq_bypass_producer *prod)
1591 {
1592         struct kvm_kernel_irqfd *irqfd =
1593                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1594
1595         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1596                                           &irqfd->irq_entry);
1597 }
1598 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1599                                       struct irq_bypass_producer *prod)
1600 {
1601         struct kvm_kernel_irqfd *irqfd =
1602                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1603
1604         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1605                                      &irqfd->irq_entry);
1606 }
1607
1608 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1609 {
1610         struct kvm_kernel_irqfd *irqfd =
1611                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1612
1613         kvm_arm_halt_guest(irqfd->kvm);
1614 }
1615
1616 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1617 {
1618         struct kvm_kernel_irqfd *irqfd =
1619                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1620
1621         kvm_arm_resume_guest(irqfd->kvm);
1622 }
1623
1624 /**
1625  * Initialize Hyp-mode and memory mappings on all CPUs.
1626  */
1627 int kvm_arch_init(void *opaque)
1628 {
1629         int err;
1630         int ret, cpu;
1631         bool in_hyp_mode;
1632
1633         if (!is_hyp_mode_available()) {
1634                 kvm_info("HYP mode not available\n");
1635                 return -ENODEV;
1636         }
1637
1638         in_hyp_mode = is_kernel_in_hyp_mode();
1639
1640         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1641                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1642                 return -ENODEV;
1643         }
1644
1645         for_each_online_cpu(cpu) {
1646                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1647                 if (ret < 0) {
1648                         kvm_err("Error, CPU %d not supported!\n", cpu);
1649                         return -ENODEV;
1650                 }
1651         }
1652
1653         err = init_common_resources();
1654         if (err)
1655                 return err;
1656
1657         if (!in_hyp_mode) {
1658                 err = init_hyp_mode();
1659                 if (err)
1660                         goto out_err;
1661         }
1662
1663         err = init_subsystems();
1664         if (err)
1665                 goto out_hyp;
1666
1667         if (in_hyp_mode)
1668                 kvm_info("VHE mode initialized successfully\n");
1669         else
1670                 kvm_info("Hyp mode initialized successfully\n");
1671
1672         return 0;
1673
1674 out_hyp:
1675         if (!in_hyp_mode)
1676                 teardown_hyp_mode();
1677 out_err:
1678         return err;
1679 }
1680
1681 /* NOP: Compiling as a module not supported */
1682 void kvm_arch_exit(void)
1683 {
1684         kvm_perf_teardown();
1685 }
1686
1687 static int arm_init(void)
1688 {
1689         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1690         return rc;
1691 }
1692
1693 module_init(arm_init);