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_host_data_t, kvm_host_data);
  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 int kvm_arch_check_processor_compat(void)
 109 {
 110         return 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_MAX_VCPU_ID:
 228                 r = KVM_MAX_VCPU_ID;
 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         kvm_host_data_t *cpu_data;
 364
 365         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 366         cpu_data = this_cpu_ptr(&kvm_host_data);
 367
 368         /*
 369          * We might get preempted before the vCPU actually runs, but
 370          * over-invalidation doesn't affect correctness.
 371          */
 372         if (*last_ran != vcpu->vcpu_id) {
 373                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 374                 *last_ran = vcpu->vcpu_id;
 375         }
 376
 377         vcpu->cpu = cpu;
 378         vcpu->arch.host_cpu_context = &cpu_data->host_ctxt;
 379
 380         kvm_arm_set_running_vcpu(vcpu);
 381         kvm_vgic_load(vcpu);
 382         kvm_timer_vcpu_load(vcpu);
 383         kvm_vcpu_load_sysregs(vcpu);
 384         kvm_arch_vcpu_load_fp(vcpu);
 385         kvm_vcpu_pmu_restore_guest(vcpu);
 386
 387         if (single_task_running())
 388                 vcpu_clear_wfe_traps(vcpu);
 389         else
 390                 vcpu_set_wfe_traps(vcpu);
 391
 392         vcpu_ptrauth_setup_lazy(vcpu);
 393 }
 394
 395 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 396 {
 397         kvm_arch_vcpu_put_fp(vcpu);
 398         kvm_vcpu_put_sysregs(vcpu);
 399         kvm_timer_vcpu_put(vcpu);
 400         kvm_vgic_put(vcpu);
 401         kvm_vcpu_pmu_restore_host(vcpu);
 402
 403         vcpu->cpu = -1;
 404
 405         kvm_arm_set_running_vcpu(NULL);
 406 }
 407
 408 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 409 {
 410         vcpu->arch.power_off = true;
 411         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 412         kvm_vcpu_kick(vcpu);
 413 }
 414
 415 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 416                                     struct kvm_mp_state *mp_state)
 417 {
 418         if (vcpu->arch.power_off)
 419                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 420         else
 421                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 422
 423         return 0;
 424 }
 425
 426 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 427                                     struct kvm_mp_state *mp_state)
 428 {
 429         int ret = 0;
 430
 431         switch (mp_state->mp_state) {
 432         case KVM_MP_STATE_RUNNABLE:
 433                 vcpu->arch.power_off = false;
 434                 break;
 435         case KVM_MP_STATE_STOPPED:
 436                 vcpu_power_off(vcpu);
 437                 break;
 438         default:
 439                 ret = -EINVAL;
 440         }
 441
 442         return ret;
 443 }
 444
 445 /**
 446  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 447  * @v:          The VCPU pointer
 448  *
 449  * If the guest CPU is not waiting for interrupts or an interrupt line is
 450  * asserted, the CPU is by definition runnable.
 451  */
 452 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 453 {
 454         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 455         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 456                 && !v->arch.power_off && !v->arch.pause);
 457 }
 458
 459 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 460 {
 461         return vcpu_mode_priv(vcpu);
 462 }
 463
 464 /* Just ensure a guest exit from a particular CPU */
 465 static void exit_vm_noop(void *info)
 466 {
 467 }
 468
 469 void force_vm_exit(const cpumask_t *mask)
 470 {
 471         preempt_disable();
 472         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 473         preempt_enable();
 474 }
 475
 476 /**
 477  * need_new_vmid_gen - check that the VMID is still valid
 478  * @vmid: The VMID to check
 479  *
 480  * return true if there is a new generation of VMIDs being used
 481  *
 482  * The hardware supports a limited set of values with the value zero reserved
 483  * for the host, so we check if an assigned value belongs to a previous
 484  * generation, which which requires us to assign a new value. If we're the
 485  * first to use a VMID for the new generation, we must flush necessary caches
 486  * and TLBs on all CPUs.
 487  */
 488 static bool need_new_vmid_gen(struct kvm_vmid *vmid)
 489 {
 490         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 491         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 492         return unlikely(READ_ONCE(vmid->vmid_gen) != current_vmid_gen);
 493 }
 494
 495 /**
 496  * update_vmid - Update the vmid with a valid VMID for the current generation
 497  * @kvm: The guest that struct vmid belongs to
 498  * @vmid: The stage-2 VMID information struct
 499  */
 500 static void update_vmid(struct kvm_vmid *vmid)
 501 {
 502         if (!need_new_vmid_gen(vmid))
 503                 return;
 504
 505         spin_lock(&kvm_vmid_lock);
 506
 507         /*
 508          * We need to re-check the vmid_gen here to ensure that if another vcpu
 509          * already allocated a valid vmid for this vm, then this vcpu should
 510          * use the same vmid.
 511          */
 512         if (!need_new_vmid_gen(vmid)) {
 513                 spin_unlock(&kvm_vmid_lock);
 514                 return;
 515         }
 516
 517         /* First user of a new VMID generation? */
 518         if (unlikely(kvm_next_vmid == 0)) {
 519                 atomic64_inc(&kvm_vmid_gen);
 520                 kvm_next_vmid = 1;
 521
 522                 /*
 523                  * On SMP we know no other CPUs can use this CPU's or each
 524                  * other's VMID after force_vm_exit returns since the
 525                  * kvm_vmid_lock blocks them from reentry to the guest.
 526                  */
 527                 force_vm_exit(cpu_all_mask);
 528                 /*
 529                  * Now broadcast TLB + ICACHE invalidation over the inner
 530                  * shareable domain to make sure all data structures are
 531                  * clean.
 532                  */
 533                 kvm_call_hyp(__kvm_flush_vm_context);
 534         }
 535
 536         vmid->vmid = kvm_next_vmid;
 537         kvm_next_vmid++;
 538         kvm_next_vmid &= (1 << kvm_get_vmid_bits()) - 1;
 539
 540         smp_wmb();
 541         WRITE_ONCE(vmid->vmid_gen, atomic64_read(&kvm_vmid_gen));
 542
 543         spin_unlock(&kvm_vmid_lock);
 544 }
 545
 546 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 547 {
 548         struct kvm *kvm = vcpu->kvm;
 549         int ret = 0;
 550
 551         if (likely(vcpu->arch.has_run_once))
 552                 return 0;
 553
 554         if (!kvm_arm_vcpu_is_finalized(vcpu))
 555                 return -EPERM;
 556
 557         vcpu->arch.has_run_once = true;
 558
 559         if (likely(irqchip_in_kernel(kvm))) {
 560                 /*
 561                  * Map the VGIC hardware resources before running a vcpu the
 562                  * first time on this VM.
 563                  */
 564                 if (unlikely(!vgic_ready(kvm))) {
 565                         ret = kvm_vgic_map_resources(kvm);
 566                         if (ret)
 567                                 return ret;
 568                 }
 569         } else {
 570                 /*
 571                  * Tell the rest of the code that there are userspace irqchip
 572                  * VMs in the wild.
 573                  */
 574                 static_branch_inc(&userspace_irqchip_in_use);
 575         }
 576
 577         ret = kvm_timer_enable(vcpu);
 578         if (ret)
 579                 return ret;
 580
 581         ret = kvm_arm_pmu_v3_enable(vcpu);
 582
 583         return ret;
 584 }
 585
 586 bool kvm_arch_intc_initialized(struct kvm *kvm)
 587 {
 588         return vgic_initialized(kvm);
 589 }
 590
 591 void kvm_arm_halt_guest(struct kvm *kvm)
 592 {
 593         int i;
 594         struct kvm_vcpu *vcpu;
 595
 596         kvm_for_each_vcpu(i, vcpu, kvm)
 597                 vcpu->arch.pause = true;
 598         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 599 }
 600
 601 void kvm_arm_resume_guest(struct kvm *kvm)
 602 {
 603         int i;
 604         struct kvm_vcpu *vcpu;
 605
 606         kvm_for_each_vcpu(i, vcpu, kvm) {
 607                 vcpu->arch.pause = false;
 608                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 609         }
 610 }
 611
 612 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 613 {
 614         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 615
 616         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 617                                        (!vcpu->arch.pause)));
 618
 619         if (vcpu->arch.power_off || vcpu->arch.pause) {
 620                 /* Awaken to handle a signal, request we sleep again later. */
 621                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 622         }
 623
 624         /*
 625          * Make sure we will observe a potential reset request if we've
 626          * observed a change to the power state. Pairs with the smp_wmb() in
 627          * kvm_psci_vcpu_on().
 628          */
 629         smp_rmb();
 630 }
 631
 632 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 633 {
 634         return vcpu->arch.target >= 0;
 635 }
 636
 637 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 638 {
 639         if (kvm_request_pending(vcpu)) {
 640                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 641                         vcpu_req_sleep(vcpu);
 642
 643                 if (kvm_check_request(KVM_REQ_VCPU_RESET, vcpu))
 644                         kvm_reset_vcpu(vcpu);
 645
 646                 /*
 647                  * Clear IRQ_PENDING requests that were made to guarantee
 648                  * that a VCPU sees new virtual interrupts.
 649                  */
 650                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 651         }
 652 }
 653
 654 /**
 655  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 656  * @vcpu:       The VCPU pointer
 657  * @run:        The kvm_run structure pointer used for userspace state exchange
 658  *
 659  * This function is called through the VCPU_RUN ioctl called from user space. It
 660  * will execute VM code in a loop until the time slice for the process is used
 661  * or some emulation is needed from user space in which case the function will
 662  * return with return value 0 and with the kvm_run structure filled in with the
 663  * required data for the requested emulation.
 664  */
 665 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 666 {
 667         int ret;
 668
 669         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 670                 return -ENOEXEC;
 671
 672         ret = kvm_vcpu_first_run_init(vcpu);
 673         if (ret)
 674                 return ret;
 675
 676         if (run->exit_reason == KVM_EXIT_MMIO) {
 677                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 678                 if (ret)
 679                         return ret;
 680         }
 681
 682         if (run->immediate_exit)
 683                 return -EINTR;
 684
 685         vcpu_load(vcpu);
 686
 687         kvm_sigset_activate(vcpu);
 688
 689         ret = 1;
 690         run->exit_reason = KVM_EXIT_UNKNOWN;
 691         while (ret > 0) {
 692                 /*
 693                  * Check conditions before entering the guest
 694                  */
 695                 cond_resched();
 696
 697                 update_vmid(&vcpu->kvm->arch.vmid);
 698
 699                 check_vcpu_requests(vcpu);
 700
 701                 /*
 702                  * Preparing the interrupts to be injected also
 703                  * involves poking the GIC, which must be done in a
 704                  * non-preemptible context.
 705                  */
 706                 preempt_disable();
 707
 708                 kvm_pmu_flush_hwstate(vcpu);
 709
 710                 local_irq_disable();
 711
 712                 kvm_vgic_flush_hwstate(vcpu);
 713
 714                 /*
 715                  * Exit if we have a signal pending so that we can deliver the
 716                  * signal to user space.
 717                  */
 718                 if (signal_pending(current)) {
 719                         ret = -EINTR;
 720                         run->exit_reason = KVM_EXIT_INTR;
 721                 }
 722
 723                 /*
 724                  * If we're using a userspace irqchip, then check if we need
 725                  * to tell a userspace irqchip about timer or PMU level
 726                  * changes and if so, exit to userspace (the actual level
 727                  * state gets updated in kvm_timer_update_run and
 728                  * kvm_pmu_update_run below).
 729                  */
 730                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 731                         if (kvm_timer_should_notify_user(vcpu) ||
 732                             kvm_pmu_should_notify_user(vcpu)) {
 733                                 ret = -EINTR;
 734                                 run->exit_reason = KVM_EXIT_INTR;
 735                         }
 736                 }
 737
 738                 /*
 739                  * Ensure we set mode to IN_GUEST_MODE after we disable
 740                  * interrupts and before the final VCPU requests check.
 741                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 742                  * Documentation/virtual/kvm/vcpu-requests.rst
 743                  */
 744                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 745
 746                 if (ret <= 0 || need_new_vmid_gen(&vcpu->kvm->arch.vmid) ||
 747                     kvm_request_pending(vcpu)) {
 748                         vcpu->mode = OUTSIDE_GUEST_MODE;
 749                         isb(); /* Ensure work in x_flush_hwstate is committed */
 750                         kvm_pmu_sync_hwstate(vcpu);
 751                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 752                                 kvm_timer_sync_hwstate(vcpu);
 753                         kvm_vgic_sync_hwstate(vcpu);
 754                         local_irq_enable();
 755                         preempt_enable();
 756                         continue;
 757                 }
 758
 759                 kvm_arm_setup_debug(vcpu);
 760
 761                 /**************************************************************
 762                  * Enter the guest
 763                  */
 764                 trace_kvm_entry(*vcpu_pc(vcpu));
 765                 guest_enter_irqoff();
 766
 767                 if (has_vhe()) {
 768                         kvm_arm_vhe_guest_enter();
 769                         ret = kvm_vcpu_run_vhe(vcpu);
 770                         kvm_arm_vhe_guest_exit();
 771                 } else {
 772                         ret = kvm_call_hyp_ret(__kvm_vcpu_run_nvhe, vcpu);
 773                 }
 774
 775                 vcpu->mode = OUTSIDE_GUEST_MODE;
 776                 vcpu->stat.exits++;
 777                 /*
 778                  * Back from guest
 779                  *************************************************************/
 780
 781                 kvm_arm_clear_debug(vcpu);
 782
 783                 /*
 784                  * We must sync the PMU state before the vgic state so
 785                  * that the vgic can properly sample the updated state of the
 786                  * interrupt line.
 787                  */
 788                 kvm_pmu_sync_hwstate(vcpu);
 789
 790                 /*
 791                  * Sync the vgic state before syncing the timer state because
 792                  * the timer code needs to know if the virtual timer
 793                  * interrupts are active.
 794                  */
 795                 kvm_vgic_sync_hwstate(vcpu);
 796
 797                 /*
 798                  * Sync the timer hardware state before enabling interrupts as
 799                  * we don't want vtimer interrupts to race with syncing the
 800                  * timer virtual interrupt state.
 801                  */
 802                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 803                         kvm_timer_sync_hwstate(vcpu);
 804
 805                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 806
 807                 /*
 808                  * We may have taken a host interrupt in HYP mode (ie
 809                  * while executing the guest). This interrupt is still
 810                  * pending, as we haven't serviced it yet!
 811                  *
 812                  * We're now back in SVC mode, with interrupts
 813                  * disabled.  Enabling the interrupts now will have
 814                  * the effect of taking the interrupt again, in SVC
 815                  * mode this time.
 816                  */
 817                 local_irq_enable();
 818
 819                 /*
 820                  * We do local_irq_enable() before calling guest_exit() so
 821                  * that if a timer interrupt hits while running the guest we
 822                  * account that tick as being spent in the guest.  We enable
 823                  * preemption after calling guest_exit() so that if we get
 824                  * preempted we make sure ticks after that is not counted as
 825                  * guest time.
 826                  */
 827                 guest_exit();
 828                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 829
 830                 /* Exit types that need handling before we can be preempted */
 831                 handle_exit_early(vcpu, run, ret);
 832
 833                 preempt_enable();
 834
 835                 ret = handle_exit(vcpu, run, ret);
 836         }
 837
 838         /* Tell userspace about in-kernel device output levels */
 839         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 840                 kvm_timer_update_run(vcpu);
 841                 kvm_pmu_update_run(vcpu);
 842         }
 843
 844         kvm_sigset_deactivate(vcpu);
 845
 846         vcpu_put(vcpu);
 847         return ret;
 848 }
 849
 850 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 851 {
 852         int bit_index;
 853         bool set;
 854         unsigned long *hcr;
 855
 856         if (number == KVM_ARM_IRQ_CPU_IRQ)
 857                 bit_index = __ffs(HCR_VI);
 858         else /* KVM_ARM_IRQ_CPU_FIQ */
 859                 bit_index = __ffs(HCR_VF);
 860
 861         hcr = vcpu_hcr(vcpu);
 862         if (level)
 863                 set = test_and_set_bit(bit_index, hcr);
 864         else
 865                 set = test_and_clear_bit(bit_index, hcr);
 866
 867         /*
 868          * If we didn't change anything, no need to wake up or kick other CPUs
 869          */
 870         if (set == level)
 871                 return 0;
 872
 873         /*
 874          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 875          * trigger a world-switch round on the running physical CPU to set the
 876          * virtual IRQ/FIQ fields in the HCR appropriately.
 877          */
 878         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 879         kvm_vcpu_kick(vcpu);
 880
 881         return 0;
 882 }
 883
 884 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 885                           bool line_status)
 886 {
 887         u32 irq = irq_level->irq;
 888         unsigned int irq_type, vcpu_idx, irq_num;
 889         int nrcpus = atomic_read(&kvm->online_vcpus);
 890         struct kvm_vcpu *vcpu = NULL;
 891         bool level = irq_level->level;
 892
 893         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 894         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 895         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 896
 897         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 898
 899         switch (irq_type) {
 900         case KVM_ARM_IRQ_TYPE_CPU:
 901                 if (irqchip_in_kernel(kvm))
 902                         return -ENXIO;
 903
 904                 if (vcpu_idx >= nrcpus)
 905                         return -EINVAL;
 906
 907                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 908                 if (!vcpu)
 909                         return -EINVAL;
 910
 911                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 912                         return -EINVAL;
 913
 914                 return vcpu_interrupt_line(vcpu, irq_num, level);
 915         case KVM_ARM_IRQ_TYPE_PPI:
 916                 if (!irqchip_in_kernel(kvm))
 917                         return -ENXIO;
 918
 919                 if (vcpu_idx >= nrcpus)
 920                         return -EINVAL;
 921
 922                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 923                 if (!vcpu)
 924                         return -EINVAL;
 925
 926                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 927                         return -EINVAL;
 928
 929                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 930         case KVM_ARM_IRQ_TYPE_SPI:
 931                 if (!irqchip_in_kernel(kvm))
 932                         return -ENXIO;
 933
 934                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 935                         return -EINVAL;
 936
 937                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 938         }
 939
 940         return -EINVAL;
 941 }
 942
 943 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 944                                const struct kvm_vcpu_init *init)
 945 {
 946         unsigned int i, ret;
 947         int phys_target = kvm_target_cpu();
 948
 949         if (init->target != phys_target)
 950                 return -EINVAL;
 951
 952         /*
 953          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 954          * use the same target.
 955          */
 956         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 957                 return -EINVAL;
 958
 959         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 960         for (i = 0; i < sizeof(init->features) * 8; i++) {
 961                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 962
 963                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 964                         return -ENOENT;
 965
 966                 /*
 967                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 968                  * use the same feature set.
 969                  */
 970                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 971                     test_bit(i, vcpu->arch.features) != set)
 972                         return -EINVAL;
 973
 974                 if (set)
 975                         set_bit(i, vcpu->arch.features);
 976         }
 977
 978         vcpu->arch.target = phys_target;
 979
 980         /* Now we know what it is, we can reset it. */
 981         ret = kvm_reset_vcpu(vcpu);
 982         if (ret) {
 983                 vcpu->arch.target = -1;
 984                 bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 985         }
 986
 987         return ret;
 988 }
 989
 990 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 991                                          struct kvm_vcpu_init *init)
 992 {
 993         int ret;
 994
 995         ret = kvm_vcpu_set_target(vcpu, init);
 996         if (ret)
 997                 return ret;
 998
 999         /*
1000          * Ensure a rebooted VM will fault in RAM pages and detect if the
1001          * guest MMU is turned off and flush the caches as needed.
1002          */
1003         if (vcpu->arch.has_run_once)
1004                 stage2_unmap_vm(vcpu->kvm);
1005
1006         vcpu_reset_hcr(vcpu);
1007
1008         /*
1009          * Handle the "start in power-off" case.
1010          */
1011         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1012                 vcpu_power_off(vcpu);
1013         else
1014                 vcpu->arch.power_off = false;
1015
1016         return 0;
1017 }
1018
1019 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1020                                  struct kvm_device_attr *attr)
1021 {
1022         int ret = -ENXIO;
1023
1024         switch (attr->group) {
1025         default:
1026                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1027                 break;
1028         }
1029
1030         return ret;
1031 }
1032
1033 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1034                                  struct kvm_device_attr *attr)
1035 {
1036         int ret = -ENXIO;
1037
1038         switch (attr->group) {
1039         default:
1040                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1041                 break;
1042         }
1043
1044         return ret;
1045 }
1046
1047 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1048                                  struct kvm_device_attr *attr)
1049 {
1050         int ret = -ENXIO;
1051
1052         switch (attr->group) {
1053         default:
1054                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1055                 break;
1056         }
1057
1058         return ret;
1059 }
1060
1061 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1062                                    struct kvm_vcpu_events *events)
1063 {
1064         memset(events, 0, sizeof(*events));
1065
1066         return __kvm_arm_vcpu_get_events(vcpu, events);
1067 }
1068
1069 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1070                                    struct kvm_vcpu_events *events)
1071 {
1072         int i;
1073
1074         /* check whether the reserved field is zero */
1075         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1076                 if (events->reserved[i])
1077                         return -EINVAL;
1078
1079         /* check whether the pad field is zero */
1080         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1081                 if (events->exception.pad[i])
1082                         return -EINVAL;
1083
1084         return __kvm_arm_vcpu_set_events(vcpu, events);
1085 }
1086
1087 long kvm_arch_vcpu_ioctl(struct file *filp,
1088                          unsigned int ioctl, unsigned long arg)
1089 {
1090         struct kvm_vcpu *vcpu = filp->private_data;
1091         void __user *argp = (void __user *)arg;
1092         struct kvm_device_attr attr;
1093         long r;
1094
1095         switch (ioctl) {
1096         case KVM_ARM_VCPU_INIT: {
1097                 struct kvm_vcpu_init init;
1098
1099                 r = -EFAULT;
1100                 if (copy_from_user(&init, argp, sizeof(init)))
1101                         break;
1102
1103                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1104                 break;
1105         }
1106         case KVM_SET_ONE_REG:
1107         case KVM_GET_ONE_REG: {
1108                 struct kvm_one_reg reg;
1109
1110                 r = -ENOEXEC;
1111                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1112                         break;
1113
1114                 r = -EFAULT;
1115                 if (copy_from_user(&reg, argp, sizeof(reg)))
1116                         break;
1117
1118                 if (ioctl == KVM_SET_ONE_REG)
1119                         r = kvm_arm_set_reg(vcpu, &reg);
1120                 else
1121                         r = kvm_arm_get_reg(vcpu, &reg);
1122                 break;
1123         }
1124         case KVM_GET_REG_LIST: {
1125                 struct kvm_reg_list __user *user_list = argp;
1126                 struct kvm_reg_list reg_list;
1127                 unsigned n;
1128
1129                 r = -ENOEXEC;
1130                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1131                         break;
1132
1133                 r = -EPERM;
1134                 if (!kvm_arm_vcpu_is_finalized(vcpu))
1135                         break;
1136
1137                 r = -EFAULT;
1138                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1139                         break;
1140                 n = reg_list.n;
1141                 reg_list.n = kvm_arm_num_regs(vcpu);
1142                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1143                         break;
1144                 r = -E2BIG;
1145                 if (n < reg_list.n)
1146                         break;
1147                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1148                 break;
1149         }
1150         case KVM_SET_DEVICE_ATTR: {
1151                 r = -EFAULT;
1152                 if (copy_from_user(&attr, argp, sizeof(attr)))
1153                         break;
1154                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1155                 break;
1156         }
1157         case KVM_GET_DEVICE_ATTR: {
1158                 r = -EFAULT;
1159                 if (copy_from_user(&attr, argp, sizeof(attr)))
1160                         break;
1161                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1162                 break;
1163         }
1164         case KVM_HAS_DEVICE_ATTR: {
1165                 r = -EFAULT;
1166                 if (copy_from_user(&attr, argp, sizeof(attr)))
1167                         break;
1168                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1169                 break;
1170         }
1171         case KVM_GET_VCPU_EVENTS: {
1172                 struct kvm_vcpu_events events;
1173
1174                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1175                         return -EINVAL;
1176
1177                 if (copy_to_user(argp, &events, sizeof(events)))
1178                         return -EFAULT;
1179
1180                 return 0;
1181         }
1182         case KVM_SET_VCPU_EVENTS: {
1183                 struct kvm_vcpu_events events;
1184
1185                 if (copy_from_user(&events, argp, sizeof(events)))
1186                         return -EFAULT;
1187
1188                 return kvm_arm_vcpu_set_events(vcpu, &events);
1189         }
1190         case KVM_ARM_VCPU_FINALIZE: {
1191                 int what;
1192
1193                 if (!kvm_vcpu_initialized(vcpu))
1194                         return -ENOEXEC;
1195
1196                 if (get_user(what, (const int __user *)argp))
1197                         return -EFAULT;
1198
1199                 return kvm_arm_vcpu_finalize(vcpu, what);
1200         }
1201         default:
1202                 r = -EINVAL;
1203         }
1204
1205         return r;
1206 }
1207
1208 /**
1209  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1210  * @kvm: kvm instance
1211  * @log: slot id and address to which we copy the log
1212  *
1213  * Steps 1-4 below provide general overview of dirty page logging. See
1214  * kvm_get_dirty_log_protect() function description for additional details.
1215  *
1216  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1217  * always flush the TLB (step 4) even if previous step failed  and the dirty
1218  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1219  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1220  * writes will be marked dirty for next log read.
1221  *
1222  *   1. Take a snapshot of the bit and clear it if needed.
1223  *   2. Write protect the corresponding page.
1224  *   3. Copy the snapshot to the userspace.
1225  *   4. Flush TLB's if needed.
1226  */
1227 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1228 {
1229         bool flush = false;
1230         int r;
1231
1232         mutex_lock(&kvm->slots_lock);
1233
1234         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1235
1236         if (flush)
1237                 kvm_flush_remote_tlbs(kvm);
1238
1239         mutex_unlock(&kvm->slots_lock);
1240         return r;
1241 }
1242
1243 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1244 {
1245         bool flush = false;
1246         int r;
1247
1248         mutex_lock(&kvm->slots_lock);
1249
1250         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1251
1252         if (flush)
1253                 kvm_flush_remote_tlbs(kvm);
1254
1255         mutex_unlock(&kvm->slots_lock);
1256         return r;
1257 }
1258
1259 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1260                                         struct kvm_arm_device_addr *dev_addr)
1261 {
1262         unsigned long dev_id, type;
1263
1264         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1265                 KVM_ARM_DEVICE_ID_SHIFT;
1266         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1267                 KVM_ARM_DEVICE_TYPE_SHIFT;
1268
1269         switch (dev_id) {
1270         case KVM_ARM_DEVICE_VGIC_V2:
1271                 if (!vgic_present)
1272                         return -ENXIO;
1273                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1274         default:
1275                 return -ENODEV;
1276         }
1277 }
1278
1279 long kvm_arch_vm_ioctl(struct file *filp,
1280                        unsigned int ioctl, unsigned long arg)
1281 {
1282         struct kvm *kvm = filp->private_data;
1283         void __user *argp = (void __user *)arg;
1284
1285         switch (ioctl) {
1286         case KVM_CREATE_IRQCHIP: {
1287                 int ret;
1288                 if (!vgic_present)
1289                         return -ENXIO;
1290                 mutex_lock(&kvm->lock);
1291                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1292                 mutex_unlock(&kvm->lock);
1293                 return ret;
1294         }
1295         case KVM_ARM_SET_DEVICE_ADDR: {
1296                 struct kvm_arm_device_addr dev_addr;
1297
1298                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1299                         return -EFAULT;
1300                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1301         }
1302         case KVM_ARM_PREFERRED_TARGET: {
1303                 int err;
1304                 struct kvm_vcpu_init init;
1305
1306                 err = kvm_vcpu_preferred_target(&init);
1307                 if (err)
1308                         return err;
1309
1310                 if (copy_to_user(argp, &init, sizeof(init)))
1311                         return -EFAULT;
1312
1313                 return 0;
1314         }
1315         default:
1316                 return -EINVAL;
1317         }
1318 }
1319
1320 static void cpu_init_hyp_mode(void *dummy)
1321 {
1322         phys_addr_t pgd_ptr;
1323         unsigned long hyp_stack_ptr;
1324         unsigned long stack_page;
1325         unsigned long vector_ptr;
1326
1327         /* Switch from the HYP stub to our own HYP init vector */
1328         __hyp_set_vectors(kvm_get_idmap_vector());
1329
1330         pgd_ptr = kvm_mmu_get_httbr();
1331         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1332         hyp_stack_ptr = stack_page + PAGE_SIZE;
1333         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1334
1335         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1336         __cpu_init_stage2();
1337 }
1338
1339 static void cpu_hyp_reset(void)
1340 {
1341         if (!is_kernel_in_hyp_mode())
1342                 __hyp_reset_vectors();
1343 }
1344
1345 static void cpu_hyp_reinit(void)
1346 {
1347         cpu_hyp_reset();
1348
1349         if (is_kernel_in_hyp_mode())
1350                 kvm_timer_init_vhe();
1351         else
1352                 cpu_init_hyp_mode(NULL);
1353
1354         kvm_arm_init_debug();
1355
1356         if (vgic_present)
1357                 kvm_vgic_init_cpu_hardware();
1358 }
1359
1360 static void _kvm_arch_hardware_enable(void *discard)
1361 {
1362         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1363                 cpu_hyp_reinit();
1364                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1365         }
1366 }
1367
1368 int kvm_arch_hardware_enable(void)
1369 {
1370         _kvm_arch_hardware_enable(NULL);
1371         return 0;
1372 }
1373
1374 static void _kvm_arch_hardware_disable(void *discard)
1375 {
1376         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1377                 cpu_hyp_reset();
1378                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1379         }
1380 }
1381
1382 void kvm_arch_hardware_disable(void)
1383 {
1384         _kvm_arch_hardware_disable(NULL);
1385 }
1386
1387 #ifdef CONFIG_CPU_PM
1388 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1389                                     unsigned long cmd,
1390                                     void *v)
1391 {
1392         /*
1393          * kvm_arm_hardware_enabled is left with its old value over
1394          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1395          * re-enable hyp.
1396          */
1397         switch (cmd) {
1398         case CPU_PM_ENTER:
1399                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1400                         /*
1401                          * don't update kvm_arm_hardware_enabled here
1402                          * so that the hardware will be re-enabled
1403                          * when we resume. See below.
1404                          */
1405                         cpu_hyp_reset();
1406
1407                 return NOTIFY_OK;
1408         case CPU_PM_ENTER_FAILED:
1409         case CPU_PM_EXIT:
1410                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1411                         /* The hardware was enabled before suspend. */
1412                         cpu_hyp_reinit();
1413
1414                 return NOTIFY_OK;
1415
1416         default:
1417                 return NOTIFY_DONE;
1418         }
1419 }
1420
1421 static struct notifier_block hyp_init_cpu_pm_nb = {
1422         .notifier_call = hyp_init_cpu_pm_notifier,
1423 };
1424
1425 static void __init hyp_cpu_pm_init(void)
1426 {
1427         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1428 }
1429 static void __init hyp_cpu_pm_exit(void)
1430 {
1431         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1432 }
1433 #else
1434 static inline void hyp_cpu_pm_init(void)
1435 {
1436 }
1437 static inline void hyp_cpu_pm_exit(void)
1438 {
1439 }
1440 #endif
1441
1442 static int init_common_resources(void)
1443 {
1444         kvm_set_ipa_limit();
1445
1446         return 0;
1447 }
1448
1449 static int init_subsystems(void)
1450 {
1451         int err = 0;
1452
1453         /*
1454          * Enable hardware so that subsystem initialisation can access EL2.
1455          */
1456         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1457
1458         /*
1459          * Register CPU lower-power notifier
1460          */
1461         hyp_cpu_pm_init();
1462
1463         /*
1464          * Init HYP view of VGIC
1465          */
1466         err = kvm_vgic_hyp_init();
1467         switch (err) {
1468         case 0:
1469                 vgic_present = true;
1470                 break;
1471         case -ENODEV:
1472         case -ENXIO:
1473                 vgic_present = false;
1474                 err = 0;
1475                 break;
1476         default:
1477                 goto out;
1478         }
1479
1480         /*
1481          * Init HYP architected timer support
1482          */
1483         err = kvm_timer_hyp_init(vgic_present);
1484         if (err)
1485                 goto out;
1486
1487         kvm_perf_init();
1488         kvm_coproc_table_init();
1489
1490 out:
1491         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1492
1493         return err;
1494 }
1495
1496 static void teardown_hyp_mode(void)
1497 {
1498         int cpu;
1499
1500         free_hyp_pgds();
1501         for_each_possible_cpu(cpu)
1502                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1503         hyp_cpu_pm_exit();
1504 }
1505
1506 /**
1507  * Inits Hyp-mode on all online CPUs
1508  */
1509 static int init_hyp_mode(void)
1510 {
1511         int cpu;
1512         int err = 0;
1513
1514         /*
1515          * Allocate Hyp PGD and setup Hyp identity mapping
1516          */
1517         err = kvm_mmu_init();
1518         if (err)
1519                 goto out_err;
1520
1521         /*
1522          * Allocate stack pages for Hypervisor-mode
1523          */
1524         for_each_possible_cpu(cpu) {
1525                 unsigned long stack_page;
1526
1527                 stack_page = __get_free_page(GFP_KERNEL);
1528                 if (!stack_page) {
1529                         err = -ENOMEM;
1530                         goto out_err;
1531                 }
1532
1533                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1534         }
1535
1536         /*
1537          * Map the Hyp-code called directly from the host
1538          */
1539         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1540                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1541         if (err) {
1542                 kvm_err("Cannot map world-switch code\n");
1543                 goto out_err;
1544         }
1545
1546         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1547                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1548         if (err) {
1549                 kvm_err("Cannot map rodata section\n");
1550                 goto out_err;
1551         }
1552
1553         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1554                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1555         if (err) {
1556                 kvm_err("Cannot map bss section\n");
1557                 goto out_err;
1558         }
1559
1560         err = kvm_map_vectors();
1561         if (err) {
1562                 kvm_err("Cannot map vectors\n");
1563                 goto out_err;
1564         }
1565
1566         /*
1567          * Map the Hyp stack pages
1568          */
1569         for_each_possible_cpu(cpu) {
1570                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1571                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1572                                           PAGE_HYP);
1573
1574                 if (err) {
1575                         kvm_err("Cannot map hyp stack\n");
1576                         goto out_err;
1577                 }
1578         }
1579
1580         for_each_possible_cpu(cpu) {
1581                 kvm_host_data_t *cpu_data;
1582
1583                 cpu_data = per_cpu_ptr(&kvm_host_data, cpu);
1584                 kvm_init_host_cpu_context(&cpu_data->host_ctxt, cpu);
1585                 err = create_hyp_mappings(cpu_data, cpu_data + 1, PAGE_HYP);
1586
1587                 if (err) {
1588                         kvm_err("Cannot map host CPU state: %d\n", err);
1589                         goto out_err;
1590                 }
1591         }
1592
1593         err = hyp_map_aux_data();
1594         if (err)
1595                 kvm_err("Cannot map host auxiliary data: %d\n", err);
1596
1597         return 0;
1598
1599 out_err:
1600         teardown_hyp_mode();
1601         kvm_err("error initializing Hyp mode: %d\n", err);
1602         return err;
1603 }
1604
1605 static void check_kvm_target_cpu(void *ret)
1606 {
1607         *(int *)ret = kvm_target_cpu();
1608 }
1609
1610 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1611 {
1612         struct kvm_vcpu *vcpu;
1613         int i;
1614
1615         mpidr &= MPIDR_HWID_BITMASK;
1616         kvm_for_each_vcpu(i, vcpu, kvm) {
1617                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1618                         return vcpu;
1619         }
1620         return NULL;
1621 }
1622
1623 bool kvm_arch_has_irq_bypass(void)
1624 {
1625         return true;
1626 }
1627
1628 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1629                                       struct irq_bypass_producer *prod)
1630 {
1631         struct kvm_kernel_irqfd *irqfd =
1632                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1633
1634         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1635                                           &irqfd->irq_entry);
1636 }
1637 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1638                                       struct irq_bypass_producer *prod)
1639 {
1640         struct kvm_kernel_irqfd *irqfd =
1641                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1642
1643         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1644                                      &irqfd->irq_entry);
1645 }
1646
1647 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1648 {
1649         struct kvm_kernel_irqfd *irqfd =
1650                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1651
1652         kvm_arm_halt_guest(irqfd->kvm);
1653 }
1654
1655 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1656 {
1657         struct kvm_kernel_irqfd *irqfd =
1658                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1659
1660         kvm_arm_resume_guest(irqfd->kvm);
1661 }
1662
1663 /**
1664  * Initialize Hyp-mode and memory mappings on all CPUs.
1665  */
1666 int kvm_arch_init(void *opaque)
1667 {
1668         int err;
1669         int ret, cpu;
1670         bool in_hyp_mode;
1671
1672         if (!is_hyp_mode_available()) {
1673                 kvm_info("HYP mode not available\n");
1674                 return -ENODEV;
1675         }
1676
1677         in_hyp_mode = is_kernel_in_hyp_mode();
1678
1679         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1680                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1681                 return -ENODEV;
1682         }
1683
1684         for_each_online_cpu(cpu) {
1685                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1686                 if (ret < 0) {
1687                         kvm_err("Error, CPU %d not supported!\n", cpu);
1688                         return -ENODEV;
1689                 }
1690         }
1691
1692         err = init_common_resources();
1693         if (err)
1694                 return err;
1695
1696         err = kvm_arm_init_sve();
1697         if (err)
1698                 return err;
1699
1700         if (!in_hyp_mode) {
1701                 err = init_hyp_mode();
1702                 if (err)
1703                         goto out_err;
1704         }
1705
1706         err = init_subsystems();
1707         if (err)
1708                 goto out_hyp;
1709
1710         if (in_hyp_mode)
1711                 kvm_info("VHE mode initialized successfully\n");
1712         else
1713                 kvm_info("Hyp mode initialized successfully\n");
1714
1715         return 0;
1716
1717 out_hyp:
1718         if (!in_hyp_mode)
1719                 teardown_hyp_mode();
1720 out_err:
1721         return err;
1722 }
1723
1724 /* NOP: Compiling as a module not supported */
1725 void kvm_arch_exit(void)
1726 {
1727         kvm_perf_teardown();
1728 }
1729
1730 static int arm_init(void)
1731 {
1732         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1733         return rc;
1734 }
1735
1736 module_init(arm_init);