virt/kvm/arm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/bug.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/list.h>
  25 #include <linux/module.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/fs.h>
  28 #include <linux/mman.h>
  29 #include <linux/sched.h>
  30 #include <linux/kvm.h>
  31 #include <linux/kvm_irqfd.h>
  32 #include <linux/irqbypass.h>
  33 #include <linux/sched/stat.h>
  34 #include <trace/events/kvm.h>
  35 #include <kvm/arm_pmu.h>
  36 #include <kvm/arm_psci.h>
  37
  38 #define CREATE_TRACE_POINTS
  39 #include "trace.h"
  40
  41 #include <linux/uaccess.h>
  42 #include <asm/ptrace.h>
  43 #include <asm/mman.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/cacheflush.h>
  46 #include <asm/cpufeature.h>
  47 #include <asm/virt.h>
  48 #include <asm/kvm_arm.h>
  49 #include <asm/kvm_asm.h>
  50 #include <asm/kvm_mmu.h>
  51 #include <asm/kvm_emulate.h>
  52 #include <asm/kvm_coproc.h>
  53 #include <asm/sections.h>
  54
  55 #ifdef REQUIRES_VIRT
  56 __asm__(".arch_extension        virt");
  57 #endif
  58
  59 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  60 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  61
  62 /* Per-CPU variable containing the currently running vcpu. */
  63 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  64
  65 /* The VMID used in the VTTBR */
  66 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  67 static u32 kvm_next_vmid;
  68 static unsigned int kvm_vmid_bits __read_mostly;
  69 static DEFINE_RWLOCK(kvm_vmid_lock);
  70
  71 static bool vgic_present;
  72
  73 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  74
  75 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  76 {
  77         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  78 }
  79
  80 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  81
  82 /**
  83  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  84  * Must be called from non-preemptible context
  85  */
  86 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  87 {
  88         return __this_cpu_read(kvm_arm_running_vcpu);
  89 }
  90
  91 /**
  92  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  93  */
  94 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  95 {
  96         return &kvm_arm_running_vcpu;
  97 }
  98
  99 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
 100 {
 101         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_check_processor_compat(void *rtn)
 110 {
 111         *(int *)rtn = 0;
 112 }
 113
 114
 115 /**
 116  * kvm_arch_init_vm - initializes a VM data structure
 117  * @kvm:        pointer to the KVM struct
 118  */
 119 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 120 {
 121         int ret, cpu;
 122
 123         if (type)
 124                 return -EINVAL;
 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_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                 r = 1;
 216                 break;
 217         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 218                 r = 1;
 219                 break;
 220         case KVM_CAP_NR_VCPUS:
 221                 r = num_online_cpus();
 222                 break;
 223         case KVM_CAP_MAX_VCPUS:
 224                 r = KVM_MAX_VCPUS;
 225                 break;
 226         case KVM_CAP_NR_MEMSLOTS:
 227                 r = KVM_USER_MEM_SLOTS;
 228                 break;
 229         case KVM_CAP_MSI_DEVID:
 230                 if (!kvm)
 231                         r = -EINVAL;
 232                 else
 233                         r = kvm->arch.vgic.msis_require_devid;
 234                 break;
 235         case KVM_CAP_ARM_USER_IRQ:
 236                 /*
 237                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 238                  * (bump this number if adding more devices)
 239                  */
 240                 r = 1;
 241                 break;
 242         default:
 243                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 244                 break;
 245         }
 246         return r;
 247 }
 248
 249 long kvm_arch_dev_ioctl(struct file *filp,
 250                         unsigned int ioctl, unsigned long arg)
 251 {
 252         return -EINVAL;
 253 }
 254
 255 struct kvm *kvm_arch_alloc_vm(void)
 256 {
 257         if (!has_vhe())
 258                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 259
 260         return vzalloc(sizeof(struct kvm));
 261 }
 262
 263 void kvm_arch_free_vm(struct kvm *kvm)
 264 {
 265         if (!has_vhe())
 266                 kfree(kvm);
 267         else
 268                 vfree(kvm);
 269 }
 270
 271 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 272 {
 273         int err;
 274         struct kvm_vcpu *vcpu;
 275
 276         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 277                 err = -EBUSY;
 278                 goto out;
 279         }
 280
 281         if (id >= kvm->arch.max_vcpus) {
 282                 err = -EINVAL;
 283                 goto out;
 284         }
 285
 286         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 287         if (!vcpu) {
 288                 err = -ENOMEM;
 289                 goto out;
 290         }
 291
 292         err = kvm_vcpu_init(vcpu, kvm, id);
 293         if (err)
 294                 goto free_vcpu;
 295
 296         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 297         if (err)
 298                 goto vcpu_uninit;
 299
 300         return vcpu;
 301 vcpu_uninit:
 302         kvm_vcpu_uninit(vcpu);
 303 free_vcpu:
 304         kmem_cache_free(kvm_vcpu_cache, vcpu);
 305 out:
 306         return ERR_PTR(err);
 307 }
 308
 309 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 310 {
 311 }
 312
 313 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 314 {
 315         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 316                 static_branch_dec(&userspace_irqchip_in_use);
 317
 318         kvm_mmu_free_memory_caches(vcpu);
 319         kvm_timer_vcpu_terminate(vcpu);
 320         kvm_pmu_vcpu_destroy(vcpu);
 321         kvm_vcpu_uninit(vcpu);
 322         kmem_cache_free(kvm_vcpu_cache, vcpu);
 323 }
 324
 325 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 326 {
 327         kvm_arch_vcpu_free(vcpu);
 328 }
 329
 330 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 331 {
 332         return kvm_timer_is_pending(vcpu);
 333 }
 334
 335 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 336 {
 337         kvm_timer_schedule(vcpu);
 338         kvm_vgic_v4_enable_doorbell(vcpu);
 339 }
 340
 341 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 342 {
 343         kvm_timer_unschedule(vcpu);
 344         kvm_vgic_v4_disable_doorbell(vcpu);
 345 }
 346
 347 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 348 {
 349         /* Force users to call KVM_ARM_VCPU_INIT */
 350         vcpu->arch.target = -1;
 351         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 352
 353         /* Set up the timer */
 354         kvm_timer_vcpu_init(vcpu);
 355
 356         kvm_arm_reset_debug_ptr(vcpu);
 357
 358         return kvm_vgic_vcpu_init(vcpu);
 359 }
 360
 361 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 362 {
 363         int *last_ran;
 364
 365         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 366
 367         /*
 368          * We might get preempted before the vCPU actually runs, but
 369          * over-invalidation doesn't affect correctness.
 370          */
 371         if (*last_ran != vcpu->vcpu_id) {
 372                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 373                 *last_ran = vcpu->vcpu_id;
 374         }
 375
 376         vcpu->cpu = cpu;
 377         vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
 378
 379         kvm_arm_set_running_vcpu(vcpu);
 380         kvm_vgic_load(vcpu);
 381         kvm_timer_vcpu_load(vcpu);
 382         kvm_vcpu_load_sysregs(vcpu);
 383         kvm_arch_vcpu_load_fp(vcpu);
 384
 385         if (single_task_running())
 386                 vcpu_clear_wfe_traps(vcpu);
 387         else
 388                 vcpu_set_wfe_traps(vcpu);
 389 }
 390
 391 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 392 {
 393         kvm_arch_vcpu_put_fp(vcpu);
 394         kvm_vcpu_put_sysregs(vcpu);
 395         kvm_timer_vcpu_put(vcpu);
 396         kvm_vgic_put(vcpu);
 397
 398         vcpu->cpu = -1;
 399
 400         kvm_arm_set_running_vcpu(NULL);
 401 }
 402
 403 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 404 {
 405         vcpu->arch.power_off = true;
 406         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 407         kvm_vcpu_kick(vcpu);
 408 }
 409
 410 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 411                                     struct kvm_mp_state *mp_state)
 412 {
 413         if (vcpu->arch.power_off)
 414                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 415         else
 416                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 417
 418         return 0;
 419 }
 420
 421 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 422                                     struct kvm_mp_state *mp_state)
 423 {
 424         int ret = 0;
 425
 426         switch (mp_state->mp_state) {
 427         case KVM_MP_STATE_RUNNABLE:
 428                 vcpu->arch.power_off = false;
 429                 break;
 430         case KVM_MP_STATE_STOPPED:
 431                 vcpu_power_off(vcpu);
 432                 break;
 433         default:
 434                 ret = -EINVAL;
 435         }
 436
 437         return ret;
 438 }
 439
 440 /**
 441  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 442  * @v:          The VCPU pointer
 443  *
 444  * If the guest CPU is not waiting for interrupts or an interrupt line is
 445  * asserted, the CPU is by definition runnable.
 446  */
 447 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 448 {
 449         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 450         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 451                 && !v->arch.power_off && !v->arch.pause);
 452 }
 453
 454 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 455 {
 456         return vcpu_mode_priv(vcpu);
 457 }
 458
 459 /* Just ensure a guest exit from a particular CPU */
 460 static void exit_vm_noop(void *info)
 461 {
 462 }
 463
 464 void force_vm_exit(const cpumask_t *mask)
 465 {
 466         preempt_disable();
 467         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 468         preempt_enable();
 469 }
 470
 471 /**
 472  * need_new_vmid_gen - check that the VMID is still valid
 473  * @kvm: The VM's VMID to check
 474  *
 475  * return true if there is a new generation of VMIDs being used
 476  *
 477  * The hardware supports only 256 values with the value zero reserved for the
 478  * host, so we check if an assigned value belongs to a previous generation,
 479  * which which requires us to assign a new value. If we're the first to use a
 480  * VMID for the new generation, we must flush necessary caches and TLBs on all
 481  * CPUs.
 482  */
 483 static bool need_new_vmid_gen(struct kvm *kvm)
 484 {
 485         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 486 }
 487
 488 /**
 489  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 490  * @kvm The guest that we are about to run
 491  *
 492  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 493  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 494  * caches and TLBs.
 495  */
 496 static void update_vttbr(struct kvm *kvm)
 497 {
 498         phys_addr_t pgd_phys;
 499         u64 vmid, cnp = kvm_cpu_has_cnp() ? VTTBR_CNP_BIT : 0;
 500         bool new_gen;
 501
 502         read_lock(&kvm_vmid_lock);
 503         new_gen = need_new_vmid_gen(kvm);
 504         read_unlock(&kvm_vmid_lock);
 505
 506         if (!new_gen)
 507                 return;
 508
 509         write_lock(&kvm_vmid_lock);
 510
 511         /*
 512          * We need to re-check the vmid_gen here to ensure that if another vcpu
 513          * already allocated a valid vmid for this vm, then this vcpu should
 514          * use the same vmid.
 515          */
 516         if (!need_new_vmid_gen(kvm)) {
 517                 write_unlock(&kvm_vmid_lock);
 518                 return;
 519         }
 520
 521         /* First user of a new VMID generation? */
 522         if (unlikely(kvm_next_vmid == 0)) {
 523                 atomic64_inc(&kvm_vmid_gen);
 524                 kvm_next_vmid = 1;
 525
 526                 /*
 527                  * On SMP we know no other CPUs can use this CPU's or each
 528                  * other's VMID after force_vm_exit returns since the
 529                  * kvm_vmid_lock blocks them from reentry to the guest.
 530                  */
 531                 force_vm_exit(cpu_all_mask);
 532                 /*
 533                  * Now broadcast TLB + ICACHE invalidation over the inner
 534                  * shareable domain to make sure all data structures are
 535                  * clean.
 536                  */
 537                 kvm_call_hyp(__kvm_flush_vm_context);
 538         }
 539
 540         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 541         kvm->arch.vmid = kvm_next_vmid;
 542         kvm_next_vmid++;
 543         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 544
 545         /* update vttbr to be used with the new vmid */
 546         pgd_phys = virt_to_phys(kvm->arch.pgd);
 547         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 548         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 549         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid | cnp;
 550
 551         write_unlock(&kvm_vmid_lock);
 552 }
 553
 554 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 555 {
 556         struct kvm *kvm = vcpu->kvm;
 557         int ret = 0;
 558
 559         if (likely(vcpu->arch.has_run_once))
 560                 return 0;
 561
 562         vcpu->arch.has_run_once = true;
 563
 564         if (likely(irqchip_in_kernel(kvm))) {
 565                 /*
 566                  * Map the VGIC hardware resources before running a vcpu the
 567                  * first time on this VM.
 568                  */
 569                 if (unlikely(!vgic_ready(kvm))) {
 570                         ret = kvm_vgic_map_resources(kvm);
 571                         if (ret)
 572                                 return ret;
 573                 }
 574         } else {
 575                 /*
 576                  * Tell the rest of the code that there are userspace irqchip
 577                  * VMs in the wild.
 578                  */
 579                 static_branch_inc(&userspace_irqchip_in_use);
 580         }
 581
 582         ret = kvm_timer_enable(vcpu);
 583         if (ret)
 584                 return ret;
 585
 586         ret = kvm_arm_pmu_v3_enable(vcpu);
 587
 588         return ret;
 589 }
 590
 591 bool kvm_arch_intc_initialized(struct kvm *kvm)
 592 {
 593         return vgic_initialized(kvm);
 594 }
 595
 596 void kvm_arm_halt_guest(struct kvm *kvm)
 597 {
 598         int i;
 599         struct kvm_vcpu *vcpu;
 600
 601         kvm_for_each_vcpu(i, vcpu, kvm)
 602                 vcpu->arch.pause = true;
 603         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 604 }
 605
 606 void kvm_arm_resume_guest(struct kvm *kvm)
 607 {
 608         int i;
 609         struct kvm_vcpu *vcpu;
 610
 611         kvm_for_each_vcpu(i, vcpu, kvm) {
 612                 vcpu->arch.pause = false;
 613                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 614         }
 615 }
 616
 617 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 618 {
 619         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 620
 621         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 622                                        (!vcpu->arch.pause)));
 623
 624         if (vcpu->arch.power_off || vcpu->arch.pause) {
 625                 /* Awaken to handle a signal, request we sleep again later. */
 626                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 627         }
 628 }
 629
 630 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 631 {
 632         return vcpu->arch.target >= 0;
 633 }
 634
 635 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 636 {
 637         if (kvm_request_pending(vcpu)) {
 638                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 639                         vcpu_req_sleep(vcpu);
 640
 641                 /*
 642                  * Clear IRQ_PENDING requests that were made to guarantee
 643                  * that a VCPU sees new virtual interrupts.
 644                  */
 645                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 646         }
 647 }
 648
 649 /**
 650  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 651  * @vcpu:       The VCPU pointer
 652  * @run:        The kvm_run structure pointer used for userspace state exchange
 653  *
 654  * This function is called through the VCPU_RUN ioctl called from user space. It
 655  * will execute VM code in a loop until the time slice for the process is used
 656  * or some emulation is needed from user space in which case the function will
 657  * return with return value 0 and with the kvm_run structure filled in with the
 658  * required data for the requested emulation.
 659  */
 660 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 661 {
 662         int ret;
 663
 664         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 665                 return -ENOEXEC;
 666
 667         ret = kvm_vcpu_first_run_init(vcpu);
 668         if (ret)
 669                 return ret;
 670
 671         if (run->exit_reason == KVM_EXIT_MMIO) {
 672                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 673                 if (ret)
 674                         return ret;
 675                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
 676                         return 0;
 677         }
 678
 679         if (run->immediate_exit)
 680                 return -EINTR;
 681
 682         vcpu_load(vcpu);
 683
 684         kvm_sigset_activate(vcpu);
 685
 686         ret = 1;
 687         run->exit_reason = KVM_EXIT_UNKNOWN;
 688         while (ret > 0) {
 689                 /*
 690                  * Check conditions before entering the guest
 691                  */
 692                 cond_resched();
 693
 694                 update_vttbr(vcpu->kvm);
 695
 696                 check_vcpu_requests(vcpu);
 697
 698                 /*
 699                  * Preparing the interrupts to be injected also
 700                  * involves poking the GIC, which must be done in a
 701                  * non-preemptible context.
 702                  */
 703                 preempt_disable();
 704
 705                 kvm_pmu_flush_hwstate(vcpu);
 706
 707                 local_irq_disable();
 708
 709                 kvm_vgic_flush_hwstate(vcpu);
 710
 711                 /*
 712                  * Exit if we have a signal pending so that we can deliver the
 713                  * signal to user space.
 714                  */
 715                 if (signal_pending(current)) {
 716                         ret = -EINTR;
 717                         run->exit_reason = KVM_EXIT_INTR;
 718                 }
 719
 720                 /*
 721                  * If we're using a userspace irqchip, then check if we need
 722                  * to tell a userspace irqchip about timer or PMU level
 723                  * changes and if so, exit to userspace (the actual level
 724                  * state gets updated in kvm_timer_update_run and
 725                  * kvm_pmu_update_run below).
 726                  */
 727                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 728                         if (kvm_timer_should_notify_user(vcpu) ||
 729                             kvm_pmu_should_notify_user(vcpu)) {
 730                                 ret = -EINTR;
 731                                 run->exit_reason = KVM_EXIT_INTR;
 732                         }
 733                 }
 734
 735                 /*
 736                  * Ensure we set mode to IN_GUEST_MODE after we disable
 737                  * interrupts and before the final VCPU requests check.
 738                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 739                  * Documentation/virtual/kvm/vcpu-requests.rst
 740                  */
 741                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 742
 743                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 744                     kvm_request_pending(vcpu)) {
 745                         vcpu->mode = OUTSIDE_GUEST_MODE;
 746                         isb(); /* Ensure work in x_flush_hwstate is committed */
 747                         kvm_pmu_sync_hwstate(vcpu);
 748                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 749                                 kvm_timer_sync_hwstate(vcpu);
 750                         kvm_vgic_sync_hwstate(vcpu);
 751                         local_irq_enable();
 752                         preempt_enable();
 753                         continue;
 754                 }
 755
 756                 kvm_arm_setup_debug(vcpu);
 757
 758                 /**************************************************************
 759                  * Enter the guest
 760                  */
 761                 trace_kvm_entry(*vcpu_pc(vcpu));
 762                 guest_enter_irqoff();
 763
 764                 if (has_vhe()) {
 765                         kvm_arm_vhe_guest_enter();
 766                         ret = kvm_vcpu_run_vhe(vcpu);
 767                         kvm_arm_vhe_guest_exit();
 768                 } else {
 769                         ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
 770                 }
 771
 772                 vcpu->mode = OUTSIDE_GUEST_MODE;
 773                 vcpu->stat.exits++;
 774                 /*
 775                  * Back from guest
 776                  *************************************************************/
 777
 778                 kvm_arm_clear_debug(vcpu);
 779
 780                 /*
 781                  * We must sync the PMU state before the vgic state so
 782                  * that the vgic can properly sample the updated state of the
 783                  * interrupt line.
 784                  */
 785                 kvm_pmu_sync_hwstate(vcpu);
 786
 787                 /*
 788                  * Sync the vgic state before syncing the timer state because
 789                  * the timer code needs to know if the virtual timer
 790                  * interrupts are active.
 791                  */
 792                 kvm_vgic_sync_hwstate(vcpu);
 793
 794                 /*
 795                  * Sync the timer hardware state before enabling interrupts as
 796                  * we don't want vtimer interrupts to race with syncing the
 797                  * timer virtual interrupt state.
 798                  */
 799                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 800                         kvm_timer_sync_hwstate(vcpu);
 801
 802                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 803
 804                 /*
 805                  * We may have taken a host interrupt in HYP mode (ie
 806                  * while executing the guest). This interrupt is still
 807                  * pending, as we haven't serviced it yet!
 808                  *
 809                  * We're now back in SVC mode, with interrupts
 810                  * disabled.  Enabling the interrupts now will have
 811                  * the effect of taking the interrupt again, in SVC
 812                  * mode this time.
 813                  */
 814                 local_irq_enable();
 815
 816                 /*
 817                  * We do local_irq_enable() before calling guest_exit() so
 818                  * that if a timer interrupt hits while running the guest we
 819                  * account that tick as being spent in the guest.  We enable
 820                  * preemption after calling guest_exit() so that if we get
 821                  * preempted we make sure ticks after that is not counted as
 822                  * guest time.
 823                  */
 824                 guest_exit();
 825                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 826
 827                 /* Exit types that need handling before we can be preempted */
 828                 handle_exit_early(vcpu, run, ret);
 829
 830                 preempt_enable();
 831
 832                 ret = handle_exit(vcpu, run, ret);
 833         }
 834
 835         /* Tell userspace about in-kernel device output levels */
 836         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 837                 kvm_timer_update_run(vcpu);
 838                 kvm_pmu_update_run(vcpu);
 839         }
 840
 841         kvm_sigset_deactivate(vcpu);
 842
 843         vcpu_put(vcpu);
 844         return ret;
 845 }
 846
 847 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 848 {
 849         int bit_index;
 850         bool set;
 851         unsigned long *hcr;
 852
 853         if (number == KVM_ARM_IRQ_CPU_IRQ)
 854                 bit_index = __ffs(HCR_VI);
 855         else /* KVM_ARM_IRQ_CPU_FIQ */
 856                 bit_index = __ffs(HCR_VF);
 857
 858         hcr = vcpu_hcr(vcpu);
 859         if (level)
 860                 set = test_and_set_bit(bit_index, hcr);
 861         else
 862                 set = test_and_clear_bit(bit_index, hcr);
 863
 864         /*
 865          * If we didn't change anything, no need to wake up or kick other CPUs
 866          */
 867         if (set == level)
 868                 return 0;
 869
 870         /*
 871          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 872          * trigger a world-switch round on the running physical CPU to set the
 873          * virtual IRQ/FIQ fields in the HCR appropriately.
 874          */
 875         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 876         kvm_vcpu_kick(vcpu);
 877
 878         return 0;
 879 }
 880
 881 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 882                           bool line_status)
 883 {
 884         u32 irq = irq_level->irq;
 885         unsigned int irq_type, vcpu_idx, irq_num;
 886         int nrcpus = atomic_read(&kvm->online_vcpus);
 887         struct kvm_vcpu *vcpu = NULL;
 888         bool level = irq_level->level;
 889
 890         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 891         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 892         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 893
 894         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 895
 896         switch (irq_type) {
 897         case KVM_ARM_IRQ_TYPE_CPU:
 898                 if (irqchip_in_kernel(kvm))
 899                         return -ENXIO;
 900
 901                 if (vcpu_idx >= nrcpus)
 902                         return -EINVAL;
 903
 904                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 905                 if (!vcpu)
 906                         return -EINVAL;
 907
 908                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 909                         return -EINVAL;
 910
 911                 return vcpu_interrupt_line(vcpu, irq_num, level);
 912         case KVM_ARM_IRQ_TYPE_PPI:
 913                 if (!irqchip_in_kernel(kvm))
 914                         return -ENXIO;
 915
 916                 if (vcpu_idx >= nrcpus)
 917                         return -EINVAL;
 918
 919                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 920                 if (!vcpu)
 921                         return -EINVAL;
 922
 923                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 924                         return -EINVAL;
 925
 926                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 927         case KVM_ARM_IRQ_TYPE_SPI:
 928                 if (!irqchip_in_kernel(kvm))
 929                         return -ENXIO;
 930
 931                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 932                         return -EINVAL;
 933
 934                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 935         }
 936
 937         return -EINVAL;
 938 }
 939
 940 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 941                                const struct kvm_vcpu_init *init)
 942 {
 943         unsigned int i;
 944         int phys_target = kvm_target_cpu();
 945
 946         if (init->target != phys_target)
 947                 return -EINVAL;
 948
 949         /*
 950          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 951          * use the same target.
 952          */
 953         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 954                 return -EINVAL;
 955
 956         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 957         for (i = 0; i < sizeof(init->features) * 8; i++) {
 958                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 959
 960                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 961                         return -ENOENT;
 962
 963                 /*
 964                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 965                  * use the same feature set.
 966                  */
 967                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 968                     test_bit(i, vcpu->arch.features) != set)
 969                         return -EINVAL;
 970
 971                 if (set)
 972                         set_bit(i, vcpu->arch.features);
 973         }
 974
 975         vcpu->arch.target = phys_target;
 976
 977         /* Now we know what it is, we can reset it. */
 978         return kvm_reset_vcpu(vcpu);
 979 }
 980
 981
 982 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 983                                          struct kvm_vcpu_init *init)
 984 {
 985         int ret;
 986
 987         ret = kvm_vcpu_set_target(vcpu, init);
 988         if (ret)
 989                 return ret;
 990
 991         /*
 992          * Ensure a rebooted VM will fault in RAM pages and detect if the
 993          * guest MMU is turned off and flush the caches as needed.
 994          */
 995         if (vcpu->arch.has_run_once)
 996                 stage2_unmap_vm(vcpu->kvm);
 997
 998         vcpu_reset_hcr(vcpu);
 999
1000         /*
1001          * Handle the "start in power-off" case.
1002          */
1003         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1004                 vcpu_power_off(vcpu);
1005         else
1006                 vcpu->arch.power_off = false;
1007
1008         return 0;
1009 }
1010
1011 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1012                                  struct kvm_device_attr *attr)
1013 {
1014         int ret = -ENXIO;
1015
1016         switch (attr->group) {
1017         default:
1018                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1019                 break;
1020         }
1021
1022         return ret;
1023 }
1024
1025 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1026                                  struct kvm_device_attr *attr)
1027 {
1028         int ret = -ENXIO;
1029
1030         switch (attr->group) {
1031         default:
1032                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1033                 break;
1034         }
1035
1036         return ret;
1037 }
1038
1039 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1040                                  struct kvm_device_attr *attr)
1041 {
1042         int ret = -ENXIO;
1043
1044         switch (attr->group) {
1045         default:
1046                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1047                 break;
1048         }
1049
1050         return ret;
1051 }
1052
1053 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1054                                    struct kvm_vcpu_events *events)
1055 {
1056         memset(events, 0, sizeof(*events));
1057
1058         return __kvm_arm_vcpu_get_events(vcpu, events);
1059 }
1060
1061 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1062                                    struct kvm_vcpu_events *events)
1063 {
1064         int i;
1065
1066         /* check whether the reserved field is zero */
1067         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1068                 if (events->reserved[i])
1069                         return -EINVAL;
1070
1071         /* check whether the pad field is zero */
1072         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1073                 if (events->exception.pad[i])
1074                         return -EINVAL;
1075
1076         return __kvm_arm_vcpu_set_events(vcpu, events);
1077 }
1078
1079 long kvm_arch_vcpu_ioctl(struct file *filp,
1080                          unsigned int ioctl, unsigned long arg)
1081 {
1082         struct kvm_vcpu *vcpu = filp->private_data;
1083         void __user *argp = (void __user *)arg;
1084         struct kvm_device_attr attr;
1085         long r;
1086
1087         switch (ioctl) {
1088         case KVM_ARM_VCPU_INIT: {
1089                 struct kvm_vcpu_init init;
1090
1091                 r = -EFAULT;
1092                 if (copy_from_user(&init, argp, sizeof(init)))
1093                         break;
1094
1095                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1096                 break;
1097         }
1098         case KVM_SET_ONE_REG:
1099         case KVM_GET_ONE_REG: {
1100                 struct kvm_one_reg reg;
1101
1102                 r = -ENOEXEC;
1103                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1104                         break;
1105
1106                 r = -EFAULT;
1107                 if (copy_from_user(&reg, argp, sizeof(reg)))
1108                         break;
1109
1110                 if (ioctl == KVM_SET_ONE_REG)
1111                         r = kvm_arm_set_reg(vcpu, &reg);
1112                 else
1113                         r = kvm_arm_get_reg(vcpu, &reg);
1114                 break;
1115         }
1116         case KVM_GET_REG_LIST: {
1117                 struct kvm_reg_list __user *user_list = argp;
1118                 struct kvm_reg_list reg_list;
1119                 unsigned n;
1120
1121                 r = -ENOEXEC;
1122                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1123                         break;
1124
1125                 r = -EFAULT;
1126                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1127                         break;
1128                 n = reg_list.n;
1129                 reg_list.n = kvm_arm_num_regs(vcpu);
1130                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1131                         break;
1132                 r = -E2BIG;
1133                 if (n < reg_list.n)
1134                         break;
1135                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1136                 break;
1137         }
1138         case KVM_SET_DEVICE_ATTR: {
1139                 r = -EFAULT;
1140                 if (copy_from_user(&attr, argp, sizeof(attr)))
1141                         break;
1142                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1143                 break;
1144         }
1145         case KVM_GET_DEVICE_ATTR: {
1146                 r = -EFAULT;
1147                 if (copy_from_user(&attr, argp, sizeof(attr)))
1148                         break;
1149                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1150                 break;
1151         }
1152         case KVM_HAS_DEVICE_ATTR: {
1153                 r = -EFAULT;
1154                 if (copy_from_user(&attr, argp, sizeof(attr)))
1155                         break;
1156                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1157                 break;
1158         }
1159         case KVM_GET_VCPU_EVENTS: {
1160                 struct kvm_vcpu_events events;
1161
1162                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1163                         return -EINVAL;
1164
1165                 if (copy_to_user(argp, &events, sizeof(events)))
1166                         return -EFAULT;
1167
1168                 return 0;
1169         }
1170         case KVM_SET_VCPU_EVENTS: {
1171                 struct kvm_vcpu_events events;
1172
1173                 if (copy_from_user(&events, argp, sizeof(events)))
1174                         return -EFAULT;
1175
1176                 return kvm_arm_vcpu_set_events(vcpu, &events);
1177         }
1178         default:
1179                 r = -EINVAL;
1180         }
1181
1182         return r;
1183 }
1184
1185 /**
1186  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1187  * @kvm: kvm instance
1188  * @log: slot id and address to which we copy the log
1189  *
1190  * Steps 1-4 below provide general overview of dirty page logging. See
1191  * kvm_get_dirty_log_protect() function description for additional details.
1192  *
1193  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1194  * always flush the TLB (step 4) even if previous step failed  and the dirty
1195  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1196  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1197  * writes will be marked dirty for next log read.
1198  *
1199  *   1. Take a snapshot of the bit and clear it if needed.
1200  *   2. Write protect the corresponding page.
1201  *   3. Copy the snapshot to the userspace.
1202  *   4. Flush TLB's if needed.
1203  */
1204 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1205 {
1206         bool is_dirty = false;
1207         int r;
1208
1209         mutex_lock(&kvm->slots_lock);
1210
1211         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1212
1213         if (is_dirty)
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         kvm_arm_init_debug();
1300 }
1301
1302 static void cpu_hyp_reset(void)
1303 {
1304         if (!is_kernel_in_hyp_mode())
1305                 __hyp_reset_vectors();
1306 }
1307
1308 static void cpu_hyp_reinit(void)
1309 {
1310         cpu_hyp_reset();
1311
1312         if (is_kernel_in_hyp_mode()) {
1313                 /*
1314                  * __cpu_init_stage2() is safe to call even if the PM
1315                  * event was cancelled before the CPU was reset.
1316                  */
1317                 __cpu_init_stage2();
1318                 kvm_timer_init_vhe();
1319         } else {
1320                 cpu_init_hyp_mode(NULL);
1321         }
1322
1323         if (vgic_present)
1324                 kvm_vgic_init_cpu_hardware();
1325 }
1326
1327 static void _kvm_arch_hardware_enable(void *discard)
1328 {
1329         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1330                 cpu_hyp_reinit();
1331                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1332         }
1333 }
1334
1335 int kvm_arch_hardware_enable(void)
1336 {
1337         _kvm_arch_hardware_enable(NULL);
1338         return 0;
1339 }
1340
1341 static void _kvm_arch_hardware_disable(void *discard)
1342 {
1343         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1344                 cpu_hyp_reset();
1345                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1346         }
1347 }
1348
1349 void kvm_arch_hardware_disable(void)
1350 {
1351         _kvm_arch_hardware_disable(NULL);
1352 }
1353
1354 #ifdef CONFIG_CPU_PM
1355 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1356                                     unsigned long cmd,
1357                                     void *v)
1358 {
1359         /*
1360          * kvm_arm_hardware_enabled is left with its old value over
1361          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1362          * re-enable hyp.
1363          */
1364         switch (cmd) {
1365         case CPU_PM_ENTER:
1366                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1367                         /*
1368                          * don't update kvm_arm_hardware_enabled here
1369                          * so that the hardware will be re-enabled
1370                          * when we resume. See below.
1371                          */
1372                         cpu_hyp_reset();
1373
1374                 return NOTIFY_OK;
1375         case CPU_PM_ENTER_FAILED:
1376         case CPU_PM_EXIT:
1377                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1378                         /* The hardware was enabled before suspend. */
1379                         cpu_hyp_reinit();
1380
1381                 return NOTIFY_OK;
1382
1383         default:
1384                 return NOTIFY_DONE;
1385         }
1386 }
1387
1388 static struct notifier_block hyp_init_cpu_pm_nb = {
1389         .notifier_call = hyp_init_cpu_pm_notifier,
1390 };
1391
1392 static void __init hyp_cpu_pm_init(void)
1393 {
1394         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1395 }
1396 static void __init hyp_cpu_pm_exit(void)
1397 {
1398         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1399 }
1400 #else
1401 static inline void hyp_cpu_pm_init(void)
1402 {
1403 }
1404 static inline void hyp_cpu_pm_exit(void)
1405 {
1406 }
1407 #endif
1408
1409 static int init_common_resources(void)
1410 {
1411         /* set size of VMID supported by CPU */
1412         kvm_vmid_bits = kvm_get_vmid_bits();
1413         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1414
1415         return 0;
1416 }
1417
1418 static int init_subsystems(void)
1419 {
1420         int err = 0;
1421
1422         /*
1423          * Enable hardware so that subsystem initialisation can access EL2.
1424          */
1425         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1426
1427         /*
1428          * Register CPU lower-power notifier
1429          */
1430         hyp_cpu_pm_init();
1431
1432         /*
1433          * Init HYP view of VGIC
1434          */
1435         err = kvm_vgic_hyp_init();
1436         switch (err) {
1437         case 0:
1438                 vgic_present = true;
1439                 break;
1440         case -ENODEV:
1441         case -ENXIO:
1442                 vgic_present = false;
1443                 err = 0;
1444                 break;
1445         default:
1446                 goto out;
1447         }
1448
1449         /*
1450          * Init HYP architected timer support
1451          */
1452         err = kvm_timer_hyp_init(vgic_present);
1453         if (err)
1454                 goto out;
1455
1456         kvm_perf_init();
1457         kvm_coproc_table_init();
1458
1459 out:
1460         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1461
1462         return err;
1463 }
1464
1465 static void teardown_hyp_mode(void)
1466 {
1467         int cpu;
1468
1469         free_hyp_pgds();
1470         for_each_possible_cpu(cpu)
1471                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1472         hyp_cpu_pm_exit();
1473 }
1474
1475 /**
1476  * Inits Hyp-mode on all online CPUs
1477  */
1478 static int init_hyp_mode(void)
1479 {
1480         int cpu;
1481         int err = 0;
1482
1483         /*
1484          * Allocate Hyp PGD and setup Hyp identity mapping
1485          */
1486         err = kvm_mmu_init();
1487         if (err)
1488                 goto out_err;
1489
1490         /*
1491          * Allocate stack pages for Hypervisor-mode
1492          */
1493         for_each_possible_cpu(cpu) {
1494                 unsigned long stack_page;
1495
1496                 stack_page = __get_free_page(GFP_KERNEL);
1497                 if (!stack_page) {
1498                         err = -ENOMEM;
1499                         goto out_err;
1500                 }
1501
1502                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1503         }
1504
1505         /*
1506          * Map the Hyp-code called directly from the host
1507          */
1508         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1509                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1510         if (err) {
1511                 kvm_err("Cannot map world-switch code\n");
1512                 goto out_err;
1513         }
1514
1515         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1516                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1517         if (err) {
1518                 kvm_err("Cannot map rodata section\n");
1519                 goto out_err;
1520         }
1521
1522         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1523                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1524         if (err) {
1525                 kvm_err("Cannot map bss section\n");
1526                 goto out_err;
1527         }
1528
1529         err = kvm_map_vectors();
1530         if (err) {
1531                 kvm_err("Cannot map vectors\n");
1532                 goto out_err;
1533         }
1534
1535         /*
1536          * Map the Hyp stack pages
1537          */
1538         for_each_possible_cpu(cpu) {
1539                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1540                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1541                                           PAGE_HYP);
1542
1543                 if (err) {
1544                         kvm_err("Cannot map hyp stack\n");
1545                         goto out_err;
1546                 }
1547         }
1548
1549         for_each_possible_cpu(cpu) {
1550                 kvm_cpu_context_t *cpu_ctxt;
1551
1552                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1553                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1554
1555                 if (err) {
1556                         kvm_err("Cannot map host CPU state: %d\n", err);
1557                         goto out_err;
1558                 }
1559         }
1560
1561         err = hyp_map_aux_data();
1562         if (err)
1563                 kvm_err("Cannot map host auxilary data: %d\n", err);
1564
1565         return 0;
1566
1567 out_err:
1568         teardown_hyp_mode();
1569         kvm_err("error initializing Hyp mode: %d\n", err);
1570         return err;
1571 }
1572
1573 static void check_kvm_target_cpu(void *ret)
1574 {
1575         *(int *)ret = kvm_target_cpu();
1576 }
1577
1578 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1579 {
1580         struct kvm_vcpu *vcpu;
1581         int i;
1582
1583         mpidr &= MPIDR_HWID_BITMASK;
1584         kvm_for_each_vcpu(i, vcpu, kvm) {
1585                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1586                         return vcpu;
1587         }
1588         return NULL;
1589 }
1590
1591 bool kvm_arch_has_irq_bypass(void)
1592 {
1593         return true;
1594 }
1595
1596 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1597                                       struct irq_bypass_producer *prod)
1598 {
1599         struct kvm_kernel_irqfd *irqfd =
1600                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1601
1602         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1603                                           &irqfd->irq_entry);
1604 }
1605 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1606                                       struct irq_bypass_producer *prod)
1607 {
1608         struct kvm_kernel_irqfd *irqfd =
1609                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1610
1611         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1612                                      &irqfd->irq_entry);
1613 }
1614
1615 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1616 {
1617         struct kvm_kernel_irqfd *irqfd =
1618                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1619
1620         kvm_arm_halt_guest(irqfd->kvm);
1621 }
1622
1623 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1624 {
1625         struct kvm_kernel_irqfd *irqfd =
1626                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1627
1628         kvm_arm_resume_guest(irqfd->kvm);
1629 }
1630
1631 /**
1632  * Initialize Hyp-mode and memory mappings on all CPUs.
1633  */
1634 int kvm_arch_init(void *opaque)
1635 {
1636         int err;
1637         int ret, cpu;
1638         bool in_hyp_mode;
1639
1640         if (!is_hyp_mode_available()) {
1641                 kvm_info("HYP mode not available\n");
1642                 return -ENODEV;
1643         }
1644
1645         if (!kvm_arch_check_sve_has_vhe()) {
1646                 kvm_pr_unimpl("SVE system without VHE unsupported.  Broken cpu?");
1647                 return -ENODEV;
1648         }
1649
1650         for_each_online_cpu(cpu) {
1651                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1652                 if (ret < 0) {
1653                         kvm_err("Error, CPU %d not supported!\n", cpu);
1654                         return -ENODEV;
1655                 }
1656         }
1657
1658         err = init_common_resources();
1659         if (err)
1660                 return err;
1661
1662         in_hyp_mode = is_kernel_in_hyp_mode();
1663
1664         if (!in_hyp_mode) {
1665                 err = init_hyp_mode();
1666                 if (err)
1667                         goto out_err;
1668         }
1669
1670         err = init_subsystems();
1671         if (err)
1672                 goto out_hyp;
1673
1674         if (in_hyp_mode)
1675                 kvm_info("VHE mode initialized successfully\n");
1676         else
1677                 kvm_info("Hyp mode initialized successfully\n");
1678
1679         return 0;
1680
1681 out_hyp:
1682         if (!in_hyp_mode)
1683                 teardown_hyp_mode();
1684 out_err:
1685         return err;
1686 }
1687
1688 /* NOP: Compiling as a module not supported */
1689 void kvm_arch_exit(void)
1690 {
1691         kvm_perf_teardown();
1692 }
1693
1694 static int arm_init(void)
1695 {
1696         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1697         return rc;
1698 }
1699
1700 module_init(arm_init);