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_SPINLOCK(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         ret = kvm_arm_setup_stage2(kvm, type);
 124         if (ret)
 125                 return ret;
 126
 127         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 128         if (!kvm->arch.last_vcpu_ran)
 129                 return -ENOMEM;
 130
 131         for_each_possible_cpu(cpu)
 132                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 133
 134         ret = kvm_alloc_stage2_pgd(kvm);
 135         if (ret)
 136                 goto out_fail_alloc;
 137
 138         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 139         if (ret)
 140                 goto out_free_stage2_pgd;
 141
 142         kvm_vgic_early_init(kvm);
 143
 144         /* Mark the initial VMID generation invalid */
 145         kvm->arch.vmid_gen = 0;
 146
 147         /* The maximum number of VCPUs is limited by the host's GIC model */
 148         kvm->arch.max_vcpus = vgic_present ?
 149                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 150
 151         return ret;
 152 out_free_stage2_pgd:
 153         kvm_free_stage2_pgd(kvm);
 154 out_fail_alloc:
 155         free_percpu(kvm->arch.last_vcpu_ran);
 156         kvm->arch.last_vcpu_ran = NULL;
 157         return ret;
 158 }
 159
 160 bool kvm_arch_has_vcpu_debugfs(void)
 161 {
 162         return false;
 163 }
 164
 165 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 166 {
 167         return 0;
 168 }
 169
 170 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 171 {
 172         return VM_FAULT_SIGBUS;
 173 }
 174
 175
 176 /**
 177  * kvm_arch_destroy_vm - destroy the VM data structure
 178  * @kvm:        pointer to the KVM struct
 179  */
 180 void kvm_arch_destroy_vm(struct kvm *kvm)
 181 {
 182         int i;
 183
 184         kvm_vgic_destroy(kvm);
 185
 186         free_percpu(kvm->arch.last_vcpu_ran);
 187         kvm->arch.last_vcpu_ran = NULL;
 188
 189         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 190                 if (kvm->vcpus[i]) {
 191                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 192                         kvm->vcpus[i] = NULL;
 193                 }
 194         }
 195         atomic_set(&kvm->online_vcpus, 0);
 196 }
 197
 198 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 199 {
 200         int r;
 201         switch (ext) {
 202         case KVM_CAP_IRQCHIP:
 203                 r = vgic_present;
 204                 break;
 205         case KVM_CAP_IOEVENTFD:
 206         case KVM_CAP_DEVICE_CTRL:
 207         case KVM_CAP_USER_MEMORY:
 208         case KVM_CAP_SYNC_MMU:
 209         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 210         case KVM_CAP_ONE_REG:
 211         case KVM_CAP_ARM_PSCI:
 212         case KVM_CAP_ARM_PSCI_0_2:
 213         case KVM_CAP_READONLY_MEM:
 214         case KVM_CAP_MP_STATE:
 215         case KVM_CAP_IMMEDIATE_EXIT:
 216         case KVM_CAP_VCPU_EVENTS:
 217                 r = 1;
 218                 break;
 219         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 220                 r = 1;
 221                 break;
 222         case KVM_CAP_NR_VCPUS:
 223                 r = num_online_cpus();
 224                 break;
 225         case KVM_CAP_MAX_VCPUS:
 226                 r = KVM_MAX_VCPUS;
 227                 break;
 228         case KVM_CAP_NR_MEMSLOTS:
 229                 r = KVM_USER_MEM_SLOTS;
 230                 break;
 231         case KVM_CAP_MSI_DEVID:
 232                 if (!kvm)
 233                         r = -EINVAL;
 234                 else
 235                         r = kvm->arch.vgic.msis_require_devid;
 236                 break;
 237         case KVM_CAP_ARM_USER_IRQ:
 238                 /*
 239                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 240                  * (bump this number if adding more devices)
 241                  */
 242                 r = 1;
 243                 break;
 244         default:
 245                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 246                 break;
 247         }
 248         return r;
 249 }
 250
 251 long kvm_arch_dev_ioctl(struct file *filp,
 252                         unsigned int ioctl, unsigned long arg)
 253 {
 254         return -EINVAL;
 255 }
 256
 257 struct kvm *kvm_arch_alloc_vm(void)
 258 {
 259         if (!has_vhe())
 260                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 261
 262         return vzalloc(sizeof(struct kvm));
 263 }
 264
 265 void kvm_arch_free_vm(struct kvm *kvm)
 266 {
 267         if (!has_vhe())
 268                 kfree(kvm);
 269         else
 270                 vfree(kvm);
 271 }
 272
 273 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 274 {
 275         int err;
 276         struct kvm_vcpu *vcpu;
 277
 278         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 279                 err = -EBUSY;
 280                 goto out;
 281         }
 282
 283         if (id >= kvm->arch.max_vcpus) {
 284                 err = -EINVAL;
 285                 goto out;
 286         }
 287
 288         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 289         if (!vcpu) {
 290                 err = -ENOMEM;
 291                 goto out;
 292         }
 293
 294         err = kvm_vcpu_init(vcpu, kvm, id);
 295         if (err)
 296                 goto free_vcpu;
 297
 298         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 299         if (err)
 300                 goto vcpu_uninit;
 301
 302         return vcpu;
 303 vcpu_uninit:
 304         kvm_vcpu_uninit(vcpu);
 305 free_vcpu:
 306         kmem_cache_free(kvm_vcpu_cache, vcpu);
 307 out:
 308         return ERR_PTR(err);
 309 }
 310
 311 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 312 {
 313 }
 314
 315 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 316 {
 317         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 318                 static_branch_dec(&userspace_irqchip_in_use);
 319
 320         kvm_mmu_free_memory_caches(vcpu);
 321         kvm_timer_vcpu_terminate(vcpu);
 322         kvm_pmu_vcpu_destroy(vcpu);
 323         kvm_vcpu_uninit(vcpu);
 324         kmem_cache_free(kvm_vcpu_cache, vcpu);
 325 }
 326
 327 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 328 {
 329         kvm_arch_vcpu_free(vcpu);
 330 }
 331
 332 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 333 {
 334         return kvm_timer_is_pending(vcpu);
 335 }
 336
 337 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 338 {
 339         kvm_timer_schedule(vcpu);
 340         kvm_vgic_v4_enable_doorbell(vcpu);
 341 }
 342
 343 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 344 {
 345         kvm_timer_unschedule(vcpu);
 346         kvm_vgic_v4_disable_doorbell(vcpu);
 347 }
 348
 349 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 350 {
 351         /* Force users to call KVM_ARM_VCPU_INIT */
 352         vcpu->arch.target = -1;
 353         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 354
 355         /* Set up the timer */
 356         kvm_timer_vcpu_init(vcpu);
 357
 358         kvm_arm_reset_debug_ptr(vcpu);
 359
 360         return kvm_vgic_vcpu_init(vcpu);
 361 }
 362
 363 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 364 {
 365         int *last_ran;
 366
 367         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 368
 369         /*
 370          * We might get preempted before the vCPU actually runs, but
 371          * over-invalidation doesn't affect correctness.
 372          */
 373         if (*last_ran != vcpu->vcpu_id) {
 374                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 375                 *last_ran = vcpu->vcpu_id;
 376         }
 377
 378         vcpu->cpu = cpu;
 379         vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
 380
 381         kvm_arm_set_running_vcpu(vcpu);
 382         kvm_vgic_load(vcpu);
 383         kvm_timer_vcpu_load(vcpu);
 384         kvm_vcpu_load_sysregs(vcpu);
 385         kvm_arch_vcpu_load_fp(vcpu);
 386
 387         if (single_task_running())
 388                 vcpu_clear_wfe_traps(vcpu);
 389         else
 390                 vcpu_set_wfe_traps(vcpu);
 391 }
 392
 393 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 394 {
 395         kvm_arch_vcpu_put_fp(vcpu);
 396         kvm_vcpu_put_sysregs(vcpu);
 397         kvm_timer_vcpu_put(vcpu);
 398         kvm_vgic_put(vcpu);
 399
 400         vcpu->cpu = -1;
 401
 402         kvm_arm_set_running_vcpu(NULL);
 403 }
 404
 405 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 406 {
 407         vcpu->arch.power_off = true;
 408         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 409         kvm_vcpu_kick(vcpu);
 410 }
 411
 412 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 413                                     struct kvm_mp_state *mp_state)
 414 {
 415         if (vcpu->arch.power_off)
 416                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 417         else
 418                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 419
 420         return 0;
 421 }
 422
 423 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 424                                     struct kvm_mp_state *mp_state)
 425 {
 426         int ret = 0;
 427
 428         switch (mp_state->mp_state) {
 429         case KVM_MP_STATE_RUNNABLE:
 430                 vcpu->arch.power_off = false;
 431                 break;
 432         case KVM_MP_STATE_STOPPED:
 433                 vcpu_power_off(vcpu);
 434                 break;
 435         default:
 436                 ret = -EINVAL;
 437         }
 438
 439         return ret;
 440 }
 441
 442 /**
 443  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 444  * @v:          The VCPU pointer
 445  *
 446  * If the guest CPU is not waiting for interrupts or an interrupt line is
 447  * asserted, the CPU is by definition runnable.
 448  */
 449 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 450 {
 451         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 452         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 453                 && !v->arch.power_off && !v->arch.pause);
 454 }
 455
 456 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 457 {
 458         return vcpu_mode_priv(vcpu);
 459 }
 460
 461 /* Just ensure a guest exit from a particular CPU */
 462 static void exit_vm_noop(void *info)
 463 {
 464 }
 465
 466 void force_vm_exit(const cpumask_t *mask)
 467 {
 468         preempt_disable();
 469         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 470         preempt_enable();
 471 }
 472
 473 /**
 474  * need_new_vmid_gen - check that the VMID is still valid
 475  * @kvm: The VM's VMID to check
 476  *
 477  * return true if there is a new generation of VMIDs being used
 478  *
 479  * The hardware supports only 256 values with the value zero reserved for the
 480  * host, so we check if an assigned value belongs to a previous generation,
 481  * which which requires us to assign a new value. If we're the first to use a
 482  * VMID for the new generation, we must flush necessary caches and TLBs on all
 483  * CPUs.
 484  */
 485 static bool need_new_vmid_gen(struct kvm *kvm)
 486 {
 487         u64 current_vmid_gen = atomic64_read(&kvm_vmid_gen);
 488         smp_rmb(); /* Orders read of kvm_vmid_gen and kvm->arch.vmid */
 489         return unlikely(READ_ONCE(kvm->arch.vmid_gen) != current_vmid_gen);
 490 }
 491
 492 /**
 493  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 494  * @kvm The guest that we are about to run
 495  *
 496  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 497  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 498  * caches and TLBs.
 499  */
 500 static void update_vttbr(struct kvm *kvm)
 501 {
 502         phys_addr_t pgd_phys;
 503         u64 vmid, cnp = kvm_cpu_has_cnp() ? VTTBR_CNP_BIT : 0;
 504
 505         if (!need_new_vmid_gen(kvm))
 506                 return;
 507
 508         spin_lock(&kvm_vmid_lock);
 509
 510         /*
 511          * We need to re-check the vmid_gen here to ensure that if another vcpu
 512          * already allocated a valid vmid for this vm, then this vcpu should
 513          * use the same vmid.
 514          */
 515         if (!need_new_vmid_gen(kvm)) {
 516                 spin_unlock(&kvm_vmid_lock);
 517                 return;
 518         }
 519
 520         /* First user of a new VMID generation? */
 521         if (unlikely(kvm_next_vmid == 0)) {
 522                 atomic64_inc(&kvm_vmid_gen);
 523                 kvm_next_vmid = 1;
 524
 525                 /*
 526                  * On SMP we know no other CPUs can use this CPU's or each
 527                  * other's VMID after force_vm_exit returns since the
 528                  * kvm_vmid_lock blocks them from reentry to the guest.
 529                  */
 530                 force_vm_exit(cpu_all_mask);
 531                 /*
 532                  * Now broadcast TLB + ICACHE invalidation over the inner
 533                  * shareable domain to make sure all data structures are
 534                  * clean.
 535                  */
 536                 kvm_call_hyp(__kvm_flush_vm_context);
 537         }
 538
 539         kvm->arch.vmid = kvm_next_vmid;
 540         kvm_next_vmid++;
 541         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 542
 543         /* update vttbr to be used with the new vmid */
 544         pgd_phys = virt_to_phys(kvm->arch.pgd);
 545         BUG_ON(pgd_phys & ~kvm_vttbr_baddr_mask(kvm));
 546         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 547         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid | cnp;
 548
 549         smp_wmb();
 550         WRITE_ONCE(kvm->arch.vmid_gen, atomic64_read(&kvm_vmid_gen));
 551
 552         spin_unlock(&kvm_vmid_lock);
 553 }
 554
 555 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 556 {
 557         struct kvm *kvm = vcpu->kvm;
 558         int ret = 0;
 559
 560         if (likely(vcpu->arch.has_run_once))
 561                 return 0;
 562
 563         vcpu->arch.has_run_once = true;
 564
 565         if (likely(irqchip_in_kernel(kvm))) {
 566                 /*
 567                  * Map the VGIC hardware resources before running a vcpu the
 568                  * first time on this VM.
 569                  */
 570                 if (unlikely(!vgic_ready(kvm))) {
 571                         ret = kvm_vgic_map_resources(kvm);
 572                         if (ret)
 573                                 return ret;
 574                 }
 575         } else {
 576                 /*
 577                  * Tell the rest of the code that there are userspace irqchip
 578                  * VMs in the wild.
 579                  */
 580                 static_branch_inc(&userspace_irqchip_in_use);
 581         }
 582
 583         ret = kvm_timer_enable(vcpu);
 584         if (ret)
 585                 return ret;
 586
 587         ret = kvm_arm_pmu_v3_enable(vcpu);
 588
 589         return ret;
 590 }
 591
 592 bool kvm_arch_intc_initialized(struct kvm *kvm)
 593 {
 594         return vgic_initialized(kvm);
 595 }
 596
 597 void kvm_arm_halt_guest(struct kvm *kvm)
 598 {
 599         int i;
 600         struct kvm_vcpu *vcpu;
 601
 602         kvm_for_each_vcpu(i, vcpu, kvm)
 603                 vcpu->arch.pause = true;
 604         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 605 }
 606
 607 void kvm_arm_resume_guest(struct kvm *kvm)
 608 {
 609         int i;
 610         struct kvm_vcpu *vcpu;
 611
 612         kvm_for_each_vcpu(i, vcpu, kvm) {
 613                 vcpu->arch.pause = false;
 614                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 615         }
 616 }
 617
 618 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 619 {
 620         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 621
 622         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 623                                        (!vcpu->arch.pause)));
 624
 625         if (vcpu->arch.power_off || vcpu->arch.pause) {
 626                 /* Awaken to handle a signal, request we sleep again later. */
 627                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 628         }
 629 }
 630
 631 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 632 {
 633         return vcpu->arch.target >= 0;
 634 }
 635
 636 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 637 {
 638         if (kvm_request_pending(vcpu)) {
 639                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 640                         vcpu_req_sleep(vcpu);
 641
 642                 /*
 643                  * Clear IRQ_PENDING requests that were made to guarantee
 644                  * that a VCPU sees new virtual interrupts.
 645                  */
 646                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 647         }
 648 }
 649
 650 /**
 651  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 652  * @vcpu:       The VCPU pointer
 653  * @run:        The kvm_run structure pointer used for userspace state exchange
 654  *
 655  * This function is called through the VCPU_RUN ioctl called from user space. It
 656  * will execute VM code in a loop until the time slice for the process is used
 657  * or some emulation is needed from user space in which case the function will
 658  * return with return value 0 and with the kvm_run structure filled in with the
 659  * required data for the requested emulation.
 660  */
 661 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 662 {
 663         int ret;
 664
 665         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 666                 return -ENOEXEC;
 667
 668         ret = kvm_vcpu_first_run_init(vcpu);
 669         if (ret)
 670                 return ret;
 671
 672         if (run->exit_reason == KVM_EXIT_MMIO) {
 673                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 674                 if (ret)
 675                         return ret;
 676         }
 677
 678         if (run->immediate_exit)
 679                 return -EINTR;
 680
 681         vcpu_load(vcpu);
 682
 683         kvm_sigset_activate(vcpu);
 684
 685         ret = 1;
 686         run->exit_reason = KVM_EXIT_UNKNOWN;
 687         while (ret > 0) {
 688                 /*
 689                  * Check conditions before entering the guest
 690                  */
 691                 cond_resched();
 692
 693                 update_vttbr(vcpu->kvm);
 694
 695                 check_vcpu_requests(vcpu);
 696
 697                 /*
 698                  * Preparing the interrupts to be injected also
 699                  * involves poking the GIC, which must be done in a
 700                  * non-preemptible context.
 701                  */
 702                 preempt_disable();
 703
 704                 kvm_pmu_flush_hwstate(vcpu);
 705
 706                 local_irq_disable();
 707
 708                 kvm_vgic_flush_hwstate(vcpu);
 709
 710                 /*
 711                  * Exit if we have a signal pending so that we can deliver the
 712                  * signal to user space.
 713                  */
 714                 if (signal_pending(current)) {
 715                         ret = -EINTR;
 716                         run->exit_reason = KVM_EXIT_INTR;
 717                 }
 718
 719                 /*
 720                  * If we're using a userspace irqchip, then check if we need
 721                  * to tell a userspace irqchip about timer or PMU level
 722                  * changes and if so, exit to userspace (the actual level
 723                  * state gets updated in kvm_timer_update_run and
 724                  * kvm_pmu_update_run below).
 725                  */
 726                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 727                         if (kvm_timer_should_notify_user(vcpu) ||
 728                             kvm_pmu_should_notify_user(vcpu)) {
 729                                 ret = -EINTR;
 730                                 run->exit_reason = KVM_EXIT_INTR;
 731                         }
 732                 }
 733
 734                 /*
 735                  * Ensure we set mode to IN_GUEST_MODE after we disable
 736                  * interrupts and before the final VCPU requests check.
 737                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 738                  * Documentation/virtual/kvm/vcpu-requests.rst
 739                  */
 740                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 741
 742                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 743                     kvm_request_pending(vcpu)) {
 744                         vcpu->mode = OUTSIDE_GUEST_MODE;
 745                         isb(); /* Ensure work in x_flush_hwstate is committed */
 746                         kvm_pmu_sync_hwstate(vcpu);
 747                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 748                                 kvm_timer_sync_hwstate(vcpu);
 749                         kvm_vgic_sync_hwstate(vcpu);
 750                         local_irq_enable();
 751                         preempt_enable();
 752                         continue;
 753                 }
 754
 755                 kvm_arm_setup_debug(vcpu);
 756
 757                 /**************************************************************
 758                  * Enter the guest
 759                  */
 760                 trace_kvm_entry(*vcpu_pc(vcpu));
 761                 guest_enter_irqoff();
 762
 763                 if (has_vhe()) {
 764                         kvm_arm_vhe_guest_enter();
 765                         ret = kvm_vcpu_run_vhe(vcpu);
 766                         kvm_arm_vhe_guest_exit();
 767                 } else {
 768                         ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
 769                 }
 770
 771                 vcpu->mode = OUTSIDE_GUEST_MODE;
 772                 vcpu->stat.exits++;
 773                 /*
 774                  * Back from guest
 775                  *************************************************************/
 776
 777                 kvm_arm_clear_debug(vcpu);
 778
 779                 /*
 780                  * We must sync the PMU state before the vgic state so
 781                  * that the vgic can properly sample the updated state of the
 782                  * interrupt line.
 783                  */
 784                 kvm_pmu_sync_hwstate(vcpu);
 785
 786                 /*
 787                  * Sync the vgic state before syncing the timer state because
 788                  * the timer code needs to know if the virtual timer
 789                  * interrupts are active.
 790                  */
 791                 kvm_vgic_sync_hwstate(vcpu);
 792
 793                 /*
 794                  * Sync the timer hardware state before enabling interrupts as
 795                  * we don't want vtimer interrupts to race with syncing the
 796                  * timer virtual interrupt state.
 797                  */
 798                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 799                         kvm_timer_sync_hwstate(vcpu);
 800
 801                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 802
 803                 /*
 804                  * We may have taken a host interrupt in HYP mode (ie
 805                  * while executing the guest). This interrupt is still
 806                  * pending, as we haven't serviced it yet!
 807                  *
 808                  * We're now back in SVC mode, with interrupts
 809                  * disabled.  Enabling the interrupts now will have
 810                  * the effect of taking the interrupt again, in SVC
 811                  * mode this time.
 812                  */
 813                 local_irq_enable();
 814
 815                 /*
 816                  * We do local_irq_enable() before calling guest_exit() so
 817                  * that if a timer interrupt hits while running the guest we
 818                  * account that tick as being spent in the guest.  We enable
 819                  * preemption after calling guest_exit() so that if we get
 820                  * preempted we make sure ticks after that is not counted as
 821                  * guest time.
 822                  */
 823                 guest_exit();
 824                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 825
 826                 /* Exit types that need handling before we can be preempted */
 827                 handle_exit_early(vcpu, run, ret);
 828
 829                 preempt_enable();
 830
 831                 ret = handle_exit(vcpu, run, ret);
 832         }
 833
 834         /* Tell userspace about in-kernel device output levels */
 835         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 836                 kvm_timer_update_run(vcpu);
 837                 kvm_pmu_update_run(vcpu);
 838         }
 839
 840         kvm_sigset_deactivate(vcpu);
 841
 842         vcpu_put(vcpu);
 843         return ret;
 844 }
 845
 846 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 847 {
 848         int bit_index;
 849         bool set;
 850         unsigned long *hcr;
 851
 852         if (number == KVM_ARM_IRQ_CPU_IRQ)
 853                 bit_index = __ffs(HCR_VI);
 854         else /* KVM_ARM_IRQ_CPU_FIQ */
 855                 bit_index = __ffs(HCR_VF);
 856
 857         hcr = vcpu_hcr(vcpu);
 858         if (level)
 859                 set = test_and_set_bit(bit_index, hcr);
 860         else
 861                 set = test_and_clear_bit(bit_index, hcr);
 862
 863         /*
 864          * If we didn't change anything, no need to wake up or kick other CPUs
 865          */
 866         if (set == level)
 867                 return 0;
 868
 869         /*
 870          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 871          * trigger a world-switch round on the running physical CPU to set the
 872          * virtual IRQ/FIQ fields in the HCR appropriately.
 873          */
 874         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 875         kvm_vcpu_kick(vcpu);
 876
 877         return 0;
 878 }
 879
 880 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 881                           bool line_status)
 882 {
 883         u32 irq = irq_level->irq;
 884         unsigned int irq_type, vcpu_idx, irq_num;
 885         int nrcpus = atomic_read(&kvm->online_vcpus);
 886         struct kvm_vcpu *vcpu = NULL;
 887         bool level = irq_level->level;
 888
 889         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 890         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 891         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 892
 893         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 894
 895         switch (irq_type) {
 896         case KVM_ARM_IRQ_TYPE_CPU:
 897                 if (irqchip_in_kernel(kvm))
 898                         return -ENXIO;
 899
 900                 if (vcpu_idx >= nrcpus)
 901                         return -EINVAL;
 902
 903                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 904                 if (!vcpu)
 905                         return -EINVAL;
 906
 907                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 908                         return -EINVAL;
 909
 910                 return vcpu_interrupt_line(vcpu, irq_num, level);
 911         case KVM_ARM_IRQ_TYPE_PPI:
 912                 if (!irqchip_in_kernel(kvm))
 913                         return -ENXIO;
 914
 915                 if (vcpu_idx >= nrcpus)
 916                         return -EINVAL;
 917
 918                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 919                 if (!vcpu)
 920                         return -EINVAL;
 921
 922                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 923                         return -EINVAL;
 924
 925                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 926         case KVM_ARM_IRQ_TYPE_SPI:
 927                 if (!irqchip_in_kernel(kvm))
 928                         return -ENXIO;
 929
 930                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 931                         return -EINVAL;
 932
 933                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 934         }
 935
 936         return -EINVAL;
 937 }
 938
 939 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 940                                const struct kvm_vcpu_init *init)
 941 {
 942         unsigned int i;
 943         int phys_target = kvm_target_cpu();
 944
 945         if (init->target != phys_target)
 946                 return -EINVAL;
 947
 948         /*
 949          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 950          * use the same target.
 951          */
 952         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 953                 return -EINVAL;
 954
 955         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 956         for (i = 0; i < sizeof(init->features) * 8; i++) {
 957                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 958
 959                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 960                         return -ENOENT;
 961
 962                 /*
 963                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 964                  * use the same feature set.
 965                  */
 966                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 967                     test_bit(i, vcpu->arch.features) != set)
 968                         return -EINVAL;
 969
 970                 if (set)
 971                         set_bit(i, vcpu->arch.features);
 972         }
 973
 974         vcpu->arch.target = phys_target;
 975
 976         /* Now we know what it is, we can reset it. */
 977         return kvm_reset_vcpu(vcpu);
 978 }
 979
 980
 981 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 982                                          struct kvm_vcpu_init *init)
 983 {
 984         int ret;
 985
 986         ret = kvm_vcpu_set_target(vcpu, init);
 987         if (ret)
 988                 return ret;
 989
 990         /*
 991          * Ensure a rebooted VM will fault in RAM pages and detect if the
 992          * guest MMU is turned off and flush the caches as needed.
 993          */
 994         if (vcpu->arch.has_run_once)
 995                 stage2_unmap_vm(vcpu->kvm);
 996
 997         vcpu_reset_hcr(vcpu);
 998
 999         /*
1000          * Handle the "start in power-off" case.
1001          */
1002         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1003                 vcpu_power_off(vcpu);
1004         else
1005                 vcpu->arch.power_off = false;
1006
1007         return 0;
1008 }
1009
1010 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1011                                  struct kvm_device_attr *attr)
1012 {
1013         int ret = -ENXIO;
1014
1015         switch (attr->group) {
1016         default:
1017                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1018                 break;
1019         }
1020
1021         return ret;
1022 }
1023
1024 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1025                                  struct kvm_device_attr *attr)
1026 {
1027         int ret = -ENXIO;
1028
1029         switch (attr->group) {
1030         default:
1031                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1032                 break;
1033         }
1034
1035         return ret;
1036 }
1037
1038 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1039                                  struct kvm_device_attr *attr)
1040 {
1041         int ret = -ENXIO;
1042
1043         switch (attr->group) {
1044         default:
1045                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1046                 break;
1047         }
1048
1049         return ret;
1050 }
1051
1052 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1053                                    struct kvm_vcpu_events *events)
1054 {
1055         memset(events, 0, sizeof(*events));
1056
1057         return __kvm_arm_vcpu_get_events(vcpu, events);
1058 }
1059
1060 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1061                                    struct kvm_vcpu_events *events)
1062 {
1063         int i;
1064
1065         /* check whether the reserved field is zero */
1066         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1067                 if (events->reserved[i])
1068                         return -EINVAL;
1069
1070         /* check whether the pad field is zero */
1071         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1072                 if (events->exception.pad[i])
1073                         return -EINVAL;
1074
1075         return __kvm_arm_vcpu_set_events(vcpu, events);
1076 }
1077
1078 long kvm_arch_vcpu_ioctl(struct file *filp,
1079                          unsigned int ioctl, unsigned long arg)
1080 {
1081         struct kvm_vcpu *vcpu = filp->private_data;
1082         void __user *argp = (void __user *)arg;
1083         struct kvm_device_attr attr;
1084         long r;
1085
1086         switch (ioctl) {
1087         case KVM_ARM_VCPU_INIT: {
1088                 struct kvm_vcpu_init init;
1089
1090                 r = -EFAULT;
1091                 if (copy_from_user(&init, argp, sizeof(init)))
1092                         break;
1093
1094                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1095                 break;
1096         }
1097         case KVM_SET_ONE_REG:
1098         case KVM_GET_ONE_REG: {
1099                 struct kvm_one_reg reg;
1100
1101                 r = -ENOEXEC;
1102                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1103                         break;
1104
1105                 r = -EFAULT;
1106                 if (copy_from_user(&reg, argp, sizeof(reg)))
1107                         break;
1108
1109                 if (ioctl == KVM_SET_ONE_REG)
1110                         r = kvm_arm_set_reg(vcpu, &reg);
1111                 else
1112                         r = kvm_arm_get_reg(vcpu, &reg);
1113                 break;
1114         }
1115         case KVM_GET_REG_LIST: {
1116                 struct kvm_reg_list __user *user_list = argp;
1117                 struct kvm_reg_list reg_list;
1118                 unsigned n;
1119
1120                 r = -ENOEXEC;
1121                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1122                         break;
1123
1124                 r = -EFAULT;
1125                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1126                         break;
1127                 n = reg_list.n;
1128                 reg_list.n = kvm_arm_num_regs(vcpu);
1129                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1130                         break;
1131                 r = -E2BIG;
1132                 if (n < reg_list.n)
1133                         break;
1134                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1135                 break;
1136         }
1137         case KVM_SET_DEVICE_ATTR: {
1138                 r = -EFAULT;
1139                 if (copy_from_user(&attr, argp, sizeof(attr)))
1140                         break;
1141                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1142                 break;
1143         }
1144         case KVM_GET_DEVICE_ATTR: {
1145                 r = -EFAULT;
1146                 if (copy_from_user(&attr, argp, sizeof(attr)))
1147                         break;
1148                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1149                 break;
1150         }
1151         case KVM_HAS_DEVICE_ATTR: {
1152                 r = -EFAULT;
1153                 if (copy_from_user(&attr, argp, sizeof(attr)))
1154                         break;
1155                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1156                 break;
1157         }
1158         case KVM_GET_VCPU_EVENTS: {
1159                 struct kvm_vcpu_events events;
1160
1161                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1162                         return -EINVAL;
1163
1164                 if (copy_to_user(argp, &events, sizeof(events)))
1165                         return -EFAULT;
1166
1167                 return 0;
1168         }
1169         case KVM_SET_VCPU_EVENTS: {
1170                 struct kvm_vcpu_events events;
1171
1172                 if (copy_from_user(&events, argp, sizeof(events)))
1173                         return -EFAULT;
1174
1175                 return kvm_arm_vcpu_set_events(vcpu, &events);
1176         }
1177         default:
1178                 r = -EINVAL;
1179         }
1180
1181         return r;
1182 }
1183
1184 /**
1185  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1186  * @kvm: kvm instance
1187  * @log: slot id and address to which we copy the log
1188  *
1189  * Steps 1-4 below provide general overview of dirty page logging. See
1190  * kvm_get_dirty_log_protect() function description for additional details.
1191  *
1192  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1193  * always flush the TLB (step 4) even if previous step failed  and the dirty
1194  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1195  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1196  * writes will be marked dirty for next log read.
1197  *
1198  *   1. Take a snapshot of the bit and clear it if needed.
1199  *   2. Write protect the corresponding page.
1200  *   3. Copy the snapshot to the userspace.
1201  *   4. Flush TLB's if needed.
1202  */
1203 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1204 {
1205         bool flush = false;
1206         int r;
1207
1208         mutex_lock(&kvm->slots_lock);
1209
1210         r = kvm_get_dirty_log_protect(kvm, log, &flush);
1211
1212         if (flush)
1213                 kvm_flush_remote_tlbs(kvm);
1214
1215         mutex_unlock(&kvm->slots_lock);
1216         return r;
1217 }
1218
1219 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
1220 {
1221         bool flush = false;
1222         int r;
1223
1224         mutex_lock(&kvm->slots_lock);
1225
1226         r = kvm_clear_dirty_log_protect(kvm, log, &flush);
1227
1228         if (flush)
1229                 kvm_flush_remote_tlbs(kvm);
1230
1231         mutex_unlock(&kvm->slots_lock);
1232         return r;
1233 }
1234
1235 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1236                                         struct kvm_arm_device_addr *dev_addr)
1237 {
1238         unsigned long dev_id, type;
1239
1240         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1241                 KVM_ARM_DEVICE_ID_SHIFT;
1242         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1243                 KVM_ARM_DEVICE_TYPE_SHIFT;
1244
1245         switch (dev_id) {
1246         case KVM_ARM_DEVICE_VGIC_V2:
1247                 if (!vgic_present)
1248                         return -ENXIO;
1249                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1250         default:
1251                 return -ENODEV;
1252         }
1253 }
1254
1255 long kvm_arch_vm_ioctl(struct file *filp,
1256                        unsigned int ioctl, unsigned long arg)
1257 {
1258         struct kvm *kvm = filp->private_data;
1259         void __user *argp = (void __user *)arg;
1260
1261         switch (ioctl) {
1262         case KVM_CREATE_IRQCHIP: {
1263                 int ret;
1264                 if (!vgic_present)
1265                         return -ENXIO;
1266                 mutex_lock(&kvm->lock);
1267                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1268                 mutex_unlock(&kvm->lock);
1269                 return ret;
1270         }
1271         case KVM_ARM_SET_DEVICE_ADDR: {
1272                 struct kvm_arm_device_addr dev_addr;
1273
1274                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1275                         return -EFAULT;
1276                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1277         }
1278         case KVM_ARM_PREFERRED_TARGET: {
1279                 int err;
1280                 struct kvm_vcpu_init init;
1281
1282                 err = kvm_vcpu_preferred_target(&init);
1283                 if (err)
1284                         return err;
1285
1286                 if (copy_to_user(argp, &init, sizeof(init)))
1287                         return -EFAULT;
1288
1289                 return 0;
1290         }
1291         default:
1292                 return -EINVAL;
1293         }
1294 }
1295
1296 static void cpu_init_hyp_mode(void *dummy)
1297 {
1298         phys_addr_t pgd_ptr;
1299         unsigned long hyp_stack_ptr;
1300         unsigned long stack_page;
1301         unsigned long vector_ptr;
1302
1303         /* Switch from the HYP stub to our own HYP init vector */
1304         __hyp_set_vectors(kvm_get_idmap_vector());
1305
1306         pgd_ptr = kvm_mmu_get_httbr();
1307         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1308         hyp_stack_ptr = stack_page + PAGE_SIZE;
1309         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1310
1311         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1312         __cpu_init_stage2();
1313 }
1314
1315 static void cpu_hyp_reset(void)
1316 {
1317         if (!is_kernel_in_hyp_mode())
1318                 __hyp_reset_vectors();
1319 }
1320
1321 static void cpu_hyp_reinit(void)
1322 {
1323         cpu_hyp_reset();
1324
1325         if (is_kernel_in_hyp_mode())
1326                 kvm_timer_init_vhe();
1327         else
1328                 cpu_init_hyp_mode(NULL);
1329
1330         kvm_arm_init_debug();
1331
1332         if (vgic_present)
1333                 kvm_vgic_init_cpu_hardware();
1334 }
1335
1336 static void _kvm_arch_hardware_enable(void *discard)
1337 {
1338         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1339                 cpu_hyp_reinit();
1340                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1341         }
1342 }
1343
1344 int kvm_arch_hardware_enable(void)
1345 {
1346         _kvm_arch_hardware_enable(NULL);
1347         return 0;
1348 }
1349
1350 static void _kvm_arch_hardware_disable(void *discard)
1351 {
1352         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1353                 cpu_hyp_reset();
1354                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1355         }
1356 }
1357
1358 void kvm_arch_hardware_disable(void)
1359 {
1360         _kvm_arch_hardware_disable(NULL);
1361 }
1362
1363 #ifdef CONFIG_CPU_PM
1364 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1365                                     unsigned long cmd,
1366                                     void *v)
1367 {
1368         /*
1369          * kvm_arm_hardware_enabled is left with its old value over
1370          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1371          * re-enable hyp.
1372          */
1373         switch (cmd) {
1374         case CPU_PM_ENTER:
1375                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1376                         /*
1377                          * don't update kvm_arm_hardware_enabled here
1378                          * so that the hardware will be re-enabled
1379                          * when we resume. See below.
1380                          */
1381                         cpu_hyp_reset();
1382
1383                 return NOTIFY_OK;
1384         case CPU_PM_ENTER_FAILED:
1385         case CPU_PM_EXIT:
1386                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1387                         /* The hardware was enabled before suspend. */
1388                         cpu_hyp_reinit();
1389
1390                 return NOTIFY_OK;
1391
1392         default:
1393                 return NOTIFY_DONE;
1394         }
1395 }
1396
1397 static struct notifier_block hyp_init_cpu_pm_nb = {
1398         .notifier_call = hyp_init_cpu_pm_notifier,
1399 };
1400
1401 static void __init hyp_cpu_pm_init(void)
1402 {
1403         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1404 }
1405 static void __init hyp_cpu_pm_exit(void)
1406 {
1407         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1408 }
1409 #else
1410 static inline void hyp_cpu_pm_init(void)
1411 {
1412 }
1413 static inline void hyp_cpu_pm_exit(void)
1414 {
1415 }
1416 #endif
1417
1418 static int init_common_resources(void)
1419 {
1420         /* set size of VMID supported by CPU */
1421         kvm_vmid_bits = kvm_get_vmid_bits();
1422         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1423
1424         kvm_set_ipa_limit();
1425
1426         return 0;
1427 }
1428
1429 static int init_subsystems(void)
1430 {
1431         int err = 0;
1432
1433         /*
1434          * Enable hardware so that subsystem initialisation can access EL2.
1435          */
1436         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1437
1438         /*
1439          * Register CPU lower-power notifier
1440          */
1441         hyp_cpu_pm_init();
1442
1443         /*
1444          * Init HYP view of VGIC
1445          */
1446         err = kvm_vgic_hyp_init();
1447         switch (err) {
1448         case 0:
1449                 vgic_present = true;
1450                 break;
1451         case -ENODEV:
1452         case -ENXIO:
1453                 vgic_present = false;
1454                 err = 0;
1455                 break;
1456         default:
1457                 goto out;
1458         }
1459
1460         /*
1461          * Init HYP architected timer support
1462          */
1463         err = kvm_timer_hyp_init(vgic_present);
1464         if (err)
1465                 goto out;
1466
1467         kvm_perf_init();
1468         kvm_coproc_table_init();
1469
1470 out:
1471         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1472
1473         return err;
1474 }
1475
1476 static void teardown_hyp_mode(void)
1477 {
1478         int cpu;
1479
1480         free_hyp_pgds();
1481         for_each_possible_cpu(cpu)
1482                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1483         hyp_cpu_pm_exit();
1484 }
1485
1486 /**
1487  * Inits Hyp-mode on all online CPUs
1488  */
1489 static int init_hyp_mode(void)
1490 {
1491         int cpu;
1492         int err = 0;
1493
1494         /*
1495          * Allocate Hyp PGD and setup Hyp identity mapping
1496          */
1497         err = kvm_mmu_init();
1498         if (err)
1499                 goto out_err;
1500
1501         /*
1502          * Allocate stack pages for Hypervisor-mode
1503          */
1504         for_each_possible_cpu(cpu) {
1505                 unsigned long stack_page;
1506
1507                 stack_page = __get_free_page(GFP_KERNEL);
1508                 if (!stack_page) {
1509                         err = -ENOMEM;
1510                         goto out_err;
1511                 }
1512
1513                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1514         }
1515
1516         /*
1517          * Map the Hyp-code called directly from the host
1518          */
1519         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1520                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1521         if (err) {
1522                 kvm_err("Cannot map world-switch code\n");
1523                 goto out_err;
1524         }
1525
1526         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1527                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1528         if (err) {
1529                 kvm_err("Cannot map rodata section\n");
1530                 goto out_err;
1531         }
1532
1533         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1534                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1535         if (err) {
1536                 kvm_err("Cannot map bss section\n");
1537                 goto out_err;
1538         }
1539
1540         err = kvm_map_vectors();
1541         if (err) {
1542                 kvm_err("Cannot map vectors\n");
1543                 goto out_err;
1544         }
1545
1546         /*
1547          * Map the Hyp stack pages
1548          */
1549         for_each_possible_cpu(cpu) {
1550                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1551                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1552                                           PAGE_HYP);
1553
1554                 if (err) {
1555                         kvm_err("Cannot map hyp stack\n");
1556                         goto out_err;
1557                 }
1558         }
1559
1560         for_each_possible_cpu(cpu) {
1561                 kvm_cpu_context_t *cpu_ctxt;
1562
1563                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1564                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1565
1566                 if (err) {
1567                         kvm_err("Cannot map host CPU state: %d\n", err);
1568                         goto out_err;
1569                 }
1570         }
1571
1572         err = hyp_map_aux_data();
1573         if (err)
1574                 kvm_err("Cannot map host auxilary data: %d\n", err);
1575
1576         return 0;
1577
1578 out_err:
1579         teardown_hyp_mode();
1580         kvm_err("error initializing Hyp mode: %d\n", err);
1581         return err;
1582 }
1583
1584 static void check_kvm_target_cpu(void *ret)
1585 {
1586         *(int *)ret = kvm_target_cpu();
1587 }
1588
1589 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1590 {
1591         struct kvm_vcpu *vcpu;
1592         int i;
1593
1594         mpidr &= MPIDR_HWID_BITMASK;
1595         kvm_for_each_vcpu(i, vcpu, kvm) {
1596                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1597                         return vcpu;
1598         }
1599         return NULL;
1600 }
1601
1602 bool kvm_arch_has_irq_bypass(void)
1603 {
1604         return true;
1605 }
1606
1607 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1608                                       struct irq_bypass_producer *prod)
1609 {
1610         struct kvm_kernel_irqfd *irqfd =
1611                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1612
1613         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1614                                           &irqfd->irq_entry);
1615 }
1616 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1617                                       struct irq_bypass_producer *prod)
1618 {
1619         struct kvm_kernel_irqfd *irqfd =
1620                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1621
1622         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1623                                      &irqfd->irq_entry);
1624 }
1625
1626 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1627 {
1628         struct kvm_kernel_irqfd *irqfd =
1629                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1630
1631         kvm_arm_halt_guest(irqfd->kvm);
1632 }
1633
1634 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1635 {
1636         struct kvm_kernel_irqfd *irqfd =
1637                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1638
1639         kvm_arm_resume_guest(irqfd->kvm);
1640 }
1641
1642 /**
1643  * Initialize Hyp-mode and memory mappings on all CPUs.
1644  */
1645 int kvm_arch_init(void *opaque)
1646 {
1647         int err;
1648         int ret, cpu;
1649         bool in_hyp_mode;
1650
1651         if (!is_hyp_mode_available()) {
1652                 kvm_info("HYP mode not available\n");
1653                 return -ENODEV;
1654         }
1655
1656         in_hyp_mode = is_kernel_in_hyp_mode();
1657
1658         if (!in_hyp_mode && kvm_arch_requires_vhe()) {
1659                 kvm_pr_unimpl("CPU unsupported in non-VHE mode, not initializing\n");
1660                 return -ENODEV;
1661         }
1662
1663         for_each_online_cpu(cpu) {
1664                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1665                 if (ret < 0) {
1666                         kvm_err("Error, CPU %d not supported!\n", cpu);
1667                         return -ENODEV;
1668                 }
1669         }
1670
1671         err = init_common_resources();
1672         if (err)
1673                 return err;
1674
1675         if (!in_hyp_mode) {
1676                 err = init_hyp_mode();
1677                 if (err)
1678                         goto out_err;
1679         }
1680
1681         err = init_subsystems();
1682         if (err)
1683                 goto out_hyp;
1684
1685         if (in_hyp_mode)
1686                 kvm_info("VHE mode initialized successfully\n");
1687         else
1688                 kvm_info("Hyp mode initialized successfully\n");
1689
1690         return 0;
1691
1692 out_hyp:
1693         if (!in_hyp_mode)
1694                 teardown_hyp_mode();
1695 out_err:
1696         return err;
1697 }
1698
1699 /* NOP: Compiling as a module not supported */
1700 void kvm_arch_exit(void)
1701 {
1702         kvm_perf_teardown();
1703 }
1704
1705 static int arm_init(void)
1706 {
1707         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1708         return rc;
1709 }
1710
1711 module_init(arm_init);