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/cpu_pm.h>
  20 #include <linux/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/list.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <linux/kvm_irqfd.h>
  31 #include <linux/irqbypass.h>
  32 #include <trace/events/kvm.h>
  33 #include <kvm/arm_pmu.h>
  34
  35 #define CREATE_TRACE_POINTS
  36 #include "trace.h"
  37
  38 #include <linux/uaccess.h>
  39 #include <asm/ptrace.h>
  40 #include <asm/mman.h>
  41 #include <asm/tlbflush.h>
  42 #include <asm/cacheflush.h>
  43 #include <asm/virt.h>
  44 #include <asm/kvm_arm.h>
  45 #include <asm/kvm_asm.h>
  46 #include <asm/kvm_mmu.h>
  47 #include <asm/kvm_emulate.h>
  48 #include <asm/kvm_coproc.h>
  49 #include <asm/kvm_psci.h>
  50 #include <asm/sections.h>
  51
  52 #ifdef REQUIRES_VIRT
  53 __asm__(".arch_extension        virt");
  54 #endif
  55
  56 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  57 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  58
  59 /* Per-CPU variable containing the currently running vcpu. */
  60 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  61
  62 /* The VMID used in the VTTBR */
  63 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  64 static u32 kvm_next_vmid;
  65 static unsigned int kvm_vmid_bits __read_mostly;
  66 static DEFINE_SPINLOCK(kvm_vmid_lock);
  67
  68 static bool vgic_present;
  69
  70 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  71
  72 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  73 {
  74         BUG_ON(preemptible());
  75         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  76 }
  77
  78 /**
  79  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  80  * Must be called from non-preemptible context
  81  */
  82 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  83 {
  84         BUG_ON(preemptible());
  85         return __this_cpu_read(kvm_arm_running_vcpu);
  86 }
  87
  88 /**
  89  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  90  */
  91 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  92 {
  93         return &kvm_arm_running_vcpu;
  94 }
  95
  96 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  97 {
  98         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  99 }
 100
 101 int kvm_arch_hardware_setup(void)
 102 {
 103         return 0;
 104 }
 105
 106 void kvm_arch_check_processor_compat(void *rtn)
 107 {
 108         *(int *)rtn = 0;
 109 }
 110
 111
 112 /**
 113  * kvm_arch_init_vm - initializes a VM data structure
 114  * @kvm:        pointer to the KVM struct
 115  */
 116 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 117 {
 118         int ret, cpu;
 119
 120         if (type)
 121                 return -EINVAL;
 122
 123         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 124         if (!kvm->arch.last_vcpu_ran)
 125                 return -ENOMEM;
 126
 127         for_each_possible_cpu(cpu)
 128                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 129
 130         ret = kvm_alloc_stage2_pgd(kvm);
 131         if (ret)
 132                 goto out_fail_alloc;
 133
 134         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 135         if (ret)
 136                 goto out_free_stage2_pgd;
 137
 138         kvm_vgic_early_init(kvm);
 139
 140         /* Mark the initial VMID generation invalid */
 141         kvm->arch.vmid_gen = 0;
 142
 143         /* The maximum number of VCPUs is limited by the host's GIC model */
 144         kvm->arch.max_vcpus = vgic_present ?
 145                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 146
 147         return ret;
 148 out_free_stage2_pgd:
 149         kvm_free_stage2_pgd(kvm);
 150 out_fail_alloc:
 151         free_percpu(kvm->arch.last_vcpu_ran);
 152         kvm->arch.last_vcpu_ran = NULL;
 153         return ret;
 154 }
 155
 156 bool kvm_arch_has_vcpu_debugfs(void)
 157 {
 158         return false;
 159 }
 160
 161 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 162 {
 163         return 0;
 164 }
 165
 166 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 167 {
 168         return VM_FAULT_SIGBUS;
 169 }
 170
 171
 172 /**
 173  * kvm_arch_destroy_vm - destroy the VM data structure
 174  * @kvm:        pointer to the KVM struct
 175  */
 176 void kvm_arch_destroy_vm(struct kvm *kvm)
 177 {
 178         int i;
 179
 180         kvm_vgic_destroy(kvm);
 181
 182         free_percpu(kvm->arch.last_vcpu_ran);
 183         kvm->arch.last_vcpu_ran = NULL;
 184
 185         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 186                 if (kvm->vcpus[i]) {
 187                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 188                         kvm->vcpus[i] = NULL;
 189                 }
 190         }
 191 }
 192
 193 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 194 {
 195         int r;
 196         switch (ext) {
 197         case KVM_CAP_IRQCHIP:
 198                 r = vgic_present;
 199                 break;
 200         case KVM_CAP_IOEVENTFD:
 201         case KVM_CAP_DEVICE_CTRL:
 202         case KVM_CAP_USER_MEMORY:
 203         case KVM_CAP_SYNC_MMU:
 204         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 205         case KVM_CAP_ONE_REG:
 206         case KVM_CAP_ARM_PSCI:
 207         case KVM_CAP_ARM_PSCI_0_2:
 208         case KVM_CAP_READONLY_MEM:
 209         case KVM_CAP_MP_STATE:
 210         case KVM_CAP_IMMEDIATE_EXIT:
 211                 r = 1;
 212                 break;
 213         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 214                 r = 1;
 215                 break;
 216         case KVM_CAP_NR_VCPUS:
 217                 r = num_online_cpus();
 218                 break;
 219         case KVM_CAP_MAX_VCPUS:
 220                 r = KVM_MAX_VCPUS;
 221                 break;
 222         case KVM_CAP_NR_MEMSLOTS:
 223                 r = KVM_USER_MEM_SLOTS;
 224                 break;
 225         case KVM_CAP_MSI_DEVID:
 226                 if (!kvm)
 227                         r = -EINVAL;
 228                 else
 229                         r = kvm->arch.vgic.msis_require_devid;
 230                 break;
 231         case KVM_CAP_ARM_USER_IRQ:
 232                 /*
 233                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 234                  * (bump this number if adding more devices)
 235                  */
 236                 r = 1;
 237                 break;
 238         default:
 239                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 240                 break;
 241         }
 242         return r;
 243 }
 244
 245 long kvm_arch_dev_ioctl(struct file *filp,
 246                         unsigned int ioctl, unsigned long arg)
 247 {
 248         return -EINVAL;
 249 }
 250
 251
 252 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 253 {
 254         int err;
 255         struct kvm_vcpu *vcpu;
 256
 257         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 258                 err = -EBUSY;
 259                 goto out;
 260         }
 261
 262         if (id >= kvm->arch.max_vcpus) {
 263                 err = -EINVAL;
 264                 goto out;
 265         }
 266
 267         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 268         if (!vcpu) {
 269                 err = -ENOMEM;
 270                 goto out;
 271         }
 272
 273         err = kvm_vcpu_init(vcpu, kvm, id);
 274         if (err)
 275                 goto free_vcpu;
 276
 277         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 278         if (err)
 279                 goto vcpu_uninit;
 280
 281         return vcpu;
 282 vcpu_uninit:
 283         kvm_vcpu_uninit(vcpu);
 284 free_vcpu:
 285         kmem_cache_free(kvm_vcpu_cache, vcpu);
 286 out:
 287         return ERR_PTR(err);
 288 }
 289
 290 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 291 {
 292         kvm_vgic_vcpu_early_init(vcpu);
 293 }
 294
 295 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 296 {
 297         kvm_mmu_free_memory_caches(vcpu);
 298         kvm_timer_vcpu_terminate(vcpu);
 299         kvm_vgic_vcpu_destroy(vcpu);
 300         kvm_pmu_vcpu_destroy(vcpu);
 301         kvm_vcpu_uninit(vcpu);
 302         kmem_cache_free(kvm_vcpu_cache, vcpu);
 303 }
 304
 305 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 306 {
 307         kvm_arch_vcpu_free(vcpu);
 308 }
 309
 310 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 311 {
 312         return kvm_timer_is_pending(vcpu);
 313 }
 314
 315 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 316 {
 317         kvm_timer_schedule(vcpu);
 318         kvm_vgic_v4_enable_doorbell(vcpu);
 319 }
 320
 321 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 322 {
 323         kvm_timer_unschedule(vcpu);
 324         kvm_vgic_v4_disable_doorbell(vcpu);
 325 }
 326
 327 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 328 {
 329         /* Force users to call KVM_ARM_VCPU_INIT */
 330         vcpu->arch.target = -1;
 331         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 332
 333         /* Set up the timer */
 334         kvm_timer_vcpu_init(vcpu);
 335
 336         kvm_arm_reset_debug_ptr(vcpu);
 337
 338         return kvm_vgic_vcpu_init(vcpu);
 339 }
 340
 341 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 342 {
 343         int *last_ran;
 344
 345         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 346
 347         /*
 348          * We might get preempted before the vCPU actually runs, but
 349          * over-invalidation doesn't affect correctness.
 350          */
 351         if (*last_ran != vcpu->vcpu_id) {
 352                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 353                 *last_ran = vcpu->vcpu_id;
 354         }
 355
 356         vcpu->cpu = cpu;
 357         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 358
 359         kvm_arm_set_running_vcpu(vcpu);
 360         kvm_vgic_load(vcpu);
 361         kvm_timer_vcpu_load(vcpu);
 362 }
 363
 364 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 365 {
 366         kvm_timer_vcpu_put(vcpu);
 367         kvm_vgic_put(vcpu);
 368
 369         vcpu->cpu = -1;
 370
 371         kvm_arm_set_running_vcpu(NULL);
 372 }
 373
 374 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 375 {
 376         vcpu->arch.power_off = true;
 377         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 378         kvm_vcpu_kick(vcpu);
 379 }
 380
 381 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 382                                     struct kvm_mp_state *mp_state)
 383 {
 384         if (vcpu->arch.power_off)
 385                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 386         else
 387                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 388
 389         return 0;
 390 }
 391
 392 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 393                                     struct kvm_mp_state *mp_state)
 394 {
 395         switch (mp_state->mp_state) {
 396         case KVM_MP_STATE_RUNNABLE:
 397                 vcpu->arch.power_off = false;
 398                 break;
 399         case KVM_MP_STATE_STOPPED:
 400                 vcpu_power_off(vcpu);
 401                 break;
 402         default:
 403                 return -EINVAL;
 404         }
 405
 406         return 0;
 407 }
 408
 409 /**
 410  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 411  * @v:          The VCPU pointer
 412  *
 413  * If the guest CPU is not waiting for interrupts or an interrupt line is
 414  * asserted, the CPU is by definition runnable.
 415  */
 416 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 417 {
 418         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
 419                 && !v->arch.power_off && !v->arch.pause);
 420 }
 421
 422 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 423 {
 424         return vcpu_mode_priv(vcpu);
 425 }
 426
 427 /* Just ensure a guest exit from a particular CPU */
 428 static void exit_vm_noop(void *info)
 429 {
 430 }
 431
 432 void force_vm_exit(const cpumask_t *mask)
 433 {
 434         preempt_disable();
 435         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 436         preempt_enable();
 437 }
 438
 439 /**
 440  * need_new_vmid_gen - check that the VMID is still valid
 441  * @kvm: The VM's VMID to check
 442  *
 443  * return true if there is a new generation of VMIDs being used
 444  *
 445  * The hardware supports only 256 values with the value zero reserved for the
 446  * host, so we check if an assigned value belongs to a previous generation,
 447  * which which requires us to assign a new value. If we're the first to use a
 448  * VMID for the new generation, we must flush necessary caches and TLBs on all
 449  * CPUs.
 450  */
 451 static bool need_new_vmid_gen(struct kvm *kvm)
 452 {
 453         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 454 }
 455
 456 /**
 457  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 458  * @kvm The guest that we are about to run
 459  *
 460  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 461  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 462  * caches and TLBs.
 463  */
 464 static void update_vttbr(struct kvm *kvm)
 465 {
 466         phys_addr_t pgd_phys;
 467         u64 vmid;
 468
 469         if (!need_new_vmid_gen(kvm))
 470                 return;
 471
 472         spin_lock(&kvm_vmid_lock);
 473
 474         /*
 475          * We need to re-check the vmid_gen here to ensure that if another vcpu
 476          * already allocated a valid vmid for this vm, then this vcpu should
 477          * use the same vmid.
 478          */
 479         if (!need_new_vmid_gen(kvm)) {
 480                 spin_unlock(&kvm_vmid_lock);
 481                 return;
 482         }
 483
 484         /* First user of a new VMID generation? */
 485         if (unlikely(kvm_next_vmid == 0)) {
 486                 atomic64_inc(&kvm_vmid_gen);
 487                 kvm_next_vmid = 1;
 488
 489                 /*
 490                  * On SMP we know no other CPUs can use this CPU's or each
 491                  * other's VMID after force_vm_exit returns since the
 492                  * kvm_vmid_lock blocks them from reentry to the guest.
 493                  */
 494                 force_vm_exit(cpu_all_mask);
 495                 /*
 496                  * Now broadcast TLB + ICACHE invalidation over the inner
 497                  * shareable domain to make sure all data structures are
 498                  * clean.
 499                  */
 500                 kvm_call_hyp(__kvm_flush_vm_context);
 501         }
 502
 503         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 504         kvm->arch.vmid = kvm_next_vmid;
 505         kvm_next_vmid++;
 506         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 507
 508         /* update vttbr to be used with the new vmid */
 509         pgd_phys = virt_to_phys(kvm->arch.pgd);
 510         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 511         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 512         kvm->arch.vttbr = pgd_phys | vmid;
 513
 514         spin_unlock(&kvm_vmid_lock);
 515 }
 516
 517 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 518 {
 519         struct kvm *kvm = vcpu->kvm;
 520         int ret = 0;
 521
 522         if (likely(vcpu->arch.has_run_once))
 523                 return 0;
 524
 525         vcpu->arch.has_run_once = true;
 526
 527         /*
 528          * Map the VGIC hardware resources before running a vcpu the first
 529          * time on this VM.
 530          */
 531         if (unlikely(irqchip_in_kernel(kvm) && !vgic_ready(kvm))) {
 532                 ret = kvm_vgic_map_resources(kvm);
 533                 if (ret)
 534                         return ret;
 535         }
 536
 537         ret = kvm_timer_enable(vcpu);
 538         if (ret)
 539                 return ret;
 540
 541         ret = kvm_arm_pmu_v3_enable(vcpu);
 542
 543         return ret;
 544 }
 545
 546 bool kvm_arch_intc_initialized(struct kvm *kvm)
 547 {
 548         return vgic_initialized(kvm);
 549 }
 550
 551 void kvm_arm_halt_guest(struct kvm *kvm)
 552 {
 553         int i;
 554         struct kvm_vcpu *vcpu;
 555
 556         kvm_for_each_vcpu(i, vcpu, kvm)
 557                 vcpu->arch.pause = true;
 558         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 559 }
 560
 561 void kvm_arm_resume_guest(struct kvm *kvm)
 562 {
 563         int i;
 564         struct kvm_vcpu *vcpu;
 565
 566         kvm_for_each_vcpu(i, vcpu, kvm) {
 567                 vcpu->arch.pause = false;
 568                 swake_up(kvm_arch_vcpu_wq(vcpu));
 569         }
 570 }
 571
 572 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 573 {
 574         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 575
 576         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 577                                        (!vcpu->arch.pause)));
 578
 579         if (vcpu->arch.power_off || vcpu->arch.pause) {
 580                 /* Awaken to handle a signal, request we sleep again later. */
 581                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 582         }
 583 }
 584
 585 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 586 {
 587         return vcpu->arch.target >= 0;
 588 }
 589
 590 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 591 {
 592         if (kvm_request_pending(vcpu)) {
 593                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 594                         vcpu_req_sleep(vcpu);
 595
 596                 /*
 597                  * Clear IRQ_PENDING requests that were made to guarantee
 598                  * that a VCPU sees new virtual interrupts.
 599                  */
 600                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 601         }
 602 }
 603
 604 /**
 605  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 606  * @vcpu:       The VCPU pointer
 607  * @run:        The kvm_run structure pointer used for userspace state exchange
 608  *
 609  * This function is called through the VCPU_RUN ioctl called from user space. It
 610  * will execute VM code in a loop until the time slice for the process is used
 611  * or some emulation is needed from user space in which case the function will
 612  * return with return value 0 and with the kvm_run structure filled in with the
 613  * required data for the requested emulation.
 614  */
 615 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 616 {
 617         int ret;
 618         sigset_t sigsaved;
 619
 620         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 621                 return -ENOEXEC;
 622
 623         ret = kvm_vcpu_first_run_init(vcpu);
 624         if (ret)
 625                 return ret;
 626
 627         if (run->exit_reason == KVM_EXIT_MMIO) {
 628                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 629                 if (ret)
 630                         return ret;
 631         }
 632
 633         if (run->immediate_exit)
 634                 return -EINTR;
 635
 636         if (vcpu->sigset_active)
 637                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 638
 639         ret = 1;
 640         run->exit_reason = KVM_EXIT_UNKNOWN;
 641         while (ret > 0) {
 642                 /*
 643                  * Check conditions before entering the guest
 644                  */
 645                 cond_resched();
 646
 647                 update_vttbr(vcpu->kvm);
 648
 649                 check_vcpu_requests(vcpu);
 650
 651                 /*
 652                  * Preparing the interrupts to be injected also
 653                  * involves poking the GIC, which must be done in a
 654                  * non-preemptible context.
 655                  */
 656                 preempt_disable();
 657
 658                 /* Flush FP/SIMD state that can't survive guest entry/exit */
 659                 kvm_fpsimd_flush_cpu_state();
 660
 661                 kvm_pmu_flush_hwstate(vcpu);
 662
 663                 local_irq_disable();
 664
 665                 kvm_vgic_flush_hwstate(vcpu);
 666
 667                 /*
 668                  * If we have a singal pending, or need to notify a userspace
 669                  * irqchip about timer or PMU level changes, then we exit (and
 670                  * update the timer level state in kvm_timer_update_run
 671                  * below).
 672                  */
 673                 if (signal_pending(current) ||
 674                     kvm_timer_should_notify_user(vcpu) ||
 675                     kvm_pmu_should_notify_user(vcpu)) {
 676                         ret = -EINTR;
 677                         run->exit_reason = KVM_EXIT_INTR;
 678                 }
 679
 680                 /*
 681                  * Ensure we set mode to IN_GUEST_MODE after we disable
 682                  * interrupts and before the final VCPU requests check.
 683                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 684                  * Documentation/virtual/kvm/vcpu-requests.rst
 685                  */
 686                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 687
 688                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 689                     kvm_request_pending(vcpu)) {
 690                         vcpu->mode = OUTSIDE_GUEST_MODE;
 691                         kvm_pmu_sync_hwstate(vcpu);
 692                         kvm_timer_sync_hwstate(vcpu);
 693                         kvm_vgic_sync_hwstate(vcpu);
 694                         local_irq_enable();
 695                         preempt_enable();
 696                         continue;
 697                 }
 698
 699                 kvm_arm_setup_debug(vcpu);
 700
 701                 /**************************************************************
 702                  * Enter the guest
 703                  */
 704                 trace_kvm_entry(*vcpu_pc(vcpu));
 705                 guest_enter_irqoff();
 706
 707                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 708
 709                 vcpu->mode = OUTSIDE_GUEST_MODE;
 710                 vcpu->stat.exits++;
 711                 /*
 712                  * Back from guest
 713                  *************************************************************/
 714
 715                 kvm_arm_clear_debug(vcpu);
 716
 717                 /*
 718                  * We must sync the PMU state before the vgic state so
 719                  * that the vgic can properly sample the updated state of the
 720                  * interrupt line.
 721                  */
 722                 kvm_pmu_sync_hwstate(vcpu);
 723
 724                 /*
 725                  * Sync the vgic state before syncing the timer state because
 726                  * the timer code needs to know if the virtual timer
 727                  * interrupts are active.
 728                  */
 729                 kvm_vgic_sync_hwstate(vcpu);
 730
 731                 /*
 732                  * Sync the timer hardware state before enabling interrupts as
 733                  * we don't want vtimer interrupts to race with syncing the
 734                  * timer virtual interrupt state.
 735                  */
 736                 kvm_timer_sync_hwstate(vcpu);
 737
 738                 /*
 739                  * We may have taken a host interrupt in HYP mode (ie
 740                  * while executing the guest). This interrupt is still
 741                  * pending, as we haven't serviced it yet!
 742                  *
 743                  * We're now back in SVC mode, with interrupts
 744                  * disabled.  Enabling the interrupts now will have
 745                  * the effect of taking the interrupt again, in SVC
 746                  * mode this time.
 747                  */
 748                 local_irq_enable();
 749
 750                 /*
 751                  * We do local_irq_enable() before calling guest_exit() so
 752                  * that if a timer interrupt hits while running the guest we
 753                  * account that tick as being spent in the guest.  We enable
 754                  * preemption after calling guest_exit() so that if we get
 755                  * preempted we make sure ticks after that is not counted as
 756                  * guest time.
 757                  */
 758                 guest_exit();
 759                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 760
 761                 preempt_enable();
 762
 763                 ret = handle_exit(vcpu, run, ret);
 764         }
 765
 766         /* Tell userspace about in-kernel device output levels */
 767         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 768                 kvm_timer_update_run(vcpu);
 769                 kvm_pmu_update_run(vcpu);
 770         }
 771
 772         if (vcpu->sigset_active)
 773                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 774         return ret;
 775 }
 776
 777 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 778 {
 779         int bit_index;
 780         bool set;
 781         unsigned long *ptr;
 782
 783         if (number == KVM_ARM_IRQ_CPU_IRQ)
 784                 bit_index = __ffs(HCR_VI);
 785         else /* KVM_ARM_IRQ_CPU_FIQ */
 786                 bit_index = __ffs(HCR_VF);
 787
 788         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 789         if (level)
 790                 set = test_and_set_bit(bit_index, ptr);
 791         else
 792                 set = test_and_clear_bit(bit_index, ptr);
 793
 794         /*
 795          * If we didn't change anything, no need to wake up or kick other CPUs
 796          */
 797         if (set == level)
 798                 return 0;
 799
 800         /*
 801          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 802          * trigger a world-switch round on the running physical CPU to set the
 803          * virtual IRQ/FIQ fields in the HCR appropriately.
 804          */
 805         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 806         kvm_vcpu_kick(vcpu);
 807
 808         return 0;
 809 }
 810
 811 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 812                           bool line_status)
 813 {
 814         u32 irq = irq_level->irq;
 815         unsigned int irq_type, vcpu_idx, irq_num;
 816         int nrcpus = atomic_read(&kvm->online_vcpus);
 817         struct kvm_vcpu *vcpu = NULL;
 818         bool level = irq_level->level;
 819
 820         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 821         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 822         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 823
 824         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 825
 826         switch (irq_type) {
 827         case KVM_ARM_IRQ_TYPE_CPU:
 828                 if (irqchip_in_kernel(kvm))
 829                         return -ENXIO;
 830
 831                 if (vcpu_idx >= nrcpus)
 832                         return -EINVAL;
 833
 834                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 835                 if (!vcpu)
 836                         return -EINVAL;
 837
 838                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 839                         return -EINVAL;
 840
 841                 return vcpu_interrupt_line(vcpu, irq_num, level);
 842         case KVM_ARM_IRQ_TYPE_PPI:
 843                 if (!irqchip_in_kernel(kvm))
 844                         return -ENXIO;
 845
 846                 if (vcpu_idx >= nrcpus)
 847                         return -EINVAL;
 848
 849                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 850                 if (!vcpu)
 851                         return -EINVAL;
 852
 853                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 854                         return -EINVAL;
 855
 856                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 857         case KVM_ARM_IRQ_TYPE_SPI:
 858                 if (!irqchip_in_kernel(kvm))
 859                         return -ENXIO;
 860
 861                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 862                         return -EINVAL;
 863
 864                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 865         }
 866
 867         return -EINVAL;
 868 }
 869
 870 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 871                                const struct kvm_vcpu_init *init)
 872 {
 873         unsigned int i;
 874         int phys_target = kvm_target_cpu();
 875
 876         if (init->target != phys_target)
 877                 return -EINVAL;
 878
 879         /*
 880          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 881          * use the same target.
 882          */
 883         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 884                 return -EINVAL;
 885
 886         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 887         for (i = 0; i < sizeof(init->features) * 8; i++) {
 888                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 889
 890                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 891                         return -ENOENT;
 892
 893                 /*
 894                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 895                  * use the same feature set.
 896                  */
 897                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 898                     test_bit(i, vcpu->arch.features) != set)
 899                         return -EINVAL;
 900
 901                 if (set)
 902                         set_bit(i, vcpu->arch.features);
 903         }
 904
 905         vcpu->arch.target = phys_target;
 906
 907         /* Now we know what it is, we can reset it. */
 908         return kvm_reset_vcpu(vcpu);
 909 }
 910
 911
 912 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 913                                          struct kvm_vcpu_init *init)
 914 {
 915         int ret;
 916
 917         ret = kvm_vcpu_set_target(vcpu, init);
 918         if (ret)
 919                 return ret;
 920
 921         /*
 922          * Ensure a rebooted VM will fault in RAM pages and detect if the
 923          * guest MMU is turned off and flush the caches as needed.
 924          */
 925         if (vcpu->arch.has_run_once)
 926                 stage2_unmap_vm(vcpu->kvm);
 927
 928         vcpu_reset_hcr(vcpu);
 929
 930         /*
 931          * Handle the "start in power-off" case.
 932          */
 933         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 934                 vcpu_power_off(vcpu);
 935         else
 936                 vcpu->arch.power_off = false;
 937
 938         return 0;
 939 }
 940
 941 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 942                                  struct kvm_device_attr *attr)
 943 {
 944         int ret = -ENXIO;
 945
 946         switch (attr->group) {
 947         default:
 948                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 949                 break;
 950         }
 951
 952         return ret;
 953 }
 954
 955 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 956                                  struct kvm_device_attr *attr)
 957 {
 958         int ret = -ENXIO;
 959
 960         switch (attr->group) {
 961         default:
 962                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
 963                 break;
 964         }
 965
 966         return ret;
 967 }
 968
 969 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
 970                                  struct kvm_device_attr *attr)
 971 {
 972         int ret = -ENXIO;
 973
 974         switch (attr->group) {
 975         default:
 976                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
 977                 break;
 978         }
 979
 980         return ret;
 981 }
 982
 983 long kvm_arch_vcpu_ioctl(struct file *filp,
 984                          unsigned int ioctl, unsigned long arg)
 985 {
 986         struct kvm_vcpu *vcpu = filp->private_data;
 987         void __user *argp = (void __user *)arg;
 988         struct kvm_device_attr attr;
 989
 990         switch (ioctl) {
 991         case KVM_ARM_VCPU_INIT: {
 992                 struct kvm_vcpu_init init;
 993
 994                 if (copy_from_user(&init, argp, sizeof(init)))
 995                         return -EFAULT;
 996
 997                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 998         }
 999         case KVM_SET_ONE_REG:
1000         case KVM_GET_ONE_REG: {
1001                 struct kvm_one_reg reg;
1002
1003                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1004                         return -ENOEXEC;
1005
1006                 if (copy_from_user(&reg, argp, sizeof(reg)))
1007                         return -EFAULT;
1008                 if (ioctl == KVM_SET_ONE_REG)
1009                         return kvm_arm_set_reg(vcpu, &reg);
1010                 else
1011                         return kvm_arm_get_reg(vcpu, &reg);
1012         }
1013         case KVM_GET_REG_LIST: {
1014                 struct kvm_reg_list __user *user_list = argp;
1015                 struct kvm_reg_list reg_list;
1016                 unsigned n;
1017
1018                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1019                         return -ENOEXEC;
1020
1021                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1022                         return -EFAULT;
1023                 n = reg_list.n;
1024                 reg_list.n = kvm_arm_num_regs(vcpu);
1025                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1026                         return -EFAULT;
1027                 if (n < reg_list.n)
1028                         return -E2BIG;
1029                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1030         }
1031         case KVM_SET_DEVICE_ATTR: {
1032                 if (copy_from_user(&attr, argp, sizeof(attr)))
1033                         return -EFAULT;
1034                 return kvm_arm_vcpu_set_attr(vcpu, &attr);
1035         }
1036         case KVM_GET_DEVICE_ATTR: {
1037                 if (copy_from_user(&attr, argp, sizeof(attr)))
1038                         return -EFAULT;
1039                 return kvm_arm_vcpu_get_attr(vcpu, &attr);
1040         }
1041         case KVM_HAS_DEVICE_ATTR: {
1042                 if (copy_from_user(&attr, argp, sizeof(attr)))
1043                         return -EFAULT;
1044                 return kvm_arm_vcpu_has_attr(vcpu, &attr);
1045         }
1046         default:
1047                 return -EINVAL;
1048         }
1049 }
1050
1051 /**
1052  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1053  * @kvm: kvm instance
1054  * @log: slot id and address to which we copy the log
1055  *
1056  * Steps 1-4 below provide general overview of dirty page logging. See
1057  * kvm_get_dirty_log_protect() function description for additional details.
1058  *
1059  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1060  * always flush the TLB (step 4) even if previous step failed  and the dirty
1061  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1062  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1063  * writes will be marked dirty for next log read.
1064  *
1065  *   1. Take a snapshot of the bit and clear it if needed.
1066  *   2. Write protect the corresponding page.
1067  *   3. Copy the snapshot to the userspace.
1068  *   4. Flush TLB's if needed.
1069  */
1070 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1071 {
1072         bool is_dirty = false;
1073         int r;
1074
1075         mutex_lock(&kvm->slots_lock);
1076
1077         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1078
1079         if (is_dirty)
1080                 kvm_flush_remote_tlbs(kvm);
1081
1082         mutex_unlock(&kvm->slots_lock);
1083         return r;
1084 }
1085
1086 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1087                                         struct kvm_arm_device_addr *dev_addr)
1088 {
1089         unsigned long dev_id, type;
1090
1091         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1092                 KVM_ARM_DEVICE_ID_SHIFT;
1093         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1094                 KVM_ARM_DEVICE_TYPE_SHIFT;
1095
1096         switch (dev_id) {
1097         case KVM_ARM_DEVICE_VGIC_V2:
1098                 if (!vgic_present)
1099                         return -ENXIO;
1100                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1101         default:
1102                 return -ENODEV;
1103         }
1104 }
1105
1106 long kvm_arch_vm_ioctl(struct file *filp,
1107                        unsigned int ioctl, unsigned long arg)
1108 {
1109         struct kvm *kvm = filp->private_data;
1110         void __user *argp = (void __user *)arg;
1111
1112         switch (ioctl) {
1113         case KVM_CREATE_IRQCHIP: {
1114                 int ret;
1115                 if (!vgic_present)
1116                         return -ENXIO;
1117                 mutex_lock(&kvm->lock);
1118                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1119                 mutex_unlock(&kvm->lock);
1120                 return ret;
1121         }
1122         case KVM_ARM_SET_DEVICE_ADDR: {
1123                 struct kvm_arm_device_addr dev_addr;
1124
1125                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1126                         return -EFAULT;
1127                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1128         }
1129         case KVM_ARM_PREFERRED_TARGET: {
1130                 int err;
1131                 struct kvm_vcpu_init init;
1132
1133                 err = kvm_vcpu_preferred_target(&init);
1134                 if (err)
1135                         return err;
1136
1137                 if (copy_to_user(argp, &init, sizeof(init)))
1138                         return -EFAULT;
1139
1140                 return 0;
1141         }
1142         default:
1143                 return -EINVAL;
1144         }
1145 }
1146
1147 static void cpu_init_hyp_mode(void *dummy)
1148 {
1149         phys_addr_t pgd_ptr;
1150         unsigned long hyp_stack_ptr;
1151         unsigned long stack_page;
1152         unsigned long vector_ptr;
1153
1154         /* Switch from the HYP stub to our own HYP init vector */
1155         __hyp_set_vectors(kvm_get_idmap_vector());
1156
1157         pgd_ptr = kvm_mmu_get_httbr();
1158         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1159         hyp_stack_ptr = stack_page + PAGE_SIZE;
1160         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1161
1162         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1163         __cpu_init_stage2();
1164
1165         kvm_arm_init_debug();
1166 }
1167
1168 static void cpu_hyp_reset(void)
1169 {
1170         if (!is_kernel_in_hyp_mode())
1171                 __hyp_reset_vectors();
1172 }
1173
1174 static void cpu_hyp_reinit(void)
1175 {
1176         cpu_hyp_reset();
1177
1178         if (is_kernel_in_hyp_mode()) {
1179                 /*
1180                  * __cpu_init_stage2() is safe to call even if the PM
1181                  * event was cancelled before the CPU was reset.
1182                  */
1183                 __cpu_init_stage2();
1184                 kvm_timer_init_vhe();
1185         } else {
1186                 cpu_init_hyp_mode(NULL);
1187         }
1188
1189         if (vgic_present)
1190                 kvm_vgic_init_cpu_hardware();
1191 }
1192
1193 static void _kvm_arch_hardware_enable(void *discard)
1194 {
1195         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1196                 cpu_hyp_reinit();
1197                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1198         }
1199 }
1200
1201 int kvm_arch_hardware_enable(void)
1202 {
1203         _kvm_arch_hardware_enable(NULL);
1204         return 0;
1205 }
1206
1207 static void _kvm_arch_hardware_disable(void *discard)
1208 {
1209         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1210                 cpu_hyp_reset();
1211                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1212         }
1213 }
1214
1215 void kvm_arch_hardware_disable(void)
1216 {
1217         _kvm_arch_hardware_disable(NULL);
1218 }
1219
1220 #ifdef CONFIG_CPU_PM
1221 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1222                                     unsigned long cmd,
1223                                     void *v)
1224 {
1225         /*
1226          * kvm_arm_hardware_enabled is left with its old value over
1227          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1228          * re-enable hyp.
1229          */
1230         switch (cmd) {
1231         case CPU_PM_ENTER:
1232                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1233                         /*
1234                          * don't update kvm_arm_hardware_enabled here
1235                          * so that the hardware will be re-enabled
1236                          * when we resume. See below.
1237                          */
1238                         cpu_hyp_reset();
1239
1240                 return NOTIFY_OK;
1241         case CPU_PM_EXIT:
1242                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1243                         /* The hardware was enabled before suspend. */
1244                         cpu_hyp_reinit();
1245
1246                 return NOTIFY_OK;
1247
1248         default:
1249                 return NOTIFY_DONE;
1250         }
1251 }
1252
1253 static struct notifier_block hyp_init_cpu_pm_nb = {
1254         .notifier_call = hyp_init_cpu_pm_notifier,
1255 };
1256
1257 static void __init hyp_cpu_pm_init(void)
1258 {
1259         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1260 }
1261 static void __init hyp_cpu_pm_exit(void)
1262 {
1263         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1264 }
1265 #else
1266 static inline void hyp_cpu_pm_init(void)
1267 {
1268 }
1269 static inline void hyp_cpu_pm_exit(void)
1270 {
1271 }
1272 #endif
1273
1274 static void teardown_common_resources(void)
1275 {
1276         free_percpu(kvm_host_cpu_state);
1277 }
1278
1279 static int init_common_resources(void)
1280 {
1281         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1282         if (!kvm_host_cpu_state) {
1283                 kvm_err("Cannot allocate host CPU state\n");
1284                 return -ENOMEM;
1285         }
1286
1287         /* set size of VMID supported by CPU */
1288         kvm_vmid_bits = kvm_get_vmid_bits();
1289         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1290
1291         return 0;
1292 }
1293
1294 static int init_subsystems(void)
1295 {
1296         int err = 0;
1297
1298         /*
1299          * Enable hardware so that subsystem initialisation can access EL2.
1300          */
1301         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1302
1303         /*
1304          * Register CPU lower-power notifier
1305          */
1306         hyp_cpu_pm_init();
1307
1308         /*
1309          * Init HYP view of VGIC
1310          */
1311         err = kvm_vgic_hyp_init();
1312         switch (err) {
1313         case 0:
1314                 vgic_present = true;
1315                 break;
1316         case -ENODEV:
1317         case -ENXIO:
1318                 vgic_present = false;
1319                 err = 0;
1320                 break;
1321         default:
1322                 goto out;
1323         }
1324
1325         /*
1326          * Init HYP architected timer support
1327          */
1328         err = kvm_timer_hyp_init();
1329         if (err)
1330                 goto out;
1331
1332         kvm_perf_init();
1333         kvm_coproc_table_init();
1334
1335 out:
1336         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1337
1338         return err;
1339 }
1340
1341 static void teardown_hyp_mode(void)
1342 {
1343         int cpu;
1344
1345         free_hyp_pgds();
1346         for_each_possible_cpu(cpu)
1347                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1348         hyp_cpu_pm_exit();
1349 }
1350
1351 /**
1352  * Inits Hyp-mode on all online CPUs
1353  */
1354 static int init_hyp_mode(void)
1355 {
1356         int cpu;
1357         int err = 0;
1358
1359         /*
1360          * Allocate Hyp PGD and setup Hyp identity mapping
1361          */
1362         err = kvm_mmu_init();
1363         if (err)
1364                 goto out_err;
1365
1366         /*
1367          * Allocate stack pages for Hypervisor-mode
1368          */
1369         for_each_possible_cpu(cpu) {
1370                 unsigned long stack_page;
1371
1372                 stack_page = __get_free_page(GFP_KERNEL);
1373                 if (!stack_page) {
1374                         err = -ENOMEM;
1375                         goto out_err;
1376                 }
1377
1378                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1379         }
1380
1381         /*
1382          * Map the Hyp-code called directly from the host
1383          */
1384         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1385                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1386         if (err) {
1387                 kvm_err("Cannot map world-switch code\n");
1388                 goto out_err;
1389         }
1390
1391         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1392                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1393         if (err) {
1394                 kvm_err("Cannot map rodata section\n");
1395                 goto out_err;
1396         }
1397
1398         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1399                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1400         if (err) {
1401                 kvm_err("Cannot map bss section\n");
1402                 goto out_err;
1403         }
1404
1405         /*
1406          * Map the Hyp stack pages
1407          */
1408         for_each_possible_cpu(cpu) {
1409                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1410                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1411                                           PAGE_HYP);
1412
1413                 if (err) {
1414                         kvm_err("Cannot map hyp stack\n");
1415                         goto out_err;
1416                 }
1417         }
1418
1419         for_each_possible_cpu(cpu) {
1420                 kvm_cpu_context_t *cpu_ctxt;
1421
1422                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1423                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1424
1425                 if (err) {
1426                         kvm_err("Cannot map host CPU state: %d\n", err);
1427                         goto out_err;
1428                 }
1429         }
1430
1431         return 0;
1432
1433 out_err:
1434         teardown_hyp_mode();
1435         kvm_err("error initializing Hyp mode: %d\n", err);
1436         return err;
1437 }
1438
1439 static void check_kvm_target_cpu(void *ret)
1440 {
1441         *(int *)ret = kvm_target_cpu();
1442 }
1443
1444 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1445 {
1446         struct kvm_vcpu *vcpu;
1447         int i;
1448
1449         mpidr &= MPIDR_HWID_BITMASK;
1450         kvm_for_each_vcpu(i, vcpu, kvm) {
1451                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1452                         return vcpu;
1453         }
1454         return NULL;
1455 }
1456
1457 bool kvm_arch_has_irq_bypass(void)
1458 {
1459         return true;
1460 }
1461
1462 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1463                                       struct irq_bypass_producer *prod)
1464 {
1465         struct kvm_kernel_irqfd *irqfd =
1466                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1467
1468         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1469                                           &irqfd->irq_entry);
1470 }
1471 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1472                                       struct irq_bypass_producer *prod)
1473 {
1474         struct kvm_kernel_irqfd *irqfd =
1475                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1476
1477         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1478                                      &irqfd->irq_entry);
1479 }
1480
1481 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1482 {
1483         struct kvm_kernel_irqfd *irqfd =
1484                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1485
1486         kvm_arm_halt_guest(irqfd->kvm);
1487 }
1488
1489 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1490 {
1491         struct kvm_kernel_irqfd *irqfd =
1492                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1493
1494         kvm_arm_resume_guest(irqfd->kvm);
1495 }
1496
1497 /**
1498  * Initialize Hyp-mode and memory mappings on all CPUs.
1499  */
1500 int kvm_arch_init(void *opaque)
1501 {
1502         int err;
1503         int ret, cpu;
1504         bool in_hyp_mode;
1505
1506         if (!is_hyp_mode_available()) {
1507                 kvm_err("HYP mode not available\n");
1508                 return -ENODEV;
1509         }
1510
1511         for_each_online_cpu(cpu) {
1512                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1513                 if (ret < 0) {
1514                         kvm_err("Error, CPU %d not supported!\n", cpu);
1515                         return -ENODEV;
1516                 }
1517         }
1518
1519         err = init_common_resources();
1520         if (err)
1521                 return err;
1522
1523         in_hyp_mode = is_kernel_in_hyp_mode();
1524
1525         if (!in_hyp_mode) {
1526                 err = init_hyp_mode();
1527                 if (err)
1528                         goto out_err;
1529         }
1530
1531         err = init_subsystems();
1532         if (err)
1533                 goto out_hyp;
1534
1535         if (in_hyp_mode)
1536                 kvm_info("VHE mode initialized successfully\n");
1537         else
1538                 kvm_info("Hyp mode initialized successfully\n");
1539
1540         return 0;
1541
1542 out_hyp:
1543         if (!in_hyp_mode)
1544                 teardown_hyp_mode();
1545 out_err:
1546         teardown_common_resources();
1547         return err;
1548 }
1549
1550 /* NOP: Compiling as a module not supported */
1551 void kvm_arch_exit(void)
1552 {
1553         kvm_perf_teardown();
1554 }
1555
1556 static int arm_init(void)
1557 {
1558         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1559         return rc;
1560 }
1561
1562 module_init(arm_init);