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         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  75 }
  76
  77 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  78
  79 /**
  80  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  81  * Must be called from non-preemptible context
  82  */
  83 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  84 {
  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         atomic_set(&kvm->online_vcpus, 0);
 192 }
 193
 194 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 195 {
 196         int r;
 197         switch (ext) {
 198         case KVM_CAP_IRQCHIP:
 199                 r = vgic_present;
 200                 break;
 201         case KVM_CAP_IOEVENTFD:
 202         case KVM_CAP_DEVICE_CTRL:
 203         case KVM_CAP_USER_MEMORY:
 204         case KVM_CAP_SYNC_MMU:
 205         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 206         case KVM_CAP_ONE_REG:
 207         case KVM_CAP_ARM_PSCI:
 208         case KVM_CAP_ARM_PSCI_0_2:
 209         case KVM_CAP_READONLY_MEM:
 210         case KVM_CAP_MP_STATE:
 211         case KVM_CAP_IMMEDIATE_EXIT:
 212                 r = 1;
 213                 break;
 214         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 215                 r = 1;
 216                 break;
 217         case KVM_CAP_NR_VCPUS:
 218                 r = num_online_cpus();
 219                 break;
 220         case KVM_CAP_MAX_VCPUS:
 221                 r = KVM_MAX_VCPUS;
 222                 break;
 223         case KVM_CAP_NR_MEMSLOTS:
 224                 r = KVM_USER_MEM_SLOTS;
 225                 break;
 226         case KVM_CAP_MSI_DEVID:
 227                 if (!kvm)
 228                         r = -EINVAL;
 229                 else
 230                         r = kvm->arch.vgic.msis_require_devid;
 231                 break;
 232         case KVM_CAP_ARM_USER_IRQ:
 233                 /*
 234                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 235                  * (bump this number if adding more devices)
 236                  */
 237                 r = 1;
 238                 break;
 239         default:
 240                 r = kvm_arch_dev_ioctl_check_extension(kvm, ext);
 241                 break;
 242         }
 243         return r;
 244 }
 245
 246 long kvm_arch_dev_ioctl(struct file *filp,
 247                         unsigned int ioctl, unsigned long arg)
 248 {
 249         return -EINVAL;
 250 }
 251
 252
 253 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 254 {
 255         int err;
 256         struct kvm_vcpu *vcpu;
 257
 258         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 259                 err = -EBUSY;
 260                 goto out;
 261         }
 262
 263         if (id >= kvm->arch.max_vcpus) {
 264                 err = -EINVAL;
 265                 goto out;
 266         }
 267
 268         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 269         if (!vcpu) {
 270                 err = -ENOMEM;
 271                 goto out;
 272         }
 273
 274         err = kvm_vcpu_init(vcpu, kvm, id);
 275         if (err)
 276                 goto free_vcpu;
 277
 278         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 279         if (err)
 280                 goto vcpu_uninit;
 281
 282         return vcpu;
 283 vcpu_uninit:
 284         kvm_vcpu_uninit(vcpu);
 285 free_vcpu:
 286         kmem_cache_free(kvm_vcpu_cache, vcpu);
 287 out:
 288         return ERR_PTR(err);
 289 }
 290
 291 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 292 {
 293         kvm_vgic_vcpu_early_init(vcpu);
 294 }
 295
 296 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 297 {
 298         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 299                 static_branch_dec(&userspace_irqchip_in_use);
 300
 301         kvm_mmu_free_memory_caches(vcpu);
 302         kvm_timer_vcpu_terminate(vcpu);
 303         kvm_pmu_vcpu_destroy(vcpu);
 304         kvm_vcpu_uninit(vcpu);
 305         kmem_cache_free(kvm_vcpu_cache, vcpu);
 306 }
 307
 308 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 309 {
 310         kvm_arch_vcpu_free(vcpu);
 311 }
 312
 313 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 314 {
 315         return kvm_timer_is_pending(vcpu);
 316 }
 317
 318 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 319 {
 320         kvm_timer_schedule(vcpu);
 321         kvm_vgic_v4_enable_doorbell(vcpu);
 322 }
 323
 324 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 325 {
 326         kvm_timer_unschedule(vcpu);
 327         kvm_vgic_v4_disable_doorbell(vcpu);
 328 }
 329
 330 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 331 {
 332         /* Force users to call KVM_ARM_VCPU_INIT */
 333         vcpu->arch.target = -1;
 334         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 335
 336         /* Set up the timer */
 337         kvm_timer_vcpu_init(vcpu);
 338
 339         kvm_arm_reset_debug_ptr(vcpu);
 340
 341         return kvm_vgic_vcpu_init(vcpu);
 342 }
 343
 344 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 345 {
 346         int *last_ran;
 347
 348         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 349
 350         /*
 351          * We might get preempted before the vCPU actually runs, but
 352          * over-invalidation doesn't affect correctness.
 353          */
 354         if (*last_ran != vcpu->vcpu_id) {
 355                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 356                 *last_ran = vcpu->vcpu_id;
 357         }
 358
 359         vcpu->cpu = cpu;
 360         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 361
 362         kvm_arm_set_running_vcpu(vcpu);
 363         kvm_vgic_load(vcpu);
 364         kvm_timer_vcpu_load(vcpu);
 365 }
 366
 367 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 368 {
 369         kvm_timer_vcpu_put(vcpu);
 370         kvm_vgic_put(vcpu);
 371
 372         vcpu->cpu = -1;
 373
 374         kvm_arm_set_running_vcpu(NULL);
 375 }
 376
 377 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 378 {
 379         vcpu->arch.power_off = true;
 380         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 381         kvm_vcpu_kick(vcpu);
 382 }
 383
 384 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 385                                     struct kvm_mp_state *mp_state)
 386 {
 387         vcpu_load(vcpu);
 388
 389         if (vcpu->arch.power_off)
 390                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 391         else
 392                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 393
 394         vcpu_put(vcpu);
 395         return 0;
 396 }
 397
 398 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 399                                     struct kvm_mp_state *mp_state)
 400 {
 401         int ret = 0;
 402
 403         vcpu_load(vcpu);
 404
 405         switch (mp_state->mp_state) {
 406         case KVM_MP_STATE_RUNNABLE:
 407                 vcpu->arch.power_off = false;
 408                 break;
 409         case KVM_MP_STATE_STOPPED:
 410                 vcpu_power_off(vcpu);
 411                 break;
 412         default:
 413                 ret = -EINVAL;
 414         }
 415
 416         vcpu_put(vcpu);
 417         return ret;
 418 }
 419
 420 /**
 421  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 422  * @v:          The VCPU pointer
 423  *
 424  * If the guest CPU is not waiting for interrupts or an interrupt line is
 425  * asserted, the CPU is by definition runnable.
 426  */
 427 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 428 {
 429         return ((!!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v))
 430                 && !v->arch.power_off && !v->arch.pause);
 431 }
 432
 433 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 434 {
 435         return vcpu_mode_priv(vcpu);
 436 }
 437
 438 /* Just ensure a guest exit from a particular CPU */
 439 static void exit_vm_noop(void *info)
 440 {
 441 }
 442
 443 void force_vm_exit(const cpumask_t *mask)
 444 {
 445         preempt_disable();
 446         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 447         preempt_enable();
 448 }
 449
 450 /**
 451  * need_new_vmid_gen - check that the VMID is still valid
 452  * @kvm: The VM's VMID to check
 453  *
 454  * return true if there is a new generation of VMIDs being used
 455  *
 456  * The hardware supports only 256 values with the value zero reserved for the
 457  * host, so we check if an assigned value belongs to a previous generation,
 458  * which which requires us to assign a new value. If we're the first to use a
 459  * VMID for the new generation, we must flush necessary caches and TLBs on all
 460  * CPUs.
 461  */
 462 static bool need_new_vmid_gen(struct kvm *kvm)
 463 {
 464         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 465 }
 466
 467 /**
 468  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 469  * @kvm The guest that we are about to run
 470  *
 471  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 472  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 473  * caches and TLBs.
 474  */
 475 static void update_vttbr(struct kvm *kvm)
 476 {
 477         phys_addr_t pgd_phys;
 478         u64 vmid;
 479
 480         if (!need_new_vmid_gen(kvm))
 481                 return;
 482
 483         spin_lock(&kvm_vmid_lock);
 484
 485         /*
 486          * We need to re-check the vmid_gen here to ensure that if another vcpu
 487          * already allocated a valid vmid for this vm, then this vcpu should
 488          * use the same vmid.
 489          */
 490         if (!need_new_vmid_gen(kvm)) {
 491                 spin_unlock(&kvm_vmid_lock);
 492                 return;
 493         }
 494
 495         /* First user of a new VMID generation? */
 496         if (unlikely(kvm_next_vmid == 0)) {
 497                 atomic64_inc(&kvm_vmid_gen);
 498                 kvm_next_vmid = 1;
 499
 500                 /*
 501                  * On SMP we know no other CPUs can use this CPU's or each
 502                  * other's VMID after force_vm_exit returns since the
 503                  * kvm_vmid_lock blocks them from reentry to the guest.
 504                  */
 505                 force_vm_exit(cpu_all_mask);
 506                 /*
 507                  * Now broadcast TLB + ICACHE invalidation over the inner
 508                  * shareable domain to make sure all data structures are
 509                  * clean.
 510                  */
 511                 kvm_call_hyp(__kvm_flush_vm_context);
 512         }
 513
 514         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 515         kvm->arch.vmid = kvm_next_vmid;
 516         kvm_next_vmid++;
 517         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 518
 519         /* update vttbr to be used with the new vmid */
 520         pgd_phys = virt_to_phys(kvm->arch.pgd);
 521         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 522         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 523         kvm->arch.vttbr = pgd_phys | vmid;
 524
 525         spin_unlock(&kvm_vmid_lock);
 526 }
 527
 528 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 529 {
 530         struct kvm *kvm = vcpu->kvm;
 531         int ret = 0;
 532
 533         if (likely(vcpu->arch.has_run_once))
 534                 return 0;
 535
 536         vcpu->arch.has_run_once = true;
 537
 538         if (likely(irqchip_in_kernel(kvm))) {
 539                 /*
 540                  * Map the VGIC hardware resources before running a vcpu the
 541                  * first time on this VM.
 542                  */
 543                 if (unlikely(!vgic_ready(kvm))) {
 544                         ret = kvm_vgic_map_resources(kvm);
 545                         if (ret)
 546                                 return ret;
 547                 }
 548         } else {
 549                 /*
 550                  * Tell the rest of the code that there are userspace irqchip
 551                  * VMs in the wild.
 552                  */
 553                 static_branch_inc(&userspace_irqchip_in_use);
 554         }
 555
 556         ret = kvm_timer_enable(vcpu);
 557         if (ret)
 558                 return ret;
 559
 560         ret = kvm_arm_pmu_v3_enable(vcpu);
 561
 562         return ret;
 563 }
 564
 565 bool kvm_arch_intc_initialized(struct kvm *kvm)
 566 {
 567         return vgic_initialized(kvm);
 568 }
 569
 570 void kvm_arm_halt_guest(struct kvm *kvm)
 571 {
 572         int i;
 573         struct kvm_vcpu *vcpu;
 574
 575         kvm_for_each_vcpu(i, vcpu, kvm)
 576                 vcpu->arch.pause = true;
 577         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 578 }
 579
 580 void kvm_arm_resume_guest(struct kvm *kvm)
 581 {
 582         int i;
 583         struct kvm_vcpu *vcpu;
 584
 585         kvm_for_each_vcpu(i, vcpu, kvm) {
 586                 vcpu->arch.pause = false;
 587                 swake_up(kvm_arch_vcpu_wq(vcpu));
 588         }
 589 }
 590
 591 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 592 {
 593         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 594
 595         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 596                                        (!vcpu->arch.pause)));
 597
 598         if (vcpu->arch.power_off || vcpu->arch.pause) {
 599                 /* Awaken to handle a signal, request we sleep again later. */
 600                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 601         }
 602 }
 603
 604 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 605 {
 606         return vcpu->arch.target >= 0;
 607 }
 608
 609 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 610 {
 611         if (kvm_request_pending(vcpu)) {
 612                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 613                         vcpu_req_sleep(vcpu);
 614
 615                 /*
 616                  * Clear IRQ_PENDING requests that were made to guarantee
 617                  * that a VCPU sees new virtual interrupts.
 618                  */
 619                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 620         }
 621 }
 622
 623 /**
 624  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 625  * @vcpu:       The VCPU pointer
 626  * @run:        The kvm_run structure pointer used for userspace state exchange
 627  *
 628  * This function is called through the VCPU_RUN ioctl called from user space. It
 629  * will execute VM code in a loop until the time slice for the process is used
 630  * or some emulation is needed from user space in which case the function will
 631  * return with return value 0 and with the kvm_run structure filled in with the
 632  * required data for the requested emulation.
 633  */
 634 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 635 {
 636         int ret;
 637
 638         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 639                 return -ENOEXEC;
 640
 641         vcpu_load(vcpu);
 642
 643         ret = kvm_vcpu_first_run_init(vcpu);
 644         if (ret)
 645                 goto out;
 646
 647         if (run->exit_reason == KVM_EXIT_MMIO) {
 648                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 649                 if (ret)
 650                         goto out;
 651                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run)) {
 652                         ret = 0;
 653                         goto out;
 654                 }
 655
 656         }
 657
 658         if (run->immediate_exit) {
 659                 ret = -EINTR;
 660                 goto out;
 661         }
 662
 663         kvm_sigset_activate(vcpu);
 664
 665         ret = 1;
 666         run->exit_reason = KVM_EXIT_UNKNOWN;
 667         while (ret > 0) {
 668                 /*
 669                  * Check conditions before entering the guest
 670                  */
 671                 cond_resched();
 672
 673                 update_vttbr(vcpu->kvm);
 674
 675                 check_vcpu_requests(vcpu);
 676
 677                 /*
 678                  * Preparing the interrupts to be injected also
 679                  * involves poking the GIC, which must be done in a
 680                  * non-preemptible context.
 681                  */
 682                 preempt_disable();
 683
 684                 /* Flush FP/SIMD state that can't survive guest entry/exit */
 685                 kvm_fpsimd_flush_cpu_state();
 686
 687                 kvm_pmu_flush_hwstate(vcpu);
 688
 689                 local_irq_disable();
 690
 691                 kvm_vgic_flush_hwstate(vcpu);
 692
 693                 /*
 694                  * Exit if we have a signal pending so that we can deliver the
 695                  * signal to user space.
 696                  */
 697                 if (signal_pending(current)) {
 698                         ret = -EINTR;
 699                         run->exit_reason = KVM_EXIT_INTR;
 700                 }
 701
 702                 /*
 703                  * If we're using a userspace irqchip, then check if we need
 704                  * to tell a userspace irqchip about timer or PMU level
 705                  * changes and if so, exit to userspace (the actual level
 706                  * state gets updated in kvm_timer_update_run and
 707                  * kvm_pmu_update_run below).
 708                  */
 709                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 710                         if (kvm_timer_should_notify_user(vcpu) ||
 711                             kvm_pmu_should_notify_user(vcpu)) {
 712                                 ret = -EINTR;
 713                                 run->exit_reason = KVM_EXIT_INTR;
 714                         }
 715                 }
 716
 717                 /*
 718                  * Ensure we set mode to IN_GUEST_MODE after we disable
 719                  * interrupts and before the final VCPU requests check.
 720                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 721                  * Documentation/virtual/kvm/vcpu-requests.rst
 722                  */
 723                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 724
 725                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 726                     kvm_request_pending(vcpu)) {
 727                         vcpu->mode = OUTSIDE_GUEST_MODE;
 728                         kvm_pmu_sync_hwstate(vcpu);
 729                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 730                                 kvm_timer_sync_hwstate(vcpu);
 731                         kvm_vgic_sync_hwstate(vcpu);
 732                         local_irq_enable();
 733                         preempt_enable();
 734                         continue;
 735                 }
 736
 737                 kvm_arm_setup_debug(vcpu);
 738
 739                 /**************************************************************
 740                  * Enter the guest
 741                  */
 742                 trace_kvm_entry(*vcpu_pc(vcpu));
 743                 guest_enter_irqoff();
 744
 745                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 746
 747                 vcpu->mode = OUTSIDE_GUEST_MODE;
 748                 vcpu->stat.exits++;
 749                 /*
 750                  * Back from guest
 751                  *************************************************************/
 752
 753                 kvm_arm_clear_debug(vcpu);
 754
 755                 /*
 756                  * We must sync the PMU state before the vgic state so
 757                  * that the vgic can properly sample the updated state of the
 758                  * interrupt line.
 759                  */
 760                 kvm_pmu_sync_hwstate(vcpu);
 761
 762                 /*
 763                  * Sync the vgic state before syncing the timer state because
 764                  * the timer code needs to know if the virtual timer
 765                  * interrupts are active.
 766                  */
 767                 kvm_vgic_sync_hwstate(vcpu);
 768
 769                 /*
 770                  * Sync the timer hardware state before enabling interrupts as
 771                  * we don't want vtimer interrupts to race with syncing the
 772                  * timer virtual interrupt state.
 773                  */
 774                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 775                         kvm_timer_sync_hwstate(vcpu);
 776
 777                 /*
 778                  * We may have taken a host interrupt in HYP mode (ie
 779                  * while executing the guest). This interrupt is still
 780                  * pending, as we haven't serviced it yet!
 781                  *
 782                  * We're now back in SVC mode, with interrupts
 783                  * disabled.  Enabling the interrupts now will have
 784                  * the effect of taking the interrupt again, in SVC
 785                  * mode this time.
 786                  */
 787                 local_irq_enable();
 788
 789                 /*
 790                  * We do local_irq_enable() before calling guest_exit() so
 791                  * that if a timer interrupt hits while running the guest we
 792                  * account that tick as being spent in the guest.  We enable
 793                  * preemption after calling guest_exit() so that if we get
 794                  * preempted we make sure ticks after that is not counted as
 795                  * guest time.
 796                  */
 797                 guest_exit();
 798                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 799
 800                 preempt_enable();
 801
 802                 ret = handle_exit(vcpu, run, ret);
 803         }
 804
 805         /* Tell userspace about in-kernel device output levels */
 806         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 807                 kvm_timer_update_run(vcpu);
 808                 kvm_pmu_update_run(vcpu);
 809         }
 810
 811         kvm_sigset_deactivate(vcpu);
 812
 813 out:
 814         vcpu_put(vcpu);
 815         return ret;
 816 }
 817
 818 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 819 {
 820         int bit_index;
 821         bool set;
 822         unsigned long *ptr;
 823
 824         if (number == KVM_ARM_IRQ_CPU_IRQ)
 825                 bit_index = __ffs(HCR_VI);
 826         else /* KVM_ARM_IRQ_CPU_FIQ */
 827                 bit_index = __ffs(HCR_VF);
 828
 829         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 830         if (level)
 831                 set = test_and_set_bit(bit_index, ptr);
 832         else
 833                 set = test_and_clear_bit(bit_index, ptr);
 834
 835         /*
 836          * If we didn't change anything, no need to wake up or kick other CPUs
 837          */
 838         if (set == level)
 839                 return 0;
 840
 841         /*
 842          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 843          * trigger a world-switch round on the running physical CPU to set the
 844          * virtual IRQ/FIQ fields in the HCR appropriately.
 845          */
 846         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 847         kvm_vcpu_kick(vcpu);
 848
 849         return 0;
 850 }
 851
 852 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 853                           bool line_status)
 854 {
 855         u32 irq = irq_level->irq;
 856         unsigned int irq_type, vcpu_idx, irq_num;
 857         int nrcpus = atomic_read(&kvm->online_vcpus);
 858         struct kvm_vcpu *vcpu = NULL;
 859         bool level = irq_level->level;
 860
 861         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 862         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 863         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 864
 865         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 866
 867         switch (irq_type) {
 868         case KVM_ARM_IRQ_TYPE_CPU:
 869                 if (irqchip_in_kernel(kvm))
 870                         return -ENXIO;
 871
 872                 if (vcpu_idx >= nrcpus)
 873                         return -EINVAL;
 874
 875                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 876                 if (!vcpu)
 877                         return -EINVAL;
 878
 879                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 880                         return -EINVAL;
 881
 882                 return vcpu_interrupt_line(vcpu, irq_num, level);
 883         case KVM_ARM_IRQ_TYPE_PPI:
 884                 if (!irqchip_in_kernel(kvm))
 885                         return -ENXIO;
 886
 887                 if (vcpu_idx >= nrcpus)
 888                         return -EINVAL;
 889
 890                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 891                 if (!vcpu)
 892                         return -EINVAL;
 893
 894                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 895                         return -EINVAL;
 896
 897                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 898         case KVM_ARM_IRQ_TYPE_SPI:
 899                 if (!irqchip_in_kernel(kvm))
 900                         return -ENXIO;
 901
 902                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 903                         return -EINVAL;
 904
 905                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 906         }
 907
 908         return -EINVAL;
 909 }
 910
 911 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 912                                const struct kvm_vcpu_init *init)
 913 {
 914         unsigned int i;
 915         int phys_target = kvm_target_cpu();
 916
 917         if (init->target != phys_target)
 918                 return -EINVAL;
 919
 920         /*
 921          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 922          * use the same target.
 923          */
 924         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 925                 return -EINVAL;
 926
 927         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 928         for (i = 0; i < sizeof(init->features) * 8; i++) {
 929                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 930
 931                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 932                         return -ENOENT;
 933
 934                 /*
 935                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 936                  * use the same feature set.
 937                  */
 938                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 939                     test_bit(i, vcpu->arch.features) != set)
 940                         return -EINVAL;
 941
 942                 if (set)
 943                         set_bit(i, vcpu->arch.features);
 944         }
 945
 946         vcpu->arch.target = phys_target;
 947
 948         /* Now we know what it is, we can reset it. */
 949         return kvm_reset_vcpu(vcpu);
 950 }
 951
 952
 953 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 954                                          struct kvm_vcpu_init *init)
 955 {
 956         int ret;
 957
 958         ret = kvm_vcpu_set_target(vcpu, init);
 959         if (ret)
 960                 return ret;
 961
 962         /*
 963          * Ensure a rebooted VM will fault in RAM pages and detect if the
 964          * guest MMU is turned off and flush the caches as needed.
 965          */
 966         if (vcpu->arch.has_run_once)
 967                 stage2_unmap_vm(vcpu->kvm);
 968
 969         vcpu_reset_hcr(vcpu);
 970
 971         /*
 972          * Handle the "start in power-off" case.
 973          */
 974         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 975                 vcpu_power_off(vcpu);
 976         else
 977                 vcpu->arch.power_off = false;
 978
 979         return 0;
 980 }
 981
 982 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 983                                  struct kvm_device_attr *attr)
 984 {
 985         int ret = -ENXIO;
 986
 987         switch (attr->group) {
 988         default:
 989                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 990                 break;
 991         }
 992
 993         return ret;
 994 }
 995
 996 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 997                                  struct kvm_device_attr *attr)
 998 {
 999         int ret = -ENXIO;
1000
1001         switch (attr->group) {
1002         default:
1003                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1004                 break;
1005         }
1006
1007         return ret;
1008 }
1009
1010 static int kvm_arm_vcpu_has_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_has_attr(vcpu, attr);
1018                 break;
1019         }
1020
1021         return ret;
1022 }
1023
1024 long kvm_arch_vcpu_ioctl(struct file *filp,
1025                          unsigned int ioctl, unsigned long arg)
1026 {
1027         struct kvm_vcpu *vcpu = filp->private_data;
1028         void __user *argp = (void __user *)arg;
1029         struct kvm_device_attr attr;
1030         long r;
1031
1032         vcpu_load(vcpu);
1033
1034         switch (ioctl) {
1035         case KVM_ARM_VCPU_INIT: {
1036                 struct kvm_vcpu_init init;
1037
1038                 r = -EFAULT;
1039                 if (copy_from_user(&init, argp, sizeof(init)))
1040                         break;
1041
1042                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1043                 break;
1044         }
1045         case KVM_SET_ONE_REG:
1046         case KVM_GET_ONE_REG: {
1047                 struct kvm_one_reg reg;
1048
1049                 r = -ENOEXEC;
1050                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1051                         break;
1052
1053                 r = -EFAULT;
1054                 if (copy_from_user(&reg, argp, sizeof(reg)))
1055                         break;
1056
1057                 if (ioctl == KVM_SET_ONE_REG)
1058                         r = kvm_arm_set_reg(vcpu, &reg);
1059                 else
1060                         r = kvm_arm_get_reg(vcpu, &reg);
1061                 break;
1062         }
1063         case KVM_GET_REG_LIST: {
1064                 struct kvm_reg_list __user *user_list = argp;
1065                 struct kvm_reg_list reg_list;
1066                 unsigned n;
1067
1068                 r = -ENOEXEC;
1069                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1070                         break;
1071
1072                 r = -EFAULT;
1073                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1074                         break;
1075                 n = reg_list.n;
1076                 reg_list.n = kvm_arm_num_regs(vcpu);
1077                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1078                         break;
1079                 r = -E2BIG;
1080                 if (n < reg_list.n)
1081                         break;
1082                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1083                 break;
1084         }
1085         case KVM_SET_DEVICE_ATTR: {
1086                 r = -EFAULT;
1087                 if (copy_from_user(&attr, argp, sizeof(attr)))
1088                         break;
1089                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1090                 break;
1091         }
1092         case KVM_GET_DEVICE_ATTR: {
1093                 r = -EFAULT;
1094                 if (copy_from_user(&attr, argp, sizeof(attr)))
1095                         break;
1096                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1097                 break;
1098         }
1099         case KVM_HAS_DEVICE_ATTR: {
1100                 r = -EFAULT;
1101                 if (copy_from_user(&attr, argp, sizeof(attr)))
1102                         break;
1103                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1104                 break;
1105         }
1106         default:
1107                 r = -EINVAL;
1108         }
1109
1110         vcpu_put(vcpu);
1111         return r;
1112 }
1113
1114 /**
1115  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1116  * @kvm: kvm instance
1117  * @log: slot id and address to which we copy the log
1118  *
1119  * Steps 1-4 below provide general overview of dirty page logging. See
1120  * kvm_get_dirty_log_protect() function description for additional details.
1121  *
1122  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1123  * always flush the TLB (step 4) even if previous step failed  and the dirty
1124  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1125  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1126  * writes will be marked dirty for next log read.
1127  *
1128  *   1. Take a snapshot of the bit and clear it if needed.
1129  *   2. Write protect the corresponding page.
1130  *   3. Copy the snapshot to the userspace.
1131  *   4. Flush TLB's if needed.
1132  */
1133 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1134 {
1135         bool is_dirty = false;
1136         int r;
1137
1138         mutex_lock(&kvm->slots_lock);
1139
1140         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1141
1142         if (is_dirty)
1143                 kvm_flush_remote_tlbs(kvm);
1144
1145         mutex_unlock(&kvm->slots_lock);
1146         return r;
1147 }
1148
1149 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1150                                         struct kvm_arm_device_addr *dev_addr)
1151 {
1152         unsigned long dev_id, type;
1153
1154         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1155                 KVM_ARM_DEVICE_ID_SHIFT;
1156         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1157                 KVM_ARM_DEVICE_TYPE_SHIFT;
1158
1159         switch (dev_id) {
1160         case KVM_ARM_DEVICE_VGIC_V2:
1161                 if (!vgic_present)
1162                         return -ENXIO;
1163                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1164         default:
1165                 return -ENODEV;
1166         }
1167 }
1168
1169 long kvm_arch_vm_ioctl(struct file *filp,
1170                        unsigned int ioctl, unsigned long arg)
1171 {
1172         struct kvm *kvm = filp->private_data;
1173         void __user *argp = (void __user *)arg;
1174
1175         switch (ioctl) {
1176         case KVM_CREATE_IRQCHIP: {
1177                 int ret;
1178                 if (!vgic_present)
1179                         return -ENXIO;
1180                 mutex_lock(&kvm->lock);
1181                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1182                 mutex_unlock(&kvm->lock);
1183                 return ret;
1184         }
1185         case KVM_ARM_SET_DEVICE_ADDR: {
1186                 struct kvm_arm_device_addr dev_addr;
1187
1188                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1189                         return -EFAULT;
1190                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1191         }
1192         case KVM_ARM_PREFERRED_TARGET: {
1193                 int err;
1194                 struct kvm_vcpu_init init;
1195
1196                 err = kvm_vcpu_preferred_target(&init);
1197                 if (err)
1198                         return err;
1199
1200                 if (copy_to_user(argp, &init, sizeof(init)))
1201                         return -EFAULT;
1202
1203                 return 0;
1204         }
1205         default:
1206                 return -EINVAL;
1207         }
1208 }
1209
1210 static void cpu_init_hyp_mode(void *dummy)
1211 {
1212         phys_addr_t pgd_ptr;
1213         unsigned long hyp_stack_ptr;
1214         unsigned long stack_page;
1215         unsigned long vector_ptr;
1216
1217         /* Switch from the HYP stub to our own HYP init vector */
1218         __hyp_set_vectors(kvm_get_idmap_vector());
1219
1220         pgd_ptr = kvm_mmu_get_httbr();
1221         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1222         hyp_stack_ptr = stack_page + PAGE_SIZE;
1223         vector_ptr = (unsigned long)kvm_ksym_ref(__kvm_hyp_vector);
1224
1225         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1226         __cpu_init_stage2();
1227
1228         kvm_arm_init_debug();
1229 }
1230
1231 static void cpu_hyp_reset(void)
1232 {
1233         if (!is_kernel_in_hyp_mode())
1234                 __hyp_reset_vectors();
1235 }
1236
1237 static void cpu_hyp_reinit(void)
1238 {
1239         cpu_hyp_reset();
1240
1241         if (is_kernel_in_hyp_mode()) {
1242                 /*
1243                  * __cpu_init_stage2() is safe to call even if the PM
1244                  * event was cancelled before the CPU was reset.
1245                  */
1246                 __cpu_init_stage2();
1247                 kvm_timer_init_vhe();
1248         } else {
1249                 cpu_init_hyp_mode(NULL);
1250         }
1251
1252         if (vgic_present)
1253                 kvm_vgic_init_cpu_hardware();
1254 }
1255
1256 static void _kvm_arch_hardware_enable(void *discard)
1257 {
1258         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1259                 cpu_hyp_reinit();
1260                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1261         }
1262 }
1263
1264 int kvm_arch_hardware_enable(void)
1265 {
1266         _kvm_arch_hardware_enable(NULL);
1267         return 0;
1268 }
1269
1270 static void _kvm_arch_hardware_disable(void *discard)
1271 {
1272         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1273                 cpu_hyp_reset();
1274                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1275         }
1276 }
1277
1278 void kvm_arch_hardware_disable(void)
1279 {
1280         _kvm_arch_hardware_disable(NULL);
1281 }
1282
1283 #ifdef CONFIG_CPU_PM
1284 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1285                                     unsigned long cmd,
1286                                     void *v)
1287 {
1288         /*
1289          * kvm_arm_hardware_enabled is left with its old value over
1290          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1291          * re-enable hyp.
1292          */
1293         switch (cmd) {
1294         case CPU_PM_ENTER:
1295                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1296                         /*
1297                          * don't update kvm_arm_hardware_enabled here
1298                          * so that the hardware will be re-enabled
1299                          * when we resume. See below.
1300                          */
1301                         cpu_hyp_reset();
1302
1303                 return NOTIFY_OK;
1304         case CPU_PM_ENTER_FAILED:
1305         case CPU_PM_EXIT:
1306                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1307                         /* The hardware was enabled before suspend. */
1308                         cpu_hyp_reinit();
1309
1310                 return NOTIFY_OK;
1311
1312         default:
1313                 return NOTIFY_DONE;
1314         }
1315 }
1316
1317 static struct notifier_block hyp_init_cpu_pm_nb = {
1318         .notifier_call = hyp_init_cpu_pm_notifier,
1319 };
1320
1321 static void __init hyp_cpu_pm_init(void)
1322 {
1323         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1324 }
1325 static void __init hyp_cpu_pm_exit(void)
1326 {
1327         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1328 }
1329 #else
1330 static inline void hyp_cpu_pm_init(void)
1331 {
1332 }
1333 static inline void hyp_cpu_pm_exit(void)
1334 {
1335 }
1336 #endif
1337
1338 static void teardown_common_resources(void)
1339 {
1340         free_percpu(kvm_host_cpu_state);
1341 }
1342
1343 static int init_common_resources(void)
1344 {
1345         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1346         if (!kvm_host_cpu_state) {
1347                 kvm_err("Cannot allocate host CPU state\n");
1348                 return -ENOMEM;
1349         }
1350
1351         /* set size of VMID supported by CPU */
1352         kvm_vmid_bits = kvm_get_vmid_bits();
1353         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1354
1355         return 0;
1356 }
1357
1358 static int init_subsystems(void)
1359 {
1360         int err = 0;
1361
1362         /*
1363          * Enable hardware so that subsystem initialisation can access EL2.
1364          */
1365         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1366
1367         /*
1368          * Register CPU lower-power notifier
1369          */
1370         hyp_cpu_pm_init();
1371
1372         /*
1373          * Init HYP view of VGIC
1374          */
1375         err = kvm_vgic_hyp_init();
1376         switch (err) {
1377         case 0:
1378                 vgic_present = true;
1379                 break;
1380         case -ENODEV:
1381         case -ENXIO:
1382                 vgic_present = false;
1383                 err = 0;
1384                 break;
1385         default:
1386                 goto out;
1387         }
1388
1389         /*
1390          * Init HYP architected timer support
1391          */
1392         err = kvm_timer_hyp_init(vgic_present);
1393         if (err)
1394                 goto out;
1395
1396         kvm_perf_init();
1397         kvm_coproc_table_init();
1398
1399 out:
1400         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1401
1402         return err;
1403 }
1404
1405 static void teardown_hyp_mode(void)
1406 {
1407         int cpu;
1408
1409         free_hyp_pgds();
1410         for_each_possible_cpu(cpu)
1411                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1412         hyp_cpu_pm_exit();
1413 }
1414
1415 /**
1416  * Inits Hyp-mode on all online CPUs
1417  */
1418 static int init_hyp_mode(void)
1419 {
1420         int cpu;
1421         int err = 0;
1422
1423         /*
1424          * Allocate Hyp PGD and setup Hyp identity mapping
1425          */
1426         err = kvm_mmu_init();
1427         if (err)
1428                 goto out_err;
1429
1430         /*
1431          * Allocate stack pages for Hypervisor-mode
1432          */
1433         for_each_possible_cpu(cpu) {
1434                 unsigned long stack_page;
1435
1436                 stack_page = __get_free_page(GFP_KERNEL);
1437                 if (!stack_page) {
1438                         err = -ENOMEM;
1439                         goto out_err;
1440                 }
1441
1442                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1443         }
1444
1445         /*
1446          * Map the Hyp-code called directly from the host
1447          */
1448         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1449                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1450         if (err) {
1451                 kvm_err("Cannot map world-switch code\n");
1452                 goto out_err;
1453         }
1454
1455         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1456                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1457         if (err) {
1458                 kvm_err("Cannot map rodata section\n");
1459                 goto out_err;
1460         }
1461
1462         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1463                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1464         if (err) {
1465                 kvm_err("Cannot map bss section\n");
1466                 goto out_err;
1467         }
1468
1469         /*
1470          * Map the Hyp stack pages
1471          */
1472         for_each_possible_cpu(cpu) {
1473                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1474                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1475                                           PAGE_HYP);
1476
1477                 if (err) {
1478                         kvm_err("Cannot map hyp stack\n");
1479                         goto out_err;
1480                 }
1481         }
1482
1483         for_each_possible_cpu(cpu) {
1484                 kvm_cpu_context_t *cpu_ctxt;
1485
1486                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1487                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1488
1489                 if (err) {
1490                         kvm_err("Cannot map host CPU state: %d\n", err);
1491                         goto out_err;
1492                 }
1493         }
1494
1495         return 0;
1496
1497 out_err:
1498         teardown_hyp_mode();
1499         kvm_err("error initializing Hyp mode: %d\n", err);
1500         return err;
1501 }
1502
1503 static void check_kvm_target_cpu(void *ret)
1504 {
1505         *(int *)ret = kvm_target_cpu();
1506 }
1507
1508 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1509 {
1510         struct kvm_vcpu *vcpu;
1511         int i;
1512
1513         mpidr &= MPIDR_HWID_BITMASK;
1514         kvm_for_each_vcpu(i, vcpu, kvm) {
1515                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1516                         return vcpu;
1517         }
1518         return NULL;
1519 }
1520
1521 bool kvm_arch_has_irq_bypass(void)
1522 {
1523         return true;
1524 }
1525
1526 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1527                                       struct irq_bypass_producer *prod)
1528 {
1529         struct kvm_kernel_irqfd *irqfd =
1530                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1531
1532         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1533                                           &irqfd->irq_entry);
1534 }
1535 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1536                                       struct irq_bypass_producer *prod)
1537 {
1538         struct kvm_kernel_irqfd *irqfd =
1539                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1540
1541         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1542                                      &irqfd->irq_entry);
1543 }
1544
1545 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1546 {
1547         struct kvm_kernel_irqfd *irqfd =
1548                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1549
1550         kvm_arm_halt_guest(irqfd->kvm);
1551 }
1552
1553 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1554 {
1555         struct kvm_kernel_irqfd *irqfd =
1556                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1557
1558         kvm_arm_resume_guest(irqfd->kvm);
1559 }
1560
1561 /**
1562  * Initialize Hyp-mode and memory mappings on all CPUs.
1563  */
1564 int kvm_arch_init(void *opaque)
1565 {
1566         int err;
1567         int ret, cpu;
1568         bool in_hyp_mode;
1569
1570         if (!is_hyp_mode_available()) {
1571                 kvm_info("HYP mode not available\n");
1572                 return -ENODEV;
1573         }
1574
1575         for_each_online_cpu(cpu) {
1576                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1577                 if (ret < 0) {
1578                         kvm_err("Error, CPU %d not supported!\n", cpu);
1579                         return -ENODEV;
1580                 }
1581         }
1582
1583         err = init_common_resources();
1584         if (err)
1585                 return err;
1586
1587         in_hyp_mode = is_kernel_in_hyp_mode();
1588
1589         if (!in_hyp_mode) {
1590                 err = init_hyp_mode();
1591                 if (err)
1592                         goto out_err;
1593         }
1594
1595         err = init_subsystems();
1596         if (err)
1597                 goto out_hyp;
1598
1599         if (in_hyp_mode)
1600                 kvm_info("VHE mode initialized successfully\n");
1601         else
1602                 kvm_info("Hyp mode initialized successfully\n");
1603
1604         return 0;
1605
1606 out_hyp:
1607         if (!in_hyp_mode)
1608                 teardown_hyp_mode();
1609 out_err:
1610         teardown_common_resources();
1611         return err;
1612 }
1613
1614 /* NOP: Compiling as a module not supported */
1615 void kvm_arch_exit(void)
1616 {
1617         kvm_perf_teardown();
1618 }
1619
1620 static int arm_init(void)
1621 {
1622         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1623         return rc;
1624 }
1625
1626 module_init(arm_init);