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 #include <kvm/arm_psci.h>
  35
  36 #define CREATE_TRACE_POINTS
  37 #include "trace.h"
  38
  39 #include <linux/uaccess.h>
  40 #include <asm/ptrace.h>
  41 #include <asm/mman.h>
  42 #include <asm/tlbflush.h>
  43 #include <asm/cacheflush.h>
  44 #include <asm/virt.h>
  45 #include <asm/kvm_arm.h>
  46 #include <asm/kvm_asm.h>
  47 #include <asm/kvm_mmu.h>
  48 #include <asm/kvm_emulate.h>
  49 #include <asm/kvm_coproc.h>
  50 #include <asm/sections.h>
  51
  52 #ifdef REQUIRES_VIRT
  53 __asm__(".arch_extension        virt");
  54 #endif
  55
  56 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  57 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  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         kvm_vcpu_load_sysregs(vcpu);
 366 }
 367
 368 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 369 {
 370         kvm_vcpu_put_sysregs(vcpu);
 371         kvm_timer_vcpu_put(vcpu);
 372         kvm_vgic_put(vcpu);
 373
 374         vcpu->cpu = -1;
 375
 376         kvm_arm_set_running_vcpu(NULL);
 377 }
 378
 379 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 380 {
 381         vcpu->arch.power_off = true;
 382         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 383         kvm_vcpu_kick(vcpu);
 384 }
 385
 386 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 387                                     struct kvm_mp_state *mp_state)
 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         return 0;
 395 }
 396
 397 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 398                                     struct kvm_mp_state *mp_state)
 399 {
 400         int ret = 0;
 401
 402         switch (mp_state->mp_state) {
 403         case KVM_MP_STATE_RUNNABLE:
 404                 vcpu->arch.power_off = false;
 405                 break;
 406         case KVM_MP_STATE_STOPPED:
 407                 vcpu_power_off(vcpu);
 408                 break;
 409         default:
 410                 ret = -EINVAL;
 411         }
 412
 413         return ret;
 414 }
 415
 416 /**
 417  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 418  * @v:          The VCPU pointer
 419  *
 420  * If the guest CPU is not waiting for interrupts or an interrupt line is
 421  * asserted, the CPU is by definition runnable.
 422  */
 423 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 424 {
 425         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 426         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 427                 && !v->arch.power_off && !v->arch.pause);
 428 }
 429
 430 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 431 {
 432         return vcpu_mode_priv(vcpu);
 433 }
 434
 435 /* Just ensure a guest exit from a particular CPU */
 436 static void exit_vm_noop(void *info)
 437 {
 438 }
 439
 440 void force_vm_exit(const cpumask_t *mask)
 441 {
 442         preempt_disable();
 443         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 444         preempt_enable();
 445 }
 446
 447 /**
 448  * need_new_vmid_gen - check that the VMID is still valid
 449  * @kvm: The VM's VMID to check
 450  *
 451  * return true if there is a new generation of VMIDs being used
 452  *
 453  * The hardware supports only 256 values with the value zero reserved for the
 454  * host, so we check if an assigned value belongs to a previous generation,
 455  * which which requires us to assign a new value. If we're the first to use a
 456  * VMID for the new generation, we must flush necessary caches and TLBs on all
 457  * CPUs.
 458  */
 459 static bool need_new_vmid_gen(struct kvm *kvm)
 460 {
 461         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 462 }
 463
 464 /**
 465  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 466  * @kvm The guest that we are about to run
 467  *
 468  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 469  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 470  * caches and TLBs.
 471  */
 472 static void update_vttbr(struct kvm *kvm)
 473 {
 474         phys_addr_t pgd_phys;
 475         u64 vmid;
 476
 477         if (!need_new_vmid_gen(kvm))
 478                 return;
 479
 480         spin_lock(&kvm_vmid_lock);
 481
 482         /*
 483          * We need to re-check the vmid_gen here to ensure that if another vcpu
 484          * already allocated a valid vmid for this vm, then this vcpu should
 485          * use the same vmid.
 486          */
 487         if (!need_new_vmid_gen(kvm)) {
 488                 spin_unlock(&kvm_vmid_lock);
 489                 return;
 490         }
 491
 492         /* First user of a new VMID generation? */
 493         if (unlikely(kvm_next_vmid == 0)) {
 494                 atomic64_inc(&kvm_vmid_gen);
 495                 kvm_next_vmid = 1;
 496
 497                 /*
 498                  * On SMP we know no other CPUs can use this CPU's or each
 499                  * other's VMID after force_vm_exit returns since the
 500                  * kvm_vmid_lock blocks them from reentry to the guest.
 501                  */
 502                 force_vm_exit(cpu_all_mask);
 503                 /*
 504                  * Now broadcast TLB + ICACHE invalidation over the inner
 505                  * shareable domain to make sure all data structures are
 506                  * clean.
 507                  */
 508                 kvm_call_hyp(__kvm_flush_vm_context);
 509         }
 510
 511         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 512         kvm->arch.vmid = kvm_next_vmid;
 513         kvm_next_vmid++;
 514         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 515
 516         /* update vttbr to be used with the new vmid */
 517         pgd_phys = virt_to_phys(kvm->arch.pgd);
 518         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 519         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 520         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid;
 521
 522         spin_unlock(&kvm_vmid_lock);
 523 }
 524
 525 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 526 {
 527         struct kvm *kvm = vcpu->kvm;
 528         int ret = 0;
 529
 530         if (likely(vcpu->arch.has_run_once))
 531                 return 0;
 532
 533         vcpu->arch.has_run_once = true;
 534
 535         if (likely(irqchip_in_kernel(kvm))) {
 536                 /*
 537                  * Map the VGIC hardware resources before running a vcpu the
 538                  * first time on this VM.
 539                  */
 540                 if (unlikely(!vgic_ready(kvm))) {
 541                         ret = kvm_vgic_map_resources(kvm);
 542                         if (ret)
 543                                 return ret;
 544                 }
 545         } else {
 546                 /*
 547                  * Tell the rest of the code that there are userspace irqchip
 548                  * VMs in the wild.
 549                  */
 550                 static_branch_inc(&userspace_irqchip_in_use);
 551         }
 552
 553         ret = kvm_timer_enable(vcpu);
 554         if (ret)
 555                 return ret;
 556
 557         ret = kvm_arm_pmu_v3_enable(vcpu);
 558
 559         return ret;
 560 }
 561
 562 bool kvm_arch_intc_initialized(struct kvm *kvm)
 563 {
 564         return vgic_initialized(kvm);
 565 }
 566
 567 void kvm_arm_halt_guest(struct kvm *kvm)
 568 {
 569         int i;
 570         struct kvm_vcpu *vcpu;
 571
 572         kvm_for_each_vcpu(i, vcpu, kvm)
 573                 vcpu->arch.pause = true;
 574         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 575 }
 576
 577 void kvm_arm_resume_guest(struct kvm *kvm)
 578 {
 579         int i;
 580         struct kvm_vcpu *vcpu;
 581
 582         kvm_for_each_vcpu(i, vcpu, kvm) {
 583                 vcpu->arch.pause = false;
 584                 swake_up(kvm_arch_vcpu_wq(vcpu));
 585         }
 586 }
 587
 588 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 589 {
 590         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 591
 592         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 593                                        (!vcpu->arch.pause)));
 594
 595         if (vcpu->arch.power_off || vcpu->arch.pause) {
 596                 /* Awaken to handle a signal, request we sleep again later. */
 597                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 598         }
 599 }
 600
 601 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 602 {
 603         return vcpu->arch.target >= 0;
 604 }
 605
 606 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 607 {
 608         if (kvm_request_pending(vcpu)) {
 609                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 610                         vcpu_req_sleep(vcpu);
 611
 612                 /*
 613                  * Clear IRQ_PENDING requests that were made to guarantee
 614                  * that a VCPU sees new virtual interrupts.
 615                  */
 616                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 617         }
 618 }
 619
 620 /**
 621  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 622  * @vcpu:       The VCPU pointer
 623  * @run:        The kvm_run structure pointer used for userspace state exchange
 624  *
 625  * This function is called through the VCPU_RUN ioctl called from user space. It
 626  * will execute VM code in a loop until the time slice for the process is used
 627  * or some emulation is needed from user space in which case the function will
 628  * return with return value 0 and with the kvm_run structure filled in with the
 629  * required data for the requested emulation.
 630  */
 631 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 632 {
 633         int ret;
 634
 635         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 636                 return -ENOEXEC;
 637
 638         ret = kvm_vcpu_first_run_init(vcpu);
 639         if (ret)
 640                 return ret;
 641
 642         if (run->exit_reason == KVM_EXIT_MMIO) {
 643                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 644                 if (ret)
 645                         return ret;
 646                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
 647                         return 0;
 648         }
 649
 650         if (run->immediate_exit)
 651                 return -EINTR;
 652
 653         vcpu_load(vcpu);
 654
 655         kvm_sigset_activate(vcpu);
 656
 657         ret = 1;
 658         run->exit_reason = KVM_EXIT_UNKNOWN;
 659         while (ret > 0) {
 660                 /*
 661                  * Check conditions before entering the guest
 662                  */
 663                 cond_resched();
 664
 665                 update_vttbr(vcpu->kvm);
 666
 667                 check_vcpu_requests(vcpu);
 668
 669                 /*
 670                  * Preparing the interrupts to be injected also
 671                  * involves poking the GIC, which must be done in a
 672                  * non-preemptible context.
 673                  */
 674                 preempt_disable();
 675
 676                 /* Flush FP/SIMD state that can't survive guest entry/exit */
 677                 kvm_fpsimd_flush_cpu_state();
 678
 679                 kvm_pmu_flush_hwstate(vcpu);
 680
 681                 local_irq_disable();
 682
 683                 kvm_vgic_flush_hwstate(vcpu);
 684
 685                 /*
 686                  * Exit if we have a signal pending so that we can deliver the
 687                  * signal to user space.
 688                  */
 689                 if (signal_pending(current)) {
 690                         ret = -EINTR;
 691                         run->exit_reason = KVM_EXIT_INTR;
 692                 }
 693
 694                 /*
 695                  * If we're using a userspace irqchip, then check if we need
 696                  * to tell a userspace irqchip about timer or PMU level
 697                  * changes and if so, exit to userspace (the actual level
 698                  * state gets updated in kvm_timer_update_run and
 699                  * kvm_pmu_update_run below).
 700                  */
 701                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 702                         if (kvm_timer_should_notify_user(vcpu) ||
 703                             kvm_pmu_should_notify_user(vcpu)) {
 704                                 ret = -EINTR;
 705                                 run->exit_reason = KVM_EXIT_INTR;
 706                         }
 707                 }
 708
 709                 /*
 710                  * Ensure we set mode to IN_GUEST_MODE after we disable
 711                  * interrupts and before the final VCPU requests check.
 712                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 713                  * Documentation/virtual/kvm/vcpu-requests.rst
 714                  */
 715                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 716
 717                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 718                     kvm_request_pending(vcpu)) {
 719                         vcpu->mode = OUTSIDE_GUEST_MODE;
 720                         isb(); /* Ensure work in x_flush_hwstate is committed */
 721                         kvm_pmu_sync_hwstate(vcpu);
 722                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 723                                 kvm_timer_sync_hwstate(vcpu);
 724                         kvm_vgic_sync_hwstate(vcpu);
 725                         local_irq_enable();
 726                         preempt_enable();
 727                         continue;
 728                 }
 729
 730                 kvm_arm_setup_debug(vcpu);
 731
 732                 /**************************************************************
 733                  * Enter the guest
 734                  */
 735                 trace_kvm_entry(*vcpu_pc(vcpu));
 736                 guest_enter_irqoff();
 737
 738                 if (has_vhe()) {
 739                         kvm_arm_vhe_guest_enter();
 740                         ret = kvm_vcpu_run_vhe(vcpu);
 741                         kvm_arm_vhe_guest_exit();
 742                 } else {
 743                         ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
 744                 }
 745
 746                 vcpu->mode = OUTSIDE_GUEST_MODE;
 747                 vcpu->stat.exits++;
 748                 /*
 749                  * Back from guest
 750                  *************************************************************/
 751
 752                 kvm_arm_clear_debug(vcpu);
 753
 754                 /*
 755                  * We must sync the PMU state before the vgic state so
 756                  * that the vgic can properly sample the updated state of the
 757                  * interrupt line.
 758                  */
 759                 kvm_pmu_sync_hwstate(vcpu);
 760
 761                 /*
 762                  * Sync the vgic state before syncing the timer state because
 763                  * the timer code needs to know if the virtual timer
 764                  * interrupts are active.
 765                  */
 766                 kvm_vgic_sync_hwstate(vcpu);
 767
 768                 /*
 769                  * Sync the timer hardware state before enabling interrupts as
 770                  * we don't want vtimer interrupts to race with syncing the
 771                  * timer virtual interrupt state.
 772                  */
 773                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 774                         kvm_timer_sync_hwstate(vcpu);
 775
 776                 /*
 777                  * We may have taken a host interrupt in HYP mode (ie
 778                  * while executing the guest). This interrupt is still
 779                  * pending, as we haven't serviced it yet!
 780                  *
 781                  * We're now back in SVC mode, with interrupts
 782                  * disabled.  Enabling the interrupts now will have
 783                  * the effect of taking the interrupt again, in SVC
 784                  * mode this time.
 785                  */
 786                 local_irq_enable();
 787
 788                 /*
 789                  * We do local_irq_enable() before calling guest_exit() so
 790                  * that if a timer interrupt hits while running the guest we
 791                  * account that tick as being spent in the guest.  We enable
 792                  * preemption after calling guest_exit() so that if we get
 793                  * preempted we make sure ticks after that is not counted as
 794                  * guest time.
 795                  */
 796                 guest_exit();
 797                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 798
 799                 /* Exit types that need handling before we can be preempted */
 800                 handle_exit_early(vcpu, run, ret);
 801
 802                 preempt_enable();
 803
 804                 ret = handle_exit(vcpu, run, ret);
 805         }
 806
 807         /* Tell userspace about in-kernel device output levels */
 808         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 809                 kvm_timer_update_run(vcpu);
 810                 kvm_pmu_update_run(vcpu);
 811         }
 812
 813         kvm_sigset_deactivate(vcpu);
 814
 815         vcpu_put(vcpu);
 816         return ret;
 817 }
 818
 819 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 820 {
 821         int bit_index;
 822         bool set;
 823         unsigned long *hcr;
 824
 825         if (number == KVM_ARM_IRQ_CPU_IRQ)
 826                 bit_index = __ffs(HCR_VI);
 827         else /* KVM_ARM_IRQ_CPU_FIQ */
 828                 bit_index = __ffs(HCR_VF);
 829
 830         hcr = vcpu_hcr(vcpu);
 831         if (level)
 832                 set = test_and_set_bit(bit_index, hcr);
 833         else
 834                 set = test_and_clear_bit(bit_index, hcr);
 835
 836         /*
 837          * If we didn't change anything, no need to wake up or kick other CPUs
 838          */
 839         if (set == level)
 840                 return 0;
 841
 842         /*
 843          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 844          * trigger a world-switch round on the running physical CPU to set the
 845          * virtual IRQ/FIQ fields in the HCR appropriately.
 846          */
 847         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 848         kvm_vcpu_kick(vcpu);
 849
 850         return 0;
 851 }
 852
 853 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 854                           bool line_status)
 855 {
 856         u32 irq = irq_level->irq;
 857         unsigned int irq_type, vcpu_idx, irq_num;
 858         int nrcpus = atomic_read(&kvm->online_vcpus);
 859         struct kvm_vcpu *vcpu = NULL;
 860         bool level = irq_level->level;
 861
 862         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 863         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 864         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 865
 866         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 867
 868         switch (irq_type) {
 869         case KVM_ARM_IRQ_TYPE_CPU:
 870                 if (irqchip_in_kernel(kvm))
 871                         return -ENXIO;
 872
 873                 if (vcpu_idx >= nrcpus)
 874                         return -EINVAL;
 875
 876                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 877                 if (!vcpu)
 878                         return -EINVAL;
 879
 880                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 881                         return -EINVAL;
 882
 883                 return vcpu_interrupt_line(vcpu, irq_num, level);
 884         case KVM_ARM_IRQ_TYPE_PPI:
 885                 if (!irqchip_in_kernel(kvm))
 886                         return -ENXIO;
 887
 888                 if (vcpu_idx >= nrcpus)
 889                         return -EINVAL;
 890
 891                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 892                 if (!vcpu)
 893                         return -EINVAL;
 894
 895                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 896                         return -EINVAL;
 897
 898                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 899         case KVM_ARM_IRQ_TYPE_SPI:
 900                 if (!irqchip_in_kernel(kvm))
 901                         return -ENXIO;
 902
 903                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 904                         return -EINVAL;
 905
 906                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 907         }
 908
 909         return -EINVAL;
 910 }
 911
 912 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 913                                const struct kvm_vcpu_init *init)
 914 {
 915         unsigned int i;
 916         int phys_target = kvm_target_cpu();
 917
 918         if (init->target != phys_target)
 919                 return -EINVAL;
 920
 921         /*
 922          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 923          * use the same target.
 924          */
 925         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 926                 return -EINVAL;
 927
 928         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 929         for (i = 0; i < sizeof(init->features) * 8; i++) {
 930                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 931
 932                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 933                         return -ENOENT;
 934
 935                 /*
 936                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 937                  * use the same feature set.
 938                  */
 939                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 940                     test_bit(i, vcpu->arch.features) != set)
 941                         return -EINVAL;
 942
 943                 if (set)
 944                         set_bit(i, vcpu->arch.features);
 945         }
 946
 947         vcpu->arch.target = phys_target;
 948
 949         /* Now we know what it is, we can reset it. */
 950         return kvm_reset_vcpu(vcpu);
 951 }
 952
 953
 954 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 955                                          struct kvm_vcpu_init *init)
 956 {
 957         int ret;
 958
 959         ret = kvm_vcpu_set_target(vcpu, init);
 960         if (ret)
 961                 return ret;
 962
 963         /*
 964          * Ensure a rebooted VM will fault in RAM pages and detect if the
 965          * guest MMU is turned off and flush the caches as needed.
 966          */
 967         if (vcpu->arch.has_run_once)
 968                 stage2_unmap_vm(vcpu->kvm);
 969
 970         vcpu_reset_hcr(vcpu);
 971
 972         /*
 973          * Handle the "start in power-off" case.
 974          */
 975         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 976                 vcpu_power_off(vcpu);
 977         else
 978                 vcpu->arch.power_off = false;
 979
 980         return 0;
 981 }
 982
 983 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 984                                  struct kvm_device_attr *attr)
 985 {
 986         int ret = -ENXIO;
 987
 988         switch (attr->group) {
 989         default:
 990                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 991                 break;
 992         }
 993
 994         return ret;
 995 }
 996
 997 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
 998                                  struct kvm_device_attr *attr)
 999 {
1000         int ret = -ENXIO;
1001
1002         switch (attr->group) {
1003         default:
1004                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1005                 break;
1006         }
1007
1008         return ret;
1009 }
1010
1011 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1012                                  struct kvm_device_attr *attr)
1013 {
1014         int ret = -ENXIO;
1015
1016         switch (attr->group) {
1017         default:
1018                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1019                 break;
1020         }
1021
1022         return ret;
1023 }
1024
1025 long kvm_arch_vcpu_ioctl(struct file *filp,
1026                          unsigned int ioctl, unsigned long arg)
1027 {
1028         struct kvm_vcpu *vcpu = filp->private_data;
1029         void __user *argp = (void __user *)arg;
1030         struct kvm_device_attr attr;
1031         long r;
1032
1033         switch (ioctl) {
1034         case KVM_ARM_VCPU_INIT: {
1035                 struct kvm_vcpu_init init;
1036
1037                 r = -EFAULT;
1038                 if (copy_from_user(&init, argp, sizeof(init)))
1039                         break;
1040
1041                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1042                 break;
1043         }
1044         case KVM_SET_ONE_REG:
1045         case KVM_GET_ONE_REG: {
1046                 struct kvm_one_reg reg;
1047
1048                 r = -ENOEXEC;
1049                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1050                         break;
1051
1052                 r = -EFAULT;
1053                 if (copy_from_user(&reg, argp, sizeof(reg)))
1054                         break;
1055
1056                 if (ioctl == KVM_SET_ONE_REG)
1057                         r = kvm_arm_set_reg(vcpu, &reg);
1058                 else
1059                         r = kvm_arm_get_reg(vcpu, &reg);
1060                 break;
1061         }
1062         case KVM_GET_REG_LIST: {
1063                 struct kvm_reg_list __user *user_list = argp;
1064                 struct kvm_reg_list reg_list;
1065                 unsigned n;
1066
1067                 r = -ENOEXEC;
1068                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1069                         break;
1070
1071                 r = -EFAULT;
1072                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1073                         break;
1074                 n = reg_list.n;
1075                 reg_list.n = kvm_arm_num_regs(vcpu);
1076                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1077                         break;
1078                 r = -E2BIG;
1079                 if (n < reg_list.n)
1080                         break;
1081                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1082                 break;
1083         }
1084         case KVM_SET_DEVICE_ATTR: {
1085                 r = -EFAULT;
1086                 if (copy_from_user(&attr, argp, sizeof(attr)))
1087                         break;
1088                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1089                 break;
1090         }
1091         case KVM_GET_DEVICE_ATTR: {
1092                 r = -EFAULT;
1093                 if (copy_from_user(&attr, argp, sizeof(attr)))
1094                         break;
1095                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1096                 break;
1097         }
1098         case KVM_HAS_DEVICE_ATTR: {
1099                 r = -EFAULT;
1100                 if (copy_from_user(&attr, argp, sizeof(attr)))
1101                         break;
1102                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1103                 break;
1104         }
1105         default:
1106                 r = -EINVAL;
1107         }
1108
1109         return r;
1110 }
1111
1112 /**
1113  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1114  * @kvm: kvm instance
1115  * @log: slot id and address to which we copy the log
1116  *
1117  * Steps 1-4 below provide general overview of dirty page logging. See
1118  * kvm_get_dirty_log_protect() function description for additional details.
1119  *
1120  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1121  * always flush the TLB (step 4) even if previous step failed  and the dirty
1122  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1123  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1124  * writes will be marked dirty for next log read.
1125  *
1126  *   1. Take a snapshot of the bit and clear it if needed.
1127  *   2. Write protect the corresponding page.
1128  *   3. Copy the snapshot to the userspace.
1129  *   4. Flush TLB's if needed.
1130  */
1131 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1132 {
1133         bool is_dirty = false;
1134         int r;
1135
1136         mutex_lock(&kvm->slots_lock);
1137
1138         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1139
1140         if (is_dirty)
1141                 kvm_flush_remote_tlbs(kvm);
1142
1143         mutex_unlock(&kvm->slots_lock);
1144         return r;
1145 }
1146
1147 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1148                                         struct kvm_arm_device_addr *dev_addr)
1149 {
1150         unsigned long dev_id, type;
1151
1152         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1153                 KVM_ARM_DEVICE_ID_SHIFT;
1154         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1155                 KVM_ARM_DEVICE_TYPE_SHIFT;
1156
1157         switch (dev_id) {
1158         case KVM_ARM_DEVICE_VGIC_V2:
1159                 if (!vgic_present)
1160                         return -ENXIO;
1161                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1162         default:
1163                 return -ENODEV;
1164         }
1165 }
1166
1167 long kvm_arch_vm_ioctl(struct file *filp,
1168                        unsigned int ioctl, unsigned long arg)
1169 {
1170         struct kvm *kvm = filp->private_data;
1171         void __user *argp = (void __user *)arg;
1172
1173         switch (ioctl) {
1174         case KVM_CREATE_IRQCHIP: {
1175                 int ret;
1176                 if (!vgic_present)
1177                         return -ENXIO;
1178                 mutex_lock(&kvm->lock);
1179                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1180                 mutex_unlock(&kvm->lock);
1181                 return ret;
1182         }
1183         case KVM_ARM_SET_DEVICE_ADDR: {
1184                 struct kvm_arm_device_addr dev_addr;
1185
1186                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1187                         return -EFAULT;
1188                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1189         }
1190         case KVM_ARM_PREFERRED_TARGET: {
1191                 int err;
1192                 struct kvm_vcpu_init init;
1193
1194                 err = kvm_vcpu_preferred_target(&init);
1195                 if (err)
1196                         return err;
1197
1198                 if (copy_to_user(argp, &init, sizeof(init)))
1199                         return -EFAULT;
1200
1201                 return 0;
1202         }
1203         default:
1204                 return -EINVAL;
1205         }
1206 }
1207
1208 static void cpu_init_hyp_mode(void *dummy)
1209 {
1210         phys_addr_t pgd_ptr;
1211         unsigned long hyp_stack_ptr;
1212         unsigned long stack_page;
1213         unsigned long vector_ptr;
1214
1215         /* Switch from the HYP stub to our own HYP init vector */
1216         __hyp_set_vectors(kvm_get_idmap_vector());
1217
1218         pgd_ptr = kvm_mmu_get_httbr();
1219         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1220         hyp_stack_ptr = stack_page + PAGE_SIZE;
1221         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1222
1223         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1224         __cpu_init_stage2();
1225
1226         kvm_arm_init_debug();
1227 }
1228
1229 static void cpu_hyp_reset(void)
1230 {
1231         if (!is_kernel_in_hyp_mode())
1232                 __hyp_reset_vectors();
1233 }
1234
1235 static void cpu_hyp_reinit(void)
1236 {
1237         cpu_hyp_reset();
1238
1239         if (is_kernel_in_hyp_mode()) {
1240                 /*
1241                  * __cpu_init_stage2() is safe to call even if the PM
1242                  * event was cancelled before the CPU was reset.
1243                  */
1244                 __cpu_init_stage2();
1245                 kvm_timer_init_vhe();
1246         } else {
1247                 cpu_init_hyp_mode(NULL);
1248         }
1249
1250         if (vgic_present)
1251                 kvm_vgic_init_cpu_hardware();
1252 }
1253
1254 static void _kvm_arch_hardware_enable(void *discard)
1255 {
1256         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1257                 cpu_hyp_reinit();
1258                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1259         }
1260 }
1261
1262 int kvm_arch_hardware_enable(void)
1263 {
1264         _kvm_arch_hardware_enable(NULL);
1265         return 0;
1266 }
1267
1268 static void _kvm_arch_hardware_disable(void *discard)
1269 {
1270         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1271                 cpu_hyp_reset();
1272                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1273         }
1274 }
1275
1276 void kvm_arch_hardware_disable(void)
1277 {
1278         _kvm_arch_hardware_disable(NULL);
1279 }
1280
1281 #ifdef CONFIG_CPU_PM
1282 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1283                                     unsigned long cmd,
1284                                     void *v)
1285 {
1286         /*
1287          * kvm_arm_hardware_enabled is left with its old value over
1288          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1289          * re-enable hyp.
1290          */
1291         switch (cmd) {
1292         case CPU_PM_ENTER:
1293                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1294                         /*
1295                          * don't update kvm_arm_hardware_enabled here
1296                          * so that the hardware will be re-enabled
1297                          * when we resume. See below.
1298                          */
1299                         cpu_hyp_reset();
1300
1301                 return NOTIFY_OK;
1302         case CPU_PM_ENTER_FAILED:
1303         case CPU_PM_EXIT:
1304                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1305                         /* The hardware was enabled before suspend. */
1306                         cpu_hyp_reinit();
1307
1308                 return NOTIFY_OK;
1309
1310         default:
1311                 return NOTIFY_DONE;
1312         }
1313 }
1314
1315 static struct notifier_block hyp_init_cpu_pm_nb = {
1316         .notifier_call = hyp_init_cpu_pm_notifier,
1317 };
1318
1319 static void __init hyp_cpu_pm_init(void)
1320 {
1321         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1322 }
1323 static void __init hyp_cpu_pm_exit(void)
1324 {
1325         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1326 }
1327 #else
1328 static inline void hyp_cpu_pm_init(void)
1329 {
1330 }
1331 static inline void hyp_cpu_pm_exit(void)
1332 {
1333 }
1334 #endif
1335
1336 static int init_common_resources(void)
1337 {
1338         /* set size of VMID supported by CPU */
1339         kvm_vmid_bits = kvm_get_vmid_bits();
1340         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1341
1342         return 0;
1343 }
1344
1345 static int init_subsystems(void)
1346 {
1347         int err = 0;
1348
1349         /*
1350          * Enable hardware so that subsystem initialisation can access EL2.
1351          */
1352         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1353
1354         /*
1355          * Register CPU lower-power notifier
1356          */
1357         hyp_cpu_pm_init();
1358
1359         /*
1360          * Init HYP view of VGIC
1361          */
1362         err = kvm_vgic_hyp_init();
1363         switch (err) {
1364         case 0:
1365                 vgic_present = true;
1366                 break;
1367         case -ENODEV:
1368         case -ENXIO:
1369                 vgic_present = false;
1370                 err = 0;
1371                 break;
1372         default:
1373                 goto out;
1374         }
1375
1376         /*
1377          * Init HYP architected timer support
1378          */
1379         err = kvm_timer_hyp_init(vgic_present);
1380         if (err)
1381                 goto out;
1382
1383         kvm_perf_init();
1384         kvm_coproc_table_init();
1385
1386 out:
1387         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1388
1389         return err;
1390 }
1391
1392 static void teardown_hyp_mode(void)
1393 {
1394         int cpu;
1395
1396         free_hyp_pgds();
1397         for_each_possible_cpu(cpu)
1398                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1399         hyp_cpu_pm_exit();
1400 }
1401
1402 /**
1403  * Inits Hyp-mode on all online CPUs
1404  */
1405 static int init_hyp_mode(void)
1406 {
1407         int cpu;
1408         int err = 0;
1409
1410         /*
1411          * Allocate Hyp PGD and setup Hyp identity mapping
1412          */
1413         err = kvm_mmu_init();
1414         if (err)
1415                 goto out_err;
1416
1417         /*
1418          * Allocate stack pages for Hypervisor-mode
1419          */
1420         for_each_possible_cpu(cpu) {
1421                 unsigned long stack_page;
1422
1423                 stack_page = __get_free_page(GFP_KERNEL);
1424                 if (!stack_page) {
1425                         err = -ENOMEM;
1426                         goto out_err;
1427                 }
1428
1429                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1430         }
1431
1432         /*
1433          * Map the Hyp-code called directly from the host
1434          */
1435         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1436                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1437         if (err) {
1438                 kvm_err("Cannot map world-switch code\n");
1439                 goto out_err;
1440         }
1441
1442         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1443                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1444         if (err) {
1445                 kvm_err("Cannot map rodata section\n");
1446                 goto out_err;
1447         }
1448
1449         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1450                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1451         if (err) {
1452                 kvm_err("Cannot map bss section\n");
1453                 goto out_err;
1454         }
1455
1456         err = kvm_map_vectors();
1457         if (err) {
1458                 kvm_err("Cannot map vectors\n");
1459                 goto out_err;
1460         }
1461
1462         /*
1463          * Map the Hyp stack pages
1464          */
1465         for_each_possible_cpu(cpu) {
1466                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1467                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1468                                           PAGE_HYP);
1469
1470                 if (err) {
1471                         kvm_err("Cannot map hyp stack\n");
1472                         goto out_err;
1473                 }
1474         }
1475
1476         for_each_possible_cpu(cpu) {
1477                 kvm_cpu_context_t *cpu_ctxt;
1478
1479                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1480                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1481
1482                 if (err) {
1483                         kvm_err("Cannot map host CPU state: %d\n", err);
1484                         goto out_err;
1485                 }
1486         }
1487
1488         return 0;
1489
1490 out_err:
1491         teardown_hyp_mode();
1492         kvm_err("error initializing Hyp mode: %d\n", err);
1493         return err;
1494 }
1495
1496 static void check_kvm_target_cpu(void *ret)
1497 {
1498         *(int *)ret = kvm_target_cpu();
1499 }
1500
1501 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1502 {
1503         struct kvm_vcpu *vcpu;
1504         int i;
1505
1506         mpidr &= MPIDR_HWID_BITMASK;
1507         kvm_for_each_vcpu(i, vcpu, kvm) {
1508                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1509                         return vcpu;
1510         }
1511         return NULL;
1512 }
1513
1514 bool kvm_arch_has_irq_bypass(void)
1515 {
1516         return true;
1517 }
1518
1519 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1520                                       struct irq_bypass_producer *prod)
1521 {
1522         struct kvm_kernel_irqfd *irqfd =
1523                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1524
1525         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1526                                           &irqfd->irq_entry);
1527 }
1528 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1529                                       struct irq_bypass_producer *prod)
1530 {
1531         struct kvm_kernel_irqfd *irqfd =
1532                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1533
1534         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1535                                      &irqfd->irq_entry);
1536 }
1537
1538 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1539 {
1540         struct kvm_kernel_irqfd *irqfd =
1541                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1542
1543         kvm_arm_halt_guest(irqfd->kvm);
1544 }
1545
1546 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1547 {
1548         struct kvm_kernel_irqfd *irqfd =
1549                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1550
1551         kvm_arm_resume_guest(irqfd->kvm);
1552 }
1553
1554 /**
1555  * Initialize Hyp-mode and memory mappings on all CPUs.
1556  */
1557 int kvm_arch_init(void *opaque)
1558 {
1559         int err;
1560         int ret, cpu;
1561         bool in_hyp_mode;
1562
1563         if (!is_hyp_mode_available()) {
1564                 kvm_info("HYP mode not available\n");
1565                 return -ENODEV;
1566         }
1567
1568         for_each_online_cpu(cpu) {
1569                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1570                 if (ret < 0) {
1571                         kvm_err("Error, CPU %d not supported!\n", cpu);
1572                         return -ENODEV;
1573                 }
1574         }
1575
1576         err = init_common_resources();
1577         if (err)
1578                 return err;
1579
1580         in_hyp_mode = is_kernel_in_hyp_mode();
1581
1582         if (!in_hyp_mode) {
1583                 err = init_hyp_mode();
1584                 if (err)
1585                         goto out_err;
1586         }
1587
1588         err = init_subsystems();
1589         if (err)
1590                 goto out_hyp;
1591
1592         if (in_hyp_mode)
1593                 kvm_info("VHE mode initialized successfully\n");
1594         else
1595                 kvm_info("Hyp mode initialized successfully\n");
1596
1597         return 0;
1598
1599 out_hyp:
1600         if (!in_hyp_mode)
1601                 teardown_hyp_mode();
1602 out_err:
1603         return err;
1604 }
1605
1606 /* NOP: Compiling as a module not supported */
1607 void kvm_arch_exit(void)
1608 {
1609         kvm_perf_teardown();
1610 }
1611
1612 static int arm_init(void)
1613 {
1614         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1615         return rc;
1616 }
1617
1618 module_init(arm_init);