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_RWLOCK(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         bool new_gen;
 477
 478         read_lock(&kvm_vmid_lock);
 479         new_gen = need_new_vmid_gen(kvm);
 480         read_unlock(&kvm_vmid_lock);
 481
 482         if (!new_gen)
 483                 return;
 484
 485         write_lock(&kvm_vmid_lock);
 486
 487         /*
 488          * We need to re-check the vmid_gen here to ensure that if another vcpu
 489          * already allocated a valid vmid for this vm, then this vcpu should
 490          * use the same vmid.
 491          */
 492         if (!need_new_vmid_gen(kvm)) {
 493                 write_unlock(&kvm_vmid_lock);
 494                 return;
 495         }
 496
 497         /* First user of a new VMID generation? */
 498         if (unlikely(kvm_next_vmid == 0)) {
 499                 atomic64_inc(&kvm_vmid_gen);
 500                 kvm_next_vmid = 1;
 501
 502                 /*
 503                  * On SMP we know no other CPUs can use this CPU's or each
 504                  * other's VMID after force_vm_exit returns since the
 505                  * kvm_vmid_lock blocks them from reentry to the guest.
 506                  */
 507                 force_vm_exit(cpu_all_mask);
 508                 /*
 509                  * Now broadcast TLB + ICACHE invalidation over the inner
 510                  * shareable domain to make sure all data structures are
 511                  * clean.
 512                  */
 513                 kvm_call_hyp(__kvm_flush_vm_context);
 514         }
 515
 516         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 517         kvm->arch.vmid = kvm_next_vmid;
 518         kvm_next_vmid++;
 519         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 520
 521         /* update vttbr to be used with the new vmid */
 522         pgd_phys = virt_to_phys(kvm->arch.pgd);
 523         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 524         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 525         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid;
 526
 527         write_unlock(&kvm_vmid_lock);
 528 }
 529
 530 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 531 {
 532         struct kvm *kvm = vcpu->kvm;
 533         int ret = 0;
 534
 535         if (likely(vcpu->arch.has_run_once))
 536                 return 0;
 537
 538         vcpu->arch.has_run_once = true;
 539
 540         if (likely(irqchip_in_kernel(kvm))) {
 541                 /*
 542                  * Map the VGIC hardware resources before running a vcpu the
 543                  * first time on this VM.
 544                  */
 545                 if (unlikely(!vgic_ready(kvm))) {
 546                         ret = kvm_vgic_map_resources(kvm);
 547                         if (ret)
 548                                 return ret;
 549                 }
 550         } else {
 551                 /*
 552                  * Tell the rest of the code that there are userspace irqchip
 553                  * VMs in the wild.
 554                  */
 555                 static_branch_inc(&userspace_irqchip_in_use);
 556         }
 557
 558         ret = kvm_timer_enable(vcpu);
 559         if (ret)
 560                 return ret;
 561
 562         ret = kvm_arm_pmu_v3_enable(vcpu);
 563
 564         return ret;
 565 }
 566
 567 bool kvm_arch_intc_initialized(struct kvm *kvm)
 568 {
 569         return vgic_initialized(kvm);
 570 }
 571
 572 void kvm_arm_halt_guest(struct kvm *kvm)
 573 {
 574         int i;
 575         struct kvm_vcpu *vcpu;
 576
 577         kvm_for_each_vcpu(i, vcpu, kvm)
 578                 vcpu->arch.pause = true;
 579         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 580 }
 581
 582 void kvm_arm_resume_guest(struct kvm *kvm)
 583 {
 584         int i;
 585         struct kvm_vcpu *vcpu;
 586
 587         kvm_for_each_vcpu(i, vcpu, kvm) {
 588                 vcpu->arch.pause = false;
 589                 swake_up(kvm_arch_vcpu_wq(vcpu));
 590         }
 591 }
 592
 593 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 594 {
 595         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 596
 597         swait_event_interruptible(*wq, ((!vcpu->arch.power_off) &&
 598                                        (!vcpu->arch.pause)));
 599
 600         if (vcpu->arch.power_off || vcpu->arch.pause) {
 601                 /* Awaken to handle a signal, request we sleep again later. */
 602                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 603         }
 604 }
 605
 606 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 607 {
 608         return vcpu->arch.target >= 0;
 609 }
 610
 611 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 612 {
 613         if (kvm_request_pending(vcpu)) {
 614                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 615                         vcpu_req_sleep(vcpu);
 616
 617                 /*
 618                  * Clear IRQ_PENDING requests that were made to guarantee
 619                  * that a VCPU sees new virtual interrupts.
 620                  */
 621                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 622         }
 623 }
 624
 625 /**
 626  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 627  * @vcpu:       The VCPU pointer
 628  * @run:        The kvm_run structure pointer used for userspace state exchange
 629  *
 630  * This function is called through the VCPU_RUN ioctl called from user space. It
 631  * will execute VM code in a loop until the time slice for the process is used
 632  * or some emulation is needed from user space in which case the function will
 633  * return with return value 0 and with the kvm_run structure filled in with the
 634  * required data for the requested emulation.
 635  */
 636 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 637 {
 638         int ret;
 639
 640         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 641                 return -ENOEXEC;
 642
 643         ret = kvm_vcpu_first_run_init(vcpu);
 644         if (ret)
 645                 return ret;
 646
 647         if (run->exit_reason == KVM_EXIT_MMIO) {
 648                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 649                 if (ret)
 650                         return ret;
 651                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
 652                         return 0;
 653         }
 654
 655         if (run->immediate_exit)
 656                 return -EINTR;
 657
 658         vcpu_load(vcpu);
 659
 660         kvm_sigset_activate(vcpu);
 661
 662         ret = 1;
 663         run->exit_reason = KVM_EXIT_UNKNOWN;
 664         while (ret > 0) {
 665                 /*
 666                  * Check conditions before entering the guest
 667                  */
 668                 cond_resched();
 669
 670                 update_vttbr(vcpu->kvm);
 671
 672                 check_vcpu_requests(vcpu);
 673
 674                 /*
 675                  * Preparing the interrupts to be injected also
 676                  * involves poking the GIC, which must be done in a
 677                  * non-preemptible context.
 678                  */
 679                 preempt_disable();
 680
 681                 /* Flush FP/SIMD state that can't survive guest entry/exit */
 682                 kvm_fpsimd_flush_cpu_state();
 683
 684                 kvm_pmu_flush_hwstate(vcpu);
 685
 686                 local_irq_disable();
 687
 688                 kvm_vgic_flush_hwstate(vcpu);
 689
 690                 /*
 691                  * Exit if we have a signal pending so that we can deliver the
 692                  * signal to user space.
 693                  */
 694                 if (signal_pending(current)) {
 695                         ret = -EINTR;
 696                         run->exit_reason = KVM_EXIT_INTR;
 697                 }
 698
 699                 /*
 700                  * If we're using a userspace irqchip, then check if we need
 701                  * to tell a userspace irqchip about timer or PMU level
 702                  * changes and if so, exit to userspace (the actual level
 703                  * state gets updated in kvm_timer_update_run and
 704                  * kvm_pmu_update_run below).
 705                  */
 706                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 707                         if (kvm_timer_should_notify_user(vcpu) ||
 708                             kvm_pmu_should_notify_user(vcpu)) {
 709                                 ret = -EINTR;
 710                                 run->exit_reason = KVM_EXIT_INTR;
 711                         }
 712                 }
 713
 714                 /*
 715                  * Ensure we set mode to IN_GUEST_MODE after we disable
 716                  * interrupts and before the final VCPU requests check.
 717                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 718                  * Documentation/virtual/kvm/vcpu-requests.rst
 719                  */
 720                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 721
 722                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 723                     kvm_request_pending(vcpu)) {
 724                         vcpu->mode = OUTSIDE_GUEST_MODE;
 725                         isb(); /* Ensure work in x_flush_hwstate is committed */
 726                         kvm_pmu_sync_hwstate(vcpu);
 727                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 728                                 kvm_timer_sync_hwstate(vcpu);
 729                         kvm_vgic_sync_hwstate(vcpu);
 730                         local_irq_enable();
 731                         preempt_enable();
 732                         continue;
 733                 }
 734
 735                 kvm_arm_setup_debug(vcpu);
 736
 737                 /**************************************************************
 738                  * Enter the guest
 739                  */
 740                 trace_kvm_entry(*vcpu_pc(vcpu));
 741                 guest_enter_irqoff();
 742
 743                 if (has_vhe()) {
 744                         kvm_arm_vhe_guest_enter();
 745                         ret = kvm_vcpu_run_vhe(vcpu);
 746                         kvm_arm_vhe_guest_exit();
 747                 } else {
 748                         ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
 749                 }
 750
 751                 vcpu->mode = OUTSIDE_GUEST_MODE;
 752                 vcpu->stat.exits++;
 753                 /*
 754                  * Back from guest
 755                  *************************************************************/
 756
 757                 kvm_arm_clear_debug(vcpu);
 758
 759                 /*
 760                  * We must sync the PMU state before the vgic state so
 761                  * that the vgic can properly sample the updated state of the
 762                  * interrupt line.
 763                  */
 764                 kvm_pmu_sync_hwstate(vcpu);
 765
 766                 /*
 767                  * Sync the vgic state before syncing the timer state because
 768                  * the timer code needs to know if the virtual timer
 769                  * interrupts are active.
 770                  */
 771                 kvm_vgic_sync_hwstate(vcpu);
 772
 773                 /*
 774                  * Sync the timer hardware state before enabling interrupts as
 775                  * we don't want vtimer interrupts to race with syncing the
 776                  * timer virtual interrupt state.
 777                  */
 778                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 779                         kvm_timer_sync_hwstate(vcpu);
 780
 781                 /*
 782                  * We may have taken a host interrupt in HYP mode (ie
 783                  * while executing the guest). This interrupt is still
 784                  * pending, as we haven't serviced it yet!
 785                  *
 786                  * We're now back in SVC mode, with interrupts
 787                  * disabled.  Enabling the interrupts now will have
 788                  * the effect of taking the interrupt again, in SVC
 789                  * mode this time.
 790                  */
 791                 local_irq_enable();
 792
 793                 /*
 794                  * We do local_irq_enable() before calling guest_exit() so
 795                  * that if a timer interrupt hits while running the guest we
 796                  * account that tick as being spent in the guest.  We enable
 797                  * preemption after calling guest_exit() so that if we get
 798                  * preempted we make sure ticks after that is not counted as
 799                  * guest time.
 800                  */
 801                 guest_exit();
 802                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 803
 804                 /* Exit types that need handling before we can be preempted */
 805                 handle_exit_early(vcpu, run, ret);
 806
 807                 preempt_enable();
 808
 809                 ret = handle_exit(vcpu, run, ret);
 810         }
 811
 812         /* Tell userspace about in-kernel device output levels */
 813         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 814                 kvm_timer_update_run(vcpu);
 815                 kvm_pmu_update_run(vcpu);
 816         }
 817
 818         kvm_sigset_deactivate(vcpu);
 819
 820         vcpu_put(vcpu);
 821         return ret;
 822 }
 823
 824 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 825 {
 826         int bit_index;
 827         bool set;
 828         unsigned long *hcr;
 829
 830         if (number == KVM_ARM_IRQ_CPU_IRQ)
 831                 bit_index = __ffs(HCR_VI);
 832         else /* KVM_ARM_IRQ_CPU_FIQ */
 833                 bit_index = __ffs(HCR_VF);
 834
 835         hcr = vcpu_hcr(vcpu);
 836         if (level)
 837                 set = test_and_set_bit(bit_index, hcr);
 838         else
 839                 set = test_and_clear_bit(bit_index, hcr);
 840
 841         /*
 842          * If we didn't change anything, no need to wake up or kick other CPUs
 843          */
 844         if (set == level)
 845                 return 0;
 846
 847         /*
 848          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 849          * trigger a world-switch round on the running physical CPU to set the
 850          * virtual IRQ/FIQ fields in the HCR appropriately.
 851          */
 852         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 853         kvm_vcpu_kick(vcpu);
 854
 855         return 0;
 856 }
 857
 858 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 859                           bool line_status)
 860 {
 861         u32 irq = irq_level->irq;
 862         unsigned int irq_type, vcpu_idx, irq_num;
 863         int nrcpus = atomic_read(&kvm->online_vcpus);
 864         struct kvm_vcpu *vcpu = NULL;
 865         bool level = irq_level->level;
 866
 867         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 868         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 869         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 870
 871         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 872
 873         switch (irq_type) {
 874         case KVM_ARM_IRQ_TYPE_CPU:
 875                 if (irqchip_in_kernel(kvm))
 876                         return -ENXIO;
 877
 878                 if (vcpu_idx >= nrcpus)
 879                         return -EINVAL;
 880
 881                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 882                 if (!vcpu)
 883                         return -EINVAL;
 884
 885                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 886                         return -EINVAL;
 887
 888                 return vcpu_interrupt_line(vcpu, irq_num, level);
 889         case KVM_ARM_IRQ_TYPE_PPI:
 890                 if (!irqchip_in_kernel(kvm))
 891                         return -ENXIO;
 892
 893                 if (vcpu_idx >= nrcpus)
 894                         return -EINVAL;
 895
 896                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 897                 if (!vcpu)
 898                         return -EINVAL;
 899
 900                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 901                         return -EINVAL;
 902
 903                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 904         case KVM_ARM_IRQ_TYPE_SPI:
 905                 if (!irqchip_in_kernel(kvm))
 906                         return -ENXIO;
 907
 908                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 909                         return -EINVAL;
 910
 911                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 912         }
 913
 914         return -EINVAL;
 915 }
 916
 917 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 918                                const struct kvm_vcpu_init *init)
 919 {
 920         unsigned int i;
 921         int phys_target = kvm_target_cpu();
 922
 923         if (init->target != phys_target)
 924                 return -EINVAL;
 925
 926         /*
 927          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 928          * use the same target.
 929          */
 930         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 931                 return -EINVAL;
 932
 933         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 934         for (i = 0; i < sizeof(init->features) * 8; i++) {
 935                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 936
 937                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 938                         return -ENOENT;
 939
 940                 /*
 941                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 942                  * use the same feature set.
 943                  */
 944                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 945                     test_bit(i, vcpu->arch.features) != set)
 946                         return -EINVAL;
 947
 948                 if (set)
 949                         set_bit(i, vcpu->arch.features);
 950         }
 951
 952         vcpu->arch.target = phys_target;
 953
 954         /* Now we know what it is, we can reset it. */
 955         return kvm_reset_vcpu(vcpu);
 956 }
 957
 958
 959 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 960                                          struct kvm_vcpu_init *init)
 961 {
 962         int ret;
 963
 964         ret = kvm_vcpu_set_target(vcpu, init);
 965         if (ret)
 966                 return ret;
 967
 968         /*
 969          * Ensure a rebooted VM will fault in RAM pages and detect if the
 970          * guest MMU is turned off and flush the caches as needed.
 971          */
 972         if (vcpu->arch.has_run_once)
 973                 stage2_unmap_vm(vcpu->kvm);
 974
 975         vcpu_reset_hcr(vcpu);
 976
 977         /*
 978          * Handle the "start in power-off" case.
 979          */
 980         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 981                 vcpu_power_off(vcpu);
 982         else
 983                 vcpu->arch.power_off = false;
 984
 985         return 0;
 986 }
 987
 988 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
 989                                  struct kvm_device_attr *attr)
 990 {
 991         int ret = -ENXIO;
 992
 993         switch (attr->group) {
 994         default:
 995                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
 996                 break;
 997         }
 998
 999         return ret;
1000 }
1001
1002 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1003                                  struct kvm_device_attr *attr)
1004 {
1005         int ret = -ENXIO;
1006
1007         switch (attr->group) {
1008         default:
1009                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1010                 break;
1011         }
1012
1013         return ret;
1014 }
1015
1016 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1017                                  struct kvm_device_attr *attr)
1018 {
1019         int ret = -ENXIO;
1020
1021         switch (attr->group) {
1022         default:
1023                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1024                 break;
1025         }
1026
1027         return ret;
1028 }
1029
1030 long kvm_arch_vcpu_ioctl(struct file *filp,
1031                          unsigned int ioctl, unsigned long arg)
1032 {
1033         struct kvm_vcpu *vcpu = filp->private_data;
1034         void __user *argp = (void __user *)arg;
1035         struct kvm_device_attr attr;
1036         long r;
1037
1038         switch (ioctl) {
1039         case KVM_ARM_VCPU_INIT: {
1040                 struct kvm_vcpu_init init;
1041
1042                 r = -EFAULT;
1043                 if (copy_from_user(&init, argp, sizeof(init)))
1044                         break;
1045
1046                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1047                 break;
1048         }
1049         case KVM_SET_ONE_REG:
1050         case KVM_GET_ONE_REG: {
1051                 struct kvm_one_reg reg;
1052
1053                 r = -ENOEXEC;
1054                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1055                         break;
1056
1057                 r = -EFAULT;
1058                 if (copy_from_user(&reg, argp, sizeof(reg)))
1059                         break;
1060
1061                 if (ioctl == KVM_SET_ONE_REG)
1062                         r = kvm_arm_set_reg(vcpu, &reg);
1063                 else
1064                         r = kvm_arm_get_reg(vcpu, &reg);
1065                 break;
1066         }
1067         case KVM_GET_REG_LIST: {
1068                 struct kvm_reg_list __user *user_list = argp;
1069                 struct kvm_reg_list reg_list;
1070                 unsigned n;
1071
1072                 r = -ENOEXEC;
1073                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1074                         break;
1075
1076                 r = -EFAULT;
1077                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1078                         break;
1079                 n = reg_list.n;
1080                 reg_list.n = kvm_arm_num_regs(vcpu);
1081                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1082                         break;
1083                 r = -E2BIG;
1084                 if (n < reg_list.n)
1085                         break;
1086                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1087                 break;
1088         }
1089         case KVM_SET_DEVICE_ATTR: {
1090                 r = -EFAULT;
1091                 if (copy_from_user(&attr, argp, sizeof(attr)))
1092                         break;
1093                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1094                 break;
1095         }
1096         case KVM_GET_DEVICE_ATTR: {
1097                 r = -EFAULT;
1098                 if (copy_from_user(&attr, argp, sizeof(attr)))
1099                         break;
1100                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1101                 break;
1102         }
1103         case KVM_HAS_DEVICE_ATTR: {
1104                 r = -EFAULT;
1105                 if (copy_from_user(&attr, argp, sizeof(attr)))
1106                         break;
1107                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1108                 break;
1109         }
1110         default:
1111                 r = -EINVAL;
1112         }
1113
1114         return r;
1115 }
1116
1117 /**
1118  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1119  * @kvm: kvm instance
1120  * @log: slot id and address to which we copy the log
1121  *
1122  * Steps 1-4 below provide general overview of dirty page logging. See
1123  * kvm_get_dirty_log_protect() function description for additional details.
1124  *
1125  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1126  * always flush the TLB (step 4) even if previous step failed  and the dirty
1127  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1128  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1129  * writes will be marked dirty for next log read.
1130  *
1131  *   1. Take a snapshot of the bit and clear it if needed.
1132  *   2. Write protect the corresponding page.
1133  *   3. Copy the snapshot to the userspace.
1134  *   4. Flush TLB's if needed.
1135  */
1136 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1137 {
1138         bool is_dirty = false;
1139         int r;
1140
1141         mutex_lock(&kvm->slots_lock);
1142
1143         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1144
1145         if (is_dirty)
1146                 kvm_flush_remote_tlbs(kvm);
1147
1148         mutex_unlock(&kvm->slots_lock);
1149         return r;
1150 }
1151
1152 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1153                                         struct kvm_arm_device_addr *dev_addr)
1154 {
1155         unsigned long dev_id, type;
1156
1157         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1158                 KVM_ARM_DEVICE_ID_SHIFT;
1159         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1160                 KVM_ARM_DEVICE_TYPE_SHIFT;
1161
1162         switch (dev_id) {
1163         case KVM_ARM_DEVICE_VGIC_V2:
1164                 if (!vgic_present)
1165                         return -ENXIO;
1166                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1167         default:
1168                 return -ENODEV;
1169         }
1170 }
1171
1172 long kvm_arch_vm_ioctl(struct file *filp,
1173                        unsigned int ioctl, unsigned long arg)
1174 {
1175         struct kvm *kvm = filp->private_data;
1176         void __user *argp = (void __user *)arg;
1177
1178         switch (ioctl) {
1179         case KVM_CREATE_IRQCHIP: {
1180                 int ret;
1181                 if (!vgic_present)
1182                         return -ENXIO;
1183                 mutex_lock(&kvm->lock);
1184                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1185                 mutex_unlock(&kvm->lock);
1186                 return ret;
1187         }
1188         case KVM_ARM_SET_DEVICE_ADDR: {
1189                 struct kvm_arm_device_addr dev_addr;
1190
1191                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1192                         return -EFAULT;
1193                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1194         }
1195         case KVM_ARM_PREFERRED_TARGET: {
1196                 int err;
1197                 struct kvm_vcpu_init init;
1198
1199                 err = kvm_vcpu_preferred_target(&init);
1200                 if (err)
1201                         return err;
1202
1203                 if (copy_to_user(argp, &init, sizeof(init)))
1204                         return -EFAULT;
1205
1206                 return 0;
1207         }
1208         default:
1209                 return -EINVAL;
1210         }
1211 }
1212
1213 static void cpu_init_hyp_mode(void *dummy)
1214 {
1215         phys_addr_t pgd_ptr;
1216         unsigned long hyp_stack_ptr;
1217         unsigned long stack_page;
1218         unsigned long vector_ptr;
1219
1220         /* Switch from the HYP stub to our own HYP init vector */
1221         __hyp_set_vectors(kvm_get_idmap_vector());
1222
1223         pgd_ptr = kvm_mmu_get_httbr();
1224         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1225         hyp_stack_ptr = stack_page + PAGE_SIZE;
1226         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1227
1228         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1229         __cpu_init_stage2();
1230
1231         kvm_arm_init_debug();
1232 }
1233
1234 static void cpu_hyp_reset(void)
1235 {
1236         if (!is_kernel_in_hyp_mode())
1237                 __hyp_reset_vectors();
1238 }
1239
1240 static void cpu_hyp_reinit(void)
1241 {
1242         cpu_hyp_reset();
1243
1244         if (is_kernel_in_hyp_mode()) {
1245                 /*
1246                  * __cpu_init_stage2() is safe to call even if the PM
1247                  * event was cancelled before the CPU was reset.
1248                  */
1249                 __cpu_init_stage2();
1250                 kvm_timer_init_vhe();
1251         } else {
1252                 cpu_init_hyp_mode(NULL);
1253         }
1254
1255         if (vgic_present)
1256                 kvm_vgic_init_cpu_hardware();
1257 }
1258
1259 static void _kvm_arch_hardware_enable(void *discard)
1260 {
1261         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1262                 cpu_hyp_reinit();
1263                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1264         }
1265 }
1266
1267 int kvm_arch_hardware_enable(void)
1268 {
1269         _kvm_arch_hardware_enable(NULL);
1270         return 0;
1271 }
1272
1273 static void _kvm_arch_hardware_disable(void *discard)
1274 {
1275         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1276                 cpu_hyp_reset();
1277                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1278         }
1279 }
1280
1281 void kvm_arch_hardware_disable(void)
1282 {
1283         _kvm_arch_hardware_disable(NULL);
1284 }
1285
1286 #ifdef CONFIG_CPU_PM
1287 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1288                                     unsigned long cmd,
1289                                     void *v)
1290 {
1291         /*
1292          * kvm_arm_hardware_enabled is left with its old value over
1293          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1294          * re-enable hyp.
1295          */
1296         switch (cmd) {
1297         case CPU_PM_ENTER:
1298                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1299                         /*
1300                          * don't update kvm_arm_hardware_enabled here
1301                          * so that the hardware will be re-enabled
1302                          * when we resume. See below.
1303                          */
1304                         cpu_hyp_reset();
1305
1306                 return NOTIFY_OK;
1307         case CPU_PM_ENTER_FAILED:
1308         case CPU_PM_EXIT:
1309                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1310                         /* The hardware was enabled before suspend. */
1311                         cpu_hyp_reinit();
1312
1313                 return NOTIFY_OK;
1314
1315         default:
1316                 return NOTIFY_DONE;
1317         }
1318 }
1319
1320 static struct notifier_block hyp_init_cpu_pm_nb = {
1321         .notifier_call = hyp_init_cpu_pm_notifier,
1322 };
1323
1324 static void __init hyp_cpu_pm_init(void)
1325 {
1326         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1327 }
1328 static void __init hyp_cpu_pm_exit(void)
1329 {
1330         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1331 }
1332 #else
1333 static inline void hyp_cpu_pm_init(void)
1334 {
1335 }
1336 static inline void hyp_cpu_pm_exit(void)
1337 {
1338 }
1339 #endif
1340
1341 static int init_common_resources(void)
1342 {
1343         /* set size of VMID supported by CPU */
1344         kvm_vmid_bits = kvm_get_vmid_bits();
1345         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1346
1347         return 0;
1348 }
1349
1350 static int init_subsystems(void)
1351 {
1352         int err = 0;
1353
1354         /*
1355          * Enable hardware so that subsystem initialisation can access EL2.
1356          */
1357         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1358
1359         /*
1360          * Register CPU lower-power notifier
1361          */
1362         hyp_cpu_pm_init();
1363
1364         /*
1365          * Init HYP view of VGIC
1366          */
1367         err = kvm_vgic_hyp_init();
1368         switch (err) {
1369         case 0:
1370                 vgic_present = true;
1371                 break;
1372         case -ENODEV:
1373         case -ENXIO:
1374                 vgic_present = false;
1375                 err = 0;
1376                 break;
1377         default:
1378                 goto out;
1379         }
1380
1381         /*
1382          * Init HYP architected timer support
1383          */
1384         err = kvm_timer_hyp_init(vgic_present);
1385         if (err)
1386                 goto out;
1387
1388         kvm_perf_init();
1389         kvm_coproc_table_init();
1390
1391 out:
1392         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1393
1394         return err;
1395 }
1396
1397 static void teardown_hyp_mode(void)
1398 {
1399         int cpu;
1400
1401         free_hyp_pgds();
1402         for_each_possible_cpu(cpu)
1403                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1404         hyp_cpu_pm_exit();
1405 }
1406
1407 /**
1408  * Inits Hyp-mode on all online CPUs
1409  */
1410 static int init_hyp_mode(void)
1411 {
1412         int cpu;
1413         int err = 0;
1414
1415         /*
1416          * Allocate Hyp PGD and setup Hyp identity mapping
1417          */
1418         err = kvm_mmu_init();
1419         if (err)
1420                 goto out_err;
1421
1422         /*
1423          * Allocate stack pages for Hypervisor-mode
1424          */
1425         for_each_possible_cpu(cpu) {
1426                 unsigned long stack_page;
1427
1428                 stack_page = __get_free_page(GFP_KERNEL);
1429                 if (!stack_page) {
1430                         err = -ENOMEM;
1431                         goto out_err;
1432                 }
1433
1434                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1435         }
1436
1437         /*
1438          * Map the Hyp-code called directly from the host
1439          */
1440         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1441                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1442         if (err) {
1443                 kvm_err("Cannot map world-switch code\n");
1444                 goto out_err;
1445         }
1446
1447         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1448                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1449         if (err) {
1450                 kvm_err("Cannot map rodata section\n");
1451                 goto out_err;
1452         }
1453
1454         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1455                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1456         if (err) {
1457                 kvm_err("Cannot map bss section\n");
1458                 goto out_err;
1459         }
1460
1461         err = kvm_map_vectors();
1462         if (err) {
1463                 kvm_err("Cannot map vectors\n");
1464                 goto out_err;
1465         }
1466
1467         /*
1468          * Map the Hyp stack pages
1469          */
1470         for_each_possible_cpu(cpu) {
1471                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1472                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1473                                           PAGE_HYP);
1474
1475                 if (err) {
1476                         kvm_err("Cannot map hyp stack\n");
1477                         goto out_err;
1478                 }
1479         }
1480
1481         for_each_possible_cpu(cpu) {
1482                 kvm_cpu_context_t *cpu_ctxt;
1483
1484                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1485                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1486
1487                 if (err) {
1488                         kvm_err("Cannot map host CPU state: %d\n", err);
1489                         goto out_err;
1490                 }
1491         }
1492
1493         err = hyp_map_aux_data();
1494         if (err)
1495                 kvm_err("Cannot map host auxilary data: %d\n", err);
1496
1497         return 0;
1498
1499 out_err:
1500         teardown_hyp_mode();
1501         kvm_err("error initializing Hyp mode: %d\n", err);
1502         return err;
1503 }
1504
1505 static void check_kvm_target_cpu(void *ret)
1506 {
1507         *(int *)ret = kvm_target_cpu();
1508 }
1509
1510 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1511 {
1512         struct kvm_vcpu *vcpu;
1513         int i;
1514
1515         mpidr &= MPIDR_HWID_BITMASK;
1516         kvm_for_each_vcpu(i, vcpu, kvm) {
1517                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1518                         return vcpu;
1519         }
1520         return NULL;
1521 }
1522
1523 bool kvm_arch_has_irq_bypass(void)
1524 {
1525         return true;
1526 }
1527
1528 int kvm_arch_irq_bypass_add_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         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1535                                           &irqfd->irq_entry);
1536 }
1537 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1538                                       struct irq_bypass_producer *prod)
1539 {
1540         struct kvm_kernel_irqfd *irqfd =
1541                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1542
1543         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1544                                      &irqfd->irq_entry);
1545 }
1546
1547 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1548 {
1549         struct kvm_kernel_irqfd *irqfd =
1550                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1551
1552         kvm_arm_halt_guest(irqfd->kvm);
1553 }
1554
1555 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1556 {
1557         struct kvm_kernel_irqfd *irqfd =
1558                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1559
1560         kvm_arm_resume_guest(irqfd->kvm);
1561 }
1562
1563 /**
1564  * Initialize Hyp-mode and memory mappings on all CPUs.
1565  */
1566 int kvm_arch_init(void *opaque)
1567 {
1568         int err;
1569         int ret, cpu;
1570         bool in_hyp_mode;
1571
1572         if (!is_hyp_mode_available()) {
1573                 kvm_info("HYP mode not available\n");
1574                 return -ENODEV;
1575         }
1576
1577         for_each_online_cpu(cpu) {
1578                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1579                 if (ret < 0) {
1580                         kvm_err("Error, CPU %d not supported!\n", cpu);
1581                         return -ENODEV;
1582                 }
1583         }
1584
1585         err = init_common_resources();
1586         if (err)
1587                 return err;
1588
1589         in_hyp_mode = is_kernel_in_hyp_mode();
1590
1591         if (!in_hyp_mode) {
1592                 err = init_hyp_mode();
1593                 if (err)
1594                         goto out_err;
1595         }
1596
1597         err = init_subsystems();
1598         if (err)
1599                 goto out_hyp;
1600
1601         if (in_hyp_mode)
1602                 kvm_info("VHE mode initialized successfully\n");
1603         else
1604                 kvm_info("Hyp mode initialized successfully\n");
1605
1606         return 0;
1607
1608 out_hyp:
1609         if (!in_hyp_mode)
1610                 teardown_hyp_mode();
1611 out_err:
1612         return err;
1613 }
1614
1615 /* NOP: Compiling as a module not supported */
1616 void kvm_arch_exit(void)
1617 {
1618         kvm_perf_teardown();
1619 }
1620
1621 static int arm_init(void)
1622 {
1623         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1624         return rc;
1625 }
1626
1627 module_init(arm_init);