virt/kvm/arm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/bug.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/list.h>
  25 #include <linux/module.h>
  26 #include <linux/vmalloc.h>
  27 #include <linux/fs.h>
  28 #include <linux/mman.h>
  29 #include <linux/sched.h>
  30 #include <linux/kvm.h>
  31 #include <linux/kvm_irqfd.h>
  32 #include <linux/irqbypass.h>
  33 #include <linux/sched/stat.h>
  34 #include <trace/events/kvm.h>
  35 #include <kvm/arm_pmu.h>
  36 #include <kvm/arm_psci.h>
  37
  38 #define CREATE_TRACE_POINTS
  39 #include "trace.h"
  40
  41 #include <linux/uaccess.h>
  42 #include <asm/ptrace.h>
  43 #include <asm/mman.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/cacheflush.h>
  46 #include <asm/cpufeature.h>
  47 #include <asm/virt.h>
  48 #include <asm/kvm_arm.h>
  49 #include <asm/kvm_asm.h>
  50 #include <asm/kvm_mmu.h>
  51 #include <asm/kvm_emulate.h>
  52 #include <asm/kvm_coproc.h>
  53 #include <asm/sections.h>
  54
  55 #ifdef REQUIRES_VIRT
  56 __asm__(".arch_extension        virt");
  57 #endif
  58
  59 DEFINE_PER_CPU(kvm_cpu_context_t, kvm_host_cpu_state);
  60 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  61
  62 /* Per-CPU variable containing the currently running vcpu. */
  63 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  64
  65 /* The VMID used in the VTTBR */
  66 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  67 static u32 kvm_next_vmid;
  68 static unsigned int kvm_vmid_bits __read_mostly;
  69 static DEFINE_RWLOCK(kvm_vmid_lock);
  70
  71 static bool vgic_present;
  72
  73 static DEFINE_PER_CPU(unsigned char, kvm_arm_hardware_enabled);
  74
  75 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  76 {
  77         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  78 }
  79
  80 DEFINE_STATIC_KEY_FALSE(userspace_irqchip_in_use);
  81
  82 /**
  83  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  84  * Must be called from non-preemptible context
  85  */
  86 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  87 {
  88         return __this_cpu_read(kvm_arm_running_vcpu);
  89 }
  90
  91 /**
  92  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  93  */
  94 struct kvm_vcpu * __percpu *kvm_get_running_vcpus(void)
  95 {
  96         return &kvm_arm_running_vcpu;
  97 }
  98
  99 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
 100 {
 101         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_check_processor_compat(void *rtn)
 110 {
 111         *(int *)rtn = 0;
 112 }
 113
 114
 115 /**
 116  * kvm_arch_init_vm - initializes a VM data structure
 117  * @kvm:        pointer to the KVM struct
 118  */
 119 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 120 {
 121         int ret, cpu;
 122
 123         ret = kvm_arm_setup_stage2(kvm, type);
 124         if (ret)
 125                 return ret;
 126
 127         kvm->arch.last_vcpu_ran = alloc_percpu(typeof(*kvm->arch.last_vcpu_ran));
 128         if (!kvm->arch.last_vcpu_ran)
 129                 return -ENOMEM;
 130
 131         for_each_possible_cpu(cpu)
 132                 *per_cpu_ptr(kvm->arch.last_vcpu_ran, cpu) = -1;
 133
 134         ret = kvm_alloc_stage2_pgd(kvm);
 135         if (ret)
 136                 goto out_fail_alloc;
 137
 138         ret = create_hyp_mappings(kvm, kvm + 1, PAGE_HYP);
 139         if (ret)
 140                 goto out_free_stage2_pgd;
 141
 142         kvm_vgic_early_init(kvm);
 143
 144         /* Mark the initial VMID generation invalid */
 145         kvm->arch.vmid_gen = 0;
 146
 147         /* The maximum number of VCPUs is limited by the host's GIC model */
 148         kvm->arch.max_vcpus = vgic_present ?
 149                                 kvm_vgic_get_max_vcpus() : KVM_MAX_VCPUS;
 150
 151         return ret;
 152 out_free_stage2_pgd:
 153         kvm_free_stage2_pgd(kvm);
 154 out_fail_alloc:
 155         free_percpu(kvm->arch.last_vcpu_ran);
 156         kvm->arch.last_vcpu_ran = NULL;
 157         return ret;
 158 }
 159
 160 bool kvm_arch_has_vcpu_debugfs(void)
 161 {
 162         return false;
 163 }
 164
 165 int kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu)
 166 {
 167         return 0;
 168 }
 169
 170 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 171 {
 172         return VM_FAULT_SIGBUS;
 173 }
 174
 175
 176 /**
 177  * kvm_arch_destroy_vm - destroy the VM data structure
 178  * @kvm:        pointer to the KVM struct
 179  */
 180 void kvm_arch_destroy_vm(struct kvm *kvm)
 181 {
 182         int i;
 183
 184         kvm_vgic_destroy(kvm);
 185
 186         free_percpu(kvm->arch.last_vcpu_ran);
 187         kvm->arch.last_vcpu_ran = NULL;
 188
 189         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 190                 if (kvm->vcpus[i]) {
 191                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 192                         kvm->vcpus[i] = NULL;
 193                 }
 194         }
 195         atomic_set(&kvm->online_vcpus, 0);
 196 }
 197
 198 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 199 {
 200         int r;
 201         switch (ext) {
 202         case KVM_CAP_IRQCHIP:
 203                 r = vgic_present;
 204                 break;
 205         case KVM_CAP_IOEVENTFD:
 206         case KVM_CAP_DEVICE_CTRL:
 207         case KVM_CAP_USER_MEMORY:
 208         case KVM_CAP_SYNC_MMU:
 209         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 210         case KVM_CAP_ONE_REG:
 211         case KVM_CAP_ARM_PSCI:
 212         case KVM_CAP_ARM_PSCI_0_2:
 213         case KVM_CAP_READONLY_MEM:
 214         case KVM_CAP_MP_STATE:
 215         case KVM_CAP_IMMEDIATE_EXIT:
 216         case KVM_CAP_VCPU_EVENTS:
 217                 r = 1;
 218                 break;
 219         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 220                 r = 1;
 221                 break;
 222         case KVM_CAP_NR_VCPUS:
 223                 r = num_online_cpus();
 224                 break;
 225         case KVM_CAP_MAX_VCPUS:
 226                 r = KVM_MAX_VCPUS;
 227                 break;
 228         case KVM_CAP_NR_MEMSLOTS:
 229                 r = KVM_USER_MEM_SLOTS;
 230                 break;
 231         case KVM_CAP_MSI_DEVID:
 232                 if (!kvm)
 233                         r = -EINVAL;
 234                 else
 235                         r = kvm->arch.vgic.msis_require_devid;
 236                 break;
 237         case KVM_CAP_ARM_USER_IRQ:
 238                 /*
 239                  * 1: EL1_VTIMER, EL1_PTIMER, and PMU.
 240                  * (bump this number if adding more devices)
 241                  */
 242                 r = 1;
 243                 break;
 244         default:
 245                 r = kvm_arch_vm_ioctl_check_extension(kvm, ext);
 246                 break;
 247         }
 248         return r;
 249 }
 250
 251 long kvm_arch_dev_ioctl(struct file *filp,
 252                         unsigned int ioctl, unsigned long arg)
 253 {
 254         return -EINVAL;
 255 }
 256
 257 struct kvm *kvm_arch_alloc_vm(void)
 258 {
 259         if (!has_vhe())
 260                 return kzalloc(sizeof(struct kvm), GFP_KERNEL);
 261
 262         return vzalloc(sizeof(struct kvm));
 263 }
 264
 265 void kvm_arch_free_vm(struct kvm *kvm)
 266 {
 267         if (!has_vhe())
 268                 kfree(kvm);
 269         else
 270                 vfree(kvm);
 271 }
 272
 273 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 274 {
 275         int err;
 276         struct kvm_vcpu *vcpu;
 277
 278         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 279                 err = -EBUSY;
 280                 goto out;
 281         }
 282
 283         if (id >= kvm->arch.max_vcpus) {
 284                 err = -EINVAL;
 285                 goto out;
 286         }
 287
 288         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 289         if (!vcpu) {
 290                 err = -ENOMEM;
 291                 goto out;
 292         }
 293
 294         err = kvm_vcpu_init(vcpu, kvm, id);
 295         if (err)
 296                 goto free_vcpu;
 297
 298         err = create_hyp_mappings(vcpu, vcpu + 1, PAGE_HYP);
 299         if (err)
 300                 goto vcpu_uninit;
 301
 302         return vcpu;
 303 vcpu_uninit:
 304         kvm_vcpu_uninit(vcpu);
 305 free_vcpu:
 306         kmem_cache_free(kvm_vcpu_cache, vcpu);
 307 out:
 308         return ERR_PTR(err);
 309 }
 310
 311 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 312 {
 313 }
 314
 315 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 316 {
 317         if (vcpu->arch.has_run_once && unlikely(!irqchip_in_kernel(vcpu->kvm)))
 318                 static_branch_dec(&userspace_irqchip_in_use);
 319
 320         kvm_mmu_free_memory_caches(vcpu);
 321         kvm_timer_vcpu_terminate(vcpu);
 322         kvm_pmu_vcpu_destroy(vcpu);
 323         kvm_vcpu_uninit(vcpu);
 324         kmem_cache_free(kvm_vcpu_cache, vcpu);
 325 }
 326
 327 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 328 {
 329         kvm_arch_vcpu_free(vcpu);
 330 }
 331
 332 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 333 {
 334         return kvm_timer_is_pending(vcpu);
 335 }
 336
 337 void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu)
 338 {
 339         kvm_timer_schedule(vcpu);
 340         kvm_vgic_v4_enable_doorbell(vcpu);
 341 }
 342
 343 void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu)
 344 {
 345         kvm_timer_unschedule(vcpu);
 346         kvm_vgic_v4_disable_doorbell(vcpu);
 347 }
 348
 349 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 350 {
 351         /* Force users to call KVM_ARM_VCPU_INIT */
 352         vcpu->arch.target = -1;
 353         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 354
 355         /* Set up the timer */
 356         kvm_timer_vcpu_init(vcpu);
 357
 358         kvm_arm_reset_debug_ptr(vcpu);
 359
 360         return kvm_vgic_vcpu_init(vcpu);
 361 }
 362
 363 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 364 {
 365         int *last_ran;
 366
 367         last_ran = this_cpu_ptr(vcpu->kvm->arch.last_vcpu_ran);
 368
 369         /*
 370          * We might get preempted before the vCPU actually runs, but
 371          * over-invalidation doesn't affect correctness.
 372          */
 373         if (*last_ran != vcpu->vcpu_id) {
 374                 kvm_call_hyp(__kvm_tlb_flush_local_vmid, vcpu);
 375                 *last_ran = vcpu->vcpu_id;
 376         }
 377
 378         vcpu->cpu = cpu;
 379         vcpu->arch.host_cpu_context = this_cpu_ptr(&kvm_host_cpu_state);
 380
 381         kvm_arm_set_running_vcpu(vcpu);
 382         kvm_vgic_load(vcpu);
 383         kvm_timer_vcpu_load(vcpu);
 384         kvm_vcpu_load_sysregs(vcpu);
 385         kvm_arch_vcpu_load_fp(vcpu);
 386
 387         if (single_task_running())
 388                 vcpu_clear_wfe_traps(vcpu);
 389         else
 390                 vcpu_set_wfe_traps(vcpu);
 391 }
 392
 393 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 394 {
 395         kvm_arch_vcpu_put_fp(vcpu);
 396         kvm_vcpu_put_sysregs(vcpu);
 397         kvm_timer_vcpu_put(vcpu);
 398         kvm_vgic_put(vcpu);
 399
 400         vcpu->cpu = -1;
 401
 402         kvm_arm_set_running_vcpu(NULL);
 403 }
 404
 405 static void vcpu_power_off(struct kvm_vcpu *vcpu)
 406 {
 407         vcpu->arch.power_off = true;
 408         kvm_make_request(KVM_REQ_SLEEP, vcpu);
 409         kvm_vcpu_kick(vcpu);
 410 }
 411
 412 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 413                                     struct kvm_mp_state *mp_state)
 414 {
 415         if (vcpu->arch.power_off)
 416                 mp_state->mp_state = KVM_MP_STATE_STOPPED;
 417         else
 418                 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
 419
 420         return 0;
 421 }
 422
 423 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 424                                     struct kvm_mp_state *mp_state)
 425 {
 426         int ret = 0;
 427
 428         switch (mp_state->mp_state) {
 429         case KVM_MP_STATE_RUNNABLE:
 430                 vcpu->arch.power_off = false;
 431                 break;
 432         case KVM_MP_STATE_STOPPED:
 433                 vcpu_power_off(vcpu);
 434                 break;
 435         default:
 436                 ret = -EINVAL;
 437         }
 438
 439         return ret;
 440 }
 441
 442 /**
 443  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 444  * @v:          The VCPU pointer
 445  *
 446  * If the guest CPU is not waiting for interrupts or an interrupt line is
 447  * asserted, the CPU is by definition runnable.
 448  */
 449 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 450 {
 451         bool irq_lines = *vcpu_hcr(v) & (HCR_VI | HCR_VF);
 452         return ((irq_lines || kvm_vgic_vcpu_pending_irq(v))
 453                 && !v->arch.power_off && !v->arch.pause);
 454 }
 455
 456 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
 457 {
 458         return vcpu_mode_priv(vcpu);
 459 }
 460
 461 /* Just ensure a guest exit from a particular CPU */
 462 static void exit_vm_noop(void *info)
 463 {
 464 }
 465
 466 void force_vm_exit(const cpumask_t *mask)
 467 {
 468         preempt_disable();
 469         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 470         preempt_enable();
 471 }
 472
 473 /**
 474  * need_new_vmid_gen - check that the VMID is still valid
 475  * @kvm: The VM's VMID to check
 476  *
 477  * return true if there is a new generation of VMIDs being used
 478  *
 479  * The hardware supports only 256 values with the value zero reserved for the
 480  * host, so we check if an assigned value belongs to a previous generation,
 481  * which which requires us to assign a new value. If we're the first to use a
 482  * VMID for the new generation, we must flush necessary caches and TLBs on all
 483  * CPUs.
 484  */
 485 static bool need_new_vmid_gen(struct kvm *kvm)
 486 {
 487         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 488 }
 489
 490 /**
 491  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 492  * @kvm The guest that we are about to run
 493  *
 494  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 495  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 496  * caches and TLBs.
 497  */
 498 static void update_vttbr(struct kvm *kvm)
 499 {
 500         phys_addr_t pgd_phys;
 501         u64 vmid, cnp = kvm_cpu_has_cnp() ? VTTBR_CNP_BIT : 0;
 502         bool new_gen;
 503
 504         read_lock(&kvm_vmid_lock);
 505         new_gen = need_new_vmid_gen(kvm);
 506         read_unlock(&kvm_vmid_lock);
 507
 508         if (!new_gen)
 509                 return;
 510
 511         write_lock(&kvm_vmid_lock);
 512
 513         /*
 514          * We need to re-check the vmid_gen here to ensure that if another vcpu
 515          * already allocated a valid vmid for this vm, then this vcpu should
 516          * use the same vmid.
 517          */
 518         if (!need_new_vmid_gen(kvm)) {
 519                 write_unlock(&kvm_vmid_lock);
 520                 return;
 521         }
 522
 523         /* First user of a new VMID generation? */
 524         if (unlikely(kvm_next_vmid == 0)) {
 525                 atomic64_inc(&kvm_vmid_gen);
 526                 kvm_next_vmid = 1;
 527
 528                 /*
 529                  * On SMP we know no other CPUs can use this CPU's or each
 530                  * other's VMID after force_vm_exit returns since the
 531                  * kvm_vmid_lock blocks them from reentry to the guest.
 532                  */
 533                 force_vm_exit(cpu_all_mask);
 534                 /*
 535                  * Now broadcast TLB + ICACHE invalidation over the inner
 536                  * shareable domain to make sure all data structures are
 537                  * clean.
 538                  */
 539                 kvm_call_hyp(__kvm_flush_vm_context);
 540         }
 541
 542         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 543         kvm->arch.vmid = kvm_next_vmid;
 544         kvm_next_vmid++;
 545         kvm_next_vmid &= (1 << kvm_vmid_bits) - 1;
 546
 547         /* update vttbr to be used with the new vmid */
 548         pgd_phys = virt_to_phys(kvm->arch.pgd);
 549         BUG_ON(pgd_phys & ~kvm_vttbr_baddr_mask(kvm));
 550         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK(kvm_vmid_bits);
 551         kvm->arch.vttbr = kvm_phys_to_vttbr(pgd_phys) | vmid | cnp;
 552
 553         write_unlock(&kvm_vmid_lock);
 554 }
 555
 556 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 557 {
 558         struct kvm *kvm = vcpu->kvm;
 559         int ret = 0;
 560
 561         if (likely(vcpu->arch.has_run_once))
 562                 return 0;
 563
 564         vcpu->arch.has_run_once = true;
 565
 566         if (likely(irqchip_in_kernel(kvm))) {
 567                 /*
 568                  * Map the VGIC hardware resources before running a vcpu the
 569                  * first time on this VM.
 570                  */
 571                 if (unlikely(!vgic_ready(kvm))) {
 572                         ret = kvm_vgic_map_resources(kvm);
 573                         if (ret)
 574                                 return ret;
 575                 }
 576         } else {
 577                 /*
 578                  * Tell the rest of the code that there are userspace irqchip
 579                  * VMs in the wild.
 580                  */
 581                 static_branch_inc(&userspace_irqchip_in_use);
 582         }
 583
 584         ret = kvm_timer_enable(vcpu);
 585         if (ret)
 586                 return ret;
 587
 588         ret = kvm_arm_pmu_v3_enable(vcpu);
 589
 590         return ret;
 591 }
 592
 593 bool kvm_arch_intc_initialized(struct kvm *kvm)
 594 {
 595         return vgic_initialized(kvm);
 596 }
 597
 598 void kvm_arm_halt_guest(struct kvm *kvm)
 599 {
 600         int i;
 601         struct kvm_vcpu *vcpu;
 602
 603         kvm_for_each_vcpu(i, vcpu, kvm)
 604                 vcpu->arch.pause = true;
 605         kvm_make_all_cpus_request(kvm, KVM_REQ_SLEEP);
 606 }
 607
 608 void kvm_arm_resume_guest(struct kvm *kvm)
 609 {
 610         int i;
 611         struct kvm_vcpu *vcpu;
 612
 613         kvm_for_each_vcpu(i, vcpu, kvm) {
 614                 vcpu->arch.pause = false;
 615                 swake_up_one(kvm_arch_vcpu_wq(vcpu));
 616         }
 617 }
 618
 619 static void vcpu_req_sleep(struct kvm_vcpu *vcpu)
 620 {
 621         struct swait_queue_head *wq = kvm_arch_vcpu_wq(vcpu);
 622
 623         swait_event_interruptible_exclusive(*wq, ((!vcpu->arch.power_off) &&
 624                                        (!vcpu->arch.pause)));
 625
 626         if (vcpu->arch.power_off || vcpu->arch.pause) {
 627                 /* Awaken to handle a signal, request we sleep again later. */
 628                 kvm_make_request(KVM_REQ_SLEEP, vcpu);
 629         }
 630 }
 631
 632 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 633 {
 634         return vcpu->arch.target >= 0;
 635 }
 636
 637 static void check_vcpu_requests(struct kvm_vcpu *vcpu)
 638 {
 639         if (kvm_request_pending(vcpu)) {
 640                 if (kvm_check_request(KVM_REQ_SLEEP, vcpu))
 641                         vcpu_req_sleep(vcpu);
 642
 643                 /*
 644                  * Clear IRQ_PENDING requests that were made to guarantee
 645                  * that a VCPU sees new virtual interrupts.
 646                  */
 647                 kvm_check_request(KVM_REQ_IRQ_PENDING, vcpu);
 648         }
 649 }
 650
 651 /**
 652  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 653  * @vcpu:       The VCPU pointer
 654  * @run:        The kvm_run structure pointer used for userspace state exchange
 655  *
 656  * This function is called through the VCPU_RUN ioctl called from user space. It
 657  * will execute VM code in a loop until the time slice for the process is used
 658  * or some emulation is needed from user space in which case the function will
 659  * return with return value 0 and with the kvm_run structure filled in with the
 660  * required data for the requested emulation.
 661  */
 662 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 663 {
 664         int ret;
 665
 666         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 667                 return -ENOEXEC;
 668
 669         ret = kvm_vcpu_first_run_init(vcpu);
 670         if (ret)
 671                 return ret;
 672
 673         if (run->exit_reason == KVM_EXIT_MMIO) {
 674                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 675                 if (ret)
 676                         return ret;
 677                 if (kvm_arm_handle_step_debug(vcpu, vcpu->run))
 678                         return 0;
 679         }
 680
 681         if (run->immediate_exit)
 682                 return -EINTR;
 683
 684         vcpu_load(vcpu);
 685
 686         kvm_sigset_activate(vcpu);
 687
 688         ret = 1;
 689         run->exit_reason = KVM_EXIT_UNKNOWN;
 690         while (ret > 0) {
 691                 /*
 692                  * Check conditions before entering the guest
 693                  */
 694                 cond_resched();
 695
 696                 update_vttbr(vcpu->kvm);
 697
 698                 check_vcpu_requests(vcpu);
 699
 700                 /*
 701                  * Preparing the interrupts to be injected also
 702                  * involves poking the GIC, which must be done in a
 703                  * non-preemptible context.
 704                  */
 705                 preempt_disable();
 706
 707                 kvm_pmu_flush_hwstate(vcpu);
 708
 709                 local_irq_disable();
 710
 711                 kvm_vgic_flush_hwstate(vcpu);
 712
 713                 /*
 714                  * Exit if we have a signal pending so that we can deliver the
 715                  * signal to user space.
 716                  */
 717                 if (signal_pending(current)) {
 718                         ret = -EINTR;
 719                         run->exit_reason = KVM_EXIT_INTR;
 720                 }
 721
 722                 /*
 723                  * If we're using a userspace irqchip, then check if we need
 724                  * to tell a userspace irqchip about timer or PMU level
 725                  * changes and if so, exit to userspace (the actual level
 726                  * state gets updated in kvm_timer_update_run and
 727                  * kvm_pmu_update_run below).
 728                  */
 729                 if (static_branch_unlikely(&userspace_irqchip_in_use)) {
 730                         if (kvm_timer_should_notify_user(vcpu) ||
 731                             kvm_pmu_should_notify_user(vcpu)) {
 732                                 ret = -EINTR;
 733                                 run->exit_reason = KVM_EXIT_INTR;
 734                         }
 735                 }
 736
 737                 /*
 738                  * Ensure we set mode to IN_GUEST_MODE after we disable
 739                  * interrupts and before the final VCPU requests check.
 740                  * See the comment in kvm_vcpu_exiting_guest_mode() and
 741                  * Documentation/virtual/kvm/vcpu-requests.rst
 742                  */
 743                 smp_store_mb(vcpu->mode, IN_GUEST_MODE);
 744
 745                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm) ||
 746                     kvm_request_pending(vcpu)) {
 747                         vcpu->mode = OUTSIDE_GUEST_MODE;
 748                         isb(); /* Ensure work in x_flush_hwstate is committed */
 749                         kvm_pmu_sync_hwstate(vcpu);
 750                         if (static_branch_unlikely(&userspace_irqchip_in_use))
 751                                 kvm_timer_sync_hwstate(vcpu);
 752                         kvm_vgic_sync_hwstate(vcpu);
 753                         local_irq_enable();
 754                         preempt_enable();
 755                         continue;
 756                 }
 757
 758                 kvm_arm_setup_debug(vcpu);
 759
 760                 /**************************************************************
 761                  * Enter the guest
 762                  */
 763                 trace_kvm_entry(*vcpu_pc(vcpu));
 764                 guest_enter_irqoff();
 765
 766                 if (has_vhe()) {
 767                         kvm_arm_vhe_guest_enter();
 768                         ret = kvm_vcpu_run_vhe(vcpu);
 769                         kvm_arm_vhe_guest_exit();
 770                 } else {
 771                         ret = kvm_call_hyp(__kvm_vcpu_run_nvhe, vcpu);
 772                 }
 773
 774                 vcpu->mode = OUTSIDE_GUEST_MODE;
 775                 vcpu->stat.exits++;
 776                 /*
 777                  * Back from guest
 778                  *************************************************************/
 779
 780                 kvm_arm_clear_debug(vcpu);
 781
 782                 /*
 783                  * We must sync the PMU state before the vgic state so
 784                  * that the vgic can properly sample the updated state of the
 785                  * interrupt line.
 786                  */
 787                 kvm_pmu_sync_hwstate(vcpu);
 788
 789                 /*
 790                  * Sync the vgic state before syncing the timer state because
 791                  * the timer code needs to know if the virtual timer
 792                  * interrupts are active.
 793                  */
 794                 kvm_vgic_sync_hwstate(vcpu);
 795
 796                 /*
 797                  * Sync the timer hardware state before enabling interrupts as
 798                  * we don't want vtimer interrupts to race with syncing the
 799                  * timer virtual interrupt state.
 800                  */
 801                 if (static_branch_unlikely(&userspace_irqchip_in_use))
 802                         kvm_timer_sync_hwstate(vcpu);
 803
 804                 kvm_arch_vcpu_ctxsync_fp(vcpu);
 805
 806                 /*
 807                  * We may have taken a host interrupt in HYP mode (ie
 808                  * while executing the guest). This interrupt is still
 809                  * pending, as we haven't serviced it yet!
 810                  *
 811                  * We're now back in SVC mode, with interrupts
 812                  * disabled.  Enabling the interrupts now will have
 813                  * the effect of taking the interrupt again, in SVC
 814                  * mode this time.
 815                  */
 816                 local_irq_enable();
 817
 818                 /*
 819                  * We do local_irq_enable() before calling guest_exit() so
 820                  * that if a timer interrupt hits while running the guest we
 821                  * account that tick as being spent in the guest.  We enable
 822                  * preemption after calling guest_exit() so that if we get
 823                  * preempted we make sure ticks after that is not counted as
 824                  * guest time.
 825                  */
 826                 guest_exit();
 827                 trace_kvm_exit(ret, kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 828
 829                 /* Exit types that need handling before we can be preempted */
 830                 handle_exit_early(vcpu, run, ret);
 831
 832                 preempt_enable();
 833
 834                 ret = handle_exit(vcpu, run, ret);
 835         }
 836
 837         /* Tell userspace about in-kernel device output levels */
 838         if (unlikely(!irqchip_in_kernel(vcpu->kvm))) {
 839                 kvm_timer_update_run(vcpu);
 840                 kvm_pmu_update_run(vcpu);
 841         }
 842
 843         kvm_sigset_deactivate(vcpu);
 844
 845         vcpu_put(vcpu);
 846         return ret;
 847 }
 848
 849 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 850 {
 851         int bit_index;
 852         bool set;
 853         unsigned long *hcr;
 854
 855         if (number == KVM_ARM_IRQ_CPU_IRQ)
 856                 bit_index = __ffs(HCR_VI);
 857         else /* KVM_ARM_IRQ_CPU_FIQ */
 858                 bit_index = __ffs(HCR_VF);
 859
 860         hcr = vcpu_hcr(vcpu);
 861         if (level)
 862                 set = test_and_set_bit(bit_index, hcr);
 863         else
 864                 set = test_and_clear_bit(bit_index, hcr);
 865
 866         /*
 867          * If we didn't change anything, no need to wake up or kick other CPUs
 868          */
 869         if (set == level)
 870                 return 0;
 871
 872         /*
 873          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 874          * trigger a world-switch round on the running physical CPU to set the
 875          * virtual IRQ/FIQ fields in the HCR appropriately.
 876          */
 877         kvm_make_request(KVM_REQ_IRQ_PENDING, vcpu);
 878         kvm_vcpu_kick(vcpu);
 879
 880         return 0;
 881 }
 882
 883 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 884                           bool line_status)
 885 {
 886         u32 irq = irq_level->irq;
 887         unsigned int irq_type, vcpu_idx, irq_num;
 888         int nrcpus = atomic_read(&kvm->online_vcpus);
 889         struct kvm_vcpu *vcpu = NULL;
 890         bool level = irq_level->level;
 891
 892         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 893         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 894         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 895
 896         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 897
 898         switch (irq_type) {
 899         case KVM_ARM_IRQ_TYPE_CPU:
 900                 if (irqchip_in_kernel(kvm))
 901                         return -ENXIO;
 902
 903                 if (vcpu_idx >= nrcpus)
 904                         return -EINVAL;
 905
 906                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 907                 if (!vcpu)
 908                         return -EINVAL;
 909
 910                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 911                         return -EINVAL;
 912
 913                 return vcpu_interrupt_line(vcpu, irq_num, level);
 914         case KVM_ARM_IRQ_TYPE_PPI:
 915                 if (!irqchip_in_kernel(kvm))
 916                         return -ENXIO;
 917
 918                 if (vcpu_idx >= nrcpus)
 919                         return -EINVAL;
 920
 921                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 922                 if (!vcpu)
 923                         return -EINVAL;
 924
 925                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 926                         return -EINVAL;
 927
 928                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level, NULL);
 929         case KVM_ARM_IRQ_TYPE_SPI:
 930                 if (!irqchip_in_kernel(kvm))
 931                         return -ENXIO;
 932
 933                 if (irq_num < VGIC_NR_PRIVATE_IRQS)
 934                         return -EINVAL;
 935
 936                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level, NULL);
 937         }
 938
 939         return -EINVAL;
 940 }
 941
 942 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 943                                const struct kvm_vcpu_init *init)
 944 {
 945         unsigned int i;
 946         int phys_target = kvm_target_cpu();
 947
 948         if (init->target != phys_target)
 949                 return -EINVAL;
 950
 951         /*
 952          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 953          * use the same target.
 954          */
 955         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 956                 return -EINVAL;
 957
 958         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 959         for (i = 0; i < sizeof(init->features) * 8; i++) {
 960                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 961
 962                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 963                         return -ENOENT;
 964
 965                 /*
 966                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 967                  * use the same feature set.
 968                  */
 969                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 970                     test_bit(i, vcpu->arch.features) != set)
 971                         return -EINVAL;
 972
 973                 if (set)
 974                         set_bit(i, vcpu->arch.features);
 975         }
 976
 977         vcpu->arch.target = phys_target;
 978
 979         /* Now we know what it is, we can reset it. */
 980         return kvm_reset_vcpu(vcpu);
 981 }
 982
 983
 984 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 985                                          struct kvm_vcpu_init *init)
 986 {
 987         int ret;
 988
 989         ret = kvm_vcpu_set_target(vcpu, init);
 990         if (ret)
 991                 return ret;
 992
 993         /*
 994          * Ensure a rebooted VM will fault in RAM pages and detect if the
 995          * guest MMU is turned off and flush the caches as needed.
 996          */
 997         if (vcpu->arch.has_run_once)
 998                 stage2_unmap_vm(vcpu->kvm);
 999
1000         vcpu_reset_hcr(vcpu);
1001
1002         /*
1003          * Handle the "start in power-off" case.
1004          */
1005         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
1006                 vcpu_power_off(vcpu);
1007         else
1008                 vcpu->arch.power_off = false;
1009
1010         return 0;
1011 }
1012
1013 static int kvm_arm_vcpu_set_attr(struct kvm_vcpu *vcpu,
1014                                  struct kvm_device_attr *attr)
1015 {
1016         int ret = -ENXIO;
1017
1018         switch (attr->group) {
1019         default:
1020                 ret = kvm_arm_vcpu_arch_set_attr(vcpu, attr);
1021                 break;
1022         }
1023
1024         return ret;
1025 }
1026
1027 static int kvm_arm_vcpu_get_attr(struct kvm_vcpu *vcpu,
1028                                  struct kvm_device_attr *attr)
1029 {
1030         int ret = -ENXIO;
1031
1032         switch (attr->group) {
1033         default:
1034                 ret = kvm_arm_vcpu_arch_get_attr(vcpu, attr);
1035                 break;
1036         }
1037
1038         return ret;
1039 }
1040
1041 static int kvm_arm_vcpu_has_attr(struct kvm_vcpu *vcpu,
1042                                  struct kvm_device_attr *attr)
1043 {
1044         int ret = -ENXIO;
1045
1046         switch (attr->group) {
1047         default:
1048                 ret = kvm_arm_vcpu_arch_has_attr(vcpu, attr);
1049                 break;
1050         }
1051
1052         return ret;
1053 }
1054
1055 static int kvm_arm_vcpu_get_events(struct kvm_vcpu *vcpu,
1056                                    struct kvm_vcpu_events *events)
1057 {
1058         memset(events, 0, sizeof(*events));
1059
1060         return __kvm_arm_vcpu_get_events(vcpu, events);
1061 }
1062
1063 static int kvm_arm_vcpu_set_events(struct kvm_vcpu *vcpu,
1064                                    struct kvm_vcpu_events *events)
1065 {
1066         int i;
1067
1068         /* check whether the reserved field is zero */
1069         for (i = 0; i < ARRAY_SIZE(events->reserved); i++)
1070                 if (events->reserved[i])
1071                         return -EINVAL;
1072
1073         /* check whether the pad field is zero */
1074         for (i = 0; i < ARRAY_SIZE(events->exception.pad); i++)
1075                 if (events->exception.pad[i])
1076                         return -EINVAL;
1077
1078         return __kvm_arm_vcpu_set_events(vcpu, events);
1079 }
1080
1081 long kvm_arch_vcpu_ioctl(struct file *filp,
1082                          unsigned int ioctl, unsigned long arg)
1083 {
1084         struct kvm_vcpu *vcpu = filp->private_data;
1085         void __user *argp = (void __user *)arg;
1086         struct kvm_device_attr attr;
1087         long r;
1088
1089         switch (ioctl) {
1090         case KVM_ARM_VCPU_INIT: {
1091                 struct kvm_vcpu_init init;
1092
1093                 r = -EFAULT;
1094                 if (copy_from_user(&init, argp, sizeof(init)))
1095                         break;
1096
1097                 r = kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
1098                 break;
1099         }
1100         case KVM_SET_ONE_REG:
1101         case KVM_GET_ONE_REG: {
1102                 struct kvm_one_reg reg;
1103
1104                 r = -ENOEXEC;
1105                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1106                         break;
1107
1108                 r = -EFAULT;
1109                 if (copy_from_user(&reg, argp, sizeof(reg)))
1110                         break;
1111
1112                 if (ioctl == KVM_SET_ONE_REG)
1113                         r = kvm_arm_set_reg(vcpu, &reg);
1114                 else
1115                         r = kvm_arm_get_reg(vcpu, &reg);
1116                 break;
1117         }
1118         case KVM_GET_REG_LIST: {
1119                 struct kvm_reg_list __user *user_list = argp;
1120                 struct kvm_reg_list reg_list;
1121                 unsigned n;
1122
1123                 r = -ENOEXEC;
1124                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
1125                         break;
1126
1127                 r = -EFAULT;
1128                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
1129                         break;
1130                 n = reg_list.n;
1131                 reg_list.n = kvm_arm_num_regs(vcpu);
1132                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
1133                         break;
1134                 r = -E2BIG;
1135                 if (n < reg_list.n)
1136                         break;
1137                 r = kvm_arm_copy_reg_indices(vcpu, user_list->reg);
1138                 break;
1139         }
1140         case KVM_SET_DEVICE_ATTR: {
1141                 r = -EFAULT;
1142                 if (copy_from_user(&attr, argp, sizeof(attr)))
1143                         break;
1144                 r = kvm_arm_vcpu_set_attr(vcpu, &attr);
1145                 break;
1146         }
1147         case KVM_GET_DEVICE_ATTR: {
1148                 r = -EFAULT;
1149                 if (copy_from_user(&attr, argp, sizeof(attr)))
1150                         break;
1151                 r = kvm_arm_vcpu_get_attr(vcpu, &attr);
1152                 break;
1153         }
1154         case KVM_HAS_DEVICE_ATTR: {
1155                 r = -EFAULT;
1156                 if (copy_from_user(&attr, argp, sizeof(attr)))
1157                         break;
1158                 r = kvm_arm_vcpu_has_attr(vcpu, &attr);
1159                 break;
1160         }
1161         case KVM_GET_VCPU_EVENTS: {
1162                 struct kvm_vcpu_events events;
1163
1164                 if (kvm_arm_vcpu_get_events(vcpu, &events))
1165                         return -EINVAL;
1166
1167                 if (copy_to_user(argp, &events, sizeof(events)))
1168                         return -EFAULT;
1169
1170                 return 0;
1171         }
1172         case KVM_SET_VCPU_EVENTS: {
1173                 struct kvm_vcpu_events events;
1174
1175                 if (copy_from_user(&events, argp, sizeof(events)))
1176                         return -EFAULT;
1177
1178                 return kvm_arm_vcpu_set_events(vcpu, &events);
1179         }
1180         default:
1181                 r = -EINVAL;
1182         }
1183
1184         return r;
1185 }
1186
1187 /**
1188  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
1189  * @kvm: kvm instance
1190  * @log: slot id and address to which we copy the log
1191  *
1192  * Steps 1-4 below provide general overview of dirty page logging. See
1193  * kvm_get_dirty_log_protect() function description for additional details.
1194  *
1195  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
1196  * always flush the TLB (step 4) even if previous step failed  and the dirty
1197  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
1198  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
1199  * writes will be marked dirty for next log read.
1200  *
1201  *   1. Take a snapshot of the bit and clear it if needed.
1202  *   2. Write protect the corresponding page.
1203  *   3. Copy the snapshot to the userspace.
1204  *   4. Flush TLB's if needed.
1205  */
1206 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
1207 {
1208         bool is_dirty = false;
1209         int r;
1210
1211         mutex_lock(&kvm->slots_lock);
1212
1213         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
1214
1215         if (is_dirty)
1216                 kvm_flush_remote_tlbs(kvm);
1217
1218         mutex_unlock(&kvm->slots_lock);
1219         return r;
1220 }
1221
1222 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
1223                                         struct kvm_arm_device_addr *dev_addr)
1224 {
1225         unsigned long dev_id, type;
1226
1227         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
1228                 KVM_ARM_DEVICE_ID_SHIFT;
1229         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
1230                 KVM_ARM_DEVICE_TYPE_SHIFT;
1231
1232         switch (dev_id) {
1233         case KVM_ARM_DEVICE_VGIC_V2:
1234                 if (!vgic_present)
1235                         return -ENXIO;
1236                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
1237         default:
1238                 return -ENODEV;
1239         }
1240 }
1241
1242 long kvm_arch_vm_ioctl(struct file *filp,
1243                        unsigned int ioctl, unsigned long arg)
1244 {
1245         struct kvm *kvm = filp->private_data;
1246         void __user *argp = (void __user *)arg;
1247
1248         switch (ioctl) {
1249         case KVM_CREATE_IRQCHIP: {
1250                 int ret;
1251                 if (!vgic_present)
1252                         return -ENXIO;
1253                 mutex_lock(&kvm->lock);
1254                 ret = kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
1255                 mutex_unlock(&kvm->lock);
1256                 return ret;
1257         }
1258         case KVM_ARM_SET_DEVICE_ADDR: {
1259                 struct kvm_arm_device_addr dev_addr;
1260
1261                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
1262                         return -EFAULT;
1263                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
1264         }
1265         case KVM_ARM_PREFERRED_TARGET: {
1266                 int err;
1267                 struct kvm_vcpu_init init;
1268
1269                 err = kvm_vcpu_preferred_target(&init);
1270                 if (err)
1271                         return err;
1272
1273                 if (copy_to_user(argp, &init, sizeof(init)))
1274                         return -EFAULT;
1275
1276                 return 0;
1277         }
1278         default:
1279                 return -EINVAL;
1280         }
1281 }
1282
1283 static void cpu_init_hyp_mode(void *dummy)
1284 {
1285         phys_addr_t pgd_ptr;
1286         unsigned long hyp_stack_ptr;
1287         unsigned long stack_page;
1288         unsigned long vector_ptr;
1289
1290         /* Switch from the HYP stub to our own HYP init vector */
1291         __hyp_set_vectors(kvm_get_idmap_vector());
1292
1293         pgd_ptr = kvm_mmu_get_httbr();
1294         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
1295         hyp_stack_ptr = stack_page + PAGE_SIZE;
1296         vector_ptr = (unsigned long)kvm_get_hyp_vector();
1297
1298         __cpu_init_hyp_mode(pgd_ptr, hyp_stack_ptr, vector_ptr);
1299         __cpu_init_stage2();
1300 }
1301
1302 static void cpu_hyp_reset(void)
1303 {
1304         if (!is_kernel_in_hyp_mode())
1305                 __hyp_reset_vectors();
1306 }
1307
1308 static void cpu_hyp_reinit(void)
1309 {
1310         cpu_hyp_reset();
1311
1312         if (is_kernel_in_hyp_mode())
1313                 kvm_timer_init_vhe();
1314         else
1315                 cpu_init_hyp_mode(NULL);
1316
1317         kvm_arm_init_debug();
1318
1319         if (vgic_present)
1320                 kvm_vgic_init_cpu_hardware();
1321 }
1322
1323 static void _kvm_arch_hardware_enable(void *discard)
1324 {
1325         if (!__this_cpu_read(kvm_arm_hardware_enabled)) {
1326                 cpu_hyp_reinit();
1327                 __this_cpu_write(kvm_arm_hardware_enabled, 1);
1328         }
1329 }
1330
1331 int kvm_arch_hardware_enable(void)
1332 {
1333         _kvm_arch_hardware_enable(NULL);
1334         return 0;
1335 }
1336
1337 static void _kvm_arch_hardware_disable(void *discard)
1338 {
1339         if (__this_cpu_read(kvm_arm_hardware_enabled)) {
1340                 cpu_hyp_reset();
1341                 __this_cpu_write(kvm_arm_hardware_enabled, 0);
1342         }
1343 }
1344
1345 void kvm_arch_hardware_disable(void)
1346 {
1347         _kvm_arch_hardware_disable(NULL);
1348 }
1349
1350 #ifdef CONFIG_CPU_PM
1351 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
1352                                     unsigned long cmd,
1353                                     void *v)
1354 {
1355         /*
1356          * kvm_arm_hardware_enabled is left with its old value over
1357          * PM_ENTER->PM_EXIT. It is used to indicate PM_EXIT should
1358          * re-enable hyp.
1359          */
1360         switch (cmd) {
1361         case CPU_PM_ENTER:
1362                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1363                         /*
1364                          * don't update kvm_arm_hardware_enabled here
1365                          * so that the hardware will be re-enabled
1366                          * when we resume. See below.
1367                          */
1368                         cpu_hyp_reset();
1369
1370                 return NOTIFY_OK;
1371         case CPU_PM_ENTER_FAILED:
1372         case CPU_PM_EXIT:
1373                 if (__this_cpu_read(kvm_arm_hardware_enabled))
1374                         /* The hardware was enabled before suspend. */
1375                         cpu_hyp_reinit();
1376
1377                 return NOTIFY_OK;
1378
1379         default:
1380                 return NOTIFY_DONE;
1381         }
1382 }
1383
1384 static struct notifier_block hyp_init_cpu_pm_nb = {
1385         .notifier_call = hyp_init_cpu_pm_notifier,
1386 };
1387
1388 static void __init hyp_cpu_pm_init(void)
1389 {
1390         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
1391 }
1392 static void __init hyp_cpu_pm_exit(void)
1393 {
1394         cpu_pm_unregister_notifier(&hyp_init_cpu_pm_nb);
1395 }
1396 #else
1397 static inline void hyp_cpu_pm_init(void)
1398 {
1399 }
1400 static inline void hyp_cpu_pm_exit(void)
1401 {
1402 }
1403 #endif
1404
1405 static int init_common_resources(void)
1406 {
1407         /* set size of VMID supported by CPU */
1408         kvm_vmid_bits = kvm_get_vmid_bits();
1409         kvm_info("%d-bit VMID\n", kvm_vmid_bits);
1410
1411         kvm_set_ipa_limit();
1412
1413         return 0;
1414 }
1415
1416 static int init_subsystems(void)
1417 {
1418         int err = 0;
1419
1420         /*
1421          * Enable hardware so that subsystem initialisation can access EL2.
1422          */
1423         on_each_cpu(_kvm_arch_hardware_enable, NULL, 1);
1424
1425         /*
1426          * Register CPU lower-power notifier
1427          */
1428         hyp_cpu_pm_init();
1429
1430         /*
1431          * Init HYP view of VGIC
1432          */
1433         err = kvm_vgic_hyp_init();
1434         switch (err) {
1435         case 0:
1436                 vgic_present = true;
1437                 break;
1438         case -ENODEV:
1439         case -ENXIO:
1440                 vgic_present = false;
1441                 err = 0;
1442                 break;
1443         default:
1444                 goto out;
1445         }
1446
1447         /*
1448          * Init HYP architected timer support
1449          */
1450         err = kvm_timer_hyp_init(vgic_present);
1451         if (err)
1452                 goto out;
1453
1454         kvm_perf_init();
1455         kvm_coproc_table_init();
1456
1457 out:
1458         on_each_cpu(_kvm_arch_hardware_disable, NULL, 1);
1459
1460         return err;
1461 }
1462
1463 static void teardown_hyp_mode(void)
1464 {
1465         int cpu;
1466
1467         free_hyp_pgds();
1468         for_each_possible_cpu(cpu)
1469                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1470         hyp_cpu_pm_exit();
1471 }
1472
1473 /**
1474  * Inits Hyp-mode on all online CPUs
1475  */
1476 static int init_hyp_mode(void)
1477 {
1478         int cpu;
1479         int err = 0;
1480
1481         /*
1482          * Allocate Hyp PGD and setup Hyp identity mapping
1483          */
1484         err = kvm_mmu_init();
1485         if (err)
1486                 goto out_err;
1487
1488         /*
1489          * Allocate stack pages for Hypervisor-mode
1490          */
1491         for_each_possible_cpu(cpu) {
1492                 unsigned long stack_page;
1493
1494                 stack_page = __get_free_page(GFP_KERNEL);
1495                 if (!stack_page) {
1496                         err = -ENOMEM;
1497                         goto out_err;
1498                 }
1499
1500                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
1501         }
1502
1503         /*
1504          * Map the Hyp-code called directly from the host
1505          */
1506         err = create_hyp_mappings(kvm_ksym_ref(__hyp_text_start),
1507                                   kvm_ksym_ref(__hyp_text_end), PAGE_HYP_EXEC);
1508         if (err) {
1509                 kvm_err("Cannot map world-switch code\n");
1510                 goto out_err;
1511         }
1512
1513         err = create_hyp_mappings(kvm_ksym_ref(__start_rodata),
1514                                   kvm_ksym_ref(__end_rodata), PAGE_HYP_RO);
1515         if (err) {
1516                 kvm_err("Cannot map rodata section\n");
1517                 goto out_err;
1518         }
1519
1520         err = create_hyp_mappings(kvm_ksym_ref(__bss_start),
1521                                   kvm_ksym_ref(__bss_stop), PAGE_HYP_RO);
1522         if (err) {
1523                 kvm_err("Cannot map bss section\n");
1524                 goto out_err;
1525         }
1526
1527         err = kvm_map_vectors();
1528         if (err) {
1529                 kvm_err("Cannot map vectors\n");
1530                 goto out_err;
1531         }
1532
1533         /*
1534          * Map the Hyp stack pages
1535          */
1536         for_each_possible_cpu(cpu) {
1537                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
1538                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE,
1539                                           PAGE_HYP);
1540
1541                 if (err) {
1542                         kvm_err("Cannot map hyp stack\n");
1543                         goto out_err;
1544                 }
1545         }
1546
1547         for_each_possible_cpu(cpu) {
1548                 kvm_cpu_context_t *cpu_ctxt;
1549
1550                 cpu_ctxt = per_cpu_ptr(&kvm_host_cpu_state, cpu);
1551                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1, PAGE_HYP);
1552
1553                 if (err) {
1554                         kvm_err("Cannot map host CPU state: %d\n", err);
1555                         goto out_err;
1556                 }
1557         }
1558
1559         err = hyp_map_aux_data();
1560         if (err)
1561                 kvm_err("Cannot map host auxilary data: %d\n", err);
1562
1563         return 0;
1564
1565 out_err:
1566         teardown_hyp_mode();
1567         kvm_err("error initializing Hyp mode: %d\n", err);
1568         return err;
1569 }
1570
1571 static void check_kvm_target_cpu(void *ret)
1572 {
1573         *(int *)ret = kvm_target_cpu();
1574 }
1575
1576 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1577 {
1578         struct kvm_vcpu *vcpu;
1579         int i;
1580
1581         mpidr &= MPIDR_HWID_BITMASK;
1582         kvm_for_each_vcpu(i, vcpu, kvm) {
1583                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1584                         return vcpu;
1585         }
1586         return NULL;
1587 }
1588
1589 bool kvm_arch_has_irq_bypass(void)
1590 {
1591         return true;
1592 }
1593
1594 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
1595                                       struct irq_bypass_producer *prod)
1596 {
1597         struct kvm_kernel_irqfd *irqfd =
1598                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1599
1600         return kvm_vgic_v4_set_forwarding(irqfd->kvm, prod->irq,
1601                                           &irqfd->irq_entry);
1602 }
1603 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
1604                                       struct irq_bypass_producer *prod)
1605 {
1606         struct kvm_kernel_irqfd *irqfd =
1607                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1608
1609         kvm_vgic_v4_unset_forwarding(irqfd->kvm, prod->irq,
1610                                      &irqfd->irq_entry);
1611 }
1612
1613 void kvm_arch_irq_bypass_stop(struct irq_bypass_consumer *cons)
1614 {
1615         struct kvm_kernel_irqfd *irqfd =
1616                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1617
1618         kvm_arm_halt_guest(irqfd->kvm);
1619 }
1620
1621 void kvm_arch_irq_bypass_start(struct irq_bypass_consumer *cons)
1622 {
1623         struct kvm_kernel_irqfd *irqfd =
1624                 container_of(cons, struct kvm_kernel_irqfd, consumer);
1625
1626         kvm_arm_resume_guest(irqfd->kvm);
1627 }
1628
1629 /**
1630  * Initialize Hyp-mode and memory mappings on all CPUs.
1631  */
1632 int kvm_arch_init(void *opaque)
1633 {
1634         int err;
1635         int ret, cpu;
1636         bool in_hyp_mode;
1637
1638         if (!is_hyp_mode_available()) {
1639                 kvm_info("HYP mode not available\n");
1640                 return -ENODEV;
1641         }
1642
1643         if (!kvm_arch_check_sve_has_vhe()) {
1644                 kvm_pr_unimpl("SVE system without VHE unsupported.  Broken cpu?");
1645                 return -ENODEV;
1646         }
1647
1648         for_each_online_cpu(cpu) {
1649                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1650                 if (ret < 0) {
1651                         kvm_err("Error, CPU %d not supported!\n", cpu);
1652                         return -ENODEV;
1653                 }
1654         }
1655
1656         err = init_common_resources();
1657         if (err)
1658                 return err;
1659
1660         in_hyp_mode = is_kernel_in_hyp_mode();
1661
1662         if (!in_hyp_mode) {
1663                 err = init_hyp_mode();
1664                 if (err)
1665                         goto out_err;
1666         }
1667
1668         err = init_subsystems();
1669         if (err)
1670                 goto out_hyp;
1671
1672         if (in_hyp_mode)
1673                 kvm_info("VHE mode initialized successfully\n");
1674         else
1675                 kvm_info("Hyp mode initialized successfully\n");
1676
1677         return 0;
1678
1679 out_hyp:
1680         if (!in_hyp_mode)
1681                 teardown_hyp_mode();
1682 out_err:
1683         return err;
1684 }
1685
1686 /* NOP: Compiling as a module not supported */
1687 void kvm_arch_exit(void)
1688 {
1689         kvm_perf_teardown();
1690 }
1691
1692 static int arm_init(void)
1693 {
1694         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1695         return rc;
1696 }
1697
1698 module_init(arm_init);