arch/arm/kvm/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.h>
  20 #include <linux/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/module.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/fs.h>
  26 #include <linux/mman.h>
  27 #include <linux/sched.h>
  28 #include <linux/kvm.h>
  29 #include <trace/events/kvm.h>
  30
  31 #define CREATE_TRACE_POINTS
  32 #include "trace.h"
  33
  34 #include <asm/uaccess.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/mman.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/cacheflush.h>
  39 #include <asm/virt.h>
  40 #include <asm/kvm_arm.h>
  41 #include <asm/kvm_asm.h>
  42 #include <asm/kvm_mmu.h>
  43 #include <asm/kvm_emulate.h>
  44 #include <asm/kvm_coproc.h>
  45 #include <asm/kvm_psci.h>
  46
  47 #ifdef REQUIRES_VIRT
  48 __asm__(".arch_extension        virt");
  49 #endif
  50
  51 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  52 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  53 static unsigned long hyp_default_vectors;
  54
  55 /* Per-CPU variable containing the currently running vcpu. */
  56 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  57
  58 /* The VMID used in the VTTBR */
  59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  60 static u8 kvm_next_vmid;
  61 static DEFINE_SPINLOCK(kvm_vmid_lock);
  62
  63 static bool vgic_present;
  64
  65 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  66 {
  67         BUG_ON(preemptible());
  68         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  69 }
  70
  71 /**
  72  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  73  * Must be called from non-preemptible context
  74  */
  75 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  76 {
  77         BUG_ON(preemptible());
  78         return __this_cpu_read(kvm_arm_running_vcpu);
  79 }
  80
  81 /**
  82  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  83  */
  84 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  85 {
  86         return &kvm_arm_running_vcpu;
  87 }
  88
  89 int kvm_arch_hardware_enable(void *garbage)
  90 {
  91         return 0;
  92 }
  93
  94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  95 {
  96         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  97 }
  98
  99 void kvm_arch_hardware_disable(void *garbage)
 100 {
 101 }
 102
 103 int kvm_arch_hardware_setup(void)
 104 {
 105         return 0;
 106 }
 107
 108 void kvm_arch_hardware_unsetup(void)
 109 {
 110 }
 111
 112 void kvm_arch_check_processor_compat(void *rtn)
 113 {
 114         *(int *)rtn = 0;
 115 }
 116
 117 void kvm_arch_sync_events(struct kvm *kvm)
 118 {
 119 }
 120
 121 /**
 122  * kvm_arch_init_vm - initializes a VM data structure
 123  * @kvm:        pointer to the KVM struct
 124  */
 125 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 126 {
 127         int ret = 0;
 128
 129         if (type)
 130                 return -EINVAL;
 131
 132         ret = kvm_alloc_stage2_pgd(kvm);
 133         if (ret)
 134                 goto out_fail_alloc;
 135
 136         ret = create_hyp_mappings(kvm, kvm + 1);
 137         if (ret)
 138                 goto out_free_stage2_pgd;
 139
 140         /* Mark the initial VMID generation invalid */
 141         kvm->arch.vmid_gen = 0;
 142
 143         return ret;
 144 out_free_stage2_pgd:
 145         kvm_free_stage2_pgd(kvm);
 146 out_fail_alloc:
 147         return ret;
 148 }
 149
 150 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 151 {
 152         return VM_FAULT_SIGBUS;
 153 }
 154
 155 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 156                            struct kvm_memory_slot *dont)
 157 {
 158 }
 159
 160 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 161                             unsigned long npages)
 162 {
 163         return 0;
 164 }
 165
 166 /**
 167  * kvm_arch_destroy_vm - destroy the VM data structure
 168  * @kvm:        pointer to the KVM struct
 169  */
 170 void kvm_arch_destroy_vm(struct kvm *kvm)
 171 {
 172         int i;
 173
 174         kvm_free_stage2_pgd(kvm);
 175
 176         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 177                 if (kvm->vcpus[i]) {
 178                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 179                         kvm->vcpus[i] = NULL;
 180                 }
 181         }
 182 }
 183
 184 int kvm_dev_ioctl_check_extension(long ext)
 185 {
 186         int r;
 187         switch (ext) {
 188         case KVM_CAP_IRQCHIP:
 189                 r = vgic_present;
 190                 break;
 191         case KVM_CAP_USER_MEMORY:
 192         case KVM_CAP_SYNC_MMU:
 193         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 194         case KVM_CAP_ONE_REG:
 195         case KVM_CAP_ARM_PSCI:
 196                 r = 1;
 197                 break;
 198         case KVM_CAP_COALESCED_MMIO:
 199                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 200                 break;
 201         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 202                 r = 1;
 203                 break;
 204         case KVM_CAP_NR_VCPUS:
 205                 r = num_online_cpus();
 206                 break;
 207         case KVM_CAP_MAX_VCPUS:
 208                 r = KVM_MAX_VCPUS;
 209                 break;
 210         default:
 211                 r = kvm_arch_dev_ioctl_check_extension(ext);
 212                 break;
 213         }
 214         return r;
 215 }
 216
 217 long kvm_arch_dev_ioctl(struct file *filp,
 218                         unsigned int ioctl, unsigned long arg)
 219 {
 220         return -EINVAL;
 221 }
 222
 223 void kvm_arch_memslots_updated(struct kvm *kvm)
 224 {
 225 }
 226
 227 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 228                                    struct kvm_memory_slot *memslot,
 229                                    struct kvm_userspace_memory_region *mem,
 230                                    enum kvm_mr_change change)
 231 {
 232         return 0;
 233 }
 234
 235 void kvm_arch_commit_memory_region(struct kvm *kvm,
 236                                    struct kvm_userspace_memory_region *mem,
 237                                    const struct kvm_memory_slot *old,
 238                                    enum kvm_mr_change change)
 239 {
 240 }
 241
 242 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 243 {
 244 }
 245
 246 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 247                                    struct kvm_memory_slot *slot)
 248 {
 249 }
 250
 251 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 252 {
 253         int err;
 254         struct kvm_vcpu *vcpu;
 255
 256         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 257         if (!vcpu) {
 258                 err = -ENOMEM;
 259                 goto out;
 260         }
 261
 262         err = kvm_vcpu_init(vcpu, kvm, id);
 263         if (err)
 264                 goto free_vcpu;
 265
 266         err = create_hyp_mappings(vcpu, vcpu + 1);
 267         if (err)
 268                 goto vcpu_uninit;
 269
 270         return vcpu;
 271 vcpu_uninit:
 272         kvm_vcpu_uninit(vcpu);
 273 free_vcpu:
 274         kmem_cache_free(kvm_vcpu_cache, vcpu);
 275 out:
 276         return ERR_PTR(err);
 277 }
 278
 279 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 280 {
 281         return 0;
 282 }
 283
 284 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 285 {
 286         kvm_mmu_free_memory_caches(vcpu);
 287         kvm_timer_vcpu_terminate(vcpu);
 288         kmem_cache_free(kvm_vcpu_cache, vcpu);
 289 }
 290
 291 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 292 {
 293         kvm_arch_vcpu_free(vcpu);
 294 }
 295
 296 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 297 {
 298         return 0;
 299 }
 300
 301 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 302 {
 303         int ret;
 304
 305         /* Force users to call KVM_ARM_VCPU_INIT */
 306         vcpu->arch.target = -1;
 307
 308         /* Set up VGIC */
 309         ret = kvm_vgic_vcpu_init(vcpu);
 310         if (ret)
 311                 return ret;
 312
 313         /* Set up the timer */
 314         kvm_timer_vcpu_init(vcpu);
 315
 316         return 0;
 317 }
 318
 319 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 320 {
 321 }
 322
 323 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 324 {
 325         vcpu->cpu = cpu;
 326         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 327
 328         /*
 329          * Check whether this vcpu requires the cache to be flushed on
 330          * this physical CPU. This is a consequence of doing dcache
 331          * operations by set/way on this vcpu. We do it here to be in
 332          * a non-preemptible section.
 333          */
 334         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 335                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 336
 337         kvm_arm_set_running_vcpu(vcpu);
 338 }
 339
 340 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 341 {
 342         kvm_arm_set_running_vcpu(NULL);
 343 }
 344
 345 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 346                                         struct kvm_guest_debug *dbg)
 347 {
 348         return -EINVAL;
 349 }
 350
 351
 352 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 353                                     struct kvm_mp_state *mp_state)
 354 {
 355         return -EINVAL;
 356 }
 357
 358 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 359                                     struct kvm_mp_state *mp_state)
 360 {
 361         return -EINVAL;
 362 }
 363
 364 /**
 365  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 366  * @v:          The VCPU pointer
 367  *
 368  * If the guest CPU is not waiting for interrupts or an interrupt line is
 369  * asserted, the CPU is by definition runnable.
 370  */
 371 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 372 {
 373         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 374 }
 375
 376 /* Just ensure a guest exit from a particular CPU */
 377 static void exit_vm_noop(void *info)
 378 {
 379 }
 380
 381 void force_vm_exit(const cpumask_t *mask)
 382 {
 383         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 384 }
 385
 386 /**
 387  * need_new_vmid_gen - check that the VMID is still valid
 388  * @kvm: The VM's VMID to checkt
 389  *
 390  * return true if there is a new generation of VMIDs being used
 391  *
 392  * The hardware supports only 256 values with the value zero reserved for the
 393  * host, so we check if an assigned value belongs to a previous generation,
 394  * which which requires us to assign a new value. If we're the first to use a
 395  * VMID for the new generation, we must flush necessary caches and TLBs on all
 396  * CPUs.
 397  */
 398 static bool need_new_vmid_gen(struct kvm *kvm)
 399 {
 400         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 401 }
 402
 403 /**
 404  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 405  * @kvm The guest that we are about to run
 406  *
 407  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 408  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 409  * caches and TLBs.
 410  */
 411 static void update_vttbr(struct kvm *kvm)
 412 {
 413         phys_addr_t pgd_phys;
 414         u64 vmid;
 415
 416         if (!need_new_vmid_gen(kvm))
 417                 return;
 418
 419         spin_lock(&kvm_vmid_lock);
 420
 421         /*
 422          * We need to re-check the vmid_gen here to ensure that if another vcpu
 423          * already allocated a valid vmid for this vm, then this vcpu should
 424          * use the same vmid.
 425          */
 426         if (!need_new_vmid_gen(kvm)) {
 427                 spin_unlock(&kvm_vmid_lock);
 428                 return;
 429         }
 430
 431         /* First user of a new VMID generation? */
 432         if (unlikely(kvm_next_vmid == 0)) {
 433                 atomic64_inc(&kvm_vmid_gen);
 434                 kvm_next_vmid = 1;
 435
 436                 /*
 437                  * On SMP we know no other CPUs can use this CPU's or each
 438                  * other's VMID after force_vm_exit returns since the
 439                  * kvm_vmid_lock blocks them from reentry to the guest.
 440                  */
 441                 force_vm_exit(cpu_all_mask);
 442                 /*
 443                  * Now broadcast TLB + ICACHE invalidation over the inner
 444                  * shareable domain to make sure all data structures are
 445                  * clean.
 446                  */
 447                 kvm_call_hyp(__kvm_flush_vm_context);
 448         }
 449
 450         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 451         kvm->arch.vmid = kvm_next_vmid;
 452         kvm_next_vmid++;
 453
 454         /* update vttbr to be used with the new vmid */
 455         pgd_phys = virt_to_phys(kvm->arch.pgd);
 456         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 457         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 458         kvm->arch.vttbr |= vmid;
 459
 460         spin_unlock(&kvm_vmid_lock);
 461 }
 462
 463 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 464 {
 465         if (likely(vcpu->arch.has_run_once))
 466                 return 0;
 467
 468         vcpu->arch.has_run_once = true;
 469
 470         /*
 471          * Initialize the VGIC before running a vcpu the first time on
 472          * this VM.
 473          */
 474         if (irqchip_in_kernel(vcpu->kvm) &&
 475             unlikely(!vgic_initialized(vcpu->kvm))) {
 476                 int ret = kvm_vgic_init(vcpu->kvm);
 477                 if (ret)
 478                         return ret;
 479         }
 480
 481         /*
 482          * Handle the "start in power-off" case by calling into the
 483          * PSCI code.
 484          */
 485         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
 486                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
 487                 kvm_psci_call(vcpu);
 488         }
 489
 490         return 0;
 491 }
 492
 493 static void vcpu_pause(struct kvm_vcpu *vcpu)
 494 {
 495         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 496
 497         wait_event_interruptible(*wq, !vcpu->arch.pause);
 498 }
 499
 500 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 501 {
 502         return vcpu->arch.target >= 0;
 503 }
 504
 505 /**
 506  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 507  * @vcpu:       The VCPU pointer
 508  * @run:        The kvm_run structure pointer used for userspace state exchange
 509  *
 510  * This function is called through the VCPU_RUN ioctl called from user space. It
 511  * will execute VM code in a loop until the time slice for the process is used
 512  * or some emulation is needed from user space in which case the function will
 513  * return with return value 0 and with the kvm_run structure filled in with the
 514  * required data for the requested emulation.
 515  */
 516 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 517 {
 518         int ret;
 519         sigset_t sigsaved;
 520
 521         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 522                 return -ENOEXEC;
 523
 524         ret = kvm_vcpu_first_run_init(vcpu);
 525         if (ret)
 526                 return ret;
 527
 528         if (run->exit_reason == KVM_EXIT_MMIO) {
 529                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 530                 if (ret)
 531                         return ret;
 532         }
 533
 534         if (vcpu->sigset_active)
 535                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 536
 537         ret = 1;
 538         run->exit_reason = KVM_EXIT_UNKNOWN;
 539         while (ret > 0) {
 540                 /*
 541                  * Check conditions before entering the guest
 542                  */
 543                 cond_resched();
 544
 545                 update_vttbr(vcpu->kvm);
 546
 547                 if (vcpu->arch.pause)
 548                         vcpu_pause(vcpu);
 549
 550                 kvm_vgic_flush_hwstate(vcpu);
 551                 kvm_timer_flush_hwstate(vcpu);
 552
 553                 local_irq_disable();
 554
 555                 /*
 556                  * Re-check atomic conditions
 557                  */
 558                 if (signal_pending(current)) {
 559                         ret = -EINTR;
 560                         run->exit_reason = KVM_EXIT_INTR;
 561                 }
 562
 563                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 564                         local_irq_enable();
 565                         kvm_timer_sync_hwstate(vcpu);
 566                         kvm_vgic_sync_hwstate(vcpu);
 567                         continue;
 568                 }
 569
 570                 /**************************************************************
 571                  * Enter the guest
 572                  */
 573                 trace_kvm_entry(*vcpu_pc(vcpu));
 574                 kvm_guest_enter();
 575                 vcpu->mode = IN_GUEST_MODE;
 576
 577                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 578
 579                 vcpu->mode = OUTSIDE_GUEST_MODE;
 580                 vcpu->arch.last_pcpu = smp_processor_id();
 581                 kvm_guest_exit();
 582                 trace_kvm_exit(*vcpu_pc(vcpu));
 583                 /*
 584                  * We may have taken a host interrupt in HYP mode (ie
 585                  * while executing the guest). This interrupt is still
 586                  * pending, as we haven't serviced it yet!
 587                  *
 588                  * We're now back in SVC mode, with interrupts
 589                  * disabled.  Enabling the interrupts now will have
 590                  * the effect of taking the interrupt again, in SVC
 591                  * mode this time.
 592                  */
 593                 local_irq_enable();
 594
 595                 /*
 596                  * Back from guest
 597                  *************************************************************/
 598
 599                 kvm_timer_sync_hwstate(vcpu);
 600                 kvm_vgic_sync_hwstate(vcpu);
 601
 602                 ret = handle_exit(vcpu, run, ret);
 603         }
 604
 605         if (vcpu->sigset_active)
 606                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 607         return ret;
 608 }
 609
 610 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 611 {
 612         int bit_index;
 613         bool set;
 614         unsigned long *ptr;
 615
 616         if (number == KVM_ARM_IRQ_CPU_IRQ)
 617                 bit_index = __ffs(HCR_VI);
 618         else /* KVM_ARM_IRQ_CPU_FIQ */
 619                 bit_index = __ffs(HCR_VF);
 620
 621         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 622         if (level)
 623                 set = test_and_set_bit(bit_index, ptr);
 624         else
 625                 set = test_and_clear_bit(bit_index, ptr);
 626
 627         /*
 628          * If we didn't change anything, no need to wake up or kick other CPUs
 629          */
 630         if (set == level)
 631                 return 0;
 632
 633         /*
 634          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 635          * trigger a world-switch round on the running physical CPU to set the
 636          * virtual IRQ/FIQ fields in the HCR appropriately.
 637          */
 638         kvm_vcpu_kick(vcpu);
 639
 640         return 0;
 641 }
 642
 643 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 644                           bool line_status)
 645 {
 646         u32 irq = irq_level->irq;
 647         unsigned int irq_type, vcpu_idx, irq_num;
 648         int nrcpus = atomic_read(&kvm->online_vcpus);
 649         struct kvm_vcpu *vcpu = NULL;
 650         bool level = irq_level->level;
 651
 652         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 653         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 654         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 655
 656         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 657
 658         switch (irq_type) {
 659         case KVM_ARM_IRQ_TYPE_CPU:
 660                 if (irqchip_in_kernel(kvm))
 661                         return -ENXIO;
 662
 663                 if (vcpu_idx >= nrcpus)
 664                         return -EINVAL;
 665
 666                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 667                 if (!vcpu)
 668                         return -EINVAL;
 669
 670                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 671                         return -EINVAL;
 672
 673                 return vcpu_interrupt_line(vcpu, irq_num, level);
 674         case KVM_ARM_IRQ_TYPE_PPI:
 675                 if (!irqchip_in_kernel(kvm))
 676                         return -ENXIO;
 677
 678                 if (vcpu_idx >= nrcpus)
 679                         return -EINVAL;
 680
 681                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 682                 if (!vcpu)
 683                         return -EINVAL;
 684
 685                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 686                         return -EINVAL;
 687
 688                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 689         case KVM_ARM_IRQ_TYPE_SPI:
 690                 if (!irqchip_in_kernel(kvm))
 691                         return -ENXIO;
 692
 693                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 694                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 695                         return -EINVAL;
 696
 697                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 698         }
 699
 700         return -EINVAL;
 701 }
 702
 703 long kvm_arch_vcpu_ioctl(struct file *filp,
 704                          unsigned int ioctl, unsigned long arg)
 705 {
 706         struct kvm_vcpu *vcpu = filp->private_data;
 707         void __user *argp = (void __user *)arg;
 708
 709         switch (ioctl) {
 710         case KVM_ARM_VCPU_INIT: {
 711                 struct kvm_vcpu_init init;
 712
 713                 if (copy_from_user(&init, argp, sizeof(init)))
 714                         return -EFAULT;
 715
 716                 return kvm_vcpu_set_target(vcpu, &init);
 717
 718         }
 719         case KVM_SET_ONE_REG:
 720         case KVM_GET_ONE_REG: {
 721                 struct kvm_one_reg reg;
 722
 723                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 724                         return -ENOEXEC;
 725
 726                 if (copy_from_user(&reg, argp, sizeof(reg)))
 727                         return -EFAULT;
 728                 if (ioctl == KVM_SET_ONE_REG)
 729                         return kvm_arm_set_reg(vcpu, &reg);
 730                 else
 731                         return kvm_arm_get_reg(vcpu, &reg);
 732         }
 733         case KVM_GET_REG_LIST: {
 734                 struct kvm_reg_list __user *user_list = argp;
 735                 struct kvm_reg_list reg_list;
 736                 unsigned n;
 737
 738                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 739                         return -ENOEXEC;
 740
 741                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 742                         return -EFAULT;
 743                 n = reg_list.n;
 744                 reg_list.n = kvm_arm_num_regs(vcpu);
 745                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 746                         return -EFAULT;
 747                 if (n < reg_list.n)
 748                         return -E2BIG;
 749                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 750         }
 751         default:
 752                 return -EINVAL;
 753         }
 754 }
 755
 756 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 757 {
 758         return -EINVAL;
 759 }
 760
 761 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 762                                         struct kvm_arm_device_addr *dev_addr)
 763 {
 764         unsigned long dev_id, type;
 765
 766         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 767                 KVM_ARM_DEVICE_ID_SHIFT;
 768         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 769                 KVM_ARM_DEVICE_TYPE_SHIFT;
 770
 771         switch (dev_id) {
 772         case KVM_ARM_DEVICE_VGIC_V2:
 773                 if (!vgic_present)
 774                         return -ENXIO;
 775                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
 776         default:
 777                 return -ENODEV;
 778         }
 779 }
 780
 781 long kvm_arch_vm_ioctl(struct file *filp,
 782                        unsigned int ioctl, unsigned long arg)
 783 {
 784         struct kvm *kvm = filp->private_data;
 785         void __user *argp = (void __user *)arg;
 786
 787         switch (ioctl) {
 788         case KVM_CREATE_IRQCHIP: {
 789                 if (vgic_present)
 790                         return kvm_vgic_create(kvm);
 791                 else
 792                         return -ENXIO;
 793         }
 794         case KVM_ARM_SET_DEVICE_ADDR: {
 795                 struct kvm_arm_device_addr dev_addr;
 796
 797                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 798                         return -EFAULT;
 799                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 800         }
 801         case KVM_ARM_PREFERRED_TARGET: {
 802                 int err;
 803                 struct kvm_vcpu_init init;
 804
 805                 err = kvm_vcpu_preferred_target(&init);
 806                 if (err)
 807                         return err;
 808
 809                 if (copy_to_user(argp, &init, sizeof(init)))
 810                         return -EFAULT;
 811
 812                 return 0;
 813         }
 814         default:
 815                 return -EINVAL;
 816         }
 817 }
 818
 819 static void cpu_init_hyp_mode(void *dummy)
 820 {
 821         phys_addr_t boot_pgd_ptr;
 822         phys_addr_t pgd_ptr;
 823         unsigned long hyp_stack_ptr;
 824         unsigned long stack_page;
 825         unsigned long vector_ptr;
 826
 827         /* Switch from the HYP stub to our own HYP init vector */
 828         __hyp_set_vectors(kvm_get_idmap_vector());
 829
 830         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 831         pgd_ptr = kvm_mmu_get_httbr();
 832         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 833         hyp_stack_ptr = stack_page + PAGE_SIZE;
 834         vector_ptr = (unsigned long)__kvm_hyp_vector;
 835
 836         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 837 }
 838
 839 static int hyp_init_cpu_notify(struct notifier_block *self,
 840                                unsigned long action, void *cpu)
 841 {
 842         switch (action) {
 843         case CPU_STARTING:
 844         case CPU_STARTING_FROZEN:
 845                 cpu_init_hyp_mode(NULL);
 846                 break;
 847         }
 848
 849         return NOTIFY_OK;
 850 }
 851
 852 static struct notifier_block hyp_init_cpu_nb = {
 853         .notifier_call = hyp_init_cpu_notify,
 854 };
 855
 856 /**
 857  * Inits Hyp-mode on all online CPUs
 858  */
 859 static int init_hyp_mode(void)
 860 {
 861         int cpu;
 862         int err = 0;
 863
 864         /*
 865          * Allocate Hyp PGD and setup Hyp identity mapping
 866          */
 867         err = kvm_mmu_init();
 868         if (err)
 869                 goto out_err;
 870
 871         /*
 872          * It is probably enough to obtain the default on one
 873          * CPU. It's unlikely to be different on the others.
 874          */
 875         hyp_default_vectors = __hyp_get_vectors();
 876
 877         /*
 878          * Allocate stack pages for Hypervisor-mode
 879          */
 880         for_each_possible_cpu(cpu) {
 881                 unsigned long stack_page;
 882
 883                 stack_page = __get_free_page(GFP_KERNEL);
 884                 if (!stack_page) {
 885                         err = -ENOMEM;
 886                         goto out_free_stack_pages;
 887                 }
 888
 889                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 890         }
 891
 892         /*
 893          * Map the Hyp-code called directly from the host
 894          */
 895         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 896         if (err) {
 897                 kvm_err("Cannot map world-switch code\n");
 898                 goto out_free_mappings;
 899         }
 900
 901         /*
 902          * Map the Hyp stack pages
 903          */
 904         for_each_possible_cpu(cpu) {
 905                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 906                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 907
 908                 if (err) {
 909                         kvm_err("Cannot map hyp stack\n");
 910                         goto out_free_mappings;
 911                 }
 912         }
 913
 914         /*
 915          * Map the host CPU structures
 916          */
 917         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 918         if (!kvm_host_cpu_state) {
 919                 err = -ENOMEM;
 920                 kvm_err("Cannot allocate host CPU state\n");
 921                 goto out_free_mappings;
 922         }
 923
 924         for_each_possible_cpu(cpu) {
 925                 kvm_cpu_context_t *cpu_ctxt;
 926
 927                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 928                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 929
 930                 if (err) {
 931                         kvm_err("Cannot map host CPU state: %d\n", err);
 932                         goto out_free_context;
 933                 }
 934         }
 935
 936         /*
 937          * Execute the init code on each CPU.
 938          */
 939         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 940
 941         /*
 942          * Init HYP view of VGIC
 943          */
 944         err = kvm_vgic_hyp_init();
 945         if (err)
 946                 goto out_free_context;
 947
 948 #ifdef CONFIG_KVM_ARM_VGIC
 949                 vgic_present = true;
 950 #endif
 951
 952         /*
 953          * Init HYP architected timer support
 954          */
 955         err = kvm_timer_hyp_init();
 956         if (err)
 957                 goto out_free_mappings;
 958
 959 #ifndef CONFIG_HOTPLUG_CPU
 960         free_boot_hyp_pgd();
 961 #endif
 962
 963         kvm_perf_init();
 964
 965         kvm_info("Hyp mode initialized successfully\n");
 966
 967         return 0;
 968 out_free_context:
 969         free_percpu(kvm_host_cpu_state);
 970 out_free_mappings:
 971         free_hyp_pgds();
 972 out_free_stack_pages:
 973         for_each_possible_cpu(cpu)
 974                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 975 out_err:
 976         kvm_err("error initializing Hyp mode: %d\n", err);
 977         return err;
 978 }
 979
 980 static void check_kvm_target_cpu(void *ret)
 981 {
 982         *(int *)ret = kvm_target_cpu();
 983 }
 984
 985 /**
 986  * Initialize Hyp-mode and memory mappings on all CPUs.
 987  */
 988 int kvm_arch_init(void *opaque)
 989 {
 990         int err;
 991         int ret, cpu;
 992
 993         if (!is_hyp_mode_available()) {
 994                 kvm_err("HYP mode not available\n");
 995                 return -ENODEV;
 996         }
 997
 998         for_each_online_cpu(cpu) {
 999                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1000                 if (ret < 0) {
1001                         kvm_err("Error, CPU %d not supported!\n", cpu);
1002                         return -ENODEV;
1003                 }
1004         }
1005
1006         err = init_hyp_mode();
1007         if (err)
1008                 goto out_err;
1009
1010         err = register_cpu_notifier(&hyp_init_cpu_nb);
1011         if (err) {
1012                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1013                 goto out_err;
1014         }
1015
1016         kvm_coproc_table_init();
1017         return 0;
1018 out_err:
1019         return err;
1020 }
1021
1022 /* NOP: Compiling as a module not supported */
1023 void kvm_arch_exit(void)
1024 {
1025         kvm_perf_teardown();
1026 }
1027
1028 static int arm_init(void)
1029 {
1030         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1031         return rc;
1032 }
1033
1034 module_init(arm_init);