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/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.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 <trace/events/kvm.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/ptrace.h>
  37 #include <asm/mman.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/virt.h>
  41 #include <asm/kvm_arm.h>
  42 #include <asm/kvm_asm.h>
  43 #include <asm/kvm_mmu.h>
  44 #include <asm/kvm_emulate.h>
  45 #include <asm/kvm_coproc.h>
  46 #include <asm/kvm_psci.h>
  47
  48 #ifdef REQUIRES_VIRT
  49 __asm__(".arch_extension        virt");
  50 #endif
  51
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* Per-CPU variable containing the currently running vcpu. */
  57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  58
  59 /* The VMID used in the VTTBR */
  60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  61 static u8 kvm_next_vmid;
  62 static DEFINE_SPINLOCK(kvm_vmid_lock);
  63
  64 static bool vgic_present;
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         BUG_ON(preemptible());
  69         __get_cpu_var(kvm_arm_running_vcpu) = vcpu;
  70 }
  71
  72 /**
  73  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  74  * Must be called from non-preemptible context
  75  */
  76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  77 {
  78         BUG_ON(preemptible());
  79         return __get_cpu_var(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_hardware_enable(void *garbage)
  91 {
  92         return 0;
  93 }
  94
  95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  96 {
  97         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  98 }
  99
 100 void kvm_arch_hardware_disable(void *garbage)
 101 {
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_hardware_unsetup(void)
 110 {
 111 }
 112
 113 void kvm_arch_check_processor_compat(void *rtn)
 114 {
 115         *(int *)rtn = 0;
 116 }
 117
 118 void kvm_arch_sync_events(struct kvm *kvm)
 119 {
 120 }
 121
 122 /**
 123  * kvm_arch_init_vm - initializes a VM data structure
 124  * @kvm:        pointer to the KVM struct
 125  */
 126 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 127 {
 128         int ret = 0;
 129
 130         if (type)
 131                 return -EINVAL;
 132
 133         ret = kvm_alloc_stage2_pgd(kvm);
 134         if (ret)
 135                 goto out_fail_alloc;
 136
 137         ret = create_hyp_mappings(kvm, kvm + 1);
 138         if (ret)
 139                 goto out_free_stage2_pgd;
 140
 141         /* Mark the initial VMID generation invalid */
 142         kvm->arch.vmid_gen = 0;
 143
 144         return ret;
 145 out_free_stage2_pgd:
 146         kvm_free_stage2_pgd(kvm);
 147 out_fail_alloc:
 148         return ret;
 149 }
 150
 151 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 152 {
 153         return VM_FAULT_SIGBUS;
 154 }
 155
 156 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
 157                            struct kvm_memory_slot *dont)
 158 {
 159 }
 160
 161 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, 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 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 224                                    struct kvm_memory_slot *memslot,
 225                                    struct kvm_userspace_memory_region *mem,
 226                                    enum kvm_mr_change change)
 227 {
 228         return 0;
 229 }
 230
 231 void kvm_arch_commit_memory_region(struct kvm *kvm,
 232                                    struct kvm_userspace_memory_region *mem,
 233                                    const struct kvm_memory_slot *old,
 234                                    enum kvm_mr_change change)
 235 {
 236 }
 237
 238 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 239 {
 240 }
 241
 242 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 243                                    struct kvm_memory_slot *slot)
 244 {
 245 }
 246
 247 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 248 {
 249         int err;
 250         struct kvm_vcpu *vcpu;
 251
 252         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 253         if (!vcpu) {
 254                 err = -ENOMEM;
 255                 goto out;
 256         }
 257
 258         err = kvm_vcpu_init(vcpu, kvm, id);
 259         if (err)
 260                 goto free_vcpu;
 261
 262         err = create_hyp_mappings(vcpu, vcpu + 1);
 263         if (err)
 264                 goto vcpu_uninit;
 265
 266         return vcpu;
 267 vcpu_uninit:
 268         kvm_vcpu_uninit(vcpu);
 269 free_vcpu:
 270         kmem_cache_free(kvm_vcpu_cache, vcpu);
 271 out:
 272         return ERR_PTR(err);
 273 }
 274
 275 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 276 {
 277         return 0;
 278 }
 279
 280 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 281 {
 282         kvm_mmu_free_memory_caches(vcpu);
 283         kvm_timer_vcpu_terminate(vcpu);
 284         kmem_cache_free(kvm_vcpu_cache, vcpu);
 285 }
 286
 287 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 288 {
 289         kvm_arch_vcpu_free(vcpu);
 290 }
 291
 292 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 293 {
 294         return 0;
 295 }
 296
 297 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 298 {
 299         int ret;
 300
 301         /* Force users to call KVM_ARM_VCPU_INIT */
 302         vcpu->arch.target = -1;
 303
 304         /* Set up VGIC */
 305         ret = kvm_vgic_vcpu_init(vcpu);
 306         if (ret)
 307                 return ret;
 308
 309         /* Set up the timer */
 310         kvm_timer_vcpu_init(vcpu);
 311
 312         return 0;
 313 }
 314
 315 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 316 {
 317 }
 318
 319 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 320 {
 321         vcpu->cpu = cpu;
 322         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 323
 324         /*
 325          * Check whether this vcpu requires the cache to be flushed on
 326          * this physical CPU. This is a consequence of doing dcache
 327          * operations by set/way on this vcpu. We do it here to be in
 328          * a non-preemptible section.
 329          */
 330         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 331                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 332
 333         kvm_arm_set_running_vcpu(vcpu);
 334 }
 335
 336 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 337 {
 338         kvm_arm_set_running_vcpu(NULL);
 339 }
 340
 341 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 342                                         struct kvm_guest_debug *dbg)
 343 {
 344         return -EINVAL;
 345 }
 346
 347
 348 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 349                                     struct kvm_mp_state *mp_state)
 350 {
 351         return -EINVAL;
 352 }
 353
 354 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 355                                     struct kvm_mp_state *mp_state)
 356 {
 357         return -EINVAL;
 358 }
 359
 360 /**
 361  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 362  * @v:          The VCPU pointer
 363  *
 364  * If the guest CPU is not waiting for interrupts or an interrupt line is
 365  * asserted, the CPU is by definition runnable.
 366  */
 367 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 368 {
 369         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 370 }
 371
 372 /* Just ensure a guest exit from a particular CPU */
 373 static void exit_vm_noop(void *info)
 374 {
 375 }
 376
 377 void force_vm_exit(const cpumask_t *mask)
 378 {
 379         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 380 }
 381
 382 /**
 383  * need_new_vmid_gen - check that the VMID is still valid
 384  * @kvm: The VM's VMID to checkt
 385  *
 386  * return true if there is a new generation of VMIDs being used
 387  *
 388  * The hardware supports only 256 values with the value zero reserved for the
 389  * host, so we check if an assigned value belongs to a previous generation,
 390  * which which requires us to assign a new value. If we're the first to use a
 391  * VMID for the new generation, we must flush necessary caches and TLBs on all
 392  * CPUs.
 393  */
 394 static bool need_new_vmid_gen(struct kvm *kvm)
 395 {
 396         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 397 }
 398
 399 /**
 400  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 401  * @kvm The guest that we are about to run
 402  *
 403  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 404  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 405  * caches and TLBs.
 406  */
 407 static void update_vttbr(struct kvm *kvm)
 408 {
 409         phys_addr_t pgd_phys;
 410         u64 vmid;
 411
 412         if (!need_new_vmid_gen(kvm))
 413                 return;
 414
 415         spin_lock(&kvm_vmid_lock);
 416
 417         /*
 418          * We need to re-check the vmid_gen here to ensure that if another vcpu
 419          * already allocated a valid vmid for this vm, then this vcpu should
 420          * use the same vmid.
 421          */
 422         if (!need_new_vmid_gen(kvm)) {
 423                 spin_unlock(&kvm_vmid_lock);
 424                 return;
 425         }
 426
 427         /* First user of a new VMID generation? */
 428         if (unlikely(kvm_next_vmid == 0)) {
 429                 atomic64_inc(&kvm_vmid_gen);
 430                 kvm_next_vmid = 1;
 431
 432                 /*
 433                  * On SMP we know no other CPUs can use this CPU's or each
 434                  * other's VMID after force_vm_exit returns since the
 435                  * kvm_vmid_lock blocks them from reentry to the guest.
 436                  */
 437                 force_vm_exit(cpu_all_mask);
 438                 /*
 439                  * Now broadcast TLB + ICACHE invalidation over the inner
 440                  * shareable domain to make sure all data structures are
 441                  * clean.
 442                  */
 443                 kvm_call_hyp(__kvm_flush_vm_context);
 444         }
 445
 446         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 447         kvm->arch.vmid = kvm_next_vmid;
 448         kvm_next_vmid++;
 449
 450         /* update vttbr to be used with the new vmid */
 451         pgd_phys = virt_to_phys(kvm->arch.pgd);
 452         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 453         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 454         kvm->arch.vttbr |= vmid;
 455
 456         spin_unlock(&kvm_vmid_lock);
 457 }
 458
 459 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 460 {
 461         if (likely(vcpu->arch.has_run_once))
 462                 return 0;
 463
 464         vcpu->arch.has_run_once = true;
 465
 466         /*
 467          * Initialize the VGIC before running a vcpu the first time on
 468          * this VM.
 469          */
 470         if (irqchip_in_kernel(vcpu->kvm) &&
 471             unlikely(!vgic_initialized(vcpu->kvm))) {
 472                 int ret = kvm_vgic_init(vcpu->kvm);
 473                 if (ret)
 474                         return ret;
 475         }
 476
 477         /*
 478          * Handle the "start in power-off" case by calling into the
 479          * PSCI code.
 480          */
 481         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
 482                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
 483                 kvm_psci_call(vcpu);
 484         }
 485
 486         return 0;
 487 }
 488
 489 static void vcpu_pause(struct kvm_vcpu *vcpu)
 490 {
 491         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 492
 493         wait_event_interruptible(*wq, !vcpu->arch.pause);
 494 }
 495
 496 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 497 {
 498         return vcpu->arch.target >= 0;
 499 }
 500
 501 /**
 502  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 503  * @vcpu:       The VCPU pointer
 504  * @run:        The kvm_run structure pointer used for userspace state exchange
 505  *
 506  * This function is called through the VCPU_RUN ioctl called from user space. It
 507  * will execute VM code in a loop until the time slice for the process is used
 508  * or some emulation is needed from user space in which case the function will
 509  * return with return value 0 and with the kvm_run structure filled in with the
 510  * required data for the requested emulation.
 511  */
 512 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 513 {
 514         int ret;
 515         sigset_t sigsaved;
 516
 517         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 518                 return -ENOEXEC;
 519
 520         ret = kvm_vcpu_first_run_init(vcpu);
 521         if (ret)
 522                 return ret;
 523
 524         if (run->exit_reason == KVM_EXIT_MMIO) {
 525                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 526                 if (ret)
 527                         return ret;
 528         }
 529
 530         if (vcpu->sigset_active)
 531                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 532
 533         ret = 1;
 534         run->exit_reason = KVM_EXIT_UNKNOWN;
 535         while (ret > 0) {
 536                 /*
 537                  * Check conditions before entering the guest
 538                  */
 539                 cond_resched();
 540
 541                 update_vttbr(vcpu->kvm);
 542
 543                 if (vcpu->arch.pause)
 544                         vcpu_pause(vcpu);
 545
 546                 kvm_vgic_flush_hwstate(vcpu);
 547                 kvm_timer_flush_hwstate(vcpu);
 548
 549                 local_irq_disable();
 550
 551                 /*
 552                  * Re-check atomic conditions
 553                  */
 554                 if (signal_pending(current)) {
 555                         ret = -EINTR;
 556                         run->exit_reason = KVM_EXIT_INTR;
 557                 }
 558
 559                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 560                         local_irq_enable();
 561                         kvm_timer_sync_hwstate(vcpu);
 562                         kvm_vgic_sync_hwstate(vcpu);
 563                         continue;
 564                 }
 565
 566                 /**************************************************************
 567                  * Enter the guest
 568                  */
 569                 trace_kvm_entry(*vcpu_pc(vcpu));
 570                 kvm_guest_enter();
 571                 vcpu->mode = IN_GUEST_MODE;
 572
 573                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 574
 575                 vcpu->mode = OUTSIDE_GUEST_MODE;
 576                 vcpu->arch.last_pcpu = smp_processor_id();
 577                 kvm_guest_exit();
 578                 trace_kvm_exit(*vcpu_pc(vcpu));
 579                 /*
 580                  * We may have taken a host interrupt in HYP mode (ie
 581                  * while executing the guest). This interrupt is still
 582                  * pending, as we haven't serviced it yet!
 583                  *
 584                  * We're now back in SVC mode, with interrupts
 585                  * disabled.  Enabling the interrupts now will have
 586                  * the effect of taking the interrupt again, in SVC
 587                  * mode this time.
 588                  */
 589                 local_irq_enable();
 590
 591                 /*
 592                  * Back from guest
 593                  *************************************************************/
 594
 595                 kvm_timer_sync_hwstate(vcpu);
 596                 kvm_vgic_sync_hwstate(vcpu);
 597
 598                 ret = handle_exit(vcpu, run, ret);
 599         }
 600
 601         if (vcpu->sigset_active)
 602                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 603         return ret;
 604 }
 605
 606 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 607 {
 608         int bit_index;
 609         bool set;
 610         unsigned long *ptr;
 611
 612         if (number == KVM_ARM_IRQ_CPU_IRQ)
 613                 bit_index = __ffs(HCR_VI);
 614         else /* KVM_ARM_IRQ_CPU_FIQ */
 615                 bit_index = __ffs(HCR_VF);
 616
 617         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 618         if (level)
 619                 set = test_and_set_bit(bit_index, ptr);
 620         else
 621                 set = test_and_clear_bit(bit_index, ptr);
 622
 623         /*
 624          * If we didn't change anything, no need to wake up or kick other CPUs
 625          */
 626         if (set == level)
 627                 return 0;
 628
 629         /*
 630          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 631          * trigger a world-switch round on the running physical CPU to set the
 632          * virtual IRQ/FIQ fields in the HCR appropriately.
 633          */
 634         kvm_vcpu_kick(vcpu);
 635
 636         return 0;
 637 }
 638
 639 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 640                           bool line_status)
 641 {
 642         u32 irq = irq_level->irq;
 643         unsigned int irq_type, vcpu_idx, irq_num;
 644         int nrcpus = atomic_read(&kvm->online_vcpus);
 645         struct kvm_vcpu *vcpu = NULL;
 646         bool level = irq_level->level;
 647
 648         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 649         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 650         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 651
 652         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 653
 654         switch (irq_type) {
 655         case KVM_ARM_IRQ_TYPE_CPU:
 656                 if (irqchip_in_kernel(kvm))
 657                         return -ENXIO;
 658
 659                 if (vcpu_idx >= nrcpus)
 660                         return -EINVAL;
 661
 662                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 663                 if (!vcpu)
 664                         return -EINVAL;
 665
 666                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 667                         return -EINVAL;
 668
 669                 return vcpu_interrupt_line(vcpu, irq_num, level);
 670         case KVM_ARM_IRQ_TYPE_PPI:
 671                 if (!irqchip_in_kernel(kvm))
 672                         return -ENXIO;
 673
 674                 if (vcpu_idx >= nrcpus)
 675                         return -EINVAL;
 676
 677                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 678                 if (!vcpu)
 679                         return -EINVAL;
 680
 681                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 682                         return -EINVAL;
 683
 684                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 685         case KVM_ARM_IRQ_TYPE_SPI:
 686                 if (!irqchip_in_kernel(kvm))
 687                         return -ENXIO;
 688
 689                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 690                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 691                         return -EINVAL;
 692
 693                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 694         }
 695
 696         return -EINVAL;
 697 }
 698
 699 long kvm_arch_vcpu_ioctl(struct file *filp,
 700                          unsigned int ioctl, unsigned long arg)
 701 {
 702         struct kvm_vcpu *vcpu = filp->private_data;
 703         void __user *argp = (void __user *)arg;
 704
 705         switch (ioctl) {
 706         case KVM_ARM_VCPU_INIT: {
 707                 struct kvm_vcpu_init init;
 708
 709                 if (copy_from_user(&init, argp, sizeof(init)))
 710                         return -EFAULT;
 711
 712                 return kvm_vcpu_set_target(vcpu, &init);
 713
 714         }
 715         case KVM_SET_ONE_REG:
 716         case KVM_GET_ONE_REG: {
 717                 struct kvm_one_reg reg;
 718
 719                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 720                         return -ENOEXEC;
 721
 722                 if (copy_from_user(&reg, argp, sizeof(reg)))
 723                         return -EFAULT;
 724                 if (ioctl == KVM_SET_ONE_REG)
 725                         return kvm_arm_set_reg(vcpu, &reg);
 726                 else
 727                         return kvm_arm_get_reg(vcpu, &reg);
 728         }
 729         case KVM_GET_REG_LIST: {
 730                 struct kvm_reg_list __user *user_list = argp;
 731                 struct kvm_reg_list reg_list;
 732                 unsigned n;
 733
 734                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 735                         return -ENOEXEC;
 736
 737                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 738                         return -EFAULT;
 739                 n = reg_list.n;
 740                 reg_list.n = kvm_arm_num_regs(vcpu);
 741                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 742                         return -EFAULT;
 743                 if (n < reg_list.n)
 744                         return -E2BIG;
 745                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 746         }
 747         default:
 748                 return -EINVAL;
 749         }
 750 }
 751
 752 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 753 {
 754         return -EINVAL;
 755 }
 756
 757 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 758                                         struct kvm_arm_device_addr *dev_addr)
 759 {
 760         unsigned long dev_id, type;
 761
 762         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 763                 KVM_ARM_DEVICE_ID_SHIFT;
 764         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 765                 KVM_ARM_DEVICE_TYPE_SHIFT;
 766
 767         switch (dev_id) {
 768         case KVM_ARM_DEVICE_VGIC_V2:
 769                 if (!vgic_present)
 770                         return -ENXIO;
 771                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
 772         default:
 773                 return -ENODEV;
 774         }
 775 }
 776
 777 long kvm_arch_vm_ioctl(struct file *filp,
 778                        unsigned int ioctl, unsigned long arg)
 779 {
 780         struct kvm *kvm = filp->private_data;
 781         void __user *argp = (void __user *)arg;
 782
 783         switch (ioctl) {
 784         case KVM_CREATE_IRQCHIP: {
 785                 if (vgic_present)
 786                         return kvm_vgic_create(kvm);
 787                 else
 788                         return -ENXIO;
 789         }
 790         case KVM_ARM_SET_DEVICE_ADDR: {
 791                 struct kvm_arm_device_addr dev_addr;
 792
 793                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 794                         return -EFAULT;
 795                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 796         }
 797         default:
 798                 return -EINVAL;
 799         }
 800 }
 801
 802 static void cpu_init_hyp_mode(void *dummy)
 803 {
 804         unsigned long long boot_pgd_ptr;
 805         unsigned long long pgd_ptr;
 806         unsigned long hyp_stack_ptr;
 807         unsigned long stack_page;
 808         unsigned long vector_ptr;
 809
 810         /* Switch from the HYP stub to our own HYP init vector */
 811         __hyp_set_vectors(kvm_get_idmap_vector());
 812
 813         boot_pgd_ptr = (unsigned long long)kvm_mmu_get_boot_httbr();
 814         pgd_ptr = (unsigned long long)kvm_mmu_get_httbr();
 815         stack_page = __get_cpu_var(kvm_arm_hyp_stack_page);
 816         hyp_stack_ptr = stack_page + PAGE_SIZE;
 817         vector_ptr = (unsigned long)__kvm_hyp_vector;
 818
 819         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 820 }
 821
 822 static int hyp_init_cpu_notify(struct notifier_block *self,
 823                                unsigned long action, void *cpu)
 824 {
 825         switch (action) {
 826         case CPU_STARTING:
 827         case CPU_STARTING_FROZEN:
 828                 cpu_init_hyp_mode(NULL);
 829                 break;
 830         }
 831
 832         return NOTIFY_OK;
 833 }
 834
 835 static struct notifier_block hyp_init_cpu_nb = {
 836         .notifier_call = hyp_init_cpu_notify,
 837 };
 838
 839 #ifdef CONFIG_CPU_PM
 840 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 841                                     unsigned long cmd,
 842                                     void *v)
 843 {
 844         if (cmd == CPU_PM_EXIT) {
 845                 cpu_init_hyp_mode(NULL);
 846                 return NOTIFY_OK;
 847         }
 848
 849         return NOTIFY_DONE;
 850 }
 851
 852 static struct notifier_block hyp_init_cpu_pm_nb = {
 853         .notifier_call = hyp_init_cpu_pm_notifier,
 854 };
 855
 856 static void __init hyp_cpu_pm_init(void)
 857 {
 858         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 859 }
 860 #else
 861 static inline void hyp_cpu_pm_init(void)
 862 {
 863 }
 864 #endif
 865
 866 /**
 867  * Inits Hyp-mode on all online CPUs
 868  */
 869 static int init_hyp_mode(void)
 870 {
 871         int cpu;
 872         int err = 0;
 873
 874         /*
 875          * Allocate Hyp PGD and setup Hyp identity mapping
 876          */
 877         err = kvm_mmu_init();
 878         if (err)
 879                 goto out_err;
 880
 881         /*
 882          * It is probably enough to obtain the default on one
 883          * CPU. It's unlikely to be different on the others.
 884          */
 885         hyp_default_vectors = __hyp_get_vectors();
 886
 887         /*
 888          * Allocate stack pages for Hypervisor-mode
 889          */
 890         for_each_possible_cpu(cpu) {
 891                 unsigned long stack_page;
 892
 893                 stack_page = __get_free_page(GFP_KERNEL);
 894                 if (!stack_page) {
 895                         err = -ENOMEM;
 896                         goto out_free_stack_pages;
 897                 }
 898
 899                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 900         }
 901
 902         /*
 903          * Map the Hyp-code called directly from the host
 904          */
 905         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 906         if (err) {
 907                 kvm_err("Cannot map world-switch code\n");
 908                 goto out_free_mappings;
 909         }
 910
 911         /*
 912          * Map the Hyp stack pages
 913          */
 914         for_each_possible_cpu(cpu) {
 915                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 916                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 917
 918                 if (err) {
 919                         kvm_err("Cannot map hyp stack\n");
 920                         goto out_free_mappings;
 921                 }
 922         }
 923
 924         /*
 925          * Map the host CPU structures
 926          */
 927         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 928         if (!kvm_host_cpu_state) {
 929                 err = -ENOMEM;
 930                 kvm_err("Cannot allocate host CPU state\n");
 931                 goto out_free_mappings;
 932         }
 933
 934         for_each_possible_cpu(cpu) {
 935                 kvm_cpu_context_t *cpu_ctxt;
 936
 937                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 938                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 939
 940                 if (err) {
 941                         kvm_err("Cannot map host CPU state: %d\n", err);
 942                         goto out_free_context;
 943                 }
 944         }
 945
 946         /*
 947          * Execute the init code on each CPU.
 948          */
 949         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 950
 951         /*
 952          * Init HYP view of VGIC
 953          */
 954         err = kvm_vgic_hyp_init();
 955         if (err)
 956                 goto out_free_context;
 957
 958 #ifdef CONFIG_KVM_ARM_VGIC
 959                 vgic_present = true;
 960 #endif
 961
 962         /*
 963          * Init HYP architected timer support
 964          */
 965         err = kvm_timer_hyp_init();
 966         if (err)
 967                 goto out_free_mappings;
 968
 969 #ifndef CONFIG_HOTPLUG_CPU
 970         free_boot_hyp_pgd();
 971 #endif
 972
 973         kvm_perf_init();
 974
 975         kvm_info("Hyp mode initialized successfully\n");
 976
 977         return 0;
 978 out_free_context:
 979         free_percpu(kvm_host_cpu_state);
 980 out_free_mappings:
 981         free_hyp_pgds();
 982 out_free_stack_pages:
 983         for_each_possible_cpu(cpu)
 984                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 985 out_err:
 986         kvm_err("error initializing Hyp mode: %d\n", err);
 987         return err;
 988 }
 989
 990 static void check_kvm_target_cpu(void *ret)
 991 {
 992         *(int *)ret = kvm_target_cpu();
 993 }
 994
 995 /**
 996  * Initialize Hyp-mode and memory mappings on all CPUs.
 997  */
 998 int kvm_arch_init(void *opaque)
 999 {
1000         int err;
1001         int ret, cpu;
1002
1003         if (!is_hyp_mode_available()) {
1004                 kvm_err("HYP mode not available\n");
1005                 return -ENODEV;
1006         }
1007
1008         for_each_online_cpu(cpu) {
1009                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1010                 if (ret < 0) {
1011                         kvm_err("Error, CPU %d not supported!\n", cpu);
1012                         return -ENODEV;
1013                 }
1014         }
1015
1016         err = init_hyp_mode();
1017         if (err)
1018                 goto out_err;
1019
1020         err = register_cpu_notifier(&hyp_init_cpu_nb);
1021         if (err) {
1022                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1023                 goto out_err;
1024         }
1025
1026         hyp_cpu_pm_init();
1027
1028         kvm_coproc_table_init();
1029         return 0;
1030 out_err:
1031         return err;
1032 }
1033
1034 /* NOP: Compiling as a module not supported */
1035 void kvm_arch_exit(void)
1036 {
1037         kvm_perf_teardown();
1038 }
1039
1040 static int arm_init(void)
1041 {
1042         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1043         return rc;
1044 }
1045
1046 module_init(arm_init);