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