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         __this_cpu_write(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 __this_cpu_read(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         kvm_timer_init(kvm);
 142
 143         /* Mark the initial VMID generation invalid */
 144         kvm->arch.vmid_gen = 0;
 145
 146         return ret;
 147 out_free_stage2_pgd:
 148         kvm_free_stage2_pgd(kvm);
 149 out_fail_alloc:
 150         return ret;
 151 }
 152
 153 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 154 {
 155         return VM_FAULT_SIGBUS;
 156 }
 157
 158
 159 /**
 160  * kvm_arch_destroy_vm - destroy the VM data structure
 161  * @kvm:        pointer to the KVM struct
 162  */
 163 void kvm_arch_destroy_vm(struct kvm *kvm)
 164 {
 165         int i;
 166
 167         kvm_free_stage2_pgd(kvm);
 168
 169         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 170                 if (kvm->vcpus[i]) {
 171                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 172                         kvm->vcpus[i] = NULL;
 173                 }
 174         }
 175 }
 176
 177 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 178 {
 179         int r;
 180         switch (ext) {
 181         case KVM_CAP_IRQCHIP:
 182                 r = vgic_present;
 183                 break;
 184         case KVM_CAP_DEVICE_CTRL:
 185         case KVM_CAP_USER_MEMORY:
 186         case KVM_CAP_SYNC_MMU:
 187         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 188         case KVM_CAP_ONE_REG:
 189         case KVM_CAP_ARM_PSCI:
 190         case KVM_CAP_ARM_PSCI_0_2:
 191         case KVM_CAP_READONLY_MEM:
 192                 r = 1;
 193                 break;
 194         case KVM_CAP_COALESCED_MMIO:
 195                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 196                 break;
 197         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 198                 r = 1;
 199                 break;
 200         case KVM_CAP_NR_VCPUS:
 201                 r = num_online_cpus();
 202                 break;
 203         case KVM_CAP_MAX_VCPUS:
 204                 r = KVM_MAX_VCPUS;
 205                 break;
 206         default:
 207                 r = kvm_arch_dev_ioctl_check_extension(ext);
 208                 break;
 209         }
 210         return r;
 211 }
 212
 213 long kvm_arch_dev_ioctl(struct file *filp,
 214                         unsigned int ioctl, unsigned long arg)
 215 {
 216         return -EINVAL;
 217 }
 218
 219
 220 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 221 {
 222         int err;
 223         struct kvm_vcpu *vcpu;
 224
 225         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 226         if (!vcpu) {
 227                 err = -ENOMEM;
 228                 goto out;
 229         }
 230
 231         err = kvm_vcpu_init(vcpu, kvm, id);
 232         if (err)
 233                 goto free_vcpu;
 234
 235         err = create_hyp_mappings(vcpu, vcpu + 1);
 236         if (err)
 237                 goto vcpu_uninit;
 238
 239         return vcpu;
 240 vcpu_uninit:
 241         kvm_vcpu_uninit(vcpu);
 242 free_vcpu:
 243         kmem_cache_free(kvm_vcpu_cache, vcpu);
 244 out:
 245         return ERR_PTR(err);
 246 }
 247
 248 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 249 {
 250         return 0;
 251 }
 252
 253 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 254 {
 255         kvm_mmu_free_memory_caches(vcpu);
 256         kvm_timer_vcpu_terminate(vcpu);
 257         kmem_cache_free(kvm_vcpu_cache, vcpu);
 258 }
 259
 260 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 261 {
 262         kvm_arch_vcpu_free(vcpu);
 263 }
 264
 265 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 266 {
 267         return 0;
 268 }
 269
 270 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 271 {
 272         int ret;
 273
 274         /* Force users to call KVM_ARM_VCPU_INIT */
 275         vcpu->arch.target = -1;
 276
 277         /* Set up VGIC */
 278         ret = kvm_vgic_vcpu_init(vcpu);
 279         if (ret)
 280                 return ret;
 281
 282         /* Set up the timer */
 283         kvm_timer_vcpu_init(vcpu);
 284
 285         return 0;
 286 }
 287
 288 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 289 {
 290 }
 291
 292 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
 293 {
 294 }
 295
 296 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 297 {
 298         vcpu->cpu = cpu;
 299         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 300
 301         /*
 302          * Check whether this vcpu requires the cache to be flushed on
 303          * this physical CPU. This is a consequence of doing dcache
 304          * operations by set/way on this vcpu. We do it here to be in
 305          * a non-preemptible section.
 306          */
 307         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 308                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 309
 310         kvm_arm_set_running_vcpu(vcpu);
 311 }
 312
 313 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 314 {
 315         /*
 316          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 317          * if the vcpu is no longer assigned to a cpu.  This is used for the
 318          * optimized make_all_cpus_request path.
 319          */
 320         vcpu->cpu = -1;
 321
 322         kvm_arm_set_running_vcpu(NULL);
 323 }
 324
 325 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 326                                         struct kvm_guest_debug *dbg)
 327 {
 328         return -EINVAL;
 329 }
 330
 331
 332 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 333                                     struct kvm_mp_state *mp_state)
 334 {
 335         return -EINVAL;
 336 }
 337
 338 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 339                                     struct kvm_mp_state *mp_state)
 340 {
 341         return -EINVAL;
 342 }
 343
 344 /**
 345  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 346  * @v:          The VCPU pointer
 347  *
 348  * If the guest CPU is not waiting for interrupts or an interrupt line is
 349  * asserted, the CPU is by definition runnable.
 350  */
 351 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 352 {
 353         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 354 }
 355
 356 /* Just ensure a guest exit from a particular CPU */
 357 static void exit_vm_noop(void *info)
 358 {
 359 }
 360
 361 void force_vm_exit(const cpumask_t *mask)
 362 {
 363         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 364 }
 365
 366 /**
 367  * need_new_vmid_gen - check that the VMID is still valid
 368  * @kvm: The VM's VMID to checkt
 369  *
 370  * return true if there is a new generation of VMIDs being used
 371  *
 372  * The hardware supports only 256 values with the value zero reserved for the
 373  * host, so we check if an assigned value belongs to a previous generation,
 374  * which which requires us to assign a new value. If we're the first to use a
 375  * VMID for the new generation, we must flush necessary caches and TLBs on all
 376  * CPUs.
 377  */
 378 static bool need_new_vmid_gen(struct kvm *kvm)
 379 {
 380         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 381 }
 382
 383 /**
 384  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 385  * @kvm The guest that we are about to run
 386  *
 387  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 388  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 389  * caches and TLBs.
 390  */
 391 static void update_vttbr(struct kvm *kvm)
 392 {
 393         phys_addr_t pgd_phys;
 394         u64 vmid;
 395
 396         if (!need_new_vmid_gen(kvm))
 397                 return;
 398
 399         spin_lock(&kvm_vmid_lock);
 400
 401         /*
 402          * We need to re-check the vmid_gen here to ensure that if another vcpu
 403          * already allocated a valid vmid for this vm, then this vcpu should
 404          * use the same vmid.
 405          */
 406         if (!need_new_vmid_gen(kvm)) {
 407                 spin_unlock(&kvm_vmid_lock);
 408                 return;
 409         }
 410
 411         /* First user of a new VMID generation? */
 412         if (unlikely(kvm_next_vmid == 0)) {
 413                 atomic64_inc(&kvm_vmid_gen);
 414                 kvm_next_vmid = 1;
 415
 416                 /*
 417                  * On SMP we know no other CPUs can use this CPU's or each
 418                  * other's VMID after force_vm_exit returns since the
 419                  * kvm_vmid_lock blocks them from reentry to the guest.
 420                  */
 421                 force_vm_exit(cpu_all_mask);
 422                 /*
 423                  * Now broadcast TLB + ICACHE invalidation over the inner
 424                  * shareable domain to make sure all data structures are
 425                  * clean.
 426                  */
 427                 kvm_call_hyp(__kvm_flush_vm_context);
 428         }
 429
 430         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 431         kvm->arch.vmid = kvm_next_vmid;
 432         kvm_next_vmid++;
 433
 434         /* update vttbr to be used with the new vmid */
 435         pgd_phys = virt_to_phys(kvm->arch.pgd);
 436         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 437         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 438         kvm->arch.vttbr |= vmid;
 439
 440         spin_unlock(&kvm_vmid_lock);
 441 }
 442
 443 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 444 {
 445         int ret;
 446
 447         if (likely(vcpu->arch.has_run_once))
 448                 return 0;
 449
 450         vcpu->arch.has_run_once = true;
 451
 452         /*
 453          * Initialize the VGIC before running a vcpu the first time on
 454          * this VM.
 455          */
 456         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 457                 ret = kvm_vgic_init(vcpu->kvm);
 458                 if (ret)
 459                         return ret;
 460         }
 461
 462         return 0;
 463 }
 464
 465 static void vcpu_pause(struct kvm_vcpu *vcpu)
 466 {
 467         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 468
 469         wait_event_interruptible(*wq, !vcpu->arch.pause);
 470 }
 471
 472 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 473 {
 474         return vcpu->arch.target >= 0;
 475 }
 476
 477 /**
 478  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 479  * @vcpu:       The VCPU pointer
 480  * @run:        The kvm_run structure pointer used for userspace state exchange
 481  *
 482  * This function is called through the VCPU_RUN ioctl called from user space. It
 483  * will execute VM code in a loop until the time slice for the process is used
 484  * or some emulation is needed from user space in which case the function will
 485  * return with return value 0 and with the kvm_run structure filled in with the
 486  * required data for the requested emulation.
 487  */
 488 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 489 {
 490         int ret;
 491         sigset_t sigsaved;
 492
 493         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 494                 return -ENOEXEC;
 495
 496         ret = kvm_vcpu_first_run_init(vcpu);
 497         if (ret)
 498                 return ret;
 499
 500         if (run->exit_reason == KVM_EXIT_MMIO) {
 501                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 502                 if (ret)
 503                         return ret;
 504         }
 505
 506         if (vcpu->sigset_active)
 507                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 508
 509         ret = 1;
 510         run->exit_reason = KVM_EXIT_UNKNOWN;
 511         while (ret > 0) {
 512                 /*
 513                  * Check conditions before entering the guest
 514                  */
 515                 cond_resched();
 516
 517                 update_vttbr(vcpu->kvm);
 518
 519                 if (vcpu->arch.pause)
 520                         vcpu_pause(vcpu);
 521
 522                 kvm_vgic_flush_hwstate(vcpu);
 523                 kvm_timer_flush_hwstate(vcpu);
 524
 525                 local_irq_disable();
 526
 527                 /*
 528                  * Re-check atomic conditions
 529                  */
 530                 if (signal_pending(current)) {
 531                         ret = -EINTR;
 532                         run->exit_reason = KVM_EXIT_INTR;
 533                 }
 534
 535                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 536                         local_irq_enable();
 537                         kvm_timer_sync_hwstate(vcpu);
 538                         kvm_vgic_sync_hwstate(vcpu);
 539                         continue;
 540                 }
 541
 542                 /**************************************************************
 543                  * Enter the guest
 544                  */
 545                 trace_kvm_entry(*vcpu_pc(vcpu));
 546                 kvm_guest_enter();
 547                 vcpu->mode = IN_GUEST_MODE;
 548
 549                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 550
 551                 vcpu->mode = OUTSIDE_GUEST_MODE;
 552                 vcpu->arch.last_pcpu = smp_processor_id();
 553                 kvm_guest_exit();
 554                 trace_kvm_exit(*vcpu_pc(vcpu));
 555                 /*
 556                  * We may have taken a host interrupt in HYP mode (ie
 557                  * while executing the guest). This interrupt is still
 558                  * pending, as we haven't serviced it yet!
 559                  *
 560                  * We're now back in SVC mode, with interrupts
 561                  * disabled.  Enabling the interrupts now will have
 562                  * the effect of taking the interrupt again, in SVC
 563                  * mode this time.
 564                  */
 565                 local_irq_enable();
 566
 567                 /*
 568                  * Back from guest
 569                  *************************************************************/
 570
 571                 kvm_timer_sync_hwstate(vcpu);
 572                 kvm_vgic_sync_hwstate(vcpu);
 573
 574                 ret = handle_exit(vcpu, run, ret);
 575         }
 576
 577         if (vcpu->sigset_active)
 578                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 579         return ret;
 580 }
 581
 582 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 583 {
 584         int bit_index;
 585         bool set;
 586         unsigned long *ptr;
 587
 588         if (number == KVM_ARM_IRQ_CPU_IRQ)
 589                 bit_index = __ffs(HCR_VI);
 590         else /* KVM_ARM_IRQ_CPU_FIQ */
 591                 bit_index = __ffs(HCR_VF);
 592
 593         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 594         if (level)
 595                 set = test_and_set_bit(bit_index, ptr);
 596         else
 597                 set = test_and_clear_bit(bit_index, ptr);
 598
 599         /*
 600          * If we didn't change anything, no need to wake up or kick other CPUs
 601          */
 602         if (set == level)
 603                 return 0;
 604
 605         /*
 606          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 607          * trigger a world-switch round on the running physical CPU to set the
 608          * virtual IRQ/FIQ fields in the HCR appropriately.
 609          */
 610         kvm_vcpu_kick(vcpu);
 611
 612         return 0;
 613 }
 614
 615 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 616                           bool line_status)
 617 {
 618         u32 irq = irq_level->irq;
 619         unsigned int irq_type, vcpu_idx, irq_num;
 620         int nrcpus = atomic_read(&kvm->online_vcpus);
 621         struct kvm_vcpu *vcpu = NULL;
 622         bool level = irq_level->level;
 623
 624         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 625         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 626         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 627
 628         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 629
 630         switch (irq_type) {
 631         case KVM_ARM_IRQ_TYPE_CPU:
 632                 if (irqchip_in_kernel(kvm))
 633                         return -ENXIO;
 634
 635                 if (vcpu_idx >= nrcpus)
 636                         return -EINVAL;
 637
 638                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 639                 if (!vcpu)
 640                         return -EINVAL;
 641
 642                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 643                         return -EINVAL;
 644
 645                 return vcpu_interrupt_line(vcpu, irq_num, level);
 646         case KVM_ARM_IRQ_TYPE_PPI:
 647                 if (!irqchip_in_kernel(kvm))
 648                         return -ENXIO;
 649
 650                 if (vcpu_idx >= nrcpus)
 651                         return -EINVAL;
 652
 653                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 654                 if (!vcpu)
 655                         return -EINVAL;
 656
 657                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 658                         return -EINVAL;
 659
 660                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 661         case KVM_ARM_IRQ_TYPE_SPI:
 662                 if (!irqchip_in_kernel(kvm))
 663                         return -ENXIO;
 664
 665                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 666                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 667                         return -EINVAL;
 668
 669                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 670         }
 671
 672         return -EINVAL;
 673 }
 674
 675 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 676                                          struct kvm_vcpu_init *init)
 677 {
 678         int ret;
 679
 680         ret = kvm_vcpu_set_target(vcpu, init);
 681         if (ret)
 682                 return ret;
 683
 684         /*
 685          * Handle the "start in power-off" case by marking the VCPU as paused.
 686          */
 687         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 688                 vcpu->arch.pause = true;
 689
 690         return 0;
 691 }
 692
 693 long kvm_arch_vcpu_ioctl(struct file *filp,
 694                          unsigned int ioctl, unsigned long arg)
 695 {
 696         struct kvm_vcpu *vcpu = filp->private_data;
 697         void __user *argp = (void __user *)arg;
 698
 699         switch (ioctl) {
 700         case KVM_ARM_VCPU_INIT: {
 701                 struct kvm_vcpu_init init;
 702
 703                 if (copy_from_user(&init, argp, sizeof(init)))
 704                         return -EFAULT;
 705
 706                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 707         }
 708         case KVM_SET_ONE_REG:
 709         case KVM_GET_ONE_REG: {
 710                 struct kvm_one_reg reg;
 711
 712                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 713                         return -ENOEXEC;
 714
 715                 if (copy_from_user(&reg, argp, sizeof(reg)))
 716                         return -EFAULT;
 717                 if (ioctl == KVM_SET_ONE_REG)
 718                         return kvm_arm_set_reg(vcpu, &reg);
 719                 else
 720                         return kvm_arm_get_reg(vcpu, &reg);
 721         }
 722         case KVM_GET_REG_LIST: {
 723                 struct kvm_reg_list __user *user_list = argp;
 724                 struct kvm_reg_list reg_list;
 725                 unsigned n;
 726
 727                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 728                         return -ENOEXEC;
 729
 730                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 731                         return -EFAULT;
 732                 n = reg_list.n;
 733                 reg_list.n = kvm_arm_num_regs(vcpu);
 734                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 735                         return -EFAULT;
 736                 if (n < reg_list.n)
 737                         return -E2BIG;
 738                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 739         }
 740         default:
 741                 return -EINVAL;
 742         }
 743 }
 744
 745 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 746 {
 747         return -EINVAL;
 748 }
 749
 750 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 751                                         struct kvm_arm_device_addr *dev_addr)
 752 {
 753         unsigned long dev_id, type;
 754
 755         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 756                 KVM_ARM_DEVICE_ID_SHIFT;
 757         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 758                 KVM_ARM_DEVICE_TYPE_SHIFT;
 759
 760         switch (dev_id) {
 761         case KVM_ARM_DEVICE_VGIC_V2:
 762                 if (!vgic_present)
 763                         return -ENXIO;
 764                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 765         default:
 766                 return -ENODEV;
 767         }
 768 }
 769
 770 long kvm_arch_vm_ioctl(struct file *filp,
 771                        unsigned int ioctl, unsigned long arg)
 772 {
 773         struct kvm *kvm = filp->private_data;
 774         void __user *argp = (void __user *)arg;
 775
 776         switch (ioctl) {
 777         case KVM_CREATE_IRQCHIP: {
 778                 if (vgic_present)
 779                         return kvm_vgic_create(kvm);
 780                 else
 781                         return -ENXIO;
 782         }
 783         case KVM_ARM_SET_DEVICE_ADDR: {
 784                 struct kvm_arm_device_addr dev_addr;
 785
 786                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 787                         return -EFAULT;
 788                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 789         }
 790         case KVM_ARM_PREFERRED_TARGET: {
 791                 int err;
 792                 struct kvm_vcpu_init init;
 793
 794                 err = kvm_vcpu_preferred_target(&init);
 795                 if (err)
 796                         return err;
 797
 798                 if (copy_to_user(argp, &init, sizeof(init)))
 799                         return -EFAULT;
 800
 801                 return 0;
 802         }
 803         default:
 804                 return -EINVAL;
 805         }
 806 }
 807
 808 static void cpu_init_hyp_mode(void *dummy)
 809 {
 810         phys_addr_t boot_pgd_ptr;
 811         phys_addr_t pgd_ptr;
 812         unsigned long hyp_stack_ptr;
 813         unsigned long stack_page;
 814         unsigned long vector_ptr;
 815
 816         /* Switch from the HYP stub to our own HYP init vector */
 817         __hyp_set_vectors(kvm_get_idmap_vector());
 818
 819         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 820         pgd_ptr = kvm_mmu_get_httbr();
 821         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 822         hyp_stack_ptr = stack_page + PAGE_SIZE;
 823         vector_ptr = (unsigned long)__kvm_hyp_vector;
 824
 825         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 826 }
 827
 828 static int hyp_init_cpu_notify(struct notifier_block *self,
 829                                unsigned long action, void *cpu)
 830 {
 831         switch (action) {
 832         case CPU_STARTING:
 833         case CPU_STARTING_FROZEN:
 834                 cpu_init_hyp_mode(NULL);
 835                 break;
 836         }
 837
 838         return NOTIFY_OK;
 839 }
 840
 841 static struct notifier_block hyp_init_cpu_nb = {
 842         .notifier_call = hyp_init_cpu_notify,
 843 };
 844
 845 #ifdef CONFIG_CPU_PM
 846 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 847                                     unsigned long cmd,
 848                                     void *v)
 849 {
 850         if (cmd == CPU_PM_EXIT &&
 851             __hyp_get_vectors() == hyp_default_vectors) {
 852                 cpu_init_hyp_mode(NULL);
 853                 return NOTIFY_OK;
 854         }
 855
 856         return NOTIFY_DONE;
 857 }
 858
 859 static struct notifier_block hyp_init_cpu_pm_nb = {
 860         .notifier_call = hyp_init_cpu_pm_notifier,
 861 };
 862
 863 static void __init hyp_cpu_pm_init(void)
 864 {
 865         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 866 }
 867 #else
 868 static inline void hyp_cpu_pm_init(void)
 869 {
 870 }
 871 #endif
 872
 873 /**
 874  * Inits Hyp-mode on all online CPUs
 875  */
 876 static int init_hyp_mode(void)
 877 {
 878         int cpu;
 879         int err = 0;
 880
 881         /*
 882          * Allocate Hyp PGD and setup Hyp identity mapping
 883          */
 884         err = kvm_mmu_init();
 885         if (err)
 886                 goto out_err;
 887
 888         /*
 889          * It is probably enough to obtain the default on one
 890          * CPU. It's unlikely to be different on the others.
 891          */
 892         hyp_default_vectors = __hyp_get_vectors();
 893
 894         /*
 895          * Allocate stack pages for Hypervisor-mode
 896          */
 897         for_each_possible_cpu(cpu) {
 898                 unsigned long stack_page;
 899
 900                 stack_page = __get_free_page(GFP_KERNEL);
 901                 if (!stack_page) {
 902                         err = -ENOMEM;
 903                         goto out_free_stack_pages;
 904                 }
 905
 906                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 907         }
 908
 909         /*
 910          * Map the Hyp-code called directly from the host
 911          */
 912         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 913         if (err) {
 914                 kvm_err("Cannot map world-switch code\n");
 915                 goto out_free_mappings;
 916         }
 917
 918         /*
 919          * Map the Hyp stack pages
 920          */
 921         for_each_possible_cpu(cpu) {
 922                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 923                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 924
 925                 if (err) {
 926                         kvm_err("Cannot map hyp stack\n");
 927                         goto out_free_mappings;
 928                 }
 929         }
 930
 931         /*
 932          * Map the host CPU structures
 933          */
 934         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 935         if (!kvm_host_cpu_state) {
 936                 err = -ENOMEM;
 937                 kvm_err("Cannot allocate host CPU state\n");
 938                 goto out_free_mappings;
 939         }
 940
 941         for_each_possible_cpu(cpu) {
 942                 kvm_cpu_context_t *cpu_ctxt;
 943
 944                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 945                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 946
 947                 if (err) {
 948                         kvm_err("Cannot map host CPU state: %d\n", err);
 949                         goto out_free_context;
 950                 }
 951         }
 952
 953         /*
 954          * Execute the init code on each CPU.
 955          */
 956         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 957
 958         /*
 959          * Init HYP view of VGIC
 960          */
 961         err = kvm_vgic_hyp_init();
 962         if (err)
 963                 goto out_free_context;
 964
 965 #ifdef CONFIG_KVM_ARM_VGIC
 966                 vgic_present = true;
 967 #endif
 968
 969         /*
 970          * Init HYP architected timer support
 971          */
 972         err = kvm_timer_hyp_init();
 973         if (err)
 974                 goto out_free_mappings;
 975
 976 #ifndef CONFIG_HOTPLUG_CPU
 977         free_boot_hyp_pgd();
 978 #endif
 979
 980         kvm_perf_init();
 981
 982         kvm_info("Hyp mode initialized successfully\n");
 983
 984         return 0;
 985 out_free_context:
 986         free_percpu(kvm_host_cpu_state);
 987 out_free_mappings:
 988         free_hyp_pgds();
 989 out_free_stack_pages:
 990         for_each_possible_cpu(cpu)
 991                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 992 out_err:
 993         kvm_err("error initializing Hyp mode: %d\n", err);
 994         return err;
 995 }
 996
 997 static void check_kvm_target_cpu(void *ret)
 998 {
 999         *(int *)ret = kvm_target_cpu();
1000 }
1001
1002 /**
1003  * Initialize Hyp-mode and memory mappings on all CPUs.
1004  */
1005 int kvm_arch_init(void *opaque)
1006 {
1007         int err;
1008         int ret, cpu;
1009
1010         if (!is_hyp_mode_available()) {
1011                 kvm_err("HYP mode not available\n");
1012                 return -ENODEV;
1013         }
1014
1015         for_each_online_cpu(cpu) {
1016                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1017                 if (ret < 0) {
1018                         kvm_err("Error, CPU %d not supported!\n", cpu);
1019                         return -ENODEV;
1020                 }
1021         }
1022
1023         cpu_notifier_register_begin();
1024
1025         err = init_hyp_mode();
1026         if (err)
1027                 goto out_err;
1028
1029         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1030         if (err) {
1031                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1032                 goto out_err;
1033         }
1034
1035         cpu_notifier_register_done();
1036
1037         hyp_cpu_pm_init();
1038
1039         kvm_coproc_table_init();
1040         return 0;
1041 out_err:
1042         cpu_notifier_register_done();
1043         return err;
1044 }
1045
1046 /* NOP: Compiling as a module not supported */
1047 void kvm_arch_exit(void)
1048 {
1049         kvm_perf_teardown();
1050 }
1051
1052 static int arm_init(void)
1053 {
1054         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1055         return rc;
1056 }
1057
1058 module_init(arm_init);