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         /* 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 *kvm, struct kvm_memory_slot *free,
 157                            struct kvm_memory_slot *dont)
 158 {
 159 }
 160
 161 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 162                             unsigned long npages)
 163 {
 164         return 0;
 165 }
 166
 167 /**
 168  * kvm_arch_destroy_vm - destroy the VM data structure
 169  * @kvm:        pointer to the KVM struct
 170  */
 171 void kvm_arch_destroy_vm(struct kvm *kvm)
 172 {
 173         int i;
 174
 175         kvm_free_stage2_pgd(kvm);
 176
 177         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 178                 if (kvm->vcpus[i]) {
 179                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 180                         kvm->vcpus[i] = NULL;
 181                 }
 182         }
 183 }
 184
 185 int kvm_dev_ioctl_check_extension(long ext)
 186 {
 187         int r;
 188         switch (ext) {
 189         case KVM_CAP_IRQCHIP:
 190                 r = vgic_present;
 191                 break;
 192         case KVM_CAP_USER_MEMORY:
 193         case KVM_CAP_SYNC_MMU:
 194         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 195         case KVM_CAP_ONE_REG:
 196         case KVM_CAP_ARM_PSCI:
 197                 r = 1;
 198                 break;
 199         case KVM_CAP_COALESCED_MMIO:
 200                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 201                 break;
 202         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 203                 r = 1;
 204                 break;
 205         case KVM_CAP_NR_VCPUS:
 206                 r = num_online_cpus();
 207                 break;
 208         case KVM_CAP_MAX_VCPUS:
 209                 r = KVM_MAX_VCPUS;
 210                 break;
 211         default:
 212                 r = kvm_arch_dev_ioctl_check_extension(ext);
 213                 break;
 214         }
 215         return r;
 216 }
 217
 218 long kvm_arch_dev_ioctl(struct file *filp,
 219                         unsigned int ioctl, unsigned long arg)
 220 {
 221         return -EINVAL;
 222 }
 223
 224 void kvm_arch_memslots_updated(struct kvm *kvm)
 225 {
 226 }
 227
 228 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 229                                    struct kvm_memory_slot *memslot,
 230                                    struct kvm_userspace_memory_region *mem,
 231                                    enum kvm_mr_change change)
 232 {
 233         return 0;
 234 }
 235
 236 void kvm_arch_commit_memory_region(struct kvm *kvm,
 237                                    struct kvm_userspace_memory_region *mem,
 238                                    const struct kvm_memory_slot *old,
 239                                    enum kvm_mr_change change)
 240 {
 241 }
 242
 243 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 244 {
 245 }
 246
 247 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 248                                    struct kvm_memory_slot *slot)
 249 {
 250 }
 251
 252 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 253 {
 254         int err;
 255         struct kvm_vcpu *vcpu;
 256
 257         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 258         if (!vcpu) {
 259                 err = -ENOMEM;
 260                 goto out;
 261         }
 262
 263         err = kvm_vcpu_init(vcpu, kvm, id);
 264         if (err)
 265                 goto free_vcpu;
 266
 267         err = create_hyp_mappings(vcpu, vcpu + 1);
 268         if (err)
 269                 goto vcpu_uninit;
 270
 271         return vcpu;
 272 vcpu_uninit:
 273         kvm_vcpu_uninit(vcpu);
 274 free_vcpu:
 275         kmem_cache_free(kvm_vcpu_cache, vcpu);
 276 out:
 277         return ERR_PTR(err);
 278 }
 279
 280 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 281 {
 282         return 0;
 283 }
 284
 285 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 286 {
 287         kvm_mmu_free_memory_caches(vcpu);
 288         kvm_timer_vcpu_terminate(vcpu);
 289         kmem_cache_free(kvm_vcpu_cache, vcpu);
 290 }
 291
 292 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 293 {
 294         kvm_arch_vcpu_free(vcpu);
 295 }
 296
 297 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 298 {
 299         return 0;
 300 }
 301
 302 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 303 {
 304         int ret;
 305
 306         /* Force users to call KVM_ARM_VCPU_INIT */
 307         vcpu->arch.target = -1;
 308
 309         /* Set up VGIC */
 310         ret = kvm_vgic_vcpu_init(vcpu);
 311         if (ret)
 312                 return ret;
 313
 314         /* Set up the timer */
 315         kvm_timer_vcpu_init(vcpu);
 316
 317         return 0;
 318 }
 319
 320 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 321 {
 322 }
 323
 324 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 325 {
 326         vcpu->cpu = cpu;
 327         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 328
 329         /*
 330          * Check whether this vcpu requires the cache to be flushed on
 331          * this physical CPU. This is a consequence of doing dcache
 332          * operations by set/way on this vcpu. We do it here to be in
 333          * a non-preemptible section.
 334          */
 335         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 336                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 337
 338         kvm_arm_set_running_vcpu(vcpu);
 339 }
 340
 341 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 342 {
 343         kvm_arm_set_running_vcpu(NULL);
 344 }
 345
 346 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 347                                         struct kvm_guest_debug *dbg)
 348 {
 349         return -EINVAL;
 350 }
 351
 352
 353 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 354                                     struct kvm_mp_state *mp_state)
 355 {
 356         return -EINVAL;
 357 }
 358
 359 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 360                                     struct kvm_mp_state *mp_state)
 361 {
 362         return -EINVAL;
 363 }
 364
 365 /**
 366  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 367  * @v:          The VCPU pointer
 368  *
 369  * If the guest CPU is not waiting for interrupts or an interrupt line is
 370  * asserted, the CPU is by definition runnable.
 371  */
 372 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 373 {
 374         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 375 }
 376
 377 /* Just ensure a guest exit from a particular CPU */
 378 static void exit_vm_noop(void *info)
 379 {
 380 }
 381
 382 void force_vm_exit(const cpumask_t *mask)
 383 {
 384         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 385 }
 386
 387 /**
 388  * need_new_vmid_gen - check that the VMID is still valid
 389  * @kvm: The VM's VMID to checkt
 390  *
 391  * return true if there is a new generation of VMIDs being used
 392  *
 393  * The hardware supports only 256 values with the value zero reserved for the
 394  * host, so we check if an assigned value belongs to a previous generation,
 395  * which which requires us to assign a new value. If we're the first to use a
 396  * VMID for the new generation, we must flush necessary caches and TLBs on all
 397  * CPUs.
 398  */
 399 static bool need_new_vmid_gen(struct kvm *kvm)
 400 {
 401         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 402 }
 403
 404 /**
 405  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 406  * @kvm The guest that we are about to run
 407  *
 408  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 409  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 410  * caches and TLBs.
 411  */
 412 static void update_vttbr(struct kvm *kvm)
 413 {
 414         phys_addr_t pgd_phys;
 415         u64 vmid;
 416
 417         if (!need_new_vmid_gen(kvm))
 418                 return;
 419
 420         spin_lock(&kvm_vmid_lock);
 421
 422         /*
 423          * We need to re-check the vmid_gen here to ensure that if another vcpu
 424          * already allocated a valid vmid for this vm, then this vcpu should
 425          * use the same vmid.
 426          */
 427         if (!need_new_vmid_gen(kvm)) {
 428                 spin_unlock(&kvm_vmid_lock);
 429                 return;
 430         }
 431
 432         /* First user of a new VMID generation? */
 433         if (unlikely(kvm_next_vmid == 0)) {
 434                 atomic64_inc(&kvm_vmid_gen);
 435                 kvm_next_vmid = 1;
 436
 437                 /*
 438                  * On SMP we know no other CPUs can use this CPU's or each
 439                  * other's VMID after force_vm_exit returns since the
 440                  * kvm_vmid_lock blocks them from reentry to the guest.
 441                  */
 442                 force_vm_exit(cpu_all_mask);
 443                 /*
 444                  * Now broadcast TLB + ICACHE invalidation over the inner
 445                  * shareable domain to make sure all data structures are
 446                  * clean.
 447                  */
 448                 kvm_call_hyp(__kvm_flush_vm_context);
 449         }
 450
 451         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 452         kvm->arch.vmid = kvm_next_vmid;
 453         kvm_next_vmid++;
 454
 455         /* update vttbr to be used with the new vmid */
 456         pgd_phys = virt_to_phys(kvm->arch.pgd);
 457         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 458         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 459         kvm->arch.vttbr |= vmid;
 460
 461         spin_unlock(&kvm_vmid_lock);
 462 }
 463
 464 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 465 {
 466         if (likely(vcpu->arch.has_run_once))
 467                 return 0;
 468
 469         vcpu->arch.has_run_once = true;
 470
 471         /*
 472          * Initialize the VGIC before running a vcpu the first time on
 473          * this VM.
 474          */
 475         if (irqchip_in_kernel(vcpu->kvm) &&
 476             unlikely(!vgic_initialized(vcpu->kvm))) {
 477                 int ret = kvm_vgic_init(vcpu->kvm);
 478                 if (ret)
 479                         return ret;
 480         }
 481
 482         /*
 483          * Handle the "start in power-off" case by calling into the
 484          * PSCI code.
 485          */
 486         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
 487                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
 488                 kvm_psci_call(vcpu);
 489         }
 490
 491         return 0;
 492 }
 493
 494 static void vcpu_pause(struct kvm_vcpu *vcpu)
 495 {
 496         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 497
 498         wait_event_interruptible(*wq, !vcpu->arch.pause);
 499 }
 500
 501 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 502 {
 503         return vcpu->arch.target >= 0;
 504 }
 505
 506 /**
 507  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 508  * @vcpu:       The VCPU pointer
 509  * @run:        The kvm_run structure pointer used for userspace state exchange
 510  *
 511  * This function is called through the VCPU_RUN ioctl called from user space. It
 512  * will execute VM code in a loop until the time slice for the process is used
 513  * or some emulation is needed from user space in which case the function will
 514  * return with return value 0 and with the kvm_run structure filled in with the
 515  * required data for the requested emulation.
 516  */
 517 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 518 {
 519         int ret;
 520         sigset_t sigsaved;
 521
 522         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 523                 return -ENOEXEC;
 524
 525         ret = kvm_vcpu_first_run_init(vcpu);
 526         if (ret)
 527                 return ret;
 528
 529         if (run->exit_reason == KVM_EXIT_MMIO) {
 530                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 531                 if (ret)
 532                         return ret;
 533         }
 534
 535         if (vcpu->sigset_active)
 536                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 537
 538         ret = 1;
 539         run->exit_reason = KVM_EXIT_UNKNOWN;
 540         while (ret > 0) {
 541                 /*
 542                  * Check conditions before entering the guest
 543                  */
 544                 cond_resched();
 545
 546                 update_vttbr(vcpu->kvm);
 547
 548                 if (vcpu->arch.pause)
 549                         vcpu_pause(vcpu);
 550
 551                 kvm_vgic_flush_hwstate(vcpu);
 552                 kvm_timer_flush_hwstate(vcpu);
 553
 554                 local_irq_disable();
 555
 556                 /*
 557                  * Re-check atomic conditions
 558                  */
 559                 if (signal_pending(current)) {
 560                         ret = -EINTR;
 561                         run->exit_reason = KVM_EXIT_INTR;
 562                 }
 563
 564                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 565                         local_irq_enable();
 566                         kvm_timer_sync_hwstate(vcpu);
 567                         kvm_vgic_sync_hwstate(vcpu);
 568                         continue;
 569                 }
 570
 571                 /**************************************************************
 572                  * Enter the guest
 573                  */
 574                 trace_kvm_entry(*vcpu_pc(vcpu));
 575                 kvm_guest_enter();
 576                 vcpu->mode = IN_GUEST_MODE;
 577
 578                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 579
 580                 vcpu->mode = OUTSIDE_GUEST_MODE;
 581                 vcpu->arch.last_pcpu = smp_processor_id();
 582                 kvm_guest_exit();
 583                 trace_kvm_exit(*vcpu_pc(vcpu));
 584                 /*
 585                  * We may have taken a host interrupt in HYP mode (ie
 586                  * while executing the guest). This interrupt is still
 587                  * pending, as we haven't serviced it yet!
 588                  *
 589                  * We're now back in SVC mode, with interrupts
 590                  * disabled.  Enabling the interrupts now will have
 591                  * the effect of taking the interrupt again, in SVC
 592                  * mode this time.
 593                  */
 594                 local_irq_enable();
 595
 596                 /*
 597                  * Back from guest
 598                  *************************************************************/
 599
 600                 kvm_timer_sync_hwstate(vcpu);
 601                 kvm_vgic_sync_hwstate(vcpu);
 602
 603                 ret = handle_exit(vcpu, run, ret);
 604         }
 605
 606         if (vcpu->sigset_active)
 607                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 608         return ret;
 609 }
 610
 611 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 612 {
 613         int bit_index;
 614         bool set;
 615         unsigned long *ptr;
 616
 617         if (number == KVM_ARM_IRQ_CPU_IRQ)
 618                 bit_index = __ffs(HCR_VI);
 619         else /* KVM_ARM_IRQ_CPU_FIQ */
 620                 bit_index = __ffs(HCR_VF);
 621
 622         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 623         if (level)
 624                 set = test_and_set_bit(bit_index, ptr);
 625         else
 626                 set = test_and_clear_bit(bit_index, ptr);
 627
 628         /*
 629          * If we didn't change anything, no need to wake up or kick other CPUs
 630          */
 631         if (set == level)
 632                 return 0;
 633
 634         /*
 635          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 636          * trigger a world-switch round on the running physical CPU to set the
 637          * virtual IRQ/FIQ fields in the HCR appropriately.
 638          */
 639         kvm_vcpu_kick(vcpu);
 640
 641         return 0;
 642 }
 643
 644 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 645                           bool line_status)
 646 {
 647         u32 irq = irq_level->irq;
 648         unsigned int irq_type, vcpu_idx, irq_num;
 649         int nrcpus = atomic_read(&kvm->online_vcpus);
 650         struct kvm_vcpu *vcpu = NULL;
 651         bool level = irq_level->level;
 652
 653         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 654         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 655         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 656
 657         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 658
 659         switch (irq_type) {
 660         case KVM_ARM_IRQ_TYPE_CPU:
 661                 if (irqchip_in_kernel(kvm))
 662                         return -ENXIO;
 663
 664                 if (vcpu_idx >= nrcpus)
 665                         return -EINVAL;
 666
 667                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 668                 if (!vcpu)
 669                         return -EINVAL;
 670
 671                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 672                         return -EINVAL;
 673
 674                 return vcpu_interrupt_line(vcpu, irq_num, level);
 675         case KVM_ARM_IRQ_TYPE_PPI:
 676                 if (!irqchip_in_kernel(kvm))
 677                         return -ENXIO;
 678
 679                 if (vcpu_idx >= nrcpus)
 680                         return -EINVAL;
 681
 682                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 683                 if (!vcpu)
 684                         return -EINVAL;
 685
 686                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 687                         return -EINVAL;
 688
 689                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 690         case KVM_ARM_IRQ_TYPE_SPI:
 691                 if (!irqchip_in_kernel(kvm))
 692                         return -ENXIO;
 693
 694                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 695                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 696                         return -EINVAL;
 697
 698                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 699         }
 700
 701         return -EINVAL;
 702 }
 703
 704 long kvm_arch_vcpu_ioctl(struct file *filp,
 705                          unsigned int ioctl, unsigned long arg)
 706 {
 707         struct kvm_vcpu *vcpu = filp->private_data;
 708         void __user *argp = (void __user *)arg;
 709
 710         switch (ioctl) {
 711         case KVM_ARM_VCPU_INIT: {
 712                 struct kvm_vcpu_init init;
 713
 714                 if (copy_from_user(&init, argp, sizeof(init)))
 715                         return -EFAULT;
 716
 717                 return kvm_vcpu_set_target(vcpu, &init);
 718
 719         }
 720         case KVM_SET_ONE_REG:
 721         case KVM_GET_ONE_REG: {
 722                 struct kvm_one_reg reg;
 723
 724                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 725                         return -ENOEXEC;
 726
 727                 if (copy_from_user(&reg, argp, sizeof(reg)))
 728                         return -EFAULT;
 729                 if (ioctl == KVM_SET_ONE_REG)
 730                         return kvm_arm_set_reg(vcpu, &reg);
 731                 else
 732                         return kvm_arm_get_reg(vcpu, &reg);
 733         }
 734         case KVM_GET_REG_LIST: {
 735                 struct kvm_reg_list __user *user_list = argp;
 736                 struct kvm_reg_list reg_list;
 737                 unsigned n;
 738
 739                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 740                         return -ENOEXEC;
 741
 742                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 743                         return -EFAULT;
 744                 n = reg_list.n;
 745                 reg_list.n = kvm_arm_num_regs(vcpu);
 746                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 747                         return -EFAULT;
 748                 if (n < reg_list.n)
 749                         return -E2BIG;
 750                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 751         }
 752         default:
 753                 return -EINVAL;
 754         }
 755 }
 756
 757 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 758 {
 759         return -EINVAL;
 760 }
 761
 762 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 763                                         struct kvm_arm_device_addr *dev_addr)
 764 {
 765         unsigned long dev_id, type;
 766
 767         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 768                 KVM_ARM_DEVICE_ID_SHIFT;
 769         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 770                 KVM_ARM_DEVICE_TYPE_SHIFT;
 771
 772         switch (dev_id) {
 773         case KVM_ARM_DEVICE_VGIC_V2:
 774                 if (!vgic_present)
 775                         return -ENXIO;
 776                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
 777         default:
 778                 return -ENODEV;
 779         }
 780 }
 781
 782 long kvm_arch_vm_ioctl(struct file *filp,
 783                        unsigned int ioctl, unsigned long arg)
 784 {
 785         struct kvm *kvm = filp->private_data;
 786         void __user *argp = (void __user *)arg;
 787
 788         switch (ioctl) {
 789         case KVM_CREATE_IRQCHIP: {
 790                 if (vgic_present)
 791                         return kvm_vgic_create(kvm);
 792                 else
 793                         return -ENXIO;
 794         }
 795         case KVM_ARM_SET_DEVICE_ADDR: {
 796                 struct kvm_arm_device_addr dev_addr;
 797
 798                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 799                         return -EFAULT;
 800                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 801         }
 802         case KVM_ARM_PREFERRED_TARGET: {
 803                 int err;
 804                 struct kvm_vcpu_init init;
 805
 806                 err = kvm_vcpu_preferred_target(&init);
 807                 if (err)
 808                         return err;
 809
 810                 if (copy_to_user(argp, &init, sizeof(init)))
 811                         return -EFAULT;
 812
 813                 return 0;
 814         }
 815         default:
 816                 return -EINVAL;
 817         }
 818 }
 819
 820 static void cpu_init_hyp_mode(void *dummy)
 821 {
 822         phys_addr_t boot_pgd_ptr;
 823         phys_addr_t pgd_ptr;
 824         unsigned long hyp_stack_ptr;
 825         unsigned long stack_page;
 826         unsigned long vector_ptr;
 827
 828         /* Switch from the HYP stub to our own HYP init vector */
 829         __hyp_set_vectors(kvm_get_idmap_vector());
 830
 831         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 832         pgd_ptr = kvm_mmu_get_httbr();
 833         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 834         hyp_stack_ptr = stack_page + PAGE_SIZE;
 835         vector_ptr = (unsigned long)__kvm_hyp_vector;
 836
 837         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 838 }
 839
 840 static int hyp_init_cpu_notify(struct notifier_block *self,
 841                                unsigned long action, void *cpu)
 842 {
 843         switch (action) {
 844         case CPU_STARTING:
 845         case CPU_STARTING_FROZEN:
 846                 cpu_init_hyp_mode(NULL);
 847                 break;
 848         }
 849
 850         return NOTIFY_OK;
 851 }
 852
 853 static struct notifier_block hyp_init_cpu_nb = {
 854         .notifier_call = hyp_init_cpu_notify,
 855 };
 856
 857 #ifdef CONFIG_CPU_PM
 858 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 859                                     unsigned long cmd,
 860                                     void *v)
 861 {
 862         if (cmd == CPU_PM_EXIT) {
 863                 cpu_init_hyp_mode(NULL);
 864                 return NOTIFY_OK;
 865         }
 866
 867         return NOTIFY_DONE;
 868 }
 869
 870 static struct notifier_block hyp_init_cpu_pm_nb = {
 871         .notifier_call = hyp_init_cpu_pm_notifier,
 872 };
 873
 874 static void __init hyp_cpu_pm_init(void)
 875 {
 876         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 877 }
 878 #else
 879 static inline void hyp_cpu_pm_init(void)
 880 {
 881 }
 882 #endif
 883
 884 /**
 885  * Inits Hyp-mode on all online CPUs
 886  */
 887 static int init_hyp_mode(void)
 888 {
 889         int cpu;
 890         int err = 0;
 891
 892         /*
 893          * Allocate Hyp PGD and setup Hyp identity mapping
 894          */
 895         err = kvm_mmu_init();
 896         if (err)
 897                 goto out_err;
 898
 899         /*
 900          * It is probably enough to obtain the default on one
 901          * CPU. It's unlikely to be different on the others.
 902          */
 903         hyp_default_vectors = __hyp_get_vectors();
 904
 905         /*
 906          * Allocate stack pages for Hypervisor-mode
 907          */
 908         for_each_possible_cpu(cpu) {
 909                 unsigned long stack_page;
 910
 911                 stack_page = __get_free_page(GFP_KERNEL);
 912                 if (!stack_page) {
 913                         err = -ENOMEM;
 914                         goto out_free_stack_pages;
 915                 }
 916
 917                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 918         }
 919
 920         /*
 921          * Map the Hyp-code called directly from the host
 922          */
 923         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 924         if (err) {
 925                 kvm_err("Cannot map world-switch code\n");
 926                 goto out_free_mappings;
 927         }
 928
 929         /*
 930          * Map the Hyp stack pages
 931          */
 932         for_each_possible_cpu(cpu) {
 933                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 934                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 935
 936                 if (err) {
 937                         kvm_err("Cannot map hyp stack\n");
 938                         goto out_free_mappings;
 939                 }
 940         }
 941
 942         /*
 943          * Map the host CPU structures
 944          */
 945         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 946         if (!kvm_host_cpu_state) {
 947                 err = -ENOMEM;
 948                 kvm_err("Cannot allocate host CPU state\n");
 949                 goto out_free_mappings;
 950         }
 951
 952         for_each_possible_cpu(cpu) {
 953                 kvm_cpu_context_t *cpu_ctxt;
 954
 955                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 956                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 957
 958                 if (err) {
 959                         kvm_err("Cannot map host CPU state: %d\n", err);
 960                         goto out_free_context;
 961                 }
 962         }
 963
 964         /*
 965          * Execute the init code on each CPU.
 966          */
 967         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 968
 969         /*
 970          * Init HYP view of VGIC
 971          */
 972         err = kvm_vgic_hyp_init();
 973         if (err)
 974                 goto out_free_context;
 975
 976 #ifdef CONFIG_KVM_ARM_VGIC
 977                 vgic_present = true;
 978 #endif
 979
 980         /*
 981          * Init HYP architected timer support
 982          */
 983         err = kvm_timer_hyp_init();
 984         if (err)
 985                 goto out_free_mappings;
 986
 987 #ifndef CONFIG_HOTPLUG_CPU
 988         free_boot_hyp_pgd();
 989 #endif
 990
 991         kvm_perf_init();
 992
 993         kvm_info("Hyp mode initialized successfully\n");
 994
 995         return 0;
 996 out_free_context:
 997         free_percpu(kvm_host_cpu_state);
 998 out_free_mappings:
 999         free_hyp_pgds();
1000 out_free_stack_pages:
1001         for_each_possible_cpu(cpu)
1002                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1003 out_err:
1004         kvm_err("error initializing Hyp mode: %d\n", err);
1005         return err;
1006 }
1007
1008 static void check_kvm_target_cpu(void *ret)
1009 {
1010         *(int *)ret = kvm_target_cpu();
1011 }
1012
1013 /**
1014  * Initialize Hyp-mode and memory mappings on all CPUs.
1015  */
1016 int kvm_arch_init(void *opaque)
1017 {
1018         int err;
1019         int ret, cpu;
1020
1021         if (!is_hyp_mode_available()) {
1022                 kvm_err("HYP mode not available\n");
1023                 return -ENODEV;
1024         }
1025
1026         for_each_online_cpu(cpu) {
1027                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1028                 if (ret < 0) {
1029                         kvm_err("Error, CPU %d not supported!\n", cpu);
1030                         return -ENODEV;
1031                 }
1032         }
1033
1034         err = init_hyp_mode();
1035         if (err)
1036                 goto out_err;
1037
1038         err = register_cpu_notifier(&hyp_init_cpu_nb);
1039         if (err) {
1040                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1041                 goto out_err;
1042         }
1043
1044         hyp_cpu_pm_init();
1045
1046         kvm_coproc_table_init();
1047         return 0;
1048 out_err:
1049         return err;
1050 }
1051
1052 /* NOP: Compiling as a module not supported */
1053 void kvm_arch_exit(void)
1054 {
1055         kvm_perf_teardown();
1056 }
1057
1058 static int arm_init(void)
1059 {
1060         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1061         return rc;
1062 }
1063
1064 module_init(arm_init);