arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/module.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/fs.h>
  26 #include <linux/mman.h>
  27 #include <linux/sched.h>
  28 #include <linux/kvm.h>
  29 #include <trace/events/kvm.h>
  30
  31 #define CREATE_TRACE_POINTS
  32 #include "trace.h"
  33
  34 #include <asm/uaccess.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/mman.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/cacheflush.h>
  39 #include <asm/virt.h>
  40 #include <asm/kvm_arm.h>
  41 #include <asm/kvm_asm.h>
  42 #include <asm/kvm_mmu.h>
  43 #include <asm/kvm_emulate.h>
  44 #include <asm/kvm_coproc.h>
  45 #include <asm/kvm_psci.h>
  46
  47 #ifdef REQUIRES_VIRT
  48 __asm__(".arch_extension        virt");
  49 #endif
  50
  51 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  52 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  53 static unsigned long hyp_default_vectors;
  54
  55 /* Per-CPU variable containing the currently running vcpu. */
  56 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  57
  58 /* The VMID used in the VTTBR */
  59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  60 static u8 kvm_next_vmid;
  61 static DEFINE_SPINLOCK(kvm_vmid_lock);
  62
  63 static bool vgic_present;
  64
  65 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  66 {
  67         BUG_ON(preemptible());
  68         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  69 }
  70
  71 /**
  72  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  73  * Must be called from non-preemptible context
  74  */
  75 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  76 {
  77         BUG_ON(preemptible());
  78         return __this_cpu_read(kvm_arm_running_vcpu);
  79 }
  80
  81 /**
  82  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  83  */
  84 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  85 {
  86         return &kvm_arm_running_vcpu;
  87 }
  88
  89 int kvm_arch_hardware_enable(void *garbage)
  90 {
  91         return 0;
  92 }
  93
  94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  95 {
  96         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  97 }
  98
  99 void kvm_arch_hardware_disable(void *garbage)
 100 {
 101 }
 102
 103 int kvm_arch_hardware_setup(void)
 104 {
 105         return 0;
 106 }
 107
 108 void kvm_arch_hardware_unsetup(void)
 109 {
 110 }
 111
 112 void kvm_arch_check_processor_compat(void *rtn)
 113 {
 114         *(int *)rtn = 0;
 115 }
 116
 117 void kvm_arch_sync_events(struct kvm *kvm)
 118 {
 119 }
 120
 121 /**
 122  * kvm_arch_init_vm - initializes a VM data structure
 123  * @kvm:        pointer to the KVM struct
 124  */
 125 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 126 {
 127         int ret = 0;
 128
 129         if (type)
 130                 return -EINVAL;
 131
 132         ret = kvm_alloc_stage2_pgd(kvm);
 133         if (ret)
 134                 goto out_fail_alloc;
 135
 136         ret = create_hyp_mappings(kvm, kvm + 1);
 137         if (ret)
 138                 goto out_free_stage2_pgd;
 139
 140         kvm_timer_init(kvm);
 141
 142         /* Mark the initial VMID generation invalid */
 143         kvm->arch.vmid_gen = 0;
 144
 145         return ret;
 146 out_free_stage2_pgd:
 147         kvm_free_stage2_pgd(kvm);
 148 out_fail_alloc:
 149         return ret;
 150 }
 151
 152 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 153 {
 154         return VM_FAULT_SIGBUS;
 155 }
 156
 157 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 158                            struct kvm_memory_slot *dont)
 159 {
 160 }
 161
 162 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 163                             unsigned long npages)
 164 {
 165         return 0;
 166 }
 167
 168 /**
 169  * kvm_arch_destroy_vm - destroy the VM data structure
 170  * @kvm:        pointer to the KVM struct
 171  */
 172 void kvm_arch_destroy_vm(struct kvm *kvm)
 173 {
 174         int i;
 175
 176         kvm_free_stage2_pgd(kvm);
 177
 178         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 179                 if (kvm->vcpus[i]) {
 180                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 181                         kvm->vcpus[i] = NULL;
 182                 }
 183         }
 184 }
 185
 186 int kvm_dev_ioctl_check_extension(long ext)
 187 {
 188         int r;
 189         switch (ext) {
 190         case KVM_CAP_IRQCHIP:
 191                 r = vgic_present;
 192                 break;
 193         case KVM_CAP_USER_MEMORY:
 194         case KVM_CAP_SYNC_MMU:
 195         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 196         case KVM_CAP_ONE_REG:
 197         case KVM_CAP_ARM_PSCI:
 198                 r = 1;
 199                 break;
 200         case KVM_CAP_COALESCED_MMIO:
 201                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 202                 break;
 203         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 204                 r = 1;
 205                 break;
 206         case KVM_CAP_NR_VCPUS:
 207                 r = num_online_cpus();
 208                 break;
 209         case KVM_CAP_MAX_VCPUS:
 210                 r = KVM_MAX_VCPUS;
 211                 break;
 212         default:
 213                 r = kvm_arch_dev_ioctl_check_extension(ext);
 214                 break;
 215         }
 216         return r;
 217 }
 218
 219 long kvm_arch_dev_ioctl(struct file *filp,
 220                         unsigned int ioctl, unsigned long arg)
 221 {
 222         return -EINVAL;
 223 }
 224
 225 void kvm_arch_memslots_updated(struct kvm *kvm)
 226 {
 227 }
 228
 229 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 230                                    struct kvm_memory_slot *memslot,
 231                                    struct kvm_userspace_memory_region *mem,
 232                                    enum kvm_mr_change change)
 233 {
 234         return 0;
 235 }
 236
 237 void kvm_arch_commit_memory_region(struct kvm *kvm,
 238                                    struct kvm_userspace_memory_region *mem,
 239                                    const struct kvm_memory_slot *old,
 240                                    enum kvm_mr_change change)
 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.host_cpu_context = this_cpu_ptr(kvm_host_cpu_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         int ret;
 468
 469         if (likely(vcpu->arch.has_run_once))
 470                 return 0;
 471
 472         vcpu->arch.has_run_once = true;
 473
 474         /*
 475          * Initialize the VGIC before running a vcpu the first time on
 476          * this VM.
 477          */
 478         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 479                 ret = kvm_vgic_init(vcpu->kvm);
 480                 if (ret)
 481                         return ret;
 482         }
 483
 484         /*
 485          * Handle the "start in power-off" case by calling into the
 486          * PSCI code.
 487          */
 488         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
 489                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
 490                 kvm_psci_call(vcpu);
 491         }
 492
 493         return 0;
 494 }
 495
 496 static void vcpu_pause(struct kvm_vcpu *vcpu)
 497 {
 498         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 499
 500         wait_event_interruptible(*wq, !vcpu->arch.pause);
 501 }
 502
 503 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 504 {
 505         return vcpu->arch.target >= 0;
 506 }
 507
 508 /**
 509  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 510  * @vcpu:       The VCPU pointer
 511  * @run:        The kvm_run structure pointer used for userspace state exchange
 512  *
 513  * This function is called through the VCPU_RUN ioctl called from user space. It
 514  * will execute VM code in a loop until the time slice for the process is used
 515  * or some emulation is needed from user space in which case the function will
 516  * return with return value 0 and with the kvm_run structure filled in with the
 517  * required data for the requested emulation.
 518  */
 519 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 520 {
 521         int ret;
 522         sigset_t sigsaved;
 523
 524         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 525                 return -ENOEXEC;
 526
 527         ret = kvm_vcpu_first_run_init(vcpu);
 528         if (ret)
 529                 return ret;
 530
 531         if (run->exit_reason == KVM_EXIT_MMIO) {
 532                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 533                 if (ret)
 534                         return ret;
 535         }
 536
 537         if (vcpu->sigset_active)
 538                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 539
 540         ret = 1;
 541         run->exit_reason = KVM_EXIT_UNKNOWN;
 542         while (ret > 0) {
 543                 /*
 544                  * Check conditions before entering the guest
 545                  */
 546                 cond_resched();
 547
 548                 update_vttbr(vcpu->kvm);
 549
 550                 if (vcpu->arch.pause)
 551                         vcpu_pause(vcpu);
 552
 553                 kvm_vgic_flush_hwstate(vcpu);
 554                 kvm_timer_flush_hwstate(vcpu);
 555
 556                 local_irq_disable();
 557
 558                 /*
 559                  * Re-check atomic conditions
 560                  */
 561                 if (signal_pending(current)) {
 562                         ret = -EINTR;
 563                         run->exit_reason = KVM_EXIT_INTR;
 564                 }
 565
 566                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 567                         local_irq_enable();
 568                         kvm_timer_sync_hwstate(vcpu);
 569                         kvm_vgic_sync_hwstate(vcpu);
 570                         continue;
 571                 }
 572
 573                 /**************************************************************
 574                  * Enter the guest
 575                  */
 576                 trace_kvm_entry(*vcpu_pc(vcpu));
 577                 kvm_guest_enter();
 578                 vcpu->mode = IN_GUEST_MODE;
 579
 580                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 581
 582                 vcpu->mode = OUTSIDE_GUEST_MODE;
 583                 vcpu->arch.last_pcpu = smp_processor_id();
 584                 kvm_guest_exit();
 585                 trace_kvm_exit(*vcpu_pc(vcpu));
 586                 /*
 587                  * We may have taken a host interrupt in HYP mode (ie
 588                  * while executing the guest). This interrupt is still
 589                  * pending, as we haven't serviced it yet!
 590                  *
 591                  * We're now back in SVC mode, with interrupts
 592                  * disabled.  Enabling the interrupts now will have
 593                  * the effect of taking the interrupt again, in SVC
 594                  * mode this time.
 595                  */
 596                 local_irq_enable();
 597
 598                 /*
 599                  * Back from guest
 600                  *************************************************************/
 601
 602                 kvm_timer_sync_hwstate(vcpu);
 603                 kvm_vgic_sync_hwstate(vcpu);
 604
 605                 ret = handle_exit(vcpu, run, ret);
 606         }
 607
 608         if (vcpu->sigset_active)
 609                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 610         return ret;
 611 }
 612
 613 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 614 {
 615         int bit_index;
 616         bool set;
 617         unsigned long *ptr;
 618
 619         if (number == KVM_ARM_IRQ_CPU_IRQ)
 620                 bit_index = __ffs(HCR_VI);
 621         else /* KVM_ARM_IRQ_CPU_FIQ */
 622                 bit_index = __ffs(HCR_VF);
 623
 624         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 625         if (level)
 626                 set = test_and_set_bit(bit_index, ptr);
 627         else
 628                 set = test_and_clear_bit(bit_index, ptr);
 629
 630         /*
 631          * If we didn't change anything, no need to wake up or kick other CPUs
 632          */
 633         if (set == level)
 634                 return 0;
 635
 636         /*
 637          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 638          * trigger a world-switch round on the running physical CPU to set the
 639          * virtual IRQ/FIQ fields in the HCR appropriately.
 640          */
 641         kvm_vcpu_kick(vcpu);
 642
 643         return 0;
 644 }
 645
 646 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 647                           bool line_status)
 648 {
 649         u32 irq = irq_level->irq;
 650         unsigned int irq_type, vcpu_idx, irq_num;
 651         int nrcpus = atomic_read(&kvm->online_vcpus);
 652         struct kvm_vcpu *vcpu = NULL;
 653         bool level = irq_level->level;
 654
 655         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 656         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 657         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 658
 659         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 660
 661         switch (irq_type) {
 662         case KVM_ARM_IRQ_TYPE_CPU:
 663                 if (irqchip_in_kernel(kvm))
 664                         return -ENXIO;
 665
 666                 if (vcpu_idx >= nrcpus)
 667                         return -EINVAL;
 668
 669                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 670                 if (!vcpu)
 671                         return -EINVAL;
 672
 673                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 674                         return -EINVAL;
 675
 676                 return vcpu_interrupt_line(vcpu, irq_num, level);
 677         case KVM_ARM_IRQ_TYPE_PPI:
 678                 if (!irqchip_in_kernel(kvm))
 679                         return -ENXIO;
 680
 681                 if (vcpu_idx >= nrcpus)
 682                         return -EINVAL;
 683
 684                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 685                 if (!vcpu)
 686                         return -EINVAL;
 687
 688                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 689                         return -EINVAL;
 690
 691                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 692         case KVM_ARM_IRQ_TYPE_SPI:
 693                 if (!irqchip_in_kernel(kvm))
 694                         return -ENXIO;
 695
 696                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 697                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 698                         return -EINVAL;
 699
 700                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 701         }
 702
 703         return -EINVAL;
 704 }
 705
 706 long kvm_arch_vcpu_ioctl(struct file *filp,
 707                          unsigned int ioctl, unsigned long arg)
 708 {
 709         struct kvm_vcpu *vcpu = filp->private_data;
 710         void __user *argp = (void __user *)arg;
 711
 712         switch (ioctl) {
 713         case KVM_ARM_VCPU_INIT: {
 714                 struct kvm_vcpu_init init;
 715
 716                 if (copy_from_user(&init, argp, sizeof(init)))
 717                         return -EFAULT;
 718
 719                 return kvm_vcpu_set_target(vcpu, &init);
 720
 721         }
 722         case KVM_SET_ONE_REG:
 723         case KVM_GET_ONE_REG: {
 724                 struct kvm_one_reg reg;
 725
 726                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 727                         return -ENOEXEC;
 728
 729                 if (copy_from_user(&reg, argp, sizeof(reg)))
 730                         return -EFAULT;
 731                 if (ioctl == KVM_SET_ONE_REG)
 732                         return kvm_arm_set_reg(vcpu, &reg);
 733                 else
 734                         return kvm_arm_get_reg(vcpu, &reg);
 735         }
 736         case KVM_GET_REG_LIST: {
 737                 struct kvm_reg_list __user *user_list = argp;
 738                 struct kvm_reg_list reg_list;
 739                 unsigned n;
 740
 741                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 742                         return -ENOEXEC;
 743
 744                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 745                         return -EFAULT;
 746                 n = reg_list.n;
 747                 reg_list.n = kvm_arm_num_regs(vcpu);
 748                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 749                         return -EFAULT;
 750                 if (n < reg_list.n)
 751                         return -E2BIG;
 752                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 753         }
 754         default:
 755                 return -EINVAL;
 756         }
 757 }
 758
 759 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 760 {
 761         return -EINVAL;
 762 }
 763
 764 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 765                                         struct kvm_arm_device_addr *dev_addr)
 766 {
 767         unsigned long dev_id, type;
 768
 769         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 770                 KVM_ARM_DEVICE_ID_SHIFT;
 771         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 772                 KVM_ARM_DEVICE_TYPE_SHIFT;
 773
 774         switch (dev_id) {
 775         case KVM_ARM_DEVICE_VGIC_V2:
 776                 if (!vgic_present)
 777                         return -ENXIO;
 778                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
 779         default:
 780                 return -ENODEV;
 781         }
 782 }
 783
 784 long kvm_arch_vm_ioctl(struct file *filp,
 785                        unsigned int ioctl, unsigned long arg)
 786 {
 787         struct kvm *kvm = filp->private_data;
 788         void __user *argp = (void __user *)arg;
 789
 790         switch (ioctl) {
 791         case KVM_CREATE_IRQCHIP: {
 792                 if (vgic_present)
 793                         return kvm_vgic_create(kvm);
 794                 else
 795                         return -ENXIO;
 796         }
 797         case KVM_ARM_SET_DEVICE_ADDR: {
 798                 struct kvm_arm_device_addr dev_addr;
 799
 800                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 801                         return -EFAULT;
 802                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 803         }
 804         case KVM_ARM_PREFERRED_TARGET: {
 805                 int err;
 806                 struct kvm_vcpu_init init;
 807
 808                 err = kvm_vcpu_preferred_target(&init);
 809                 if (err)
 810                         return err;
 811
 812                 if (copy_to_user(argp, &init, sizeof(init)))
 813                         return -EFAULT;
 814
 815                 return 0;
 816         }
 817         default:
 818                 return -EINVAL;
 819         }
 820 }
 821
 822 static void cpu_init_hyp_mode(void *dummy)
 823 {
 824         phys_addr_t boot_pgd_ptr;
 825         phys_addr_t pgd_ptr;
 826         unsigned long hyp_stack_ptr;
 827         unsigned long stack_page;
 828         unsigned long vector_ptr;
 829
 830         /* Switch from the HYP stub to our own HYP init vector */
 831         __hyp_set_vectors(kvm_get_idmap_vector());
 832
 833         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 834         pgd_ptr = kvm_mmu_get_httbr();
 835         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 836         hyp_stack_ptr = stack_page + PAGE_SIZE;
 837         vector_ptr = (unsigned long)__kvm_hyp_vector;
 838
 839         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 840 }
 841
 842 static int hyp_init_cpu_notify(struct notifier_block *self,
 843                                unsigned long action, void *cpu)
 844 {
 845         switch (action) {
 846         case CPU_STARTING:
 847         case CPU_STARTING_FROZEN:
 848                 cpu_init_hyp_mode(NULL);
 849                 break;
 850         }
 851
 852         return NOTIFY_OK;
 853 }
 854
 855 static struct notifier_block hyp_init_cpu_nb = {
 856         .notifier_call = hyp_init_cpu_notify,
 857 };
 858
 859 /**
 860  * Inits Hyp-mode on all online CPUs
 861  */
 862 static int init_hyp_mode(void)
 863 {
 864         int cpu;
 865         int err = 0;
 866
 867         /*
 868          * Allocate Hyp PGD and setup Hyp identity mapping
 869          */
 870         err = kvm_mmu_init();
 871         if (err)
 872                 goto out_err;
 873
 874         /*
 875          * It is probably enough to obtain the default on one
 876          * CPU. It's unlikely to be different on the others.
 877          */
 878         hyp_default_vectors = __hyp_get_vectors();
 879
 880         /*
 881          * Allocate stack pages for Hypervisor-mode
 882          */
 883         for_each_possible_cpu(cpu) {
 884                 unsigned long stack_page;
 885
 886                 stack_page = __get_free_page(GFP_KERNEL);
 887                 if (!stack_page) {
 888                         err = -ENOMEM;
 889                         goto out_free_stack_pages;
 890                 }
 891
 892                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 893         }
 894
 895         /*
 896          * Map the Hyp-code called directly from the host
 897          */
 898         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 899         if (err) {
 900                 kvm_err("Cannot map world-switch code\n");
 901                 goto out_free_mappings;
 902         }
 903
 904         /*
 905          * Map the Hyp stack pages
 906          */
 907         for_each_possible_cpu(cpu) {
 908                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 909                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 910
 911                 if (err) {
 912                         kvm_err("Cannot map hyp stack\n");
 913                         goto out_free_mappings;
 914                 }
 915         }
 916
 917         /*
 918          * Map the host CPU structures
 919          */
 920         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 921         if (!kvm_host_cpu_state) {
 922                 err = -ENOMEM;
 923                 kvm_err("Cannot allocate host CPU state\n");
 924                 goto out_free_mappings;
 925         }
 926
 927         for_each_possible_cpu(cpu) {
 928                 kvm_cpu_context_t *cpu_ctxt;
 929
 930                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 931                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 932
 933                 if (err) {
 934                         kvm_err("Cannot map host CPU state: %d\n", err);
 935                         goto out_free_context;
 936                 }
 937         }
 938
 939         /*
 940          * Execute the init code on each CPU.
 941          */
 942         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 943
 944         /*
 945          * Init HYP view of VGIC
 946          */
 947         err = kvm_vgic_hyp_init();
 948         if (err)
 949                 goto out_free_context;
 950
 951 #ifdef CONFIG_KVM_ARM_VGIC
 952                 vgic_present = true;
 953 #endif
 954
 955         /*
 956          * Init HYP architected timer support
 957          */
 958         err = kvm_timer_hyp_init();
 959         if (err)
 960                 goto out_free_mappings;
 961
 962 #ifndef CONFIG_HOTPLUG_CPU
 963         free_boot_hyp_pgd();
 964 #endif
 965
 966         kvm_perf_init();
 967
 968         kvm_info("Hyp mode initialized successfully\n");
 969
 970         return 0;
 971 out_free_context:
 972         free_percpu(kvm_host_cpu_state);
 973 out_free_mappings:
 974         free_hyp_pgds();
 975 out_free_stack_pages:
 976         for_each_possible_cpu(cpu)
 977                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 978 out_err:
 979         kvm_err("error initializing Hyp mode: %d\n", err);
 980         return err;
 981 }
 982
 983 static void check_kvm_target_cpu(void *ret)
 984 {
 985         *(int *)ret = kvm_target_cpu();
 986 }
 987
 988 /**
 989  * Initialize Hyp-mode and memory mappings on all CPUs.
 990  */
 991 int kvm_arch_init(void *opaque)
 992 {
 993         int err;
 994         int ret, cpu;
 995
 996         if (!is_hyp_mode_available()) {
 997                 kvm_err("HYP mode not available\n");
 998                 return -ENODEV;
 999         }
1000
1001         for_each_online_cpu(cpu) {
1002                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1003                 if (ret < 0) {
1004                         kvm_err("Error, CPU %d not supported!\n", cpu);
1005                         return -ENODEV;
1006                 }
1007         }
1008
1009         err = init_hyp_mode();
1010         if (err)
1011                 goto out_err;
1012
1013         err = register_cpu_notifier(&hyp_init_cpu_nb);
1014         if (err) {
1015                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1016                 goto out_err;
1017         }
1018
1019         kvm_coproc_table_init();
1020         return 0;
1021 out_err:
1022         return err;
1023 }
1024
1025 /* NOP: Compiling as a module not supported */
1026 void kvm_arch_exit(void)
1027 {
1028         kvm_perf_teardown();
1029 }
1030
1031 static int arm_init(void)
1032 {
1033         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1034         return rc;
1035 }
1036
1037 module_init(arm_init);