arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/ptrace.h>
  37 #include <asm/mman.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/virt.h>
  41 #include <asm/kvm_arm.h>
  42 #include <asm/kvm_asm.h>
  43 #include <asm/kvm_mmu.h>
  44 #include <asm/kvm_emulate.h>
  45 #include <asm/kvm_coproc.h>
  46 #include <asm/kvm_psci.h>
  47
  48 #ifdef REQUIRES_VIRT
  49 __asm__(".arch_extension        virt");
  50 #endif
  51
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* Per-CPU variable containing the currently running vcpu. */
  57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  58
  59 /* The VMID used in the VTTBR */
  60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  61 static u8 kvm_next_vmid;
  62 static DEFINE_SPINLOCK(kvm_vmid_lock);
  63
  64 static bool vgic_present;
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         BUG_ON(preemptible());
  69         __get_cpu_var(kvm_arm_running_vcpu) = vcpu;
  70 }
  71
  72 /**
  73  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  74  * Must be called from non-preemptible context
  75  */
  76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  77 {
  78         BUG_ON(preemptible());
  79         return __get_cpu_var(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_hardware_enable(void *garbage)
  91 {
  92         return 0;
  93 }
  94
  95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  96 {
  97         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  98 }
  99
 100 void kvm_arch_hardware_disable(void *garbage)
 101 {
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_hardware_unsetup(void)
 110 {
 111 }
 112
 113 void kvm_arch_check_processor_compat(void *rtn)
 114 {
 115         *(int *)rtn = 0;
 116 }
 117
 118 void kvm_arch_sync_events(struct kvm *kvm)
 119 {
 120 }
 121
 122 /**
 123  * kvm_arch_init_vm - initializes a VM data structure
 124  * @kvm:        pointer to the KVM struct
 125  */
 126 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 127 {
 128         int ret = 0;
 129
 130         if (type)
 131                 return -EINVAL;
 132
 133         ret = kvm_alloc_stage2_pgd(kvm);
 134         if (ret)
 135                 goto out_fail_alloc;
 136
 137         ret = create_hyp_mappings(kvm, kvm + 1);
 138         if (ret)
 139                 goto out_free_stage2_pgd;
 140
 141         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 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 159                            struct kvm_memory_slot *dont)
 160 {
 161 }
 162
 163 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 164                             unsigned long npages)
 165 {
 166         return 0;
 167 }
 168
 169 /**
 170  * kvm_arch_destroy_vm - destroy the VM data structure
 171  * @kvm:        pointer to the KVM struct
 172  */
 173 void kvm_arch_destroy_vm(struct kvm *kvm)
 174 {
 175         int i;
 176
 177         kvm_free_stage2_pgd(kvm);
 178
 179         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 180                 if (kvm->vcpus[i]) {
 181                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 182                         kvm->vcpus[i] = NULL;
 183                 }
 184         }
 185 }
 186
 187 int kvm_dev_ioctl_check_extension(long ext)
 188 {
 189         int r;
 190         switch (ext) {
 191         case KVM_CAP_IRQCHIP:
 192                 r = vgic_present;
 193                 break;
 194         case KVM_CAP_USER_MEMORY:
 195         case KVM_CAP_SYNC_MMU:
 196         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 197         case KVM_CAP_ONE_REG:
 198         case KVM_CAP_ARM_PSCI:
 199                 r = 1;
 200                 break;
 201         case KVM_CAP_COALESCED_MMIO:
 202                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 203                 break;
 204         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 205                 r = 1;
 206                 break;
 207         case KVM_CAP_NR_VCPUS:
 208                 r = num_online_cpus();
 209                 break;
 210         case KVM_CAP_MAX_VCPUS:
 211                 r = KVM_MAX_VCPUS;
 212                 break;
 213         default:
 214                 r = kvm_arch_dev_ioctl_check_extension(ext);
 215                 break;
 216         }
 217         return r;
 218 }
 219
 220 long kvm_arch_dev_ioctl(struct file *filp,
 221                         unsigned int ioctl, unsigned long arg)
 222 {
 223         return -EINVAL;
 224 }
 225
 226 void kvm_arch_memslots_updated(struct kvm *kvm)
 227 {
 228 }
 229
 230 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 231                                    struct kvm_memory_slot *memslot,
 232                                    struct kvm_userspace_memory_region *mem,
 233                                    enum kvm_mr_change change)
 234 {
 235         return 0;
 236 }
 237
 238 void kvm_arch_commit_memory_region(struct kvm *kvm,
 239                                    struct kvm_userspace_memory_region *mem,
 240                                    const struct kvm_memory_slot *old,
 241                                    enum kvm_mr_change change)
 242 {
 243 }
 244
 245 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 246 {
 247 }
 248
 249 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 250                                    struct kvm_memory_slot *slot)
 251 {
 252 }
 253
 254 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 255 {
 256         int err;
 257         struct kvm_vcpu *vcpu;
 258
 259         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 260         if (!vcpu) {
 261                 err = -ENOMEM;
 262                 goto out;
 263         }
 264
 265         err = kvm_vcpu_init(vcpu, kvm, id);
 266         if (err)
 267                 goto free_vcpu;
 268
 269         err = create_hyp_mappings(vcpu, vcpu + 1);
 270         if (err)
 271                 goto vcpu_uninit;
 272
 273         return vcpu;
 274 vcpu_uninit:
 275         kvm_vcpu_uninit(vcpu);
 276 free_vcpu:
 277         kmem_cache_free(kvm_vcpu_cache, vcpu);
 278 out:
 279         return ERR_PTR(err);
 280 }
 281
 282 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 283 {
 284         return 0;
 285 }
 286
 287 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 288 {
 289         kvm_mmu_free_memory_caches(vcpu);
 290         kvm_timer_vcpu_terminate(vcpu);
 291         kmem_cache_free(kvm_vcpu_cache, vcpu);
 292 }
 293
 294 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 295 {
 296         kvm_arch_vcpu_free(vcpu);
 297 }
 298
 299 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 300 {
 301         return 0;
 302 }
 303
 304 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 305 {
 306         int ret;
 307
 308         /* Force users to call KVM_ARM_VCPU_INIT */
 309         vcpu->arch.target = -1;
 310
 311         /* Set up VGIC */
 312         ret = kvm_vgic_vcpu_init(vcpu);
 313         if (ret)
 314                 return ret;
 315
 316         /* Set up the timer */
 317         kvm_timer_vcpu_init(vcpu);
 318
 319         return 0;
 320 }
 321
 322 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 323 {
 324 }
 325
 326 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 327 {
 328         vcpu->cpu = cpu;
 329         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 330
 331         /*
 332          * Check whether this vcpu requires the cache to be flushed on
 333          * this physical CPU. This is a consequence of doing dcache
 334          * operations by set/way on this vcpu. We do it here to be in
 335          * a non-preemptible section.
 336          */
 337         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 338                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 339
 340         kvm_arm_set_running_vcpu(vcpu);
 341 }
 342
 343 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 344 {
 345         kvm_arm_set_running_vcpu(NULL);
 346 }
 347
 348 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 349                                         struct kvm_guest_debug *dbg)
 350 {
 351         return -EINVAL;
 352 }
 353
 354
 355 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 356                                     struct kvm_mp_state *mp_state)
 357 {
 358         return -EINVAL;
 359 }
 360
 361 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 362                                     struct kvm_mp_state *mp_state)
 363 {
 364         return -EINVAL;
 365 }
 366
 367 /**
 368  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 369  * @v:          The VCPU pointer
 370  *
 371  * If the guest CPU is not waiting for interrupts or an interrupt line is
 372  * asserted, the CPU is by definition runnable.
 373  */
 374 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 375 {
 376         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 377 }
 378
 379 /* Just ensure a guest exit from a particular CPU */
 380 static void exit_vm_noop(void *info)
 381 {
 382 }
 383
 384 void force_vm_exit(const cpumask_t *mask)
 385 {
 386         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 387 }
 388
 389 /**
 390  * need_new_vmid_gen - check that the VMID is still valid
 391  * @kvm: The VM's VMID to checkt
 392  *
 393  * return true if there is a new generation of VMIDs being used
 394  *
 395  * The hardware supports only 256 values with the value zero reserved for the
 396  * host, so we check if an assigned value belongs to a previous generation,
 397  * which which requires us to assign a new value. If we're the first to use a
 398  * VMID for the new generation, we must flush necessary caches and TLBs on all
 399  * CPUs.
 400  */
 401 static bool need_new_vmid_gen(struct kvm *kvm)
 402 {
 403         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 404 }
 405
 406 /**
 407  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 408  * @kvm The guest that we are about to run
 409  *
 410  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 411  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 412  * caches and TLBs.
 413  */
 414 static void update_vttbr(struct kvm *kvm)
 415 {
 416         phys_addr_t pgd_phys;
 417         u64 vmid;
 418
 419         if (!need_new_vmid_gen(kvm))
 420                 return;
 421
 422         spin_lock(&kvm_vmid_lock);
 423
 424         /*
 425          * We need to re-check the vmid_gen here to ensure that if another vcpu
 426          * already allocated a valid vmid for this vm, then this vcpu should
 427          * use the same vmid.
 428          */
 429         if (!need_new_vmid_gen(kvm)) {
 430                 spin_unlock(&kvm_vmid_lock);
 431                 return;
 432         }
 433
 434         /* First user of a new VMID generation? */
 435         if (unlikely(kvm_next_vmid == 0)) {
 436                 atomic64_inc(&kvm_vmid_gen);
 437                 kvm_next_vmid = 1;
 438
 439                 /*
 440                  * On SMP we know no other CPUs can use this CPU's or each
 441                  * other's VMID after force_vm_exit returns since the
 442                  * kvm_vmid_lock blocks them from reentry to the guest.
 443                  */
 444                 force_vm_exit(cpu_all_mask);
 445                 /*
 446                  * Now broadcast TLB + ICACHE invalidation over the inner
 447                  * shareable domain to make sure all data structures are
 448                  * clean.
 449                  */
 450                 kvm_call_hyp(__kvm_flush_vm_context);
 451         }
 452
 453         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 454         kvm->arch.vmid = kvm_next_vmid;
 455         kvm_next_vmid++;
 456
 457         /* update vttbr to be used with the new vmid */
 458         pgd_phys = virt_to_phys(kvm->arch.pgd);
 459         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 460         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 461         kvm->arch.vttbr |= vmid;
 462
 463         spin_unlock(&kvm_vmid_lock);
 464 }
 465
 466 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 467 {
 468         int ret;
 469
 470         if (likely(vcpu->arch.has_run_once))
 471                 return 0;
 472
 473         vcpu->arch.has_run_once = true;
 474
 475         /*
 476          * Initialize the VGIC before running a vcpu the first time on
 477          * this VM.
 478          */
 479         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 480                 ret = kvm_vgic_init(vcpu->kvm);
 481                 if (ret)
 482                         return ret;
 483         }
 484
 485         return 0;
 486 }
 487
 488 static void vcpu_pause(struct kvm_vcpu *vcpu)
 489 {
 490         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 491
 492         wait_event_interruptible(*wq, !vcpu->arch.pause);
 493 }
 494
 495 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 496 {
 497         return vcpu->arch.target >= 0;
 498 }
 499
 500 /**
 501  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 502  * @vcpu:       The VCPU pointer
 503  * @run:        The kvm_run structure pointer used for userspace state exchange
 504  *
 505  * This function is called through the VCPU_RUN ioctl called from user space. It
 506  * will execute VM code in a loop until the time slice for the process is used
 507  * or some emulation is needed from user space in which case the function will
 508  * return with return value 0 and with the kvm_run structure filled in with the
 509  * required data for the requested emulation.
 510  */
 511 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 512 {
 513         int ret;
 514         sigset_t sigsaved;
 515
 516         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 517                 return -ENOEXEC;
 518
 519         ret = kvm_vcpu_first_run_init(vcpu);
 520         if (ret)
 521                 return ret;
 522
 523         if (run->exit_reason == KVM_EXIT_MMIO) {
 524                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 525                 if (ret)
 526                         return ret;
 527         }
 528
 529         if (vcpu->sigset_active)
 530                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 531
 532         ret = 1;
 533         run->exit_reason = KVM_EXIT_UNKNOWN;
 534         while (ret > 0) {
 535                 /*
 536                  * Check conditions before entering the guest
 537                  */
 538                 cond_resched();
 539
 540                 update_vttbr(vcpu->kvm);
 541
 542                 if (vcpu->arch.pause)
 543                         vcpu_pause(vcpu);
 544
 545                 kvm_vgic_flush_hwstate(vcpu);
 546                 kvm_timer_flush_hwstate(vcpu);
 547
 548                 local_irq_disable();
 549
 550                 /*
 551                  * Re-check atomic conditions
 552                  */
 553                 if (signal_pending(current)) {
 554                         ret = -EINTR;
 555                         run->exit_reason = KVM_EXIT_INTR;
 556                 }
 557
 558                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 559                         local_irq_enable();
 560                         kvm_timer_sync_hwstate(vcpu);
 561                         kvm_vgic_sync_hwstate(vcpu);
 562                         continue;
 563                 }
 564
 565                 /**************************************************************
 566                  * Enter the guest
 567                  */
 568                 trace_kvm_entry(*vcpu_pc(vcpu));
 569                 kvm_guest_enter();
 570                 vcpu->mode = IN_GUEST_MODE;
 571
 572                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 573
 574                 vcpu->mode = OUTSIDE_GUEST_MODE;
 575                 vcpu->arch.last_pcpu = smp_processor_id();
 576                 kvm_guest_exit();
 577                 trace_kvm_exit(*vcpu_pc(vcpu));
 578                 /*
 579                  * We may have taken a host interrupt in HYP mode (ie
 580                  * while executing the guest). This interrupt is still
 581                  * pending, as we haven't serviced it yet!
 582                  *
 583                  * We're now back in SVC mode, with interrupts
 584                  * disabled.  Enabling the interrupts now will have
 585                  * the effect of taking the interrupt again, in SVC
 586                  * mode this time.
 587                  */
 588                 local_irq_enable();
 589
 590                 /*
 591                  * Back from guest
 592                  *************************************************************/
 593
 594                 kvm_timer_sync_hwstate(vcpu);
 595                 kvm_vgic_sync_hwstate(vcpu);
 596
 597                 ret = handle_exit(vcpu, run, ret);
 598         }
 599
 600         if (vcpu->sigset_active)
 601                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 602         return ret;
 603 }
 604
 605 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 606 {
 607         int bit_index;
 608         bool set;
 609         unsigned long *ptr;
 610
 611         if (number == KVM_ARM_IRQ_CPU_IRQ)
 612                 bit_index = __ffs(HCR_VI);
 613         else /* KVM_ARM_IRQ_CPU_FIQ */
 614                 bit_index = __ffs(HCR_VF);
 615
 616         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 617         if (level)
 618                 set = test_and_set_bit(bit_index, ptr);
 619         else
 620                 set = test_and_clear_bit(bit_index, ptr);
 621
 622         /*
 623          * If we didn't change anything, no need to wake up or kick other CPUs
 624          */
 625         if (set == level)
 626                 return 0;
 627
 628         /*
 629          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 630          * trigger a world-switch round on the running physical CPU to set the
 631          * virtual IRQ/FIQ fields in the HCR appropriately.
 632          */
 633         kvm_vcpu_kick(vcpu);
 634
 635         return 0;
 636 }
 637
 638 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 639                           bool line_status)
 640 {
 641         u32 irq = irq_level->irq;
 642         unsigned int irq_type, vcpu_idx, irq_num;
 643         int nrcpus = atomic_read(&kvm->online_vcpus);
 644         struct kvm_vcpu *vcpu = NULL;
 645         bool level = irq_level->level;
 646
 647         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 648         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 649         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 650
 651         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 652
 653         switch (irq_type) {
 654         case KVM_ARM_IRQ_TYPE_CPU:
 655                 if (irqchip_in_kernel(kvm))
 656                         return -ENXIO;
 657
 658                 if (vcpu_idx >= nrcpus)
 659                         return -EINVAL;
 660
 661                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 662                 if (!vcpu)
 663                         return -EINVAL;
 664
 665                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 666                         return -EINVAL;
 667
 668                 return vcpu_interrupt_line(vcpu, irq_num, level);
 669         case KVM_ARM_IRQ_TYPE_PPI:
 670                 if (!irqchip_in_kernel(kvm))
 671                         return -ENXIO;
 672
 673                 if (vcpu_idx >= nrcpus)
 674                         return -EINVAL;
 675
 676                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 677                 if (!vcpu)
 678                         return -EINVAL;
 679
 680                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 681                         return -EINVAL;
 682
 683                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 684         case KVM_ARM_IRQ_TYPE_SPI:
 685                 if (!irqchip_in_kernel(kvm))
 686                         return -ENXIO;
 687
 688                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 689                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 690                         return -EINVAL;
 691
 692                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 693         }
 694
 695         return -EINVAL;
 696 }
 697
 698 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 699                                          struct kvm_vcpu_init *init)
 700 {
 701         int ret;
 702
 703         ret = kvm_vcpu_set_target(vcpu, init);
 704         if (ret)
 705                 return ret;
 706
 707         /*
 708          * Handle the "start in power-off" case by marking the VCPU as paused.
 709          */
 710         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 711                 vcpu->arch.pause = true;
 712
 713         return 0;
 714 }
 715
 716 long kvm_arch_vcpu_ioctl(struct file *filp,
 717                          unsigned int ioctl, unsigned long arg)
 718 {
 719         struct kvm_vcpu *vcpu = filp->private_data;
 720         void __user *argp = (void __user *)arg;
 721
 722         switch (ioctl) {
 723         case KVM_ARM_VCPU_INIT: {
 724                 struct kvm_vcpu_init init;
 725
 726                 if (copy_from_user(&init, argp, sizeof(init)))
 727                         return -EFAULT;
 728
 729                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 730         }
 731         case KVM_SET_ONE_REG:
 732         case KVM_GET_ONE_REG: {
 733                 struct kvm_one_reg reg;
 734
 735                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 736                         return -ENOEXEC;
 737
 738                 if (copy_from_user(&reg, argp, sizeof(reg)))
 739                         return -EFAULT;
 740                 if (ioctl == KVM_SET_ONE_REG)
 741                         return kvm_arm_set_reg(vcpu, &reg);
 742                 else
 743                         return kvm_arm_get_reg(vcpu, &reg);
 744         }
 745         case KVM_GET_REG_LIST: {
 746                 struct kvm_reg_list __user *user_list = argp;
 747                 struct kvm_reg_list reg_list;
 748                 unsigned n;
 749
 750                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 751                         return -ENOEXEC;
 752
 753                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 754                         return -EFAULT;
 755                 n = reg_list.n;
 756                 reg_list.n = kvm_arm_num_regs(vcpu);
 757                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 758                         return -EFAULT;
 759                 if (n < reg_list.n)
 760                         return -E2BIG;
 761                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 762         }
 763         default:
 764                 return -EINVAL;
 765         }
 766 }
 767
 768 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 769 {
 770         return -EINVAL;
 771 }
 772
 773 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 774                                         struct kvm_arm_device_addr *dev_addr)
 775 {
 776         unsigned long dev_id, type;
 777
 778         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 779                 KVM_ARM_DEVICE_ID_SHIFT;
 780         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 781                 KVM_ARM_DEVICE_TYPE_SHIFT;
 782
 783         switch (dev_id) {
 784         case KVM_ARM_DEVICE_VGIC_V2:
 785                 if (!vgic_present)
 786                         return -ENXIO;
 787                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
 788         default:
 789                 return -ENODEV;
 790         }
 791 }
 792
 793 long kvm_arch_vm_ioctl(struct file *filp,
 794                        unsigned int ioctl, unsigned long arg)
 795 {
 796         struct kvm *kvm = filp->private_data;
 797         void __user *argp = (void __user *)arg;
 798
 799         switch (ioctl) {
 800         case KVM_CREATE_IRQCHIP: {
 801                 if (vgic_present)
 802                         return kvm_vgic_create(kvm);
 803                 else
 804                         return -ENXIO;
 805         }
 806         case KVM_ARM_SET_DEVICE_ADDR: {
 807                 struct kvm_arm_device_addr dev_addr;
 808
 809                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 810                         return -EFAULT;
 811                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 812         }
 813         case KVM_ARM_PREFERRED_TARGET: {
 814                 int err;
 815                 struct kvm_vcpu_init init;
 816
 817                 err = kvm_vcpu_preferred_target(&init);
 818                 if (err)
 819                         return err;
 820
 821                 if (copy_to_user(argp, &init, sizeof(init)))
 822                         return -EFAULT;
 823
 824                 return 0;
 825         }
 826         default:
 827                 return -EINVAL;
 828         }
 829 }
 830
 831 static void cpu_init_hyp_mode(void *dummy)
 832 {
 833         phys_addr_t boot_pgd_ptr;
 834         phys_addr_t pgd_ptr;
 835         unsigned long hyp_stack_ptr;
 836         unsigned long stack_page;
 837         unsigned long vector_ptr;
 838
 839         /* Switch from the HYP stub to our own HYP init vector */
 840         __hyp_set_vectors(kvm_get_idmap_vector());
 841
 842         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 843         pgd_ptr = kvm_mmu_get_httbr();
 844         stack_page = __get_cpu_var(kvm_arm_hyp_stack_page);
 845         hyp_stack_ptr = stack_page + PAGE_SIZE;
 846         vector_ptr = (unsigned long)__kvm_hyp_vector;
 847
 848         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 849 }
 850
 851 static int hyp_init_cpu_notify(struct notifier_block *self,
 852                                unsigned long action, void *cpu)
 853 {
 854         switch (action) {
 855         case CPU_STARTING:
 856         case CPU_STARTING_FROZEN:
 857                 cpu_init_hyp_mode(NULL);
 858                 break;
 859         }
 860
 861         return NOTIFY_OK;
 862 }
 863
 864 static struct notifier_block hyp_init_cpu_nb = {
 865         .notifier_call = hyp_init_cpu_notify,
 866 };
 867
 868 #ifdef CONFIG_CPU_PM
 869 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 870                                     unsigned long cmd,
 871                                     void *v)
 872 {
 873         if (cmd == CPU_PM_EXIT) {
 874                 cpu_init_hyp_mode(NULL);
 875                 return NOTIFY_OK;
 876         }
 877
 878         return NOTIFY_DONE;
 879 }
 880
 881 static struct notifier_block hyp_init_cpu_pm_nb = {
 882         .notifier_call = hyp_init_cpu_pm_notifier,
 883 };
 884
 885 static void __init hyp_cpu_pm_init(void)
 886 {
 887         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 888 }
 889 #else
 890 static inline void hyp_cpu_pm_init(void)
 891 {
 892 }
 893 #endif
 894
 895 /**
 896  * Inits Hyp-mode on all online CPUs
 897  */
 898 static int init_hyp_mode(void)
 899 {
 900         int cpu;
 901         int err = 0;
 902
 903         /*
 904          * Allocate Hyp PGD and setup Hyp identity mapping
 905          */
 906         err = kvm_mmu_init();
 907         if (err)
 908                 goto out_err;
 909
 910         /*
 911          * It is probably enough to obtain the default on one
 912          * CPU. It's unlikely to be different on the others.
 913          */
 914         hyp_default_vectors = __hyp_get_vectors();
 915
 916         /*
 917          * Allocate stack pages for Hypervisor-mode
 918          */
 919         for_each_possible_cpu(cpu) {
 920                 unsigned long stack_page;
 921
 922                 stack_page = __get_free_page(GFP_KERNEL);
 923                 if (!stack_page) {
 924                         err = -ENOMEM;
 925                         goto out_free_stack_pages;
 926                 }
 927
 928                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 929         }
 930
 931         /*
 932          * Map the Hyp-code called directly from the host
 933          */
 934         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 935         if (err) {
 936                 kvm_err("Cannot map world-switch code\n");
 937                 goto out_free_mappings;
 938         }
 939
 940         /*
 941          * Map the Hyp stack pages
 942          */
 943         for_each_possible_cpu(cpu) {
 944                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 945                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 946
 947                 if (err) {
 948                         kvm_err("Cannot map hyp stack\n");
 949                         goto out_free_mappings;
 950                 }
 951         }
 952
 953         /*
 954          * Map the host CPU structures
 955          */
 956         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 957         if (!kvm_host_cpu_state) {
 958                 err = -ENOMEM;
 959                 kvm_err("Cannot allocate host CPU state\n");
 960                 goto out_free_mappings;
 961         }
 962
 963         for_each_possible_cpu(cpu) {
 964                 kvm_cpu_context_t *cpu_ctxt;
 965
 966                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 967                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 968
 969                 if (err) {
 970                         kvm_err("Cannot map host CPU state: %d\n", err);
 971                         goto out_free_context;
 972                 }
 973         }
 974
 975         /*
 976          * Execute the init code on each CPU.
 977          */
 978         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 979
 980         /*
 981          * Init HYP view of VGIC
 982          */
 983         err = kvm_vgic_hyp_init();
 984         if (err)
 985                 goto out_free_context;
 986
 987 #ifdef CONFIG_KVM_ARM_VGIC
 988                 vgic_present = true;
 989 #endif
 990
 991         /*
 992          * Init HYP architected timer support
 993          */
 994         err = kvm_timer_hyp_init();
 995         if (err)
 996                 goto out_free_mappings;
 997
 998 #ifndef CONFIG_HOTPLUG_CPU
 999         free_boot_hyp_pgd();
1000 #endif
1001
1002         kvm_perf_init();
1003
1004         kvm_info("Hyp mode initialized successfully\n");
1005
1006         return 0;
1007 out_free_context:
1008         free_percpu(kvm_host_cpu_state);
1009 out_free_mappings:
1010         free_hyp_pgds();
1011 out_free_stack_pages:
1012         for_each_possible_cpu(cpu)
1013                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1014 out_err:
1015         kvm_err("error initializing Hyp mode: %d\n", err);
1016         return err;
1017 }
1018
1019 static void check_kvm_target_cpu(void *ret)
1020 {
1021         *(int *)ret = kvm_target_cpu();
1022 }
1023
1024 /**
1025  * Initialize Hyp-mode and memory mappings on all CPUs.
1026  */
1027 int kvm_arch_init(void *opaque)
1028 {
1029         int err;
1030         int ret, cpu;
1031
1032         if (!is_hyp_mode_available()) {
1033                 kvm_err("HYP mode not available\n");
1034                 return -ENODEV;
1035         }
1036
1037         for_each_online_cpu(cpu) {
1038                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1039                 if (ret < 0) {
1040                         kvm_err("Error, CPU %d not supported!\n", cpu);
1041                         return -ENODEV;
1042                 }
1043         }
1044
1045         err = init_hyp_mode();
1046         if (err)
1047                 goto out_err;
1048
1049         err = register_cpu_notifier(&hyp_init_cpu_nb);
1050         if (err) {
1051                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1052                 goto out_err;
1053         }
1054
1055         hyp_cpu_pm_init();
1056
1057         kvm_coproc_table_init();
1058         return 0;
1059 out_err:
1060         return err;
1061 }
1062
1063 /* NOP: Compiling as a module not supported */
1064 void kvm_arch_exit(void)
1065 {
1066         kvm_perf_teardown();
1067 }
1068
1069 static int arm_init(void)
1070 {
1071         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1072         return rc;
1073 }
1074
1075 module_init(arm_init);