arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/ptrace.h>
  37 #include <asm/mman.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/virt.h>
  41 #include <asm/kvm_arm.h>
  42 #include <asm/kvm_asm.h>
  43 #include <asm/kvm_mmu.h>
  44 #include <asm/kvm_emulate.h>
  45 #include <asm/kvm_coproc.h>
  46 #include <asm/kvm_psci.h>
  47
  48 #ifdef REQUIRES_VIRT
  49 __asm__(".arch_extension        virt");
  50 #endif
  51
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* Per-CPU variable containing the currently running vcpu. */
  57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  58
  59 /* The VMID used in the VTTBR */
  60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  61 static u8 kvm_next_vmid;
  62 static DEFINE_SPINLOCK(kvm_vmid_lock);
  63
  64 static bool vgic_present;
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         BUG_ON(preemptible());
  69         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  70 }
  71
  72 /**
  73  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  74  * Must be called from non-preemptible context
  75  */
  76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  77 {
  78         BUG_ON(preemptible());
  79         return __this_cpu_read(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_hardware_enable(void *garbage)
  91 {
  92         return 0;
  93 }
  94
  95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  96 {
  97         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  98 }
  99
 100 void kvm_arch_hardware_disable(void *garbage)
 101 {
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_hardware_unsetup(void)
 110 {
 111 }
 112
 113 void kvm_arch_check_processor_compat(void *rtn)
 114 {
 115         *(int *)rtn = 0;
 116 }
 117
 118 void kvm_arch_sync_events(struct kvm *kvm)
 119 {
 120 }
 121
 122 /**
 123  * kvm_arch_init_vm - initializes a VM data structure
 124  * @kvm:        pointer to the KVM struct
 125  */
 126 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 127 {
 128         int ret = 0;
 129
 130         if (type)
 131                 return -EINVAL;
 132
 133         ret = kvm_alloc_stage2_pgd(kvm);
 134         if (ret)
 135                 goto out_fail_alloc;
 136
 137         ret = create_hyp_mappings(kvm, kvm + 1);
 138         if (ret)
 139                 goto out_free_stage2_pgd;
 140
 141         kvm_timer_init(kvm);
 142
 143         /* Mark the initial VMID generation invalid */
 144         kvm->arch.vmid_gen = 0;
 145
 146         return ret;
 147 out_free_stage2_pgd:
 148         kvm_free_stage2_pgd(kvm);
 149 out_fail_alloc:
 150         return ret;
 151 }
 152
 153 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 154 {
 155         return VM_FAULT_SIGBUS;
 156 }
 157
 158 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_DEVICE_CTRL:
 195         case KVM_CAP_USER_MEMORY:
 196         case KVM_CAP_SYNC_MMU:
 197         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 198         case KVM_CAP_ONE_REG:
 199         case KVM_CAP_ARM_PSCI:
 200         case KVM_CAP_ARM_PSCI_0_2:
 201                 r = 1;
 202                 break;
 203         case KVM_CAP_COALESCED_MMIO:
 204                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 205                 break;
 206         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 207                 r = 1;
 208                 break;
 209         case KVM_CAP_NR_VCPUS:
 210                 r = num_online_cpus();
 211                 break;
 212         case KVM_CAP_MAX_VCPUS:
 213                 r = KVM_MAX_VCPUS;
 214                 break;
 215         default:
 216                 r = kvm_arch_dev_ioctl_check_extension(ext);
 217                 break;
 218         }
 219         return r;
 220 }
 221
 222 long kvm_arch_dev_ioctl(struct file *filp,
 223                         unsigned int ioctl, unsigned long arg)
 224 {
 225         return -EINVAL;
 226 }
 227
 228 void kvm_arch_memslots_updated(struct kvm *kvm)
 229 {
 230 }
 231
 232 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 233                                    struct kvm_memory_slot *memslot,
 234                                    struct kvm_userspace_memory_region *mem,
 235                                    enum kvm_mr_change change)
 236 {
 237         return 0;
 238 }
 239
 240 void kvm_arch_commit_memory_region(struct kvm *kvm,
 241                                    struct kvm_userspace_memory_region *mem,
 242                                    const struct kvm_memory_slot *old,
 243                                    enum kvm_mr_change change)
 244 {
 245 }
 246
 247 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 248 {
 249 }
 250
 251 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 252                                    struct kvm_memory_slot *slot)
 253 {
 254 }
 255
 256 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 257 {
 258         int err;
 259         struct kvm_vcpu *vcpu;
 260
 261         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 262         if (!vcpu) {
 263                 err = -ENOMEM;
 264                 goto out;
 265         }
 266
 267         err = kvm_vcpu_init(vcpu, kvm, id);
 268         if (err)
 269                 goto free_vcpu;
 270
 271         err = create_hyp_mappings(vcpu, vcpu + 1);
 272         if (err)
 273                 goto vcpu_uninit;
 274
 275         return vcpu;
 276 vcpu_uninit:
 277         kvm_vcpu_uninit(vcpu);
 278 free_vcpu:
 279         kmem_cache_free(kvm_vcpu_cache, vcpu);
 280 out:
 281         return ERR_PTR(err);
 282 }
 283
 284 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 285 {
 286         return 0;
 287 }
 288
 289 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 290 {
 291         kvm_mmu_free_memory_caches(vcpu);
 292         kvm_timer_vcpu_terminate(vcpu);
 293         kmem_cache_free(kvm_vcpu_cache, vcpu);
 294 }
 295
 296 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 297 {
 298         kvm_arch_vcpu_free(vcpu);
 299 }
 300
 301 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 302 {
 303         return 0;
 304 }
 305
 306 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 307 {
 308         int ret;
 309
 310         /* Force users to call KVM_ARM_VCPU_INIT */
 311         vcpu->arch.target = -1;
 312
 313         /* Set up VGIC */
 314         ret = kvm_vgic_vcpu_init(vcpu);
 315         if (ret)
 316                 return ret;
 317
 318         /* Set up the timer */
 319         kvm_timer_vcpu_init(vcpu);
 320
 321         return 0;
 322 }
 323
 324 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 325 {
 326 }
 327
 328 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 329 {
 330         vcpu->cpu = cpu;
 331         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 332
 333         /*
 334          * Check whether this vcpu requires the cache to be flushed on
 335          * this physical CPU. This is a consequence of doing dcache
 336          * operations by set/way on this vcpu. We do it here to be in
 337          * a non-preemptible section.
 338          */
 339         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 340                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 341
 342         kvm_arm_set_running_vcpu(vcpu);
 343 }
 344
 345 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 346 {
 347         /*
 348          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 349          * if the vcpu is no longer assigned to a cpu.  This is used for the
 350          * optimized make_all_cpus_request path.
 351          */
 352         vcpu->cpu = -1;
 353
 354         kvm_arm_set_running_vcpu(NULL);
 355 }
 356
 357 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 358                                         struct kvm_guest_debug *dbg)
 359 {
 360         return -EINVAL;
 361 }
 362
 363
 364 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 365                                     struct kvm_mp_state *mp_state)
 366 {
 367         return -EINVAL;
 368 }
 369
 370 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 371                                     struct kvm_mp_state *mp_state)
 372 {
 373         return -EINVAL;
 374 }
 375
 376 /**
 377  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 378  * @v:          The VCPU pointer
 379  *
 380  * If the guest CPU is not waiting for interrupts or an interrupt line is
 381  * asserted, the CPU is by definition runnable.
 382  */
 383 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 384 {
 385         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 386 }
 387
 388 /* Just ensure a guest exit from a particular CPU */
 389 static void exit_vm_noop(void *info)
 390 {
 391 }
 392
 393 void force_vm_exit(const cpumask_t *mask)
 394 {
 395         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 396 }
 397
 398 /**
 399  * need_new_vmid_gen - check that the VMID is still valid
 400  * @kvm: The VM's VMID to checkt
 401  *
 402  * return true if there is a new generation of VMIDs being used
 403  *
 404  * The hardware supports only 256 values with the value zero reserved for the
 405  * host, so we check if an assigned value belongs to a previous generation,
 406  * which which requires us to assign a new value. If we're the first to use a
 407  * VMID for the new generation, we must flush necessary caches and TLBs on all
 408  * CPUs.
 409  */
 410 static bool need_new_vmid_gen(struct kvm *kvm)
 411 {
 412         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 413 }
 414
 415 /**
 416  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 417  * @kvm The guest that we are about to run
 418  *
 419  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 420  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 421  * caches and TLBs.
 422  */
 423 static void update_vttbr(struct kvm *kvm)
 424 {
 425         phys_addr_t pgd_phys;
 426         u64 vmid;
 427
 428         if (!need_new_vmid_gen(kvm))
 429                 return;
 430
 431         spin_lock(&kvm_vmid_lock);
 432
 433         /*
 434          * We need to re-check the vmid_gen here to ensure that if another vcpu
 435          * already allocated a valid vmid for this vm, then this vcpu should
 436          * use the same vmid.
 437          */
 438         if (!need_new_vmid_gen(kvm)) {
 439                 spin_unlock(&kvm_vmid_lock);
 440                 return;
 441         }
 442
 443         /* First user of a new VMID generation? */
 444         if (unlikely(kvm_next_vmid == 0)) {
 445                 atomic64_inc(&kvm_vmid_gen);
 446                 kvm_next_vmid = 1;
 447
 448                 /*
 449                  * On SMP we know no other CPUs can use this CPU's or each
 450                  * other's VMID after force_vm_exit returns since the
 451                  * kvm_vmid_lock blocks them from reentry to the guest.
 452                  */
 453                 force_vm_exit(cpu_all_mask);
 454                 /*
 455                  * Now broadcast TLB + ICACHE invalidation over the inner
 456                  * shareable domain to make sure all data structures are
 457                  * clean.
 458                  */
 459                 kvm_call_hyp(__kvm_flush_vm_context);
 460         }
 461
 462         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 463         kvm->arch.vmid = kvm_next_vmid;
 464         kvm_next_vmid++;
 465
 466         /* update vttbr to be used with the new vmid */
 467         pgd_phys = virt_to_phys(kvm->arch.pgd);
 468         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 469         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 470         kvm->arch.vttbr |= vmid;
 471
 472         spin_unlock(&kvm_vmid_lock);
 473 }
 474
 475 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 476 {
 477         int ret;
 478
 479         if (likely(vcpu->arch.has_run_once))
 480                 return 0;
 481
 482         vcpu->arch.has_run_once = true;
 483
 484         /*
 485          * Initialize the VGIC before running a vcpu the first time on
 486          * this VM.
 487          */
 488         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 489                 ret = kvm_vgic_init(vcpu->kvm);
 490                 if (ret)
 491                         return ret;
 492         }
 493
 494         return 0;
 495 }
 496
 497 static void vcpu_pause(struct kvm_vcpu *vcpu)
 498 {
 499         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 500
 501         wait_event_interruptible(*wq, !vcpu->arch.pause);
 502 }
 503
 504 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 505 {
 506         return vcpu->arch.target >= 0;
 507 }
 508
 509 /**
 510  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 511  * @vcpu:       The VCPU pointer
 512  * @run:        The kvm_run structure pointer used for userspace state exchange
 513  *
 514  * This function is called through the VCPU_RUN ioctl called from user space. It
 515  * will execute VM code in a loop until the time slice for the process is used
 516  * or some emulation is needed from user space in which case the function will
 517  * return with return value 0 and with the kvm_run structure filled in with the
 518  * required data for the requested emulation.
 519  */
 520 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 521 {
 522         int ret;
 523         sigset_t sigsaved;
 524
 525         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 526                 return -ENOEXEC;
 527
 528         ret = kvm_vcpu_first_run_init(vcpu);
 529         if (ret)
 530                 return ret;
 531
 532         if (run->exit_reason == KVM_EXIT_MMIO) {
 533                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 534                 if (ret)
 535                         return ret;
 536         }
 537
 538         if (vcpu->sigset_active)
 539                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 540
 541         ret = 1;
 542         run->exit_reason = KVM_EXIT_UNKNOWN;
 543         while (ret > 0) {
 544                 /*
 545                  * Check conditions before entering the guest
 546                  */
 547                 cond_resched();
 548
 549                 update_vttbr(vcpu->kvm);
 550
 551                 if (vcpu->arch.pause)
 552                         vcpu_pause(vcpu);
 553
 554                 kvm_vgic_flush_hwstate(vcpu);
 555                 kvm_timer_flush_hwstate(vcpu);
 556
 557                 local_irq_disable();
 558
 559                 /*
 560                  * Re-check atomic conditions
 561                  */
 562                 if (signal_pending(current)) {
 563                         ret = -EINTR;
 564                         run->exit_reason = KVM_EXIT_INTR;
 565                 }
 566
 567                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 568                         local_irq_enable();
 569                         kvm_timer_sync_hwstate(vcpu);
 570                         kvm_vgic_sync_hwstate(vcpu);
 571                         continue;
 572                 }
 573
 574                 /**************************************************************
 575                  * Enter the guest
 576                  */
 577                 trace_kvm_entry(*vcpu_pc(vcpu));
 578                 kvm_guest_enter();
 579                 vcpu->mode = IN_GUEST_MODE;
 580
 581                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 582
 583                 vcpu->mode = OUTSIDE_GUEST_MODE;
 584                 vcpu->arch.last_pcpu = smp_processor_id();
 585                 kvm_guest_exit();
 586                 trace_kvm_exit(*vcpu_pc(vcpu));
 587                 /*
 588                  * We may have taken a host interrupt in HYP mode (ie
 589                  * while executing the guest). This interrupt is still
 590                  * pending, as we haven't serviced it yet!
 591                  *
 592                  * We're now back in SVC mode, with interrupts
 593                  * disabled.  Enabling the interrupts now will have
 594                  * the effect of taking the interrupt again, in SVC
 595                  * mode this time.
 596                  */
 597                 local_irq_enable();
 598
 599                 /*
 600                  * Back from guest
 601                  *************************************************************/
 602
 603                 kvm_timer_sync_hwstate(vcpu);
 604                 kvm_vgic_sync_hwstate(vcpu);
 605
 606                 ret = handle_exit(vcpu, run, ret);
 607         }
 608
 609         if (vcpu->sigset_active)
 610                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 611         return ret;
 612 }
 613
 614 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 615 {
 616         int bit_index;
 617         bool set;
 618         unsigned long *ptr;
 619
 620         if (number == KVM_ARM_IRQ_CPU_IRQ)
 621                 bit_index = __ffs(HCR_VI);
 622         else /* KVM_ARM_IRQ_CPU_FIQ */
 623                 bit_index = __ffs(HCR_VF);
 624
 625         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 626         if (level)
 627                 set = test_and_set_bit(bit_index, ptr);
 628         else
 629                 set = test_and_clear_bit(bit_index, ptr);
 630
 631         /*
 632          * If we didn't change anything, no need to wake up or kick other CPUs
 633          */
 634         if (set == level)
 635                 return 0;
 636
 637         /*
 638          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 639          * trigger a world-switch round on the running physical CPU to set the
 640          * virtual IRQ/FIQ fields in the HCR appropriately.
 641          */
 642         kvm_vcpu_kick(vcpu);
 643
 644         return 0;
 645 }
 646
 647 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 648                           bool line_status)
 649 {
 650         u32 irq = irq_level->irq;
 651         unsigned int irq_type, vcpu_idx, irq_num;
 652         int nrcpus = atomic_read(&kvm->online_vcpus);
 653         struct kvm_vcpu *vcpu = NULL;
 654         bool level = irq_level->level;
 655
 656         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 657         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 658         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 659
 660         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 661
 662         switch (irq_type) {
 663         case KVM_ARM_IRQ_TYPE_CPU:
 664                 if (irqchip_in_kernel(kvm))
 665                         return -ENXIO;
 666
 667                 if (vcpu_idx >= nrcpus)
 668                         return -EINVAL;
 669
 670                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 671                 if (!vcpu)
 672                         return -EINVAL;
 673
 674                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 675                         return -EINVAL;
 676
 677                 return vcpu_interrupt_line(vcpu, irq_num, level);
 678         case KVM_ARM_IRQ_TYPE_PPI:
 679                 if (!irqchip_in_kernel(kvm))
 680                         return -ENXIO;
 681
 682                 if (vcpu_idx >= nrcpus)
 683                         return -EINVAL;
 684
 685                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 686                 if (!vcpu)
 687                         return -EINVAL;
 688
 689                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 690                         return -EINVAL;
 691
 692                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 693         case KVM_ARM_IRQ_TYPE_SPI:
 694                 if (!irqchip_in_kernel(kvm))
 695                         return -ENXIO;
 696
 697                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 698                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 699                         return -EINVAL;
 700
 701                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 702         }
 703
 704         return -EINVAL;
 705 }
 706
 707 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 708                                          struct kvm_vcpu_init *init)
 709 {
 710         int ret;
 711
 712         ret = kvm_vcpu_set_target(vcpu, init);
 713         if (ret)
 714                 return ret;
 715
 716         /*
 717          * Handle the "start in power-off" case by marking the VCPU as paused.
 718          */
 719         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 720                 vcpu->arch.pause = true;
 721
 722         return 0;
 723 }
 724
 725 long kvm_arch_vcpu_ioctl(struct file *filp,
 726                          unsigned int ioctl, unsigned long arg)
 727 {
 728         struct kvm_vcpu *vcpu = filp->private_data;
 729         void __user *argp = (void __user *)arg;
 730
 731         switch (ioctl) {
 732         case KVM_ARM_VCPU_INIT: {
 733                 struct kvm_vcpu_init init;
 734
 735                 if (copy_from_user(&init, argp, sizeof(init)))
 736                         return -EFAULT;
 737
 738                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 739         }
 740         case KVM_SET_ONE_REG:
 741         case KVM_GET_ONE_REG: {
 742                 struct kvm_one_reg reg;
 743
 744                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 745                         return -ENOEXEC;
 746
 747                 if (copy_from_user(&reg, argp, sizeof(reg)))
 748                         return -EFAULT;
 749                 if (ioctl == KVM_SET_ONE_REG)
 750                         return kvm_arm_set_reg(vcpu, &reg);
 751                 else
 752                         return kvm_arm_get_reg(vcpu, &reg);
 753         }
 754         case KVM_GET_REG_LIST: {
 755                 struct kvm_reg_list __user *user_list = argp;
 756                 struct kvm_reg_list reg_list;
 757                 unsigned n;
 758
 759                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 760                         return -ENOEXEC;
 761
 762                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 763                         return -EFAULT;
 764                 n = reg_list.n;
 765                 reg_list.n = kvm_arm_num_regs(vcpu);
 766                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 767                         return -EFAULT;
 768                 if (n < reg_list.n)
 769                         return -E2BIG;
 770                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 771         }
 772         default:
 773                 return -EINVAL;
 774         }
 775 }
 776
 777 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 778 {
 779         return -EINVAL;
 780 }
 781
 782 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 783                                         struct kvm_arm_device_addr *dev_addr)
 784 {
 785         unsigned long dev_id, type;
 786
 787         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 788                 KVM_ARM_DEVICE_ID_SHIFT;
 789         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 790                 KVM_ARM_DEVICE_TYPE_SHIFT;
 791
 792         switch (dev_id) {
 793         case KVM_ARM_DEVICE_VGIC_V2:
 794                 if (!vgic_present)
 795                         return -ENXIO;
 796                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 797         default:
 798                 return -ENODEV;
 799         }
 800 }
 801
 802 long kvm_arch_vm_ioctl(struct file *filp,
 803                        unsigned int ioctl, unsigned long arg)
 804 {
 805         struct kvm *kvm = filp->private_data;
 806         void __user *argp = (void __user *)arg;
 807
 808         switch (ioctl) {
 809         case KVM_CREATE_IRQCHIP: {
 810                 if (vgic_present)
 811                         return kvm_vgic_create(kvm);
 812                 else
 813                         return -ENXIO;
 814         }
 815         case KVM_ARM_SET_DEVICE_ADDR: {
 816                 struct kvm_arm_device_addr dev_addr;
 817
 818                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 819                         return -EFAULT;
 820                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 821         }
 822         case KVM_ARM_PREFERRED_TARGET: {
 823                 int err;
 824                 struct kvm_vcpu_init init;
 825
 826                 err = kvm_vcpu_preferred_target(&init);
 827                 if (err)
 828                         return err;
 829
 830                 if (copy_to_user(argp, &init, sizeof(init)))
 831                         return -EFAULT;
 832
 833                 return 0;
 834         }
 835         default:
 836                 return -EINVAL;
 837         }
 838 }
 839
 840 static void cpu_init_hyp_mode(void *dummy)
 841 {
 842         phys_addr_t boot_pgd_ptr;
 843         phys_addr_t pgd_ptr;
 844         unsigned long hyp_stack_ptr;
 845         unsigned long stack_page;
 846         unsigned long vector_ptr;
 847
 848         /* Switch from the HYP stub to our own HYP init vector */
 849         __hyp_set_vectors(kvm_get_idmap_vector());
 850
 851         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 852         pgd_ptr = kvm_mmu_get_httbr();
 853         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 854         hyp_stack_ptr = stack_page + PAGE_SIZE;
 855         vector_ptr = (unsigned long)__kvm_hyp_vector;
 856
 857         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 858 }
 859
 860 static int hyp_init_cpu_notify(struct notifier_block *self,
 861                                unsigned long action, void *cpu)
 862 {
 863         switch (action) {
 864         case CPU_STARTING:
 865         case CPU_STARTING_FROZEN:
 866                 cpu_init_hyp_mode(NULL);
 867                 break;
 868         }
 869
 870         return NOTIFY_OK;
 871 }
 872
 873 static struct notifier_block hyp_init_cpu_nb = {
 874         .notifier_call = hyp_init_cpu_notify,
 875 };
 876
 877 #ifdef CONFIG_CPU_PM
 878 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 879                                     unsigned long cmd,
 880                                     void *v)
 881 {
 882         if (cmd == CPU_PM_EXIT &&
 883             __hyp_get_vectors() == hyp_default_vectors) {
 884                 cpu_init_hyp_mode(NULL);
 885                 return NOTIFY_OK;
 886         }
 887
 888         return NOTIFY_DONE;
 889 }
 890
 891 static struct notifier_block hyp_init_cpu_pm_nb = {
 892         .notifier_call = hyp_init_cpu_pm_notifier,
 893 };
 894
 895 static void __init hyp_cpu_pm_init(void)
 896 {
 897         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 898 }
 899 #else
 900 static inline void hyp_cpu_pm_init(void)
 901 {
 902 }
 903 #endif
 904
 905 /**
 906  * Inits Hyp-mode on all online CPUs
 907  */
 908 static int init_hyp_mode(void)
 909 {
 910         int cpu;
 911         int err = 0;
 912
 913         /*
 914          * Allocate Hyp PGD and setup Hyp identity mapping
 915          */
 916         err = kvm_mmu_init();
 917         if (err)
 918                 goto out_err;
 919
 920         /*
 921          * It is probably enough to obtain the default on one
 922          * CPU. It's unlikely to be different on the others.
 923          */
 924         hyp_default_vectors = __hyp_get_vectors();
 925
 926         /*
 927          * Allocate stack pages for Hypervisor-mode
 928          */
 929         for_each_possible_cpu(cpu) {
 930                 unsigned long stack_page;
 931
 932                 stack_page = __get_free_page(GFP_KERNEL);
 933                 if (!stack_page) {
 934                         err = -ENOMEM;
 935                         goto out_free_stack_pages;
 936                 }
 937
 938                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 939         }
 940
 941         /*
 942          * Map the Hyp-code called directly from the host
 943          */
 944         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 945         if (err) {
 946                 kvm_err("Cannot map world-switch code\n");
 947                 goto out_free_mappings;
 948         }
 949
 950         /*
 951          * Map the Hyp stack pages
 952          */
 953         for_each_possible_cpu(cpu) {
 954                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 955                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 956
 957                 if (err) {
 958                         kvm_err("Cannot map hyp stack\n");
 959                         goto out_free_mappings;
 960                 }
 961         }
 962
 963         /*
 964          * Map the host CPU structures
 965          */
 966         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 967         if (!kvm_host_cpu_state) {
 968                 err = -ENOMEM;
 969                 kvm_err("Cannot allocate host CPU state\n");
 970                 goto out_free_mappings;
 971         }
 972
 973         for_each_possible_cpu(cpu) {
 974                 kvm_cpu_context_t *cpu_ctxt;
 975
 976                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 977                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 978
 979                 if (err) {
 980                         kvm_err("Cannot map host CPU state: %d\n", err);
 981                         goto out_free_context;
 982                 }
 983         }
 984
 985         /*
 986          * Execute the init code on each CPU.
 987          */
 988         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 989
 990         /*
 991          * Init HYP view of VGIC
 992          */
 993         err = kvm_vgic_hyp_init();
 994         if (err)
 995                 goto out_free_context;
 996
 997 #ifdef CONFIG_KVM_ARM_VGIC
 998                 vgic_present = true;
 999 #endif
1000
1001         /*
1002          * Init HYP architected timer support
1003          */
1004         err = kvm_timer_hyp_init();
1005         if (err)
1006                 goto out_free_mappings;
1007
1008 #ifndef CONFIG_HOTPLUG_CPU
1009         free_boot_hyp_pgd();
1010 #endif
1011
1012         kvm_perf_init();
1013
1014         kvm_info("Hyp mode initialized successfully\n");
1015
1016         return 0;
1017 out_free_context:
1018         free_percpu(kvm_host_cpu_state);
1019 out_free_mappings:
1020         free_hyp_pgds();
1021 out_free_stack_pages:
1022         for_each_possible_cpu(cpu)
1023                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1024 out_err:
1025         kvm_err("error initializing Hyp mode: %d\n", err);
1026         return err;
1027 }
1028
1029 static void check_kvm_target_cpu(void *ret)
1030 {
1031         *(int *)ret = kvm_target_cpu();
1032 }
1033
1034 /**
1035  * Initialize Hyp-mode and memory mappings on all CPUs.
1036  */
1037 int kvm_arch_init(void *opaque)
1038 {
1039         int err;
1040         int ret, cpu;
1041
1042         if (!is_hyp_mode_available()) {
1043                 kvm_err("HYP mode not available\n");
1044                 return -ENODEV;
1045         }
1046
1047         for_each_online_cpu(cpu) {
1048                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1049                 if (ret < 0) {
1050                         kvm_err("Error, CPU %d not supported!\n", cpu);
1051                         return -ENODEV;
1052                 }
1053         }
1054
1055         cpu_notifier_register_begin();
1056
1057         err = init_hyp_mode();
1058         if (err)
1059                 goto out_err;
1060
1061         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1062         if (err) {
1063                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1064                 goto out_err;
1065         }
1066
1067         cpu_notifier_register_done();
1068
1069         hyp_cpu_pm_init();
1070
1071         kvm_coproc_table_init();
1072         return 0;
1073 out_err:
1074         cpu_notifier_register_done();
1075         return err;
1076 }
1077
1078 /* NOP: Compiling as a module not supported */
1079 void kvm_arch_exit(void)
1080 {
1081         kvm_perf_teardown();
1082 }
1083
1084 static int arm_init(void)
1085 {
1086         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1087         return rc;
1088 }
1089
1090 module_init(arm_init);