arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/ptrace.h>
  37 #include <asm/mman.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/virt.h>
  41 #include <asm/kvm_arm.h>
  42 #include <asm/kvm_asm.h>
  43 #include <asm/kvm_mmu.h>
  44 #include <asm/kvm_emulate.h>
  45 #include <asm/kvm_coproc.h>
  46 #include <asm/kvm_psci.h>
  47
  48 #ifdef REQUIRES_VIRT
  49 __asm__(".arch_extension        virt");
  50 #endif
  51
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* Per-CPU variable containing the currently running vcpu. */
  57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  58
  59 /* The VMID used in the VTTBR */
  60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  61 static u8 kvm_next_vmid;
  62 static DEFINE_SPINLOCK(kvm_vmid_lock);
  63
  64 static bool vgic_present;
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         BUG_ON(preemptible());
  69         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  70 }
  71
  72 /**
  73  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  74  * Must be called from non-preemptible context
  75  */
  76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  77 {
  78         BUG_ON(preemptible());
  79         return __this_cpu_read(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_hardware_enable(void *garbage)
  91 {
  92         return 0;
  93 }
  94
  95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  96 {
  97         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  98 }
  99
 100 void kvm_arch_hardware_disable(void *garbage)
 101 {
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_hardware_unsetup(void)
 110 {
 111 }
 112
 113 void kvm_arch_check_processor_compat(void *rtn)
 114 {
 115         *(int *)rtn = 0;
 116 }
 117
 118 void kvm_arch_sync_events(struct kvm *kvm)
 119 {
 120 }
 121
 122 /**
 123  * kvm_arch_init_vm - initializes a VM data structure
 124  * @kvm:        pointer to the KVM struct
 125  */
 126 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 127 {
 128         int ret = 0;
 129
 130         if (type)
 131                 return -EINVAL;
 132
 133         ret = kvm_alloc_stage2_pgd(kvm);
 134         if (ret)
 135                 goto out_fail_alloc;
 136
 137         ret = create_hyp_mappings(kvm, kvm + 1);
 138         if (ret)
 139                 goto out_free_stage2_pgd;
 140
 141         kvm_timer_init(kvm);
 142
 143         /* Mark the initial VMID generation invalid */
 144         kvm->arch.vmid_gen = 0;
 145
 146         return ret;
 147 out_free_stage2_pgd:
 148         kvm_free_stage2_pgd(kvm);
 149 out_fail_alloc:
 150         return ret;
 151 }
 152
 153 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 154 {
 155         return VM_FAULT_SIGBUS;
 156 }
 157
 158
 159 /**
 160  * kvm_arch_destroy_vm - destroy the VM data structure
 161  * @kvm:        pointer to the KVM struct
 162  */
 163 void kvm_arch_destroy_vm(struct kvm *kvm)
 164 {
 165         int i;
 166
 167         kvm_free_stage2_pgd(kvm);
 168
 169         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 170                 if (kvm->vcpus[i]) {
 171                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 172                         kvm->vcpus[i] = NULL;
 173                 }
 174         }
 175 }
 176
 177 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 178 {
 179         int r;
 180         switch (ext) {
 181         case KVM_CAP_IRQCHIP:
 182                 r = vgic_present;
 183                 break;
 184         case KVM_CAP_DEVICE_CTRL:
 185         case KVM_CAP_USER_MEMORY:
 186         case KVM_CAP_SYNC_MMU:
 187         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 188         case KVM_CAP_ONE_REG:
 189         case KVM_CAP_ARM_PSCI:
 190         case KVM_CAP_ARM_PSCI_0_2:
 191                 r = 1;
 192                 break;
 193         case KVM_CAP_COALESCED_MMIO:
 194                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 195                 break;
 196         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 197                 r = 1;
 198                 break;
 199         case KVM_CAP_NR_VCPUS:
 200                 r = num_online_cpus();
 201                 break;
 202         case KVM_CAP_MAX_VCPUS:
 203                 r = KVM_MAX_VCPUS;
 204                 break;
 205         default:
 206                 r = kvm_arch_dev_ioctl_check_extension(ext);
 207                 break;
 208         }
 209         return r;
 210 }
 211
 212 long kvm_arch_dev_ioctl(struct file *filp,
 213                         unsigned int ioctl, unsigned long arg)
 214 {
 215         return -EINVAL;
 216 }
 217
 218
 219 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 220 {
 221         int err;
 222         struct kvm_vcpu *vcpu;
 223
 224         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 225         if (!vcpu) {
 226                 err = -ENOMEM;
 227                 goto out;
 228         }
 229
 230         err = kvm_vcpu_init(vcpu, kvm, id);
 231         if (err)
 232                 goto free_vcpu;
 233
 234         err = create_hyp_mappings(vcpu, vcpu + 1);
 235         if (err)
 236                 goto vcpu_uninit;
 237
 238         return vcpu;
 239 vcpu_uninit:
 240         kvm_vcpu_uninit(vcpu);
 241 free_vcpu:
 242         kmem_cache_free(kvm_vcpu_cache, vcpu);
 243 out:
 244         return ERR_PTR(err);
 245 }
 246
 247 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 248 {
 249         return 0;
 250 }
 251
 252 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 253 {
 254         kvm_mmu_free_memory_caches(vcpu);
 255         kvm_timer_vcpu_terminate(vcpu);
 256         kmem_cache_free(kvm_vcpu_cache, vcpu);
 257 }
 258
 259 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 260 {
 261         kvm_arch_vcpu_free(vcpu);
 262 }
 263
 264 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 265 {
 266         return 0;
 267 }
 268
 269 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 270 {
 271         int ret;
 272
 273         /* Force users to call KVM_ARM_VCPU_INIT */
 274         vcpu->arch.target = -1;
 275
 276         /* Set up VGIC */
 277         ret = kvm_vgic_vcpu_init(vcpu);
 278         if (ret)
 279                 return ret;
 280
 281         /* Set up the timer */
 282         kvm_timer_vcpu_init(vcpu);
 283
 284         return 0;
 285 }
 286
 287 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 288 {
 289 }
 290
 291 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
 292 {
 293 }
 294
 295 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 296 {
 297         vcpu->cpu = cpu;
 298         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 299
 300         /*
 301          * Check whether this vcpu requires the cache to be flushed on
 302          * this physical CPU. This is a consequence of doing dcache
 303          * operations by set/way on this vcpu. We do it here to be in
 304          * a non-preemptible section.
 305          */
 306         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 307                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 308
 309         kvm_arm_set_running_vcpu(vcpu);
 310 }
 311
 312 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 313 {
 314         /*
 315          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 316          * if the vcpu is no longer assigned to a cpu.  This is used for the
 317          * optimized make_all_cpus_request path.
 318          */
 319         vcpu->cpu = -1;
 320
 321         kvm_arm_set_running_vcpu(NULL);
 322 }
 323
 324 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 325                                         struct kvm_guest_debug *dbg)
 326 {
 327         return -EINVAL;
 328 }
 329
 330
 331 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 332                                     struct kvm_mp_state *mp_state)
 333 {
 334         return -EINVAL;
 335 }
 336
 337 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 338                                     struct kvm_mp_state *mp_state)
 339 {
 340         return -EINVAL;
 341 }
 342
 343 /**
 344  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 345  * @v:          The VCPU pointer
 346  *
 347  * If the guest CPU is not waiting for interrupts or an interrupt line is
 348  * asserted, the CPU is by definition runnable.
 349  */
 350 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 351 {
 352         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 353 }
 354
 355 /* Just ensure a guest exit from a particular CPU */
 356 static void exit_vm_noop(void *info)
 357 {
 358 }
 359
 360 void force_vm_exit(const cpumask_t *mask)
 361 {
 362         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 363 }
 364
 365 /**
 366  * need_new_vmid_gen - check that the VMID is still valid
 367  * @kvm: The VM's VMID to checkt
 368  *
 369  * return true if there is a new generation of VMIDs being used
 370  *
 371  * The hardware supports only 256 values with the value zero reserved for the
 372  * host, so we check if an assigned value belongs to a previous generation,
 373  * which which requires us to assign a new value. If we're the first to use a
 374  * VMID for the new generation, we must flush necessary caches and TLBs on all
 375  * CPUs.
 376  */
 377 static bool need_new_vmid_gen(struct kvm *kvm)
 378 {
 379         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 380 }
 381
 382 /**
 383  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 384  * @kvm The guest that we are about to run
 385  *
 386  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 387  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 388  * caches and TLBs.
 389  */
 390 static void update_vttbr(struct kvm *kvm)
 391 {
 392         phys_addr_t pgd_phys;
 393         u64 vmid;
 394
 395         if (!need_new_vmid_gen(kvm))
 396                 return;
 397
 398         spin_lock(&kvm_vmid_lock);
 399
 400         /*
 401          * We need to re-check the vmid_gen here to ensure that if another vcpu
 402          * already allocated a valid vmid for this vm, then this vcpu should
 403          * use the same vmid.
 404          */
 405         if (!need_new_vmid_gen(kvm)) {
 406                 spin_unlock(&kvm_vmid_lock);
 407                 return;
 408         }
 409
 410         /* First user of a new VMID generation? */
 411         if (unlikely(kvm_next_vmid == 0)) {
 412                 atomic64_inc(&kvm_vmid_gen);
 413                 kvm_next_vmid = 1;
 414
 415                 /*
 416                  * On SMP we know no other CPUs can use this CPU's or each
 417                  * other's VMID after force_vm_exit returns since the
 418                  * kvm_vmid_lock blocks them from reentry to the guest.
 419                  */
 420                 force_vm_exit(cpu_all_mask);
 421                 /*
 422                  * Now broadcast TLB + ICACHE invalidation over the inner
 423                  * shareable domain to make sure all data structures are
 424                  * clean.
 425                  */
 426                 kvm_call_hyp(__kvm_flush_vm_context);
 427         }
 428
 429         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 430         kvm->arch.vmid = kvm_next_vmid;
 431         kvm_next_vmid++;
 432
 433         /* update vttbr to be used with the new vmid */
 434         pgd_phys = virt_to_phys(kvm->arch.pgd);
 435         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 436         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 437         kvm->arch.vttbr |= vmid;
 438
 439         spin_unlock(&kvm_vmid_lock);
 440 }
 441
 442 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 443 {
 444         int ret;
 445
 446         if (likely(vcpu->arch.has_run_once))
 447                 return 0;
 448
 449         vcpu->arch.has_run_once = true;
 450
 451         /*
 452          * Initialize the VGIC before running a vcpu the first time on
 453          * this VM.
 454          */
 455         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 456                 ret = kvm_vgic_init(vcpu->kvm);
 457                 if (ret)
 458                         return ret;
 459         }
 460
 461         return 0;
 462 }
 463
 464 static void vcpu_pause(struct kvm_vcpu *vcpu)
 465 {
 466         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 467
 468         wait_event_interruptible(*wq, !vcpu->arch.pause);
 469 }
 470
 471 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 472 {
 473         return vcpu->arch.target >= 0;
 474 }
 475
 476 /**
 477  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 478  * @vcpu:       The VCPU pointer
 479  * @run:        The kvm_run structure pointer used for userspace state exchange
 480  *
 481  * This function is called through the VCPU_RUN ioctl called from user space. It
 482  * will execute VM code in a loop until the time slice for the process is used
 483  * or some emulation is needed from user space in which case the function will
 484  * return with return value 0 and with the kvm_run structure filled in with the
 485  * required data for the requested emulation.
 486  */
 487 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 488 {
 489         int ret;
 490         sigset_t sigsaved;
 491
 492         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 493                 return -ENOEXEC;
 494
 495         ret = kvm_vcpu_first_run_init(vcpu);
 496         if (ret)
 497                 return ret;
 498
 499         if (run->exit_reason == KVM_EXIT_MMIO) {
 500                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 501                 if (ret)
 502                         return ret;
 503         }
 504
 505         if (vcpu->sigset_active)
 506                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 507
 508         ret = 1;
 509         run->exit_reason = KVM_EXIT_UNKNOWN;
 510         while (ret > 0) {
 511                 /*
 512                  * Check conditions before entering the guest
 513                  */
 514                 cond_resched();
 515
 516                 update_vttbr(vcpu->kvm);
 517
 518                 if (vcpu->arch.pause)
 519                         vcpu_pause(vcpu);
 520
 521                 kvm_vgic_flush_hwstate(vcpu);
 522                 kvm_timer_flush_hwstate(vcpu);
 523
 524                 local_irq_disable();
 525
 526                 /*
 527                  * Re-check atomic conditions
 528                  */
 529                 if (signal_pending(current)) {
 530                         ret = -EINTR;
 531                         run->exit_reason = KVM_EXIT_INTR;
 532                 }
 533
 534                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 535                         local_irq_enable();
 536                         kvm_timer_sync_hwstate(vcpu);
 537                         kvm_vgic_sync_hwstate(vcpu);
 538                         continue;
 539                 }
 540
 541                 /**************************************************************
 542                  * Enter the guest
 543                  */
 544                 trace_kvm_entry(*vcpu_pc(vcpu));
 545                 kvm_guest_enter();
 546                 vcpu->mode = IN_GUEST_MODE;
 547
 548                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 549
 550                 vcpu->mode = OUTSIDE_GUEST_MODE;
 551                 vcpu->arch.last_pcpu = smp_processor_id();
 552                 kvm_guest_exit();
 553                 trace_kvm_exit(*vcpu_pc(vcpu));
 554                 /*
 555                  * We may have taken a host interrupt in HYP mode (ie
 556                  * while executing the guest). This interrupt is still
 557                  * pending, as we haven't serviced it yet!
 558                  *
 559                  * We're now back in SVC mode, with interrupts
 560                  * disabled.  Enabling the interrupts now will have
 561                  * the effect of taking the interrupt again, in SVC
 562                  * mode this time.
 563                  */
 564                 local_irq_enable();
 565
 566                 /*
 567                  * Back from guest
 568                  *************************************************************/
 569
 570                 kvm_timer_sync_hwstate(vcpu);
 571                 kvm_vgic_sync_hwstate(vcpu);
 572
 573                 ret = handle_exit(vcpu, run, ret);
 574         }
 575
 576         if (vcpu->sigset_active)
 577                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 578         return ret;
 579 }
 580
 581 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 582 {
 583         int bit_index;
 584         bool set;
 585         unsigned long *ptr;
 586
 587         if (number == KVM_ARM_IRQ_CPU_IRQ)
 588                 bit_index = __ffs(HCR_VI);
 589         else /* KVM_ARM_IRQ_CPU_FIQ */
 590                 bit_index = __ffs(HCR_VF);
 591
 592         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 593         if (level)
 594                 set = test_and_set_bit(bit_index, ptr);
 595         else
 596                 set = test_and_clear_bit(bit_index, ptr);
 597
 598         /*
 599          * If we didn't change anything, no need to wake up or kick other CPUs
 600          */
 601         if (set == level)
 602                 return 0;
 603
 604         /*
 605          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 606          * trigger a world-switch round on the running physical CPU to set the
 607          * virtual IRQ/FIQ fields in the HCR appropriately.
 608          */
 609         kvm_vcpu_kick(vcpu);
 610
 611         return 0;
 612 }
 613
 614 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 615                           bool line_status)
 616 {
 617         u32 irq = irq_level->irq;
 618         unsigned int irq_type, vcpu_idx, irq_num;
 619         int nrcpus = atomic_read(&kvm->online_vcpus);
 620         struct kvm_vcpu *vcpu = NULL;
 621         bool level = irq_level->level;
 622
 623         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 624         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 625         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 626
 627         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 628
 629         switch (irq_type) {
 630         case KVM_ARM_IRQ_TYPE_CPU:
 631                 if (irqchip_in_kernel(kvm))
 632                         return -ENXIO;
 633
 634                 if (vcpu_idx >= nrcpus)
 635                         return -EINVAL;
 636
 637                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 638                 if (!vcpu)
 639                         return -EINVAL;
 640
 641                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 642                         return -EINVAL;
 643
 644                 return vcpu_interrupt_line(vcpu, irq_num, level);
 645         case KVM_ARM_IRQ_TYPE_PPI:
 646                 if (!irqchip_in_kernel(kvm))
 647                         return -ENXIO;
 648
 649                 if (vcpu_idx >= nrcpus)
 650                         return -EINVAL;
 651
 652                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 653                 if (!vcpu)
 654                         return -EINVAL;
 655
 656                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 657                         return -EINVAL;
 658
 659                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 660         case KVM_ARM_IRQ_TYPE_SPI:
 661                 if (!irqchip_in_kernel(kvm))
 662                         return -ENXIO;
 663
 664                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 665                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 666                         return -EINVAL;
 667
 668                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 669         }
 670
 671         return -EINVAL;
 672 }
 673
 674 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 675                                          struct kvm_vcpu_init *init)
 676 {
 677         int ret;
 678
 679         ret = kvm_vcpu_set_target(vcpu, init);
 680         if (ret)
 681                 return ret;
 682
 683         /*
 684          * Handle the "start in power-off" case by marking the VCPU as paused.
 685          */
 686         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 687                 vcpu->arch.pause = true;
 688
 689         return 0;
 690 }
 691
 692 long kvm_arch_vcpu_ioctl(struct file *filp,
 693                          unsigned int ioctl, unsigned long arg)
 694 {
 695         struct kvm_vcpu *vcpu = filp->private_data;
 696         void __user *argp = (void __user *)arg;
 697
 698         switch (ioctl) {
 699         case KVM_ARM_VCPU_INIT: {
 700                 struct kvm_vcpu_init init;
 701
 702                 if (copy_from_user(&init, argp, sizeof(init)))
 703                         return -EFAULT;
 704
 705                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 706         }
 707         case KVM_SET_ONE_REG:
 708         case KVM_GET_ONE_REG: {
 709                 struct kvm_one_reg reg;
 710
 711                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 712                         return -ENOEXEC;
 713
 714                 if (copy_from_user(&reg, argp, sizeof(reg)))
 715                         return -EFAULT;
 716                 if (ioctl == KVM_SET_ONE_REG)
 717                         return kvm_arm_set_reg(vcpu, &reg);
 718                 else
 719                         return kvm_arm_get_reg(vcpu, &reg);
 720         }
 721         case KVM_GET_REG_LIST: {
 722                 struct kvm_reg_list __user *user_list = argp;
 723                 struct kvm_reg_list reg_list;
 724                 unsigned n;
 725
 726                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 727                         return -ENOEXEC;
 728
 729                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 730                         return -EFAULT;
 731                 n = reg_list.n;
 732                 reg_list.n = kvm_arm_num_regs(vcpu);
 733                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 734                         return -EFAULT;
 735                 if (n < reg_list.n)
 736                         return -E2BIG;
 737                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 738         }
 739         default:
 740                 return -EINVAL;
 741         }
 742 }
 743
 744 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 745 {
 746         return -EINVAL;
 747 }
 748
 749 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 750                                         struct kvm_arm_device_addr *dev_addr)
 751 {
 752         unsigned long dev_id, type;
 753
 754         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 755                 KVM_ARM_DEVICE_ID_SHIFT;
 756         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 757                 KVM_ARM_DEVICE_TYPE_SHIFT;
 758
 759         switch (dev_id) {
 760         case KVM_ARM_DEVICE_VGIC_V2:
 761                 if (!vgic_present)
 762                         return -ENXIO;
 763                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 764         default:
 765                 return -ENODEV;
 766         }
 767 }
 768
 769 long kvm_arch_vm_ioctl(struct file *filp,
 770                        unsigned int ioctl, unsigned long arg)
 771 {
 772         struct kvm *kvm = filp->private_data;
 773         void __user *argp = (void __user *)arg;
 774
 775         switch (ioctl) {
 776         case KVM_CREATE_IRQCHIP: {
 777                 if (vgic_present)
 778                         return kvm_vgic_create(kvm);
 779                 else
 780                         return -ENXIO;
 781         }
 782         case KVM_ARM_SET_DEVICE_ADDR: {
 783                 struct kvm_arm_device_addr dev_addr;
 784
 785                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 786                         return -EFAULT;
 787                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 788         }
 789         case KVM_ARM_PREFERRED_TARGET: {
 790                 int err;
 791                 struct kvm_vcpu_init init;
 792
 793                 err = kvm_vcpu_preferred_target(&init);
 794                 if (err)
 795                         return err;
 796
 797                 if (copy_to_user(argp, &init, sizeof(init)))
 798                         return -EFAULT;
 799
 800                 return 0;
 801         }
 802         default:
 803                 return -EINVAL;
 804         }
 805 }
 806
 807 static void cpu_init_hyp_mode(void *dummy)
 808 {
 809         phys_addr_t boot_pgd_ptr;
 810         phys_addr_t pgd_ptr;
 811         unsigned long hyp_stack_ptr;
 812         unsigned long stack_page;
 813         unsigned long vector_ptr;
 814
 815         /* Switch from the HYP stub to our own HYP init vector */
 816         __hyp_set_vectors(kvm_get_idmap_vector());
 817
 818         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 819         pgd_ptr = kvm_mmu_get_httbr();
 820         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 821         hyp_stack_ptr = stack_page + PAGE_SIZE;
 822         vector_ptr = (unsigned long)__kvm_hyp_vector;
 823
 824         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 825 }
 826
 827 static int hyp_init_cpu_notify(struct notifier_block *self,
 828                                unsigned long action, void *cpu)
 829 {
 830         switch (action) {
 831         case CPU_STARTING:
 832         case CPU_STARTING_FROZEN:
 833                 cpu_init_hyp_mode(NULL);
 834                 break;
 835         }
 836
 837         return NOTIFY_OK;
 838 }
 839
 840 static struct notifier_block hyp_init_cpu_nb = {
 841         .notifier_call = hyp_init_cpu_notify,
 842 };
 843
 844 #ifdef CONFIG_CPU_PM
 845 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 846                                     unsigned long cmd,
 847                                     void *v)
 848 {
 849         if (cmd == CPU_PM_EXIT &&
 850             __hyp_get_vectors() == hyp_default_vectors) {
 851                 cpu_init_hyp_mode(NULL);
 852                 return NOTIFY_OK;
 853         }
 854
 855         return NOTIFY_DONE;
 856 }
 857
 858 static struct notifier_block hyp_init_cpu_pm_nb = {
 859         .notifier_call = hyp_init_cpu_pm_notifier,
 860 };
 861
 862 static void __init hyp_cpu_pm_init(void)
 863 {
 864         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 865 }
 866 #else
 867 static inline void hyp_cpu_pm_init(void)
 868 {
 869 }
 870 #endif
 871
 872 /**
 873  * Inits Hyp-mode on all online CPUs
 874  */
 875 static int init_hyp_mode(void)
 876 {
 877         int cpu;
 878         int err = 0;
 879
 880         /*
 881          * Allocate Hyp PGD and setup Hyp identity mapping
 882          */
 883         err = kvm_mmu_init();
 884         if (err)
 885                 goto out_err;
 886
 887         /*
 888          * It is probably enough to obtain the default on one
 889          * CPU. It's unlikely to be different on the others.
 890          */
 891         hyp_default_vectors = __hyp_get_vectors();
 892
 893         /*
 894          * Allocate stack pages for Hypervisor-mode
 895          */
 896         for_each_possible_cpu(cpu) {
 897                 unsigned long stack_page;
 898
 899                 stack_page = __get_free_page(GFP_KERNEL);
 900                 if (!stack_page) {
 901                         err = -ENOMEM;
 902                         goto out_free_stack_pages;
 903                 }
 904
 905                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 906         }
 907
 908         /*
 909          * Map the Hyp-code called directly from the host
 910          */
 911         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 912         if (err) {
 913                 kvm_err("Cannot map world-switch code\n");
 914                 goto out_free_mappings;
 915         }
 916
 917         /*
 918          * Map the Hyp stack pages
 919          */
 920         for_each_possible_cpu(cpu) {
 921                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 922                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 923
 924                 if (err) {
 925                         kvm_err("Cannot map hyp stack\n");
 926                         goto out_free_mappings;
 927                 }
 928         }
 929
 930         /*
 931          * Map the host CPU structures
 932          */
 933         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 934         if (!kvm_host_cpu_state) {
 935                 err = -ENOMEM;
 936                 kvm_err("Cannot allocate host CPU state\n");
 937                 goto out_free_mappings;
 938         }
 939
 940         for_each_possible_cpu(cpu) {
 941                 kvm_cpu_context_t *cpu_ctxt;
 942
 943                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 944                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 945
 946                 if (err) {
 947                         kvm_err("Cannot map host CPU state: %d\n", err);
 948                         goto out_free_context;
 949                 }
 950         }
 951
 952         /*
 953          * Execute the init code on each CPU.
 954          */
 955         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 956
 957         /*
 958          * Init HYP view of VGIC
 959          */
 960         err = kvm_vgic_hyp_init();
 961         if (err)
 962                 goto out_free_context;
 963
 964 #ifdef CONFIG_KVM_ARM_VGIC
 965                 vgic_present = true;
 966 #endif
 967
 968         /*
 969          * Init HYP architected timer support
 970          */
 971         err = kvm_timer_hyp_init();
 972         if (err)
 973                 goto out_free_mappings;
 974
 975 #ifndef CONFIG_HOTPLUG_CPU
 976         free_boot_hyp_pgd();
 977 #endif
 978
 979         kvm_perf_init();
 980
 981         kvm_info("Hyp mode initialized successfully\n");
 982
 983         return 0;
 984 out_free_context:
 985         free_percpu(kvm_host_cpu_state);
 986 out_free_mappings:
 987         free_hyp_pgds();
 988 out_free_stack_pages:
 989         for_each_possible_cpu(cpu)
 990                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 991 out_err:
 992         kvm_err("error initializing Hyp mode: %d\n", err);
 993         return err;
 994 }
 995
 996 static void check_kvm_target_cpu(void *ret)
 997 {
 998         *(int *)ret = kvm_target_cpu();
 999 }
1000
1001 /**
1002  * Initialize Hyp-mode and memory mappings on all CPUs.
1003  */
1004 int kvm_arch_init(void *opaque)
1005 {
1006         int err;
1007         int ret, cpu;
1008
1009         if (!is_hyp_mode_available()) {
1010                 kvm_err("HYP mode not available\n");
1011                 return -ENODEV;
1012         }
1013
1014         for_each_online_cpu(cpu) {
1015                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1016                 if (ret < 0) {
1017                         kvm_err("Error, CPU %d not supported!\n", cpu);
1018                         return -ENODEV;
1019                 }
1020         }
1021
1022         cpu_notifier_register_begin();
1023
1024         err = init_hyp_mode();
1025         if (err)
1026                 goto out_err;
1027
1028         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1029         if (err) {
1030                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1031                 goto out_err;
1032         }
1033
1034         cpu_notifier_register_done();
1035
1036         hyp_cpu_pm_init();
1037
1038         kvm_coproc_table_init();
1039         return 0;
1040 out_err:
1041         cpu_notifier_register_done();
1042         return err;
1043 }
1044
1045 /* NOP: Compiling as a module not supported */
1046 void kvm_arch_exit(void)
1047 {
1048         kvm_perf_teardown();
1049 }
1050
1051 static int arm_init(void)
1052 {
1053         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1054         return rc;
1055 }
1056
1057 module_init(arm_init);