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