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)
  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         /* Force users to call KVM_ARM_VCPU_INIT */
 265         vcpu->arch.target = -1;
 266
 267         /* Set up the timer */
 268         kvm_timer_vcpu_init(vcpu);
 269
 270         return 0;
 271 }
 272
 273 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 274 {
 275         vcpu->cpu = cpu;
 276         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 277
 278         /*
 279          * Check whether this vcpu requires the cache to be flushed on
 280          * this physical CPU. This is a consequence of doing dcache
 281          * operations by set/way on this vcpu. We do it here to be in
 282          * a non-preemptible section.
 283          */
 284         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 285                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 286
 287         kvm_arm_set_running_vcpu(vcpu);
 288 }
 289
 290 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 291 {
 292         /*
 293          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 294          * if the vcpu is no longer assigned to a cpu.  This is used for the
 295          * optimized make_all_cpus_request path.
 296          */
 297         vcpu->cpu = -1;
 298
 299         kvm_arm_set_running_vcpu(NULL);
 300 }
 301
 302 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 303                                         struct kvm_guest_debug *dbg)
 304 {
 305         return -EINVAL;
 306 }
 307
 308
 309 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 310                                     struct kvm_mp_state *mp_state)
 311 {
 312         return -EINVAL;
 313 }
 314
 315 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 316                                     struct kvm_mp_state *mp_state)
 317 {
 318         return -EINVAL;
 319 }
 320
 321 /**
 322  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 323  * @v:          The VCPU pointer
 324  *
 325  * If the guest CPU is not waiting for interrupts or an interrupt line is
 326  * asserted, the CPU is by definition runnable.
 327  */
 328 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 329 {
 330         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 331 }
 332
 333 /* Just ensure a guest exit from a particular CPU */
 334 static void exit_vm_noop(void *info)
 335 {
 336 }
 337
 338 void force_vm_exit(const cpumask_t *mask)
 339 {
 340         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 341 }
 342
 343 /**
 344  * need_new_vmid_gen - check that the VMID is still valid
 345  * @kvm: The VM's VMID to checkt
 346  *
 347  * return true if there is a new generation of VMIDs being used
 348  *
 349  * The hardware supports only 256 values with the value zero reserved for the
 350  * host, so we check if an assigned value belongs to a previous generation,
 351  * which which requires us to assign a new value. If we're the first to use a
 352  * VMID for the new generation, we must flush necessary caches and TLBs on all
 353  * CPUs.
 354  */
 355 static bool need_new_vmid_gen(struct kvm *kvm)
 356 {
 357         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 358 }
 359
 360 /**
 361  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 362  * @kvm The guest that we are about to run
 363  *
 364  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 365  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 366  * caches and TLBs.
 367  */
 368 static void update_vttbr(struct kvm *kvm)
 369 {
 370         phys_addr_t pgd_phys;
 371         u64 vmid;
 372
 373         if (!need_new_vmid_gen(kvm))
 374                 return;
 375
 376         spin_lock(&kvm_vmid_lock);
 377
 378         /*
 379          * We need to re-check the vmid_gen here to ensure that if another vcpu
 380          * already allocated a valid vmid for this vm, then this vcpu should
 381          * use the same vmid.
 382          */
 383         if (!need_new_vmid_gen(kvm)) {
 384                 spin_unlock(&kvm_vmid_lock);
 385                 return;
 386         }
 387
 388         /* First user of a new VMID generation? */
 389         if (unlikely(kvm_next_vmid == 0)) {
 390                 atomic64_inc(&kvm_vmid_gen);
 391                 kvm_next_vmid = 1;
 392
 393                 /*
 394                  * On SMP we know no other CPUs can use this CPU's or each
 395                  * other's VMID after force_vm_exit returns since the
 396                  * kvm_vmid_lock blocks them from reentry to the guest.
 397                  */
 398                 force_vm_exit(cpu_all_mask);
 399                 /*
 400                  * Now broadcast TLB + ICACHE invalidation over the inner
 401                  * shareable domain to make sure all data structures are
 402                  * clean.
 403                  */
 404                 kvm_call_hyp(__kvm_flush_vm_context);
 405         }
 406
 407         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 408         kvm->arch.vmid = kvm_next_vmid;
 409         kvm_next_vmid++;
 410
 411         /* update vttbr to be used with the new vmid */
 412         pgd_phys = virt_to_phys(kvm->arch.pgd);
 413         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 414         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 415         kvm->arch.vttbr = pgd_phys | vmid;
 416
 417         spin_unlock(&kvm_vmid_lock);
 418 }
 419
 420 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 421 {
 422         int ret;
 423
 424         if (likely(vcpu->arch.has_run_once))
 425                 return 0;
 426
 427         vcpu->arch.has_run_once = true;
 428
 429         /*
 430          * Initialize the VGIC before running a vcpu the first time on
 431          * this VM.
 432          */
 433         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 434                 ret = kvm_vgic_init(vcpu->kvm);
 435                 if (ret)
 436                         return ret;
 437         }
 438
 439         return 0;
 440 }
 441
 442 static void vcpu_pause(struct kvm_vcpu *vcpu)
 443 {
 444         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 445
 446         wait_event_interruptible(*wq, !vcpu->arch.pause);
 447 }
 448
 449 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 450 {
 451         return vcpu->arch.target >= 0;
 452 }
 453
 454 /**
 455  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 456  * @vcpu:       The VCPU pointer
 457  * @run:        The kvm_run structure pointer used for userspace state exchange
 458  *
 459  * This function is called through the VCPU_RUN ioctl called from user space. It
 460  * will execute VM code in a loop until the time slice for the process is used
 461  * or some emulation is needed from user space in which case the function will
 462  * return with return value 0 and with the kvm_run structure filled in with the
 463  * required data for the requested emulation.
 464  */
 465 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 466 {
 467         int ret;
 468         sigset_t sigsaved;
 469
 470         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 471                 return -ENOEXEC;
 472
 473         ret = kvm_vcpu_first_run_init(vcpu);
 474         if (ret)
 475                 return ret;
 476
 477         if (run->exit_reason == KVM_EXIT_MMIO) {
 478                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 479                 if (ret)
 480                         return ret;
 481         }
 482
 483         if (vcpu->sigset_active)
 484                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 485
 486         ret = 1;
 487         run->exit_reason = KVM_EXIT_UNKNOWN;
 488         while (ret > 0) {
 489                 /*
 490                  * Check conditions before entering the guest
 491                  */
 492                 cond_resched();
 493
 494                 update_vttbr(vcpu->kvm);
 495
 496                 if (vcpu->arch.pause)
 497                         vcpu_pause(vcpu);
 498
 499                 kvm_vgic_flush_hwstate(vcpu);
 500                 kvm_timer_flush_hwstate(vcpu);
 501
 502                 local_irq_disable();
 503
 504                 /*
 505                  * Re-check atomic conditions
 506                  */
 507                 if (signal_pending(current)) {
 508                         ret = -EINTR;
 509                         run->exit_reason = KVM_EXIT_INTR;
 510                 }
 511
 512                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 513                         local_irq_enable();
 514                         kvm_timer_sync_hwstate(vcpu);
 515                         kvm_vgic_sync_hwstate(vcpu);
 516                         continue;
 517                 }
 518
 519                 /**************************************************************
 520                  * Enter the guest
 521                  */
 522                 trace_kvm_entry(*vcpu_pc(vcpu));
 523                 kvm_guest_enter();
 524                 vcpu->mode = IN_GUEST_MODE;
 525
 526                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 527
 528                 vcpu->mode = OUTSIDE_GUEST_MODE;
 529                 vcpu->arch.last_pcpu = smp_processor_id();
 530                 kvm_guest_exit();
 531                 trace_kvm_exit(*vcpu_pc(vcpu));
 532                 /*
 533                  * We may have taken a host interrupt in HYP mode (ie
 534                  * while executing the guest). This interrupt is still
 535                  * pending, as we haven't serviced it yet!
 536                  *
 537                  * We're now back in SVC mode, with interrupts
 538                  * disabled.  Enabling the interrupts now will have
 539                  * the effect of taking the interrupt again, in SVC
 540                  * mode this time.
 541                  */
 542                 local_irq_enable();
 543
 544                 /*
 545                  * Back from guest
 546                  *************************************************************/
 547
 548                 kvm_timer_sync_hwstate(vcpu);
 549                 kvm_vgic_sync_hwstate(vcpu);
 550
 551                 ret = handle_exit(vcpu, run, ret);
 552         }
 553
 554         if (vcpu->sigset_active)
 555                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 556         return ret;
 557 }
 558
 559 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 560 {
 561         int bit_index;
 562         bool set;
 563         unsigned long *ptr;
 564
 565         if (number == KVM_ARM_IRQ_CPU_IRQ)
 566                 bit_index = __ffs(HCR_VI);
 567         else /* KVM_ARM_IRQ_CPU_FIQ */
 568                 bit_index = __ffs(HCR_VF);
 569
 570         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 571         if (level)
 572                 set = test_and_set_bit(bit_index, ptr);
 573         else
 574                 set = test_and_clear_bit(bit_index, ptr);
 575
 576         /*
 577          * If we didn't change anything, no need to wake up or kick other CPUs
 578          */
 579         if (set == level)
 580                 return 0;
 581
 582         /*
 583          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 584          * trigger a world-switch round on the running physical CPU to set the
 585          * virtual IRQ/FIQ fields in the HCR appropriately.
 586          */
 587         kvm_vcpu_kick(vcpu);
 588
 589         return 0;
 590 }
 591
 592 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 593                           bool line_status)
 594 {
 595         u32 irq = irq_level->irq;
 596         unsigned int irq_type, vcpu_idx, irq_num;
 597         int nrcpus = atomic_read(&kvm->online_vcpus);
 598         struct kvm_vcpu *vcpu = NULL;
 599         bool level = irq_level->level;
 600
 601         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 602         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 603         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 604
 605         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 606
 607         switch (irq_type) {
 608         case KVM_ARM_IRQ_TYPE_CPU:
 609                 if (irqchip_in_kernel(kvm))
 610                         return -ENXIO;
 611
 612                 if (vcpu_idx >= nrcpus)
 613                         return -EINVAL;
 614
 615                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 616                 if (!vcpu)
 617                         return -EINVAL;
 618
 619                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 620                         return -EINVAL;
 621
 622                 return vcpu_interrupt_line(vcpu, irq_num, level);
 623         case KVM_ARM_IRQ_TYPE_PPI:
 624                 if (!irqchip_in_kernel(kvm))
 625                         return -ENXIO;
 626
 627                 if (vcpu_idx >= nrcpus)
 628                         return -EINVAL;
 629
 630                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 631                 if (!vcpu)
 632                         return -EINVAL;
 633
 634                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 635                         return -EINVAL;
 636
 637                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 638         case KVM_ARM_IRQ_TYPE_SPI:
 639                 if (!irqchip_in_kernel(kvm))
 640                         return -ENXIO;
 641
 642                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 643                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 644                         return -EINVAL;
 645
 646                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 647         }
 648
 649         return -EINVAL;
 650 }
 651
 652 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 653                                          struct kvm_vcpu_init *init)
 654 {
 655         int ret;
 656
 657         ret = kvm_vcpu_set_target(vcpu, init);
 658         if (ret)
 659                 return ret;
 660
 661         /*
 662          * Handle the "start in power-off" case by marking the VCPU as paused.
 663          */
 664         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 665                 vcpu->arch.pause = true;
 666
 667         return 0;
 668 }
 669
 670 long kvm_arch_vcpu_ioctl(struct file *filp,
 671                          unsigned int ioctl, unsigned long arg)
 672 {
 673         struct kvm_vcpu *vcpu = filp->private_data;
 674         void __user *argp = (void __user *)arg;
 675
 676         switch (ioctl) {
 677         case KVM_ARM_VCPU_INIT: {
 678                 struct kvm_vcpu_init init;
 679
 680                 if (copy_from_user(&init, argp, sizeof(init)))
 681                         return -EFAULT;
 682
 683                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 684         }
 685         case KVM_SET_ONE_REG:
 686         case KVM_GET_ONE_REG: {
 687                 struct kvm_one_reg reg;
 688
 689                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 690                         return -ENOEXEC;
 691
 692                 if (copy_from_user(&reg, argp, sizeof(reg)))
 693                         return -EFAULT;
 694                 if (ioctl == KVM_SET_ONE_REG)
 695                         return kvm_arm_set_reg(vcpu, &reg);
 696                 else
 697                         return kvm_arm_get_reg(vcpu, &reg);
 698         }
 699         case KVM_GET_REG_LIST: {
 700                 struct kvm_reg_list __user *user_list = argp;
 701                 struct kvm_reg_list reg_list;
 702                 unsigned n;
 703
 704                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 705                         return -ENOEXEC;
 706
 707                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 708                         return -EFAULT;
 709                 n = reg_list.n;
 710                 reg_list.n = kvm_arm_num_regs(vcpu);
 711                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 712                         return -EFAULT;
 713                 if (n < reg_list.n)
 714                         return -E2BIG;
 715                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 716         }
 717         default:
 718                 return -EINVAL;
 719         }
 720 }
 721
 722 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 723 {
 724         return -EINVAL;
 725 }
 726
 727 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 728                                         struct kvm_arm_device_addr *dev_addr)
 729 {
 730         unsigned long dev_id, type;
 731
 732         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 733                 KVM_ARM_DEVICE_ID_SHIFT;
 734         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 735                 KVM_ARM_DEVICE_TYPE_SHIFT;
 736
 737         switch (dev_id) {
 738         case KVM_ARM_DEVICE_VGIC_V2:
 739                 if (!vgic_present)
 740                         return -ENXIO;
 741                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 742         default:
 743                 return -ENODEV;
 744         }
 745 }
 746
 747 long kvm_arch_vm_ioctl(struct file *filp,
 748                        unsigned int ioctl, unsigned long arg)
 749 {
 750         struct kvm *kvm = filp->private_data;
 751         void __user *argp = (void __user *)arg;
 752
 753         switch (ioctl) {
 754         case KVM_CREATE_IRQCHIP: {
 755                 if (vgic_present)
 756                         return kvm_vgic_create(kvm);
 757                 else
 758                         return -ENXIO;
 759         }
 760         case KVM_ARM_SET_DEVICE_ADDR: {
 761                 struct kvm_arm_device_addr dev_addr;
 762
 763                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 764                         return -EFAULT;
 765                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 766         }
 767         case KVM_ARM_PREFERRED_TARGET: {
 768                 int err;
 769                 struct kvm_vcpu_init init;
 770
 771                 err = kvm_vcpu_preferred_target(&init);
 772                 if (err)
 773                         return err;
 774
 775                 if (copy_to_user(argp, &init, sizeof(init)))
 776                         return -EFAULT;
 777
 778                 return 0;
 779         }
 780         default:
 781                 return -EINVAL;
 782         }
 783 }
 784
 785 static void cpu_init_hyp_mode(void *dummy)
 786 {
 787         phys_addr_t boot_pgd_ptr;
 788         phys_addr_t pgd_ptr;
 789         unsigned long hyp_stack_ptr;
 790         unsigned long stack_page;
 791         unsigned long vector_ptr;
 792
 793         /* Switch from the HYP stub to our own HYP init vector */
 794         __hyp_set_vectors(kvm_get_idmap_vector());
 795
 796         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 797         pgd_ptr = kvm_mmu_get_httbr();
 798         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 799         hyp_stack_ptr = stack_page + PAGE_SIZE;
 800         vector_ptr = (unsigned long)__kvm_hyp_vector;
 801
 802         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 803 }
 804
 805 static int hyp_init_cpu_notify(struct notifier_block *self,
 806                                unsigned long action, void *cpu)
 807 {
 808         switch (action) {
 809         case CPU_STARTING:
 810         case CPU_STARTING_FROZEN:
 811                 cpu_init_hyp_mode(NULL);
 812                 break;
 813         }
 814
 815         return NOTIFY_OK;
 816 }
 817
 818 static struct notifier_block hyp_init_cpu_nb = {
 819         .notifier_call = hyp_init_cpu_notify,
 820 };
 821
 822 #ifdef CONFIG_CPU_PM
 823 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 824                                     unsigned long cmd,
 825                                     void *v)
 826 {
 827         if (cmd == CPU_PM_EXIT &&
 828             __hyp_get_vectors() == hyp_default_vectors) {
 829                 cpu_init_hyp_mode(NULL);
 830                 return NOTIFY_OK;
 831         }
 832
 833         return NOTIFY_DONE;
 834 }
 835
 836 static struct notifier_block hyp_init_cpu_pm_nb = {
 837         .notifier_call = hyp_init_cpu_pm_notifier,
 838 };
 839
 840 static void __init hyp_cpu_pm_init(void)
 841 {
 842         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 843 }
 844 #else
 845 static inline void hyp_cpu_pm_init(void)
 846 {
 847 }
 848 #endif
 849
 850 /**
 851  * Inits Hyp-mode on all online CPUs
 852  */
 853 static int init_hyp_mode(void)
 854 {
 855         int cpu;
 856         int err = 0;
 857
 858         /*
 859          * Allocate Hyp PGD and setup Hyp identity mapping
 860          */
 861         err = kvm_mmu_init();
 862         if (err)
 863                 goto out_err;
 864
 865         /*
 866          * It is probably enough to obtain the default on one
 867          * CPU. It's unlikely to be different on the others.
 868          */
 869         hyp_default_vectors = __hyp_get_vectors();
 870
 871         /*
 872          * Allocate stack pages for Hypervisor-mode
 873          */
 874         for_each_possible_cpu(cpu) {
 875                 unsigned long stack_page;
 876
 877                 stack_page = __get_free_page(GFP_KERNEL);
 878                 if (!stack_page) {
 879                         err = -ENOMEM;
 880                         goto out_free_stack_pages;
 881                 }
 882
 883                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 884         }
 885
 886         /*
 887          * Map the Hyp-code called directly from the host
 888          */
 889         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 890         if (err) {
 891                 kvm_err("Cannot map world-switch code\n");
 892                 goto out_free_mappings;
 893         }
 894
 895         /*
 896          * Map the Hyp stack pages
 897          */
 898         for_each_possible_cpu(cpu) {
 899                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 900                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 901
 902                 if (err) {
 903                         kvm_err("Cannot map hyp stack\n");
 904                         goto out_free_mappings;
 905                 }
 906         }
 907
 908         /*
 909          * Map the host CPU structures
 910          */
 911         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 912         if (!kvm_host_cpu_state) {
 913                 err = -ENOMEM;
 914                 kvm_err("Cannot allocate host CPU state\n");
 915                 goto out_free_mappings;
 916         }
 917
 918         for_each_possible_cpu(cpu) {
 919                 kvm_cpu_context_t *cpu_ctxt;
 920
 921                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 922                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 923
 924                 if (err) {
 925                         kvm_err("Cannot map host CPU state: %d\n", err);
 926                         goto out_free_context;
 927                 }
 928         }
 929
 930         /*
 931          * Execute the init code on each CPU.
 932          */
 933         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 934
 935         /*
 936          * Init HYP view of VGIC
 937          */
 938         err = kvm_vgic_hyp_init();
 939         if (err)
 940                 goto out_free_context;
 941
 942 #ifdef CONFIG_KVM_ARM_VGIC
 943                 vgic_present = true;
 944 #endif
 945
 946         /*
 947          * Init HYP architected timer support
 948          */
 949         err = kvm_timer_hyp_init();
 950         if (err)
 951                 goto out_free_mappings;
 952
 953 #ifndef CONFIG_HOTPLUG_CPU
 954         free_boot_hyp_pgd();
 955 #endif
 956
 957         kvm_perf_init();
 958
 959         kvm_info("Hyp mode initialized successfully\n");
 960
 961         return 0;
 962 out_free_context:
 963         free_percpu(kvm_host_cpu_state);
 964 out_free_mappings:
 965         free_hyp_pgds();
 966 out_free_stack_pages:
 967         for_each_possible_cpu(cpu)
 968                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 969 out_err:
 970         kvm_err("error initializing Hyp mode: %d\n", err);
 971         return err;
 972 }
 973
 974 static void check_kvm_target_cpu(void *ret)
 975 {
 976         *(int *)ret = kvm_target_cpu();
 977 }
 978
 979 /**
 980  * Initialize Hyp-mode and memory mappings on all CPUs.
 981  */
 982 int kvm_arch_init(void *opaque)
 983 {
 984         int err;
 985         int ret, cpu;
 986
 987         if (!is_hyp_mode_available()) {
 988                 kvm_err("HYP mode not available\n");
 989                 return -ENODEV;
 990         }
 991
 992         for_each_online_cpu(cpu) {
 993                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
 994                 if (ret < 0) {
 995                         kvm_err("Error, CPU %d not supported!\n", cpu);
 996                         return -ENODEV;
 997                 }
 998         }
 999
1000         cpu_notifier_register_begin();
1001
1002         err = init_hyp_mode();
1003         if (err)
1004                 goto out_err;
1005
1006         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1007         if (err) {
1008                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1009                 goto out_err;
1010         }
1011
1012         cpu_notifier_register_done();
1013
1014         hyp_cpu_pm_init();
1015
1016         kvm_coproc_table_init();
1017         return 0;
1018 out_err:
1019         cpu_notifier_register_done();
1020         return err;
1021 }
1022
1023 /* NOP: Compiling as a module not supported */
1024 void kvm_arch_exit(void)
1025 {
1026         kvm_perf_teardown();
1027 }
1028
1029 static int arm_init(void)
1030 {
1031         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1032         return rc;
1033 }
1034
1035 module_init(arm_init);