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         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 267
 268         /* Set up the timer */
 269         kvm_timer_vcpu_init(vcpu);
 270
 271         return 0;
 272 }
 273
 274 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 275 {
 276         vcpu->cpu = cpu;
 277         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 278
 279         /*
 280          * Check whether this vcpu requires the cache to be flushed on
 281          * this physical CPU. This is a consequence of doing dcache
 282          * operations by set/way on this vcpu. We do it here to be in
 283          * a non-preemptible section.
 284          */
 285         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 286                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 287
 288         kvm_arm_set_running_vcpu(vcpu);
 289 }
 290
 291 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 292 {
 293         /*
 294          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 295          * if the vcpu is no longer assigned to a cpu.  This is used for the
 296          * optimized make_all_cpus_request path.
 297          */
 298         vcpu->cpu = -1;
 299
 300         kvm_arm_set_running_vcpu(NULL);
 301 }
 302
 303 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 304                                         struct kvm_guest_debug *dbg)
 305 {
 306         return -EINVAL;
 307 }
 308
 309
 310 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 311                                     struct kvm_mp_state *mp_state)
 312 {
 313         return -EINVAL;
 314 }
 315
 316 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 317                                     struct kvm_mp_state *mp_state)
 318 {
 319         return -EINVAL;
 320 }
 321
 322 /**
 323  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 324  * @v:          The VCPU pointer
 325  *
 326  * If the guest CPU is not waiting for interrupts or an interrupt line is
 327  * asserted, the CPU is by definition runnable.
 328  */
 329 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 330 {
 331         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 332 }
 333
 334 /* Just ensure a guest exit from a particular CPU */
 335 static void exit_vm_noop(void *info)
 336 {
 337 }
 338
 339 void force_vm_exit(const cpumask_t *mask)
 340 {
 341         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 342 }
 343
 344 /**
 345  * need_new_vmid_gen - check that the VMID is still valid
 346  * @kvm: The VM's VMID to checkt
 347  *
 348  * return true if there is a new generation of VMIDs being used
 349  *
 350  * The hardware supports only 256 values with the value zero reserved for the
 351  * host, so we check if an assigned value belongs to a previous generation,
 352  * which which requires us to assign a new value. If we're the first to use a
 353  * VMID for the new generation, we must flush necessary caches and TLBs on all
 354  * CPUs.
 355  */
 356 static bool need_new_vmid_gen(struct kvm *kvm)
 357 {
 358         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 359 }
 360
 361 /**
 362  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 363  * @kvm The guest that we are about to run
 364  *
 365  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 366  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 367  * caches and TLBs.
 368  */
 369 static void update_vttbr(struct kvm *kvm)
 370 {
 371         phys_addr_t pgd_phys;
 372         u64 vmid;
 373
 374         if (!need_new_vmid_gen(kvm))
 375                 return;
 376
 377         spin_lock(&kvm_vmid_lock);
 378
 379         /*
 380          * We need to re-check the vmid_gen here to ensure that if another vcpu
 381          * already allocated a valid vmid for this vm, then this vcpu should
 382          * use the same vmid.
 383          */
 384         if (!need_new_vmid_gen(kvm)) {
 385                 spin_unlock(&kvm_vmid_lock);
 386                 return;
 387         }
 388
 389         /* First user of a new VMID generation? */
 390         if (unlikely(kvm_next_vmid == 0)) {
 391                 atomic64_inc(&kvm_vmid_gen);
 392                 kvm_next_vmid = 1;
 393
 394                 /*
 395                  * On SMP we know no other CPUs can use this CPU's or each
 396                  * other's VMID after force_vm_exit returns since the
 397                  * kvm_vmid_lock blocks them from reentry to the guest.
 398                  */
 399                 force_vm_exit(cpu_all_mask);
 400                 /*
 401                  * Now broadcast TLB + ICACHE invalidation over the inner
 402                  * shareable domain to make sure all data structures are
 403                  * clean.
 404                  */
 405                 kvm_call_hyp(__kvm_flush_vm_context);
 406         }
 407
 408         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 409         kvm->arch.vmid = kvm_next_vmid;
 410         kvm_next_vmid++;
 411
 412         /* update vttbr to be used with the new vmid */
 413         pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
 414         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 415         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 416         kvm->arch.vttbr = pgd_phys | vmid;
 417
 418         spin_unlock(&kvm_vmid_lock);
 419 }
 420
 421 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 422 {
 423         int ret;
 424
 425         if (likely(vcpu->arch.has_run_once))
 426                 return 0;
 427
 428         vcpu->arch.has_run_once = true;
 429
 430         /*
 431          * Initialize the VGIC before running a vcpu the first time on
 432          * this VM.
 433          */
 434         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 435                 ret = kvm_vgic_init(vcpu->kvm);
 436                 if (ret)
 437                         return ret;
 438         }
 439
 440         return 0;
 441 }
 442
 443 static void vcpu_pause(struct kvm_vcpu *vcpu)
 444 {
 445         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 446
 447         wait_event_interruptible(*wq, !vcpu->arch.pause);
 448 }
 449
 450 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 451 {
 452         return vcpu->arch.target >= 0;
 453 }
 454
 455 /**
 456  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 457  * @vcpu:       The VCPU pointer
 458  * @run:        The kvm_run structure pointer used for userspace state exchange
 459  *
 460  * This function is called through the VCPU_RUN ioctl called from user space. It
 461  * will execute VM code in a loop until the time slice for the process is used
 462  * or some emulation is needed from user space in which case the function will
 463  * return with return value 0 and with the kvm_run structure filled in with the
 464  * required data for the requested emulation.
 465  */
 466 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 467 {
 468         int ret;
 469         sigset_t sigsaved;
 470
 471         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 472                 return -ENOEXEC;
 473
 474         ret = kvm_vcpu_first_run_init(vcpu);
 475         if (ret)
 476                 return ret;
 477
 478         if (run->exit_reason == KVM_EXIT_MMIO) {
 479                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 480                 if (ret)
 481                         return ret;
 482         }
 483
 484         if (vcpu->sigset_active)
 485                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 486
 487         ret = 1;
 488         run->exit_reason = KVM_EXIT_UNKNOWN;
 489         while (ret > 0) {
 490                 /*
 491                  * Check conditions before entering the guest
 492                  */
 493                 cond_resched();
 494
 495                 update_vttbr(vcpu->kvm);
 496
 497                 if (vcpu->arch.pause)
 498                         vcpu_pause(vcpu);
 499
 500                 kvm_vgic_flush_hwstate(vcpu);
 501                 kvm_timer_flush_hwstate(vcpu);
 502
 503                 local_irq_disable();
 504
 505                 /*
 506                  * Re-check atomic conditions
 507                  */
 508                 if (signal_pending(current)) {
 509                         ret = -EINTR;
 510                         run->exit_reason = KVM_EXIT_INTR;
 511                 }
 512
 513                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 514                         local_irq_enable();
 515                         kvm_timer_sync_hwstate(vcpu);
 516                         kvm_vgic_sync_hwstate(vcpu);
 517                         continue;
 518                 }
 519
 520                 /**************************************************************
 521                  * Enter the guest
 522                  */
 523                 trace_kvm_entry(*vcpu_pc(vcpu));
 524                 kvm_guest_enter();
 525                 vcpu->mode = IN_GUEST_MODE;
 526
 527                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 528
 529                 vcpu->mode = OUTSIDE_GUEST_MODE;
 530                 vcpu->arch.last_pcpu = smp_processor_id();
 531                 kvm_guest_exit();
 532                 trace_kvm_exit(*vcpu_pc(vcpu));
 533                 /*
 534                  * We may have taken a host interrupt in HYP mode (ie
 535                  * while executing the guest). This interrupt is still
 536                  * pending, as we haven't serviced it yet!
 537                  *
 538                  * We're now back in SVC mode, with interrupts
 539                  * disabled.  Enabling the interrupts now will have
 540                  * the effect of taking the interrupt again, in SVC
 541                  * mode this time.
 542                  */
 543                 local_irq_enable();
 544
 545                 /*
 546                  * Back from guest
 547                  *************************************************************/
 548
 549                 kvm_timer_sync_hwstate(vcpu);
 550                 kvm_vgic_sync_hwstate(vcpu);
 551
 552                 ret = handle_exit(vcpu, run, ret);
 553         }
 554
 555         if (vcpu->sigset_active)
 556                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 557         return ret;
 558 }
 559
 560 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 561 {
 562         int bit_index;
 563         bool set;
 564         unsigned long *ptr;
 565
 566         if (number == KVM_ARM_IRQ_CPU_IRQ)
 567                 bit_index = __ffs(HCR_VI);
 568         else /* KVM_ARM_IRQ_CPU_FIQ */
 569                 bit_index = __ffs(HCR_VF);
 570
 571         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 572         if (level)
 573                 set = test_and_set_bit(bit_index, ptr);
 574         else
 575                 set = test_and_clear_bit(bit_index, ptr);
 576
 577         /*
 578          * If we didn't change anything, no need to wake up or kick other CPUs
 579          */
 580         if (set == level)
 581                 return 0;
 582
 583         /*
 584          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 585          * trigger a world-switch round on the running physical CPU to set the
 586          * virtual IRQ/FIQ fields in the HCR appropriately.
 587          */
 588         kvm_vcpu_kick(vcpu);
 589
 590         return 0;
 591 }
 592
 593 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 594                           bool line_status)
 595 {
 596         u32 irq = irq_level->irq;
 597         unsigned int irq_type, vcpu_idx, irq_num;
 598         int nrcpus = atomic_read(&kvm->online_vcpus);
 599         struct kvm_vcpu *vcpu = NULL;
 600         bool level = irq_level->level;
 601
 602         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 603         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 604         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 605
 606         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 607
 608         switch (irq_type) {
 609         case KVM_ARM_IRQ_TYPE_CPU:
 610                 if (irqchip_in_kernel(kvm))
 611                         return -ENXIO;
 612
 613                 if (vcpu_idx >= nrcpus)
 614                         return -EINVAL;
 615
 616                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 617                 if (!vcpu)
 618                         return -EINVAL;
 619
 620                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 621                         return -EINVAL;
 622
 623                 return vcpu_interrupt_line(vcpu, irq_num, level);
 624         case KVM_ARM_IRQ_TYPE_PPI:
 625                 if (!irqchip_in_kernel(kvm))
 626                         return -ENXIO;
 627
 628                 if (vcpu_idx >= nrcpus)
 629                         return -EINVAL;
 630
 631                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 632                 if (!vcpu)
 633                         return -EINVAL;
 634
 635                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 636                         return -EINVAL;
 637
 638                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 639         case KVM_ARM_IRQ_TYPE_SPI:
 640                 if (!irqchip_in_kernel(kvm))
 641                         return -ENXIO;
 642
 643                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 644                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 645                         return -EINVAL;
 646
 647                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 648         }
 649
 650         return -EINVAL;
 651 }
 652
 653 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 654                                const struct kvm_vcpu_init *init)
 655 {
 656         unsigned int i;
 657         int phys_target = kvm_target_cpu();
 658
 659         if (init->target != phys_target)
 660                 return -EINVAL;
 661
 662         /*
 663          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 664          * use the same target.
 665          */
 666         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 667                 return -EINVAL;
 668
 669         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 670         for (i = 0; i < sizeof(init->features) * 8; i++) {
 671                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 672
 673                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 674                         return -ENOENT;
 675
 676                 /*
 677                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 678                  * use the same feature set.
 679                  */
 680                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 681                     test_bit(i, vcpu->arch.features) != set)
 682                         return -EINVAL;
 683
 684                 if (set)
 685                         set_bit(i, vcpu->arch.features);
 686         }
 687
 688         vcpu->arch.target = phys_target;
 689
 690         /* Now we know what it is, we can reset it. */
 691         return kvm_reset_vcpu(vcpu);
 692 }
 693
 694
 695 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 696                                          struct kvm_vcpu_init *init)
 697 {
 698         int ret;
 699
 700         ret = kvm_vcpu_set_target(vcpu, init);
 701         if (ret)
 702                 return ret;
 703
 704         /*
 705          * Ensure a rebooted VM will fault in RAM pages and detect if the
 706          * guest MMU is turned off and flush the caches as needed.
 707          */
 708         if (vcpu->arch.has_run_once)
 709                 stage2_unmap_vm(vcpu->kvm);
 710
 711         vcpu_reset_hcr(vcpu);
 712
 713         /*
 714          * Handle the "start in power-off" case by marking the VCPU as paused.
 715          */
 716         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 717                 vcpu->arch.pause = true;
 718         else
 719                 vcpu->arch.pause = false;
 720
 721         return 0;
 722 }
 723
 724 long kvm_arch_vcpu_ioctl(struct file *filp,
 725                          unsigned int ioctl, unsigned long arg)
 726 {
 727         struct kvm_vcpu *vcpu = filp->private_data;
 728         void __user *argp = (void __user *)arg;
 729
 730         switch (ioctl) {
 731         case KVM_ARM_VCPU_INIT: {
 732                 struct kvm_vcpu_init init;
 733
 734                 if (copy_from_user(&init, argp, sizeof(init)))
 735                         return -EFAULT;
 736
 737                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 738         }
 739         case KVM_SET_ONE_REG:
 740         case KVM_GET_ONE_REG: {
 741                 struct kvm_one_reg reg;
 742
 743                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 744                         return -ENOEXEC;
 745
 746                 if (copy_from_user(&reg, argp, sizeof(reg)))
 747                         return -EFAULT;
 748                 if (ioctl == KVM_SET_ONE_REG)
 749                         return kvm_arm_set_reg(vcpu, &reg);
 750                 else
 751                         return kvm_arm_get_reg(vcpu, &reg);
 752         }
 753         case KVM_GET_REG_LIST: {
 754                 struct kvm_reg_list __user *user_list = argp;
 755                 struct kvm_reg_list reg_list;
 756                 unsigned n;
 757
 758                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 759                         return -ENOEXEC;
 760
 761                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 762                         return -EFAULT;
 763                 n = reg_list.n;
 764                 reg_list.n = kvm_arm_num_regs(vcpu);
 765                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 766                         return -EFAULT;
 767                 if (n < reg_list.n)
 768                         return -E2BIG;
 769                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 770         }
 771         default:
 772                 return -EINVAL;
 773         }
 774 }
 775
 776 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 777 {
 778         return -EINVAL;
 779 }
 780
 781 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 782                                         struct kvm_arm_device_addr *dev_addr)
 783 {
 784         unsigned long dev_id, type;
 785
 786         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 787                 KVM_ARM_DEVICE_ID_SHIFT;
 788         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 789                 KVM_ARM_DEVICE_TYPE_SHIFT;
 790
 791         switch (dev_id) {
 792         case KVM_ARM_DEVICE_VGIC_V2:
 793                 if (!vgic_present)
 794                         return -ENXIO;
 795                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 796         default:
 797                 return -ENODEV;
 798         }
 799 }
 800
 801 long kvm_arch_vm_ioctl(struct file *filp,
 802                        unsigned int ioctl, unsigned long arg)
 803 {
 804         struct kvm *kvm = filp->private_data;
 805         void __user *argp = (void __user *)arg;
 806
 807         switch (ioctl) {
 808         case KVM_CREATE_IRQCHIP: {
 809                 if (vgic_present)
 810                         return kvm_vgic_create(kvm);
 811                 else
 812                         return -ENXIO;
 813         }
 814         case KVM_ARM_SET_DEVICE_ADDR: {
 815                 struct kvm_arm_device_addr dev_addr;
 816
 817                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 818                         return -EFAULT;
 819                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 820         }
 821         case KVM_ARM_PREFERRED_TARGET: {
 822                 int err;
 823                 struct kvm_vcpu_init init;
 824
 825                 err = kvm_vcpu_preferred_target(&init);
 826                 if (err)
 827                         return err;
 828
 829                 if (copy_to_user(argp, &init, sizeof(init)))
 830                         return -EFAULT;
 831
 832                 return 0;
 833         }
 834         default:
 835                 return -EINVAL;
 836         }
 837 }
 838
 839 static void cpu_init_hyp_mode(void *dummy)
 840 {
 841         phys_addr_t boot_pgd_ptr;
 842         phys_addr_t pgd_ptr;
 843         unsigned long hyp_stack_ptr;
 844         unsigned long stack_page;
 845         unsigned long vector_ptr;
 846
 847         /* Switch from the HYP stub to our own HYP init vector */
 848         __hyp_set_vectors(kvm_get_idmap_vector());
 849
 850         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 851         pgd_ptr = kvm_mmu_get_httbr();
 852         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 853         hyp_stack_ptr = stack_page + PAGE_SIZE;
 854         vector_ptr = (unsigned long)__kvm_hyp_vector;
 855
 856         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 857 }
 858
 859 static int hyp_init_cpu_notify(struct notifier_block *self,
 860                                unsigned long action, void *cpu)
 861 {
 862         switch (action) {
 863         case CPU_STARTING:
 864         case CPU_STARTING_FROZEN:
 865                 if (__hyp_get_vectors() == hyp_default_vectors)
 866                         cpu_init_hyp_mode(NULL);
 867                 break;
 868         }
 869
 870         return NOTIFY_OK;
 871 }
 872
 873 static struct notifier_block hyp_init_cpu_nb = {
 874         .notifier_call = hyp_init_cpu_notify,
 875 };
 876
 877 #ifdef CONFIG_CPU_PM
 878 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 879                                     unsigned long cmd,
 880                                     void *v)
 881 {
 882         if (cmd == CPU_PM_EXIT &&
 883             __hyp_get_vectors() == hyp_default_vectors) {
 884                 cpu_init_hyp_mode(NULL);
 885                 return NOTIFY_OK;
 886         }
 887
 888         return NOTIFY_DONE;
 889 }
 890
 891 static struct notifier_block hyp_init_cpu_pm_nb = {
 892         .notifier_call = hyp_init_cpu_pm_notifier,
 893 };
 894
 895 static void __init hyp_cpu_pm_init(void)
 896 {
 897         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 898 }
 899 #else
 900 static inline void hyp_cpu_pm_init(void)
 901 {
 902 }
 903 #endif
 904
 905 /**
 906  * Inits Hyp-mode on all online CPUs
 907  */
 908 static int init_hyp_mode(void)
 909 {
 910         int cpu;
 911         int err = 0;
 912
 913         /*
 914          * Allocate Hyp PGD and setup Hyp identity mapping
 915          */
 916         err = kvm_mmu_init();
 917         if (err)
 918                 goto out_err;
 919
 920         /*
 921          * It is probably enough to obtain the default on one
 922          * CPU. It's unlikely to be different on the others.
 923          */
 924         hyp_default_vectors = __hyp_get_vectors();
 925
 926         /*
 927          * Allocate stack pages for Hypervisor-mode
 928          */
 929         for_each_possible_cpu(cpu) {
 930                 unsigned long stack_page;
 931
 932                 stack_page = __get_free_page(GFP_KERNEL);
 933                 if (!stack_page) {
 934                         err = -ENOMEM;
 935                         goto out_free_stack_pages;
 936                 }
 937
 938                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 939         }
 940
 941         /*
 942          * Map the Hyp-code called directly from the host
 943          */
 944         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 945         if (err) {
 946                 kvm_err("Cannot map world-switch code\n");
 947                 goto out_free_mappings;
 948         }
 949
 950         /*
 951          * Map the Hyp stack pages
 952          */
 953         for_each_possible_cpu(cpu) {
 954                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 955                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 956
 957                 if (err) {
 958                         kvm_err("Cannot map hyp stack\n");
 959                         goto out_free_mappings;
 960                 }
 961         }
 962
 963         /*
 964          * Map the host CPU structures
 965          */
 966         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 967         if (!kvm_host_cpu_state) {
 968                 err = -ENOMEM;
 969                 kvm_err("Cannot allocate host CPU state\n");
 970                 goto out_free_mappings;
 971         }
 972
 973         for_each_possible_cpu(cpu) {
 974                 kvm_cpu_context_t *cpu_ctxt;
 975
 976                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 977                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 978
 979                 if (err) {
 980                         kvm_err("Cannot map host CPU state: %d\n", err);
 981                         goto out_free_context;
 982                 }
 983         }
 984
 985         /*
 986          * Execute the init code on each CPU.
 987          */
 988         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 989
 990         /*
 991          * Init HYP view of VGIC
 992          */
 993         err = kvm_vgic_hyp_init();
 994         if (err)
 995                 goto out_free_context;
 996
 997 #ifdef CONFIG_KVM_ARM_VGIC
 998                 vgic_present = true;
 999 #endif
1000
1001         /*
1002          * Init HYP architected timer support
1003          */
1004         err = kvm_timer_hyp_init();
1005         if (err)
1006                 goto out_free_mappings;
1007
1008 #ifndef CONFIG_HOTPLUG_CPU
1009         free_boot_hyp_pgd();
1010 #endif
1011
1012         kvm_perf_init();
1013
1014         kvm_info("Hyp mode initialized successfully\n");
1015
1016         return 0;
1017 out_free_context:
1018         free_percpu(kvm_host_cpu_state);
1019 out_free_mappings:
1020         free_hyp_pgds();
1021 out_free_stack_pages:
1022         for_each_possible_cpu(cpu)
1023                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1024 out_err:
1025         kvm_err("error initializing Hyp mode: %d\n", err);
1026         return err;
1027 }
1028
1029 static void check_kvm_target_cpu(void *ret)
1030 {
1031         *(int *)ret = kvm_target_cpu();
1032 }
1033
1034 /**
1035  * Initialize Hyp-mode and memory mappings on all CPUs.
1036  */
1037 int kvm_arch_init(void *opaque)
1038 {
1039         int err;
1040         int ret, cpu;
1041
1042         if (!is_hyp_mode_available()) {
1043                 kvm_err("HYP mode not available\n");
1044                 return -ENODEV;
1045         }
1046
1047         for_each_online_cpu(cpu) {
1048                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1049                 if (ret < 0) {
1050                         kvm_err("Error, CPU %d not supported!\n", cpu);
1051                         return -ENODEV;
1052                 }
1053         }
1054
1055         cpu_notifier_register_begin();
1056
1057         err = init_hyp_mode();
1058         if (err)
1059                 goto out_err;
1060
1061         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1062         if (err) {
1063                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1064                 goto out_err;
1065         }
1066
1067         cpu_notifier_register_done();
1068
1069         hyp_cpu_pm_init();
1070
1071         kvm_coproc_table_init();
1072         return 0;
1073 out_err:
1074         cpu_notifier_register_done();
1075         return err;
1076 }
1077
1078 /* NOP: Compiling as a module not supported */
1079 void kvm_arch_exit(void)
1080 {
1081         kvm_perf_teardown();
1082 }
1083
1084 static int arm_init(void)
1085 {
1086         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1087         return rc;
1088 }
1089
1090 module_init(arm_init);