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         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 217                 err = -EBUSY;
 218                 goto out;
 219         }
 220
 221         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 222         if (!vcpu) {
 223                 err = -ENOMEM;
 224                 goto out;
 225         }
 226
 227         err = kvm_vcpu_init(vcpu, kvm, id);
 228         if (err)
 229                 goto free_vcpu;
 230
 231         err = create_hyp_mappings(vcpu, vcpu + 1);
 232         if (err)
 233                 goto vcpu_uninit;
 234
 235         return vcpu;
 236 vcpu_uninit:
 237         kvm_vcpu_uninit(vcpu);
 238 free_vcpu:
 239         kmem_cache_free(kvm_vcpu_cache, vcpu);
 240 out:
 241         return ERR_PTR(err);
 242 }
 243
 244 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 245 {
 246         return 0;
 247 }
 248
 249 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 250 {
 251         kvm_mmu_free_memory_caches(vcpu);
 252         kvm_timer_vcpu_terminate(vcpu);
 253         kvm_vgic_vcpu_destroy(vcpu);
 254         kmem_cache_free(kvm_vcpu_cache, vcpu);
 255 }
 256
 257 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 258 {
 259         kvm_arch_vcpu_free(vcpu);
 260 }
 261
 262 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 263 {
 264         return 0;
 265 }
 266
 267 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 268 {
 269         /* Force users to call KVM_ARM_VCPU_INIT */
 270         vcpu->arch.target = -1;
 271         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 272
 273         /* Set up the timer */
 274         kvm_timer_vcpu_init(vcpu);
 275
 276         return 0;
 277 }
 278
 279 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 280 {
 281         vcpu->cpu = cpu;
 282         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 283
 284         /*
 285          * Check whether this vcpu requires the cache to be flushed on
 286          * this physical CPU. This is a consequence of doing dcache
 287          * operations by set/way on this vcpu. We do it here to be in
 288          * a non-preemptible section.
 289          */
 290         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 291                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 292
 293         kvm_arm_set_running_vcpu(vcpu);
 294 }
 295
 296 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 297 {
 298         /*
 299          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 300          * if the vcpu is no longer assigned to a cpu.  This is used for the
 301          * optimized make_all_cpus_request path.
 302          */
 303         vcpu->cpu = -1;
 304
 305         kvm_arm_set_running_vcpu(NULL);
 306 }
 307
 308 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 309                                         struct kvm_guest_debug *dbg)
 310 {
 311         return -EINVAL;
 312 }
 313
 314
 315 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 316                                     struct kvm_mp_state *mp_state)
 317 {
 318         return -EINVAL;
 319 }
 320
 321 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 322                                     struct kvm_mp_state *mp_state)
 323 {
 324         return -EINVAL;
 325 }
 326
 327 /**
 328  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 329  * @v:          The VCPU pointer
 330  *
 331  * If the guest CPU is not waiting for interrupts or an interrupt line is
 332  * asserted, the CPU is by definition runnable.
 333  */
 334 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 335 {
 336         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 337 }
 338
 339 /* Just ensure a guest exit from a particular CPU */
 340 static void exit_vm_noop(void *info)
 341 {
 342 }
 343
 344 void force_vm_exit(const cpumask_t *mask)
 345 {
 346         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 347 }
 348
 349 /**
 350  * need_new_vmid_gen - check that the VMID is still valid
 351  * @kvm: The VM's VMID to checkt
 352  *
 353  * return true if there is a new generation of VMIDs being used
 354  *
 355  * The hardware supports only 256 values with the value zero reserved for the
 356  * host, so we check if an assigned value belongs to a previous generation,
 357  * which which requires us to assign a new value. If we're the first to use a
 358  * VMID for the new generation, we must flush necessary caches and TLBs on all
 359  * CPUs.
 360  */
 361 static bool need_new_vmid_gen(struct kvm *kvm)
 362 {
 363         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 364 }
 365
 366 /**
 367  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 368  * @kvm The guest that we are about to run
 369  *
 370  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 371  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 372  * caches and TLBs.
 373  */
 374 static void update_vttbr(struct kvm *kvm)
 375 {
 376         phys_addr_t pgd_phys;
 377         u64 vmid;
 378
 379         if (!need_new_vmid_gen(kvm))
 380                 return;
 381
 382         spin_lock(&kvm_vmid_lock);
 383
 384         /*
 385          * We need to re-check the vmid_gen here to ensure that if another vcpu
 386          * already allocated a valid vmid for this vm, then this vcpu should
 387          * use the same vmid.
 388          */
 389         if (!need_new_vmid_gen(kvm)) {
 390                 spin_unlock(&kvm_vmid_lock);
 391                 return;
 392         }
 393
 394         /* First user of a new VMID generation? */
 395         if (unlikely(kvm_next_vmid == 0)) {
 396                 atomic64_inc(&kvm_vmid_gen);
 397                 kvm_next_vmid = 1;
 398
 399                 /*
 400                  * On SMP we know no other CPUs can use this CPU's or each
 401                  * other's VMID after force_vm_exit returns since the
 402                  * kvm_vmid_lock blocks them from reentry to the guest.
 403                  */
 404                 force_vm_exit(cpu_all_mask);
 405                 /*
 406                  * Now broadcast TLB + ICACHE invalidation over the inner
 407                  * shareable domain to make sure all data structures are
 408                  * clean.
 409                  */
 410                 kvm_call_hyp(__kvm_flush_vm_context);
 411         }
 412
 413         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 414         kvm->arch.vmid = kvm_next_vmid;
 415         kvm_next_vmid++;
 416
 417         /* update vttbr to be used with the new vmid */
 418         pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
 419         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 420         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 421         kvm->arch.vttbr = pgd_phys | vmid;
 422
 423         spin_unlock(&kvm_vmid_lock);
 424 }
 425
 426 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 427 {
 428         int ret;
 429
 430         if (likely(vcpu->arch.has_run_once))
 431                 return 0;
 432
 433         vcpu->arch.has_run_once = true;
 434
 435         /*
 436          * Map the VGIC hardware resources before running a vcpu the first
 437          * time on this VM.
 438          */
 439         if (unlikely(!vgic_ready(vcpu->kvm))) {
 440                 ret = kvm_vgic_map_resources(vcpu->kvm);
 441                 if (ret)
 442                         return ret;
 443         }
 444
 445         return 0;
 446 }
 447
 448 static void vcpu_pause(struct kvm_vcpu *vcpu)
 449 {
 450         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 451
 452         wait_event_interruptible(*wq, !vcpu->arch.pause);
 453 }
 454
 455 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 456 {
 457         return vcpu->arch.target >= 0;
 458 }
 459
 460 /**
 461  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 462  * @vcpu:       The VCPU pointer
 463  * @run:        The kvm_run structure pointer used for userspace state exchange
 464  *
 465  * This function is called through the VCPU_RUN ioctl called from user space. It
 466  * will execute VM code in a loop until the time slice for the process is used
 467  * or some emulation is needed from user space in which case the function will
 468  * return with return value 0 and with the kvm_run structure filled in with the
 469  * required data for the requested emulation.
 470  */
 471 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 472 {
 473         int ret;
 474         sigset_t sigsaved;
 475
 476         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 477                 return -ENOEXEC;
 478
 479         ret = kvm_vcpu_first_run_init(vcpu);
 480         if (ret)
 481                 return ret;
 482
 483         if (run->exit_reason == KVM_EXIT_MMIO) {
 484                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 485                 if (ret)
 486                         return ret;
 487         }
 488
 489         if (vcpu->sigset_active)
 490                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 491
 492         ret = 1;
 493         run->exit_reason = KVM_EXIT_UNKNOWN;
 494         while (ret > 0) {
 495                 /*
 496                  * Check conditions before entering the guest
 497                  */
 498                 cond_resched();
 499
 500                 update_vttbr(vcpu->kvm);
 501
 502                 if (vcpu->arch.pause)
 503                         vcpu_pause(vcpu);
 504
 505                 kvm_vgic_flush_hwstate(vcpu);
 506                 kvm_timer_flush_hwstate(vcpu);
 507
 508                 local_irq_disable();
 509
 510                 /*
 511                  * Re-check atomic conditions
 512                  */
 513                 if (signal_pending(current)) {
 514                         ret = -EINTR;
 515                         run->exit_reason = KVM_EXIT_INTR;
 516                 }
 517
 518                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 519                         local_irq_enable();
 520                         kvm_timer_sync_hwstate(vcpu);
 521                         kvm_vgic_sync_hwstate(vcpu);
 522                         continue;
 523                 }
 524
 525                 /**************************************************************
 526                  * Enter the guest
 527                  */
 528                 trace_kvm_entry(*vcpu_pc(vcpu));
 529                 kvm_guest_enter();
 530                 vcpu->mode = IN_GUEST_MODE;
 531
 532                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 533
 534                 vcpu->mode = OUTSIDE_GUEST_MODE;
 535                 vcpu->arch.last_pcpu = smp_processor_id();
 536                 kvm_guest_exit();
 537                 trace_kvm_exit(*vcpu_pc(vcpu));
 538                 /*
 539                  * We may have taken a host interrupt in HYP mode (ie
 540                  * while executing the guest). This interrupt is still
 541                  * pending, as we haven't serviced it yet!
 542                  *
 543                  * We're now back in SVC mode, with interrupts
 544                  * disabled.  Enabling the interrupts now will have
 545                  * the effect of taking the interrupt again, in SVC
 546                  * mode this time.
 547                  */
 548                 local_irq_enable();
 549
 550                 /*
 551                  * Back from guest
 552                  *************************************************************/
 553
 554                 kvm_timer_sync_hwstate(vcpu);
 555                 kvm_vgic_sync_hwstate(vcpu);
 556
 557                 ret = handle_exit(vcpu, run, ret);
 558         }
 559
 560         if (vcpu->sigset_active)
 561                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 562         return ret;
 563 }
 564
 565 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 566 {
 567         int bit_index;
 568         bool set;
 569         unsigned long *ptr;
 570
 571         if (number == KVM_ARM_IRQ_CPU_IRQ)
 572                 bit_index = __ffs(HCR_VI);
 573         else /* KVM_ARM_IRQ_CPU_FIQ */
 574                 bit_index = __ffs(HCR_VF);
 575
 576         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 577         if (level)
 578                 set = test_and_set_bit(bit_index, ptr);
 579         else
 580                 set = test_and_clear_bit(bit_index, ptr);
 581
 582         /*
 583          * If we didn't change anything, no need to wake up or kick other CPUs
 584          */
 585         if (set == level)
 586                 return 0;
 587
 588         /*
 589          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 590          * trigger a world-switch round on the running physical CPU to set the
 591          * virtual IRQ/FIQ fields in the HCR appropriately.
 592          */
 593         kvm_vcpu_kick(vcpu);
 594
 595         return 0;
 596 }
 597
 598 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 599                           bool line_status)
 600 {
 601         u32 irq = irq_level->irq;
 602         unsigned int irq_type, vcpu_idx, irq_num;
 603         int nrcpus = atomic_read(&kvm->online_vcpus);
 604         struct kvm_vcpu *vcpu = NULL;
 605         bool level = irq_level->level;
 606
 607         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 608         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 609         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 610
 611         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 612
 613         switch (irq_type) {
 614         case KVM_ARM_IRQ_TYPE_CPU:
 615                 if (irqchip_in_kernel(kvm))
 616                         return -ENXIO;
 617
 618                 if (vcpu_idx >= nrcpus)
 619                         return -EINVAL;
 620
 621                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 622                 if (!vcpu)
 623                         return -EINVAL;
 624
 625                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 626                         return -EINVAL;
 627
 628                 return vcpu_interrupt_line(vcpu, irq_num, level);
 629         case KVM_ARM_IRQ_TYPE_PPI:
 630                 if (!irqchip_in_kernel(kvm))
 631                         return -ENXIO;
 632
 633                 if (vcpu_idx >= nrcpus)
 634                         return -EINVAL;
 635
 636                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 637                 if (!vcpu)
 638                         return -EINVAL;
 639
 640                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 641                         return -EINVAL;
 642
 643                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 644         case KVM_ARM_IRQ_TYPE_SPI:
 645                 if (!irqchip_in_kernel(kvm))
 646                         return -ENXIO;
 647
 648                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 649                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 650                         return -EINVAL;
 651
 652                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 653         }
 654
 655         return -EINVAL;
 656 }
 657
 658 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 659                                const struct kvm_vcpu_init *init)
 660 {
 661         unsigned int i;
 662         int phys_target = kvm_target_cpu();
 663
 664         if (init->target != phys_target)
 665                 return -EINVAL;
 666
 667         /*
 668          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 669          * use the same target.
 670          */
 671         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 672                 return -EINVAL;
 673
 674         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 675         for (i = 0; i < sizeof(init->features) * 8; i++) {
 676                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 677
 678                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 679                         return -ENOENT;
 680
 681                 /*
 682                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 683                  * use the same feature set.
 684                  */
 685                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 686                     test_bit(i, vcpu->arch.features) != set)
 687                         return -EINVAL;
 688
 689                 if (set)
 690                         set_bit(i, vcpu->arch.features);
 691         }
 692
 693         vcpu->arch.target = phys_target;
 694
 695         /* Now we know what it is, we can reset it. */
 696         return kvm_reset_vcpu(vcpu);
 697 }
 698
 699
 700 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 701                                          struct kvm_vcpu_init *init)
 702 {
 703         int ret;
 704
 705         ret = kvm_vcpu_set_target(vcpu, init);
 706         if (ret)
 707                 return ret;
 708
 709         /*
 710          * Ensure a rebooted VM will fault in RAM pages and detect if the
 711          * guest MMU is turned off and flush the caches as needed.
 712          */
 713         if (vcpu->arch.has_run_once)
 714                 stage2_unmap_vm(vcpu->kvm);
 715
 716         vcpu_reset_hcr(vcpu);
 717
 718         /*
 719          * Handle the "start in power-off" case by marking the VCPU as paused.
 720          */
 721         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 722                 vcpu->arch.pause = true;
 723         else
 724                 vcpu->arch.pause = false;
 725
 726         return 0;
 727 }
 728
 729 long kvm_arch_vcpu_ioctl(struct file *filp,
 730                          unsigned int ioctl, unsigned long arg)
 731 {
 732         struct kvm_vcpu *vcpu = filp->private_data;
 733         void __user *argp = (void __user *)arg;
 734
 735         switch (ioctl) {
 736         case KVM_ARM_VCPU_INIT: {
 737                 struct kvm_vcpu_init init;
 738
 739                 if (copy_from_user(&init, argp, sizeof(init)))
 740                         return -EFAULT;
 741
 742                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 743         }
 744         case KVM_SET_ONE_REG:
 745         case KVM_GET_ONE_REG: {
 746                 struct kvm_one_reg reg;
 747
 748                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 749                         return -ENOEXEC;
 750
 751                 if (copy_from_user(&reg, argp, sizeof(reg)))
 752                         return -EFAULT;
 753                 if (ioctl == KVM_SET_ONE_REG)
 754                         return kvm_arm_set_reg(vcpu, &reg);
 755                 else
 756                         return kvm_arm_get_reg(vcpu, &reg);
 757         }
 758         case KVM_GET_REG_LIST: {
 759                 struct kvm_reg_list __user *user_list = argp;
 760                 struct kvm_reg_list reg_list;
 761                 unsigned n;
 762
 763                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 764                         return -ENOEXEC;
 765
 766                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 767                         return -EFAULT;
 768                 n = reg_list.n;
 769                 reg_list.n = kvm_arm_num_regs(vcpu);
 770                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 771                         return -EFAULT;
 772                 if (n < reg_list.n)
 773                         return -E2BIG;
 774                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 775         }
 776         default:
 777                 return -EINVAL;
 778         }
 779 }
 780
 781 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 782 {
 783         return -EINVAL;
 784 }
 785
 786 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 787                                         struct kvm_arm_device_addr *dev_addr)
 788 {
 789         unsigned long dev_id, type;
 790
 791         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 792                 KVM_ARM_DEVICE_ID_SHIFT;
 793         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 794                 KVM_ARM_DEVICE_TYPE_SHIFT;
 795
 796         switch (dev_id) {
 797         case KVM_ARM_DEVICE_VGIC_V2:
 798                 if (!vgic_present)
 799                         return -ENXIO;
 800                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 801         default:
 802                 return -ENODEV;
 803         }
 804 }
 805
 806 long kvm_arch_vm_ioctl(struct file *filp,
 807                        unsigned int ioctl, unsigned long arg)
 808 {
 809         struct kvm *kvm = filp->private_data;
 810         void __user *argp = (void __user *)arg;
 811
 812         switch (ioctl) {
 813         case KVM_CREATE_IRQCHIP: {
 814                 if (vgic_present)
 815                         return kvm_vgic_create(kvm);
 816                 else
 817                         return -ENXIO;
 818         }
 819         case KVM_ARM_SET_DEVICE_ADDR: {
 820                 struct kvm_arm_device_addr dev_addr;
 821
 822                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 823                         return -EFAULT;
 824                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 825         }
 826         case KVM_ARM_PREFERRED_TARGET: {
 827                 int err;
 828                 struct kvm_vcpu_init init;
 829
 830                 err = kvm_vcpu_preferred_target(&init);
 831                 if (err)
 832                         return err;
 833
 834                 if (copy_to_user(argp, &init, sizeof(init)))
 835                         return -EFAULT;
 836
 837                 return 0;
 838         }
 839         default:
 840                 return -EINVAL;
 841         }
 842 }
 843
 844 static void cpu_init_hyp_mode(void *dummy)
 845 {
 846         phys_addr_t boot_pgd_ptr;
 847         phys_addr_t pgd_ptr;
 848         unsigned long hyp_stack_ptr;
 849         unsigned long stack_page;
 850         unsigned long vector_ptr;
 851
 852         /* Switch from the HYP stub to our own HYP init vector */
 853         __hyp_set_vectors(kvm_get_idmap_vector());
 854
 855         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 856         pgd_ptr = kvm_mmu_get_httbr();
 857         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 858         hyp_stack_ptr = stack_page + PAGE_SIZE;
 859         vector_ptr = (unsigned long)__kvm_hyp_vector;
 860
 861         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 862 }
 863
 864 static int hyp_init_cpu_notify(struct notifier_block *self,
 865                                unsigned long action, void *cpu)
 866 {
 867         switch (action) {
 868         case CPU_STARTING:
 869         case CPU_STARTING_FROZEN:
 870                 if (__hyp_get_vectors() == hyp_default_vectors)
 871                         cpu_init_hyp_mode(NULL);
 872                 break;
 873         }
 874
 875         return NOTIFY_OK;
 876 }
 877
 878 static struct notifier_block hyp_init_cpu_nb = {
 879         .notifier_call = hyp_init_cpu_notify,
 880 };
 881
 882 #ifdef CONFIG_CPU_PM
 883 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 884                                     unsigned long cmd,
 885                                     void *v)
 886 {
 887         if (cmd == CPU_PM_EXIT &&
 888             __hyp_get_vectors() == hyp_default_vectors) {
 889                 cpu_init_hyp_mode(NULL);
 890                 return NOTIFY_OK;
 891         }
 892
 893         return NOTIFY_DONE;
 894 }
 895
 896 static struct notifier_block hyp_init_cpu_pm_nb = {
 897         .notifier_call = hyp_init_cpu_pm_notifier,
 898 };
 899
 900 static void __init hyp_cpu_pm_init(void)
 901 {
 902         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 903 }
 904 #else
 905 static inline void hyp_cpu_pm_init(void)
 906 {
 907 }
 908 #endif
 909
 910 /**
 911  * Inits Hyp-mode on all online CPUs
 912  */
 913 static int init_hyp_mode(void)
 914 {
 915         int cpu;
 916         int err = 0;
 917
 918         /*
 919          * Allocate Hyp PGD and setup Hyp identity mapping
 920          */
 921         err = kvm_mmu_init();
 922         if (err)
 923                 goto out_err;
 924
 925         /*
 926          * It is probably enough to obtain the default on one
 927          * CPU. It's unlikely to be different on the others.
 928          */
 929         hyp_default_vectors = __hyp_get_vectors();
 930
 931         /*
 932          * Allocate stack pages for Hypervisor-mode
 933          */
 934         for_each_possible_cpu(cpu) {
 935                 unsigned long stack_page;
 936
 937                 stack_page = __get_free_page(GFP_KERNEL);
 938                 if (!stack_page) {
 939                         err = -ENOMEM;
 940                         goto out_free_stack_pages;
 941                 }
 942
 943                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 944         }
 945
 946         /*
 947          * Map the Hyp-code called directly from the host
 948          */
 949         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 950         if (err) {
 951                 kvm_err("Cannot map world-switch code\n");
 952                 goto out_free_mappings;
 953         }
 954
 955         /*
 956          * Map the Hyp stack pages
 957          */
 958         for_each_possible_cpu(cpu) {
 959                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 960                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 961
 962                 if (err) {
 963                         kvm_err("Cannot map hyp stack\n");
 964                         goto out_free_mappings;
 965                 }
 966         }
 967
 968         /*
 969          * Map the host CPU structures
 970          */
 971         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 972         if (!kvm_host_cpu_state) {
 973                 err = -ENOMEM;
 974                 kvm_err("Cannot allocate host CPU state\n");
 975                 goto out_free_mappings;
 976         }
 977
 978         for_each_possible_cpu(cpu) {
 979                 kvm_cpu_context_t *cpu_ctxt;
 980
 981                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 982                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 983
 984                 if (err) {
 985                         kvm_err("Cannot map host CPU state: %d\n", err);
 986                         goto out_free_context;
 987                 }
 988         }
 989
 990         /*
 991          * Execute the init code on each CPU.
 992          */
 993         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 994
 995         /*
 996          * Init HYP view of VGIC
 997          */
 998         err = kvm_vgic_hyp_init();
 999         if (err)
1000                 goto out_free_context;
1001
1002 #ifdef CONFIG_KVM_ARM_VGIC
1003                 vgic_present = true;
1004 #endif
1005
1006         /*
1007          * Init HYP architected timer support
1008          */
1009         err = kvm_timer_hyp_init();
1010         if (err)
1011                 goto out_free_mappings;
1012
1013 #ifndef CONFIG_HOTPLUG_CPU
1014         free_boot_hyp_pgd();
1015 #endif
1016
1017         kvm_perf_init();
1018
1019         kvm_info("Hyp mode initialized successfully\n");
1020
1021         return 0;
1022 out_free_context:
1023         free_percpu(kvm_host_cpu_state);
1024 out_free_mappings:
1025         free_hyp_pgds();
1026 out_free_stack_pages:
1027         for_each_possible_cpu(cpu)
1028                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1029 out_err:
1030         kvm_err("error initializing Hyp mode: %d\n", err);
1031         return err;
1032 }
1033
1034 static void check_kvm_target_cpu(void *ret)
1035 {
1036         *(int *)ret = kvm_target_cpu();
1037 }
1038
1039 /**
1040  * Initialize Hyp-mode and memory mappings on all CPUs.
1041  */
1042 int kvm_arch_init(void *opaque)
1043 {
1044         int err;
1045         int ret, cpu;
1046
1047         if (!is_hyp_mode_available()) {
1048                 kvm_err("HYP mode not available\n");
1049                 return -ENODEV;
1050         }
1051
1052         for_each_online_cpu(cpu) {
1053                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1054                 if (ret < 0) {
1055                         kvm_err("Error, CPU %d not supported!\n", cpu);
1056                         return -ENODEV;
1057                 }
1058         }
1059
1060         cpu_notifier_register_begin();
1061
1062         err = init_hyp_mode();
1063         if (err)
1064                 goto out_err;
1065
1066         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1067         if (err) {
1068                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1069                 goto out_err;
1070         }
1071
1072         cpu_notifier_register_done();
1073
1074         hyp_cpu_pm_init();
1075
1076         kvm_coproc_table_init();
1077         return 0;
1078 out_err:
1079         cpu_notifier_register_done();
1080         return err;
1081 }
1082
1083 /* NOP: Compiling as a module not supported */
1084 void kvm_arch_exit(void)
1085 {
1086         kvm_perf_teardown();
1087 }
1088
1089 static int arm_init(void)
1090 {
1091         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1092         return rc;
1093 }
1094
1095 module_init(arm_init);