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         /* The maximum number of VCPUs is limited by the host's GIC model */
 136         kvm->arch.max_vcpus = kvm_vgic_get_max_vcpus();
 137
 138         return ret;
 139 out_free_stage2_pgd:
 140         kvm_free_stage2_pgd(kvm);
 141 out_fail_alloc:
 142         return ret;
 143 }
 144
 145 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 146 {
 147         return VM_FAULT_SIGBUS;
 148 }
 149
 150
 151 /**
 152  * kvm_arch_destroy_vm - destroy the VM data structure
 153  * @kvm:        pointer to the KVM struct
 154  */
 155 void kvm_arch_destroy_vm(struct kvm *kvm)
 156 {
 157         int i;
 158
 159         kvm_free_stage2_pgd(kvm);
 160
 161         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 162                 if (kvm->vcpus[i]) {
 163                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 164                         kvm->vcpus[i] = NULL;
 165                 }
 166         }
 167
 168         kvm_vgic_destroy(kvm);
 169 }
 170
 171 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 172 {
 173         int r;
 174         switch (ext) {
 175         case KVM_CAP_IRQCHIP:
 176                 r = vgic_present;
 177                 break;
 178         case KVM_CAP_DEVICE_CTRL:
 179         case KVM_CAP_USER_MEMORY:
 180         case KVM_CAP_SYNC_MMU:
 181         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 182         case KVM_CAP_ONE_REG:
 183         case KVM_CAP_ARM_PSCI:
 184         case KVM_CAP_ARM_PSCI_0_2:
 185         case KVM_CAP_READONLY_MEM:
 186                 r = 1;
 187                 break;
 188         case KVM_CAP_COALESCED_MMIO:
 189                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 190                 break;
 191         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 192                 r = 1;
 193                 break;
 194         case KVM_CAP_NR_VCPUS:
 195                 r = num_online_cpus();
 196                 break;
 197         case KVM_CAP_MAX_VCPUS:
 198                 r = KVM_MAX_VCPUS;
 199                 break;
 200         default:
 201                 r = kvm_arch_dev_ioctl_check_extension(ext);
 202                 break;
 203         }
 204         return r;
 205 }
 206
 207 long kvm_arch_dev_ioctl(struct file *filp,
 208                         unsigned int ioctl, unsigned long arg)
 209 {
 210         return -EINVAL;
 211 }
 212
 213
 214 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 215 {
 216         int err;
 217         struct kvm_vcpu *vcpu;
 218
 219         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm)) {
 220                 err = -EBUSY;
 221                 goto out;
 222         }
 223
 224         if (id >= kvm->arch.max_vcpus) {
 225                 err = -EINVAL;
 226                 goto out;
 227         }
 228
 229         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 230         if (!vcpu) {
 231                 err = -ENOMEM;
 232                 goto out;
 233         }
 234
 235         err = kvm_vcpu_init(vcpu, kvm, id);
 236         if (err)
 237                 goto free_vcpu;
 238
 239         err = create_hyp_mappings(vcpu, vcpu + 1);
 240         if (err)
 241                 goto vcpu_uninit;
 242
 243         return vcpu;
 244 vcpu_uninit:
 245         kvm_vcpu_uninit(vcpu);
 246 free_vcpu:
 247         kmem_cache_free(kvm_vcpu_cache, vcpu);
 248 out:
 249         return ERR_PTR(err);
 250 }
 251
 252 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 253 {
 254 }
 255
 256 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 257 {
 258         kvm_mmu_free_memory_caches(vcpu);
 259         kvm_timer_vcpu_terminate(vcpu);
 260         kvm_vgic_vcpu_destroy(vcpu);
 261         kmem_cache_free(kvm_vcpu_cache, vcpu);
 262 }
 263
 264 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 265 {
 266         kvm_arch_vcpu_free(vcpu);
 267 }
 268
 269 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 270 {
 271         return 0;
 272 }
 273
 274 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 275 {
 276         /* Force users to call KVM_ARM_VCPU_INIT */
 277         vcpu->arch.target = -1;
 278         bitmap_zero(vcpu->arch.features, KVM_VCPU_MAX_FEATURES);
 279
 280         /* Set up the timer */
 281         kvm_timer_vcpu_init(vcpu);
 282
 283         return 0;
 284 }
 285
 286 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 287 {
 288         vcpu->cpu = cpu;
 289         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 290
 291         kvm_arm_set_running_vcpu(vcpu);
 292 }
 293
 294 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 295 {
 296         /*
 297          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 298          * if the vcpu is no longer assigned to a cpu.  This is used for the
 299          * optimized make_all_cpus_request path.
 300          */
 301         vcpu->cpu = -1;
 302
 303         kvm_arm_set_running_vcpu(NULL);
 304 }
 305
 306 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 307                                         struct kvm_guest_debug *dbg)
 308 {
 309         return -EINVAL;
 310 }
 311
 312
 313 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 314                                     struct kvm_mp_state *mp_state)
 315 {
 316         return -EINVAL;
 317 }
 318
 319 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 320                                     struct kvm_mp_state *mp_state)
 321 {
 322         return -EINVAL;
 323 }
 324
 325 /**
 326  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 327  * @v:          The VCPU pointer
 328  *
 329  * If the guest CPU is not waiting for interrupts or an interrupt line is
 330  * asserted, the CPU is by definition runnable.
 331  */
 332 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 333 {
 334         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 335 }
 336
 337 /* Just ensure a guest exit from a particular CPU */
 338 static void exit_vm_noop(void *info)
 339 {
 340 }
 341
 342 void force_vm_exit(const cpumask_t *mask)
 343 {
 344         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 345 }
 346
 347 /**
 348  * need_new_vmid_gen - check that the VMID is still valid
 349  * @kvm: The VM's VMID to checkt
 350  *
 351  * return true if there is a new generation of VMIDs being used
 352  *
 353  * The hardware supports only 256 values with the value zero reserved for the
 354  * host, so we check if an assigned value belongs to a previous generation,
 355  * which which requires us to assign a new value. If we're the first to use a
 356  * VMID for the new generation, we must flush necessary caches and TLBs on all
 357  * CPUs.
 358  */
 359 static bool need_new_vmid_gen(struct kvm *kvm)
 360 {
 361         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 362 }
 363
 364 /**
 365  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 366  * @kvm The guest that we are about to run
 367  *
 368  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 369  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 370  * caches and TLBs.
 371  */
 372 static void update_vttbr(struct kvm *kvm)
 373 {
 374         phys_addr_t pgd_phys;
 375         u64 vmid;
 376
 377         if (!need_new_vmid_gen(kvm))
 378                 return;
 379
 380         spin_lock(&kvm_vmid_lock);
 381
 382         /*
 383          * We need to re-check the vmid_gen here to ensure that if another vcpu
 384          * already allocated a valid vmid for this vm, then this vcpu should
 385          * use the same vmid.
 386          */
 387         if (!need_new_vmid_gen(kvm)) {
 388                 spin_unlock(&kvm_vmid_lock);
 389                 return;
 390         }
 391
 392         /* First user of a new VMID generation? */
 393         if (unlikely(kvm_next_vmid == 0)) {
 394                 atomic64_inc(&kvm_vmid_gen);
 395                 kvm_next_vmid = 1;
 396
 397                 /*
 398                  * On SMP we know no other CPUs can use this CPU's or each
 399                  * other's VMID after force_vm_exit returns since the
 400                  * kvm_vmid_lock blocks them from reentry to the guest.
 401                  */
 402                 force_vm_exit(cpu_all_mask);
 403                 /*
 404                  * Now broadcast TLB + ICACHE invalidation over the inner
 405                  * shareable domain to make sure all data structures are
 406                  * clean.
 407                  */
 408                 kvm_call_hyp(__kvm_flush_vm_context);
 409         }
 410
 411         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 412         kvm->arch.vmid = kvm_next_vmid;
 413         kvm_next_vmid++;
 414
 415         /* update vttbr to be used with the new vmid */
 416         pgd_phys = virt_to_phys(kvm_get_hwpgd(kvm));
 417         BUG_ON(pgd_phys & ~VTTBR_BADDR_MASK);
 418         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 419         kvm->arch.vttbr = pgd_phys | vmid;
 420
 421         spin_unlock(&kvm_vmid_lock);
 422 }
 423
 424 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 425 {
 426         struct kvm *kvm = vcpu->kvm;
 427         int ret;
 428
 429         if (likely(vcpu->arch.has_run_once))
 430                 return 0;
 431
 432         vcpu->arch.has_run_once = true;
 433
 434         /*
 435          * Map the VGIC hardware resources before running a vcpu the first
 436          * time on this VM.
 437          */
 438         if (unlikely(!vgic_ready(kvm))) {
 439                 ret = kvm_vgic_map_resources(kvm);
 440                 if (ret)
 441                         return ret;
 442         }
 443
 444         /*
 445          * Enable the arch timers only if we have an in-kernel VGIC
 446          * and it has been properly initialized, since we cannot handle
 447          * interrupts from the virtual timer with a userspace gic.
 448          */
 449         if (irqchip_in_kernel(kvm) && vgic_initialized(kvm))
 450                 kvm_timer_enable(kvm);
 451
 452         return 0;
 453 }
 454
 455 static void vcpu_pause(struct kvm_vcpu *vcpu)
 456 {
 457         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 458
 459         wait_event_interruptible(*wq, !vcpu->arch.pause);
 460 }
 461
 462 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 463 {
 464         return vcpu->arch.target >= 0;
 465 }
 466
 467 /**
 468  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 469  * @vcpu:       The VCPU pointer
 470  * @run:        The kvm_run structure pointer used for userspace state exchange
 471  *
 472  * This function is called through the VCPU_RUN ioctl called from user space. It
 473  * will execute VM code in a loop until the time slice for the process is used
 474  * or some emulation is needed from user space in which case the function will
 475  * return with return value 0 and with the kvm_run structure filled in with the
 476  * required data for the requested emulation.
 477  */
 478 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 479 {
 480         int ret;
 481         sigset_t sigsaved;
 482
 483         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 484                 return -ENOEXEC;
 485
 486         ret = kvm_vcpu_first_run_init(vcpu);
 487         if (ret)
 488                 return ret;
 489
 490         if (run->exit_reason == KVM_EXIT_MMIO) {
 491                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 492                 if (ret)
 493                         return ret;
 494         }
 495
 496         if (vcpu->sigset_active)
 497                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 498
 499         ret = 1;
 500         run->exit_reason = KVM_EXIT_UNKNOWN;
 501         while (ret > 0) {
 502                 /*
 503                  * Check conditions before entering the guest
 504                  */
 505                 cond_resched();
 506
 507                 update_vttbr(vcpu->kvm);
 508
 509                 if (vcpu->arch.pause)
 510                         vcpu_pause(vcpu);
 511
 512                 kvm_vgic_flush_hwstate(vcpu);
 513                 kvm_timer_flush_hwstate(vcpu);
 514
 515                 local_irq_disable();
 516
 517                 /*
 518                  * Re-check atomic conditions
 519                  */
 520                 if (signal_pending(current)) {
 521                         ret = -EINTR;
 522                         run->exit_reason = KVM_EXIT_INTR;
 523                 }
 524
 525                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 526                         local_irq_enable();
 527                         kvm_timer_sync_hwstate(vcpu);
 528                         kvm_vgic_sync_hwstate(vcpu);
 529                         continue;
 530                 }
 531
 532                 /**************************************************************
 533                  * Enter the guest
 534                  */
 535                 trace_kvm_entry(*vcpu_pc(vcpu));
 536                 kvm_guest_enter();
 537                 vcpu->mode = IN_GUEST_MODE;
 538
 539                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 540
 541                 vcpu->mode = OUTSIDE_GUEST_MODE;
 542                 kvm_guest_exit();
 543                 trace_kvm_exit(kvm_vcpu_trap_get_class(vcpu), *vcpu_pc(vcpu));
 544                 /*
 545                  * We may have taken a host interrupt in HYP mode (ie
 546                  * while executing the guest). This interrupt is still
 547                  * pending, as we haven't serviced it yet!
 548                  *
 549                  * We're now back in SVC mode, with interrupts
 550                  * disabled.  Enabling the interrupts now will have
 551                  * the effect of taking the interrupt again, in SVC
 552                  * mode this time.
 553                  */
 554                 local_irq_enable();
 555
 556                 /*
 557                  * Back from guest
 558                  *************************************************************/
 559
 560                 kvm_timer_sync_hwstate(vcpu);
 561                 kvm_vgic_sync_hwstate(vcpu);
 562
 563                 ret = handle_exit(vcpu, run, ret);
 564         }
 565
 566         if (vcpu->sigset_active)
 567                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 568         return ret;
 569 }
 570
 571 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 572 {
 573         int bit_index;
 574         bool set;
 575         unsigned long *ptr;
 576
 577         if (number == KVM_ARM_IRQ_CPU_IRQ)
 578                 bit_index = __ffs(HCR_VI);
 579         else /* KVM_ARM_IRQ_CPU_FIQ */
 580                 bit_index = __ffs(HCR_VF);
 581
 582         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 583         if (level)
 584                 set = test_and_set_bit(bit_index, ptr);
 585         else
 586                 set = test_and_clear_bit(bit_index, ptr);
 587
 588         /*
 589          * If we didn't change anything, no need to wake up or kick other CPUs
 590          */
 591         if (set == level)
 592                 return 0;
 593
 594         /*
 595          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 596          * trigger a world-switch round on the running physical CPU to set the
 597          * virtual IRQ/FIQ fields in the HCR appropriately.
 598          */
 599         kvm_vcpu_kick(vcpu);
 600
 601         return 0;
 602 }
 603
 604 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 605                           bool line_status)
 606 {
 607         u32 irq = irq_level->irq;
 608         unsigned int irq_type, vcpu_idx, irq_num;
 609         int nrcpus = atomic_read(&kvm->online_vcpus);
 610         struct kvm_vcpu *vcpu = NULL;
 611         bool level = irq_level->level;
 612
 613         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 614         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 615         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 616
 617         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 618
 619         switch (irq_type) {
 620         case KVM_ARM_IRQ_TYPE_CPU:
 621                 if (irqchip_in_kernel(kvm))
 622                         return -ENXIO;
 623
 624                 if (vcpu_idx >= nrcpus)
 625                         return -EINVAL;
 626
 627                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 628                 if (!vcpu)
 629                         return -EINVAL;
 630
 631                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 632                         return -EINVAL;
 633
 634                 return vcpu_interrupt_line(vcpu, irq_num, level);
 635         case KVM_ARM_IRQ_TYPE_PPI:
 636                 if (!irqchip_in_kernel(kvm))
 637                         return -ENXIO;
 638
 639                 if (vcpu_idx >= nrcpus)
 640                         return -EINVAL;
 641
 642                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 643                 if (!vcpu)
 644                         return -EINVAL;
 645
 646                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 647                         return -EINVAL;
 648
 649                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 650         case KVM_ARM_IRQ_TYPE_SPI:
 651                 if (!irqchip_in_kernel(kvm))
 652                         return -ENXIO;
 653
 654                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 655                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 656                         return -EINVAL;
 657
 658                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 659         }
 660
 661         return -EINVAL;
 662 }
 663
 664 static int kvm_vcpu_set_target(struct kvm_vcpu *vcpu,
 665                                const struct kvm_vcpu_init *init)
 666 {
 667         unsigned int i;
 668         int phys_target = kvm_target_cpu();
 669
 670         if (init->target != phys_target)
 671                 return -EINVAL;
 672
 673         /*
 674          * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 675          * use the same target.
 676          */
 677         if (vcpu->arch.target != -1 && vcpu->arch.target != init->target)
 678                 return -EINVAL;
 679
 680         /* -ENOENT for unknown features, -EINVAL for invalid combinations. */
 681         for (i = 0; i < sizeof(init->features) * 8; i++) {
 682                 bool set = (init->features[i / 32] & (1 << (i % 32)));
 683
 684                 if (set && i >= KVM_VCPU_MAX_FEATURES)
 685                         return -ENOENT;
 686
 687                 /*
 688                  * Secondary and subsequent calls to KVM_ARM_VCPU_INIT must
 689                  * use the same feature set.
 690                  */
 691                 if (vcpu->arch.target != -1 && i < KVM_VCPU_MAX_FEATURES &&
 692                     test_bit(i, vcpu->arch.features) != set)
 693                         return -EINVAL;
 694
 695                 if (set)
 696                         set_bit(i, vcpu->arch.features);
 697         }
 698
 699         vcpu->arch.target = phys_target;
 700
 701         /* Now we know what it is, we can reset it. */
 702         return kvm_reset_vcpu(vcpu);
 703 }
 704
 705
 706 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 707                                          struct kvm_vcpu_init *init)
 708 {
 709         int ret;
 710
 711         ret = kvm_vcpu_set_target(vcpu, init);
 712         if (ret)
 713                 return ret;
 714
 715         /*
 716          * Ensure a rebooted VM will fault in RAM pages and detect if the
 717          * guest MMU is turned off and flush the caches as needed.
 718          */
 719         if (vcpu->arch.has_run_once)
 720                 stage2_unmap_vm(vcpu->kvm);
 721
 722         vcpu_reset_hcr(vcpu);
 723
 724         /*
 725          * Handle the "start in power-off" case by marking the VCPU as paused.
 726          */
 727         if (test_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 728                 vcpu->arch.pause = true;
 729         else
 730                 vcpu->arch.pause = false;
 731
 732         return 0;
 733 }
 734
 735 long kvm_arch_vcpu_ioctl(struct file *filp,
 736                          unsigned int ioctl, unsigned long arg)
 737 {
 738         struct kvm_vcpu *vcpu = filp->private_data;
 739         void __user *argp = (void __user *)arg;
 740
 741         switch (ioctl) {
 742         case KVM_ARM_VCPU_INIT: {
 743                 struct kvm_vcpu_init init;
 744
 745                 if (copy_from_user(&init, argp, sizeof(init)))
 746                         return -EFAULT;
 747
 748                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 749         }
 750         case KVM_SET_ONE_REG:
 751         case KVM_GET_ONE_REG: {
 752                 struct kvm_one_reg reg;
 753
 754                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 755                         return -ENOEXEC;
 756
 757                 if (copy_from_user(&reg, argp, sizeof(reg)))
 758                         return -EFAULT;
 759                 if (ioctl == KVM_SET_ONE_REG)
 760                         return kvm_arm_set_reg(vcpu, &reg);
 761                 else
 762                         return kvm_arm_get_reg(vcpu, &reg);
 763         }
 764         case KVM_GET_REG_LIST: {
 765                 struct kvm_reg_list __user *user_list = argp;
 766                 struct kvm_reg_list reg_list;
 767                 unsigned n;
 768
 769                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 770                         return -ENOEXEC;
 771
 772                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 773                         return -EFAULT;
 774                 n = reg_list.n;
 775                 reg_list.n = kvm_arm_num_regs(vcpu);
 776                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 777                         return -EFAULT;
 778                 if (n < reg_list.n)
 779                         return -E2BIG;
 780                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 781         }
 782         default:
 783                 return -EINVAL;
 784         }
 785 }
 786
 787 /**
 788  * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
 789  * @kvm: kvm instance
 790  * @log: slot id and address to which we copy the log
 791  *
 792  * Steps 1-4 below provide general overview of dirty page logging. See
 793  * kvm_get_dirty_log_protect() function description for additional details.
 794  *
 795  * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
 796  * always flush the TLB (step 4) even if previous step failed  and the dirty
 797  * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
 798  * does not preclude user space subsequent dirty log read. Flushing TLB ensures
 799  * writes will be marked dirty for next log read.
 800  *
 801  *   1. Take a snapshot of the bit and clear it if needed.
 802  *   2. Write protect the corresponding page.
 803  *   3. Copy the snapshot to the userspace.
 804  *   4. Flush TLB's if needed.
 805  */
 806 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 807 {
 808         bool is_dirty = false;
 809         int r;
 810
 811         mutex_lock(&kvm->slots_lock);
 812
 813         r = kvm_get_dirty_log_protect(kvm, log, &is_dirty);
 814
 815         if (is_dirty)
 816                 kvm_flush_remote_tlbs(kvm);
 817
 818         mutex_unlock(&kvm->slots_lock);
 819         return r;
 820 }
 821
 822 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 823                                         struct kvm_arm_device_addr *dev_addr)
 824 {
 825         unsigned long dev_id, type;
 826
 827         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 828                 KVM_ARM_DEVICE_ID_SHIFT;
 829         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 830                 KVM_ARM_DEVICE_TYPE_SHIFT;
 831
 832         switch (dev_id) {
 833         case KVM_ARM_DEVICE_VGIC_V2:
 834                 if (!vgic_present)
 835                         return -ENXIO;
 836                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 837         default:
 838                 return -ENODEV;
 839         }
 840 }
 841
 842 long kvm_arch_vm_ioctl(struct file *filp,
 843                        unsigned int ioctl, unsigned long arg)
 844 {
 845         struct kvm *kvm = filp->private_data;
 846         void __user *argp = (void __user *)arg;
 847
 848         switch (ioctl) {
 849         case KVM_CREATE_IRQCHIP: {
 850                 if (vgic_present)
 851                         return kvm_vgic_create(kvm, KVM_DEV_TYPE_ARM_VGIC_V2);
 852                 else
 853                         return -ENXIO;
 854         }
 855         case KVM_ARM_SET_DEVICE_ADDR: {
 856                 struct kvm_arm_device_addr dev_addr;
 857
 858                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 859                         return -EFAULT;
 860                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 861         }
 862         case KVM_ARM_PREFERRED_TARGET: {
 863                 int err;
 864                 struct kvm_vcpu_init init;
 865
 866                 err = kvm_vcpu_preferred_target(&init);
 867                 if (err)
 868                         return err;
 869
 870                 if (copy_to_user(argp, &init, sizeof(init)))
 871                         return -EFAULT;
 872
 873                 return 0;
 874         }
 875         default:
 876                 return -EINVAL;
 877         }
 878 }
 879
 880 static void cpu_init_hyp_mode(void *dummy)
 881 {
 882         phys_addr_t boot_pgd_ptr;
 883         phys_addr_t pgd_ptr;
 884         unsigned long hyp_stack_ptr;
 885         unsigned long stack_page;
 886         unsigned long vector_ptr;
 887
 888         /* Switch from the HYP stub to our own HYP init vector */
 889         __hyp_set_vectors(kvm_get_idmap_vector());
 890
 891         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 892         pgd_ptr = kvm_mmu_get_httbr();
 893         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 894         hyp_stack_ptr = stack_page + PAGE_SIZE;
 895         vector_ptr = (unsigned long)__kvm_hyp_vector;
 896
 897         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 898 }
 899
 900 static int hyp_init_cpu_notify(struct notifier_block *self,
 901                                unsigned long action, void *cpu)
 902 {
 903         switch (action) {
 904         case CPU_STARTING:
 905         case CPU_STARTING_FROZEN:
 906                 if (__hyp_get_vectors() == hyp_default_vectors)
 907                         cpu_init_hyp_mode(NULL);
 908                 break;
 909         }
 910
 911         return NOTIFY_OK;
 912 }
 913
 914 static struct notifier_block hyp_init_cpu_nb = {
 915         .notifier_call = hyp_init_cpu_notify,
 916 };
 917
 918 #ifdef CONFIG_CPU_PM
 919 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 920                                     unsigned long cmd,
 921                                     void *v)
 922 {
 923         if (cmd == CPU_PM_EXIT &&
 924             __hyp_get_vectors() == hyp_default_vectors) {
 925                 cpu_init_hyp_mode(NULL);
 926                 return NOTIFY_OK;
 927         }
 928
 929         return NOTIFY_DONE;
 930 }
 931
 932 static struct notifier_block hyp_init_cpu_pm_nb = {
 933         .notifier_call = hyp_init_cpu_pm_notifier,
 934 };
 935
 936 static void __init hyp_cpu_pm_init(void)
 937 {
 938         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 939 }
 940 #else
 941 static inline void hyp_cpu_pm_init(void)
 942 {
 943 }
 944 #endif
 945
 946 /**
 947  * Inits Hyp-mode on all online CPUs
 948  */
 949 static int init_hyp_mode(void)
 950 {
 951         int cpu;
 952         int err = 0;
 953
 954         /*
 955          * Allocate Hyp PGD and setup Hyp identity mapping
 956          */
 957         err = kvm_mmu_init();
 958         if (err)
 959                 goto out_err;
 960
 961         /*
 962          * It is probably enough to obtain the default on one
 963          * CPU. It's unlikely to be different on the others.
 964          */
 965         hyp_default_vectors = __hyp_get_vectors();
 966
 967         /*
 968          * Allocate stack pages for Hypervisor-mode
 969          */
 970         for_each_possible_cpu(cpu) {
 971                 unsigned long stack_page;
 972
 973                 stack_page = __get_free_page(GFP_KERNEL);
 974                 if (!stack_page) {
 975                         err = -ENOMEM;
 976                         goto out_free_stack_pages;
 977                 }
 978
 979                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 980         }
 981
 982         /*
 983          * Map the Hyp-code called directly from the host
 984          */
 985         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 986         if (err) {
 987                 kvm_err("Cannot map world-switch code\n");
 988                 goto out_free_mappings;
 989         }
 990
 991         /*
 992          * Map the Hyp stack pages
 993          */
 994         for_each_possible_cpu(cpu) {
 995                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 996                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 997
 998                 if (err) {
 999                         kvm_err("Cannot map hyp stack\n");
1000                         goto out_free_mappings;
1001                 }
1002         }
1003
1004         /*
1005          * Map the host CPU structures
1006          */
1007         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
1008         if (!kvm_host_cpu_state) {
1009                 err = -ENOMEM;
1010                 kvm_err("Cannot allocate host CPU state\n");
1011                 goto out_free_mappings;
1012         }
1013
1014         for_each_possible_cpu(cpu) {
1015                 kvm_cpu_context_t *cpu_ctxt;
1016
1017                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
1018                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
1019
1020                 if (err) {
1021                         kvm_err("Cannot map host CPU state: %d\n", err);
1022                         goto out_free_context;
1023                 }
1024         }
1025
1026         /*
1027          * Execute the init code on each CPU.
1028          */
1029         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
1030
1031         /*
1032          * Init HYP view of VGIC
1033          */
1034         err = kvm_vgic_hyp_init();
1035         if (err)
1036                 goto out_free_context;
1037
1038 #ifdef CONFIG_KVM_ARM_VGIC
1039                 vgic_present = true;
1040 #endif
1041
1042         /*
1043          * Init HYP architected timer support
1044          */
1045         err = kvm_timer_hyp_init();
1046         if (err)
1047                 goto out_free_mappings;
1048
1049 #ifndef CONFIG_HOTPLUG_CPU
1050         free_boot_hyp_pgd();
1051 #endif
1052
1053         kvm_perf_init();
1054
1055         kvm_info("Hyp mode initialized successfully\n");
1056
1057         return 0;
1058 out_free_context:
1059         free_percpu(kvm_host_cpu_state);
1060 out_free_mappings:
1061         free_hyp_pgds();
1062 out_free_stack_pages:
1063         for_each_possible_cpu(cpu)
1064                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1065 out_err:
1066         kvm_err("error initializing Hyp mode: %d\n", err);
1067         return err;
1068 }
1069
1070 static void check_kvm_target_cpu(void *ret)
1071 {
1072         *(int *)ret = kvm_target_cpu();
1073 }
1074
1075 struct kvm_vcpu *kvm_mpidr_to_vcpu(struct kvm *kvm, unsigned long mpidr)
1076 {
1077         struct kvm_vcpu *vcpu;
1078         int i;
1079
1080         mpidr &= MPIDR_HWID_BITMASK;
1081         kvm_for_each_vcpu(i, vcpu, kvm) {
1082                 if (mpidr == kvm_vcpu_get_mpidr_aff(vcpu))
1083                         return vcpu;
1084         }
1085         return NULL;
1086 }
1087
1088 /**
1089  * Initialize Hyp-mode and memory mappings on all CPUs.
1090  */
1091 int kvm_arch_init(void *opaque)
1092 {
1093         int err;
1094         int ret, cpu;
1095
1096         if (!is_hyp_mode_available()) {
1097                 kvm_err("HYP mode not available\n");
1098                 return -ENODEV;
1099         }
1100
1101         for_each_online_cpu(cpu) {
1102                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1103                 if (ret < 0) {
1104                         kvm_err("Error, CPU %d not supported!\n", cpu);
1105                         return -ENODEV;
1106                 }
1107         }
1108
1109         cpu_notifier_register_begin();
1110
1111         err = init_hyp_mode();
1112         if (err)
1113                 goto out_err;
1114
1115         err = __register_cpu_notifier(&hyp_init_cpu_nb);
1116         if (err) {
1117                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1118                 goto out_err;
1119         }
1120
1121         cpu_notifier_register_done();
1122
1123         hyp_cpu_pm_init();
1124
1125         kvm_coproc_table_init();
1126         return 0;
1127 out_err:
1128         cpu_notifier_register_done();
1129         return err;
1130 }
1131
1132 /* NOP: Compiling as a module not supported */
1133 void kvm_arch_exit(void)
1134 {
1135         kvm_perf_teardown();
1136 }
1137
1138 static int arm_init(void)
1139 {
1140         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1141         return rc;
1142 }
1143
1144 module_init(arm_init);