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