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/errno.h>
  21 #include <linux/err.h>
  22 #include <linux/kvm_host.h>
  23 #include <linux/module.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/fs.h>
  26 #include <linux/mman.h>
  27 #include <linux/sched.h>
  28 #include <linux/kvm.h>
  29 #include <trace/events/kvm.h>
  30
  31 #define CREATE_TRACE_POINTS
  32 #include "trace.h"
  33
  34 #include <asm/uaccess.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/mman.h>
  37 #include <asm/tlbflush.h>
  38 #include <asm/cacheflush.h>
  39 #include <asm/virt.h>
  40 #include <asm/kvm_arm.h>
  41 #include <asm/kvm_asm.h>
  42 #include <asm/kvm_mmu.h>
  43 #include <asm/kvm_emulate.h>
  44 #include <asm/kvm_coproc.h>
  45 #include <asm/kvm_psci.h>
  46
  47 #ifdef REQUIRES_VIRT
  48 __asm__(".arch_extension        virt");
  49 #endif
  50
  51 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  52 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  53 static unsigned long hyp_default_vectors;
  54
  55 /* Per-CPU variable containing the currently running vcpu. */
  56 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  57
  58 /* The VMID used in the VTTBR */
  59 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  60 static u8 kvm_next_vmid;
  61 static DEFINE_SPINLOCK(kvm_vmid_lock);
  62
  63 static bool vgic_present;
  64
  65 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  66 {
  67         BUG_ON(preemptible());
  68         __this_cpu_write(kvm_arm_running_vcpu, vcpu);
  69 }
  70
  71 /**
  72  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  73  * Must be called from non-preemptible context
  74  */
  75 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  76 {
  77         BUG_ON(preemptible());
  78         return __this_cpu_read(kvm_arm_running_vcpu);
  79 }
  80
  81 /**
  82  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  83  */
  84 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  85 {
  86         return &kvm_arm_running_vcpu;
  87 }
  88
  89 int kvm_arch_hardware_enable(void *garbage)
  90 {
  91         return 0;
  92 }
  93
  94 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  95 {
  96         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  97 }
  98
  99 void kvm_arch_hardware_disable(void *garbage)
 100 {
 101 }
 102
 103 int kvm_arch_hardware_setup(void)
 104 {
 105         return 0;
 106 }
 107
 108 void kvm_arch_hardware_unsetup(void)
 109 {
 110 }
 111
 112 void kvm_arch_check_processor_compat(void *rtn)
 113 {
 114         *(int *)rtn = 0;
 115 }
 116
 117 void kvm_arch_sync_events(struct kvm *kvm)
 118 {
 119 }
 120
 121 /**
 122  * kvm_arch_init_vm - initializes a VM data structure
 123  * @kvm:        pointer to the KVM struct
 124  */
 125 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 126 {
 127         int ret = 0;
 128
 129         if (type)
 130                 return -EINVAL;
 131
 132         ret = kvm_alloc_stage2_pgd(kvm);
 133         if (ret)
 134                 goto out_fail_alloc;
 135
 136         ret = create_hyp_mappings(kvm, kvm + 1);
 137         if (ret)
 138                 goto out_free_stage2_pgd;
 139
 140         kvm_timer_init(kvm);
 141
 142         /* Mark the initial VMID generation invalid */
 143         kvm->arch.vmid_gen = 0;
 144
 145         return ret;
 146 out_free_stage2_pgd:
 147         kvm_free_stage2_pgd(kvm);
 148 out_fail_alloc:
 149         return ret;
 150 }
 151
 152 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 153 {
 154         return VM_FAULT_SIGBUS;
 155 }
 156
 157 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 158                            struct kvm_memory_slot *dont)
 159 {
 160 }
 161
 162 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 163                             unsigned long npages)
 164 {
 165         return 0;
 166 }
 167
 168 /**
 169  * kvm_arch_destroy_vm - destroy the VM data structure
 170  * @kvm:        pointer to the KVM struct
 171  */
 172 void kvm_arch_destroy_vm(struct kvm *kvm)
 173 {
 174         int i;
 175
 176         kvm_free_stage2_pgd(kvm);
 177
 178         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 179                 if (kvm->vcpus[i]) {
 180                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 181                         kvm->vcpus[i] = NULL;
 182                 }
 183         }
 184 }
 185
 186 int kvm_dev_ioctl_check_extension(long ext)
 187 {
 188         int r;
 189         switch (ext) {
 190         case KVM_CAP_IRQCHIP:
 191                 r = vgic_present;
 192                 break;
 193         case KVM_CAP_DEVICE_CTRL:
 194         case KVM_CAP_USER_MEMORY:
 195         case KVM_CAP_SYNC_MMU:
 196         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 197         case KVM_CAP_ONE_REG:
 198         case KVM_CAP_ARM_PSCI:
 199                 r = 1;
 200                 break;
 201         case KVM_CAP_COALESCED_MMIO:
 202                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 203                 break;
 204         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 205                 r = 1;
 206                 break;
 207         case KVM_CAP_NR_VCPUS:
 208                 r = num_online_cpus();
 209                 break;
 210         case KVM_CAP_MAX_VCPUS:
 211                 r = KVM_MAX_VCPUS;
 212                 break;
 213         default:
 214                 r = kvm_arch_dev_ioctl_check_extension(ext);
 215                 break;
 216         }
 217         return r;
 218 }
 219
 220 long kvm_arch_dev_ioctl(struct file *filp,
 221                         unsigned int ioctl, unsigned long arg)
 222 {
 223         return -EINVAL;
 224 }
 225
 226 void kvm_arch_memslots_updated(struct kvm *kvm)
 227 {
 228 }
 229
 230 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 231                                    struct kvm_memory_slot *memslot,
 232                                    struct kvm_userspace_memory_region *mem,
 233                                    enum kvm_mr_change change)
 234 {
 235         return 0;
 236 }
 237
 238 void kvm_arch_commit_memory_region(struct kvm *kvm,
 239                                    struct kvm_userspace_memory_region *mem,
 240                                    const struct kvm_memory_slot *old,
 241                                    enum kvm_mr_change change)
 242 {
 243 }
 244
 245 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 246 {
 247 }
 248
 249 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 250                                    struct kvm_memory_slot *slot)
 251 {
 252 }
 253
 254 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 255 {
 256         int err;
 257         struct kvm_vcpu *vcpu;
 258
 259         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 260         if (!vcpu) {
 261                 err = -ENOMEM;
 262                 goto out;
 263         }
 264
 265         err = kvm_vcpu_init(vcpu, kvm, id);
 266         if (err)
 267                 goto free_vcpu;
 268
 269         err = create_hyp_mappings(vcpu, vcpu + 1);
 270         if (err)
 271                 goto vcpu_uninit;
 272
 273         return vcpu;
 274 vcpu_uninit:
 275         kvm_vcpu_uninit(vcpu);
 276 free_vcpu:
 277         kmem_cache_free(kvm_vcpu_cache, vcpu);
 278 out:
 279         return ERR_PTR(err);
 280 }
 281
 282 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 283 {
 284         return 0;
 285 }
 286
 287 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 288 {
 289         kvm_mmu_free_memory_caches(vcpu);
 290         kvm_timer_vcpu_terminate(vcpu);
 291         kmem_cache_free(kvm_vcpu_cache, vcpu);
 292 }
 293
 294 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 295 {
 296         kvm_arch_vcpu_free(vcpu);
 297 }
 298
 299 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 300 {
 301         return 0;
 302 }
 303
 304 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 305 {
 306         int ret;
 307
 308         /* Force users to call KVM_ARM_VCPU_INIT */
 309         vcpu->arch.target = -1;
 310
 311         /* Set up VGIC */
 312         ret = kvm_vgic_vcpu_init(vcpu);
 313         if (ret)
 314                 return ret;
 315
 316         /* Set up the timer */
 317         kvm_timer_vcpu_init(vcpu);
 318
 319         return 0;
 320 }
 321
 322 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 323 {
 324 }
 325
 326 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 327 {
 328         vcpu->cpu = cpu;
 329         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 330
 331         /*
 332          * Check whether this vcpu requires the cache to be flushed on
 333          * this physical CPU. This is a consequence of doing dcache
 334          * operations by set/way on this vcpu. We do it here to be in
 335          * a non-preemptible section.
 336          */
 337         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 338                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 339
 340         kvm_arm_set_running_vcpu(vcpu);
 341 }
 342
 343 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 344 {
 345         /*
 346          * The arch-generic KVM code expects the cpu field of a vcpu to be -1
 347          * if the vcpu is no longer assigned to a cpu.  This is used for the
 348          * optimized make_all_cpus_request path.
 349          */
 350         vcpu->cpu = -1;
 351
 352         kvm_arm_set_running_vcpu(NULL);
 353 }
 354
 355 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 356                                         struct kvm_guest_debug *dbg)
 357 {
 358         return -EINVAL;
 359 }
 360
 361
 362 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 363                                     struct kvm_mp_state *mp_state)
 364 {
 365         return -EINVAL;
 366 }
 367
 368 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 369                                     struct kvm_mp_state *mp_state)
 370 {
 371         return -EINVAL;
 372 }
 373
 374 /**
 375  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 376  * @v:          The VCPU pointer
 377  *
 378  * If the guest CPU is not waiting for interrupts or an interrupt line is
 379  * asserted, the CPU is by definition runnable.
 380  */
 381 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 382 {
 383         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 384 }
 385
 386 /* Just ensure a guest exit from a particular CPU */
 387 static void exit_vm_noop(void *info)
 388 {
 389 }
 390
 391 void force_vm_exit(const cpumask_t *mask)
 392 {
 393         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 394 }
 395
 396 /**
 397  * need_new_vmid_gen - check that the VMID is still valid
 398  * @kvm: The VM's VMID to checkt
 399  *
 400  * return true if there is a new generation of VMIDs being used
 401  *
 402  * The hardware supports only 256 values with the value zero reserved for the
 403  * host, so we check if an assigned value belongs to a previous generation,
 404  * which which requires us to assign a new value. If we're the first to use a
 405  * VMID for the new generation, we must flush necessary caches and TLBs on all
 406  * CPUs.
 407  */
 408 static bool need_new_vmid_gen(struct kvm *kvm)
 409 {
 410         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 411 }
 412
 413 /**
 414  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 415  * @kvm The guest that we are about to run
 416  *
 417  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 418  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 419  * caches and TLBs.
 420  */
 421 static void update_vttbr(struct kvm *kvm)
 422 {
 423         phys_addr_t pgd_phys;
 424         u64 vmid;
 425
 426         if (!need_new_vmid_gen(kvm))
 427                 return;
 428
 429         spin_lock(&kvm_vmid_lock);
 430
 431         /*
 432          * We need to re-check the vmid_gen here to ensure that if another vcpu
 433          * already allocated a valid vmid for this vm, then this vcpu should
 434          * use the same vmid.
 435          */
 436         if (!need_new_vmid_gen(kvm)) {
 437                 spin_unlock(&kvm_vmid_lock);
 438                 return;
 439         }
 440
 441         /* First user of a new VMID generation? */
 442         if (unlikely(kvm_next_vmid == 0)) {
 443                 atomic64_inc(&kvm_vmid_gen);
 444                 kvm_next_vmid = 1;
 445
 446                 /*
 447                  * On SMP we know no other CPUs can use this CPU's or each
 448                  * other's VMID after force_vm_exit returns since the
 449                  * kvm_vmid_lock blocks them from reentry to the guest.
 450                  */
 451                 force_vm_exit(cpu_all_mask);
 452                 /*
 453                  * Now broadcast TLB + ICACHE invalidation over the inner
 454                  * shareable domain to make sure all data structures are
 455                  * clean.
 456                  */
 457                 kvm_call_hyp(__kvm_flush_vm_context);
 458         }
 459
 460         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 461         kvm->arch.vmid = kvm_next_vmid;
 462         kvm_next_vmid++;
 463
 464         /* update vttbr to be used with the new vmid */
 465         pgd_phys = virt_to_phys(kvm->arch.pgd);
 466         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 467         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 468         kvm->arch.vttbr |= vmid;
 469
 470         spin_unlock(&kvm_vmid_lock);
 471 }
 472
 473 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 474 {
 475         int ret;
 476
 477         if (likely(vcpu->arch.has_run_once))
 478                 return 0;
 479
 480         vcpu->arch.has_run_once = true;
 481
 482         /*
 483          * Initialize the VGIC before running a vcpu the first time on
 484          * this VM.
 485          */
 486         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 487                 ret = kvm_vgic_init(vcpu->kvm);
 488                 if (ret)
 489                         return ret;
 490         }
 491
 492         /*
 493          * Handle the "start in power-off" case by calling into the
 494          * PSCI code.
 495          */
 496         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
 497                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
 498                 kvm_psci_call(vcpu);
 499         }
 500
 501         return 0;
 502 }
 503
 504 static void vcpu_pause(struct kvm_vcpu *vcpu)
 505 {
 506         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 507
 508         wait_event_interruptible(*wq, !vcpu->arch.pause);
 509 }
 510
 511 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 512 {
 513         return vcpu->arch.target >= 0;
 514 }
 515
 516 /**
 517  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 518  * @vcpu:       The VCPU pointer
 519  * @run:        The kvm_run structure pointer used for userspace state exchange
 520  *
 521  * This function is called through the VCPU_RUN ioctl called from user space. It
 522  * will execute VM code in a loop until the time slice for the process is used
 523  * or some emulation is needed from user space in which case the function will
 524  * return with return value 0 and with the kvm_run structure filled in with the
 525  * required data for the requested emulation.
 526  */
 527 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 528 {
 529         int ret;
 530         sigset_t sigsaved;
 531
 532         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 533                 return -ENOEXEC;
 534
 535         ret = kvm_vcpu_first_run_init(vcpu);
 536         if (ret)
 537                 return ret;
 538
 539         if (run->exit_reason == KVM_EXIT_MMIO) {
 540                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 541                 if (ret)
 542                         return ret;
 543         }
 544
 545         if (vcpu->sigset_active)
 546                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 547
 548         ret = 1;
 549         run->exit_reason = KVM_EXIT_UNKNOWN;
 550         while (ret > 0) {
 551                 /*
 552                  * Check conditions before entering the guest
 553                  */
 554                 cond_resched();
 555
 556                 update_vttbr(vcpu->kvm);
 557
 558                 if (vcpu->arch.pause)
 559                         vcpu_pause(vcpu);
 560
 561                 kvm_vgic_flush_hwstate(vcpu);
 562                 kvm_timer_flush_hwstate(vcpu);
 563
 564                 local_irq_disable();
 565
 566                 /*
 567                  * Re-check atomic conditions
 568                  */
 569                 if (signal_pending(current)) {
 570                         ret = -EINTR;
 571                         run->exit_reason = KVM_EXIT_INTR;
 572                 }
 573
 574                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 575                         local_irq_enable();
 576                         kvm_timer_sync_hwstate(vcpu);
 577                         kvm_vgic_sync_hwstate(vcpu);
 578                         continue;
 579                 }
 580
 581                 /**************************************************************
 582                  * Enter the guest
 583                  */
 584                 trace_kvm_entry(*vcpu_pc(vcpu));
 585                 kvm_guest_enter();
 586                 vcpu->mode = IN_GUEST_MODE;
 587
 588                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 589
 590                 vcpu->mode = OUTSIDE_GUEST_MODE;
 591                 vcpu->arch.last_pcpu = smp_processor_id();
 592                 kvm_guest_exit();
 593                 trace_kvm_exit(*vcpu_pc(vcpu));
 594                 /*
 595                  * We may have taken a host interrupt in HYP mode (ie
 596                  * while executing the guest). This interrupt is still
 597                  * pending, as we haven't serviced it yet!
 598                  *
 599                  * We're now back in SVC mode, with interrupts
 600                  * disabled.  Enabling the interrupts now will have
 601                  * the effect of taking the interrupt again, in SVC
 602                  * mode this time.
 603                  */
 604                 local_irq_enable();
 605
 606                 /*
 607                  * Back from guest
 608                  *************************************************************/
 609
 610                 kvm_timer_sync_hwstate(vcpu);
 611                 kvm_vgic_sync_hwstate(vcpu);
 612
 613                 ret = handle_exit(vcpu, run, ret);
 614         }
 615
 616         if (vcpu->sigset_active)
 617                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 618         return ret;
 619 }
 620
 621 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 622 {
 623         int bit_index;
 624         bool set;
 625         unsigned long *ptr;
 626
 627         if (number == KVM_ARM_IRQ_CPU_IRQ)
 628                 bit_index = __ffs(HCR_VI);
 629         else /* KVM_ARM_IRQ_CPU_FIQ */
 630                 bit_index = __ffs(HCR_VF);
 631
 632         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 633         if (level)
 634                 set = test_and_set_bit(bit_index, ptr);
 635         else
 636                 set = test_and_clear_bit(bit_index, ptr);
 637
 638         /*
 639          * If we didn't change anything, no need to wake up or kick other CPUs
 640          */
 641         if (set == level)
 642                 return 0;
 643
 644         /*
 645          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 646          * trigger a world-switch round on the running physical CPU to set the
 647          * virtual IRQ/FIQ fields in the HCR appropriately.
 648          */
 649         kvm_vcpu_kick(vcpu);
 650
 651         return 0;
 652 }
 653
 654 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 655                           bool line_status)
 656 {
 657         u32 irq = irq_level->irq;
 658         unsigned int irq_type, vcpu_idx, irq_num;
 659         int nrcpus = atomic_read(&kvm->online_vcpus);
 660         struct kvm_vcpu *vcpu = NULL;
 661         bool level = irq_level->level;
 662
 663         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 664         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 665         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 666
 667         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 668
 669         switch (irq_type) {
 670         case KVM_ARM_IRQ_TYPE_CPU:
 671                 if (irqchip_in_kernel(kvm))
 672                         return -ENXIO;
 673
 674                 if (vcpu_idx >= nrcpus)
 675                         return -EINVAL;
 676
 677                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 678                 if (!vcpu)
 679                         return -EINVAL;
 680
 681                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 682                         return -EINVAL;
 683
 684                 return vcpu_interrupt_line(vcpu, irq_num, level);
 685         case KVM_ARM_IRQ_TYPE_PPI:
 686                 if (!irqchip_in_kernel(kvm))
 687                         return -ENXIO;
 688
 689                 if (vcpu_idx >= nrcpus)
 690                         return -EINVAL;
 691
 692                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 693                 if (!vcpu)
 694                         return -EINVAL;
 695
 696                 if (irq_num < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 697                         return -EINVAL;
 698
 699                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 700         case KVM_ARM_IRQ_TYPE_SPI:
 701                 if (!irqchip_in_kernel(kvm))
 702                         return -ENXIO;
 703
 704                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 705                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 706                         return -EINVAL;
 707
 708                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 709         }
 710
 711         return -EINVAL;
 712 }
 713
 714 long kvm_arch_vcpu_ioctl(struct file *filp,
 715                          unsigned int ioctl, unsigned long arg)
 716 {
 717         struct kvm_vcpu *vcpu = filp->private_data;
 718         void __user *argp = (void __user *)arg;
 719
 720         switch (ioctl) {
 721         case KVM_ARM_VCPU_INIT: {
 722                 struct kvm_vcpu_init init;
 723
 724                 if (copy_from_user(&init, argp, sizeof(init)))
 725                         return -EFAULT;
 726
 727                 return kvm_vcpu_set_target(vcpu, &init);
 728
 729         }
 730         case KVM_SET_ONE_REG:
 731         case KVM_GET_ONE_REG: {
 732                 struct kvm_one_reg reg;
 733
 734                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 735                         return -ENOEXEC;
 736
 737                 if (copy_from_user(&reg, argp, sizeof(reg)))
 738                         return -EFAULT;
 739                 if (ioctl == KVM_SET_ONE_REG)
 740                         return kvm_arm_set_reg(vcpu, &reg);
 741                 else
 742                         return kvm_arm_get_reg(vcpu, &reg);
 743         }
 744         case KVM_GET_REG_LIST: {
 745                 struct kvm_reg_list __user *user_list = argp;
 746                 struct kvm_reg_list reg_list;
 747                 unsigned n;
 748
 749                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 750                         return -ENOEXEC;
 751
 752                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 753                         return -EFAULT;
 754                 n = reg_list.n;
 755                 reg_list.n = kvm_arm_num_regs(vcpu);
 756                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 757                         return -EFAULT;
 758                 if (n < reg_list.n)
 759                         return -E2BIG;
 760                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 761         }
 762         default:
 763                 return -EINVAL;
 764         }
 765 }
 766
 767 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 768 {
 769         return -EINVAL;
 770 }
 771
 772 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 773                                         struct kvm_arm_device_addr *dev_addr)
 774 {
 775         unsigned long dev_id, type;
 776
 777         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 778                 KVM_ARM_DEVICE_ID_SHIFT;
 779         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 780                 KVM_ARM_DEVICE_TYPE_SHIFT;
 781
 782         switch (dev_id) {
 783         case KVM_ARM_DEVICE_VGIC_V2:
 784                 if (!vgic_present)
 785                         return -ENXIO;
 786                 return kvm_vgic_addr(kvm, type, &dev_addr->addr, true);
 787         default:
 788                 return -ENODEV;
 789         }
 790 }
 791
 792 long kvm_arch_vm_ioctl(struct file *filp,
 793                        unsigned int ioctl, unsigned long arg)
 794 {
 795         struct kvm *kvm = filp->private_data;
 796         void __user *argp = (void __user *)arg;
 797
 798         switch (ioctl) {
 799         case KVM_CREATE_IRQCHIP: {
 800                 if (vgic_present)
 801                         return kvm_vgic_create(kvm);
 802                 else
 803                         return -ENXIO;
 804         }
 805         case KVM_ARM_SET_DEVICE_ADDR: {
 806                 struct kvm_arm_device_addr dev_addr;
 807
 808                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 809                         return -EFAULT;
 810                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 811         }
 812         case KVM_ARM_PREFERRED_TARGET: {
 813                 int err;
 814                 struct kvm_vcpu_init init;
 815
 816                 err = kvm_vcpu_preferred_target(&init);
 817                 if (err)
 818                         return err;
 819
 820                 if (copy_to_user(argp, &init, sizeof(init)))
 821                         return -EFAULT;
 822
 823                 return 0;
 824         }
 825         default:
 826                 return -EINVAL;
 827         }
 828 }
 829
 830 static void cpu_init_hyp_mode(void *dummy)
 831 {
 832         phys_addr_t boot_pgd_ptr;
 833         phys_addr_t pgd_ptr;
 834         unsigned long hyp_stack_ptr;
 835         unsigned long stack_page;
 836         unsigned long vector_ptr;
 837
 838         /* Switch from the HYP stub to our own HYP init vector */
 839         __hyp_set_vectors(kvm_get_idmap_vector());
 840
 841         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 842         pgd_ptr = kvm_mmu_get_httbr();
 843         stack_page = __this_cpu_read(kvm_arm_hyp_stack_page);
 844         hyp_stack_ptr = stack_page + PAGE_SIZE;
 845         vector_ptr = (unsigned long)__kvm_hyp_vector;
 846
 847         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 848 }
 849
 850 static int hyp_init_cpu_notify(struct notifier_block *self,
 851                                unsigned long action, void *cpu)
 852 {
 853         switch (action) {
 854         case CPU_STARTING:
 855         case CPU_STARTING_FROZEN:
 856                 cpu_init_hyp_mode(NULL);
 857                 break;
 858         }
 859
 860         return NOTIFY_OK;
 861 }
 862
 863 static struct notifier_block hyp_init_cpu_nb = {
 864         .notifier_call = hyp_init_cpu_notify,
 865 };
 866
 867 /**
 868  * Inits Hyp-mode on all online CPUs
 869  */
 870 static int init_hyp_mode(void)
 871 {
 872         int cpu;
 873         int err = 0;
 874
 875         /*
 876          * Allocate Hyp PGD and setup Hyp identity mapping
 877          */
 878         err = kvm_mmu_init();
 879         if (err)
 880                 goto out_err;
 881
 882         /*
 883          * It is probably enough to obtain the default on one
 884          * CPU. It's unlikely to be different on the others.
 885          */
 886         hyp_default_vectors = __hyp_get_vectors();
 887
 888         /*
 889          * Allocate stack pages for Hypervisor-mode
 890          */
 891         for_each_possible_cpu(cpu) {
 892                 unsigned long stack_page;
 893
 894                 stack_page = __get_free_page(GFP_KERNEL);
 895                 if (!stack_page) {
 896                         err = -ENOMEM;
 897                         goto out_free_stack_pages;
 898                 }
 899
 900                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 901         }
 902
 903         /*
 904          * Map the Hyp-code called directly from the host
 905          */
 906         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 907         if (err) {
 908                 kvm_err("Cannot map world-switch code\n");
 909                 goto out_free_mappings;
 910         }
 911
 912         /*
 913          * Map the Hyp stack pages
 914          */
 915         for_each_possible_cpu(cpu) {
 916                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 917                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 918
 919                 if (err) {
 920                         kvm_err("Cannot map hyp stack\n");
 921                         goto out_free_mappings;
 922                 }
 923         }
 924
 925         /*
 926          * Map the host CPU structures
 927          */
 928         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 929         if (!kvm_host_cpu_state) {
 930                 err = -ENOMEM;
 931                 kvm_err("Cannot allocate host CPU state\n");
 932                 goto out_free_mappings;
 933         }
 934
 935         for_each_possible_cpu(cpu) {
 936                 kvm_cpu_context_t *cpu_ctxt;
 937
 938                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 939                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 940
 941                 if (err) {
 942                         kvm_err("Cannot map host CPU state: %d\n", err);
 943                         goto out_free_context;
 944                 }
 945         }
 946
 947         /*
 948          * Execute the init code on each CPU.
 949          */
 950         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 951
 952         /*
 953          * Init HYP view of VGIC
 954          */
 955         err = kvm_vgic_hyp_init();
 956         if (err)
 957                 goto out_free_context;
 958
 959 #ifdef CONFIG_KVM_ARM_VGIC
 960                 vgic_present = true;
 961 #endif
 962
 963         /*
 964          * Init HYP architected timer support
 965          */
 966         err = kvm_timer_hyp_init();
 967         if (err)
 968                 goto out_free_mappings;
 969
 970 #ifndef CONFIG_HOTPLUG_CPU
 971         free_boot_hyp_pgd();
 972 #endif
 973
 974         kvm_perf_init();
 975
 976         kvm_info("Hyp mode initialized successfully\n");
 977
 978         return 0;
 979 out_free_context:
 980         free_percpu(kvm_host_cpu_state);
 981 out_free_mappings:
 982         free_hyp_pgds();
 983 out_free_stack_pages:
 984         for_each_possible_cpu(cpu)
 985                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 986 out_err:
 987         kvm_err("error initializing Hyp mode: %d\n", err);
 988         return err;
 989 }
 990
 991 static void check_kvm_target_cpu(void *ret)
 992 {
 993         *(int *)ret = kvm_target_cpu();
 994 }
 995
 996 /**
 997  * Initialize Hyp-mode and memory mappings on all CPUs.
 998  */
 999 int kvm_arch_init(void *opaque)
1000 {
1001         int err;
1002         int ret, cpu;
1003
1004         if (!is_hyp_mode_available()) {
1005                 kvm_err("HYP mode not available\n");
1006                 return -ENODEV;
1007         }
1008
1009         for_each_online_cpu(cpu) {
1010                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1011                 if (ret < 0) {
1012                         kvm_err("Error, CPU %d not supported!\n", cpu);
1013                         return -ENODEV;
1014                 }
1015         }
1016
1017         err = init_hyp_mode();
1018         if (err)
1019                 goto out_err;
1020
1021         err = register_cpu_notifier(&hyp_init_cpu_nb);
1022         if (err) {
1023                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1024                 goto out_err;
1025         }
1026
1027         kvm_coproc_table_init();
1028         return 0;
1029 out_err:
1030         return err;
1031 }
1032
1033 /* NOP: Compiling as a module not supported */
1034 void kvm_arch_exit(void)
1035 {
1036         kvm_perf_teardown();
1037 }
1038
1039 static int arm_init(void)
1040 {
1041         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1042         return rc;
1043 }
1044
1045 module_init(arm_init);