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/errno.h>
  20 #include <linux/err.h>
  21 #include <linux/kvm_host.h>
  22 #include <linux/module.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/fs.h>
  25 #include <linux/mman.h>
  26 #include <linux/sched.h>
  27 #include <linux/kvm.h>
  28 #include <trace/events/kvm.h>
  29
  30 #define CREATE_TRACE_POINTS
  31 #include "trace.h"
  32
  33 #include <asm/unified.h>
  34 #include <asm/uaccess.h>
  35 #include <asm/ptrace.h>
  36 #include <asm/mman.h>
  37 #include <asm/cputype.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/opcodes.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 struct vfp_hard_struct __percpu *kvm_host_vfp_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* The VMID used in the VTTBR */
  57 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  58 static u8 kvm_next_vmid;
  59 static DEFINE_SPINLOCK(kvm_vmid_lock);
  60
  61 int kvm_arch_hardware_enable(void *garbage)
  62 {
  63         return 0;
  64 }
  65
  66 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  67 {
  68         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  69 }
  70
  71 void kvm_arch_hardware_disable(void *garbage)
  72 {
  73 }
  74
  75 int kvm_arch_hardware_setup(void)
  76 {
  77         return 0;
  78 }
  79
  80 void kvm_arch_hardware_unsetup(void)
  81 {
  82 }
  83
  84 void kvm_arch_check_processor_compat(void *rtn)
  85 {
  86         *(int *)rtn = 0;
  87 }
  88
  89 void kvm_arch_sync_events(struct kvm *kvm)
  90 {
  91 }
  92
  93 /**
  94  * kvm_arch_init_vm - initializes a VM data structure
  95  * @kvm:        pointer to the KVM struct
  96  */
  97 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
  98 {
  99         int ret = 0;
 100
 101         if (type)
 102                 return -EINVAL;
 103
 104         ret = kvm_alloc_stage2_pgd(kvm);
 105         if (ret)
 106                 goto out_fail_alloc;
 107
 108         ret = create_hyp_mappings(kvm, kvm + 1);
 109         if (ret)
 110                 goto out_free_stage2_pgd;
 111
 112         /* Mark the initial VMID generation invalid */
 113         kvm->arch.vmid_gen = 0;
 114
 115         return ret;
 116 out_free_stage2_pgd:
 117         kvm_free_stage2_pgd(kvm);
 118 out_fail_alloc:
 119         return ret;
 120 }
 121
 122 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 123 {
 124         return VM_FAULT_SIGBUS;
 125 }
 126
 127 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
 128                            struct kvm_memory_slot *dont)
 129 {
 130 }
 131
 132 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
 133 {
 134         return 0;
 135 }
 136
 137 /**
 138  * kvm_arch_destroy_vm - destroy the VM data structure
 139  * @kvm:        pointer to the KVM struct
 140  */
 141 void kvm_arch_destroy_vm(struct kvm *kvm)
 142 {
 143         int i;
 144
 145         kvm_free_stage2_pgd(kvm);
 146
 147         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 148                 if (kvm->vcpus[i]) {
 149                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 150                         kvm->vcpus[i] = NULL;
 151                 }
 152         }
 153 }
 154
 155 int kvm_dev_ioctl_check_extension(long ext)
 156 {
 157         int r;
 158         switch (ext) {
 159         case KVM_CAP_USER_MEMORY:
 160         case KVM_CAP_SYNC_MMU:
 161         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 162         case KVM_CAP_ONE_REG:
 163                 r = 1;
 164                 break;
 165         case KVM_CAP_COALESCED_MMIO:
 166                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 167                 break;
 168         case KVM_CAP_NR_VCPUS:
 169                 r = num_online_cpus();
 170                 break;
 171         case KVM_CAP_MAX_VCPUS:
 172                 r = KVM_MAX_VCPUS;
 173                 break;
 174         default:
 175                 r = 0;
 176                 break;
 177         }
 178         return r;
 179 }
 180
 181 long kvm_arch_dev_ioctl(struct file *filp,
 182                         unsigned int ioctl, unsigned long arg)
 183 {
 184         return -EINVAL;
 185 }
 186
 187 int kvm_arch_set_memory_region(struct kvm *kvm,
 188                                struct kvm_userspace_memory_region *mem,
 189                                struct kvm_memory_slot old,
 190                                int user_alloc)
 191 {
 192         return 0;
 193 }
 194
 195 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 196                                    struct kvm_memory_slot *memslot,
 197                                    struct kvm_memory_slot old,
 198                                    struct kvm_userspace_memory_region *mem,
 199                                    int user_alloc)
 200 {
 201         return 0;
 202 }
 203
 204 void kvm_arch_commit_memory_region(struct kvm *kvm,
 205                                    struct kvm_userspace_memory_region *mem,
 206                                    struct kvm_memory_slot old,
 207                                    int user_alloc)
 208 {
 209 }
 210
 211 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 212 {
 213 }
 214
 215 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 216                                    struct kvm_memory_slot *slot)
 217 {
 218 }
 219
 220 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 221 {
 222         int err;
 223         struct kvm_vcpu *vcpu;
 224
 225         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 226         if (!vcpu) {
 227                 err = -ENOMEM;
 228                 goto out;
 229         }
 230
 231         err = kvm_vcpu_init(vcpu, kvm, id);
 232         if (err)
 233                 goto free_vcpu;
 234
 235         err = create_hyp_mappings(vcpu, vcpu + 1);
 236         if (err)
 237                 goto vcpu_uninit;
 238
 239         return vcpu;
 240 vcpu_uninit:
 241         kvm_vcpu_uninit(vcpu);
 242 free_vcpu:
 243         kmem_cache_free(kvm_vcpu_cache, vcpu);
 244 out:
 245         return ERR_PTR(err);
 246 }
 247
 248 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 249 {
 250         return 0;
 251 }
 252
 253 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 254 {
 255         kvm_mmu_free_memory_caches(vcpu);
 256         kmem_cache_free(kvm_vcpu_cache, vcpu);
 257 }
 258
 259 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 260 {
 261         kvm_arch_vcpu_free(vcpu);
 262 }
 263
 264 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 265 {
 266         return 0;
 267 }
 268
 269 int __attribute_const__ kvm_target_cpu(void)
 270 {
 271         unsigned long implementor = read_cpuid_implementor();
 272         unsigned long part_number = read_cpuid_part_number();
 273
 274         if (implementor != ARM_CPU_IMP_ARM)
 275                 return -EINVAL;
 276
 277         switch (part_number) {
 278         case ARM_CPU_PART_CORTEX_A15:
 279                 return KVM_ARM_TARGET_CORTEX_A15;
 280         default:
 281                 return -EINVAL;
 282         }
 283 }
 284
 285 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 286 {
 287         /* Force users to call KVM_ARM_VCPU_INIT */
 288         vcpu->arch.target = -1;
 289         return 0;
 290 }
 291
 292 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 293 {
 294 }
 295
 296 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 297 {
 298         vcpu->cpu = cpu;
 299         vcpu->arch.vfp_host = this_cpu_ptr(kvm_host_vfp_state);
 300
 301         /*
 302          * Check whether this vcpu requires the cache to be flushed on
 303          * this physical CPU. This is a consequence of doing dcache
 304          * operations by set/way on this vcpu. We do it here to be in
 305          * a non-preemptible section.
 306          */
 307         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 308                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 309 }
 310
 311 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 312 {
 313 }
 314
 315 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 316                                         struct kvm_guest_debug *dbg)
 317 {
 318         return -EINVAL;
 319 }
 320
 321
 322 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 323                                     struct kvm_mp_state *mp_state)
 324 {
 325         return -EINVAL;
 326 }
 327
 328 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 329                                     struct kvm_mp_state *mp_state)
 330 {
 331         return -EINVAL;
 332 }
 333
 334 /**
 335  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 336  * @v:          The VCPU pointer
 337  *
 338  * If the guest CPU is not waiting for interrupts or an interrupt line is
 339  * asserted, the CPU is by definition runnable.
 340  */
 341 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 342 {
 343         return !!v->arch.irq_lines;
 344 }
 345
 346 /* Just ensure a guest exit from a particular CPU */
 347 static void exit_vm_noop(void *info)
 348 {
 349 }
 350
 351 void force_vm_exit(const cpumask_t *mask)
 352 {
 353         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 354 }
 355
 356 /**
 357  * need_new_vmid_gen - check that the VMID is still valid
 358  * @kvm: The VM's VMID to checkt
 359  *
 360  * return true if there is a new generation of VMIDs being used
 361  *
 362  * The hardware supports only 256 values with the value zero reserved for the
 363  * host, so we check if an assigned value belongs to a previous generation,
 364  * which which requires us to assign a new value. If we're the first to use a
 365  * VMID for the new generation, we must flush necessary caches and TLBs on all
 366  * CPUs.
 367  */
 368 static bool need_new_vmid_gen(struct kvm *kvm)
 369 {
 370         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 371 }
 372
 373 /**
 374  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 375  * @kvm The guest that we are about to run
 376  *
 377  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 378  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 379  * caches and TLBs.
 380  */
 381 static void update_vttbr(struct kvm *kvm)
 382 {
 383         phys_addr_t pgd_phys;
 384         u64 vmid;
 385
 386         if (!need_new_vmid_gen(kvm))
 387                 return;
 388
 389         spin_lock(&kvm_vmid_lock);
 390
 391         /*
 392          * We need to re-check the vmid_gen here to ensure that if another vcpu
 393          * already allocated a valid vmid for this vm, then this vcpu should
 394          * use the same vmid.
 395          */
 396         if (!need_new_vmid_gen(kvm)) {
 397                 spin_unlock(&kvm_vmid_lock);
 398                 return;
 399         }
 400
 401         /* First user of a new VMID generation? */
 402         if (unlikely(kvm_next_vmid == 0)) {
 403                 atomic64_inc(&kvm_vmid_gen);
 404                 kvm_next_vmid = 1;
 405
 406                 /*
 407                  * On SMP we know no other CPUs can use this CPU's or each
 408                  * other's VMID after force_vm_exit returns since the
 409                  * kvm_vmid_lock blocks them from reentry to the guest.
 410                  */
 411                 force_vm_exit(cpu_all_mask);
 412                 /*
 413                  * Now broadcast TLB + ICACHE invalidation over the inner
 414                  * shareable domain to make sure all data structures are
 415                  * clean.
 416                  */
 417                 kvm_call_hyp(__kvm_flush_vm_context);
 418         }
 419
 420         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 421         kvm->arch.vmid = kvm_next_vmid;
 422         kvm_next_vmid++;
 423
 424         /* update vttbr to be used with the new vmid */
 425         pgd_phys = virt_to_phys(kvm->arch.pgd);
 426         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 427         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 428         kvm->arch.vttbr |= vmid;
 429
 430         spin_unlock(&kvm_vmid_lock);
 431 }
 432
 433 static int handle_svc_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run)
 434 {
 435         /* SVC called from Hyp mode should never get here */
 436         kvm_debug("SVC called from Hyp mode shouldn't go here\n");
 437         BUG();
 438         return -EINVAL; /* Squash warning */
 439 }
 440
 441 static int handle_hvc(struct kvm_vcpu *vcpu, struct kvm_run *run)
 442 {
 443         trace_kvm_hvc(*vcpu_pc(vcpu), *vcpu_reg(vcpu, 0),
 444                       vcpu->arch.hsr & HSR_HVC_IMM_MASK);
 445
 446         kvm_inject_undefined(vcpu);
 447         return 1;
 448 }
 449
 450 static int handle_smc(struct kvm_vcpu *vcpu, struct kvm_run *run)
 451 {
 452         /* We don't support SMC; don't do that. */
 453         kvm_debug("smc: at %08x", *vcpu_pc(vcpu));
 454         kvm_inject_undefined(vcpu);
 455         return 1;
 456 }
 457
 458 static int handle_pabt_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run)
 459 {
 460         /* The hypervisor should never cause aborts */
 461         kvm_err("Prefetch Abort taken from Hyp mode at %#08x (HSR: %#08x)\n",
 462                 vcpu->arch.hxfar, vcpu->arch.hsr);
 463         return -EFAULT;
 464 }
 465
 466 static int handle_dabt_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run)
 467 {
 468         /* This is either an error in the ws. code or an external abort */
 469         kvm_err("Data Abort taken from Hyp mode at %#08x (HSR: %#08x)\n",
 470                 vcpu->arch.hxfar, vcpu->arch.hsr);
 471         return -EFAULT;
 472 }
 473
 474 typedef int (*exit_handle_fn)(struct kvm_vcpu *, struct kvm_run *);
 475 static exit_handle_fn arm_exit_handlers[] = {
 476         [HSR_EC_WFI]            = kvm_handle_wfi,
 477         [HSR_EC_CP15_32]        = kvm_handle_cp15_32,
 478         [HSR_EC_CP15_64]        = kvm_handle_cp15_64,
 479         [HSR_EC_CP14_MR]        = kvm_handle_cp14_access,
 480         [HSR_EC_CP14_LS]        = kvm_handle_cp14_load_store,
 481         [HSR_EC_CP14_64]        = kvm_handle_cp14_access,
 482         [HSR_EC_CP_0_13]        = kvm_handle_cp_0_13_access,
 483         [HSR_EC_CP10_ID]        = kvm_handle_cp10_id,
 484         [HSR_EC_SVC_HYP]        = handle_svc_hyp,
 485         [HSR_EC_HVC]            = handle_hvc,
 486         [HSR_EC_SMC]            = handle_smc,
 487         [HSR_EC_IABT]           = kvm_handle_guest_abort,
 488         [HSR_EC_IABT_HYP]       = handle_pabt_hyp,
 489         [HSR_EC_DABT]           = kvm_handle_guest_abort,
 490         [HSR_EC_DABT_HYP]       = handle_dabt_hyp,
 491 };
 492
 493 /*
 494  * A conditional instruction is allowed to trap, even though it
 495  * wouldn't be executed.  So let's re-implement the hardware, in
 496  * software!
 497  */
 498 static bool kvm_condition_valid(struct kvm_vcpu *vcpu)
 499 {
 500         unsigned long cpsr, cond, insn;
 501
 502         /*
 503          * Exception Code 0 can only happen if we set HCR.TGE to 1, to
 504          * catch undefined instructions, and then we won't get past
 505          * the arm_exit_handlers test anyway.
 506          */
 507         BUG_ON(((vcpu->arch.hsr & HSR_EC) >> HSR_EC_SHIFT) == 0);
 508
 509         /* Top two bits non-zero?  Unconditional. */
 510         if (vcpu->arch.hsr >> 30)
 511                 return true;
 512
 513         cpsr = *vcpu_cpsr(vcpu);
 514
 515         /* Is condition field valid? */
 516         if ((vcpu->arch.hsr & HSR_CV) >> HSR_CV_SHIFT)
 517                 cond = (vcpu->arch.hsr & HSR_COND) >> HSR_COND_SHIFT;
 518         else {
 519                 /* This can happen in Thumb mode: examine IT state. */
 520                 unsigned long it;
 521
 522                 it = ((cpsr >> 8) & 0xFC) | ((cpsr >> 25) & 0x3);
 523
 524                 /* it == 0 => unconditional. */
 525                 if (it == 0)
 526                         return true;
 527
 528                 /* The cond for this insn works out as the top 4 bits. */
 529                 cond = (it >> 4);
 530         }
 531
 532         /* Shift makes it look like an ARM-mode instruction */
 533         insn = cond << 28;
 534         return arm_check_condition(insn, cpsr) != ARM_OPCODE_CONDTEST_FAIL;
 535 }
 536
 537 /*
 538  * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on
 539  * proper exit to QEMU.
 540  */
 541 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
 542                        int exception_index)
 543 {
 544         unsigned long hsr_ec;
 545
 546         switch (exception_index) {
 547         case ARM_EXCEPTION_IRQ:
 548                 return 1;
 549         case ARM_EXCEPTION_UNDEFINED:
 550                 kvm_err("Undefined exception in Hyp mode at: %#08x\n",
 551                         vcpu->arch.hyp_pc);
 552                 BUG();
 553                 panic("KVM: Hypervisor undefined exception!\n");
 554         case ARM_EXCEPTION_DATA_ABORT:
 555         case ARM_EXCEPTION_PREF_ABORT:
 556         case ARM_EXCEPTION_HVC:
 557                 hsr_ec = (vcpu->arch.hsr & HSR_EC) >> HSR_EC_SHIFT;
 558
 559                 if (hsr_ec >= ARRAY_SIZE(arm_exit_handlers)
 560                     || !arm_exit_handlers[hsr_ec]) {
 561                         kvm_err("Unkown exception class: %#08lx, "
 562                                 "hsr: %#08x\n", hsr_ec,
 563                                 (unsigned int)vcpu->arch.hsr);
 564                         BUG();
 565                 }
 566
 567                 /*
 568                  * See ARM ARM B1.14.1: "Hyp traps on instructions
 569                  * that fail their condition code check"
 570                  */
 571                 if (!kvm_condition_valid(vcpu)) {
 572                         bool is_wide = vcpu->arch.hsr & HSR_IL;
 573                         kvm_skip_instr(vcpu, is_wide);
 574                         return 1;
 575                 }
 576
 577                 return arm_exit_handlers[hsr_ec](vcpu, run);
 578         default:
 579                 kvm_pr_unimpl("Unsupported exception type: %d",
 580                               exception_index);
 581                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 582                 return 0;
 583         }
 584 }
 585
 586 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 587 {
 588         if (likely(vcpu->arch.has_run_once))
 589                 return 0;
 590
 591         vcpu->arch.has_run_once = true;
 592         return 0;
 593 }
 594
 595 /**
 596  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 597  * @vcpu:       The VCPU pointer
 598  * @run:        The kvm_run structure pointer used for userspace state exchange
 599  *
 600  * This function is called through the VCPU_RUN ioctl called from user space. It
 601  * will execute VM code in a loop until the time slice for the process is used
 602  * or some emulation is needed from user space in which case the function will
 603  * return with return value 0 and with the kvm_run structure filled in with the
 604  * required data for the requested emulation.
 605  */
 606 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 607 {
 608         int ret;
 609         sigset_t sigsaved;
 610
 611         /* Make sure they initialize the vcpu with KVM_ARM_VCPU_INIT */
 612         if (unlikely(vcpu->arch.target < 0))
 613                 return -ENOEXEC;
 614
 615         ret = kvm_vcpu_first_run_init(vcpu);
 616         if (ret)
 617                 return ret;
 618
 619         if (vcpu->sigset_active)
 620                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 621
 622         ret = 1;
 623         run->exit_reason = KVM_EXIT_UNKNOWN;
 624         while (ret > 0) {
 625                 /*
 626                  * Check conditions before entering the guest
 627                  */
 628                 cond_resched();
 629
 630                 update_vttbr(vcpu->kvm);
 631
 632                 local_irq_disable();
 633
 634                 /*
 635                  * Re-check atomic conditions
 636                  */
 637                 if (signal_pending(current)) {
 638                         ret = -EINTR;
 639                         run->exit_reason = KVM_EXIT_INTR;
 640                 }
 641
 642                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 643                         local_irq_enable();
 644                         continue;
 645                 }
 646
 647                 /**************************************************************
 648                  * Enter the guest
 649                  */
 650                 trace_kvm_entry(*vcpu_pc(vcpu));
 651                 kvm_guest_enter();
 652                 vcpu->mode = IN_GUEST_MODE;
 653
 654                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 655
 656                 vcpu->mode = OUTSIDE_GUEST_MODE;
 657                 vcpu->arch.last_pcpu = smp_processor_id();
 658                 kvm_guest_exit();
 659                 trace_kvm_exit(*vcpu_pc(vcpu));
 660                 /*
 661                  * We may have taken a host interrupt in HYP mode (ie
 662                  * while executing the guest). This interrupt is still
 663                  * pending, as we haven't serviced it yet!
 664                  *
 665                  * We're now back in SVC mode, with interrupts
 666                  * disabled.  Enabling the interrupts now will have
 667                  * the effect of taking the interrupt again, in SVC
 668                  * mode this time.
 669                  */
 670                 local_irq_enable();
 671
 672                 /*
 673                  * Back from guest
 674                  *************************************************************/
 675
 676                 ret = handle_exit(vcpu, run, ret);
 677         }
 678
 679         if (vcpu->sigset_active)
 680                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 681         return ret;
 682 }
 683
 684 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 685 {
 686         int bit_index;
 687         bool set;
 688         unsigned long *ptr;
 689
 690         if (number == KVM_ARM_IRQ_CPU_IRQ)
 691                 bit_index = __ffs(HCR_VI);
 692         else /* KVM_ARM_IRQ_CPU_FIQ */
 693                 bit_index = __ffs(HCR_VF);
 694
 695         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 696         if (level)
 697                 set = test_and_set_bit(bit_index, ptr);
 698         else
 699                 set = test_and_clear_bit(bit_index, ptr);
 700
 701         /*
 702          * If we didn't change anything, no need to wake up or kick other CPUs
 703          */
 704         if (set == level)
 705                 return 0;
 706
 707         /*
 708          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 709          * trigger a world-switch round on the running physical CPU to set the
 710          * virtual IRQ/FIQ fields in the HCR appropriately.
 711          */
 712         kvm_vcpu_kick(vcpu);
 713
 714         return 0;
 715 }
 716
 717 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level)
 718 {
 719         u32 irq = irq_level->irq;
 720         unsigned int irq_type, vcpu_idx, irq_num;
 721         int nrcpus = atomic_read(&kvm->online_vcpus);
 722         struct kvm_vcpu *vcpu = NULL;
 723         bool level = irq_level->level;
 724
 725         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 726         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 727         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 728
 729         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 730
 731         if (irq_type != KVM_ARM_IRQ_TYPE_CPU)
 732                 return -EINVAL;
 733
 734         if (vcpu_idx >= nrcpus)
 735                 return -EINVAL;
 736
 737         vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 738         if (!vcpu)
 739                 return -EINVAL;
 740
 741         if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 742                 return -EINVAL;
 743
 744         return vcpu_interrupt_line(vcpu, irq_num, level);
 745 }
 746
 747 long kvm_arch_vcpu_ioctl(struct file *filp,
 748                          unsigned int ioctl, unsigned long arg)
 749 {
 750         struct kvm_vcpu *vcpu = filp->private_data;
 751         void __user *argp = (void __user *)arg;
 752
 753         switch (ioctl) {
 754         case KVM_ARM_VCPU_INIT: {
 755                 struct kvm_vcpu_init init;
 756
 757                 if (copy_from_user(&init, argp, sizeof(init)))
 758                         return -EFAULT;
 759
 760                 return kvm_vcpu_set_target(vcpu, &init);
 761
 762         }
 763         case KVM_SET_ONE_REG:
 764         case KVM_GET_ONE_REG: {
 765                 struct kvm_one_reg reg;
 766                 if (copy_from_user(&reg, argp, sizeof(reg)))
 767                         return -EFAULT;
 768                 if (ioctl == KVM_SET_ONE_REG)
 769                         return kvm_arm_set_reg(vcpu, &reg);
 770                 else
 771                         return kvm_arm_get_reg(vcpu, &reg);
 772         }
 773         case KVM_GET_REG_LIST: {
 774                 struct kvm_reg_list __user *user_list = argp;
 775                 struct kvm_reg_list reg_list;
 776                 unsigned n;
 777
 778                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 779                         return -EFAULT;
 780                 n = reg_list.n;
 781                 reg_list.n = kvm_arm_num_regs(vcpu);
 782                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 783                         return -EFAULT;
 784                 if (n < reg_list.n)
 785                         return -E2BIG;
 786                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 787         }
 788         default:
 789                 return -EINVAL;
 790         }
 791 }
 792
 793 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 794 {
 795         return -EINVAL;
 796 }
 797
 798 long kvm_arch_vm_ioctl(struct file *filp,
 799                        unsigned int ioctl, unsigned long arg)
 800 {
 801         return -EINVAL;
 802 }
 803
 804 static void cpu_init_hyp_mode(void *vector)
 805 {
 806         unsigned long long pgd_ptr;
 807         unsigned long pgd_low, pgd_high;
 808         unsigned long hyp_stack_ptr;
 809         unsigned long stack_page;
 810         unsigned long vector_ptr;
 811
 812         /* Switch from the HYP stub to our own HYP init vector */
 813         __hyp_set_vectors((unsigned long)vector);
 814
 815         pgd_ptr = (unsigned long long)kvm_mmu_get_httbr();
 816         pgd_low = (pgd_ptr & ((1ULL << 32) - 1));
 817         pgd_high = (pgd_ptr >> 32ULL);
 818         stack_page = __get_cpu_var(kvm_arm_hyp_stack_page);
 819         hyp_stack_ptr = stack_page + PAGE_SIZE;
 820         vector_ptr = (unsigned long)__kvm_hyp_vector;
 821
 822         /*
 823          * Call initialization code, and switch to the full blown
 824          * HYP code. The init code doesn't need to preserve these registers as
 825          * r1-r3 and r12 are already callee save according to the AAPCS.
 826          * Note that we slightly misuse the prototype by casing the pgd_low to
 827          * a void *.
 828          */
 829         kvm_call_hyp((void *)pgd_low, pgd_high, hyp_stack_ptr, vector_ptr);
 830 }
 831
 832 /**
 833  * Inits Hyp-mode on all online CPUs
 834  */
 835 static int init_hyp_mode(void)
 836 {
 837         phys_addr_t init_phys_addr;
 838         int cpu;
 839         int err = 0;
 840
 841         /*
 842          * Allocate Hyp PGD and setup Hyp identity mapping
 843          */
 844         err = kvm_mmu_init();
 845         if (err)
 846                 goto out_err;
 847
 848         /*
 849          * It is probably enough to obtain the default on one
 850          * CPU. It's unlikely to be different on the others.
 851          */
 852         hyp_default_vectors = __hyp_get_vectors();
 853
 854         /*
 855          * Allocate stack pages for Hypervisor-mode
 856          */
 857         for_each_possible_cpu(cpu) {
 858                 unsigned long stack_page;
 859
 860                 stack_page = __get_free_page(GFP_KERNEL);
 861                 if (!stack_page) {
 862                         err = -ENOMEM;
 863                         goto out_free_stack_pages;
 864                 }
 865
 866                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 867         }
 868
 869         /*
 870          * Execute the init code on each CPU.
 871          *
 872          * Note: The stack is not mapped yet, so don't do anything else than
 873          * initializing the hypervisor mode on each CPU using a local stack
 874          * space for temporary storage.
 875          */
 876         init_phys_addr = virt_to_phys(__kvm_hyp_init);
 877         for_each_online_cpu(cpu) {
 878                 smp_call_function_single(cpu, cpu_init_hyp_mode,
 879                                          (void *)(long)init_phys_addr, 1);
 880         }
 881
 882         /*
 883          * Unmap the identity mapping
 884          */
 885         kvm_clear_hyp_idmap();
 886
 887         /*
 888          * Map the Hyp-code called directly from the host
 889          */
 890         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 891         if (err) {
 892                 kvm_err("Cannot map world-switch code\n");
 893                 goto out_free_mappings;
 894         }
 895
 896         /*
 897          * Map the Hyp stack pages
 898          */
 899         for_each_possible_cpu(cpu) {
 900                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 901                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 902
 903                 if (err) {
 904                         kvm_err("Cannot map hyp stack\n");
 905                         goto out_free_mappings;
 906                 }
 907         }
 908
 909         /*
 910          * Map the host VFP structures
 911          */
 912         kvm_host_vfp_state = alloc_percpu(struct vfp_hard_struct);
 913         if (!kvm_host_vfp_state) {
 914                 err = -ENOMEM;
 915                 kvm_err("Cannot allocate host VFP state\n");
 916                 goto out_free_mappings;
 917         }
 918
 919         for_each_possible_cpu(cpu) {
 920                 struct vfp_hard_struct *vfp;
 921
 922                 vfp = per_cpu_ptr(kvm_host_vfp_state, cpu);
 923                 err = create_hyp_mappings(vfp, vfp + 1);
 924
 925                 if (err) {
 926                         kvm_err("Cannot map host VFP state: %d\n", err);
 927                         goto out_free_vfp;
 928                 }
 929         }
 930
 931         kvm_info("Hyp mode initialized successfully\n");
 932         return 0;
 933 out_free_vfp:
 934         free_percpu(kvm_host_vfp_state);
 935 out_free_mappings:
 936         free_hyp_pmds();
 937 out_free_stack_pages:
 938         for_each_possible_cpu(cpu)
 939                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 940 out_err:
 941         kvm_err("error initializing Hyp mode: %d\n", err);
 942         return err;
 943 }
 944
 945 /**
 946  * Initialize Hyp-mode and memory mappings on all CPUs.
 947  */
 948 int kvm_arch_init(void *opaque)
 949 {
 950         int err;
 951
 952         if (!is_hyp_mode_available()) {
 953                 kvm_err("HYP mode not available\n");
 954                 return -ENODEV;
 955         }
 956
 957         if (kvm_target_cpu() < 0) {
 958                 kvm_err("Target CPU not supported!\n");
 959                 return -ENODEV;
 960         }
 961
 962         err = init_hyp_mode();
 963         if (err)
 964                 goto out_err;
 965
 966         kvm_coproc_table_init();
 967         return 0;
 968 out_err:
 969         return err;
 970 }
 971
 972 /* NOP: Compiling as a module not supported */
 973 void kvm_arch_exit(void)
 974 {
 975 }
 976
 977 static int arm_init(void)
 978 {
 979         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
 980         return rc;
 981 }
 982
 983 module_init(arm_init);