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/kvm_psci.h>
  47 #include <asm/opcodes.h>
  48
  49 #ifdef REQUIRES_VIRT
  50 __asm__(".arch_extension        virt");
  51 #endif
  52
  53 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  54 static struct vfp_hard_struct __percpu *kvm_host_vfp_state;
  55 static unsigned long hyp_default_vectors;
  56
  57 /* The VMID used in the VTTBR */
  58 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  59 static u8 kvm_next_vmid;
  60 static DEFINE_SPINLOCK(kvm_vmid_lock);
  61
  62 int kvm_arch_hardware_enable(void *garbage)
  63 {
  64         return 0;
  65 }
  66
  67 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  68 {
  69         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  70 }
  71
  72 void kvm_arch_hardware_disable(void *garbage)
  73 {
  74 }
  75
  76 int kvm_arch_hardware_setup(void)
  77 {
  78         return 0;
  79 }
  80
  81 void kvm_arch_hardware_unsetup(void)
  82 {
  83 }
  84
  85 void kvm_arch_check_processor_compat(void *rtn)
  86 {
  87         *(int *)rtn = 0;
  88 }
  89
  90 void kvm_arch_sync_events(struct kvm *kvm)
  91 {
  92 }
  93
  94 /**
  95  * kvm_arch_init_vm - initializes a VM data structure
  96  * @kvm:        pointer to the KVM struct
  97  */
  98 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
  99 {
 100         int ret = 0;
 101
 102         if (type)
 103                 return -EINVAL;
 104
 105         ret = kvm_alloc_stage2_pgd(kvm);
 106         if (ret)
 107                 goto out_fail_alloc;
 108
 109         ret = create_hyp_mappings(kvm, kvm + 1);
 110         if (ret)
 111                 goto out_free_stage2_pgd;
 112
 113         /* Mark the initial VMID generation invalid */
 114         kvm->arch.vmid_gen = 0;
 115
 116         return ret;
 117 out_free_stage2_pgd:
 118         kvm_free_stage2_pgd(kvm);
 119 out_fail_alloc:
 120         return ret;
 121 }
 122
 123 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 124 {
 125         return VM_FAULT_SIGBUS;
 126 }
 127
 128 void kvm_arch_free_memslot(struct kvm_memory_slot *free,
 129                            struct kvm_memory_slot *dont)
 130 {
 131 }
 132
 133 int kvm_arch_create_memslot(struct kvm_memory_slot *slot, unsigned long npages)
 134 {
 135         return 0;
 136 }
 137
 138 /**
 139  * kvm_arch_destroy_vm - destroy the VM data structure
 140  * @kvm:        pointer to the KVM struct
 141  */
 142 void kvm_arch_destroy_vm(struct kvm *kvm)
 143 {
 144         int i;
 145
 146         kvm_free_stage2_pgd(kvm);
 147
 148         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 149                 if (kvm->vcpus[i]) {
 150                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 151                         kvm->vcpus[i] = NULL;
 152                 }
 153         }
 154 }
 155
 156 int kvm_dev_ioctl_check_extension(long ext)
 157 {
 158         int r;
 159         switch (ext) {
 160         case KVM_CAP_USER_MEMORY:
 161         case KVM_CAP_SYNC_MMU:
 162         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 163         case KVM_CAP_ONE_REG:
 164         case KVM_CAP_ARM_PSCI:
 165                 r = 1;
 166                 break;
 167         case KVM_CAP_COALESCED_MMIO:
 168                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 169                 break;
 170         case KVM_CAP_NR_VCPUS:
 171                 r = num_online_cpus();
 172                 break;
 173         case KVM_CAP_MAX_VCPUS:
 174                 r = KVM_MAX_VCPUS;
 175                 break;
 176         default:
 177                 r = 0;
 178                 break;
 179         }
 180         return r;
 181 }
 182
 183 long kvm_arch_dev_ioctl(struct file *filp,
 184                         unsigned int ioctl, unsigned long arg)
 185 {
 186         return -EINVAL;
 187 }
 188
 189 int kvm_arch_set_memory_region(struct kvm *kvm,
 190                                struct kvm_userspace_memory_region *mem,
 191                                struct kvm_memory_slot old,
 192                                int user_alloc)
 193 {
 194         return 0;
 195 }
 196
 197 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 198                                    struct kvm_memory_slot *memslot,
 199                                    struct kvm_memory_slot old,
 200                                    struct kvm_userspace_memory_region *mem,
 201                                    int user_alloc)
 202 {
 203         return 0;
 204 }
 205
 206 void kvm_arch_commit_memory_region(struct kvm *kvm,
 207                                    struct kvm_userspace_memory_region *mem,
 208                                    struct kvm_memory_slot old,
 209                                    int user_alloc)
 210 {
 211 }
 212
 213 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 214 {
 215 }
 216
 217 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 218                                    struct kvm_memory_slot *slot)
 219 {
 220 }
 221
 222 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 223 {
 224         int err;
 225         struct kvm_vcpu *vcpu;
 226
 227         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 228         if (!vcpu) {
 229                 err = -ENOMEM;
 230                 goto out;
 231         }
 232
 233         err = kvm_vcpu_init(vcpu, kvm, id);
 234         if (err)
 235                 goto free_vcpu;
 236
 237         err = create_hyp_mappings(vcpu, vcpu + 1);
 238         if (err)
 239                 goto vcpu_uninit;
 240
 241         return vcpu;
 242 vcpu_uninit:
 243         kvm_vcpu_uninit(vcpu);
 244 free_vcpu:
 245         kmem_cache_free(kvm_vcpu_cache, vcpu);
 246 out:
 247         return ERR_PTR(err);
 248 }
 249
 250 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 251 {
 252         return 0;
 253 }
 254
 255 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 256 {
 257         kvm_mmu_free_memory_caches(vcpu);
 258         kmem_cache_free(kvm_vcpu_cache, vcpu);
 259 }
 260
 261 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 262 {
 263         kvm_arch_vcpu_free(vcpu);
 264 }
 265
 266 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 267 {
 268         return 0;
 269 }
 270
 271 int __attribute_const__ kvm_target_cpu(void)
 272 {
 273         unsigned long implementor = read_cpuid_implementor();
 274         unsigned long part_number = read_cpuid_part_number();
 275
 276         if (implementor != ARM_CPU_IMP_ARM)
 277                 return -EINVAL;
 278
 279         switch (part_number) {
 280         case ARM_CPU_PART_CORTEX_A15:
 281                 return KVM_ARM_TARGET_CORTEX_A15;
 282         default:
 283                 return -EINVAL;
 284         }
 285 }
 286
 287 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 288 {
 289         /* Force users to call KVM_ARM_VCPU_INIT */
 290         vcpu->arch.target = -1;
 291         return 0;
 292 }
 293
 294 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 295 {
 296 }
 297
 298 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 299 {
 300         vcpu->cpu = cpu;
 301         vcpu->arch.vfp_host = this_cpu_ptr(kvm_host_vfp_state);
 302
 303         /*
 304          * Check whether this vcpu requires the cache to be flushed on
 305          * this physical CPU. This is a consequence of doing dcache
 306          * operations by set/way on this vcpu. We do it here to be in
 307          * a non-preemptible section.
 308          */
 309         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 310                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 311 }
 312
 313 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 314 {
 315 }
 316
 317 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 318                                         struct kvm_guest_debug *dbg)
 319 {
 320         return -EINVAL;
 321 }
 322
 323
 324 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 325                                     struct kvm_mp_state *mp_state)
 326 {
 327         return -EINVAL;
 328 }
 329
 330 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 331                                     struct kvm_mp_state *mp_state)
 332 {
 333         return -EINVAL;
 334 }
 335
 336 /**
 337  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 338  * @v:          The VCPU pointer
 339  *
 340  * If the guest CPU is not waiting for interrupts or an interrupt line is
 341  * asserted, the CPU is by definition runnable.
 342  */
 343 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 344 {
 345         return !!v->arch.irq_lines;
 346 }
 347
 348 /* Just ensure a guest exit from a particular CPU */
 349 static void exit_vm_noop(void *info)
 350 {
 351 }
 352
 353 void force_vm_exit(const cpumask_t *mask)
 354 {
 355         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 356 }
 357
 358 /**
 359  * need_new_vmid_gen - check that the VMID is still valid
 360  * @kvm: The VM's VMID to checkt
 361  *
 362  * return true if there is a new generation of VMIDs being used
 363  *
 364  * The hardware supports only 256 values with the value zero reserved for the
 365  * host, so we check if an assigned value belongs to a previous generation,
 366  * which which requires us to assign a new value. If we're the first to use a
 367  * VMID for the new generation, we must flush necessary caches and TLBs on all
 368  * CPUs.
 369  */
 370 static bool need_new_vmid_gen(struct kvm *kvm)
 371 {
 372         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 373 }
 374
 375 /**
 376  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 377  * @kvm The guest that we are about to run
 378  *
 379  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 380  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 381  * caches and TLBs.
 382  */
 383 static void update_vttbr(struct kvm *kvm)
 384 {
 385         phys_addr_t pgd_phys;
 386         u64 vmid;
 387
 388         if (!need_new_vmid_gen(kvm))
 389                 return;
 390
 391         spin_lock(&kvm_vmid_lock);
 392
 393         /*
 394          * We need to re-check the vmid_gen here to ensure that if another vcpu
 395          * already allocated a valid vmid for this vm, then this vcpu should
 396          * use the same vmid.
 397          */
 398         if (!need_new_vmid_gen(kvm)) {
 399                 spin_unlock(&kvm_vmid_lock);
 400                 return;
 401         }
 402
 403         /* First user of a new VMID generation? */
 404         if (unlikely(kvm_next_vmid == 0)) {
 405                 atomic64_inc(&kvm_vmid_gen);
 406                 kvm_next_vmid = 1;
 407
 408                 /*
 409                  * On SMP we know no other CPUs can use this CPU's or each
 410                  * other's VMID after force_vm_exit returns since the
 411                  * kvm_vmid_lock blocks them from reentry to the guest.
 412                  */
 413                 force_vm_exit(cpu_all_mask);
 414                 /*
 415                  * Now broadcast TLB + ICACHE invalidation over the inner
 416                  * shareable domain to make sure all data structures are
 417                  * clean.
 418                  */
 419                 kvm_call_hyp(__kvm_flush_vm_context);
 420         }
 421
 422         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 423         kvm->arch.vmid = kvm_next_vmid;
 424         kvm_next_vmid++;
 425
 426         /* update vttbr to be used with the new vmid */
 427         pgd_phys = virt_to_phys(kvm->arch.pgd);
 428         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 429         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 430         kvm->arch.vttbr |= vmid;
 431
 432         spin_unlock(&kvm_vmid_lock);
 433 }
 434
 435 static int handle_svc_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run)
 436 {
 437         /* SVC called from Hyp mode should never get here */
 438         kvm_debug("SVC called from Hyp mode shouldn't go here\n");
 439         BUG();
 440         return -EINVAL; /* Squash warning */
 441 }
 442
 443 static int handle_hvc(struct kvm_vcpu *vcpu, struct kvm_run *run)
 444 {
 445         trace_kvm_hvc(*vcpu_pc(vcpu), *vcpu_reg(vcpu, 0),
 446                       vcpu->arch.hsr & HSR_HVC_IMM_MASK);
 447
 448         if (kvm_psci_call(vcpu))
 449                 return 1;
 450
 451         kvm_inject_undefined(vcpu);
 452         return 1;
 453 }
 454
 455 static int handle_smc(struct kvm_vcpu *vcpu, struct kvm_run *run)
 456 {
 457         if (kvm_psci_call(vcpu))
 458                 return 1;
 459
 460         kvm_inject_undefined(vcpu);
 461         return 1;
 462 }
 463
 464 static int handle_pabt_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run)
 465 {
 466         /* The hypervisor should never cause aborts */
 467         kvm_err("Prefetch Abort taken from Hyp mode at %#08x (HSR: %#08x)\n",
 468                 vcpu->arch.hxfar, vcpu->arch.hsr);
 469         return -EFAULT;
 470 }
 471
 472 static int handle_dabt_hyp(struct kvm_vcpu *vcpu, struct kvm_run *run)
 473 {
 474         /* This is either an error in the ws. code or an external abort */
 475         kvm_err("Data Abort taken from Hyp mode at %#08x (HSR: %#08x)\n",
 476                 vcpu->arch.hxfar, vcpu->arch.hsr);
 477         return -EFAULT;
 478 }
 479
 480 typedef int (*exit_handle_fn)(struct kvm_vcpu *, struct kvm_run *);
 481 static exit_handle_fn arm_exit_handlers[] = {
 482         [HSR_EC_WFI]            = kvm_handle_wfi,
 483         [HSR_EC_CP15_32]        = kvm_handle_cp15_32,
 484         [HSR_EC_CP15_64]        = kvm_handle_cp15_64,
 485         [HSR_EC_CP14_MR]        = kvm_handle_cp14_access,
 486         [HSR_EC_CP14_LS]        = kvm_handle_cp14_load_store,
 487         [HSR_EC_CP14_64]        = kvm_handle_cp14_access,
 488         [HSR_EC_CP_0_13]        = kvm_handle_cp_0_13_access,
 489         [HSR_EC_CP10_ID]        = kvm_handle_cp10_id,
 490         [HSR_EC_SVC_HYP]        = handle_svc_hyp,
 491         [HSR_EC_HVC]            = handle_hvc,
 492         [HSR_EC_SMC]            = handle_smc,
 493         [HSR_EC_IABT]           = kvm_handle_guest_abort,
 494         [HSR_EC_IABT_HYP]       = handle_pabt_hyp,
 495         [HSR_EC_DABT]           = kvm_handle_guest_abort,
 496         [HSR_EC_DABT_HYP]       = handle_dabt_hyp,
 497 };
 498
 499 /*
 500  * A conditional instruction is allowed to trap, even though it
 501  * wouldn't be executed.  So let's re-implement the hardware, in
 502  * software!
 503  */
 504 static bool kvm_condition_valid(struct kvm_vcpu *vcpu)
 505 {
 506         unsigned long cpsr, cond, insn;
 507
 508         /*
 509          * Exception Code 0 can only happen if we set HCR.TGE to 1, to
 510          * catch undefined instructions, and then we won't get past
 511          * the arm_exit_handlers test anyway.
 512          */
 513         BUG_ON(((vcpu->arch.hsr & HSR_EC) >> HSR_EC_SHIFT) == 0);
 514
 515         /* Top two bits non-zero?  Unconditional. */
 516         if (vcpu->arch.hsr >> 30)
 517                 return true;
 518
 519         cpsr = *vcpu_cpsr(vcpu);
 520
 521         /* Is condition field valid? */
 522         if ((vcpu->arch.hsr & HSR_CV) >> HSR_CV_SHIFT)
 523                 cond = (vcpu->arch.hsr & HSR_COND) >> HSR_COND_SHIFT;
 524         else {
 525                 /* This can happen in Thumb mode: examine IT state. */
 526                 unsigned long it;
 527
 528                 it = ((cpsr >> 8) & 0xFC) | ((cpsr >> 25) & 0x3);
 529
 530                 /* it == 0 => unconditional. */
 531                 if (it == 0)
 532                         return true;
 533
 534                 /* The cond for this insn works out as the top 4 bits. */
 535                 cond = (it >> 4);
 536         }
 537
 538         /* Shift makes it look like an ARM-mode instruction */
 539         insn = cond << 28;
 540         return arm_check_condition(insn, cpsr) != ARM_OPCODE_CONDTEST_FAIL;
 541 }
 542
 543 /*
 544  * Return > 0 to return to guest, < 0 on error, 0 (and set exit_reason) on
 545  * proper exit to QEMU.
 546  */
 547 static int handle_exit(struct kvm_vcpu *vcpu, struct kvm_run *run,
 548                        int exception_index)
 549 {
 550         unsigned long hsr_ec;
 551
 552         switch (exception_index) {
 553         case ARM_EXCEPTION_IRQ:
 554                 return 1;
 555         case ARM_EXCEPTION_UNDEFINED:
 556                 kvm_err("Undefined exception in Hyp mode at: %#08x\n",
 557                         vcpu->arch.hyp_pc);
 558                 BUG();
 559                 panic("KVM: Hypervisor undefined exception!\n");
 560         case ARM_EXCEPTION_DATA_ABORT:
 561         case ARM_EXCEPTION_PREF_ABORT:
 562         case ARM_EXCEPTION_HVC:
 563                 hsr_ec = (vcpu->arch.hsr & HSR_EC) >> HSR_EC_SHIFT;
 564
 565                 if (hsr_ec >= ARRAY_SIZE(arm_exit_handlers)
 566                     || !arm_exit_handlers[hsr_ec]) {
 567                         kvm_err("Unkown exception class: %#08lx, "
 568                                 "hsr: %#08x\n", hsr_ec,
 569                                 (unsigned int)vcpu->arch.hsr);
 570                         BUG();
 571                 }
 572
 573                 /*
 574                  * See ARM ARM B1.14.1: "Hyp traps on instructions
 575                  * that fail their condition code check"
 576                  */
 577                 if (!kvm_condition_valid(vcpu)) {
 578                         bool is_wide = vcpu->arch.hsr & HSR_IL;
 579                         kvm_skip_instr(vcpu, is_wide);
 580                         return 1;
 581                 }
 582
 583                 return arm_exit_handlers[hsr_ec](vcpu, run);
 584         default:
 585                 kvm_pr_unimpl("Unsupported exception type: %d",
 586                               exception_index);
 587                 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
 588                 return 0;
 589         }
 590 }
 591
 592 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 593 {
 594         if (likely(vcpu->arch.has_run_once))
 595                 return 0;
 596
 597         vcpu->arch.has_run_once = true;
 598
 599         /*
 600          * Handle the "start in power-off" case by calling into the
 601          * PSCI code.
 602          */
 603         if (test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features)) {
 604                 *vcpu_reg(vcpu, 0) = KVM_PSCI_FN_CPU_OFF;
 605                 kvm_psci_call(vcpu);
 606         }
 607
 608         return 0;
 609 }
 610
 611 static void vcpu_pause(struct kvm_vcpu *vcpu)
 612 {
 613         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 614
 615         wait_event_interruptible(*wq, !vcpu->arch.pause);
 616 }
 617
 618 /**
 619  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 620  * @vcpu:       The VCPU pointer
 621  * @run:        The kvm_run structure pointer used for userspace state exchange
 622  *
 623  * This function is called through the VCPU_RUN ioctl called from user space. It
 624  * will execute VM code in a loop until the time slice for the process is used
 625  * or some emulation is needed from user space in which case the function will
 626  * return with return value 0 and with the kvm_run structure filled in with the
 627  * required data for the requested emulation.
 628  */
 629 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 630 {
 631         int ret;
 632         sigset_t sigsaved;
 633
 634         /* Make sure they initialize the vcpu with KVM_ARM_VCPU_INIT */
 635         if (unlikely(vcpu->arch.target < 0))
 636                 return -ENOEXEC;
 637
 638         ret = kvm_vcpu_first_run_init(vcpu);
 639         if (ret)
 640                 return ret;
 641
 642         if (run->exit_reason == KVM_EXIT_MMIO) {
 643                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 644                 if (ret)
 645                         return ret;
 646         }
 647
 648         if (vcpu->sigset_active)
 649                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 650
 651         ret = 1;
 652         run->exit_reason = KVM_EXIT_UNKNOWN;
 653         while (ret > 0) {
 654                 /*
 655                  * Check conditions before entering the guest
 656                  */
 657                 cond_resched();
 658
 659                 update_vttbr(vcpu->kvm);
 660
 661                 if (vcpu->arch.pause)
 662                         vcpu_pause(vcpu);
 663
 664                 local_irq_disable();
 665
 666                 /*
 667                  * Re-check atomic conditions
 668                  */
 669                 if (signal_pending(current)) {
 670                         ret = -EINTR;
 671                         run->exit_reason = KVM_EXIT_INTR;
 672                 }
 673
 674                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 675                         local_irq_enable();
 676                         continue;
 677                 }
 678
 679                 /**************************************************************
 680                  * Enter the guest
 681                  */
 682                 trace_kvm_entry(*vcpu_pc(vcpu));
 683                 kvm_guest_enter();
 684                 vcpu->mode = IN_GUEST_MODE;
 685
 686                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 687
 688                 vcpu->mode = OUTSIDE_GUEST_MODE;
 689                 vcpu->arch.last_pcpu = smp_processor_id();
 690                 kvm_guest_exit();
 691                 trace_kvm_exit(*vcpu_pc(vcpu));
 692                 /*
 693                  * We may have taken a host interrupt in HYP mode (ie
 694                  * while executing the guest). This interrupt is still
 695                  * pending, as we haven't serviced it yet!
 696                  *
 697                  * We're now back in SVC mode, with interrupts
 698                  * disabled.  Enabling the interrupts now will have
 699                  * the effect of taking the interrupt again, in SVC
 700                  * mode this time.
 701                  */
 702                 local_irq_enable();
 703
 704                 /*
 705                  * Back from guest
 706                  *************************************************************/
 707
 708                 ret = handle_exit(vcpu, run, ret);
 709         }
 710
 711         if (vcpu->sigset_active)
 712                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 713         return ret;
 714 }
 715
 716 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 717 {
 718         int bit_index;
 719         bool set;
 720         unsigned long *ptr;
 721
 722         if (number == KVM_ARM_IRQ_CPU_IRQ)
 723                 bit_index = __ffs(HCR_VI);
 724         else /* KVM_ARM_IRQ_CPU_FIQ */
 725                 bit_index = __ffs(HCR_VF);
 726
 727         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 728         if (level)
 729                 set = test_and_set_bit(bit_index, ptr);
 730         else
 731                 set = test_and_clear_bit(bit_index, ptr);
 732
 733         /*
 734          * If we didn't change anything, no need to wake up or kick other CPUs
 735          */
 736         if (set == level)
 737                 return 0;
 738
 739         /*
 740          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 741          * trigger a world-switch round on the running physical CPU to set the
 742          * virtual IRQ/FIQ fields in the HCR appropriately.
 743          */
 744         kvm_vcpu_kick(vcpu);
 745
 746         return 0;
 747 }
 748
 749 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level)
 750 {
 751         u32 irq = irq_level->irq;
 752         unsigned int irq_type, vcpu_idx, irq_num;
 753         int nrcpus = atomic_read(&kvm->online_vcpus);
 754         struct kvm_vcpu *vcpu = NULL;
 755         bool level = irq_level->level;
 756
 757         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 758         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 759         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 760
 761         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 762
 763         if (irq_type != KVM_ARM_IRQ_TYPE_CPU)
 764                 return -EINVAL;
 765
 766         if (vcpu_idx >= nrcpus)
 767                 return -EINVAL;
 768
 769         vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 770         if (!vcpu)
 771                 return -EINVAL;
 772
 773         if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 774                 return -EINVAL;
 775
 776         return vcpu_interrupt_line(vcpu, irq_num, level);
 777 }
 778
 779 long kvm_arch_vcpu_ioctl(struct file *filp,
 780                          unsigned int ioctl, unsigned long arg)
 781 {
 782         struct kvm_vcpu *vcpu = filp->private_data;
 783         void __user *argp = (void __user *)arg;
 784
 785         switch (ioctl) {
 786         case KVM_ARM_VCPU_INIT: {
 787                 struct kvm_vcpu_init init;
 788
 789                 if (copy_from_user(&init, argp, sizeof(init)))
 790                         return -EFAULT;
 791
 792                 return kvm_vcpu_set_target(vcpu, &init);
 793
 794         }
 795         case KVM_SET_ONE_REG:
 796         case KVM_GET_ONE_REG: {
 797                 struct kvm_one_reg reg;
 798                 if (copy_from_user(&reg, argp, sizeof(reg)))
 799                         return -EFAULT;
 800                 if (ioctl == KVM_SET_ONE_REG)
 801                         return kvm_arm_set_reg(vcpu, &reg);
 802                 else
 803                         return kvm_arm_get_reg(vcpu, &reg);
 804         }
 805         case KVM_GET_REG_LIST: {
 806                 struct kvm_reg_list __user *user_list = argp;
 807                 struct kvm_reg_list reg_list;
 808                 unsigned n;
 809
 810                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 811                         return -EFAULT;
 812                 n = reg_list.n;
 813                 reg_list.n = kvm_arm_num_regs(vcpu);
 814                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 815                         return -EFAULT;
 816                 if (n < reg_list.n)
 817                         return -E2BIG;
 818                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 819         }
 820         default:
 821                 return -EINVAL;
 822         }
 823 }
 824
 825 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 826 {
 827         return -EINVAL;
 828 }
 829
 830 long kvm_arch_vm_ioctl(struct file *filp,
 831                        unsigned int ioctl, unsigned long arg)
 832 {
 833         return -EINVAL;
 834 }
 835
 836 static void cpu_init_hyp_mode(void *vector)
 837 {
 838         unsigned long long pgd_ptr;
 839         unsigned long pgd_low, pgd_high;
 840         unsigned long hyp_stack_ptr;
 841         unsigned long stack_page;
 842         unsigned long vector_ptr;
 843
 844         /* Switch from the HYP stub to our own HYP init vector */
 845         __hyp_set_vectors((unsigned long)vector);
 846
 847         pgd_ptr = (unsigned long long)kvm_mmu_get_httbr();
 848         pgd_low = (pgd_ptr & ((1ULL << 32) - 1));
 849         pgd_high = (pgd_ptr >> 32ULL);
 850         stack_page = __get_cpu_var(kvm_arm_hyp_stack_page);
 851         hyp_stack_ptr = stack_page + PAGE_SIZE;
 852         vector_ptr = (unsigned long)__kvm_hyp_vector;
 853
 854         /*
 855          * Call initialization code, and switch to the full blown
 856          * HYP code. The init code doesn't need to preserve these registers as
 857          * r1-r3 and r12 are already callee save according to the AAPCS.
 858          * Note that we slightly misuse the prototype by casing the pgd_low to
 859          * a void *.
 860          */
 861         kvm_call_hyp((void *)pgd_low, pgd_high, hyp_stack_ptr, vector_ptr);
 862 }
 863
 864 /**
 865  * Inits Hyp-mode on all online CPUs
 866  */
 867 static int init_hyp_mode(void)
 868 {
 869         phys_addr_t init_phys_addr;
 870         int cpu;
 871         int err = 0;
 872
 873         /*
 874          * Allocate Hyp PGD and setup Hyp identity mapping
 875          */
 876         err = kvm_mmu_init();
 877         if (err)
 878                 goto out_err;
 879
 880         /*
 881          * It is probably enough to obtain the default on one
 882          * CPU. It's unlikely to be different on the others.
 883          */
 884         hyp_default_vectors = __hyp_get_vectors();
 885
 886         /*
 887          * Allocate stack pages for Hypervisor-mode
 888          */
 889         for_each_possible_cpu(cpu) {
 890                 unsigned long stack_page;
 891
 892                 stack_page = __get_free_page(GFP_KERNEL);
 893                 if (!stack_page) {
 894                         err = -ENOMEM;
 895                         goto out_free_stack_pages;
 896                 }
 897
 898                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 899         }
 900
 901         /*
 902          * Execute the init code on each CPU.
 903          *
 904          * Note: The stack is not mapped yet, so don't do anything else than
 905          * initializing the hypervisor mode on each CPU using a local stack
 906          * space for temporary storage.
 907          */
 908         init_phys_addr = virt_to_phys(__kvm_hyp_init);
 909         for_each_online_cpu(cpu) {
 910                 smp_call_function_single(cpu, cpu_init_hyp_mode,
 911                                          (void *)(long)init_phys_addr, 1);
 912         }
 913
 914         /*
 915          * Unmap the identity mapping
 916          */
 917         kvm_clear_hyp_idmap();
 918
 919         /*
 920          * Map the Hyp-code called directly from the host
 921          */
 922         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 923         if (err) {
 924                 kvm_err("Cannot map world-switch code\n");
 925                 goto out_free_mappings;
 926         }
 927
 928         /*
 929          * Map the Hyp stack pages
 930          */
 931         for_each_possible_cpu(cpu) {
 932                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 933                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 934
 935                 if (err) {
 936                         kvm_err("Cannot map hyp stack\n");
 937                         goto out_free_mappings;
 938                 }
 939         }
 940
 941         /*
 942          * Map the host VFP structures
 943          */
 944         kvm_host_vfp_state = alloc_percpu(struct vfp_hard_struct);
 945         if (!kvm_host_vfp_state) {
 946                 err = -ENOMEM;
 947                 kvm_err("Cannot allocate host VFP state\n");
 948                 goto out_free_mappings;
 949         }
 950
 951         for_each_possible_cpu(cpu) {
 952                 struct vfp_hard_struct *vfp;
 953
 954                 vfp = per_cpu_ptr(kvm_host_vfp_state, cpu);
 955                 err = create_hyp_mappings(vfp, vfp + 1);
 956
 957                 if (err) {
 958                         kvm_err("Cannot map host VFP state: %d\n", err);
 959                         goto out_free_vfp;
 960                 }
 961         }
 962
 963         kvm_info("Hyp mode initialized successfully\n");
 964         return 0;
 965 out_free_vfp:
 966         free_percpu(kvm_host_vfp_state);
 967 out_free_mappings:
 968         free_hyp_pmds();
 969 out_free_stack_pages:
 970         for_each_possible_cpu(cpu)
 971                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
 972 out_err:
 973         kvm_err("error initializing Hyp mode: %d\n", err);
 974         return err;
 975 }
 976
 977 /**
 978  * Initialize Hyp-mode and memory mappings on all CPUs.
 979  */
 980 int kvm_arch_init(void *opaque)
 981 {
 982         int err;
 983
 984         if (!is_hyp_mode_available()) {
 985                 kvm_err("HYP mode not available\n");
 986                 return -ENODEV;
 987         }
 988
 989         if (kvm_target_cpu() < 0) {
 990                 kvm_err("Target CPU not supported!\n");
 991                 return -ENODEV;
 992         }
 993
 994         err = init_hyp_mode();
 995         if (err)
 996                 goto out_err;
 997
 998         kvm_coproc_table_init();
 999         return 0;
1000 out_err:
1001         return err;
1002 }
1003
1004 /* NOP: Compiling as a module not supported */
1005 void kvm_arch_exit(void)
1006 {
1007 }
1008
1009 static int arm_init(void)
1010 {
1011         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1012         return rc;
1013 }
1014
1015 module_init(arm_init);