arch/arm/kvm/arm.c

   1 /*
   2  * Copyright (C) 2012 - Virtual Open Systems and Columbia University
   3  * Author: Christoffer Dall <c.dall@virtualopensystems.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write to the Free Software
  16  * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301, USA.
  17  */
  18
  19 #include <linux/cpu.h>
  20 #include <linux/cpu_pm.h>
  21 #include <linux/errno.h>
  22 #include <linux/err.h>
  23 #include <linux/kvm_host.h>
  24 #include <linux/module.h>
  25 #include <linux/vmalloc.h>
  26 #include <linux/fs.h>
  27 #include <linux/mman.h>
  28 #include <linux/sched.h>
  29 #include <linux/kvm.h>
  30 #include <trace/events/kvm.h>
  31
  32 #define CREATE_TRACE_POINTS
  33 #include "trace.h"
  34
  35 #include <asm/uaccess.h>
  36 #include <asm/ptrace.h>
  37 #include <asm/mman.h>
  38 #include <asm/tlbflush.h>
  39 #include <asm/cacheflush.h>
  40 #include <asm/virt.h>
  41 #include <asm/kvm_arm.h>
  42 #include <asm/kvm_asm.h>
  43 #include <asm/kvm_mmu.h>
  44 #include <asm/kvm_emulate.h>
  45 #include <asm/kvm_coproc.h>
  46 #include <asm/kvm_psci.h>
  47
  48 #ifdef REQUIRES_VIRT
  49 __asm__(".arch_extension        virt");
  50 #endif
  51
  52 static DEFINE_PER_CPU(unsigned long, kvm_arm_hyp_stack_page);
  53 static kvm_cpu_context_t __percpu *kvm_host_cpu_state;
  54 static unsigned long hyp_default_vectors;
  55
  56 /* Per-CPU variable containing the currently running vcpu. */
  57 static DEFINE_PER_CPU(struct kvm_vcpu *, kvm_arm_running_vcpu);
  58
  59 /* The VMID used in the VTTBR */
  60 static atomic64_t kvm_vmid_gen = ATOMIC64_INIT(1);
  61 static u8 kvm_next_vmid;
  62 static DEFINE_SPINLOCK(kvm_vmid_lock);
  63
  64 static bool vgic_present;
  65
  66 static void kvm_arm_set_running_vcpu(struct kvm_vcpu *vcpu)
  67 {
  68         BUG_ON(preemptible());
  69         __get_cpu_var(kvm_arm_running_vcpu) = vcpu;
  70 }
  71
  72 /**
  73  * kvm_arm_get_running_vcpu - get the vcpu running on the current CPU.
  74  * Must be called from non-preemptible context
  75  */
  76 struct kvm_vcpu *kvm_arm_get_running_vcpu(void)
  77 {
  78         BUG_ON(preemptible());
  79         return __get_cpu_var(kvm_arm_running_vcpu);
  80 }
  81
  82 /**
  83  * kvm_arm_get_running_vcpus - get the per-CPU array of currently running vcpus.
  84  */
  85 struct kvm_vcpu __percpu **kvm_get_running_vcpus(void)
  86 {
  87         return &kvm_arm_running_vcpu;
  88 }
  89
  90 int kvm_arch_hardware_enable(void *garbage)
  91 {
  92         return 0;
  93 }
  94
  95 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
  96 {
  97         return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
  98 }
  99
 100 void kvm_arch_hardware_disable(void *garbage)
 101 {
 102 }
 103
 104 int kvm_arch_hardware_setup(void)
 105 {
 106         return 0;
 107 }
 108
 109 void kvm_arch_hardware_unsetup(void)
 110 {
 111 }
 112
 113 void kvm_arch_check_processor_compat(void *rtn)
 114 {
 115         *(int *)rtn = 0;
 116 }
 117
 118 void kvm_arch_sync_events(struct kvm *kvm)
 119 {
 120 }
 121
 122 /**
 123  * kvm_arch_init_vm - initializes a VM data structure
 124  * @kvm:        pointer to the KVM struct
 125  */
 126 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
 127 {
 128         int ret = 0;
 129
 130         if (type)
 131                 return -EINVAL;
 132
 133         ret = kvm_alloc_stage2_pgd(kvm);
 134         if (ret)
 135                 goto out_fail_alloc;
 136
 137         ret = create_hyp_mappings(kvm, kvm + 1);
 138         if (ret)
 139                 goto out_free_stage2_pgd;
 140
 141         kvm_timer_init(kvm);
 142
 143         /* Mark the initial VMID generation invalid */
 144         kvm->arch.vmid_gen = 0;
 145
 146         return ret;
 147 out_free_stage2_pgd:
 148         kvm_free_stage2_pgd(kvm);
 149 out_fail_alloc:
 150         return ret;
 151 }
 152
 153 int kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
 154 {
 155         return VM_FAULT_SIGBUS;
 156 }
 157
 158 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
 159                            struct kvm_memory_slot *dont)
 160 {
 161 }
 162
 163 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
 164                             unsigned long npages)
 165 {
 166         return 0;
 167 }
 168
 169 /**
 170  * kvm_arch_destroy_vm - destroy the VM data structure
 171  * @kvm:        pointer to the KVM struct
 172  */
 173 void kvm_arch_destroy_vm(struct kvm *kvm)
 174 {
 175         int i;
 176
 177         kvm_free_stage2_pgd(kvm);
 178
 179         for (i = 0; i < KVM_MAX_VCPUS; ++i) {
 180                 if (kvm->vcpus[i]) {
 181                         kvm_arch_vcpu_free(kvm->vcpus[i]);
 182                         kvm->vcpus[i] = NULL;
 183                 }
 184         }
 185 }
 186
 187 int kvm_dev_ioctl_check_extension(long ext)
 188 {
 189         int r;
 190         switch (ext) {
 191         case KVM_CAP_IRQCHIP:
 192                 r = vgic_present;
 193                 break;
 194         case KVM_CAP_DEVICE_CTRL:
 195         case KVM_CAP_USER_MEMORY:
 196         case KVM_CAP_SYNC_MMU:
 197         case KVM_CAP_DESTROY_MEMORY_REGION_WORKS:
 198         case KVM_CAP_ONE_REG:
 199         case KVM_CAP_ARM_PSCI:
 200                 r = 1;
 201                 break;
 202         case KVM_CAP_COALESCED_MMIO:
 203                 r = KVM_COALESCED_MMIO_PAGE_OFFSET;
 204                 break;
 205         case KVM_CAP_ARM_SET_DEVICE_ADDR:
 206                 r = 1;
 207                 break;
 208         case KVM_CAP_NR_VCPUS:
 209                 r = num_online_cpus();
 210                 break;
 211         case KVM_CAP_MAX_VCPUS:
 212                 r = KVM_MAX_VCPUS;
 213                 break;
 214         default:
 215                 r = kvm_arch_dev_ioctl_check_extension(ext);
 216                 break;
 217         }
 218         return r;
 219 }
 220
 221 long kvm_arch_dev_ioctl(struct file *filp,
 222                         unsigned int ioctl, unsigned long arg)
 223 {
 224         return -EINVAL;
 225 }
 226
 227 void kvm_arch_memslots_updated(struct kvm *kvm)
 228 {
 229 }
 230
 231 int kvm_arch_prepare_memory_region(struct kvm *kvm,
 232                                    struct kvm_memory_slot *memslot,
 233                                    struct kvm_userspace_memory_region *mem,
 234                                    enum kvm_mr_change change)
 235 {
 236         return 0;
 237 }
 238
 239 void kvm_arch_commit_memory_region(struct kvm *kvm,
 240                                    struct kvm_userspace_memory_region *mem,
 241                                    const struct kvm_memory_slot *old,
 242                                    enum kvm_mr_change change)
 243 {
 244 }
 245
 246 void kvm_arch_flush_shadow_all(struct kvm *kvm)
 247 {
 248 }
 249
 250 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
 251                                    struct kvm_memory_slot *slot)
 252 {
 253 }
 254
 255 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm, unsigned int id)
 256 {
 257         int err;
 258         struct kvm_vcpu *vcpu;
 259
 260         vcpu = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
 261         if (!vcpu) {
 262                 err = -ENOMEM;
 263                 goto out;
 264         }
 265
 266         err = kvm_vcpu_init(vcpu, kvm, id);
 267         if (err)
 268                 goto free_vcpu;
 269
 270         err = create_hyp_mappings(vcpu, vcpu + 1);
 271         if (err)
 272                 goto vcpu_uninit;
 273
 274         return vcpu;
 275 vcpu_uninit:
 276         kvm_vcpu_uninit(vcpu);
 277 free_vcpu:
 278         kmem_cache_free(kvm_vcpu_cache, vcpu);
 279 out:
 280         return ERR_PTR(err);
 281 }
 282
 283 int kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
 284 {
 285         return 0;
 286 }
 287
 288 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
 289 {
 290         kvm_mmu_free_memory_caches(vcpu);
 291         kvm_timer_vcpu_terminate(vcpu);
 292         kmem_cache_free(kvm_vcpu_cache, vcpu);
 293 }
 294
 295 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
 296 {
 297         kvm_arch_vcpu_free(vcpu);
 298 }
 299
 300 int kvm_cpu_has_pending_timer(struct kvm_vcpu *vcpu)
 301 {
 302         return 0;
 303 }
 304
 305 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
 306 {
 307         int ret;
 308
 309         /* Force users to call KVM_ARM_VCPU_INIT */
 310         vcpu->arch.target = -1;
 311
 312         /* Set up VGIC */
 313         ret = kvm_vgic_vcpu_init(vcpu);
 314         if (ret)
 315                 return ret;
 316
 317         /* Set up the timer */
 318         kvm_timer_vcpu_init(vcpu);
 319
 320         return 0;
 321 }
 322
 323 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
 324 {
 325 }
 326
 327 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
 328 {
 329         vcpu->cpu = cpu;
 330         vcpu->arch.host_cpu_context = this_cpu_ptr(kvm_host_cpu_state);
 331
 332         /*
 333          * Check whether this vcpu requires the cache to be flushed on
 334          * this physical CPU. This is a consequence of doing dcache
 335          * operations by set/way on this vcpu. We do it here to be in
 336          * a non-preemptible section.
 337          */
 338         if (cpumask_test_and_clear_cpu(cpu, &vcpu->arch.require_dcache_flush))
 339                 flush_cache_all(); /* We'd really want v7_flush_dcache_all() */
 340
 341         kvm_arm_set_running_vcpu(vcpu);
 342 }
 343
 344 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
 345 {
 346         kvm_arm_set_running_vcpu(NULL);
 347 }
 348
 349 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
 350                                         struct kvm_guest_debug *dbg)
 351 {
 352         return -EINVAL;
 353 }
 354
 355
 356 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
 357                                     struct kvm_mp_state *mp_state)
 358 {
 359         return -EINVAL;
 360 }
 361
 362 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
 363                                     struct kvm_mp_state *mp_state)
 364 {
 365         return -EINVAL;
 366 }
 367
 368 /**
 369  * kvm_arch_vcpu_runnable - determine if the vcpu can be scheduled
 370  * @v:          The VCPU pointer
 371  *
 372  * If the guest CPU is not waiting for interrupts or an interrupt line is
 373  * asserted, the CPU is by definition runnable.
 374  */
 375 int kvm_arch_vcpu_runnable(struct kvm_vcpu *v)
 376 {
 377         return !!v->arch.irq_lines || kvm_vgic_vcpu_pending_irq(v);
 378 }
 379
 380 /* Just ensure a guest exit from a particular CPU */
 381 static void exit_vm_noop(void *info)
 382 {
 383 }
 384
 385 void force_vm_exit(const cpumask_t *mask)
 386 {
 387         smp_call_function_many(mask, exit_vm_noop, NULL, true);
 388 }
 389
 390 /**
 391  * need_new_vmid_gen - check that the VMID is still valid
 392  * @kvm: The VM's VMID to checkt
 393  *
 394  * return true if there is a new generation of VMIDs being used
 395  *
 396  * The hardware supports only 256 values with the value zero reserved for the
 397  * host, so we check if an assigned value belongs to a previous generation,
 398  * which which requires us to assign a new value. If we're the first to use a
 399  * VMID for the new generation, we must flush necessary caches and TLBs on all
 400  * CPUs.
 401  */
 402 static bool need_new_vmid_gen(struct kvm *kvm)
 403 {
 404         return unlikely(kvm->arch.vmid_gen != atomic64_read(&kvm_vmid_gen));
 405 }
 406
 407 /**
 408  * update_vttbr - Update the VTTBR with a valid VMID before the guest runs
 409  * @kvm The guest that we are about to run
 410  *
 411  * Called from kvm_arch_vcpu_ioctl_run before entering the guest to ensure the
 412  * VM has a valid VMID, otherwise assigns a new one and flushes corresponding
 413  * caches and TLBs.
 414  */
 415 static void update_vttbr(struct kvm *kvm)
 416 {
 417         phys_addr_t pgd_phys;
 418         u64 vmid;
 419
 420         if (!need_new_vmid_gen(kvm))
 421                 return;
 422
 423         spin_lock(&kvm_vmid_lock);
 424
 425         /*
 426          * We need to re-check the vmid_gen here to ensure that if another vcpu
 427          * already allocated a valid vmid for this vm, then this vcpu should
 428          * use the same vmid.
 429          */
 430         if (!need_new_vmid_gen(kvm)) {
 431                 spin_unlock(&kvm_vmid_lock);
 432                 return;
 433         }
 434
 435         /* First user of a new VMID generation? */
 436         if (unlikely(kvm_next_vmid == 0)) {
 437                 atomic64_inc(&kvm_vmid_gen);
 438                 kvm_next_vmid = 1;
 439
 440                 /*
 441                  * On SMP we know no other CPUs can use this CPU's or each
 442                  * other's VMID after force_vm_exit returns since the
 443                  * kvm_vmid_lock blocks them from reentry to the guest.
 444                  */
 445                 force_vm_exit(cpu_all_mask);
 446                 /*
 447                  * Now broadcast TLB + ICACHE invalidation over the inner
 448                  * shareable domain to make sure all data structures are
 449                  * clean.
 450                  */
 451                 kvm_call_hyp(__kvm_flush_vm_context);
 452         }
 453
 454         kvm->arch.vmid_gen = atomic64_read(&kvm_vmid_gen);
 455         kvm->arch.vmid = kvm_next_vmid;
 456         kvm_next_vmid++;
 457
 458         /* update vttbr to be used with the new vmid */
 459         pgd_phys = virt_to_phys(kvm->arch.pgd);
 460         vmid = ((u64)(kvm->arch.vmid) << VTTBR_VMID_SHIFT) & VTTBR_VMID_MASK;
 461         kvm->arch.vttbr = pgd_phys & VTTBR_BADDR_MASK;
 462         kvm->arch.vttbr |= vmid;
 463
 464         spin_unlock(&kvm_vmid_lock);
 465 }
 466
 467 static int kvm_vcpu_first_run_init(struct kvm_vcpu *vcpu)
 468 {
 469         int ret;
 470
 471         if (likely(vcpu->arch.has_run_once))
 472                 return 0;
 473
 474         vcpu->arch.has_run_once = true;
 475
 476         /*
 477          * Initialize the VGIC before running a vcpu the first time on
 478          * this VM.
 479          */
 480         if (unlikely(!vgic_initialized(vcpu->kvm))) {
 481                 ret = kvm_vgic_init(vcpu->kvm);
 482                 if (ret)
 483                         return ret;
 484         }
 485
 486         return 0;
 487 }
 488
 489 static void vcpu_pause(struct kvm_vcpu *vcpu)
 490 {
 491         wait_queue_head_t *wq = kvm_arch_vcpu_wq(vcpu);
 492
 493         wait_event_interruptible(*wq, !vcpu->arch.pause);
 494 }
 495
 496 static int kvm_vcpu_initialized(struct kvm_vcpu *vcpu)
 497 {
 498         return vcpu->arch.target >= 0;
 499 }
 500
 501 /**
 502  * kvm_arch_vcpu_ioctl_run - the main VCPU run function to execute guest code
 503  * @vcpu:       The VCPU pointer
 504  * @run:        The kvm_run structure pointer used for userspace state exchange
 505  *
 506  * This function is called through the VCPU_RUN ioctl called from user space. It
 507  * will execute VM code in a loop until the time slice for the process is used
 508  * or some emulation is needed from user space in which case the function will
 509  * return with return value 0 and with the kvm_run structure filled in with the
 510  * required data for the requested emulation.
 511  */
 512 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *run)
 513 {
 514         int ret;
 515         sigset_t sigsaved;
 516
 517         if (unlikely(!kvm_vcpu_initialized(vcpu)))
 518                 return -ENOEXEC;
 519
 520         ret = kvm_vcpu_first_run_init(vcpu);
 521         if (ret)
 522                 return ret;
 523
 524         if (run->exit_reason == KVM_EXIT_MMIO) {
 525                 ret = kvm_handle_mmio_return(vcpu, vcpu->run);
 526                 if (ret)
 527                         return ret;
 528         }
 529
 530         if (vcpu->sigset_active)
 531                 sigprocmask(SIG_SETMASK, &vcpu->sigset, &sigsaved);
 532
 533         ret = 1;
 534         run->exit_reason = KVM_EXIT_UNKNOWN;
 535         while (ret > 0) {
 536                 /*
 537                  * Check conditions before entering the guest
 538                  */
 539                 cond_resched();
 540
 541                 update_vttbr(vcpu->kvm);
 542
 543                 if (vcpu->arch.pause)
 544                         vcpu_pause(vcpu);
 545
 546                 kvm_vgic_flush_hwstate(vcpu);
 547                 kvm_timer_flush_hwstate(vcpu);
 548
 549                 local_irq_disable();
 550
 551                 /*
 552                  * Re-check atomic conditions
 553                  */
 554                 if (signal_pending(current)) {
 555                         ret = -EINTR;
 556                         run->exit_reason = KVM_EXIT_INTR;
 557                 }
 558
 559                 if (ret <= 0 || need_new_vmid_gen(vcpu->kvm)) {
 560                         local_irq_enable();
 561                         kvm_timer_sync_hwstate(vcpu);
 562                         kvm_vgic_sync_hwstate(vcpu);
 563                         continue;
 564                 }
 565
 566                 /**************************************************************
 567                  * Enter the guest
 568                  */
 569                 trace_kvm_entry(*vcpu_pc(vcpu));
 570                 kvm_guest_enter();
 571                 vcpu->mode = IN_GUEST_MODE;
 572
 573                 ret = kvm_call_hyp(__kvm_vcpu_run, vcpu);
 574
 575                 vcpu->mode = OUTSIDE_GUEST_MODE;
 576                 vcpu->arch.last_pcpu = smp_processor_id();
 577                 kvm_guest_exit();
 578                 trace_kvm_exit(*vcpu_pc(vcpu));
 579                 /*
 580                  * We may have taken a host interrupt in HYP mode (ie
 581                  * while executing the guest). This interrupt is still
 582                  * pending, as we haven't serviced it yet!
 583                  *
 584                  * We're now back in SVC mode, with interrupts
 585                  * disabled.  Enabling the interrupts now will have
 586                  * the effect of taking the interrupt again, in SVC
 587                  * mode this time.
 588                  */
 589                 local_irq_enable();
 590
 591                 /*
 592                  * Back from guest
 593                  *************************************************************/
 594
 595                 kvm_timer_sync_hwstate(vcpu);
 596                 kvm_vgic_sync_hwstate(vcpu);
 597
 598                 ret = handle_exit(vcpu, run, ret);
 599         }
 600
 601         if (vcpu->sigset_active)
 602                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 603         return ret;
 604 }
 605
 606 static int vcpu_interrupt_line(struct kvm_vcpu *vcpu, int number, bool level)
 607 {
 608         int bit_index;
 609         bool set;
 610         unsigned long *ptr;
 611
 612         if (number == KVM_ARM_IRQ_CPU_IRQ)
 613                 bit_index = __ffs(HCR_VI);
 614         else /* KVM_ARM_IRQ_CPU_FIQ */
 615                 bit_index = __ffs(HCR_VF);
 616
 617         ptr = (unsigned long *)&vcpu->arch.irq_lines;
 618         if (level)
 619                 set = test_and_set_bit(bit_index, ptr);
 620         else
 621                 set = test_and_clear_bit(bit_index, ptr);
 622
 623         /*
 624          * If we didn't change anything, no need to wake up or kick other CPUs
 625          */
 626         if (set == level)
 627                 return 0;
 628
 629         /*
 630          * The vcpu irq_lines field was updated, wake up sleeping VCPUs and
 631          * trigger a world-switch round on the running physical CPU to set the
 632          * virtual IRQ/FIQ fields in the HCR appropriately.
 633          */
 634         kvm_vcpu_kick(vcpu);
 635
 636         return 0;
 637 }
 638
 639 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_level,
 640                           bool line_status)
 641 {
 642         u32 irq = irq_level->irq;
 643         unsigned int irq_type, vcpu_idx, irq_num;
 644         int nrcpus = atomic_read(&kvm->online_vcpus);
 645         struct kvm_vcpu *vcpu = NULL;
 646         bool level = irq_level->level;
 647
 648         irq_type = (irq >> KVM_ARM_IRQ_TYPE_SHIFT) & KVM_ARM_IRQ_TYPE_MASK;
 649         vcpu_idx = (irq >> KVM_ARM_IRQ_VCPU_SHIFT) & KVM_ARM_IRQ_VCPU_MASK;
 650         irq_num = (irq >> KVM_ARM_IRQ_NUM_SHIFT) & KVM_ARM_IRQ_NUM_MASK;
 651
 652         trace_kvm_irq_line(irq_type, vcpu_idx, irq_num, irq_level->level);
 653
 654         switch (irq_type) {
 655         case KVM_ARM_IRQ_TYPE_CPU:
 656                 if (irqchip_in_kernel(kvm))
 657                         return -ENXIO;
 658
 659                 if (vcpu_idx >= nrcpus)
 660                         return -EINVAL;
 661
 662                 vcpu = kvm_get_vcpu(kvm, vcpu_idx);
 663                 if (!vcpu)
 664                         return -EINVAL;
 665
 666                 if (irq_num > KVM_ARM_IRQ_CPU_FIQ)
 667                         return -EINVAL;
 668
 669                 return vcpu_interrupt_line(vcpu, irq_num, level);
 670         case KVM_ARM_IRQ_TYPE_PPI:
 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 < VGIC_NR_SGIS || irq_num >= VGIC_NR_PRIVATE_IRQS)
 682                         return -EINVAL;
 683
 684                 return kvm_vgic_inject_irq(kvm, vcpu->vcpu_id, irq_num, level);
 685         case KVM_ARM_IRQ_TYPE_SPI:
 686                 if (!irqchip_in_kernel(kvm))
 687                         return -ENXIO;
 688
 689                 if (irq_num < VGIC_NR_PRIVATE_IRQS ||
 690                     irq_num > KVM_ARM_IRQ_GIC_MAX)
 691                         return -EINVAL;
 692
 693                 return kvm_vgic_inject_irq(kvm, 0, irq_num, level);
 694         }
 695
 696         return -EINVAL;
 697 }
 698
 699 static int kvm_arch_vcpu_ioctl_vcpu_init(struct kvm_vcpu *vcpu,
 700                                          struct kvm_vcpu_init *init)
 701 {
 702         int ret;
 703
 704         ret = kvm_vcpu_set_target(vcpu, init);
 705         if (ret)
 706                 return ret;
 707
 708         /*
 709          * Handle the "start in power-off" case by marking the VCPU as paused.
 710          */
 711         if (__test_and_clear_bit(KVM_ARM_VCPU_POWER_OFF, vcpu->arch.features))
 712                 vcpu->arch.pause = true;
 713
 714         return 0;
 715 }
 716
 717 long kvm_arch_vcpu_ioctl(struct file *filp,
 718                          unsigned int ioctl, unsigned long arg)
 719 {
 720         struct kvm_vcpu *vcpu = filp->private_data;
 721         void __user *argp = (void __user *)arg;
 722
 723         switch (ioctl) {
 724         case KVM_ARM_VCPU_INIT: {
 725                 struct kvm_vcpu_init init;
 726
 727                 if (copy_from_user(&init, argp, sizeof(init)))
 728                         return -EFAULT;
 729
 730                 return kvm_arch_vcpu_ioctl_vcpu_init(vcpu, &init);
 731         }
 732         case KVM_SET_ONE_REG:
 733         case KVM_GET_ONE_REG: {
 734                 struct kvm_one_reg reg;
 735
 736                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 737                         return -ENOEXEC;
 738
 739                 if (copy_from_user(&reg, argp, sizeof(reg)))
 740                         return -EFAULT;
 741                 if (ioctl == KVM_SET_ONE_REG)
 742                         return kvm_arm_set_reg(vcpu, &reg);
 743                 else
 744                         return kvm_arm_get_reg(vcpu, &reg);
 745         }
 746         case KVM_GET_REG_LIST: {
 747                 struct kvm_reg_list __user *user_list = argp;
 748                 struct kvm_reg_list reg_list;
 749                 unsigned n;
 750
 751                 if (unlikely(!kvm_vcpu_initialized(vcpu)))
 752                         return -ENOEXEC;
 753
 754                 if (copy_from_user(&reg_list, user_list, sizeof(reg_list)))
 755                         return -EFAULT;
 756                 n = reg_list.n;
 757                 reg_list.n = kvm_arm_num_regs(vcpu);
 758                 if (copy_to_user(user_list, &reg_list, sizeof(reg_list)))
 759                         return -EFAULT;
 760                 if (n < reg_list.n)
 761                         return -E2BIG;
 762                 return kvm_arm_copy_reg_indices(vcpu, user_list->reg);
 763         }
 764         default:
 765                 return -EINVAL;
 766         }
 767 }
 768
 769 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
 770 {
 771         return -EINVAL;
 772 }
 773
 774 static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
 775                                         struct kvm_arm_device_addr *dev_addr)
 776 {
 777         unsigned long dev_id, type;
 778
 779         dev_id = (dev_addr->id & KVM_ARM_DEVICE_ID_MASK) >>
 780                 KVM_ARM_DEVICE_ID_SHIFT;
 781         type = (dev_addr->id & KVM_ARM_DEVICE_TYPE_MASK) >>
 782                 KVM_ARM_DEVICE_TYPE_SHIFT;
 783
 784         switch (dev_id) {
 785         case KVM_ARM_DEVICE_VGIC_V2:
 786                 if (!vgic_present)
 787                         return -ENXIO;
 788                 return kvm_vgic_set_addr(kvm, type, dev_addr->addr);
 789         default:
 790                 return -ENODEV;
 791         }
 792 }
 793
 794 long kvm_arch_vm_ioctl(struct file *filp,
 795                        unsigned int ioctl, unsigned long arg)
 796 {
 797         struct kvm *kvm = filp->private_data;
 798         void __user *argp = (void __user *)arg;
 799
 800         switch (ioctl) {
 801         case KVM_CREATE_IRQCHIP: {
 802                 if (vgic_present)
 803                         return kvm_vgic_create(kvm);
 804                 else
 805                         return -ENXIO;
 806         }
 807         case KVM_ARM_SET_DEVICE_ADDR: {
 808                 struct kvm_arm_device_addr dev_addr;
 809
 810                 if (copy_from_user(&dev_addr, argp, sizeof(dev_addr)))
 811                         return -EFAULT;
 812                 return kvm_vm_ioctl_set_device_addr(kvm, &dev_addr);
 813         }
 814         case KVM_ARM_PREFERRED_TARGET: {
 815                 int err;
 816                 struct kvm_vcpu_init init;
 817
 818                 err = kvm_vcpu_preferred_target(&init);
 819                 if (err)
 820                         return err;
 821
 822                 if (copy_to_user(argp, &init, sizeof(init)))
 823                         return -EFAULT;
 824
 825                 return 0;
 826         }
 827         default:
 828                 return -EINVAL;
 829         }
 830 }
 831
 832 static void cpu_init_hyp_mode(void *dummy)
 833 {
 834         phys_addr_t boot_pgd_ptr;
 835         phys_addr_t pgd_ptr;
 836         unsigned long hyp_stack_ptr;
 837         unsigned long stack_page;
 838         unsigned long vector_ptr;
 839
 840         /* Switch from the HYP stub to our own HYP init vector */
 841         __hyp_set_vectors(kvm_get_idmap_vector());
 842
 843         boot_pgd_ptr = kvm_mmu_get_boot_httbr();
 844         pgd_ptr = kvm_mmu_get_httbr();
 845         stack_page = __get_cpu_var(kvm_arm_hyp_stack_page);
 846         hyp_stack_ptr = stack_page + PAGE_SIZE;
 847         vector_ptr = (unsigned long)__kvm_hyp_vector;
 848
 849         __cpu_init_hyp_mode(boot_pgd_ptr, pgd_ptr, hyp_stack_ptr, vector_ptr);
 850 }
 851
 852 static int hyp_init_cpu_notify(struct notifier_block *self,
 853                                unsigned long action, void *cpu)
 854 {
 855         switch (action) {
 856         case CPU_STARTING:
 857         case CPU_STARTING_FROZEN:
 858                 cpu_init_hyp_mode(NULL);
 859                 break;
 860         }
 861
 862         return NOTIFY_OK;
 863 }
 864
 865 static struct notifier_block hyp_init_cpu_nb = {
 866         .notifier_call = hyp_init_cpu_notify,
 867 };
 868
 869 #ifdef CONFIG_CPU_PM
 870 static int hyp_init_cpu_pm_notifier(struct notifier_block *self,
 871                                     unsigned long cmd,
 872                                     void *v)
 873 {
 874         if (cmd == CPU_PM_EXIT) {
 875                 cpu_init_hyp_mode(NULL);
 876                 return NOTIFY_OK;
 877         }
 878
 879         return NOTIFY_DONE;
 880 }
 881
 882 static struct notifier_block hyp_init_cpu_pm_nb = {
 883         .notifier_call = hyp_init_cpu_pm_notifier,
 884 };
 885
 886 static void __init hyp_cpu_pm_init(void)
 887 {
 888         cpu_pm_register_notifier(&hyp_init_cpu_pm_nb);
 889 }
 890 #else
 891 static inline void hyp_cpu_pm_init(void)
 892 {
 893 }
 894 #endif
 895
 896 /**
 897  * Inits Hyp-mode on all online CPUs
 898  */
 899 static int init_hyp_mode(void)
 900 {
 901         int cpu;
 902         int err = 0;
 903
 904         /*
 905          * Allocate Hyp PGD and setup Hyp identity mapping
 906          */
 907         err = kvm_mmu_init();
 908         if (err)
 909                 goto out_err;
 910
 911         /*
 912          * It is probably enough to obtain the default on one
 913          * CPU. It's unlikely to be different on the others.
 914          */
 915         hyp_default_vectors = __hyp_get_vectors();
 916
 917         /*
 918          * Allocate stack pages for Hypervisor-mode
 919          */
 920         for_each_possible_cpu(cpu) {
 921                 unsigned long stack_page;
 922
 923                 stack_page = __get_free_page(GFP_KERNEL);
 924                 if (!stack_page) {
 925                         err = -ENOMEM;
 926                         goto out_free_stack_pages;
 927                 }
 928
 929                 per_cpu(kvm_arm_hyp_stack_page, cpu) = stack_page;
 930         }
 931
 932         /*
 933          * Map the Hyp-code called directly from the host
 934          */
 935         err = create_hyp_mappings(__kvm_hyp_code_start, __kvm_hyp_code_end);
 936         if (err) {
 937                 kvm_err("Cannot map world-switch code\n");
 938                 goto out_free_mappings;
 939         }
 940
 941         /*
 942          * Map the Hyp stack pages
 943          */
 944         for_each_possible_cpu(cpu) {
 945                 char *stack_page = (char *)per_cpu(kvm_arm_hyp_stack_page, cpu);
 946                 err = create_hyp_mappings(stack_page, stack_page + PAGE_SIZE);
 947
 948                 if (err) {
 949                         kvm_err("Cannot map hyp stack\n");
 950                         goto out_free_mappings;
 951                 }
 952         }
 953
 954         /*
 955          * Map the host CPU structures
 956          */
 957         kvm_host_cpu_state = alloc_percpu(kvm_cpu_context_t);
 958         if (!kvm_host_cpu_state) {
 959                 err = -ENOMEM;
 960                 kvm_err("Cannot allocate host CPU state\n");
 961                 goto out_free_mappings;
 962         }
 963
 964         for_each_possible_cpu(cpu) {
 965                 kvm_cpu_context_t *cpu_ctxt;
 966
 967                 cpu_ctxt = per_cpu_ptr(kvm_host_cpu_state, cpu);
 968                 err = create_hyp_mappings(cpu_ctxt, cpu_ctxt + 1);
 969
 970                 if (err) {
 971                         kvm_err("Cannot map host CPU state: %d\n", err);
 972                         goto out_free_context;
 973                 }
 974         }
 975
 976         /*
 977          * Execute the init code on each CPU.
 978          */
 979         on_each_cpu(cpu_init_hyp_mode, NULL, 1);
 980
 981         /*
 982          * Init HYP view of VGIC
 983          */
 984         err = kvm_vgic_hyp_init();
 985         if (err)
 986                 goto out_free_context;
 987
 988 #ifdef CONFIG_KVM_ARM_VGIC
 989                 vgic_present = true;
 990 #endif
 991
 992         /*
 993          * Init HYP architected timer support
 994          */
 995         err = kvm_timer_hyp_init();
 996         if (err)
 997                 goto out_free_mappings;
 998
 999 #ifndef CONFIG_HOTPLUG_CPU
1000         free_boot_hyp_pgd();
1001 #endif
1002
1003         kvm_perf_init();
1004
1005         kvm_info("Hyp mode initialized successfully\n");
1006
1007         return 0;
1008 out_free_context:
1009         free_percpu(kvm_host_cpu_state);
1010 out_free_mappings:
1011         free_hyp_pgds();
1012 out_free_stack_pages:
1013         for_each_possible_cpu(cpu)
1014                 free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
1015 out_err:
1016         kvm_err("error initializing Hyp mode: %d\n", err);
1017         return err;
1018 }
1019
1020 static void check_kvm_target_cpu(void *ret)
1021 {
1022         *(int *)ret = kvm_target_cpu();
1023 }
1024
1025 /**
1026  * Initialize Hyp-mode and memory mappings on all CPUs.
1027  */
1028 int kvm_arch_init(void *opaque)
1029 {
1030         int err;
1031         int ret, cpu;
1032
1033         if (!is_hyp_mode_available()) {
1034                 kvm_err("HYP mode not available\n");
1035                 return -ENODEV;
1036         }
1037
1038         for_each_online_cpu(cpu) {
1039                 smp_call_function_single(cpu, check_kvm_target_cpu, &ret, 1);
1040                 if (ret < 0) {
1041                         kvm_err("Error, CPU %d not supported!\n", cpu);
1042                         return -ENODEV;
1043                 }
1044         }
1045
1046         err = init_hyp_mode();
1047         if (err)
1048                 goto out_err;
1049
1050         err = register_cpu_notifier(&hyp_init_cpu_nb);
1051         if (err) {
1052                 kvm_err("Cannot register HYP init CPU notifier (%d)\n", err);
1053                 goto out_err;
1054         }
1055
1056         hyp_cpu_pm_init();
1057
1058         kvm_coproc_table_init();
1059         return 0;
1060 out_err:
1061         return err;
1062 }
1063
1064 /* NOP: Compiling as a module not supported */
1065 void kvm_arch_exit(void)
1066 {
1067         kvm_perf_teardown();
1068 }
1069
1070 static int arm_init(void)
1071 {
1072         int rc = kvm_init(NULL, sizeof(struct kvm_vcpu), 0, THIS_MODULE);
1073         return rc;
1074 }
1075
1076 module_init(arm_init);